1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE = 0,
50 CHUNK_ALLOC_LIMITED = 1,
51 CHUNK_ALLOC_FORCE = 2,
55 * Declare a helper function to detect underflow of various space info members
57 #define DECLARE_SPACE_INFO_UPDATE(name) \
58 static inline void update_##name(struct btrfs_space_info *sinfo, \
61 if (bytes < 0 && sinfo->name < -bytes) { \
66 sinfo->name += bytes; \
69 DECLARE_SPACE_INFO_UPDATE(bytes_may_use);
70 DECLARE_SPACE_INFO_UPDATE(bytes_pinned);
72 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
73 struct btrfs_delayed_ref_node *node, u64 parent,
74 u64 root_objectid, u64 owner_objectid,
75 u64 owner_offset, int refs_to_drop,
76 struct btrfs_delayed_extent_op *extra_op);
77 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
78 struct extent_buffer *leaf,
79 struct btrfs_extent_item *ei);
80 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
81 u64 parent, u64 root_objectid,
82 u64 flags, u64 owner, u64 offset,
83 struct btrfs_key *ins, int ref_mod);
84 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
85 struct btrfs_delayed_ref_node *node,
86 struct btrfs_delayed_extent_op *extent_op);
87 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
96 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
97 struct btrfs_space_info *space_info,
99 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
100 struct btrfs_space_info *space_info,
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED ||
108 cache->cached == BTRFS_CACHE_ERROR;
111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
113 return (cache->flags & bits) == bits;
116 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
118 atomic_inc(&cache->count);
121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
123 if (atomic_dec_and_test(&cache->count)) {
124 WARN_ON(cache->pinned > 0);
125 WARN_ON(cache->reserved > 0);
128 * If not empty, someone is still holding mutex of
129 * full_stripe_lock, which can only be released by caller.
130 * And it will definitely cause use-after-free when caller
131 * tries to release full stripe lock.
133 * No better way to resolve, but only to warn.
135 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
136 kfree(cache->free_space_ctl);
142 * this adds the block group to the fs_info rb tree for the block group
145 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
146 struct btrfs_block_group_cache *block_group)
149 struct rb_node *parent = NULL;
150 struct btrfs_block_group_cache *cache;
152 spin_lock(&info->block_group_cache_lock);
153 p = &info->block_group_cache_tree.rb_node;
157 cache = rb_entry(parent, struct btrfs_block_group_cache,
159 if (block_group->key.objectid < cache->key.objectid) {
161 } else if (block_group->key.objectid > cache->key.objectid) {
164 spin_unlock(&info->block_group_cache_lock);
169 rb_link_node(&block_group->cache_node, parent, p);
170 rb_insert_color(&block_group->cache_node,
171 &info->block_group_cache_tree);
173 if (info->first_logical_byte > block_group->key.objectid)
174 info->first_logical_byte = block_group->key.objectid;
176 spin_unlock(&info->block_group_cache_lock);
182 * This will return the block group at or after bytenr if contains is 0, else
183 * it will return the block group that contains the bytenr
185 static struct btrfs_block_group_cache *
186 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 struct btrfs_block_group_cache *cache, *ret = NULL;
193 spin_lock(&info->block_group_cache_lock);
194 n = info->block_group_cache_tree.rb_node;
197 cache = rb_entry(n, struct btrfs_block_group_cache,
199 end = cache->key.objectid + cache->key.offset - 1;
200 start = cache->key.objectid;
202 if (bytenr < start) {
203 if (!contains && (!ret || start < ret->key.objectid))
206 } else if (bytenr > start) {
207 if (contains && bytenr <= end) {
218 btrfs_get_block_group(ret);
219 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
220 info->first_logical_byte = ret->key.objectid;
222 spin_unlock(&info->block_group_cache_lock);
227 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
228 u64 start, u64 num_bytes)
230 u64 end = start + num_bytes - 1;
231 set_extent_bits(&fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE);
233 set_extent_bits(&fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE);
238 static void free_excluded_extents(struct btrfs_block_group_cache *cache)
240 struct btrfs_fs_info *fs_info = cache->fs_info;
243 start = cache->key.objectid;
244 end = start + cache->key.offset - 1;
246 clear_extent_bits(&fs_info->freed_extents[0],
247 start, end, EXTENT_UPTODATE);
248 clear_extent_bits(&fs_info->freed_extents[1],
249 start, end, EXTENT_UPTODATE);
252 static int exclude_super_stripes(struct btrfs_block_group_cache *cache)
254 struct btrfs_fs_info *fs_info = cache->fs_info;
260 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
261 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(fs_info, cache->key.objectid,
269 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
270 bytenr = btrfs_sb_offset(i);
271 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
272 bytenr, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(fs_info, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (!cache->caching_ctl) {
316 spin_unlock(&cache->lock);
320 ctl = cache->caching_ctl;
321 refcount_inc(&ctl->count);
322 spin_unlock(&cache->lock);
326 static void put_caching_control(struct btrfs_caching_control *ctl)
328 if (refcount_dec_and_test(&ctl->count))
332 #ifdef CONFIG_BTRFS_DEBUG
333 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
335 struct btrfs_fs_info *fs_info = block_group->fs_info;
336 u64 start = block_group->key.objectid;
337 u64 len = block_group->key.offset;
338 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
339 fs_info->nodesize : fs_info->sectorsize;
340 u64 step = chunk << 1;
342 while (len > chunk) {
343 btrfs_remove_free_space(block_group, start, chunk);
354 * this is only called by cache_block_group, since we could have freed extents
355 * we need to check the pinned_extents for any extents that can't be used yet
356 * since their free space will be released as soon as the transaction commits.
358 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
361 struct btrfs_fs_info *info = block_group->fs_info;
362 u64 extent_start, extent_end, size, total_added = 0;
365 while (start < end) {
366 ret = find_first_extent_bit(info->pinned_extents, start,
367 &extent_start, &extent_end,
368 EXTENT_DIRTY | EXTENT_UPTODATE,
373 if (extent_start <= start) {
374 start = extent_end + 1;
375 } else if (extent_start > start && extent_start < end) {
376 size = extent_start - start;
378 ret = btrfs_add_free_space(block_group, start,
380 BUG_ON(ret); /* -ENOMEM or logic error */
381 start = extent_end + 1;
390 ret = btrfs_add_free_space(block_group, start, size);
391 BUG_ON(ret); /* -ENOMEM or logic error */
397 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
399 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
400 struct btrfs_fs_info *fs_info = block_group->fs_info;
401 struct btrfs_root *extent_root = fs_info->extent_root;
402 struct btrfs_path *path;
403 struct extent_buffer *leaf;
404 struct btrfs_key key;
411 path = btrfs_alloc_path();
415 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
417 #ifdef CONFIG_BTRFS_DEBUG
419 * If we're fragmenting we don't want to make anybody think we can
420 * allocate from this block group until we've had a chance to fragment
423 if (btrfs_should_fragment_free_space(block_group))
427 * We don't want to deadlock with somebody trying to allocate a new
428 * extent for the extent root while also trying to search the extent
429 * root to add free space. So we skip locking and search the commit
430 * root, since its read-only
432 path->skip_locking = 1;
433 path->search_commit_root = 1;
434 path->reada = READA_FORWARD;
438 key.type = BTRFS_EXTENT_ITEM_KEY;
441 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
445 leaf = path->nodes[0];
446 nritems = btrfs_header_nritems(leaf);
449 if (btrfs_fs_closing(fs_info) > 1) {
454 if (path->slots[0] < nritems) {
455 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
457 ret = find_next_key(path, 0, &key);
461 if (need_resched() ||
462 rwsem_is_contended(&fs_info->commit_root_sem)) {
464 caching_ctl->progress = last;
465 btrfs_release_path(path);
466 up_read(&fs_info->commit_root_sem);
467 mutex_unlock(&caching_ctl->mutex);
469 mutex_lock(&caching_ctl->mutex);
470 down_read(&fs_info->commit_root_sem);
474 ret = btrfs_next_leaf(extent_root, path);
479 leaf = path->nodes[0];
480 nritems = btrfs_header_nritems(leaf);
484 if (key.objectid < last) {
487 key.type = BTRFS_EXTENT_ITEM_KEY;
490 caching_ctl->progress = last;
491 btrfs_release_path(path);
495 if (key.objectid < block_group->key.objectid) {
500 if (key.objectid >= block_group->key.objectid +
501 block_group->key.offset)
504 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
505 key.type == BTRFS_METADATA_ITEM_KEY) {
506 total_found += add_new_free_space(block_group, last,
508 if (key.type == BTRFS_METADATA_ITEM_KEY)
509 last = key.objectid +
512 last = key.objectid + key.offset;
514 if (total_found > CACHING_CTL_WAKE_UP) {
517 wake_up(&caching_ctl->wait);
524 total_found += add_new_free_space(block_group, last,
525 block_group->key.objectid +
526 block_group->key.offset);
527 caching_ctl->progress = (u64)-1;
530 btrfs_free_path(path);
534 static noinline void caching_thread(struct btrfs_work *work)
536 struct btrfs_block_group_cache *block_group;
537 struct btrfs_fs_info *fs_info;
538 struct btrfs_caching_control *caching_ctl;
541 caching_ctl = container_of(work, struct btrfs_caching_control, work);
542 block_group = caching_ctl->block_group;
543 fs_info = block_group->fs_info;
545 mutex_lock(&caching_ctl->mutex);
546 down_read(&fs_info->commit_root_sem);
548 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
549 ret = load_free_space_tree(caching_ctl);
551 ret = load_extent_tree_free(caching_ctl);
553 spin_lock(&block_group->lock);
554 block_group->caching_ctl = NULL;
555 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
556 spin_unlock(&block_group->lock);
558 #ifdef CONFIG_BTRFS_DEBUG
559 if (btrfs_should_fragment_free_space(block_group)) {
562 spin_lock(&block_group->space_info->lock);
563 spin_lock(&block_group->lock);
564 bytes_used = block_group->key.offset -
565 btrfs_block_group_used(&block_group->item);
566 block_group->space_info->bytes_used += bytes_used >> 1;
567 spin_unlock(&block_group->lock);
568 spin_unlock(&block_group->space_info->lock);
569 fragment_free_space(block_group);
573 caching_ctl->progress = (u64)-1;
575 up_read(&fs_info->commit_root_sem);
576 free_excluded_extents(block_group);
577 mutex_unlock(&caching_ctl->mutex);
579 wake_up(&caching_ctl->wait);
581 put_caching_control(caching_ctl);
582 btrfs_put_block_group(block_group);
585 static int cache_block_group(struct btrfs_block_group_cache *cache,
589 struct btrfs_fs_info *fs_info = cache->fs_info;
590 struct btrfs_caching_control *caching_ctl;
593 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
597 INIT_LIST_HEAD(&caching_ctl->list);
598 mutex_init(&caching_ctl->mutex);
599 init_waitqueue_head(&caching_ctl->wait);
600 caching_ctl->block_group = cache;
601 caching_ctl->progress = cache->key.objectid;
602 refcount_set(&caching_ctl->count, 1);
603 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
604 caching_thread, NULL, NULL);
606 spin_lock(&cache->lock);
608 * This should be a rare occasion, but this could happen I think in the
609 * case where one thread starts to load the space cache info, and then
610 * some other thread starts a transaction commit which tries to do an
611 * allocation while the other thread is still loading the space cache
612 * info. The previous loop should have kept us from choosing this block
613 * group, but if we've moved to the state where we will wait on caching
614 * block groups we need to first check if we're doing a fast load here,
615 * so we can wait for it to finish, otherwise we could end up allocating
616 * from a block group who's cache gets evicted for one reason or
619 while (cache->cached == BTRFS_CACHE_FAST) {
620 struct btrfs_caching_control *ctl;
622 ctl = cache->caching_ctl;
623 refcount_inc(&ctl->count);
624 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
625 spin_unlock(&cache->lock);
629 finish_wait(&ctl->wait, &wait);
630 put_caching_control(ctl);
631 spin_lock(&cache->lock);
634 if (cache->cached != BTRFS_CACHE_NO) {
635 spin_unlock(&cache->lock);
639 WARN_ON(cache->caching_ctl);
640 cache->caching_ctl = caching_ctl;
641 cache->cached = BTRFS_CACHE_FAST;
642 spin_unlock(&cache->lock);
644 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
645 mutex_lock(&caching_ctl->mutex);
646 ret = load_free_space_cache(fs_info, cache);
648 spin_lock(&cache->lock);
650 cache->caching_ctl = NULL;
651 cache->cached = BTRFS_CACHE_FINISHED;
652 cache->last_byte_to_unpin = (u64)-1;
653 caching_ctl->progress = (u64)-1;
655 if (load_cache_only) {
656 cache->caching_ctl = NULL;
657 cache->cached = BTRFS_CACHE_NO;
659 cache->cached = BTRFS_CACHE_STARTED;
660 cache->has_caching_ctl = 1;
663 spin_unlock(&cache->lock);
664 #ifdef CONFIG_BTRFS_DEBUG
666 btrfs_should_fragment_free_space(cache)) {
669 spin_lock(&cache->space_info->lock);
670 spin_lock(&cache->lock);
671 bytes_used = cache->key.offset -
672 btrfs_block_group_used(&cache->item);
673 cache->space_info->bytes_used += bytes_used >> 1;
674 spin_unlock(&cache->lock);
675 spin_unlock(&cache->space_info->lock);
676 fragment_free_space(cache);
679 mutex_unlock(&caching_ctl->mutex);
681 wake_up(&caching_ctl->wait);
683 put_caching_control(caching_ctl);
684 free_excluded_extents(cache);
689 * We're either using the free space tree or no caching at all.
690 * Set cached to the appropriate value and wakeup any waiters.
692 spin_lock(&cache->lock);
693 if (load_cache_only) {
694 cache->caching_ctl = NULL;
695 cache->cached = BTRFS_CACHE_NO;
697 cache->cached = BTRFS_CACHE_STARTED;
698 cache->has_caching_ctl = 1;
700 spin_unlock(&cache->lock);
701 wake_up(&caching_ctl->wait);
704 if (load_cache_only) {
705 put_caching_control(caching_ctl);
709 down_write(&fs_info->commit_root_sem);
710 refcount_inc(&caching_ctl->count);
711 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
712 up_write(&fs_info->commit_root_sem);
714 btrfs_get_block_group(cache);
716 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
722 * return the block group that starts at or after bytenr
724 static struct btrfs_block_group_cache *
725 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
727 return block_group_cache_tree_search(info, bytenr, 0);
731 * return the block group that contains the given bytenr
733 struct btrfs_block_group_cache *btrfs_lookup_block_group(
734 struct btrfs_fs_info *info,
737 return block_group_cache_tree_search(info, bytenr, 1);
740 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
743 struct list_head *head = &info->space_info;
744 struct btrfs_space_info *found;
746 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
749 list_for_each_entry_rcu(found, head, list) {
750 if (found->flags & flags) {
759 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
760 bool metadata, u64 root_objectid)
762 struct btrfs_space_info *space_info;
766 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
767 flags = BTRFS_BLOCK_GROUP_SYSTEM;
769 flags = BTRFS_BLOCK_GROUP_METADATA;
771 flags = BTRFS_BLOCK_GROUP_DATA;
774 space_info = __find_space_info(fs_info, flags);
776 percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes,
777 BTRFS_TOTAL_BYTES_PINNED_BATCH);
781 * after adding space to the filesystem, we need to clear the full flags
782 * on all the space infos.
784 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
786 struct list_head *head = &info->space_info;
787 struct btrfs_space_info *found;
790 list_for_each_entry_rcu(found, head, list)
795 /* simple helper to search for an existing data extent at a given offset */
796 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
799 struct btrfs_key key;
800 struct btrfs_path *path;
802 path = btrfs_alloc_path();
806 key.objectid = start;
808 key.type = BTRFS_EXTENT_ITEM_KEY;
809 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
810 btrfs_free_path(path);
815 * helper function to lookup reference count and flags of a tree block.
817 * the head node for delayed ref is used to store the sum of all the
818 * reference count modifications queued up in the rbtree. the head
819 * node may also store the extent flags to set. This way you can check
820 * to see what the reference count and extent flags would be if all of
821 * the delayed refs are not processed.
823 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info, u64 bytenr,
825 u64 offset, int metadata, u64 *refs, u64 *flags)
827 struct btrfs_delayed_ref_head *head;
828 struct btrfs_delayed_ref_root *delayed_refs;
829 struct btrfs_path *path;
830 struct btrfs_extent_item *ei;
831 struct extent_buffer *leaf;
832 struct btrfs_key key;
839 * If we don't have skinny metadata, don't bother doing anything
842 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
843 offset = fs_info->nodesize;
847 path = btrfs_alloc_path();
852 path->skip_locking = 1;
853 path->search_commit_root = 1;
857 key.objectid = bytenr;
860 key.type = BTRFS_METADATA_ITEM_KEY;
862 key.type = BTRFS_EXTENT_ITEM_KEY;
864 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
868 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
869 if (path->slots[0]) {
871 btrfs_item_key_to_cpu(path->nodes[0], &key,
873 if (key.objectid == bytenr &&
874 key.type == BTRFS_EXTENT_ITEM_KEY &&
875 key.offset == fs_info->nodesize)
881 leaf = path->nodes[0];
882 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
883 if (item_size >= sizeof(*ei)) {
884 ei = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_extent_item);
886 num_refs = btrfs_extent_refs(leaf, ei);
887 extent_flags = btrfs_extent_flags(leaf, ei);
890 btrfs_print_v0_err(fs_info);
892 btrfs_abort_transaction(trans, ret);
894 btrfs_handle_fs_error(fs_info, ret, NULL);
899 BUG_ON(num_refs == 0);
909 delayed_refs = &trans->transaction->delayed_refs;
910 spin_lock(&delayed_refs->lock);
911 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
913 if (!mutex_trylock(&head->mutex)) {
914 refcount_inc(&head->refs);
915 spin_unlock(&delayed_refs->lock);
917 btrfs_release_path(path);
920 * Mutex was contended, block until it's released and try
923 mutex_lock(&head->mutex);
924 mutex_unlock(&head->mutex);
925 btrfs_put_delayed_ref_head(head);
928 spin_lock(&head->lock);
929 if (head->extent_op && head->extent_op->update_flags)
930 extent_flags |= head->extent_op->flags_to_set;
932 BUG_ON(num_refs == 0);
934 num_refs += head->ref_mod;
935 spin_unlock(&head->lock);
936 mutex_unlock(&head->mutex);
938 spin_unlock(&delayed_refs->lock);
940 WARN_ON(num_refs == 0);
944 *flags = extent_flags;
946 btrfs_free_path(path);
951 * Back reference rules. Back refs have three main goals:
953 * 1) differentiate between all holders of references to an extent so that
954 * when a reference is dropped we can make sure it was a valid reference
955 * before freeing the extent.
957 * 2) Provide enough information to quickly find the holders of an extent
958 * if we notice a given block is corrupted or bad.
960 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
961 * maintenance. This is actually the same as #2, but with a slightly
962 * different use case.
964 * There are two kinds of back refs. The implicit back refs is optimized
965 * for pointers in non-shared tree blocks. For a given pointer in a block,
966 * back refs of this kind provide information about the block's owner tree
967 * and the pointer's key. These information allow us to find the block by
968 * b-tree searching. The full back refs is for pointers in tree blocks not
969 * referenced by their owner trees. The location of tree block is recorded
970 * in the back refs. Actually the full back refs is generic, and can be
971 * used in all cases the implicit back refs is used. The major shortcoming
972 * of the full back refs is its overhead. Every time a tree block gets
973 * COWed, we have to update back refs entry for all pointers in it.
975 * For a newly allocated tree block, we use implicit back refs for
976 * pointers in it. This means most tree related operations only involve
977 * implicit back refs. For a tree block created in old transaction, the
978 * only way to drop a reference to it is COW it. So we can detect the
979 * event that tree block loses its owner tree's reference and do the
980 * back refs conversion.
982 * When a tree block is COWed through a tree, there are four cases:
984 * The reference count of the block is one and the tree is the block's
985 * owner tree. Nothing to do in this case.
987 * The reference count of the block is one and the tree is not the
988 * block's owner tree. In this case, full back refs is used for pointers
989 * in the block. Remove these full back refs, add implicit back refs for
990 * every pointers in the new block.
992 * The reference count of the block is greater than one and the tree is
993 * the block's owner tree. In this case, implicit back refs is used for
994 * pointers in the block. Add full back refs for every pointers in the
995 * block, increase lower level extents' reference counts. The original
996 * implicit back refs are entailed to the new block.
998 * The reference count of the block is greater than one and the tree is
999 * not the block's owner tree. Add implicit back refs for every pointer in
1000 * the new block, increase lower level extents' reference count.
1002 * Back Reference Key composing:
1004 * The key objectid corresponds to the first byte in the extent,
1005 * The key type is used to differentiate between types of back refs.
1006 * There are different meanings of the key offset for different types
1009 * File extents can be referenced by:
1011 * - multiple snapshots, subvolumes, or different generations in one subvol
1012 * - different files inside a single subvolume
1013 * - different offsets inside a file (bookend extents in file.c)
1015 * The extent ref structure for the implicit back refs has fields for:
1017 * - Objectid of the subvolume root
1018 * - objectid of the file holding the reference
1019 * - original offset in the file
1020 * - how many bookend extents
1022 * The key offset for the implicit back refs is hash of the first
1025 * The extent ref structure for the full back refs has field for:
1027 * - number of pointers in the tree leaf
1029 * The key offset for the implicit back refs is the first byte of
1032 * When a file extent is allocated, The implicit back refs is used.
1033 * the fields are filled in:
1035 * (root_key.objectid, inode objectid, offset in file, 1)
1037 * When a file extent is removed file truncation, we find the
1038 * corresponding implicit back refs and check the following fields:
1040 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 * Btree extents can be referenced by:
1044 * - Different subvolumes
1046 * Both the implicit back refs and the full back refs for tree blocks
1047 * only consist of key. The key offset for the implicit back refs is
1048 * objectid of block's owner tree. The key offset for the full back refs
1049 * is the first byte of parent block.
1051 * When implicit back refs is used, information about the lowest key and
1052 * level of the tree block are required. These information are stored in
1053 * tree block info structure.
1057 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1058 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
1059 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1061 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1062 struct btrfs_extent_inline_ref *iref,
1063 enum btrfs_inline_ref_type is_data)
1065 int type = btrfs_extent_inline_ref_type(eb, iref);
1066 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1068 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1069 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1070 type == BTRFS_SHARED_DATA_REF_KEY ||
1071 type == BTRFS_EXTENT_DATA_REF_KEY) {
1072 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1073 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1075 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1076 ASSERT(eb->fs_info);
1078 * Every shared one has parent tree
1079 * block, which must be aligned to
1083 IS_ALIGNED(offset, eb->fs_info->nodesize))
1086 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1087 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1089 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1090 ASSERT(eb->fs_info);
1092 * Every shared one has parent tree
1093 * block, which must be aligned to
1097 IS_ALIGNED(offset, eb->fs_info->nodesize))
1101 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1106 btrfs_print_leaf((struct extent_buffer *)eb);
1107 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1111 return BTRFS_REF_TYPE_INVALID;
1114 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1116 u32 high_crc = ~(u32)0;
1117 u32 low_crc = ~(u32)0;
1120 lenum = cpu_to_le64(root_objectid);
1121 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1122 lenum = cpu_to_le64(owner);
1123 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1124 lenum = cpu_to_le64(offset);
1125 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1127 return ((u64)high_crc << 31) ^ (u64)low_crc;
1130 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1131 struct btrfs_extent_data_ref *ref)
1133 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1134 btrfs_extent_data_ref_objectid(leaf, ref),
1135 btrfs_extent_data_ref_offset(leaf, ref));
1138 static int match_extent_data_ref(struct extent_buffer *leaf,
1139 struct btrfs_extent_data_ref *ref,
1140 u64 root_objectid, u64 owner, u64 offset)
1142 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1143 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1144 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1149 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_path *path,
1151 u64 bytenr, u64 parent,
1153 u64 owner, u64 offset)
1155 struct btrfs_root *root = trans->fs_info->extent_root;
1156 struct btrfs_key key;
1157 struct btrfs_extent_data_ref *ref;
1158 struct extent_buffer *leaf;
1164 key.objectid = bytenr;
1166 key.type = BTRFS_SHARED_DATA_REF_KEY;
1167 key.offset = parent;
1169 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1170 key.offset = hash_extent_data_ref(root_objectid,
1175 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1187 leaf = path->nodes[0];
1188 nritems = btrfs_header_nritems(leaf);
1190 if (path->slots[0] >= nritems) {
1191 ret = btrfs_next_leaf(root, path);
1197 leaf = path->nodes[0];
1198 nritems = btrfs_header_nritems(leaf);
1202 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1203 if (key.objectid != bytenr ||
1204 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1207 ref = btrfs_item_ptr(leaf, path->slots[0],
1208 struct btrfs_extent_data_ref);
1210 if (match_extent_data_ref(leaf, ref, root_objectid,
1213 btrfs_release_path(path);
1225 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1226 struct btrfs_path *path,
1227 u64 bytenr, u64 parent,
1228 u64 root_objectid, u64 owner,
1229 u64 offset, int refs_to_add)
1231 struct btrfs_root *root = trans->fs_info->extent_root;
1232 struct btrfs_key key;
1233 struct extent_buffer *leaf;
1238 key.objectid = bytenr;
1240 key.type = BTRFS_SHARED_DATA_REF_KEY;
1241 key.offset = parent;
1242 size = sizeof(struct btrfs_shared_data_ref);
1244 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1245 key.offset = hash_extent_data_ref(root_objectid,
1247 size = sizeof(struct btrfs_extent_data_ref);
1250 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1251 if (ret && ret != -EEXIST)
1254 leaf = path->nodes[0];
1256 struct btrfs_shared_data_ref *ref;
1257 ref = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1260 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1262 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1263 num_refs += refs_to_add;
1264 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1267 struct btrfs_extent_data_ref *ref;
1268 while (ret == -EEXIST) {
1269 ref = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_data_ref);
1271 if (match_extent_data_ref(leaf, ref, root_objectid,
1274 btrfs_release_path(path);
1276 ret = btrfs_insert_empty_item(trans, root, path, &key,
1278 if (ret && ret != -EEXIST)
1281 leaf = path->nodes[0];
1283 ref = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_data_ref);
1286 btrfs_set_extent_data_ref_root(leaf, ref,
1288 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1289 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1290 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1292 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1293 num_refs += refs_to_add;
1294 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1297 btrfs_mark_buffer_dirty(leaf);
1300 btrfs_release_path(path);
1304 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1305 struct btrfs_path *path,
1306 int refs_to_drop, int *last_ref)
1308 struct btrfs_key key;
1309 struct btrfs_extent_data_ref *ref1 = NULL;
1310 struct btrfs_shared_data_ref *ref2 = NULL;
1311 struct extent_buffer *leaf;
1315 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1318 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1319 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1322 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1323 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1324 struct btrfs_shared_data_ref);
1325 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1326 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
1327 btrfs_print_v0_err(trans->fs_info);
1328 btrfs_abort_transaction(trans, -EINVAL);
1334 BUG_ON(num_refs < refs_to_drop);
1335 num_refs -= refs_to_drop;
1337 if (num_refs == 0) {
1338 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1341 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1342 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1343 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1344 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1345 btrfs_mark_buffer_dirty(leaf);
1350 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1351 struct btrfs_extent_inline_ref *iref)
1353 struct btrfs_key key;
1354 struct extent_buffer *leaf;
1355 struct btrfs_extent_data_ref *ref1;
1356 struct btrfs_shared_data_ref *ref2;
1360 leaf = path->nodes[0];
1361 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1363 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
1366 * If type is invalid, we should have bailed out earlier than
1369 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1370 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1371 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1372 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1373 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1375 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1376 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1378 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1379 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_extent_data_ref);
1381 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1382 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1383 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1384 struct btrfs_shared_data_ref);
1385 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1392 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1393 struct btrfs_path *path,
1394 u64 bytenr, u64 parent,
1397 struct btrfs_root *root = trans->fs_info->extent_root;
1398 struct btrfs_key key;
1401 key.objectid = bytenr;
1403 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1404 key.offset = parent;
1406 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1407 key.offset = root_objectid;
1410 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1416 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1417 struct btrfs_path *path,
1418 u64 bytenr, u64 parent,
1421 struct btrfs_key key;
1424 key.objectid = bytenr;
1426 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1427 key.offset = parent;
1429 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1430 key.offset = root_objectid;
1433 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1435 btrfs_release_path(path);
1439 static inline int extent_ref_type(u64 parent, u64 owner)
1442 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1444 type = BTRFS_SHARED_BLOCK_REF_KEY;
1446 type = BTRFS_TREE_BLOCK_REF_KEY;
1449 type = BTRFS_SHARED_DATA_REF_KEY;
1451 type = BTRFS_EXTENT_DATA_REF_KEY;
1456 static int find_next_key(struct btrfs_path *path, int level,
1457 struct btrfs_key *key)
1460 for (; level < BTRFS_MAX_LEVEL; level++) {
1461 if (!path->nodes[level])
1463 if (path->slots[level] + 1 >=
1464 btrfs_header_nritems(path->nodes[level]))
1467 btrfs_item_key_to_cpu(path->nodes[level], key,
1468 path->slots[level] + 1);
1470 btrfs_node_key_to_cpu(path->nodes[level], key,
1471 path->slots[level] + 1);
1478 * look for inline back ref. if back ref is found, *ref_ret is set
1479 * to the address of inline back ref, and 0 is returned.
1481 * if back ref isn't found, *ref_ret is set to the address where it
1482 * should be inserted, and -ENOENT is returned.
1484 * if insert is true and there are too many inline back refs, the path
1485 * points to the extent item, and -EAGAIN is returned.
1487 * NOTE: inline back refs are ordered in the same way that back ref
1488 * items in the tree are ordered.
1490 static noinline_for_stack
1491 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1492 struct btrfs_path *path,
1493 struct btrfs_extent_inline_ref **ref_ret,
1494 u64 bytenr, u64 num_bytes,
1495 u64 parent, u64 root_objectid,
1496 u64 owner, u64 offset, int insert)
1498 struct btrfs_fs_info *fs_info = trans->fs_info;
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1501 struct extent_buffer *leaf;
1502 struct btrfs_extent_item *ei;
1503 struct btrfs_extent_inline_ref *iref;
1513 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1516 key.objectid = bytenr;
1517 key.type = BTRFS_EXTENT_ITEM_KEY;
1518 key.offset = num_bytes;
1520 want = extent_ref_type(parent, owner);
1522 extra_size = btrfs_extent_inline_ref_size(want);
1523 path->keep_locks = 1;
1528 * Owner is our level, so we can just add one to get the level for the
1529 * block we are interested in.
1531 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1532 key.type = BTRFS_METADATA_ITEM_KEY;
1537 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1544 * We may be a newly converted file system which still has the old fat
1545 * extent entries for metadata, so try and see if we have one of those.
1547 if (ret > 0 && skinny_metadata) {
1548 skinny_metadata = false;
1549 if (path->slots[0]) {
1551 btrfs_item_key_to_cpu(path->nodes[0], &key,
1553 if (key.objectid == bytenr &&
1554 key.type == BTRFS_EXTENT_ITEM_KEY &&
1555 key.offset == num_bytes)
1559 key.objectid = bytenr;
1560 key.type = BTRFS_EXTENT_ITEM_KEY;
1561 key.offset = num_bytes;
1562 btrfs_release_path(path);
1567 if (ret && !insert) {
1570 } else if (WARN_ON(ret)) {
1575 leaf = path->nodes[0];
1576 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1577 if (unlikely(item_size < sizeof(*ei))) {
1579 btrfs_print_v0_err(fs_info);
1580 btrfs_abort_transaction(trans, err);
1584 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1585 flags = btrfs_extent_flags(leaf, ei);
1587 ptr = (unsigned long)(ei + 1);
1588 end = (unsigned long)ei + item_size;
1590 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1591 ptr += sizeof(struct btrfs_tree_block_info);
1595 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1596 needed = BTRFS_REF_TYPE_DATA;
1598 needed = BTRFS_REF_TYPE_BLOCK;
1606 iref = (struct btrfs_extent_inline_ref *)ptr;
1607 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1608 if (type == BTRFS_REF_TYPE_INVALID) {
1616 ptr += btrfs_extent_inline_ref_size(type);
1620 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1621 struct btrfs_extent_data_ref *dref;
1622 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1623 if (match_extent_data_ref(leaf, dref, root_objectid,
1628 if (hash_extent_data_ref_item(leaf, dref) <
1629 hash_extent_data_ref(root_objectid, owner, offset))
1633 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1635 if (parent == ref_offset) {
1639 if (ref_offset < parent)
1642 if (root_objectid == ref_offset) {
1646 if (ref_offset < root_objectid)
1650 ptr += btrfs_extent_inline_ref_size(type);
1652 if (err == -ENOENT && insert) {
1653 if (item_size + extra_size >=
1654 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1659 * To add new inline back ref, we have to make sure
1660 * there is no corresponding back ref item.
1661 * For simplicity, we just do not add new inline back
1662 * ref if there is any kind of item for this block
1664 if (find_next_key(path, 0, &key) == 0 &&
1665 key.objectid == bytenr &&
1666 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1671 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1674 path->keep_locks = 0;
1675 btrfs_unlock_up_safe(path, 1);
1681 * helper to add new inline back ref
1683 static noinline_for_stack
1684 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1685 struct btrfs_path *path,
1686 struct btrfs_extent_inline_ref *iref,
1687 u64 parent, u64 root_objectid,
1688 u64 owner, u64 offset, int refs_to_add,
1689 struct btrfs_delayed_extent_op *extent_op)
1691 struct extent_buffer *leaf;
1692 struct btrfs_extent_item *ei;
1695 unsigned long item_offset;
1700 leaf = path->nodes[0];
1701 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1702 item_offset = (unsigned long)iref - (unsigned long)ei;
1704 type = extent_ref_type(parent, owner);
1705 size = btrfs_extent_inline_ref_size(type);
1707 btrfs_extend_item(fs_info, path, size);
1709 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710 refs = btrfs_extent_refs(leaf, ei);
1711 refs += refs_to_add;
1712 btrfs_set_extent_refs(leaf, ei, refs);
1714 __run_delayed_extent_op(extent_op, leaf, ei);
1716 ptr = (unsigned long)ei + item_offset;
1717 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1718 if (ptr < end - size)
1719 memmove_extent_buffer(leaf, ptr + size, ptr,
1722 iref = (struct btrfs_extent_inline_ref *)ptr;
1723 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1724 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1725 struct btrfs_extent_data_ref *dref;
1726 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1727 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1728 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1729 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1730 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1731 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1732 struct btrfs_shared_data_ref *sref;
1733 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1734 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1735 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1736 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1737 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1739 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1741 btrfs_mark_buffer_dirty(leaf);
1744 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1745 struct btrfs_path *path,
1746 struct btrfs_extent_inline_ref **ref_ret,
1747 u64 bytenr, u64 num_bytes, u64 parent,
1748 u64 root_objectid, u64 owner, u64 offset)
1752 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1753 num_bytes, parent, root_objectid,
1758 btrfs_release_path(path);
1761 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1762 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1765 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1766 root_objectid, owner, offset);
1772 * helper to update/remove inline back ref
1774 static noinline_for_stack
1775 void update_inline_extent_backref(struct btrfs_path *path,
1776 struct btrfs_extent_inline_ref *iref,
1778 struct btrfs_delayed_extent_op *extent_op,
1781 struct extent_buffer *leaf = path->nodes[0];
1782 struct btrfs_fs_info *fs_info = leaf->fs_info;
1783 struct btrfs_extent_item *ei;
1784 struct btrfs_extent_data_ref *dref = NULL;
1785 struct btrfs_shared_data_ref *sref = NULL;
1793 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1794 refs = btrfs_extent_refs(leaf, ei);
1795 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1796 refs += refs_to_mod;
1797 btrfs_set_extent_refs(leaf, ei, refs);
1799 __run_delayed_extent_op(extent_op, leaf, ei);
1802 * If type is invalid, we should have bailed out after
1803 * lookup_inline_extent_backref().
1805 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1806 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1808 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1809 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1810 refs = btrfs_extent_data_ref_count(leaf, dref);
1811 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1812 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1813 refs = btrfs_shared_data_ref_count(leaf, sref);
1816 BUG_ON(refs_to_mod != -1);
1819 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1820 refs += refs_to_mod;
1823 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1824 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1826 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1829 size = btrfs_extent_inline_ref_size(type);
1830 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1831 ptr = (unsigned long)iref;
1832 end = (unsigned long)ei + item_size;
1833 if (ptr + size < end)
1834 memmove_extent_buffer(leaf, ptr, ptr + size,
1837 btrfs_truncate_item(fs_info, path, item_size, 1);
1839 btrfs_mark_buffer_dirty(leaf);
1842 static noinline_for_stack
1843 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1844 struct btrfs_path *path,
1845 u64 bytenr, u64 num_bytes, u64 parent,
1846 u64 root_objectid, u64 owner,
1847 u64 offset, int refs_to_add,
1848 struct btrfs_delayed_extent_op *extent_op)
1850 struct btrfs_extent_inline_ref *iref;
1853 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1854 num_bytes, parent, root_objectid,
1857 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1858 update_inline_extent_backref(path, iref, refs_to_add,
1860 } else if (ret == -ENOENT) {
1861 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1862 root_objectid, owner, offset,
1863 refs_to_add, extent_op);
1869 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1870 struct btrfs_path *path,
1871 u64 bytenr, u64 parent, u64 root_objectid,
1872 u64 owner, u64 offset, int refs_to_add)
1875 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1876 BUG_ON(refs_to_add != 1);
1877 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1880 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1881 root_objectid, owner, offset,
1887 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1888 struct btrfs_path *path,
1889 struct btrfs_extent_inline_ref *iref,
1890 int refs_to_drop, int is_data, int *last_ref)
1894 BUG_ON(!is_data && refs_to_drop != 1);
1896 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1898 } else if (is_data) {
1899 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1903 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1908 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1909 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1910 u64 *discarded_bytes)
1913 u64 bytes_left, end;
1914 u64 aligned_start = ALIGN(start, 1 << 9);
1916 if (WARN_ON(start != aligned_start)) {
1917 len -= aligned_start - start;
1918 len = round_down(len, 1 << 9);
1919 start = aligned_start;
1922 *discarded_bytes = 0;
1930 /* Skip any superblocks on this device. */
1931 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1932 u64 sb_start = btrfs_sb_offset(j);
1933 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1934 u64 size = sb_start - start;
1936 if (!in_range(sb_start, start, bytes_left) &&
1937 !in_range(sb_end, start, bytes_left) &&
1938 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1942 * Superblock spans beginning of range. Adjust start and
1945 if (sb_start <= start) {
1946 start += sb_end - start;
1951 bytes_left = end - start;
1956 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1959 *discarded_bytes += size;
1960 else if (ret != -EOPNOTSUPP)
1969 bytes_left = end - start;
1973 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1976 *discarded_bytes += bytes_left;
1981 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1982 u64 num_bytes, u64 *actual_bytes)
1985 u64 discarded_bytes = 0;
1986 struct btrfs_bio *bbio = NULL;
1990 * Avoid races with device replace and make sure our bbio has devices
1991 * associated to its stripes that don't go away while we are discarding.
1993 btrfs_bio_counter_inc_blocked(fs_info);
1994 /* Tell the block device(s) that the sectors can be discarded */
1995 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
1997 /* Error condition is -ENOMEM */
1999 struct btrfs_bio_stripe *stripe = bbio->stripes;
2003 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2005 struct request_queue *req_q;
2007 if (!stripe->dev->bdev) {
2008 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2011 req_q = bdev_get_queue(stripe->dev->bdev);
2012 if (!blk_queue_discard(req_q))
2015 ret = btrfs_issue_discard(stripe->dev->bdev,
2020 discarded_bytes += bytes;
2021 else if (ret != -EOPNOTSUPP)
2022 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2025 * Just in case we get back EOPNOTSUPP for some reason,
2026 * just ignore the return value so we don't screw up
2027 * people calling discard_extent.
2031 btrfs_put_bbio(bbio);
2033 btrfs_bio_counter_dec(fs_info);
2036 *actual_bytes = discarded_bytes;
2039 if (ret == -EOPNOTSUPP)
2044 /* Can return -ENOMEM */
2045 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2046 struct btrfs_root *root,
2047 u64 bytenr, u64 num_bytes, u64 parent,
2048 u64 root_objectid, u64 owner, u64 offset)
2050 struct btrfs_fs_info *fs_info = root->fs_info;
2051 int old_ref_mod, new_ref_mod;
2054 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2055 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2057 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2058 owner, offset, BTRFS_ADD_DELAYED_REF);
2060 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2061 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
2063 root_objectid, (int)owner,
2064 BTRFS_ADD_DELAYED_REF, NULL,
2065 &old_ref_mod, &new_ref_mod);
2067 ret = btrfs_add_delayed_data_ref(trans, bytenr,
2069 root_objectid, owner, offset,
2070 0, BTRFS_ADD_DELAYED_REF,
2071 &old_ref_mod, &new_ref_mod);
2074 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2075 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2077 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2084 * __btrfs_inc_extent_ref - insert backreference for a given extent
2086 * @trans: Handle of transaction
2088 * @node: The delayed ref node used to get the bytenr/length for
2089 * extent whose references are incremented.
2091 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2092 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2093 * bytenr of the parent block. Since new extents are always
2094 * created with indirect references, this will only be the case
2095 * when relocating a shared extent. In that case, root_objectid
2096 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2099 * @root_objectid: The id of the root where this modification has originated,
2100 * this can be either one of the well-known metadata trees or
2101 * the subvolume id which references this extent.
2103 * @owner: For data extents it is the inode number of the owning file.
2104 * For metadata extents this parameter holds the level in the
2105 * tree of the extent.
2107 * @offset: For metadata extents the offset is ignored and is currently
2108 * always passed as 0. For data extents it is the fileoffset
2109 * this extent belongs to.
2111 * @refs_to_add Number of references to add
2113 * @extent_op Pointer to a structure, holding information necessary when
2114 * updating a tree block's flags
2117 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2118 struct btrfs_delayed_ref_node *node,
2119 u64 parent, u64 root_objectid,
2120 u64 owner, u64 offset, int refs_to_add,
2121 struct btrfs_delayed_extent_op *extent_op)
2123 struct btrfs_path *path;
2124 struct extent_buffer *leaf;
2125 struct btrfs_extent_item *item;
2126 struct btrfs_key key;
2127 u64 bytenr = node->bytenr;
2128 u64 num_bytes = node->num_bytes;
2132 path = btrfs_alloc_path();
2136 path->reada = READA_FORWARD;
2137 path->leave_spinning = 1;
2138 /* this will setup the path even if it fails to insert the back ref */
2139 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2140 parent, root_objectid, owner,
2141 offset, refs_to_add, extent_op);
2142 if ((ret < 0 && ret != -EAGAIN) || !ret)
2146 * Ok we had -EAGAIN which means we didn't have space to insert and
2147 * inline extent ref, so just update the reference count and add a
2150 leaf = path->nodes[0];
2151 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2152 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2153 refs = btrfs_extent_refs(leaf, item);
2154 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2156 __run_delayed_extent_op(extent_op, leaf, item);
2158 btrfs_mark_buffer_dirty(leaf);
2159 btrfs_release_path(path);
2161 path->reada = READA_FORWARD;
2162 path->leave_spinning = 1;
2163 /* now insert the actual backref */
2164 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2165 owner, offset, refs_to_add);
2167 btrfs_abort_transaction(trans, ret);
2169 btrfs_free_path(path);
2173 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2174 struct btrfs_delayed_ref_node *node,
2175 struct btrfs_delayed_extent_op *extent_op,
2176 int insert_reserved)
2179 struct btrfs_delayed_data_ref *ref;
2180 struct btrfs_key ins;
2185 ins.objectid = node->bytenr;
2186 ins.offset = node->num_bytes;
2187 ins.type = BTRFS_EXTENT_ITEM_KEY;
2189 ref = btrfs_delayed_node_to_data_ref(node);
2190 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2192 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2193 parent = ref->parent;
2194 ref_root = ref->root;
2196 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2198 flags |= extent_op->flags_to_set;
2199 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2200 flags, ref->objectid,
2203 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2204 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2205 ref->objectid, ref->offset,
2206 node->ref_mod, extent_op);
2207 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2208 ret = __btrfs_free_extent(trans, node, parent,
2209 ref_root, ref->objectid,
2210 ref->offset, node->ref_mod,
2218 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2219 struct extent_buffer *leaf,
2220 struct btrfs_extent_item *ei)
2222 u64 flags = btrfs_extent_flags(leaf, ei);
2223 if (extent_op->update_flags) {
2224 flags |= extent_op->flags_to_set;
2225 btrfs_set_extent_flags(leaf, ei, flags);
2228 if (extent_op->update_key) {
2229 struct btrfs_tree_block_info *bi;
2230 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2231 bi = (struct btrfs_tree_block_info *)(ei + 1);
2232 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2236 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2237 struct btrfs_delayed_ref_head *head,
2238 struct btrfs_delayed_extent_op *extent_op)
2240 struct btrfs_fs_info *fs_info = trans->fs_info;
2241 struct btrfs_key key;
2242 struct btrfs_path *path;
2243 struct btrfs_extent_item *ei;
2244 struct extent_buffer *leaf;
2248 int metadata = !extent_op->is_data;
2253 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2256 path = btrfs_alloc_path();
2260 key.objectid = head->bytenr;
2263 key.type = BTRFS_METADATA_ITEM_KEY;
2264 key.offset = extent_op->level;
2266 key.type = BTRFS_EXTENT_ITEM_KEY;
2267 key.offset = head->num_bytes;
2271 path->reada = READA_FORWARD;
2272 path->leave_spinning = 1;
2273 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2280 if (path->slots[0] > 0) {
2282 btrfs_item_key_to_cpu(path->nodes[0], &key,
2284 if (key.objectid == head->bytenr &&
2285 key.type == BTRFS_EXTENT_ITEM_KEY &&
2286 key.offset == head->num_bytes)
2290 btrfs_release_path(path);
2293 key.objectid = head->bytenr;
2294 key.offset = head->num_bytes;
2295 key.type = BTRFS_EXTENT_ITEM_KEY;
2304 leaf = path->nodes[0];
2305 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2307 if (unlikely(item_size < sizeof(*ei))) {
2309 btrfs_print_v0_err(fs_info);
2310 btrfs_abort_transaction(trans, err);
2314 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2315 __run_delayed_extent_op(extent_op, leaf, ei);
2317 btrfs_mark_buffer_dirty(leaf);
2319 btrfs_free_path(path);
2323 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2324 struct btrfs_delayed_ref_node *node,
2325 struct btrfs_delayed_extent_op *extent_op,
2326 int insert_reserved)
2329 struct btrfs_delayed_tree_ref *ref;
2333 ref = btrfs_delayed_node_to_tree_ref(node);
2334 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2336 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2337 parent = ref->parent;
2338 ref_root = ref->root;
2340 if (node->ref_mod != 1) {
2341 btrfs_err(trans->fs_info,
2342 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2343 node->bytenr, node->ref_mod, node->action, ref_root,
2347 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2348 BUG_ON(!extent_op || !extent_op->update_flags);
2349 ret = alloc_reserved_tree_block(trans, node, extent_op);
2350 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2351 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2352 ref->level, 0, 1, extent_op);
2353 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2354 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2355 ref->level, 0, 1, extent_op);
2362 /* helper function to actually process a single delayed ref entry */
2363 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2364 struct btrfs_delayed_ref_node *node,
2365 struct btrfs_delayed_extent_op *extent_op,
2366 int insert_reserved)
2370 if (trans->aborted) {
2371 if (insert_reserved)
2372 btrfs_pin_extent(trans->fs_info, node->bytenr,
2373 node->num_bytes, 1);
2377 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2378 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2379 ret = run_delayed_tree_ref(trans, node, extent_op,
2381 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2382 node->type == BTRFS_SHARED_DATA_REF_KEY)
2383 ret = run_delayed_data_ref(trans, node, extent_op,
2387 if (ret && insert_reserved)
2388 btrfs_pin_extent(trans->fs_info, node->bytenr,
2389 node->num_bytes, 1);
2393 static inline struct btrfs_delayed_ref_node *
2394 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2396 struct btrfs_delayed_ref_node *ref;
2398 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
2402 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2403 * This is to prevent a ref count from going down to zero, which deletes
2404 * the extent item from the extent tree, when there still are references
2405 * to add, which would fail because they would not find the extent item.
2407 if (!list_empty(&head->ref_add_list))
2408 return list_first_entry(&head->ref_add_list,
2409 struct btrfs_delayed_ref_node, add_list);
2411 ref = rb_entry(rb_first_cached(&head->ref_tree),
2412 struct btrfs_delayed_ref_node, ref_node);
2413 ASSERT(list_empty(&ref->add_list));
2417 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2418 struct btrfs_delayed_ref_head *head)
2420 spin_lock(&delayed_refs->lock);
2421 head->processing = 0;
2422 delayed_refs->num_heads_ready++;
2423 spin_unlock(&delayed_refs->lock);
2424 btrfs_delayed_ref_unlock(head);
2427 static struct btrfs_delayed_extent_op *cleanup_extent_op(
2428 struct btrfs_delayed_ref_head *head)
2430 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2435 if (head->must_insert_reserved) {
2436 head->extent_op = NULL;
2437 btrfs_free_delayed_extent_op(extent_op);
2443 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
2444 struct btrfs_delayed_ref_head *head)
2446 struct btrfs_delayed_extent_op *extent_op;
2449 extent_op = cleanup_extent_op(head);
2452 head->extent_op = NULL;
2453 spin_unlock(&head->lock);
2454 ret = run_delayed_extent_op(trans, head, extent_op);
2455 btrfs_free_delayed_extent_op(extent_op);
2456 return ret ? ret : 1;
2459 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
2460 struct btrfs_delayed_ref_root *delayed_refs,
2461 struct btrfs_delayed_ref_head *head)
2463 int nr_items = 1; /* Dropping this ref head update. */
2465 if (head->total_ref_mod < 0) {
2466 struct btrfs_space_info *space_info;
2470 flags = BTRFS_BLOCK_GROUP_DATA;
2471 else if (head->is_system)
2472 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2474 flags = BTRFS_BLOCK_GROUP_METADATA;
2475 space_info = __find_space_info(fs_info, flags);
2477 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2479 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2482 * We had csum deletions accounted for in our delayed refs rsv,
2483 * we need to drop the csum leaves for this update from our
2486 if (head->is_data) {
2487 spin_lock(&delayed_refs->lock);
2488 delayed_refs->pending_csums -= head->num_bytes;
2489 spin_unlock(&delayed_refs->lock);
2490 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
2495 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
2498 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2499 struct btrfs_delayed_ref_head *head)
2502 struct btrfs_fs_info *fs_info = trans->fs_info;
2503 struct btrfs_delayed_ref_root *delayed_refs;
2506 delayed_refs = &trans->transaction->delayed_refs;
2508 ret = run_and_cleanup_extent_op(trans, head);
2510 unselect_delayed_ref_head(delayed_refs, head);
2511 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2518 * Need to drop our head ref lock and re-acquire the delayed ref lock
2519 * and then re-check to make sure nobody got added.
2521 spin_unlock(&head->lock);
2522 spin_lock(&delayed_refs->lock);
2523 spin_lock(&head->lock);
2524 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2525 spin_unlock(&head->lock);
2526 spin_unlock(&delayed_refs->lock);
2529 btrfs_delete_ref_head(delayed_refs, head);
2530 spin_unlock(&head->lock);
2531 spin_unlock(&delayed_refs->lock);
2533 if (head->must_insert_reserved) {
2534 btrfs_pin_extent(fs_info, head->bytenr,
2535 head->num_bytes, 1);
2536 if (head->is_data) {
2537 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2542 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
2544 trace_run_delayed_ref_head(fs_info, head, 0);
2545 btrfs_delayed_ref_unlock(head);
2546 btrfs_put_delayed_ref_head(head);
2550 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2551 struct btrfs_trans_handle *trans)
2553 struct btrfs_delayed_ref_root *delayed_refs =
2554 &trans->transaction->delayed_refs;
2555 struct btrfs_delayed_ref_head *head = NULL;
2558 spin_lock(&delayed_refs->lock);
2559 head = btrfs_select_ref_head(delayed_refs);
2561 spin_unlock(&delayed_refs->lock);
2566 * Grab the lock that says we are going to process all the refs for
2569 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2570 spin_unlock(&delayed_refs->lock);
2573 * We may have dropped the spin lock to get the head mutex lock, and
2574 * that might have given someone else time to free the head. If that's
2575 * true, it has been removed from our list and we can move on.
2578 head = ERR_PTR(-EAGAIN);
2583 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2584 struct btrfs_delayed_ref_head *locked_ref,
2585 unsigned long *run_refs)
2587 struct btrfs_fs_info *fs_info = trans->fs_info;
2588 struct btrfs_delayed_ref_root *delayed_refs;
2589 struct btrfs_delayed_extent_op *extent_op;
2590 struct btrfs_delayed_ref_node *ref;
2591 int must_insert_reserved = 0;
2594 delayed_refs = &trans->transaction->delayed_refs;
2596 lockdep_assert_held(&locked_ref->mutex);
2597 lockdep_assert_held(&locked_ref->lock);
2599 while ((ref = select_delayed_ref(locked_ref))) {
2601 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2602 spin_unlock(&locked_ref->lock);
2603 unselect_delayed_ref_head(delayed_refs, locked_ref);
2609 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2610 RB_CLEAR_NODE(&ref->ref_node);
2611 if (!list_empty(&ref->add_list))
2612 list_del(&ref->add_list);
2614 * When we play the delayed ref, also correct the ref_mod on
2617 switch (ref->action) {
2618 case BTRFS_ADD_DELAYED_REF:
2619 case BTRFS_ADD_DELAYED_EXTENT:
2620 locked_ref->ref_mod -= ref->ref_mod;
2622 case BTRFS_DROP_DELAYED_REF:
2623 locked_ref->ref_mod += ref->ref_mod;
2628 atomic_dec(&delayed_refs->num_entries);
2631 * Record the must_insert_reserved flag before we drop the
2634 must_insert_reserved = locked_ref->must_insert_reserved;
2635 locked_ref->must_insert_reserved = 0;
2637 extent_op = locked_ref->extent_op;
2638 locked_ref->extent_op = NULL;
2639 spin_unlock(&locked_ref->lock);
2641 ret = run_one_delayed_ref(trans, ref, extent_op,
2642 must_insert_reserved);
2644 btrfs_free_delayed_extent_op(extent_op);
2646 unselect_delayed_ref_head(delayed_refs, locked_ref);
2647 btrfs_put_delayed_ref(ref);
2648 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2653 btrfs_put_delayed_ref(ref);
2656 spin_lock(&locked_ref->lock);
2657 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2664 * Returns 0 on success or if called with an already aborted transaction.
2665 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2667 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2670 struct btrfs_fs_info *fs_info = trans->fs_info;
2671 struct btrfs_delayed_ref_root *delayed_refs;
2672 struct btrfs_delayed_ref_head *locked_ref = NULL;
2673 ktime_t start = ktime_get();
2675 unsigned long count = 0;
2676 unsigned long actual_count = 0;
2678 delayed_refs = &trans->transaction->delayed_refs;
2681 locked_ref = btrfs_obtain_ref_head(trans);
2682 if (IS_ERR_OR_NULL(locked_ref)) {
2683 if (PTR_ERR(locked_ref) == -EAGAIN) {
2692 * We need to try and merge add/drops of the same ref since we
2693 * can run into issues with relocate dropping the implicit ref
2694 * and then it being added back again before the drop can
2695 * finish. If we merged anything we need to re-loop so we can
2697 * Or we can get node references of the same type that weren't
2698 * merged when created due to bumps in the tree mod seq, and
2699 * we need to merge them to prevent adding an inline extent
2700 * backref before dropping it (triggering a BUG_ON at
2701 * insert_inline_extent_backref()).
2703 spin_lock(&locked_ref->lock);
2704 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2706 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2708 if (ret < 0 && ret != -EAGAIN) {
2710 * Error, btrfs_run_delayed_refs_for_head already
2711 * unlocked everything so just bail out
2716 * Success, perform the usual cleanup of a processed
2719 ret = cleanup_ref_head(trans, locked_ref);
2721 /* We dropped our lock, we need to loop. */
2730 * Either success case or btrfs_run_delayed_refs_for_head
2731 * returned -EAGAIN, meaning we need to select another head
2736 } while ((nr != -1 && count < nr) || locked_ref);
2739 * We don't want to include ref heads since we can have empty ref heads
2740 * and those will drastically skew our runtime down since we just do
2741 * accounting, no actual extent tree updates.
2743 if (actual_count > 0) {
2744 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2748 * We weigh the current average higher than our current runtime
2749 * to avoid large swings in the average.
2751 spin_lock(&delayed_refs->lock);
2752 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2753 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2754 spin_unlock(&delayed_refs->lock);
2759 #ifdef SCRAMBLE_DELAYED_REFS
2761 * Normally delayed refs get processed in ascending bytenr order. This
2762 * correlates in most cases to the order added. To expose dependencies on this
2763 * order, we start to process the tree in the middle instead of the beginning
2765 static u64 find_middle(struct rb_root *root)
2767 struct rb_node *n = root->rb_node;
2768 struct btrfs_delayed_ref_node *entry;
2771 u64 first = 0, last = 0;
2775 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2776 first = entry->bytenr;
2780 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2781 last = entry->bytenr;
2786 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2787 WARN_ON(!entry->in_tree);
2789 middle = entry->bytenr;
2802 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2806 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2807 sizeof(struct btrfs_extent_inline_ref));
2808 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2809 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2812 * We don't ever fill up leaves all the way so multiply by 2 just to be
2813 * closer to what we're really going to want to use.
2815 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2819 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2820 * would require to store the csums for that many bytes.
2822 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2825 u64 num_csums_per_leaf;
2828 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2829 num_csums_per_leaf = div64_u64(csum_size,
2830 (u64)btrfs_super_csum_size(fs_info->super_copy));
2831 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2832 num_csums += num_csums_per_leaf - 1;
2833 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2837 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info)
2839 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
2840 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2844 spin_lock(&global_rsv->lock);
2845 reserved = global_rsv->reserved;
2846 spin_unlock(&global_rsv->lock);
2849 * Since the global reserve is just kind of magic we don't really want
2850 * to rely on it to save our bacon, so if our size is more than the
2851 * delayed_refs_rsv and the global rsv then it's time to think about
2854 spin_lock(&delayed_refs_rsv->lock);
2855 reserved += delayed_refs_rsv->reserved;
2856 if (delayed_refs_rsv->size >= reserved)
2858 spin_unlock(&delayed_refs_rsv->lock);
2862 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2865 atomic_read(&trans->transaction->delayed_refs.num_entries);
2870 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2871 val = num_entries * avg_runtime;
2872 if (val >= NSEC_PER_SEC)
2874 if (val >= NSEC_PER_SEC / 2)
2877 return btrfs_check_space_for_delayed_refs(trans->fs_info);
2880 struct async_delayed_refs {
2881 struct btrfs_root *root;
2886 struct completion wait;
2887 struct btrfs_work work;
2890 static inline struct async_delayed_refs *
2891 to_async_delayed_refs(struct btrfs_work *work)
2893 return container_of(work, struct async_delayed_refs, work);
2896 static void delayed_ref_async_start(struct btrfs_work *work)
2898 struct async_delayed_refs *async = to_async_delayed_refs(work);
2899 struct btrfs_trans_handle *trans;
2900 struct btrfs_fs_info *fs_info = async->root->fs_info;
2903 /* if the commit is already started, we don't need to wait here */
2904 if (btrfs_transaction_blocked(fs_info))
2907 trans = btrfs_join_transaction(async->root);
2908 if (IS_ERR(trans)) {
2909 async->error = PTR_ERR(trans);
2913 /* Don't bother flushing if we got into a different transaction */
2914 if (trans->transid > async->transid)
2917 ret = btrfs_run_delayed_refs(trans, async->count);
2921 ret = btrfs_end_transaction(trans);
2922 if (ret && !async->error)
2926 complete(&async->wait);
2931 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2932 unsigned long count, u64 transid, int wait)
2934 struct async_delayed_refs *async;
2937 async = kmalloc(sizeof(*async), GFP_NOFS);
2941 async->root = fs_info->tree_root;
2942 async->count = count;
2944 async->transid = transid;
2949 init_completion(&async->wait);
2951 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2952 delayed_ref_async_start, NULL, NULL);
2954 btrfs_queue_work(fs_info->extent_workers, &async->work);
2957 wait_for_completion(&async->wait);
2966 * this starts processing the delayed reference count updates and
2967 * extent insertions we have queued up so far. count can be
2968 * 0, which means to process everything in the tree at the start
2969 * of the run (but not newly added entries), or it can be some target
2970 * number you'd like to process.
2972 * Returns 0 on success or if called with an aborted transaction
2973 * Returns <0 on error and aborts the transaction
2975 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2976 unsigned long count)
2978 struct btrfs_fs_info *fs_info = trans->fs_info;
2979 struct rb_node *node;
2980 struct btrfs_delayed_ref_root *delayed_refs;
2981 struct btrfs_delayed_ref_head *head;
2983 int run_all = count == (unsigned long)-1;
2985 /* We'll clean this up in btrfs_cleanup_transaction */
2989 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2992 delayed_refs = &trans->transaction->delayed_refs;
2994 count = atomic_read(&delayed_refs->num_entries) * 2;
2997 #ifdef SCRAMBLE_DELAYED_REFS
2998 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3000 ret = __btrfs_run_delayed_refs(trans, count);
3002 btrfs_abort_transaction(trans, ret);
3007 btrfs_create_pending_block_groups(trans);
3009 spin_lock(&delayed_refs->lock);
3010 node = rb_first_cached(&delayed_refs->href_root);
3012 spin_unlock(&delayed_refs->lock);
3015 head = rb_entry(node, struct btrfs_delayed_ref_head,
3017 refcount_inc(&head->refs);
3018 spin_unlock(&delayed_refs->lock);
3020 /* Mutex was contended, block until it's released and retry. */
3021 mutex_lock(&head->mutex);
3022 mutex_unlock(&head->mutex);
3024 btrfs_put_delayed_ref_head(head);
3032 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3033 struct btrfs_fs_info *fs_info,
3034 u64 bytenr, u64 num_bytes, u64 flags,
3035 int level, int is_data)
3037 struct btrfs_delayed_extent_op *extent_op;
3040 extent_op = btrfs_alloc_delayed_extent_op();
3044 extent_op->flags_to_set = flags;
3045 extent_op->update_flags = true;
3046 extent_op->update_key = false;
3047 extent_op->is_data = is_data ? true : false;
3048 extent_op->level = level;
3050 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3051 num_bytes, extent_op);
3053 btrfs_free_delayed_extent_op(extent_op);
3057 static noinline int check_delayed_ref(struct btrfs_root *root,
3058 struct btrfs_path *path,
3059 u64 objectid, u64 offset, u64 bytenr)
3061 struct btrfs_delayed_ref_head *head;
3062 struct btrfs_delayed_ref_node *ref;
3063 struct btrfs_delayed_data_ref *data_ref;
3064 struct btrfs_delayed_ref_root *delayed_refs;
3065 struct btrfs_transaction *cur_trans;
3066 struct rb_node *node;
3069 spin_lock(&root->fs_info->trans_lock);
3070 cur_trans = root->fs_info->running_transaction;
3072 refcount_inc(&cur_trans->use_count);
3073 spin_unlock(&root->fs_info->trans_lock);
3077 delayed_refs = &cur_trans->delayed_refs;
3078 spin_lock(&delayed_refs->lock);
3079 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3081 spin_unlock(&delayed_refs->lock);
3082 btrfs_put_transaction(cur_trans);
3086 if (!mutex_trylock(&head->mutex)) {
3087 refcount_inc(&head->refs);
3088 spin_unlock(&delayed_refs->lock);
3090 btrfs_release_path(path);
3093 * Mutex was contended, block until it's released and let
3096 mutex_lock(&head->mutex);
3097 mutex_unlock(&head->mutex);
3098 btrfs_put_delayed_ref_head(head);
3099 btrfs_put_transaction(cur_trans);
3102 spin_unlock(&delayed_refs->lock);
3104 spin_lock(&head->lock);
3106 * XXX: We should replace this with a proper search function in the
3109 for (node = rb_first_cached(&head->ref_tree); node;
3110 node = rb_next(node)) {
3111 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3112 /* If it's a shared ref we know a cross reference exists */
3113 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3118 data_ref = btrfs_delayed_node_to_data_ref(ref);
3121 * If our ref doesn't match the one we're currently looking at
3122 * then we have a cross reference.
3124 if (data_ref->root != root->root_key.objectid ||
3125 data_ref->objectid != objectid ||
3126 data_ref->offset != offset) {
3131 spin_unlock(&head->lock);
3132 mutex_unlock(&head->mutex);
3133 btrfs_put_transaction(cur_trans);
3137 static noinline int check_committed_ref(struct btrfs_root *root,
3138 struct btrfs_path *path,
3139 u64 objectid, u64 offset, u64 bytenr)
3141 struct btrfs_fs_info *fs_info = root->fs_info;
3142 struct btrfs_root *extent_root = fs_info->extent_root;
3143 struct extent_buffer *leaf;
3144 struct btrfs_extent_data_ref *ref;
3145 struct btrfs_extent_inline_ref *iref;
3146 struct btrfs_extent_item *ei;
3147 struct btrfs_key key;
3152 key.objectid = bytenr;
3153 key.offset = (u64)-1;
3154 key.type = BTRFS_EXTENT_ITEM_KEY;
3156 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3159 BUG_ON(ret == 0); /* Corruption */
3162 if (path->slots[0] == 0)
3166 leaf = path->nodes[0];
3167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3169 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3173 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3174 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3176 if (item_size != sizeof(*ei) +
3177 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3180 if (btrfs_extent_generation(leaf, ei) <=
3181 btrfs_root_last_snapshot(&root->root_item))
3184 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3186 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3187 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3190 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3191 if (btrfs_extent_refs(leaf, ei) !=
3192 btrfs_extent_data_ref_count(leaf, ref) ||
3193 btrfs_extent_data_ref_root(leaf, ref) !=
3194 root->root_key.objectid ||
3195 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3196 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3204 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3207 struct btrfs_path *path;
3210 path = btrfs_alloc_path();
3215 ret = check_committed_ref(root, path, objectid,
3217 if (ret && ret != -ENOENT)
3220 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3221 } while (ret == -EAGAIN);
3224 btrfs_free_path(path);
3225 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3230 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3231 struct btrfs_root *root,
3232 struct extent_buffer *buf,
3233 int full_backref, int inc)
3235 struct btrfs_fs_info *fs_info = root->fs_info;
3241 struct btrfs_key key;
3242 struct btrfs_file_extent_item *fi;
3246 int (*process_func)(struct btrfs_trans_handle *,
3247 struct btrfs_root *,
3248 u64, u64, u64, u64, u64, u64);
3251 if (btrfs_is_testing(fs_info))
3254 ref_root = btrfs_header_owner(buf);
3255 nritems = btrfs_header_nritems(buf);
3256 level = btrfs_header_level(buf);
3258 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3262 process_func = btrfs_inc_extent_ref;
3264 process_func = btrfs_free_extent;
3267 parent = buf->start;
3271 for (i = 0; i < nritems; i++) {
3273 btrfs_item_key_to_cpu(buf, &key, i);
3274 if (key.type != BTRFS_EXTENT_DATA_KEY)
3276 fi = btrfs_item_ptr(buf, i,
3277 struct btrfs_file_extent_item);
3278 if (btrfs_file_extent_type(buf, fi) ==
3279 BTRFS_FILE_EXTENT_INLINE)
3281 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3285 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3286 key.offset -= btrfs_file_extent_offset(buf, fi);
3287 ret = process_func(trans, root, bytenr, num_bytes,
3288 parent, ref_root, key.objectid,
3293 bytenr = btrfs_node_blockptr(buf, i);
3294 num_bytes = fs_info->nodesize;
3295 ret = process_func(trans, root, bytenr, num_bytes,
3296 parent, ref_root, level - 1, 0);
3306 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3307 struct extent_buffer *buf, int full_backref)
3309 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3312 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3313 struct extent_buffer *buf, int full_backref)
3315 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3318 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3319 struct btrfs_fs_info *fs_info,
3320 struct btrfs_path *path,
3321 struct btrfs_block_group_cache *cache)
3324 struct btrfs_root *extent_root = fs_info->extent_root;
3326 struct extent_buffer *leaf;
3328 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3335 leaf = path->nodes[0];
3336 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3337 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3338 btrfs_mark_buffer_dirty(leaf);
3340 btrfs_release_path(path);
3345 static struct btrfs_block_group_cache *
3346 next_block_group(struct btrfs_fs_info *fs_info,
3347 struct btrfs_block_group_cache *cache)
3349 struct rb_node *node;
3351 spin_lock(&fs_info->block_group_cache_lock);
3353 /* If our block group was removed, we need a full search. */
3354 if (RB_EMPTY_NODE(&cache->cache_node)) {
3355 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3357 spin_unlock(&fs_info->block_group_cache_lock);
3358 btrfs_put_block_group(cache);
3359 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3361 node = rb_next(&cache->cache_node);
3362 btrfs_put_block_group(cache);
3364 cache = rb_entry(node, struct btrfs_block_group_cache,
3366 btrfs_get_block_group(cache);
3369 spin_unlock(&fs_info->block_group_cache_lock);
3373 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3374 struct btrfs_trans_handle *trans,
3375 struct btrfs_path *path)
3377 struct btrfs_fs_info *fs_info = block_group->fs_info;
3378 struct btrfs_root *root = fs_info->tree_root;
3379 struct inode *inode = NULL;
3380 struct extent_changeset *data_reserved = NULL;
3382 int dcs = BTRFS_DC_ERROR;
3388 * If this block group is smaller than 100 megs don't bother caching the
3391 if (block_group->key.offset < (100 * SZ_1M)) {
3392 spin_lock(&block_group->lock);
3393 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3394 spin_unlock(&block_group->lock);
3401 inode = lookup_free_space_inode(fs_info, block_group, path);
3402 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3403 ret = PTR_ERR(inode);
3404 btrfs_release_path(path);
3408 if (IS_ERR(inode)) {
3412 if (block_group->ro)
3415 ret = create_free_space_inode(fs_info, trans, block_group,
3423 * We want to set the generation to 0, that way if anything goes wrong
3424 * from here on out we know not to trust this cache when we load up next
3427 BTRFS_I(inode)->generation = 0;
3428 ret = btrfs_update_inode(trans, root, inode);
3431 * So theoretically we could recover from this, simply set the
3432 * super cache generation to 0 so we know to invalidate the
3433 * cache, but then we'd have to keep track of the block groups
3434 * that fail this way so we know we _have_ to reset this cache
3435 * before the next commit or risk reading stale cache. So to
3436 * limit our exposure to horrible edge cases lets just abort the
3437 * transaction, this only happens in really bad situations
3440 btrfs_abort_transaction(trans, ret);
3445 /* We've already setup this transaction, go ahead and exit */
3446 if (block_group->cache_generation == trans->transid &&
3447 i_size_read(inode)) {
3448 dcs = BTRFS_DC_SETUP;
3452 if (i_size_read(inode) > 0) {
3453 ret = btrfs_check_trunc_cache_free_space(fs_info,
3454 &fs_info->global_block_rsv);
3458 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3463 spin_lock(&block_group->lock);
3464 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3465 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3467 * don't bother trying to write stuff out _if_
3468 * a) we're not cached,
3469 * b) we're with nospace_cache mount option,
3470 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3472 dcs = BTRFS_DC_WRITTEN;
3473 spin_unlock(&block_group->lock);
3476 spin_unlock(&block_group->lock);
3479 * We hit an ENOSPC when setting up the cache in this transaction, just
3480 * skip doing the setup, we've already cleared the cache so we're safe.
3482 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3488 * Try to preallocate enough space based on how big the block group is.
3489 * Keep in mind this has to include any pinned space which could end up
3490 * taking up quite a bit since it's not folded into the other space
3493 num_pages = div_u64(block_group->key.offset, SZ_256M);
3498 num_pages *= PAGE_SIZE;
3500 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3504 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3505 num_pages, num_pages,
3508 * Our cache requires contiguous chunks so that we don't modify a bunch
3509 * of metadata or split extents when writing the cache out, which means
3510 * we can enospc if we are heavily fragmented in addition to just normal
3511 * out of space conditions. So if we hit this just skip setting up any
3512 * other block groups for this transaction, maybe we'll unpin enough
3513 * space the next time around.
3516 dcs = BTRFS_DC_SETUP;
3517 else if (ret == -ENOSPC)
3518 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3523 btrfs_release_path(path);
3525 spin_lock(&block_group->lock);
3526 if (!ret && dcs == BTRFS_DC_SETUP)
3527 block_group->cache_generation = trans->transid;
3528 block_group->disk_cache_state = dcs;
3529 spin_unlock(&block_group->lock);
3531 extent_changeset_free(data_reserved);
3535 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3536 struct btrfs_fs_info *fs_info)
3538 struct btrfs_block_group_cache *cache, *tmp;
3539 struct btrfs_transaction *cur_trans = trans->transaction;
3540 struct btrfs_path *path;
3542 if (list_empty(&cur_trans->dirty_bgs) ||
3543 !btrfs_test_opt(fs_info, SPACE_CACHE))
3546 path = btrfs_alloc_path();
3550 /* Could add new block groups, use _safe just in case */
3551 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3553 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3554 cache_save_setup(cache, trans, path);
3557 btrfs_free_path(path);
3562 * transaction commit does final block group cache writeback during a
3563 * critical section where nothing is allowed to change the FS. This is
3564 * required in order for the cache to actually match the block group,
3565 * but can introduce a lot of latency into the commit.
3567 * So, btrfs_start_dirty_block_groups is here to kick off block group
3568 * cache IO. There's a chance we'll have to redo some of it if the
3569 * block group changes again during the commit, but it greatly reduces
3570 * the commit latency by getting rid of the easy block groups while
3571 * we're still allowing others to join the commit.
3573 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3575 struct btrfs_fs_info *fs_info = trans->fs_info;
3576 struct btrfs_block_group_cache *cache;
3577 struct btrfs_transaction *cur_trans = trans->transaction;
3580 struct btrfs_path *path = NULL;
3582 struct list_head *io = &cur_trans->io_bgs;
3583 int num_started = 0;
3586 spin_lock(&cur_trans->dirty_bgs_lock);
3587 if (list_empty(&cur_trans->dirty_bgs)) {
3588 spin_unlock(&cur_trans->dirty_bgs_lock);
3591 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3592 spin_unlock(&cur_trans->dirty_bgs_lock);
3596 * make sure all the block groups on our dirty list actually
3599 btrfs_create_pending_block_groups(trans);
3602 path = btrfs_alloc_path();
3608 * cache_write_mutex is here only to save us from balance or automatic
3609 * removal of empty block groups deleting this block group while we are
3610 * writing out the cache
3612 mutex_lock(&trans->transaction->cache_write_mutex);
3613 while (!list_empty(&dirty)) {
3614 bool drop_reserve = true;
3616 cache = list_first_entry(&dirty,
3617 struct btrfs_block_group_cache,
3620 * this can happen if something re-dirties a block
3621 * group that is already under IO. Just wait for it to
3622 * finish and then do it all again
3624 if (!list_empty(&cache->io_list)) {
3625 list_del_init(&cache->io_list);
3626 btrfs_wait_cache_io(trans, cache, path);
3627 btrfs_put_block_group(cache);
3632 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3633 * if it should update the cache_state. Don't delete
3634 * until after we wait.
3636 * Since we're not running in the commit critical section
3637 * we need the dirty_bgs_lock to protect from update_block_group
3639 spin_lock(&cur_trans->dirty_bgs_lock);
3640 list_del_init(&cache->dirty_list);
3641 spin_unlock(&cur_trans->dirty_bgs_lock);
3645 cache_save_setup(cache, trans, path);
3647 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3648 cache->io_ctl.inode = NULL;
3649 ret = btrfs_write_out_cache(fs_info, trans,
3651 if (ret == 0 && cache->io_ctl.inode) {
3656 * The cache_write_mutex is protecting the
3657 * io_list, also refer to the definition of
3658 * btrfs_transaction::io_bgs for more details
3660 list_add_tail(&cache->io_list, io);
3663 * if we failed to write the cache, the
3664 * generation will be bad and life goes on
3670 ret = write_one_cache_group(trans, fs_info,
3673 * Our block group might still be attached to the list
3674 * of new block groups in the transaction handle of some
3675 * other task (struct btrfs_trans_handle->new_bgs). This
3676 * means its block group item isn't yet in the extent
3677 * tree. If this happens ignore the error, as we will
3678 * try again later in the critical section of the
3679 * transaction commit.
3681 if (ret == -ENOENT) {
3683 spin_lock(&cur_trans->dirty_bgs_lock);
3684 if (list_empty(&cache->dirty_list)) {
3685 list_add_tail(&cache->dirty_list,
3686 &cur_trans->dirty_bgs);
3687 btrfs_get_block_group(cache);
3688 drop_reserve = false;
3690 spin_unlock(&cur_trans->dirty_bgs_lock);
3692 btrfs_abort_transaction(trans, ret);
3696 /* if it's not on the io list, we need to put the block group */
3698 btrfs_put_block_group(cache);
3700 btrfs_delayed_refs_rsv_release(fs_info, 1);
3706 * Avoid blocking other tasks for too long. It might even save
3707 * us from writing caches for block groups that are going to be
3710 mutex_unlock(&trans->transaction->cache_write_mutex);
3711 mutex_lock(&trans->transaction->cache_write_mutex);
3713 mutex_unlock(&trans->transaction->cache_write_mutex);
3716 * go through delayed refs for all the stuff we've just kicked off
3717 * and then loop back (just once)
3719 ret = btrfs_run_delayed_refs(trans, 0);
3720 if (!ret && loops == 0) {
3722 spin_lock(&cur_trans->dirty_bgs_lock);
3723 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3725 * dirty_bgs_lock protects us from concurrent block group
3726 * deletes too (not just cache_write_mutex).
3728 if (!list_empty(&dirty)) {
3729 spin_unlock(&cur_trans->dirty_bgs_lock);
3732 spin_unlock(&cur_trans->dirty_bgs_lock);
3733 } else if (ret < 0) {
3734 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3737 btrfs_free_path(path);
3741 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3742 struct btrfs_fs_info *fs_info)
3744 struct btrfs_block_group_cache *cache;
3745 struct btrfs_transaction *cur_trans = trans->transaction;
3748 struct btrfs_path *path;
3749 struct list_head *io = &cur_trans->io_bgs;
3750 int num_started = 0;
3752 path = btrfs_alloc_path();
3757 * Even though we are in the critical section of the transaction commit,
3758 * we can still have concurrent tasks adding elements to this
3759 * transaction's list of dirty block groups. These tasks correspond to
3760 * endio free space workers started when writeback finishes for a
3761 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3762 * allocate new block groups as a result of COWing nodes of the root
3763 * tree when updating the free space inode. The writeback for the space
3764 * caches is triggered by an earlier call to
3765 * btrfs_start_dirty_block_groups() and iterations of the following
3767 * Also we want to do the cache_save_setup first and then run the
3768 * delayed refs to make sure we have the best chance at doing this all
3771 spin_lock(&cur_trans->dirty_bgs_lock);
3772 while (!list_empty(&cur_trans->dirty_bgs)) {
3773 cache = list_first_entry(&cur_trans->dirty_bgs,
3774 struct btrfs_block_group_cache,
3778 * this can happen if cache_save_setup re-dirties a block
3779 * group that is already under IO. Just wait for it to
3780 * finish and then do it all again
3782 if (!list_empty(&cache->io_list)) {
3783 spin_unlock(&cur_trans->dirty_bgs_lock);
3784 list_del_init(&cache->io_list);
3785 btrfs_wait_cache_io(trans, cache, path);
3786 btrfs_put_block_group(cache);
3787 spin_lock(&cur_trans->dirty_bgs_lock);
3791 * don't remove from the dirty list until after we've waited
3794 list_del_init(&cache->dirty_list);
3795 spin_unlock(&cur_trans->dirty_bgs_lock);
3798 cache_save_setup(cache, trans, path);
3801 ret = btrfs_run_delayed_refs(trans,
3802 (unsigned long) -1);
3804 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3805 cache->io_ctl.inode = NULL;
3806 ret = btrfs_write_out_cache(fs_info, trans,
3808 if (ret == 0 && cache->io_ctl.inode) {
3811 list_add_tail(&cache->io_list, io);
3814 * if we failed to write the cache, the
3815 * generation will be bad and life goes on
3821 ret = write_one_cache_group(trans, fs_info,
3824 * One of the free space endio workers might have
3825 * created a new block group while updating a free space
3826 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3827 * and hasn't released its transaction handle yet, in
3828 * which case the new block group is still attached to
3829 * its transaction handle and its creation has not
3830 * finished yet (no block group item in the extent tree
3831 * yet, etc). If this is the case, wait for all free
3832 * space endio workers to finish and retry. This is a
3833 * a very rare case so no need for a more efficient and
3836 if (ret == -ENOENT) {
3837 wait_event(cur_trans->writer_wait,
3838 atomic_read(&cur_trans->num_writers) == 1);
3839 ret = write_one_cache_group(trans, fs_info,
3843 btrfs_abort_transaction(trans, ret);
3846 /* if its not on the io list, we need to put the block group */
3848 btrfs_put_block_group(cache);
3849 btrfs_delayed_refs_rsv_release(fs_info, 1);
3850 spin_lock(&cur_trans->dirty_bgs_lock);
3852 spin_unlock(&cur_trans->dirty_bgs_lock);
3855 * Refer to the definition of io_bgs member for details why it's safe
3856 * to use it without any locking
3858 while (!list_empty(io)) {
3859 cache = list_first_entry(io, struct btrfs_block_group_cache,
3861 list_del_init(&cache->io_list);
3862 btrfs_wait_cache_io(trans, cache, path);
3863 btrfs_put_block_group(cache);
3866 btrfs_free_path(path);
3870 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3872 struct btrfs_block_group_cache *block_group;
3875 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3876 if (!block_group || block_group->ro)
3879 btrfs_put_block_group(block_group);
3883 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3885 struct btrfs_block_group_cache *bg;
3888 bg = btrfs_lookup_block_group(fs_info, bytenr);
3892 spin_lock(&bg->lock);
3896 atomic_inc(&bg->nocow_writers);
3897 spin_unlock(&bg->lock);
3899 /* no put on block group, done by btrfs_dec_nocow_writers */
3901 btrfs_put_block_group(bg);
3907 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3909 struct btrfs_block_group_cache *bg;
3911 bg = btrfs_lookup_block_group(fs_info, bytenr);
3913 if (atomic_dec_and_test(&bg->nocow_writers))
3914 wake_up_var(&bg->nocow_writers);
3916 * Once for our lookup and once for the lookup done by a previous call
3917 * to btrfs_inc_nocow_writers()
3919 btrfs_put_block_group(bg);
3920 btrfs_put_block_group(bg);
3923 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3925 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3928 static const char *alloc_name(u64 flags)
3931 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3933 case BTRFS_BLOCK_GROUP_METADATA:
3935 case BTRFS_BLOCK_GROUP_DATA:
3937 case BTRFS_BLOCK_GROUP_SYSTEM:
3941 return "invalid-combination";
3945 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3948 struct btrfs_space_info *space_info;
3952 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3956 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3963 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3964 INIT_LIST_HEAD(&space_info->block_groups[i]);
3965 init_rwsem(&space_info->groups_sem);
3966 spin_lock_init(&space_info->lock);
3967 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3968 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3969 init_waitqueue_head(&space_info->wait);
3970 INIT_LIST_HEAD(&space_info->ro_bgs);
3971 INIT_LIST_HEAD(&space_info->tickets);
3972 INIT_LIST_HEAD(&space_info->priority_tickets);
3974 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3975 info->space_info_kobj, "%s",
3976 alloc_name(space_info->flags));
3978 percpu_counter_destroy(&space_info->total_bytes_pinned);
3983 list_add_rcu(&space_info->list, &info->space_info);
3984 if (flags & BTRFS_BLOCK_GROUP_DATA)
3985 info->data_sinfo = space_info;
3990 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3991 u64 total_bytes, u64 bytes_used,
3993 struct btrfs_space_info **space_info)
3995 struct btrfs_space_info *found;
3998 factor = btrfs_bg_type_to_factor(flags);
4000 found = __find_space_info(info, flags);
4002 spin_lock(&found->lock);
4003 found->total_bytes += total_bytes;
4004 found->disk_total += total_bytes * factor;
4005 found->bytes_used += bytes_used;
4006 found->disk_used += bytes_used * factor;
4007 found->bytes_readonly += bytes_readonly;
4008 if (total_bytes > 0)
4010 space_info_add_new_bytes(info, found, total_bytes -
4011 bytes_used - bytes_readonly);
4012 spin_unlock(&found->lock);
4013 *space_info = found;
4016 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4018 u64 extra_flags = chunk_to_extended(flags) &
4019 BTRFS_EXTENDED_PROFILE_MASK;
4021 write_seqlock(&fs_info->profiles_lock);
4022 if (flags & BTRFS_BLOCK_GROUP_DATA)
4023 fs_info->avail_data_alloc_bits |= extra_flags;
4024 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4025 fs_info->avail_metadata_alloc_bits |= extra_flags;
4026 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4027 fs_info->avail_system_alloc_bits |= extra_flags;
4028 write_sequnlock(&fs_info->profiles_lock);
4032 * returns target flags in extended format or 0 if restripe for this
4033 * chunk_type is not in progress
4035 * should be called with balance_lock held
4037 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4039 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4045 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4046 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4047 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4048 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4049 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4050 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4051 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4052 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4053 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4060 * @flags: available profiles in extended format (see ctree.h)
4062 * Returns reduced profile in chunk format. If profile changing is in
4063 * progress (either running or paused) picks the target profile (if it's
4064 * already available), otherwise falls back to plain reducing.
4066 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4068 u64 num_devices = fs_info->fs_devices->rw_devices;
4074 * see if restripe for this chunk_type is in progress, if so
4075 * try to reduce to the target profile
4077 spin_lock(&fs_info->balance_lock);
4078 target = get_restripe_target(fs_info, flags);
4080 /* pick target profile only if it's already available */
4081 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4082 spin_unlock(&fs_info->balance_lock);
4083 return extended_to_chunk(target);
4086 spin_unlock(&fs_info->balance_lock);
4088 /* First, mask out the RAID levels which aren't possible */
4089 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4090 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4091 allowed |= btrfs_raid_array[raid_type].bg_flag;
4095 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4096 allowed = BTRFS_BLOCK_GROUP_RAID6;
4097 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4098 allowed = BTRFS_BLOCK_GROUP_RAID5;
4099 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4100 allowed = BTRFS_BLOCK_GROUP_RAID10;
4101 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4102 allowed = BTRFS_BLOCK_GROUP_RAID1;
4103 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4104 allowed = BTRFS_BLOCK_GROUP_RAID0;
4106 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4108 return extended_to_chunk(flags | allowed);
4111 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4118 seq = read_seqbegin(&fs_info->profiles_lock);
4120 if (flags & BTRFS_BLOCK_GROUP_DATA)
4121 flags |= fs_info->avail_data_alloc_bits;
4122 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4123 flags |= fs_info->avail_system_alloc_bits;
4124 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4125 flags |= fs_info->avail_metadata_alloc_bits;
4126 } while (read_seqretry(&fs_info->profiles_lock, seq));
4128 return btrfs_reduce_alloc_profile(fs_info, flags);
4131 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4133 struct btrfs_fs_info *fs_info = root->fs_info;
4138 flags = BTRFS_BLOCK_GROUP_DATA;
4139 else if (root == fs_info->chunk_root)
4140 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4142 flags = BTRFS_BLOCK_GROUP_METADATA;
4144 ret = get_alloc_profile(fs_info, flags);
4148 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4150 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4153 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4155 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4158 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4160 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4163 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4164 bool may_use_included)
4167 return s_info->bytes_used + s_info->bytes_reserved +
4168 s_info->bytes_pinned + s_info->bytes_readonly +
4169 (may_use_included ? s_info->bytes_may_use : 0);
4172 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4174 struct btrfs_root *root = inode->root;
4175 struct btrfs_fs_info *fs_info = root->fs_info;
4176 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4179 int need_commit = 2;
4180 int have_pinned_space;
4182 /* make sure bytes are sectorsize aligned */
4183 bytes = ALIGN(bytes, fs_info->sectorsize);
4185 if (btrfs_is_free_space_inode(inode)) {
4187 ASSERT(current->journal_info);
4191 /* make sure we have enough space to handle the data first */
4192 spin_lock(&data_sinfo->lock);
4193 used = btrfs_space_info_used(data_sinfo, true);
4195 if (used + bytes > data_sinfo->total_bytes) {
4196 struct btrfs_trans_handle *trans;
4199 * if we don't have enough free bytes in this space then we need
4200 * to alloc a new chunk.
4202 if (!data_sinfo->full) {
4205 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4206 spin_unlock(&data_sinfo->lock);
4208 alloc_target = btrfs_data_alloc_profile(fs_info);
4210 * It is ugly that we don't call nolock join
4211 * transaction for the free space inode case here.
4212 * But it is safe because we only do the data space
4213 * reservation for the free space cache in the
4214 * transaction context, the common join transaction
4215 * just increase the counter of the current transaction
4216 * handler, doesn't try to acquire the trans_lock of
4219 trans = btrfs_join_transaction(root);
4221 return PTR_ERR(trans);
4223 ret = do_chunk_alloc(trans, alloc_target,
4224 CHUNK_ALLOC_NO_FORCE);
4225 btrfs_end_transaction(trans);
4230 have_pinned_space = 1;
4239 * If we don't have enough pinned space to deal with this
4240 * allocation, and no removed chunk in current transaction,
4241 * don't bother committing the transaction.
4243 have_pinned_space = __percpu_counter_compare(
4244 &data_sinfo->total_bytes_pinned,
4245 used + bytes - data_sinfo->total_bytes,
4246 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4247 spin_unlock(&data_sinfo->lock);
4249 /* commit the current transaction and try again */
4254 if (need_commit > 0) {
4255 btrfs_start_delalloc_roots(fs_info, -1);
4256 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4260 trans = btrfs_join_transaction(root);
4262 return PTR_ERR(trans);
4263 if (have_pinned_space >= 0 ||
4264 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4265 &trans->transaction->flags) ||
4267 ret = btrfs_commit_transaction(trans);
4271 * The cleaner kthread might still be doing iput
4272 * operations. Wait for it to finish so that
4273 * more space is released. We don't need to
4274 * explicitly run the delayed iputs here because
4275 * the commit_transaction would have woken up
4278 ret = btrfs_wait_on_delayed_iputs(fs_info);
4283 btrfs_end_transaction(trans);
4287 trace_btrfs_space_reservation(fs_info,
4288 "space_info:enospc",
4289 data_sinfo->flags, bytes, 1);
4292 update_bytes_may_use(data_sinfo, bytes);
4293 trace_btrfs_space_reservation(fs_info, "space_info",
4294 data_sinfo->flags, bytes, 1);
4295 spin_unlock(&data_sinfo->lock);
4300 int btrfs_check_data_free_space(struct inode *inode,
4301 struct extent_changeset **reserved, u64 start, u64 len)
4303 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4306 /* align the range */
4307 len = round_up(start + len, fs_info->sectorsize) -
4308 round_down(start, fs_info->sectorsize);
4309 start = round_down(start, fs_info->sectorsize);
4311 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4315 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4316 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4318 btrfs_free_reserved_data_space_noquota(inode, start, len);
4325 * Called if we need to clear a data reservation for this inode
4326 * Normally in a error case.
4328 * This one will *NOT* use accurate qgroup reserved space API, just for case
4329 * which we can't sleep and is sure it won't affect qgroup reserved space.
4330 * Like clear_bit_hook().
4332 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4335 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4336 struct btrfs_space_info *data_sinfo;
4338 /* Make sure the range is aligned to sectorsize */
4339 len = round_up(start + len, fs_info->sectorsize) -
4340 round_down(start, fs_info->sectorsize);
4341 start = round_down(start, fs_info->sectorsize);
4343 data_sinfo = fs_info->data_sinfo;
4344 spin_lock(&data_sinfo->lock);
4345 update_bytes_may_use(data_sinfo, -len);
4346 trace_btrfs_space_reservation(fs_info, "space_info",
4347 data_sinfo->flags, len, 0);
4348 spin_unlock(&data_sinfo->lock);
4352 * Called if we need to clear a data reservation for this inode
4353 * Normally in a error case.
4355 * This one will handle the per-inode data rsv map for accurate reserved
4358 void btrfs_free_reserved_data_space(struct inode *inode,
4359 struct extent_changeset *reserved, u64 start, u64 len)
4361 struct btrfs_root *root = BTRFS_I(inode)->root;
4363 /* Make sure the range is aligned to sectorsize */
4364 len = round_up(start + len, root->fs_info->sectorsize) -
4365 round_down(start, root->fs_info->sectorsize);
4366 start = round_down(start, root->fs_info->sectorsize);
4368 btrfs_free_reserved_data_space_noquota(inode, start, len);
4369 btrfs_qgroup_free_data(inode, reserved, start, len);
4372 static void force_metadata_allocation(struct btrfs_fs_info *info)
4374 struct list_head *head = &info->space_info;
4375 struct btrfs_space_info *found;
4378 list_for_each_entry_rcu(found, head, list) {
4379 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4380 found->force_alloc = CHUNK_ALLOC_FORCE;
4385 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4387 return (global->size << 1);
4390 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4391 struct btrfs_space_info *sinfo, int force)
4393 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4396 if (force == CHUNK_ALLOC_FORCE)
4400 * in limited mode, we want to have some free space up to
4401 * about 1% of the FS size.
4403 if (force == CHUNK_ALLOC_LIMITED) {
4404 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4405 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4407 if (sinfo->total_bytes - bytes_used < thresh)
4411 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4416 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4420 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4421 BTRFS_BLOCK_GROUP_RAID0 |
4422 BTRFS_BLOCK_GROUP_RAID5 |
4423 BTRFS_BLOCK_GROUP_RAID6))
4424 num_dev = fs_info->fs_devices->rw_devices;
4425 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4428 num_dev = 1; /* DUP or single */
4434 * If @is_allocation is true, reserve space in the system space info necessary
4435 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4438 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4440 struct btrfs_fs_info *fs_info = trans->fs_info;
4441 struct btrfs_space_info *info;
4448 * Needed because we can end up allocating a system chunk and for an
4449 * atomic and race free space reservation in the chunk block reserve.
4451 lockdep_assert_held(&fs_info->chunk_mutex);
4453 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4454 spin_lock(&info->lock);
4455 left = info->total_bytes - btrfs_space_info_used(info, true);
4456 spin_unlock(&info->lock);
4458 num_devs = get_profile_num_devs(fs_info, type);
4460 /* num_devs device items to update and 1 chunk item to add or remove */
4461 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4462 btrfs_calc_trans_metadata_size(fs_info, 1);
4464 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4465 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4466 left, thresh, type);
4467 dump_space_info(fs_info, info, 0, 0);
4470 if (left < thresh) {
4471 u64 flags = btrfs_system_alloc_profile(fs_info);
4474 * Ignore failure to create system chunk. We might end up not
4475 * needing it, as we might not need to COW all nodes/leafs from
4476 * the paths we visit in the chunk tree (they were already COWed
4477 * or created in the current transaction for example).
4479 ret = btrfs_alloc_chunk(trans, flags);
4483 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4484 &fs_info->chunk_block_rsv,
4485 thresh, BTRFS_RESERVE_NO_FLUSH);
4487 trans->chunk_bytes_reserved += thresh;
4492 * If force is CHUNK_ALLOC_FORCE:
4493 * - return 1 if it successfully allocates a chunk,
4494 * - return errors including -ENOSPC otherwise.
4495 * If force is NOT CHUNK_ALLOC_FORCE:
4496 * - return 0 if it doesn't need to allocate a new chunk,
4497 * - return 1 if it successfully allocates a chunk,
4498 * - return errors including -ENOSPC otherwise.
4500 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4503 struct btrfs_fs_info *fs_info = trans->fs_info;
4504 struct btrfs_space_info *space_info;
4505 bool wait_for_alloc = false;
4506 bool should_alloc = false;
4509 /* Don't re-enter if we're already allocating a chunk */
4510 if (trans->allocating_chunk)
4513 space_info = __find_space_info(fs_info, flags);
4517 spin_lock(&space_info->lock);
4518 if (force < space_info->force_alloc)
4519 force = space_info->force_alloc;
4520 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4521 if (space_info->full) {
4522 /* No more free physical space */
4527 spin_unlock(&space_info->lock);
4529 } else if (!should_alloc) {
4530 spin_unlock(&space_info->lock);
4532 } else if (space_info->chunk_alloc) {
4534 * Someone is already allocating, so we need to block
4535 * until this someone is finished and then loop to
4536 * recheck if we should continue with our allocation
4539 wait_for_alloc = true;
4540 spin_unlock(&space_info->lock);
4541 mutex_lock(&fs_info->chunk_mutex);
4542 mutex_unlock(&fs_info->chunk_mutex);
4544 /* Proceed with allocation */
4545 space_info->chunk_alloc = 1;
4546 wait_for_alloc = false;
4547 spin_unlock(&space_info->lock);
4551 } while (wait_for_alloc);
4553 mutex_lock(&fs_info->chunk_mutex);
4554 trans->allocating_chunk = true;
4557 * If we have mixed data/metadata chunks we want to make sure we keep
4558 * allocating mixed chunks instead of individual chunks.
4560 if (btrfs_mixed_space_info(space_info))
4561 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4564 * if we're doing a data chunk, go ahead and make sure that
4565 * we keep a reasonable number of metadata chunks allocated in the
4568 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4569 fs_info->data_chunk_allocations++;
4570 if (!(fs_info->data_chunk_allocations %
4571 fs_info->metadata_ratio))
4572 force_metadata_allocation(fs_info);
4576 * Check if we have enough space in SYSTEM chunk because we may need
4577 * to update devices.
4579 check_system_chunk(trans, flags);
4581 ret = btrfs_alloc_chunk(trans, flags);
4582 trans->allocating_chunk = false;
4584 spin_lock(&space_info->lock);
4587 space_info->full = 1;
4592 space_info->max_extent_size = 0;
4595 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4597 space_info->chunk_alloc = 0;
4598 spin_unlock(&space_info->lock);
4599 mutex_unlock(&fs_info->chunk_mutex);
4601 * When we allocate a new chunk we reserve space in the chunk block
4602 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4603 * add new nodes/leafs to it if we end up needing to do it when
4604 * inserting the chunk item and updating device items as part of the
4605 * second phase of chunk allocation, performed by
4606 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4607 * large number of new block groups to create in our transaction
4608 * handle's new_bgs list to avoid exhausting the chunk block reserve
4609 * in extreme cases - like having a single transaction create many new
4610 * block groups when starting to write out the free space caches of all
4611 * the block groups that were made dirty during the lifetime of the
4614 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4615 btrfs_create_pending_block_groups(trans);
4620 static int can_overcommit(struct btrfs_fs_info *fs_info,
4621 struct btrfs_space_info *space_info, u64 bytes,
4622 enum btrfs_reserve_flush_enum flush,
4625 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4632 /* Don't overcommit when in mixed mode. */
4633 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4637 profile = btrfs_system_alloc_profile(fs_info);
4639 profile = btrfs_metadata_alloc_profile(fs_info);
4641 used = btrfs_space_info_used(space_info, false);
4644 * We only want to allow over committing if we have lots of actual space
4645 * free, but if we don't have enough space to handle the global reserve
4646 * space then we could end up having a real enospc problem when trying
4647 * to allocate a chunk or some other such important allocation.
4649 spin_lock(&global_rsv->lock);
4650 space_size = calc_global_rsv_need_space(global_rsv);
4651 spin_unlock(&global_rsv->lock);
4652 if (used + space_size >= space_info->total_bytes)
4655 used += space_info->bytes_may_use;
4657 avail = atomic64_read(&fs_info->free_chunk_space);
4660 * If we have dup, raid1 or raid10 then only half of the free
4661 * space is actually usable. For raid56, the space info used
4662 * doesn't include the parity drive, so we don't have to
4665 factor = btrfs_bg_type_to_factor(profile);
4666 avail = div_u64(avail, factor);
4669 * If we aren't flushing all things, let us overcommit up to
4670 * 1/2th of the space. If we can flush, don't let us overcommit
4671 * too much, let it overcommit up to 1/8 of the space.
4673 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4678 if (used + bytes < space_info->total_bytes + avail)
4683 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4684 unsigned long nr_pages, int nr_items)
4686 struct super_block *sb = fs_info->sb;
4688 if (down_read_trylock(&sb->s_umount)) {
4689 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4690 up_read(&sb->s_umount);
4693 * We needn't worry the filesystem going from r/w to r/o though
4694 * we don't acquire ->s_umount mutex, because the filesystem
4695 * should guarantee the delalloc inodes list be empty after
4696 * the filesystem is readonly(all dirty pages are written to
4699 btrfs_start_delalloc_roots(fs_info, nr_items);
4700 if (!current->journal_info)
4701 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4705 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4711 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4712 nr = div64_u64(to_reclaim, bytes);
4718 #define EXTENT_SIZE_PER_ITEM SZ_256K
4721 * shrink metadata reservation for delalloc
4723 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4724 u64 orig, bool wait_ordered)
4726 struct btrfs_space_info *space_info;
4727 struct btrfs_trans_handle *trans;
4732 unsigned long nr_pages;
4735 /* Calc the number of the pages we need flush for space reservation */
4736 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4737 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4739 trans = (struct btrfs_trans_handle *)current->journal_info;
4740 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4742 delalloc_bytes = percpu_counter_sum_positive(
4743 &fs_info->delalloc_bytes);
4744 if (delalloc_bytes == 0) {
4748 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4753 while (delalloc_bytes && loops < 3) {
4754 nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT;
4757 * Triggers inode writeback for up to nr_pages. This will invoke
4758 * ->writepages callback and trigger delalloc filling
4759 * (btrfs_run_delalloc_range()).
4761 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4764 * We need to wait for the compressed pages to start before
4767 async_pages = atomic_read(&fs_info->async_delalloc_pages);
4772 * Calculate how many compressed pages we want to be written
4773 * before we continue. I.e if there are more async pages than we
4774 * require wait_event will wait until nr_pages are written.
4776 if (async_pages <= nr_pages)
4779 async_pages -= nr_pages;
4781 wait_event(fs_info->async_submit_wait,
4782 atomic_read(&fs_info->async_delalloc_pages) <=
4785 spin_lock(&space_info->lock);
4786 if (list_empty(&space_info->tickets) &&
4787 list_empty(&space_info->priority_tickets)) {
4788 spin_unlock(&space_info->lock);
4791 spin_unlock(&space_info->lock);
4794 if (wait_ordered && !trans) {
4795 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4797 time_left = schedule_timeout_killable(1);
4801 delalloc_bytes = percpu_counter_sum_positive(
4802 &fs_info->delalloc_bytes);
4806 struct reserve_ticket {
4810 struct list_head list;
4811 wait_queue_head_t wait;
4815 * maybe_commit_transaction - possibly commit the transaction if its ok to
4816 * @root - the root we're allocating for
4817 * @bytes - the number of bytes we want to reserve
4818 * @force - force the commit
4820 * This will check to make sure that committing the transaction will actually
4821 * get us somewhere and then commit the transaction if it does. Otherwise it
4822 * will return -ENOSPC.
4824 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4825 struct btrfs_space_info *space_info)
4827 struct reserve_ticket *ticket = NULL;
4828 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4829 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4830 struct btrfs_trans_handle *trans;
4832 u64 reclaim_bytes = 0;
4834 trans = (struct btrfs_trans_handle *)current->journal_info;
4838 spin_lock(&space_info->lock);
4839 if (!list_empty(&space_info->priority_tickets))
4840 ticket = list_first_entry(&space_info->priority_tickets,
4841 struct reserve_ticket, list);
4842 else if (!list_empty(&space_info->tickets))
4843 ticket = list_first_entry(&space_info->tickets,
4844 struct reserve_ticket, list);
4845 bytes_needed = (ticket) ? ticket->bytes : 0;
4846 spin_unlock(&space_info->lock);
4851 trans = btrfs_join_transaction(fs_info->extent_root);
4853 return PTR_ERR(trans);
4856 * See if there is enough pinned space to make this reservation, or if
4857 * we have block groups that are going to be freed, allowing us to
4858 * possibly do a chunk allocation the next loop through.
4860 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) ||
4861 __percpu_counter_compare(&space_info->total_bytes_pinned,
4863 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4867 * See if there is some space in the delayed insertion reservation for
4870 if (space_info != delayed_rsv->space_info)
4873 spin_lock(&delayed_rsv->lock);
4874 reclaim_bytes += delayed_rsv->reserved;
4875 spin_unlock(&delayed_rsv->lock);
4877 spin_lock(&delayed_refs_rsv->lock);
4878 reclaim_bytes += delayed_refs_rsv->reserved;
4879 spin_unlock(&delayed_refs_rsv->lock);
4880 if (reclaim_bytes >= bytes_needed)
4882 bytes_needed -= reclaim_bytes;
4884 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4886 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0)
4890 return btrfs_commit_transaction(trans);
4892 btrfs_end_transaction(trans);
4897 * Try to flush some data based on policy set by @state. This is only advisory
4898 * and may fail for various reasons. The caller is supposed to examine the
4899 * state of @space_info to detect the outcome.
4901 static void flush_space(struct btrfs_fs_info *fs_info,
4902 struct btrfs_space_info *space_info, u64 num_bytes,
4905 struct btrfs_root *root = fs_info->extent_root;
4906 struct btrfs_trans_handle *trans;
4911 case FLUSH_DELAYED_ITEMS_NR:
4912 case FLUSH_DELAYED_ITEMS:
4913 if (state == FLUSH_DELAYED_ITEMS_NR)
4914 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4918 trans = btrfs_join_transaction(root);
4919 if (IS_ERR(trans)) {
4920 ret = PTR_ERR(trans);
4923 ret = btrfs_run_delayed_items_nr(trans, nr);
4924 btrfs_end_transaction(trans);
4926 case FLUSH_DELALLOC:
4927 case FLUSH_DELALLOC_WAIT:
4928 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4929 state == FLUSH_DELALLOC_WAIT);
4931 case FLUSH_DELAYED_REFS_NR:
4932 case FLUSH_DELAYED_REFS:
4933 trans = btrfs_join_transaction(root);
4934 if (IS_ERR(trans)) {
4935 ret = PTR_ERR(trans);
4938 if (state == FLUSH_DELAYED_REFS_NR)
4939 nr = calc_reclaim_items_nr(fs_info, num_bytes);
4942 btrfs_run_delayed_refs(trans, nr);
4943 btrfs_end_transaction(trans);
4946 case ALLOC_CHUNK_FORCE:
4947 trans = btrfs_join_transaction(root);
4948 if (IS_ERR(trans)) {
4949 ret = PTR_ERR(trans);
4952 ret = do_chunk_alloc(trans,
4953 btrfs_metadata_alloc_profile(fs_info),
4954 (state == ALLOC_CHUNK) ?
4955 CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE);
4956 btrfs_end_transaction(trans);
4957 if (ret > 0 || ret == -ENOSPC)
4962 * If we have pending delayed iputs then we could free up a
4963 * bunch of pinned space, so make sure we run the iputs before
4964 * we do our pinned bytes check below.
4966 btrfs_run_delayed_iputs(fs_info);
4967 btrfs_wait_on_delayed_iputs(fs_info);
4969 ret = may_commit_transaction(fs_info, space_info);
4976 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4982 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4983 struct btrfs_space_info *space_info,
4986 struct reserve_ticket *ticket;
4991 list_for_each_entry(ticket, &space_info->tickets, list)
4992 to_reclaim += ticket->bytes;
4993 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4994 to_reclaim += ticket->bytes;
4998 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4999 if (can_overcommit(fs_info, space_info, to_reclaim,
5000 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5003 used = btrfs_space_info_used(space_info, true);
5005 if (can_overcommit(fs_info, space_info, SZ_1M,
5006 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5007 expected = div_factor_fine(space_info->total_bytes, 95);
5009 expected = div_factor_fine(space_info->total_bytes, 90);
5011 if (used > expected)
5012 to_reclaim = used - expected;
5015 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5016 space_info->bytes_reserved);
5020 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5021 struct btrfs_space_info *space_info,
5022 u64 used, bool system_chunk)
5024 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5026 /* If we're just plain full then async reclaim just slows us down. */
5027 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5030 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5034 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5035 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5038 static bool wake_all_tickets(struct list_head *head)
5040 struct reserve_ticket *ticket;
5042 while (!list_empty(head)) {
5043 ticket = list_first_entry(head, struct reserve_ticket, list);
5044 list_del_init(&ticket->list);
5045 ticket->error = -ENOSPC;
5046 wake_up(&ticket->wait);
5047 if (ticket->bytes != ticket->orig_bytes)
5054 * This is for normal flushers, we can wait all goddamned day if we want to. We
5055 * will loop and continuously try to flush as long as we are making progress.
5056 * We count progress as clearing off tickets each time we have to loop.
5058 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5060 struct btrfs_fs_info *fs_info;
5061 struct btrfs_space_info *space_info;
5064 int commit_cycles = 0;
5065 u64 last_tickets_id;
5067 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5068 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5070 spin_lock(&space_info->lock);
5071 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5074 space_info->flush = 0;
5075 spin_unlock(&space_info->lock);
5078 last_tickets_id = space_info->tickets_id;
5079 spin_unlock(&space_info->lock);
5081 flush_state = FLUSH_DELAYED_ITEMS_NR;
5083 flush_space(fs_info, space_info, to_reclaim, flush_state);
5084 spin_lock(&space_info->lock);
5085 if (list_empty(&space_info->tickets)) {
5086 space_info->flush = 0;
5087 spin_unlock(&space_info->lock);
5090 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5093 if (last_tickets_id == space_info->tickets_id) {
5096 last_tickets_id = space_info->tickets_id;
5097 flush_state = FLUSH_DELAYED_ITEMS_NR;
5103 * We don't want to force a chunk allocation until we've tried
5104 * pretty hard to reclaim space. Think of the case where we
5105 * freed up a bunch of space and so have a lot of pinned space
5106 * to reclaim. We would rather use that than possibly create a
5107 * underutilized metadata chunk. So if this is our first run
5108 * through the flushing state machine skip ALLOC_CHUNK_FORCE and
5109 * commit the transaction. If nothing has changed the next go
5110 * around then we can force a chunk allocation.
5112 if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles)
5115 if (flush_state > COMMIT_TRANS) {
5117 if (commit_cycles > 2) {
5118 if (wake_all_tickets(&space_info->tickets)) {
5119 flush_state = FLUSH_DELAYED_ITEMS_NR;
5122 space_info->flush = 0;
5125 flush_state = FLUSH_DELAYED_ITEMS_NR;
5128 spin_unlock(&space_info->lock);
5129 } while (flush_state <= COMMIT_TRANS);
5132 void btrfs_init_async_reclaim_work(struct work_struct *work)
5134 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5137 static const enum btrfs_flush_state priority_flush_states[] = {
5138 FLUSH_DELAYED_ITEMS_NR,
5139 FLUSH_DELAYED_ITEMS,
5143 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5144 struct btrfs_space_info *space_info,
5145 struct reserve_ticket *ticket)
5150 spin_lock(&space_info->lock);
5151 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5154 spin_unlock(&space_info->lock);
5157 spin_unlock(&space_info->lock);
5161 flush_space(fs_info, space_info, to_reclaim,
5162 priority_flush_states[flush_state]);
5164 spin_lock(&space_info->lock);
5165 if (ticket->bytes == 0) {
5166 spin_unlock(&space_info->lock);
5169 spin_unlock(&space_info->lock);
5170 } while (flush_state < ARRAY_SIZE(priority_flush_states));
5173 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5174 struct btrfs_space_info *space_info,
5175 struct reserve_ticket *ticket)
5179 u64 reclaim_bytes = 0;
5182 spin_lock(&space_info->lock);
5183 while (ticket->bytes > 0 && ticket->error == 0) {
5184 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5189 spin_unlock(&space_info->lock);
5193 finish_wait(&ticket->wait, &wait);
5194 spin_lock(&space_info->lock);
5197 ret = ticket->error;
5198 if (!list_empty(&ticket->list))
5199 list_del_init(&ticket->list);
5200 if (ticket->bytes && ticket->bytes < ticket->orig_bytes)
5201 reclaim_bytes = ticket->orig_bytes - ticket->bytes;
5202 spin_unlock(&space_info->lock);
5205 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5210 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5211 * @root - the root we're allocating for
5212 * @space_info - the space info we want to allocate from
5213 * @orig_bytes - the number of bytes we want
5214 * @flush - whether or not we can flush to make our reservation
5216 * This will reserve orig_bytes number of bytes from the space info associated
5217 * with the block_rsv. If there is not enough space it will make an attempt to
5218 * flush out space to make room. It will do this by flushing delalloc if
5219 * possible or committing the transaction. If flush is 0 then no attempts to
5220 * regain reservations will be made and this will fail if there is not enough
5223 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5224 struct btrfs_space_info *space_info,
5226 enum btrfs_reserve_flush_enum flush,
5229 struct reserve_ticket ticket;
5231 u64 reclaim_bytes = 0;
5235 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5237 spin_lock(&space_info->lock);
5239 used = btrfs_space_info_used(space_info, true);
5242 * If we have enough space then hooray, make our reservation and carry
5243 * on. If not see if we can overcommit, and if we can, hooray carry on.
5244 * If not things get more complicated.
5246 if (used + orig_bytes <= space_info->total_bytes) {
5247 update_bytes_may_use(space_info, orig_bytes);
5248 trace_btrfs_space_reservation(fs_info, "space_info",
5249 space_info->flags, orig_bytes, 1);
5251 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5253 update_bytes_may_use(space_info, orig_bytes);
5254 trace_btrfs_space_reservation(fs_info, "space_info",
5255 space_info->flags, orig_bytes, 1);
5260 * If we couldn't make a reservation then setup our reservation ticket
5261 * and kick the async worker if it's not already running.
5263 * If we are a priority flusher then we just need to add our ticket to
5264 * the list and we will do our own flushing further down.
5266 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5267 ticket.orig_bytes = orig_bytes;
5268 ticket.bytes = orig_bytes;
5270 init_waitqueue_head(&ticket.wait);
5271 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5272 list_add_tail(&ticket.list, &space_info->tickets);
5273 if (!space_info->flush) {
5274 space_info->flush = 1;
5275 trace_btrfs_trigger_flush(fs_info,
5279 queue_work(system_unbound_wq,
5280 &fs_info->async_reclaim_work);
5283 list_add_tail(&ticket.list,
5284 &space_info->priority_tickets);
5286 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5289 * We will do the space reservation dance during log replay,
5290 * which means we won't have fs_info->fs_root set, so don't do
5291 * the async reclaim as we will panic.
5293 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5294 need_do_async_reclaim(fs_info, space_info,
5295 used, system_chunk) &&
5296 !work_busy(&fs_info->async_reclaim_work)) {
5297 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5298 orig_bytes, flush, "preempt");
5299 queue_work(system_unbound_wq,
5300 &fs_info->async_reclaim_work);
5303 spin_unlock(&space_info->lock);
5304 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5307 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5308 return wait_reserve_ticket(fs_info, space_info, &ticket);
5311 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5312 spin_lock(&space_info->lock);
5314 if (ticket.bytes < orig_bytes)
5315 reclaim_bytes = orig_bytes - ticket.bytes;
5316 list_del_init(&ticket.list);
5319 spin_unlock(&space_info->lock);
5322 space_info_add_old_bytes(fs_info, space_info, reclaim_bytes);
5323 ASSERT(list_empty(&ticket.list));
5328 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5329 * @root - the root we're allocating for
5330 * @block_rsv - the block_rsv we're allocating for
5331 * @orig_bytes - the number of bytes we want
5332 * @flush - whether or not we can flush to make our reservation
5334 * This will reserve orig_bytes number of bytes from the space info associated
5335 * with the block_rsv. If there is not enough space it will make an attempt to
5336 * flush out space to make room. It will do this by flushing delalloc if
5337 * possible or committing the transaction. If flush is 0 then no attempts to
5338 * regain reservations will be made and this will fail if there is not enough
5341 static int reserve_metadata_bytes(struct btrfs_root *root,
5342 struct btrfs_block_rsv *block_rsv,
5344 enum btrfs_reserve_flush_enum flush)
5346 struct btrfs_fs_info *fs_info = root->fs_info;
5347 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5349 bool system_chunk = (root == fs_info->chunk_root);
5351 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5352 orig_bytes, flush, system_chunk);
5353 if (ret == -ENOSPC &&
5354 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5355 if (block_rsv != global_rsv &&
5356 !block_rsv_use_bytes(global_rsv, orig_bytes))
5359 if (ret == -ENOSPC) {
5360 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5361 block_rsv->space_info->flags,
5364 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5365 dump_space_info(fs_info, block_rsv->space_info,
5371 static struct btrfs_block_rsv *get_block_rsv(
5372 const struct btrfs_trans_handle *trans,
5373 const struct btrfs_root *root)
5375 struct btrfs_fs_info *fs_info = root->fs_info;
5376 struct btrfs_block_rsv *block_rsv = NULL;
5378 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5379 (root == fs_info->csum_root && trans->adding_csums) ||
5380 (root == fs_info->uuid_root))
5381 block_rsv = trans->block_rsv;
5384 block_rsv = root->block_rsv;
5387 block_rsv = &fs_info->empty_block_rsv;
5392 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5396 spin_lock(&block_rsv->lock);
5397 if (block_rsv->reserved >= num_bytes) {
5398 block_rsv->reserved -= num_bytes;
5399 if (block_rsv->reserved < block_rsv->size)
5400 block_rsv->full = 0;
5403 spin_unlock(&block_rsv->lock);
5407 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5408 u64 num_bytes, bool update_size)
5410 spin_lock(&block_rsv->lock);
5411 block_rsv->reserved += num_bytes;
5413 block_rsv->size += num_bytes;
5414 else if (block_rsv->reserved >= block_rsv->size)
5415 block_rsv->full = 1;
5416 spin_unlock(&block_rsv->lock);
5419 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5420 struct btrfs_block_rsv *dest, u64 num_bytes,
5423 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5426 if (global_rsv->space_info != dest->space_info)
5429 spin_lock(&global_rsv->lock);
5430 min_bytes = div_factor(global_rsv->size, min_factor);
5431 if (global_rsv->reserved < min_bytes + num_bytes) {
5432 spin_unlock(&global_rsv->lock);
5435 global_rsv->reserved -= num_bytes;
5436 if (global_rsv->reserved < global_rsv->size)
5437 global_rsv->full = 0;
5438 spin_unlock(&global_rsv->lock);
5440 block_rsv_add_bytes(dest, num_bytes, true);
5445 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5446 * @fs_info - the fs info for our fs.
5447 * @src - the source block rsv to transfer from.
5448 * @num_bytes - the number of bytes to transfer.
5450 * This transfers up to the num_bytes amount from the src rsv to the
5451 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5453 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5454 struct btrfs_block_rsv *src,
5457 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5460 spin_lock(&src->lock);
5461 src->reserved -= num_bytes;
5462 src->size -= num_bytes;
5463 spin_unlock(&src->lock);
5465 spin_lock(&delayed_refs_rsv->lock);
5466 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5467 u64 delta = delayed_refs_rsv->size -
5468 delayed_refs_rsv->reserved;
5469 if (num_bytes > delta) {
5470 to_free = num_bytes - delta;
5474 to_free = num_bytes;
5479 delayed_refs_rsv->reserved += num_bytes;
5480 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5481 delayed_refs_rsv->full = 1;
5482 spin_unlock(&delayed_refs_rsv->lock);
5485 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5488 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5493 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5494 * @fs_info - the fs_info for our fs.
5495 * @flush - control how we can flush for this reservation.
5497 * This will refill the delayed block_rsv up to 1 items size worth of space and
5498 * will return -ENOSPC if we can't make the reservation.
5500 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5501 enum btrfs_reserve_flush_enum flush)
5503 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5504 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5508 spin_lock(&block_rsv->lock);
5509 if (block_rsv->reserved < block_rsv->size) {
5510 num_bytes = block_rsv->size - block_rsv->reserved;
5511 num_bytes = min(num_bytes, limit);
5513 spin_unlock(&block_rsv->lock);
5518 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5522 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5523 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5529 * This is for space we already have accounted in space_info->bytes_may_use, so
5530 * basically when we're returning space from block_rsv's.
5532 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5533 struct btrfs_space_info *space_info,
5536 struct reserve_ticket *ticket;
5537 struct list_head *head;
5539 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5540 bool check_overcommit = false;
5542 spin_lock(&space_info->lock);
5543 head = &space_info->priority_tickets;
5546 * If we are over our limit then we need to check and see if we can
5547 * overcommit, and if we can't then we just need to free up our space
5548 * and not satisfy any requests.
5550 used = btrfs_space_info_used(space_info, true);
5551 if (used - num_bytes >= space_info->total_bytes)
5552 check_overcommit = true;
5554 while (!list_empty(head) && num_bytes) {
5555 ticket = list_first_entry(head, struct reserve_ticket,
5558 * We use 0 bytes because this space is already reserved, so
5559 * adding the ticket space would be a double count.
5561 if (check_overcommit &&
5562 !can_overcommit(fs_info, space_info, 0, flush, false))
5564 if (num_bytes >= ticket->bytes) {
5565 list_del_init(&ticket->list);
5566 num_bytes -= ticket->bytes;
5568 space_info->tickets_id++;
5569 wake_up(&ticket->wait);
5571 ticket->bytes -= num_bytes;
5576 if (num_bytes && head == &space_info->priority_tickets) {
5577 head = &space_info->tickets;
5578 flush = BTRFS_RESERVE_FLUSH_ALL;
5581 update_bytes_may_use(space_info, -num_bytes);
5582 trace_btrfs_space_reservation(fs_info, "space_info",
5583 space_info->flags, num_bytes, 0);
5584 spin_unlock(&space_info->lock);
5588 * This is for newly allocated space that isn't accounted in
5589 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5590 * we use this helper.
5592 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5593 struct btrfs_space_info *space_info,
5596 struct reserve_ticket *ticket;
5597 struct list_head *head = &space_info->priority_tickets;
5600 while (!list_empty(head) && num_bytes) {
5601 ticket = list_first_entry(head, struct reserve_ticket,
5603 if (num_bytes >= ticket->bytes) {
5604 trace_btrfs_space_reservation(fs_info, "space_info",
5607 list_del_init(&ticket->list);
5608 num_bytes -= ticket->bytes;
5609 update_bytes_may_use(space_info, ticket->bytes);
5611 space_info->tickets_id++;
5612 wake_up(&ticket->wait);
5614 trace_btrfs_space_reservation(fs_info, "space_info",
5617 update_bytes_may_use(space_info, num_bytes);
5618 ticket->bytes -= num_bytes;
5623 if (num_bytes && head == &space_info->priority_tickets) {
5624 head = &space_info->tickets;
5629 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5630 struct btrfs_block_rsv *block_rsv,
5631 struct btrfs_block_rsv *dest, u64 num_bytes,
5632 u64 *qgroup_to_release_ret)
5634 struct btrfs_space_info *space_info = block_rsv->space_info;
5635 u64 qgroup_to_release = 0;
5638 spin_lock(&block_rsv->lock);
5639 if (num_bytes == (u64)-1) {
5640 num_bytes = block_rsv->size;
5641 qgroup_to_release = block_rsv->qgroup_rsv_size;
5643 block_rsv->size -= num_bytes;
5644 if (block_rsv->reserved >= block_rsv->size) {
5645 num_bytes = block_rsv->reserved - block_rsv->size;
5646 block_rsv->reserved = block_rsv->size;
5647 block_rsv->full = 1;
5651 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5652 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5653 block_rsv->qgroup_rsv_size;
5654 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5656 qgroup_to_release = 0;
5658 spin_unlock(&block_rsv->lock);
5661 if (num_bytes > 0) {
5663 spin_lock(&dest->lock);
5667 bytes_to_add = dest->size - dest->reserved;
5668 bytes_to_add = min(num_bytes, bytes_to_add);
5669 dest->reserved += bytes_to_add;
5670 if (dest->reserved >= dest->size)
5672 num_bytes -= bytes_to_add;
5674 spin_unlock(&dest->lock);
5677 space_info_add_old_bytes(fs_info, space_info,
5680 if (qgroup_to_release_ret)
5681 *qgroup_to_release_ret = qgroup_to_release;
5685 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5686 struct btrfs_block_rsv *dst, u64 num_bytes,
5691 ret = block_rsv_use_bytes(src, num_bytes);
5695 block_rsv_add_bytes(dst, num_bytes, update_size);
5699 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5701 memset(rsv, 0, sizeof(*rsv));
5702 spin_lock_init(&rsv->lock);
5706 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5707 struct btrfs_block_rsv *rsv,
5708 unsigned short type)
5710 btrfs_init_block_rsv(rsv, type);
5711 rsv->space_info = __find_space_info(fs_info,
5712 BTRFS_BLOCK_GROUP_METADATA);
5715 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5716 unsigned short type)
5718 struct btrfs_block_rsv *block_rsv;
5720 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5724 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5728 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5729 struct btrfs_block_rsv *rsv)
5733 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5737 int btrfs_block_rsv_add(struct btrfs_root *root,
5738 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5739 enum btrfs_reserve_flush_enum flush)
5746 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5748 block_rsv_add_bytes(block_rsv, num_bytes, true);
5753 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5761 spin_lock(&block_rsv->lock);
5762 num_bytes = div_factor(block_rsv->size, min_factor);
5763 if (block_rsv->reserved >= num_bytes)
5765 spin_unlock(&block_rsv->lock);
5770 int btrfs_block_rsv_refill(struct btrfs_root *root,
5771 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5772 enum btrfs_reserve_flush_enum flush)
5780 spin_lock(&block_rsv->lock);
5781 num_bytes = min_reserved;
5782 if (block_rsv->reserved >= num_bytes)
5785 num_bytes -= block_rsv->reserved;
5786 spin_unlock(&block_rsv->lock);
5791 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5793 block_rsv_add_bytes(block_rsv, num_bytes, false);
5800 static void calc_refill_bytes(struct btrfs_block_rsv *block_rsv,
5801 u64 *metadata_bytes, u64 *qgroup_bytes)
5803 *metadata_bytes = 0;
5806 spin_lock(&block_rsv->lock);
5807 if (block_rsv->reserved < block_rsv->size)
5808 *metadata_bytes = block_rsv->size - block_rsv->reserved;
5809 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5810 *qgroup_bytes = block_rsv->qgroup_rsv_size -
5811 block_rsv->qgroup_rsv_reserved;
5812 spin_unlock(&block_rsv->lock);
5816 * btrfs_inode_rsv_refill - refill the inode block rsv.
5817 * @inode - the inode we are refilling.
5818 * @flush - the flushing restriction.
5820 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5821 * block_rsv->size as the minimum size. We'll either refill the missing amount
5822 * or return if we already have enough space. This will also handle the reserve
5823 * tracepoint for the reserved amount.
5825 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5826 enum btrfs_reserve_flush_enum flush)
5828 struct btrfs_root *root = inode->root;
5829 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5830 u64 num_bytes, last = 0;
5831 u64 qgroup_num_bytes;
5834 calc_refill_bytes(block_rsv, &num_bytes, &qgroup_num_bytes);
5839 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes,
5843 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5845 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5848 * If we are fragmented we can end up with a lot of
5849 * outstanding extents which will make our size be much
5850 * larger than our reserved amount.
5852 * If the reservation happens here, it might be very
5853 * big though not needed in the end, if the delalloc
5856 * If this is the case try and do the reserve again.
5858 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5859 calc_refill_bytes(block_rsv, &num_bytes,
5864 } while (ret && last != num_bytes);
5867 block_rsv_add_bytes(block_rsv, num_bytes, false);
5868 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5869 btrfs_ino(inode), num_bytes, 1);
5871 /* Don't forget to increase qgroup_rsv_reserved */
5872 spin_lock(&block_rsv->lock);
5873 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5874 spin_unlock(&block_rsv->lock);
5879 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5880 struct btrfs_block_rsv *block_rsv,
5881 u64 num_bytes, u64 *qgroup_to_release)
5883 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5884 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5885 struct btrfs_block_rsv *target = delayed_rsv;
5887 if (target->full || target == block_rsv)
5888 target = global_rsv;
5890 if (block_rsv->space_info != target->space_info)
5893 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5897 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5898 struct btrfs_block_rsv *block_rsv,
5901 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5905 * btrfs_inode_rsv_release - release any excessive reservation.
5906 * @inode - the inode we need to release from.
5907 * @qgroup_free - free or convert qgroup meta.
5908 * Unlike normal operation, qgroup meta reservation needs to know if we are
5909 * freeing qgroup reservation or just converting it into per-trans. Normally
5910 * @qgroup_free is true for error handling, and false for normal release.
5912 * This is the same as btrfs_block_rsv_release, except that it handles the
5913 * tracepoint for the reservation.
5915 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5917 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5918 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5920 u64 qgroup_to_release = 0;
5923 * Since we statically set the block_rsv->size we just want to say we
5924 * are releasing 0 bytes, and then we'll just get the reservation over
5927 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5928 &qgroup_to_release);
5930 trace_btrfs_space_reservation(fs_info, "delalloc",
5931 btrfs_ino(inode), released, 0);
5933 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5935 btrfs_qgroup_convert_reserved_meta(inode->root,
5940 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5941 * @fs_info - the fs_info for our fs.
5942 * @nr - the number of items to drop.
5944 * This drops the delayed ref head's count from the delayed refs rsv and frees
5945 * any excess reservation we had.
5947 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5949 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5950 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5951 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5954 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5957 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5961 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5963 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5964 struct btrfs_space_info *sinfo = block_rsv->space_info;
5968 * The global block rsv is based on the size of the extent tree, the
5969 * checksum tree and the root tree. If the fs is empty we want to set
5970 * it to a minimal amount for safety.
5972 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5973 btrfs_root_used(&fs_info->csum_root->root_item) +
5974 btrfs_root_used(&fs_info->tree_root->root_item);
5975 num_bytes = max_t(u64, num_bytes, SZ_16M);
5977 spin_lock(&sinfo->lock);
5978 spin_lock(&block_rsv->lock);
5980 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5982 if (block_rsv->reserved < block_rsv->size) {
5983 num_bytes = btrfs_space_info_used(sinfo, true);
5984 if (sinfo->total_bytes > num_bytes) {
5985 num_bytes = sinfo->total_bytes - num_bytes;
5986 num_bytes = min(num_bytes,
5987 block_rsv->size - block_rsv->reserved);
5988 block_rsv->reserved += num_bytes;
5989 update_bytes_may_use(sinfo, num_bytes);
5990 trace_btrfs_space_reservation(fs_info, "space_info",
5991 sinfo->flags, num_bytes,
5994 } else if (block_rsv->reserved > block_rsv->size) {
5995 num_bytes = block_rsv->reserved - block_rsv->size;
5996 update_bytes_may_use(sinfo, -num_bytes);
5997 trace_btrfs_space_reservation(fs_info, "space_info",
5998 sinfo->flags, num_bytes, 0);
5999 block_rsv->reserved = block_rsv->size;
6002 if (block_rsv->reserved == block_rsv->size)
6003 block_rsv->full = 1;
6005 block_rsv->full = 0;
6007 spin_unlock(&block_rsv->lock);
6008 spin_unlock(&sinfo->lock);
6011 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
6013 struct btrfs_space_info *space_info;
6015 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
6016 fs_info->chunk_block_rsv.space_info = space_info;
6018 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
6019 fs_info->global_block_rsv.space_info = space_info;
6020 fs_info->trans_block_rsv.space_info = space_info;
6021 fs_info->empty_block_rsv.space_info = space_info;
6022 fs_info->delayed_block_rsv.space_info = space_info;
6023 fs_info->delayed_refs_rsv.space_info = space_info;
6025 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
6026 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
6027 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
6028 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
6029 if (fs_info->quota_root)
6030 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
6031 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
6033 update_global_block_rsv(fs_info);
6036 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
6038 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
6040 WARN_ON(fs_info->trans_block_rsv.size > 0);
6041 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
6042 WARN_ON(fs_info->chunk_block_rsv.size > 0);
6043 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
6044 WARN_ON(fs_info->delayed_block_rsv.size > 0);
6045 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
6046 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
6047 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
6051 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
6052 * @trans - the trans that may have generated delayed refs
6054 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
6055 * it'll calculate the additional size and add it to the delayed_refs_rsv.
6057 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
6059 struct btrfs_fs_info *fs_info = trans->fs_info;
6060 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
6063 if (!trans->delayed_ref_updates)
6066 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
6067 trans->delayed_ref_updates);
6068 spin_lock(&delayed_rsv->lock);
6069 delayed_rsv->size += num_bytes;
6070 delayed_rsv->full = 0;
6071 spin_unlock(&delayed_rsv->lock);
6072 trans->delayed_ref_updates = 0;
6076 * To be called after all the new block groups attached to the transaction
6077 * handle have been created (btrfs_create_pending_block_groups()).
6079 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
6081 struct btrfs_fs_info *fs_info = trans->fs_info;
6083 if (!trans->chunk_bytes_reserved)
6086 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
6088 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
6089 trans->chunk_bytes_reserved, NULL);
6090 trans->chunk_bytes_reserved = 0;
6094 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
6095 * root: the root of the parent directory
6096 * rsv: block reservation
6097 * items: the number of items that we need do reservation
6098 * use_global_rsv: allow fallback to the global block reservation
6100 * This function is used to reserve the space for snapshot/subvolume
6101 * creation and deletion. Those operations are different with the
6102 * common file/directory operations, they change two fs/file trees
6103 * and root tree, the number of items that the qgroup reserves is
6104 * different with the free space reservation. So we can not use
6105 * the space reservation mechanism in start_transaction().
6107 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6108 struct btrfs_block_rsv *rsv, int items,
6109 bool use_global_rsv)
6111 u64 qgroup_num_bytes = 0;
6114 struct btrfs_fs_info *fs_info = root->fs_info;
6115 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6117 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6118 /* One for parent inode, two for dir entries */
6119 qgroup_num_bytes = 3 * fs_info->nodesize;
6120 ret = btrfs_qgroup_reserve_meta_prealloc(root,
6121 qgroup_num_bytes, true);
6126 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6127 rsv->space_info = __find_space_info(fs_info,
6128 BTRFS_BLOCK_GROUP_METADATA);
6129 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6130 BTRFS_RESERVE_FLUSH_ALL);
6132 if (ret == -ENOSPC && use_global_rsv)
6133 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
6135 if (ret && qgroup_num_bytes)
6136 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
6141 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6142 struct btrfs_block_rsv *rsv)
6144 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6147 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6148 struct btrfs_inode *inode)
6150 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6151 u64 reserve_size = 0;
6152 u64 qgroup_rsv_size = 0;
6154 unsigned outstanding_extents;
6156 lockdep_assert_held(&inode->lock);
6157 outstanding_extents = inode->outstanding_extents;
6158 if (outstanding_extents)
6159 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6160 outstanding_extents + 1);
6161 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6163 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6166 * For qgroup rsv, the calculation is very simple:
6167 * account one nodesize for each outstanding extent
6169 * This is overestimating in most cases.
6171 qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize;
6173 spin_lock(&block_rsv->lock);
6174 block_rsv->size = reserve_size;
6175 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6176 spin_unlock(&block_rsv->lock);
6179 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6181 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6182 unsigned nr_extents;
6183 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6185 bool delalloc_lock = true;
6187 /* If we are a free space inode we need to not flush since we will be in
6188 * the middle of a transaction commit. We also don't need the delalloc
6189 * mutex since we won't race with anybody. We need this mostly to make
6190 * lockdep shut its filthy mouth.
6192 * If we have a transaction open (can happen if we call truncate_block
6193 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6195 if (btrfs_is_free_space_inode(inode)) {
6196 flush = BTRFS_RESERVE_NO_FLUSH;
6197 delalloc_lock = false;
6199 if (current->journal_info)
6200 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6202 if (btrfs_transaction_in_commit(fs_info))
6203 schedule_timeout(1);
6207 mutex_lock(&inode->delalloc_mutex);
6209 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6211 /* Add our new extents and calculate the new rsv size. */
6212 spin_lock(&inode->lock);
6213 nr_extents = count_max_extents(num_bytes);
6214 btrfs_mod_outstanding_extents(inode, nr_extents);
6215 inode->csum_bytes += num_bytes;
6216 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6217 spin_unlock(&inode->lock);
6219 ret = btrfs_inode_rsv_refill(inode, flush);
6224 mutex_unlock(&inode->delalloc_mutex);
6228 spin_lock(&inode->lock);
6229 nr_extents = count_max_extents(num_bytes);
6230 btrfs_mod_outstanding_extents(inode, -nr_extents);
6231 inode->csum_bytes -= num_bytes;
6232 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6233 spin_unlock(&inode->lock);
6235 btrfs_inode_rsv_release(inode, true);
6237 mutex_unlock(&inode->delalloc_mutex);
6242 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6243 * @inode: the inode to release the reservation for.
6244 * @num_bytes: the number of bytes we are releasing.
6245 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6247 * This will release the metadata reservation for an inode. This can be called
6248 * once we complete IO for a given set of bytes to release their metadata
6249 * reservations, or on error for the same reason.
6251 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6254 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6256 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6257 spin_lock(&inode->lock);
6258 inode->csum_bytes -= num_bytes;
6259 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6260 spin_unlock(&inode->lock);
6262 if (btrfs_is_testing(fs_info))
6265 btrfs_inode_rsv_release(inode, qgroup_free);
6269 * btrfs_delalloc_release_extents - release our outstanding_extents
6270 * @inode: the inode to balance the reservation for.
6271 * @num_bytes: the number of bytes we originally reserved with
6272 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6274 * When we reserve space we increase outstanding_extents for the extents we may
6275 * add. Once we've set the range as delalloc or created our ordered extents we
6276 * have outstanding_extents to track the real usage, so we use this to free our
6277 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6278 * with btrfs_delalloc_reserve_metadata.
6280 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6283 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6284 unsigned num_extents;
6286 spin_lock(&inode->lock);
6287 num_extents = count_max_extents(num_bytes);
6288 btrfs_mod_outstanding_extents(inode, -num_extents);
6289 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6290 spin_unlock(&inode->lock);
6292 if (btrfs_is_testing(fs_info))
6295 btrfs_inode_rsv_release(inode, qgroup_free);
6299 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6301 * @inode: inode we're writing to
6302 * @start: start range we are writing to
6303 * @len: how long the range we are writing to
6304 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6305 * current reservation.
6307 * This will do the following things
6309 * o reserve space in data space info for num bytes
6310 * and reserve precious corresponding qgroup space
6311 * (Done in check_data_free_space)
6313 * o reserve space for metadata space, based on the number of outstanding
6314 * extents and how much csums will be needed
6315 * also reserve metadata space in a per root over-reserve method.
6316 * o add to the inodes->delalloc_bytes
6317 * o add it to the fs_info's delalloc inodes list.
6318 * (Above 3 all done in delalloc_reserve_metadata)
6320 * Return 0 for success
6321 * Return <0 for error(-ENOSPC or -EQUOT)
6323 int btrfs_delalloc_reserve_space(struct inode *inode,
6324 struct extent_changeset **reserved, u64 start, u64 len)
6328 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6331 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6333 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6338 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6339 * @inode: inode we're releasing space for
6340 * @start: start position of the space already reserved
6341 * @len: the len of the space already reserved
6342 * @release_bytes: the len of the space we consumed or didn't use
6344 * This function will release the metadata space that was not used and will
6345 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6346 * list if there are no delalloc bytes left.
6347 * Also it will handle the qgroup reserved space.
6349 void btrfs_delalloc_release_space(struct inode *inode,
6350 struct extent_changeset *reserved,
6351 u64 start, u64 len, bool qgroup_free)
6353 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6354 btrfs_free_reserved_data_space(inode, reserved, start, len);
6357 static int update_block_group(struct btrfs_trans_handle *trans,
6358 struct btrfs_fs_info *info, u64 bytenr,
6359 u64 num_bytes, int alloc)
6361 struct btrfs_block_group_cache *cache = NULL;
6362 u64 total = num_bytes;
6368 /* block accounting for super block */
6369 spin_lock(&info->delalloc_root_lock);
6370 old_val = btrfs_super_bytes_used(info->super_copy);
6372 old_val += num_bytes;
6374 old_val -= num_bytes;
6375 btrfs_set_super_bytes_used(info->super_copy, old_val);
6376 spin_unlock(&info->delalloc_root_lock);
6379 cache = btrfs_lookup_block_group(info, bytenr);
6384 factor = btrfs_bg_type_to_factor(cache->flags);
6387 * If this block group has free space cache written out, we
6388 * need to make sure to load it if we are removing space. This
6389 * is because we need the unpinning stage to actually add the
6390 * space back to the block group, otherwise we will leak space.
6392 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6393 cache_block_group(cache, 1);
6395 byte_in_group = bytenr - cache->key.objectid;
6396 WARN_ON(byte_in_group > cache->key.offset);
6398 spin_lock(&cache->space_info->lock);
6399 spin_lock(&cache->lock);
6401 if (btrfs_test_opt(info, SPACE_CACHE) &&
6402 cache->disk_cache_state < BTRFS_DC_CLEAR)
6403 cache->disk_cache_state = BTRFS_DC_CLEAR;
6405 old_val = btrfs_block_group_used(&cache->item);
6406 num_bytes = min(total, cache->key.offset - byte_in_group);
6408 old_val += num_bytes;
6409 btrfs_set_block_group_used(&cache->item, old_val);
6410 cache->reserved -= num_bytes;
6411 cache->space_info->bytes_reserved -= num_bytes;
6412 cache->space_info->bytes_used += num_bytes;
6413 cache->space_info->disk_used += num_bytes * factor;
6414 spin_unlock(&cache->lock);
6415 spin_unlock(&cache->space_info->lock);
6417 old_val -= num_bytes;
6418 btrfs_set_block_group_used(&cache->item, old_val);
6419 cache->pinned += num_bytes;
6420 update_bytes_pinned(cache->space_info, num_bytes);
6421 cache->space_info->bytes_used -= num_bytes;
6422 cache->space_info->disk_used -= num_bytes * factor;
6423 spin_unlock(&cache->lock);
6424 spin_unlock(&cache->space_info->lock);
6426 trace_btrfs_space_reservation(info, "pinned",
6427 cache->space_info->flags,
6429 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6431 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6432 set_extent_dirty(info->pinned_extents,
6433 bytenr, bytenr + num_bytes - 1,
6434 GFP_NOFS | __GFP_NOFAIL);
6437 spin_lock(&trans->transaction->dirty_bgs_lock);
6438 if (list_empty(&cache->dirty_list)) {
6439 list_add_tail(&cache->dirty_list,
6440 &trans->transaction->dirty_bgs);
6441 trans->transaction->num_dirty_bgs++;
6442 trans->delayed_ref_updates++;
6443 btrfs_get_block_group(cache);
6445 spin_unlock(&trans->transaction->dirty_bgs_lock);
6448 * No longer have used bytes in this block group, queue it for
6449 * deletion. We do this after adding the block group to the
6450 * dirty list to avoid races between cleaner kthread and space
6453 if (!alloc && old_val == 0)
6454 btrfs_mark_bg_unused(cache);
6456 btrfs_put_block_group(cache);
6458 bytenr += num_bytes;
6461 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6462 btrfs_update_delayed_refs_rsv(trans);
6466 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6468 struct btrfs_block_group_cache *cache;
6471 spin_lock(&fs_info->block_group_cache_lock);
6472 bytenr = fs_info->first_logical_byte;
6473 spin_unlock(&fs_info->block_group_cache_lock);
6475 if (bytenr < (u64)-1)
6478 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6482 bytenr = cache->key.objectid;
6483 btrfs_put_block_group(cache);
6488 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6489 struct btrfs_block_group_cache *cache,
6490 u64 bytenr, u64 num_bytes, int reserved)
6492 spin_lock(&cache->space_info->lock);
6493 spin_lock(&cache->lock);
6494 cache->pinned += num_bytes;
6495 update_bytes_pinned(cache->space_info, num_bytes);
6497 cache->reserved -= num_bytes;
6498 cache->space_info->bytes_reserved -= num_bytes;
6500 spin_unlock(&cache->lock);
6501 spin_unlock(&cache->space_info->lock);
6503 trace_btrfs_space_reservation(fs_info, "pinned",
6504 cache->space_info->flags, num_bytes, 1);
6505 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6506 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6507 set_extent_dirty(fs_info->pinned_extents, bytenr,
6508 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6513 * this function must be called within transaction
6515 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6516 u64 bytenr, u64 num_bytes, int reserved)
6518 struct btrfs_block_group_cache *cache;
6520 cache = btrfs_lookup_block_group(fs_info, bytenr);
6521 BUG_ON(!cache); /* Logic error */
6523 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6525 btrfs_put_block_group(cache);
6530 * this function must be called within transaction
6532 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6533 u64 bytenr, u64 num_bytes)
6535 struct btrfs_block_group_cache *cache;
6538 cache = btrfs_lookup_block_group(fs_info, bytenr);
6543 * pull in the free space cache (if any) so that our pin
6544 * removes the free space from the cache. We have load_only set
6545 * to one because the slow code to read in the free extents does check
6546 * the pinned extents.
6548 cache_block_group(cache, 1);
6550 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6552 /* remove us from the free space cache (if we're there at all) */
6553 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6554 btrfs_put_block_group(cache);
6558 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6559 u64 start, u64 num_bytes)
6562 struct btrfs_block_group_cache *block_group;
6563 struct btrfs_caching_control *caching_ctl;
6565 block_group = btrfs_lookup_block_group(fs_info, start);
6569 cache_block_group(block_group, 0);
6570 caching_ctl = get_caching_control(block_group);
6574 BUG_ON(!block_group_cache_done(block_group));
6575 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6577 mutex_lock(&caching_ctl->mutex);
6579 if (start >= caching_ctl->progress) {
6580 ret = add_excluded_extent(fs_info, start, num_bytes);
6581 } else if (start + num_bytes <= caching_ctl->progress) {
6582 ret = btrfs_remove_free_space(block_group,
6585 num_bytes = caching_ctl->progress - start;
6586 ret = btrfs_remove_free_space(block_group,
6591 num_bytes = (start + num_bytes) -
6592 caching_ctl->progress;
6593 start = caching_ctl->progress;
6594 ret = add_excluded_extent(fs_info, start, num_bytes);
6597 mutex_unlock(&caching_ctl->mutex);
6598 put_caching_control(caching_ctl);
6600 btrfs_put_block_group(block_group);
6604 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
6606 struct btrfs_fs_info *fs_info = eb->fs_info;
6607 struct btrfs_file_extent_item *item;
6608 struct btrfs_key key;
6613 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6616 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6617 btrfs_item_key_to_cpu(eb, &key, i);
6618 if (key.type != BTRFS_EXTENT_DATA_KEY)
6620 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6621 found_type = btrfs_file_extent_type(eb, item);
6622 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6624 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6626 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6627 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6628 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6637 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6639 atomic_inc(&bg->reservations);
6642 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6645 struct btrfs_block_group_cache *bg;
6647 bg = btrfs_lookup_block_group(fs_info, start);
6649 if (atomic_dec_and_test(&bg->reservations))
6650 wake_up_var(&bg->reservations);
6651 btrfs_put_block_group(bg);
6654 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6656 struct btrfs_space_info *space_info = bg->space_info;
6660 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6664 * Our block group is read only but before we set it to read only,
6665 * some task might have had allocated an extent from it already, but it
6666 * has not yet created a respective ordered extent (and added it to a
6667 * root's list of ordered extents).
6668 * Therefore wait for any task currently allocating extents, since the
6669 * block group's reservations counter is incremented while a read lock
6670 * on the groups' semaphore is held and decremented after releasing
6671 * the read access on that semaphore and creating the ordered extent.
6673 down_write(&space_info->groups_sem);
6674 up_write(&space_info->groups_sem);
6676 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6680 * btrfs_add_reserved_bytes - update the block_group and space info counters
6681 * @cache: The cache we are manipulating
6682 * @ram_bytes: The number of bytes of file content, and will be same to
6683 * @num_bytes except for the compress path.
6684 * @num_bytes: The number of bytes in question
6685 * @delalloc: The blocks are allocated for the delalloc write
6687 * This is called by the allocator when it reserves space. If this is a
6688 * reservation and the block group has become read only we cannot make the
6689 * reservation and return -EAGAIN, otherwise this function always succeeds.
6691 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6692 u64 ram_bytes, u64 num_bytes, int delalloc)
6694 struct btrfs_space_info *space_info = cache->space_info;
6697 spin_lock(&space_info->lock);
6698 spin_lock(&cache->lock);
6702 cache->reserved += num_bytes;
6703 space_info->bytes_reserved += num_bytes;
6704 update_bytes_may_use(space_info, -ram_bytes);
6706 cache->delalloc_bytes += num_bytes;
6708 spin_unlock(&cache->lock);
6709 spin_unlock(&space_info->lock);
6714 * btrfs_free_reserved_bytes - update the block_group and space info counters
6715 * @cache: The cache we are manipulating
6716 * @num_bytes: The number of bytes in question
6717 * @delalloc: The blocks are allocated for the delalloc write
6719 * This is called by somebody who is freeing space that was never actually used
6720 * on disk. For example if you reserve some space for a new leaf in transaction
6721 * A and before transaction A commits you free that leaf, you call this with
6722 * reserve set to 0 in order to clear the reservation.
6725 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6726 u64 num_bytes, int delalloc)
6728 struct btrfs_space_info *space_info = cache->space_info;
6730 spin_lock(&space_info->lock);
6731 spin_lock(&cache->lock);
6733 space_info->bytes_readonly += num_bytes;
6734 cache->reserved -= num_bytes;
6735 space_info->bytes_reserved -= num_bytes;
6736 space_info->max_extent_size = 0;
6739 cache->delalloc_bytes -= num_bytes;
6740 spin_unlock(&cache->lock);
6741 spin_unlock(&space_info->lock);
6743 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6745 struct btrfs_caching_control *next;
6746 struct btrfs_caching_control *caching_ctl;
6747 struct btrfs_block_group_cache *cache;
6749 down_write(&fs_info->commit_root_sem);
6751 list_for_each_entry_safe(caching_ctl, next,
6752 &fs_info->caching_block_groups, list) {
6753 cache = caching_ctl->block_group;
6754 if (block_group_cache_done(cache)) {
6755 cache->last_byte_to_unpin = (u64)-1;
6756 list_del_init(&caching_ctl->list);
6757 put_caching_control(caching_ctl);
6759 cache->last_byte_to_unpin = caching_ctl->progress;
6763 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6764 fs_info->pinned_extents = &fs_info->freed_extents[1];
6766 fs_info->pinned_extents = &fs_info->freed_extents[0];
6768 up_write(&fs_info->commit_root_sem);
6770 update_global_block_rsv(fs_info);
6774 * Returns the free cluster for the given space info and sets empty_cluster to
6775 * what it should be based on the mount options.
6777 static struct btrfs_free_cluster *
6778 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6779 struct btrfs_space_info *space_info, u64 *empty_cluster)
6781 struct btrfs_free_cluster *ret = NULL;
6784 if (btrfs_mixed_space_info(space_info))
6787 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6788 ret = &fs_info->meta_alloc_cluster;
6789 if (btrfs_test_opt(fs_info, SSD))
6790 *empty_cluster = SZ_2M;
6792 *empty_cluster = SZ_64K;
6793 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6794 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6795 *empty_cluster = SZ_2M;
6796 ret = &fs_info->data_alloc_cluster;
6802 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6804 const bool return_free_space)
6806 struct btrfs_block_group_cache *cache = NULL;
6807 struct btrfs_space_info *space_info;
6808 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6809 struct btrfs_free_cluster *cluster = NULL;
6811 u64 total_unpinned = 0;
6812 u64 empty_cluster = 0;
6815 while (start <= end) {
6818 start >= cache->key.objectid + cache->key.offset) {
6820 btrfs_put_block_group(cache);
6822 cache = btrfs_lookup_block_group(fs_info, start);
6823 BUG_ON(!cache); /* Logic error */
6825 cluster = fetch_cluster_info(fs_info,
6828 empty_cluster <<= 1;
6831 len = cache->key.objectid + cache->key.offset - start;
6832 len = min(len, end + 1 - start);
6834 if (start < cache->last_byte_to_unpin) {
6835 len = min(len, cache->last_byte_to_unpin - start);
6836 if (return_free_space)
6837 btrfs_add_free_space(cache, start, len);
6841 total_unpinned += len;
6842 space_info = cache->space_info;
6845 * If this space cluster has been marked as fragmented and we've
6846 * unpinned enough in this block group to potentially allow a
6847 * cluster to be created inside of it go ahead and clear the
6850 if (cluster && cluster->fragmented &&
6851 total_unpinned > empty_cluster) {
6852 spin_lock(&cluster->lock);
6853 cluster->fragmented = 0;
6854 spin_unlock(&cluster->lock);
6857 spin_lock(&space_info->lock);
6858 spin_lock(&cache->lock);
6859 cache->pinned -= len;
6860 update_bytes_pinned(space_info, -len);
6862 trace_btrfs_space_reservation(fs_info, "pinned",
6863 space_info->flags, len, 0);
6864 space_info->max_extent_size = 0;
6865 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6866 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6868 space_info->bytes_readonly += len;
6871 spin_unlock(&cache->lock);
6872 if (!readonly && return_free_space &&
6873 global_rsv->space_info == space_info) {
6876 spin_lock(&global_rsv->lock);
6877 if (!global_rsv->full) {
6878 to_add = min(len, global_rsv->size -
6879 global_rsv->reserved);
6880 global_rsv->reserved += to_add;
6881 update_bytes_may_use(space_info, to_add);
6882 if (global_rsv->reserved >= global_rsv->size)
6883 global_rsv->full = 1;
6884 trace_btrfs_space_reservation(fs_info,
6890 spin_unlock(&global_rsv->lock);
6891 /* Add to any tickets we may have */
6893 space_info_add_new_bytes(fs_info, space_info,
6896 spin_unlock(&space_info->lock);
6900 btrfs_put_block_group(cache);
6904 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6906 struct btrfs_fs_info *fs_info = trans->fs_info;
6907 struct btrfs_block_group_cache *block_group, *tmp;
6908 struct list_head *deleted_bgs;
6909 struct extent_io_tree *unpin;
6914 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6915 unpin = &fs_info->freed_extents[1];
6917 unpin = &fs_info->freed_extents[0];
6919 while (!trans->aborted) {
6920 struct extent_state *cached_state = NULL;
6922 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6923 ret = find_first_extent_bit(unpin, 0, &start, &end,
6924 EXTENT_DIRTY, &cached_state);
6926 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6930 if (btrfs_test_opt(fs_info, DISCARD))
6931 ret = btrfs_discard_extent(fs_info, start,
6932 end + 1 - start, NULL);
6934 clear_extent_dirty(unpin, start, end, &cached_state);
6935 unpin_extent_range(fs_info, start, end, true);
6936 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6937 free_extent_state(cached_state);
6942 * Transaction is finished. We don't need the lock anymore. We
6943 * do need to clean up the block groups in case of a transaction
6946 deleted_bgs = &trans->transaction->deleted_bgs;
6947 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6951 if (!trans->aborted)
6952 ret = btrfs_discard_extent(fs_info,
6953 block_group->key.objectid,
6954 block_group->key.offset,
6957 list_del_init(&block_group->bg_list);
6958 btrfs_put_block_group_trimming(block_group);
6959 btrfs_put_block_group(block_group);
6962 const char *errstr = btrfs_decode_error(ret);
6964 "discard failed while removing blockgroup: errno=%d %s",
6972 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6973 struct btrfs_delayed_ref_node *node, u64 parent,
6974 u64 root_objectid, u64 owner_objectid,
6975 u64 owner_offset, int refs_to_drop,
6976 struct btrfs_delayed_extent_op *extent_op)
6978 struct btrfs_fs_info *info = trans->fs_info;
6979 struct btrfs_key key;
6980 struct btrfs_path *path;
6981 struct btrfs_root *extent_root = info->extent_root;
6982 struct extent_buffer *leaf;
6983 struct btrfs_extent_item *ei;
6984 struct btrfs_extent_inline_ref *iref;
6987 int extent_slot = 0;
6988 int found_extent = 0;
6992 u64 bytenr = node->bytenr;
6993 u64 num_bytes = node->num_bytes;
6995 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6997 path = btrfs_alloc_path();
7001 path->reada = READA_FORWARD;
7002 path->leave_spinning = 1;
7004 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
7005 BUG_ON(!is_data && refs_to_drop != 1);
7008 skinny_metadata = false;
7010 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
7011 parent, root_objectid, owner_objectid,
7014 extent_slot = path->slots[0];
7015 while (extent_slot >= 0) {
7016 btrfs_item_key_to_cpu(path->nodes[0], &key,
7018 if (key.objectid != bytenr)
7020 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
7021 key.offset == num_bytes) {
7025 if (key.type == BTRFS_METADATA_ITEM_KEY &&
7026 key.offset == owner_objectid) {
7030 if (path->slots[0] - extent_slot > 5)
7035 if (!found_extent) {
7037 ret = remove_extent_backref(trans, path, NULL,
7039 is_data, &last_ref);
7041 btrfs_abort_transaction(trans, ret);
7044 btrfs_release_path(path);
7045 path->leave_spinning = 1;
7047 key.objectid = bytenr;
7048 key.type = BTRFS_EXTENT_ITEM_KEY;
7049 key.offset = num_bytes;
7051 if (!is_data && skinny_metadata) {
7052 key.type = BTRFS_METADATA_ITEM_KEY;
7053 key.offset = owner_objectid;
7056 ret = btrfs_search_slot(trans, extent_root,
7058 if (ret > 0 && skinny_metadata && path->slots[0]) {
7060 * Couldn't find our skinny metadata item,
7061 * see if we have ye olde extent item.
7064 btrfs_item_key_to_cpu(path->nodes[0], &key,
7066 if (key.objectid == bytenr &&
7067 key.type == BTRFS_EXTENT_ITEM_KEY &&
7068 key.offset == num_bytes)
7072 if (ret > 0 && skinny_metadata) {
7073 skinny_metadata = false;
7074 key.objectid = bytenr;
7075 key.type = BTRFS_EXTENT_ITEM_KEY;
7076 key.offset = num_bytes;
7077 btrfs_release_path(path);
7078 ret = btrfs_search_slot(trans, extent_root,
7084 "umm, got %d back from search, was looking for %llu",
7087 btrfs_print_leaf(path->nodes[0]);
7090 btrfs_abort_transaction(trans, ret);
7093 extent_slot = path->slots[0];
7095 } else if (WARN_ON(ret == -ENOENT)) {
7096 btrfs_print_leaf(path->nodes[0]);
7098 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7099 bytenr, parent, root_objectid, owner_objectid,
7101 btrfs_abort_transaction(trans, ret);
7104 btrfs_abort_transaction(trans, ret);
7108 leaf = path->nodes[0];
7109 item_size = btrfs_item_size_nr(leaf, extent_slot);
7110 if (unlikely(item_size < sizeof(*ei))) {
7112 btrfs_print_v0_err(info);
7113 btrfs_abort_transaction(trans, ret);
7116 ei = btrfs_item_ptr(leaf, extent_slot,
7117 struct btrfs_extent_item);
7118 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7119 key.type == BTRFS_EXTENT_ITEM_KEY) {
7120 struct btrfs_tree_block_info *bi;
7121 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7122 bi = (struct btrfs_tree_block_info *)(ei + 1);
7123 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7126 refs = btrfs_extent_refs(leaf, ei);
7127 if (refs < refs_to_drop) {
7129 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7130 refs_to_drop, refs, bytenr);
7132 btrfs_abort_transaction(trans, ret);
7135 refs -= refs_to_drop;
7139 __run_delayed_extent_op(extent_op, leaf, ei);
7141 * In the case of inline back ref, reference count will
7142 * be updated by remove_extent_backref
7145 BUG_ON(!found_extent);
7147 btrfs_set_extent_refs(leaf, ei, refs);
7148 btrfs_mark_buffer_dirty(leaf);
7151 ret = remove_extent_backref(trans, path, iref,
7152 refs_to_drop, is_data,
7155 btrfs_abort_transaction(trans, ret);
7161 BUG_ON(is_data && refs_to_drop !=
7162 extent_data_ref_count(path, iref));
7164 BUG_ON(path->slots[0] != extent_slot);
7166 BUG_ON(path->slots[0] != extent_slot + 1);
7167 path->slots[0] = extent_slot;
7173 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7176 btrfs_abort_transaction(trans, ret);
7179 btrfs_release_path(path);
7182 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7184 btrfs_abort_transaction(trans, ret);
7189 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7191 btrfs_abort_transaction(trans, ret);
7195 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7197 btrfs_abort_transaction(trans, ret);
7201 btrfs_release_path(path);
7204 btrfs_free_path(path);
7209 * when we free an block, it is possible (and likely) that we free the last
7210 * delayed ref for that extent as well. This searches the delayed ref tree for
7211 * a given extent, and if there are no other delayed refs to be processed, it
7212 * removes it from the tree.
7214 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7217 struct btrfs_delayed_ref_head *head;
7218 struct btrfs_delayed_ref_root *delayed_refs;
7221 delayed_refs = &trans->transaction->delayed_refs;
7222 spin_lock(&delayed_refs->lock);
7223 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7225 goto out_delayed_unlock;
7227 spin_lock(&head->lock);
7228 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7231 if (cleanup_extent_op(head) != NULL)
7235 * waiting for the lock here would deadlock. If someone else has it
7236 * locked they are already in the process of dropping it anyway
7238 if (!mutex_trylock(&head->mutex))
7241 btrfs_delete_ref_head(delayed_refs, head);
7242 head->processing = 0;
7244 spin_unlock(&head->lock);
7245 spin_unlock(&delayed_refs->lock);
7247 BUG_ON(head->extent_op);
7248 if (head->must_insert_reserved)
7251 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
7252 mutex_unlock(&head->mutex);
7253 btrfs_put_delayed_ref_head(head);
7256 spin_unlock(&head->lock);
7259 spin_unlock(&delayed_refs->lock);
7263 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7264 struct btrfs_root *root,
7265 struct extent_buffer *buf,
7266 u64 parent, int last_ref)
7268 struct btrfs_fs_info *fs_info = root->fs_info;
7272 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7273 int old_ref_mod, new_ref_mod;
7275 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7276 root->root_key.objectid,
7277 btrfs_header_level(buf), 0,
7278 BTRFS_DROP_DELAYED_REF);
7279 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7281 root->root_key.objectid,
7282 btrfs_header_level(buf),
7283 BTRFS_DROP_DELAYED_REF, NULL,
7284 &old_ref_mod, &new_ref_mod);
7285 BUG_ON(ret); /* -ENOMEM */
7286 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7289 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7290 struct btrfs_block_group_cache *cache;
7292 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7293 ret = check_ref_cleanup(trans, buf->start);
7299 cache = btrfs_lookup_block_group(fs_info, buf->start);
7301 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7302 pin_down_extent(fs_info, cache, buf->start,
7304 btrfs_put_block_group(cache);
7308 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7310 btrfs_add_free_space(cache, buf->start, buf->len);
7311 btrfs_free_reserved_bytes(cache, buf->len, 0);
7312 btrfs_put_block_group(cache);
7313 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7317 add_pinned_bytes(fs_info, buf->len, true,
7318 root->root_key.objectid);
7322 * Deleting the buffer, clear the corrupt flag since it doesn't
7325 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7329 /* Can return -ENOMEM */
7330 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7331 struct btrfs_root *root,
7332 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7333 u64 owner, u64 offset)
7335 struct btrfs_fs_info *fs_info = root->fs_info;
7336 int old_ref_mod, new_ref_mod;
7339 if (btrfs_is_testing(fs_info))
7342 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7343 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7344 root_objectid, owner, offset,
7345 BTRFS_DROP_DELAYED_REF);
7348 * tree log blocks never actually go into the extent allocation
7349 * tree, just update pinning info and exit early.
7351 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7352 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7353 /* unlocks the pinned mutex */
7354 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7355 old_ref_mod = new_ref_mod = 0;
7357 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7358 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7360 root_objectid, (int)owner,
7361 BTRFS_DROP_DELAYED_REF, NULL,
7362 &old_ref_mod, &new_ref_mod);
7364 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7366 root_objectid, owner, offset,
7367 0, BTRFS_DROP_DELAYED_REF,
7368 &old_ref_mod, &new_ref_mod);
7371 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7372 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7374 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7381 * when we wait for progress in the block group caching, its because
7382 * our allocation attempt failed at least once. So, we must sleep
7383 * and let some progress happen before we try again.
7385 * This function will sleep at least once waiting for new free space to
7386 * show up, and then it will check the block group free space numbers
7387 * for our min num_bytes. Another option is to have it go ahead
7388 * and look in the rbtree for a free extent of a given size, but this
7391 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7392 * any of the information in this block group.
7394 static noinline void
7395 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7398 struct btrfs_caching_control *caching_ctl;
7400 caching_ctl = get_caching_control(cache);
7404 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7405 (cache->free_space_ctl->free_space >= num_bytes));
7407 put_caching_control(caching_ctl);
7411 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7413 struct btrfs_caching_control *caching_ctl;
7416 caching_ctl = get_caching_control(cache);
7418 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7420 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7421 if (cache->cached == BTRFS_CACHE_ERROR)
7423 put_caching_control(caching_ctl);
7427 enum btrfs_loop_type {
7428 LOOP_CACHING_NOWAIT = 0,
7429 LOOP_CACHING_WAIT = 1,
7430 LOOP_ALLOC_CHUNK = 2,
7431 LOOP_NO_EMPTY_SIZE = 3,
7435 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7439 down_read(&cache->data_rwsem);
7443 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7446 btrfs_get_block_group(cache);
7448 down_read(&cache->data_rwsem);
7451 static struct btrfs_block_group_cache *
7452 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7453 struct btrfs_free_cluster *cluster,
7456 struct btrfs_block_group_cache *used_bg = NULL;
7458 spin_lock(&cluster->refill_lock);
7460 used_bg = cluster->block_group;
7464 if (used_bg == block_group)
7467 btrfs_get_block_group(used_bg);
7472 if (down_read_trylock(&used_bg->data_rwsem))
7475 spin_unlock(&cluster->refill_lock);
7477 /* We should only have one-level nested. */
7478 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7480 spin_lock(&cluster->refill_lock);
7481 if (used_bg == cluster->block_group)
7484 up_read(&used_bg->data_rwsem);
7485 btrfs_put_block_group(used_bg);
7490 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7494 up_read(&cache->data_rwsem);
7495 btrfs_put_block_group(cache);
7499 * Structure used internally for find_free_extent() function. Wraps needed
7502 struct find_free_extent_ctl {
7503 /* Basic allocation info */
7510 /* Where to start the search inside the bg */
7513 /* For clustered allocation */
7516 bool have_caching_bg;
7517 bool orig_have_caching_bg;
7519 /* RAID index, converted from flags */
7523 * Current loop number, check find_free_extent_update_loop() for details
7528 * Whether we're refilling a cluster, if true we need to re-search
7529 * current block group but don't try to refill the cluster again.
7531 bool retry_clustered;
7534 * Whether we're updating free space cache, if true we need to re-search
7535 * current block group but don't try updating free space cache again.
7537 bool retry_unclustered;
7539 /* If current block group is cached */
7542 /* Max contiguous hole found */
7543 u64 max_extent_size;
7545 /* Total free space from free space cache, not always contiguous */
7546 u64 total_free_space;
7554 * Helper function for find_free_extent().
7556 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7557 * Return -EAGAIN to inform caller that we need to re-search this block group
7558 * Return >0 to inform caller that we find nothing
7559 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7561 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7562 struct btrfs_free_cluster *last_ptr,
7563 struct find_free_extent_ctl *ffe_ctl,
7564 struct btrfs_block_group_cache **cluster_bg_ret)
7566 struct btrfs_fs_info *fs_info = bg->fs_info;
7567 struct btrfs_block_group_cache *cluster_bg;
7568 u64 aligned_cluster;
7572 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7574 goto refill_cluster;
7575 if (cluster_bg != bg && (cluster_bg->ro ||
7576 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7577 goto release_cluster;
7579 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7580 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7581 &ffe_ctl->max_extent_size);
7583 /* We have a block, we're done */
7584 spin_unlock(&last_ptr->refill_lock);
7585 trace_btrfs_reserve_extent_cluster(cluster_bg,
7586 ffe_ctl->search_start, ffe_ctl->num_bytes);
7587 *cluster_bg_ret = cluster_bg;
7588 ffe_ctl->found_offset = offset;
7591 WARN_ON(last_ptr->block_group != cluster_bg);
7595 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7596 * lets just skip it and let the allocator find whatever block it can
7597 * find. If we reach this point, we will have tried the cluster
7598 * allocator plenty of times and not have found anything, so we are
7599 * likely way too fragmented for the clustering stuff to find anything.
7601 * However, if the cluster is taken from the current block group,
7602 * release the cluster first, so that we stand a better chance of
7603 * succeeding in the unclustered allocation.
7605 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7606 spin_unlock(&last_ptr->refill_lock);
7607 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7611 /* This cluster didn't work out, free it and start over */
7612 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7614 if (cluster_bg != bg)
7615 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7618 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7619 spin_unlock(&last_ptr->refill_lock);
7623 aligned_cluster = max_t(u64,
7624 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7625 bg->full_stripe_len);
7626 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7627 ffe_ctl->search_start, ffe_ctl->num_bytes,
7630 /* Now pull our allocation out of this cluster */
7631 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7632 ffe_ctl->num_bytes, ffe_ctl->search_start,
7633 &ffe_ctl->max_extent_size);
7635 /* We found one, proceed */
7636 spin_unlock(&last_ptr->refill_lock);
7637 trace_btrfs_reserve_extent_cluster(bg,
7638 ffe_ctl->search_start,
7639 ffe_ctl->num_bytes);
7640 ffe_ctl->found_offset = offset;
7643 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7644 !ffe_ctl->retry_clustered) {
7645 spin_unlock(&last_ptr->refill_lock);
7647 ffe_ctl->retry_clustered = true;
7648 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7649 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7653 * At this point we either didn't find a cluster or we weren't able to
7654 * allocate a block from our cluster. Free the cluster we've been
7655 * trying to use, and go to the next block group.
7657 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7658 spin_unlock(&last_ptr->refill_lock);
7663 * Return >0 to inform caller that we find nothing
7664 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7665 * Return -EAGAIN to inform caller that we need to re-search this block group
7667 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7668 struct btrfs_free_cluster *last_ptr,
7669 struct find_free_extent_ctl *ffe_ctl)
7674 * We are doing an unclustered allocation, set the fragmented flag so
7675 * we don't bother trying to setup a cluster again until we get more
7678 if (unlikely(last_ptr)) {
7679 spin_lock(&last_ptr->lock);
7680 last_ptr->fragmented = 1;
7681 spin_unlock(&last_ptr->lock);
7683 if (ffe_ctl->cached) {
7684 struct btrfs_free_space_ctl *free_space_ctl;
7686 free_space_ctl = bg->free_space_ctl;
7687 spin_lock(&free_space_ctl->tree_lock);
7688 if (free_space_ctl->free_space <
7689 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7690 ffe_ctl->empty_size) {
7691 ffe_ctl->total_free_space = max_t(u64,
7692 ffe_ctl->total_free_space,
7693 free_space_ctl->free_space);
7694 spin_unlock(&free_space_ctl->tree_lock);
7697 spin_unlock(&free_space_ctl->tree_lock);
7700 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7701 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7702 &ffe_ctl->max_extent_size);
7705 * If we didn't find a chunk, and we haven't failed on this block group
7706 * before, and this block group is in the middle of caching and we are
7707 * ok with waiting, then go ahead and wait for progress to be made, and
7708 * set @retry_unclustered to true.
7710 * If @retry_unclustered is true then we've already waited on this
7711 * block group once and should move on to the next block group.
7713 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7714 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7715 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7716 ffe_ctl->empty_size);
7717 ffe_ctl->retry_unclustered = true;
7719 } else if (!offset) {
7722 ffe_ctl->found_offset = offset;
7727 * Return >0 means caller needs to re-search for free extent
7728 * Return 0 means we have the needed free extent.
7729 * Return <0 means we failed to locate any free extent.
7731 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7732 struct btrfs_free_cluster *last_ptr,
7733 struct btrfs_key *ins,
7734 struct find_free_extent_ctl *ffe_ctl,
7735 int full_search, bool use_cluster)
7737 struct btrfs_root *root = fs_info->extent_root;
7740 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7741 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7742 ffe_ctl->orig_have_caching_bg = true;
7744 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7745 ffe_ctl->have_caching_bg)
7748 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7751 if (ins->objectid) {
7752 if (!use_cluster && last_ptr) {
7753 spin_lock(&last_ptr->lock);
7754 last_ptr->window_start = ins->objectid;
7755 spin_unlock(&last_ptr->lock);
7761 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7762 * caching kthreads as we move along
7763 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7764 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7765 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7768 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7770 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7772 * We want to skip the LOOP_CACHING_WAIT step if we
7773 * don't have any uncached bgs and we've already done a
7774 * full search through.
7776 if (ffe_ctl->orig_have_caching_bg || !full_search)
7777 ffe_ctl->loop = LOOP_CACHING_WAIT;
7779 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7784 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7785 struct btrfs_trans_handle *trans;
7788 trans = current->journal_info;
7792 trans = btrfs_join_transaction(root);
7794 if (IS_ERR(trans)) {
7795 ret = PTR_ERR(trans);
7799 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7803 * If we can't allocate a new chunk we've already looped
7804 * through at least once, move on to the NO_EMPTY_SIZE
7808 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7810 /* Do not bail out on ENOSPC since we can do more. */
7811 if (ret < 0 && ret != -ENOSPC)
7812 btrfs_abort_transaction(trans, ret);
7816 btrfs_end_transaction(trans);
7821 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7823 * Don't loop again if we already have no empty_size and
7826 if (ffe_ctl->empty_size == 0 &&
7827 ffe_ctl->empty_cluster == 0)
7829 ffe_ctl->empty_size = 0;
7830 ffe_ctl->empty_cluster = 0;
7838 * walks the btree of allocated extents and find a hole of a given size.
7839 * The key ins is changed to record the hole:
7840 * ins->objectid == start position
7841 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7842 * ins->offset == the size of the hole.
7843 * Any available blocks before search_start are skipped.
7845 * If there is no suitable free space, we will record the max size of
7846 * the free space extent currently.
7848 * The overall logic and call chain:
7850 * find_free_extent()
7851 * |- Iterate through all block groups
7852 * | |- Get a valid block group
7853 * | |- Try to do clustered allocation in that block group
7854 * | |- Try to do unclustered allocation in that block group
7855 * | |- Check if the result is valid
7856 * | | |- If valid, then exit
7857 * | |- Jump to next block group
7859 * |- Push harder to find free extents
7860 * |- If not found, re-iterate all block groups
7862 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7863 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7864 u64 hint_byte, struct btrfs_key *ins,
7865 u64 flags, int delalloc)
7868 struct btrfs_free_cluster *last_ptr = NULL;
7869 struct btrfs_block_group_cache *block_group = NULL;
7870 struct find_free_extent_ctl ffe_ctl = {0};
7871 struct btrfs_space_info *space_info;
7872 bool use_cluster = true;
7873 bool full_search = false;
7875 WARN_ON(num_bytes < fs_info->sectorsize);
7877 ffe_ctl.ram_bytes = ram_bytes;
7878 ffe_ctl.num_bytes = num_bytes;
7879 ffe_ctl.empty_size = empty_size;
7880 ffe_ctl.flags = flags;
7881 ffe_ctl.search_start = 0;
7882 ffe_ctl.retry_clustered = false;
7883 ffe_ctl.retry_unclustered = false;
7884 ffe_ctl.delalloc = delalloc;
7885 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7886 ffe_ctl.have_caching_bg = false;
7887 ffe_ctl.orig_have_caching_bg = false;
7888 ffe_ctl.found_offset = 0;
7890 ins->type = BTRFS_EXTENT_ITEM_KEY;
7894 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7896 space_info = __find_space_info(fs_info, flags);
7898 btrfs_err(fs_info, "No space info for %llu", flags);
7903 * If our free space is heavily fragmented we may not be able to make
7904 * big contiguous allocations, so instead of doing the expensive search
7905 * for free space, simply return ENOSPC with our max_extent_size so we
7906 * can go ahead and search for a more manageable chunk.
7908 * If our max_extent_size is large enough for our allocation simply
7909 * disable clustering since we will likely not be able to find enough
7910 * space to create a cluster and induce latency trying.
7912 if (unlikely(space_info->max_extent_size)) {
7913 spin_lock(&space_info->lock);
7914 if (space_info->max_extent_size &&
7915 num_bytes > space_info->max_extent_size) {
7916 ins->offset = space_info->max_extent_size;
7917 spin_unlock(&space_info->lock);
7919 } else if (space_info->max_extent_size) {
7920 use_cluster = false;
7922 spin_unlock(&space_info->lock);
7925 last_ptr = fetch_cluster_info(fs_info, space_info,
7926 &ffe_ctl.empty_cluster);
7928 spin_lock(&last_ptr->lock);
7929 if (last_ptr->block_group)
7930 hint_byte = last_ptr->window_start;
7931 if (last_ptr->fragmented) {
7933 * We still set window_start so we can keep track of the
7934 * last place we found an allocation to try and save
7937 hint_byte = last_ptr->window_start;
7938 use_cluster = false;
7940 spin_unlock(&last_ptr->lock);
7943 ffe_ctl.search_start = max(ffe_ctl.search_start,
7944 first_logical_byte(fs_info, 0));
7945 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7946 if (ffe_ctl.search_start == hint_byte) {
7947 block_group = btrfs_lookup_block_group(fs_info,
7948 ffe_ctl.search_start);
7950 * we don't want to use the block group if it doesn't match our
7951 * allocation bits, or if its not cached.
7953 * However if we are re-searching with an ideal block group
7954 * picked out then we don't care that the block group is cached.
7956 if (block_group && block_group_bits(block_group, flags) &&
7957 block_group->cached != BTRFS_CACHE_NO) {
7958 down_read(&space_info->groups_sem);
7959 if (list_empty(&block_group->list) ||
7962 * someone is removing this block group,
7963 * we can't jump into the have_block_group
7964 * target because our list pointers are not
7967 btrfs_put_block_group(block_group);
7968 up_read(&space_info->groups_sem);
7970 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7971 block_group->flags);
7972 btrfs_lock_block_group(block_group, delalloc);
7973 goto have_block_group;
7975 } else if (block_group) {
7976 btrfs_put_block_group(block_group);
7980 ffe_ctl.have_caching_bg = false;
7981 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7984 down_read(&space_info->groups_sem);
7985 list_for_each_entry(block_group,
7986 &space_info->block_groups[ffe_ctl.index], list) {
7987 /* If the block group is read-only, we can skip it entirely. */
7988 if (unlikely(block_group->ro))
7991 btrfs_grab_block_group(block_group, delalloc);
7992 ffe_ctl.search_start = block_group->key.objectid;
7995 * this can happen if we end up cycling through all the
7996 * raid types, but we want to make sure we only allocate
7997 * for the proper type.
7999 if (!block_group_bits(block_group, flags)) {
8000 u64 extra = BTRFS_BLOCK_GROUP_DUP |
8001 BTRFS_BLOCK_GROUP_RAID1 |
8002 BTRFS_BLOCK_GROUP_RAID5 |
8003 BTRFS_BLOCK_GROUP_RAID6 |
8004 BTRFS_BLOCK_GROUP_RAID10;
8007 * if they asked for extra copies and this block group
8008 * doesn't provide them, bail. This does allow us to
8009 * fill raid0 from raid1.
8011 if ((flags & extra) && !(block_group->flags & extra))
8016 ffe_ctl.cached = block_group_cache_done(block_group);
8017 if (unlikely(!ffe_ctl.cached)) {
8018 ffe_ctl.have_caching_bg = true;
8019 ret = cache_block_group(block_group, 0);
8024 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
8028 * Ok we want to try and use the cluster allocator, so
8031 if (last_ptr && use_cluster) {
8032 struct btrfs_block_group_cache *cluster_bg = NULL;
8034 ret = find_free_extent_clustered(block_group, last_ptr,
8035 &ffe_ctl, &cluster_bg);
8038 if (cluster_bg && cluster_bg != block_group) {
8039 btrfs_release_block_group(block_group,
8041 block_group = cluster_bg;
8044 } else if (ret == -EAGAIN) {
8045 goto have_block_group;
8046 } else if (ret > 0) {
8049 /* ret == -ENOENT case falls through */
8052 ret = find_free_extent_unclustered(block_group, last_ptr,
8055 goto have_block_group;
8058 /* ret == 0 case falls through */
8060 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
8061 fs_info->stripesize);
8063 /* move on to the next group */
8064 if (ffe_ctl.search_start + num_bytes >
8065 block_group->key.objectid + block_group->key.offset) {
8066 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8071 if (ffe_ctl.found_offset < ffe_ctl.search_start)
8072 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8073 ffe_ctl.search_start - ffe_ctl.found_offset);
8075 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
8076 num_bytes, delalloc);
8077 if (ret == -EAGAIN) {
8078 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8082 btrfs_inc_block_group_reservations(block_group);
8084 /* we are all good, lets return */
8085 ins->objectid = ffe_ctl.search_start;
8086 ins->offset = num_bytes;
8088 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
8090 btrfs_release_block_group(block_group, delalloc);
8093 ffe_ctl.retry_clustered = false;
8094 ffe_ctl.retry_unclustered = false;
8095 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
8097 btrfs_release_block_group(block_group, delalloc);
8100 up_read(&space_info->groups_sem);
8102 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
8103 full_search, use_cluster);
8107 if (ret == -ENOSPC) {
8109 * Use ffe_ctl->total_free_space as fallback if we can't find
8110 * any contiguous hole.
8112 if (!ffe_ctl.max_extent_size)
8113 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
8114 spin_lock(&space_info->lock);
8115 space_info->max_extent_size = ffe_ctl.max_extent_size;
8116 spin_unlock(&space_info->lock);
8117 ins->offset = ffe_ctl.max_extent_size;
8122 #define DUMP_BLOCK_RSV(fs_info, rsv_name) \
8124 struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \
8125 spin_lock(&__rsv->lock); \
8126 btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \
8127 __rsv->size, __rsv->reserved); \
8128 spin_unlock(&__rsv->lock); \
8131 static void dump_space_info(struct btrfs_fs_info *fs_info,
8132 struct btrfs_space_info *info, u64 bytes,
8133 int dump_block_groups)
8135 struct btrfs_block_group_cache *cache;
8138 spin_lock(&info->lock);
8139 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8141 info->total_bytes - btrfs_space_info_used(info, true),
8142 info->full ? "" : "not ");
8144 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8145 info->total_bytes, info->bytes_used, info->bytes_pinned,
8146 info->bytes_reserved, info->bytes_may_use,
8147 info->bytes_readonly);
8148 spin_unlock(&info->lock);
8150 DUMP_BLOCK_RSV(fs_info, global_block_rsv);
8151 DUMP_BLOCK_RSV(fs_info, trans_block_rsv);
8152 DUMP_BLOCK_RSV(fs_info, chunk_block_rsv);
8153 DUMP_BLOCK_RSV(fs_info, delayed_block_rsv);
8154 DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv);
8156 if (!dump_block_groups)
8159 down_read(&info->groups_sem);
8161 list_for_each_entry(cache, &info->block_groups[index], list) {
8162 spin_lock(&cache->lock);
8164 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8165 cache->key.objectid, cache->key.offset,
8166 btrfs_block_group_used(&cache->item), cache->pinned,
8167 cache->reserved, cache->ro ? "[readonly]" : "");
8168 btrfs_dump_free_space(cache, bytes);
8169 spin_unlock(&cache->lock);
8171 if (++index < BTRFS_NR_RAID_TYPES)
8173 up_read(&info->groups_sem);
8177 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8178 * hole that is at least as big as @num_bytes.
8180 * @root - The root that will contain this extent
8182 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8183 * is used for accounting purposes. This value differs
8184 * from @num_bytes only in the case of compressed extents.
8186 * @num_bytes - Number of bytes to allocate on-disk.
8188 * @min_alloc_size - Indicates the minimum amount of space that the
8189 * allocator should try to satisfy. In some cases
8190 * @num_bytes may be larger than what is required and if
8191 * the filesystem is fragmented then allocation fails.
8192 * However, the presence of @min_alloc_size gives a
8193 * chance to try and satisfy the smaller allocation.
8195 * @empty_size - A hint that you plan on doing more COW. This is the
8196 * size in bytes the allocator should try to find free
8197 * next to the block it returns. This is just a hint and
8198 * may be ignored by the allocator.
8200 * @hint_byte - Hint to the allocator to start searching above the byte
8201 * address passed. It might be ignored.
8203 * @ins - This key is modified to record the found hole. It will
8204 * have the following values:
8205 * ins->objectid == start position
8206 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8207 * ins->offset == the size of the hole.
8209 * @is_data - Boolean flag indicating whether an extent is
8210 * allocated for data (true) or metadata (false)
8212 * @delalloc - Boolean flag indicating whether this allocation is for
8213 * delalloc or not. If 'true' data_rwsem of block groups
8214 * is going to be acquired.
8217 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8218 * case -ENOSPC is returned then @ins->offset will contain the size of the
8219 * largest available hole the allocator managed to find.
8221 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8222 u64 num_bytes, u64 min_alloc_size,
8223 u64 empty_size, u64 hint_byte,
8224 struct btrfs_key *ins, int is_data, int delalloc)
8226 struct btrfs_fs_info *fs_info = root->fs_info;
8227 bool final_tried = num_bytes == min_alloc_size;
8231 flags = get_alloc_profile_by_root(root, is_data);
8233 WARN_ON(num_bytes < fs_info->sectorsize);
8234 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8235 hint_byte, ins, flags, delalloc);
8236 if (!ret && !is_data) {
8237 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8238 } else if (ret == -ENOSPC) {
8239 if (!final_tried && ins->offset) {
8240 num_bytes = min(num_bytes >> 1, ins->offset);
8241 num_bytes = round_down(num_bytes,
8242 fs_info->sectorsize);
8243 num_bytes = max(num_bytes, min_alloc_size);
8244 ram_bytes = num_bytes;
8245 if (num_bytes == min_alloc_size)
8248 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8249 struct btrfs_space_info *sinfo;
8251 sinfo = __find_space_info(fs_info, flags);
8253 "allocation failed flags %llu, wanted %llu",
8256 dump_space_info(fs_info, sinfo, num_bytes, 1);
8263 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8265 int pin, int delalloc)
8267 struct btrfs_block_group_cache *cache;
8270 cache = btrfs_lookup_block_group(fs_info, start);
8272 btrfs_err(fs_info, "Unable to find block group for %llu",
8278 pin_down_extent(fs_info, cache, start, len, 1);
8280 if (btrfs_test_opt(fs_info, DISCARD))
8281 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8282 btrfs_add_free_space(cache, start, len);
8283 btrfs_free_reserved_bytes(cache, len, delalloc);
8284 trace_btrfs_reserved_extent_free(fs_info, start, len);
8287 btrfs_put_block_group(cache);
8291 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8292 u64 start, u64 len, int delalloc)
8294 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8297 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8300 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8303 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8304 u64 parent, u64 root_objectid,
8305 u64 flags, u64 owner, u64 offset,
8306 struct btrfs_key *ins, int ref_mod)
8308 struct btrfs_fs_info *fs_info = trans->fs_info;
8310 struct btrfs_extent_item *extent_item;
8311 struct btrfs_extent_inline_ref *iref;
8312 struct btrfs_path *path;
8313 struct extent_buffer *leaf;
8318 type = BTRFS_SHARED_DATA_REF_KEY;
8320 type = BTRFS_EXTENT_DATA_REF_KEY;
8322 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8324 path = btrfs_alloc_path();
8328 path->leave_spinning = 1;
8329 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8332 btrfs_free_path(path);
8336 leaf = path->nodes[0];
8337 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8338 struct btrfs_extent_item);
8339 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8340 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8341 btrfs_set_extent_flags(leaf, extent_item,
8342 flags | BTRFS_EXTENT_FLAG_DATA);
8344 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8345 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8347 struct btrfs_shared_data_ref *ref;
8348 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8349 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8350 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8352 struct btrfs_extent_data_ref *ref;
8353 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8354 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8355 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8356 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8357 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8360 btrfs_mark_buffer_dirty(path->nodes[0]);
8361 btrfs_free_path(path);
8363 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8367 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8368 if (ret) { /* -ENOENT, logic error */
8369 btrfs_err(fs_info, "update block group failed for %llu %llu",
8370 ins->objectid, ins->offset);
8373 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8377 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8378 struct btrfs_delayed_ref_node *node,
8379 struct btrfs_delayed_extent_op *extent_op)
8381 struct btrfs_fs_info *fs_info = trans->fs_info;
8383 struct btrfs_extent_item *extent_item;
8384 struct btrfs_key extent_key;
8385 struct btrfs_tree_block_info *block_info;
8386 struct btrfs_extent_inline_ref *iref;
8387 struct btrfs_path *path;
8388 struct extent_buffer *leaf;
8389 struct btrfs_delayed_tree_ref *ref;
8390 u32 size = sizeof(*extent_item) + sizeof(*iref);
8392 u64 flags = extent_op->flags_to_set;
8393 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8395 ref = btrfs_delayed_node_to_tree_ref(node);
8397 extent_key.objectid = node->bytenr;
8398 if (skinny_metadata) {
8399 extent_key.offset = ref->level;
8400 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8401 num_bytes = fs_info->nodesize;
8403 extent_key.offset = node->num_bytes;
8404 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8405 size += sizeof(*block_info);
8406 num_bytes = node->num_bytes;
8409 path = btrfs_alloc_path();
8413 path->leave_spinning = 1;
8414 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8417 btrfs_free_path(path);
8421 leaf = path->nodes[0];
8422 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8423 struct btrfs_extent_item);
8424 btrfs_set_extent_refs(leaf, extent_item, 1);
8425 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8426 btrfs_set_extent_flags(leaf, extent_item,
8427 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8429 if (skinny_metadata) {
8430 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8432 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8433 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8434 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8435 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8438 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8439 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8440 btrfs_set_extent_inline_ref_type(leaf, iref,
8441 BTRFS_SHARED_BLOCK_REF_KEY);
8442 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8444 btrfs_set_extent_inline_ref_type(leaf, iref,
8445 BTRFS_TREE_BLOCK_REF_KEY);
8446 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8449 btrfs_mark_buffer_dirty(leaf);
8450 btrfs_free_path(path);
8452 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8457 ret = update_block_group(trans, fs_info, extent_key.objectid,
8458 fs_info->nodesize, 1);
8459 if (ret) { /* -ENOENT, logic error */
8460 btrfs_err(fs_info, "update block group failed for %llu %llu",
8461 extent_key.objectid, extent_key.offset);
8465 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8470 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8471 struct btrfs_root *root, u64 owner,
8472 u64 offset, u64 ram_bytes,
8473 struct btrfs_key *ins)
8477 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8479 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8480 root->root_key.objectid, owner, offset,
8481 BTRFS_ADD_DELAYED_EXTENT);
8483 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8485 root->root_key.objectid, owner,
8487 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8492 * this is used by the tree logging recovery code. It records that
8493 * an extent has been allocated and makes sure to clear the free
8494 * space cache bits as well
8496 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8497 u64 root_objectid, u64 owner, u64 offset,
8498 struct btrfs_key *ins)
8500 struct btrfs_fs_info *fs_info = trans->fs_info;
8502 struct btrfs_block_group_cache *block_group;
8503 struct btrfs_space_info *space_info;
8506 * Mixed block groups will exclude before processing the log so we only
8507 * need to do the exclude dance if this fs isn't mixed.
8509 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8510 ret = __exclude_logged_extent(fs_info, ins->objectid,
8516 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8520 space_info = block_group->space_info;
8521 spin_lock(&space_info->lock);
8522 spin_lock(&block_group->lock);
8523 space_info->bytes_reserved += ins->offset;
8524 block_group->reserved += ins->offset;
8525 spin_unlock(&block_group->lock);
8526 spin_unlock(&space_info->lock);
8528 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8530 btrfs_put_block_group(block_group);
8534 static struct extent_buffer *
8535 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8536 u64 bytenr, int level, u64 owner)
8538 struct btrfs_fs_info *fs_info = root->fs_info;
8539 struct extent_buffer *buf;
8541 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8546 * Extra safety check in case the extent tree is corrupted and extent
8547 * allocator chooses to use a tree block which is already used and
8550 if (buf->lock_owner == current->pid) {
8551 btrfs_err_rl(fs_info,
8552 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8553 buf->start, btrfs_header_owner(buf), current->pid);
8554 free_extent_buffer(buf);
8555 return ERR_PTR(-EUCLEAN);
8558 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8559 btrfs_tree_lock(buf);
8560 clean_tree_block(fs_info, buf);
8561 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8563 btrfs_set_lock_blocking_write(buf);
8564 set_extent_buffer_uptodate(buf);
8566 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8567 btrfs_set_header_level(buf, level);
8568 btrfs_set_header_bytenr(buf, buf->start);
8569 btrfs_set_header_generation(buf, trans->transid);
8570 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8571 btrfs_set_header_owner(buf, owner);
8572 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8573 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8574 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8575 buf->log_index = root->log_transid % 2;
8577 * we allow two log transactions at a time, use different
8578 * EXTENT bit to differentiate dirty pages.
8580 if (buf->log_index == 0)
8581 set_extent_dirty(&root->dirty_log_pages, buf->start,
8582 buf->start + buf->len - 1, GFP_NOFS);
8584 set_extent_new(&root->dirty_log_pages, buf->start,
8585 buf->start + buf->len - 1);
8587 buf->log_index = -1;
8588 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8589 buf->start + buf->len - 1, GFP_NOFS);
8591 trans->dirty = true;
8592 /* this returns a buffer locked for blocking */
8596 static struct btrfs_block_rsv *
8597 use_block_rsv(struct btrfs_trans_handle *trans,
8598 struct btrfs_root *root, u32 blocksize)
8600 struct btrfs_fs_info *fs_info = root->fs_info;
8601 struct btrfs_block_rsv *block_rsv;
8602 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8604 bool global_updated = false;
8606 block_rsv = get_block_rsv(trans, root);
8608 if (unlikely(block_rsv->size == 0))
8611 ret = block_rsv_use_bytes(block_rsv, blocksize);
8615 if (block_rsv->failfast)
8616 return ERR_PTR(ret);
8618 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8619 global_updated = true;
8620 update_global_block_rsv(fs_info);
8625 * The global reserve still exists to save us from ourselves, so don't
8626 * warn_on if we are short on our delayed refs reserve.
8628 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8629 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8630 static DEFINE_RATELIMIT_STATE(_rs,
8631 DEFAULT_RATELIMIT_INTERVAL * 10,
8632 /*DEFAULT_RATELIMIT_BURST*/ 1);
8633 if (__ratelimit(&_rs))
8635 "BTRFS: block rsv returned %d\n", ret);
8638 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8639 BTRFS_RESERVE_NO_FLUSH);
8643 * If we couldn't reserve metadata bytes try and use some from
8644 * the global reserve if its space type is the same as the global
8647 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8648 block_rsv->space_info == global_rsv->space_info) {
8649 ret = block_rsv_use_bytes(global_rsv, blocksize);
8653 return ERR_PTR(ret);
8656 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8657 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8659 block_rsv_add_bytes(block_rsv, blocksize, false);
8660 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8664 * finds a free extent and does all the dirty work required for allocation
8665 * returns the tree buffer or an ERR_PTR on error.
8667 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8668 struct btrfs_root *root,
8669 u64 parent, u64 root_objectid,
8670 const struct btrfs_disk_key *key,
8671 int level, u64 hint,
8674 struct btrfs_fs_info *fs_info = root->fs_info;
8675 struct btrfs_key ins;
8676 struct btrfs_block_rsv *block_rsv;
8677 struct extent_buffer *buf;
8678 struct btrfs_delayed_extent_op *extent_op;
8681 u32 blocksize = fs_info->nodesize;
8682 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8684 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8685 if (btrfs_is_testing(fs_info)) {
8686 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8687 level, root_objectid);
8689 root->alloc_bytenr += blocksize;
8694 block_rsv = use_block_rsv(trans, root, blocksize);
8695 if (IS_ERR(block_rsv))
8696 return ERR_CAST(block_rsv);
8698 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8699 empty_size, hint, &ins, 0, 0);
8703 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8707 goto out_free_reserved;
8710 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8712 parent = ins.objectid;
8713 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8717 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8718 extent_op = btrfs_alloc_delayed_extent_op();
8724 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8726 memset(&extent_op->key, 0, sizeof(extent_op->key));
8727 extent_op->flags_to_set = flags;
8728 extent_op->update_key = skinny_metadata ? false : true;
8729 extent_op->update_flags = true;
8730 extent_op->is_data = false;
8731 extent_op->level = level;
8733 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8734 root_objectid, level, 0,
8735 BTRFS_ADD_DELAYED_EXTENT);
8736 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8738 root_objectid, level,
8739 BTRFS_ADD_DELAYED_EXTENT,
8740 extent_op, NULL, NULL);
8742 goto out_free_delayed;
8747 btrfs_free_delayed_extent_op(extent_op);
8749 free_extent_buffer(buf);
8751 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8753 unuse_block_rsv(fs_info, block_rsv, blocksize);
8754 return ERR_PTR(ret);
8757 struct walk_control {
8758 u64 refs[BTRFS_MAX_LEVEL];
8759 u64 flags[BTRFS_MAX_LEVEL];
8760 struct btrfs_key update_progress;
8761 struct btrfs_key drop_progress;
8773 #define DROP_REFERENCE 1
8774 #define UPDATE_BACKREF 2
8776 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8777 struct btrfs_root *root,
8778 struct walk_control *wc,
8779 struct btrfs_path *path)
8781 struct btrfs_fs_info *fs_info = root->fs_info;
8787 struct btrfs_key key;
8788 struct extent_buffer *eb;
8793 if (path->slots[wc->level] < wc->reada_slot) {
8794 wc->reada_count = wc->reada_count * 2 / 3;
8795 wc->reada_count = max(wc->reada_count, 2);
8797 wc->reada_count = wc->reada_count * 3 / 2;
8798 wc->reada_count = min_t(int, wc->reada_count,
8799 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8802 eb = path->nodes[wc->level];
8803 nritems = btrfs_header_nritems(eb);
8805 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8806 if (nread >= wc->reada_count)
8810 bytenr = btrfs_node_blockptr(eb, slot);
8811 generation = btrfs_node_ptr_generation(eb, slot);
8813 if (slot == path->slots[wc->level])
8816 if (wc->stage == UPDATE_BACKREF &&
8817 generation <= root->root_key.offset)
8820 /* We don't lock the tree block, it's OK to be racy here */
8821 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8822 wc->level - 1, 1, &refs,
8824 /* We don't care about errors in readahead. */
8829 if (wc->stage == DROP_REFERENCE) {
8833 if (wc->level == 1 &&
8834 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8836 if (!wc->update_ref ||
8837 generation <= root->root_key.offset)
8839 btrfs_node_key_to_cpu(eb, &key, slot);
8840 ret = btrfs_comp_cpu_keys(&key,
8841 &wc->update_progress);
8845 if (wc->level == 1 &&
8846 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8850 readahead_tree_block(fs_info, bytenr);
8853 wc->reada_slot = slot;
8857 * helper to process tree block while walking down the tree.
8859 * when wc->stage == UPDATE_BACKREF, this function updates
8860 * back refs for pointers in the block.
8862 * NOTE: return value 1 means we should stop walking down.
8864 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8865 struct btrfs_root *root,
8866 struct btrfs_path *path,
8867 struct walk_control *wc, int lookup_info)
8869 struct btrfs_fs_info *fs_info = root->fs_info;
8870 int level = wc->level;
8871 struct extent_buffer *eb = path->nodes[level];
8872 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8875 if (wc->stage == UPDATE_BACKREF &&
8876 btrfs_header_owner(eb) != root->root_key.objectid)
8880 * when reference count of tree block is 1, it won't increase
8881 * again. once full backref flag is set, we never clear it.
8884 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8885 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8886 BUG_ON(!path->locks[level]);
8887 ret = btrfs_lookup_extent_info(trans, fs_info,
8888 eb->start, level, 1,
8891 BUG_ON(ret == -ENOMEM);
8894 BUG_ON(wc->refs[level] == 0);
8897 if (wc->stage == DROP_REFERENCE) {
8898 if (wc->refs[level] > 1)
8901 if (path->locks[level] && !wc->keep_locks) {
8902 btrfs_tree_unlock_rw(eb, path->locks[level]);
8903 path->locks[level] = 0;
8908 /* wc->stage == UPDATE_BACKREF */
8909 if (!(wc->flags[level] & flag)) {
8910 BUG_ON(!path->locks[level]);
8911 ret = btrfs_inc_ref(trans, root, eb, 1);
8912 BUG_ON(ret); /* -ENOMEM */
8913 ret = btrfs_dec_ref(trans, root, eb, 0);
8914 BUG_ON(ret); /* -ENOMEM */
8915 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8917 btrfs_header_level(eb), 0);
8918 BUG_ON(ret); /* -ENOMEM */
8919 wc->flags[level] |= flag;
8923 * the block is shared by multiple trees, so it's not good to
8924 * keep the tree lock
8926 if (path->locks[level] && level > 0) {
8927 btrfs_tree_unlock_rw(eb, path->locks[level]);
8928 path->locks[level] = 0;
8934 * This is used to verify a ref exists for this root to deal with a bug where we
8935 * would have a drop_progress key that hadn't been updated properly.
8937 static int check_ref_exists(struct btrfs_trans_handle *trans,
8938 struct btrfs_root *root, u64 bytenr, u64 parent,
8941 struct btrfs_path *path;
8942 struct btrfs_extent_inline_ref *iref;
8945 path = btrfs_alloc_path();
8949 ret = lookup_extent_backref(trans, path, &iref, bytenr,
8950 root->fs_info->nodesize, parent,
8951 root->root_key.objectid, level, 0);
8952 btrfs_free_path(path);
8961 * helper to process tree block pointer.
8963 * when wc->stage == DROP_REFERENCE, this function checks
8964 * reference count of the block pointed to. if the block
8965 * is shared and we need update back refs for the subtree
8966 * rooted at the block, this function changes wc->stage to
8967 * UPDATE_BACKREF. if the block is shared and there is no
8968 * need to update back, this function drops the reference
8971 * NOTE: return value 1 means we should stop walking down.
8973 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8974 struct btrfs_root *root,
8975 struct btrfs_path *path,
8976 struct walk_control *wc, int *lookup_info)
8978 struct btrfs_fs_info *fs_info = root->fs_info;
8982 struct btrfs_key key;
8983 struct btrfs_key first_key;
8984 struct extent_buffer *next;
8985 int level = wc->level;
8988 bool need_account = false;
8990 generation = btrfs_node_ptr_generation(path->nodes[level],
8991 path->slots[level]);
8993 * if the lower level block was created before the snapshot
8994 * was created, we know there is no need to update back refs
8997 if (wc->stage == UPDATE_BACKREF &&
8998 generation <= root->root_key.offset) {
9003 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
9004 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
9005 path->slots[level]);
9007 next = find_extent_buffer(fs_info, bytenr);
9009 next = btrfs_find_create_tree_block(fs_info, bytenr);
9011 return PTR_ERR(next);
9013 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
9017 btrfs_tree_lock(next);
9018 btrfs_set_lock_blocking_write(next);
9020 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
9021 &wc->refs[level - 1],
9022 &wc->flags[level - 1]);
9026 if (unlikely(wc->refs[level - 1] == 0)) {
9027 btrfs_err(fs_info, "Missing references.");
9033 if (wc->stage == DROP_REFERENCE) {
9034 if (wc->refs[level - 1] > 1) {
9035 need_account = true;
9037 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9040 if (!wc->update_ref ||
9041 generation <= root->root_key.offset)
9044 btrfs_node_key_to_cpu(path->nodes[level], &key,
9045 path->slots[level]);
9046 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
9050 wc->stage = UPDATE_BACKREF;
9051 wc->shared_level = level - 1;
9055 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
9059 if (!btrfs_buffer_uptodate(next, generation, 0)) {
9060 btrfs_tree_unlock(next);
9061 free_extent_buffer(next);
9067 if (reada && level == 1)
9068 reada_walk_down(trans, root, wc, path);
9069 next = read_tree_block(fs_info, bytenr, generation, level - 1,
9072 return PTR_ERR(next);
9073 } else if (!extent_buffer_uptodate(next)) {
9074 free_extent_buffer(next);
9077 btrfs_tree_lock(next);
9078 btrfs_set_lock_blocking_write(next);
9082 ASSERT(level == btrfs_header_level(next));
9083 if (level != btrfs_header_level(next)) {
9084 btrfs_err(root->fs_info, "mismatched level");
9088 path->nodes[level] = next;
9089 path->slots[level] = 0;
9090 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9096 wc->refs[level - 1] = 0;
9097 wc->flags[level - 1] = 0;
9098 if (wc->stage == DROP_REFERENCE) {
9099 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9100 parent = path->nodes[level]->start;
9102 ASSERT(root->root_key.objectid ==
9103 btrfs_header_owner(path->nodes[level]));
9104 if (root->root_key.objectid !=
9105 btrfs_header_owner(path->nodes[level])) {
9106 btrfs_err(root->fs_info,
9107 "mismatched block owner");
9115 * If we had a drop_progress we need to verify the refs are set
9116 * as expected. If we find our ref then we know that from here
9117 * on out everything should be correct, and we can clear the
9120 if (wc->restarted) {
9121 ret = check_ref_exists(trans, root, bytenr, parent,
9132 * Reloc tree doesn't contribute to qgroup numbers, and we have
9133 * already accounted them at merge time (replace_path),
9134 * thus we could skip expensive subtree trace here.
9136 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9138 ret = btrfs_qgroup_trace_subtree(trans, next,
9139 generation, level - 1);
9141 btrfs_err_rl(fs_info,
9142 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9148 * We need to update the next key in our walk control so we can
9149 * update the drop_progress key accordingly. We don't care if
9150 * find_next_key doesn't find a key because that means we're at
9151 * the end and are going to clean up now.
9153 wc->drop_level = level;
9154 find_next_key(path, level, &wc->drop_progress);
9156 ret = btrfs_free_extent(trans, root, bytenr, fs_info->nodesize,
9157 parent, root->root_key.objectid,
9167 btrfs_tree_unlock(next);
9168 free_extent_buffer(next);
9174 * helper to process tree block while walking up the tree.
9176 * when wc->stage == DROP_REFERENCE, this function drops
9177 * reference count on the block.
9179 * when wc->stage == UPDATE_BACKREF, this function changes
9180 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9181 * to UPDATE_BACKREF previously while processing the block.
9183 * NOTE: return value 1 means we should stop walking up.
9185 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9186 struct btrfs_root *root,
9187 struct btrfs_path *path,
9188 struct walk_control *wc)
9190 struct btrfs_fs_info *fs_info = root->fs_info;
9192 int level = wc->level;
9193 struct extent_buffer *eb = path->nodes[level];
9196 if (wc->stage == UPDATE_BACKREF) {
9197 BUG_ON(wc->shared_level < level);
9198 if (level < wc->shared_level)
9201 ret = find_next_key(path, level + 1, &wc->update_progress);
9205 wc->stage = DROP_REFERENCE;
9206 wc->shared_level = -1;
9207 path->slots[level] = 0;
9210 * check reference count again if the block isn't locked.
9211 * we should start walking down the tree again if reference
9214 if (!path->locks[level]) {
9216 btrfs_tree_lock(eb);
9217 btrfs_set_lock_blocking_write(eb);
9218 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9220 ret = btrfs_lookup_extent_info(trans, fs_info,
9221 eb->start, level, 1,
9225 btrfs_tree_unlock_rw(eb, path->locks[level]);
9226 path->locks[level] = 0;
9229 BUG_ON(wc->refs[level] == 0);
9230 if (wc->refs[level] == 1) {
9231 btrfs_tree_unlock_rw(eb, path->locks[level]);
9232 path->locks[level] = 0;
9238 /* wc->stage == DROP_REFERENCE */
9239 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9241 if (wc->refs[level] == 1) {
9243 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9244 ret = btrfs_dec_ref(trans, root, eb, 1);
9246 ret = btrfs_dec_ref(trans, root, eb, 0);
9247 BUG_ON(ret); /* -ENOMEM */
9248 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9250 btrfs_err_rl(fs_info,
9251 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9255 /* make block locked assertion in clean_tree_block happy */
9256 if (!path->locks[level] &&
9257 btrfs_header_generation(eb) == trans->transid) {
9258 btrfs_tree_lock(eb);
9259 btrfs_set_lock_blocking_write(eb);
9260 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9262 clean_tree_block(fs_info, eb);
9265 if (eb == root->node) {
9266 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9268 else if (root->root_key.objectid != btrfs_header_owner(eb))
9269 goto owner_mismatch;
9271 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9272 parent = path->nodes[level + 1]->start;
9273 else if (root->root_key.objectid !=
9274 btrfs_header_owner(path->nodes[level + 1]))
9275 goto owner_mismatch;
9278 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9280 wc->refs[level] = 0;
9281 wc->flags[level] = 0;
9285 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9286 btrfs_header_owner(eb), root->root_key.objectid);
9290 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9291 struct btrfs_root *root,
9292 struct btrfs_path *path,
9293 struct walk_control *wc)
9295 int level = wc->level;
9296 int lookup_info = 1;
9299 while (level >= 0) {
9300 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9307 if (path->slots[level] >=
9308 btrfs_header_nritems(path->nodes[level]))
9311 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9313 path->slots[level]++;
9322 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9323 struct btrfs_root *root,
9324 struct btrfs_path *path,
9325 struct walk_control *wc, int max_level)
9327 int level = wc->level;
9330 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9331 while (level < max_level && path->nodes[level]) {
9333 if (path->slots[level] + 1 <
9334 btrfs_header_nritems(path->nodes[level])) {
9335 path->slots[level]++;
9338 ret = walk_up_proc(trans, root, path, wc);
9344 if (path->locks[level]) {
9345 btrfs_tree_unlock_rw(path->nodes[level],
9346 path->locks[level]);
9347 path->locks[level] = 0;
9349 free_extent_buffer(path->nodes[level]);
9350 path->nodes[level] = NULL;
9358 * drop a subvolume tree.
9360 * this function traverses the tree freeing any blocks that only
9361 * referenced by the tree.
9363 * when a shared tree block is found. this function decreases its
9364 * reference count by one. if update_ref is true, this function
9365 * also make sure backrefs for the shared block and all lower level
9366 * blocks are properly updated.
9368 * If called with for_reloc == 0, may exit early with -EAGAIN
9370 int btrfs_drop_snapshot(struct btrfs_root *root,
9371 struct btrfs_block_rsv *block_rsv, int update_ref,
9374 struct btrfs_fs_info *fs_info = root->fs_info;
9375 struct btrfs_path *path;
9376 struct btrfs_trans_handle *trans;
9377 struct btrfs_root *tree_root = fs_info->tree_root;
9378 struct btrfs_root_item *root_item = &root->root_item;
9379 struct walk_control *wc;
9380 struct btrfs_key key;
9384 bool root_dropped = false;
9386 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9388 path = btrfs_alloc_path();
9394 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9396 btrfs_free_path(path);
9401 trans = btrfs_start_transaction(tree_root, 0);
9402 if (IS_ERR(trans)) {
9403 err = PTR_ERR(trans);
9407 err = btrfs_run_delayed_items(trans);
9412 trans->block_rsv = block_rsv;
9415 * This will help us catch people modifying the fs tree while we're
9416 * dropping it. It is unsafe to mess with the fs tree while it's being
9417 * dropped as we unlock the root node and parent nodes as we walk down
9418 * the tree, assuming nothing will change. If something does change
9419 * then we'll have stale information and drop references to blocks we've
9422 set_bit(BTRFS_ROOT_DELETING, &root->state);
9423 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9424 level = btrfs_header_level(root->node);
9425 path->nodes[level] = btrfs_lock_root_node(root);
9426 btrfs_set_lock_blocking_write(path->nodes[level]);
9427 path->slots[level] = 0;
9428 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9429 memset(&wc->update_progress, 0,
9430 sizeof(wc->update_progress));
9432 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9433 memcpy(&wc->update_progress, &key,
9434 sizeof(wc->update_progress));
9436 level = root_item->drop_level;
9438 path->lowest_level = level;
9439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9440 path->lowest_level = 0;
9448 * unlock our path, this is safe because only this
9449 * function is allowed to delete this snapshot
9451 btrfs_unlock_up_safe(path, 0);
9453 level = btrfs_header_level(root->node);
9455 btrfs_tree_lock(path->nodes[level]);
9456 btrfs_set_lock_blocking_write(path->nodes[level]);
9457 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9459 ret = btrfs_lookup_extent_info(trans, fs_info,
9460 path->nodes[level]->start,
9461 level, 1, &wc->refs[level],
9467 BUG_ON(wc->refs[level] == 0);
9469 if (level == root_item->drop_level)
9472 btrfs_tree_unlock(path->nodes[level]);
9473 path->locks[level] = 0;
9474 WARN_ON(wc->refs[level] != 1);
9479 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
9481 wc->shared_level = -1;
9482 wc->stage = DROP_REFERENCE;
9483 wc->update_ref = update_ref;
9485 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9489 ret = walk_down_tree(trans, root, path, wc);
9495 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9502 BUG_ON(wc->stage != DROP_REFERENCE);
9506 if (wc->stage == DROP_REFERENCE) {
9507 wc->drop_level = wc->level;
9508 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
9510 path->slots[wc->drop_level]);
9512 btrfs_cpu_key_to_disk(&root_item->drop_progress,
9513 &wc->drop_progress);
9514 root_item->drop_level = wc->drop_level;
9516 BUG_ON(wc->level == 0);
9517 if (btrfs_should_end_transaction(trans) ||
9518 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9519 ret = btrfs_update_root(trans, tree_root,
9523 btrfs_abort_transaction(trans, ret);
9528 btrfs_end_transaction_throttle(trans);
9529 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9530 btrfs_debug(fs_info,
9531 "drop snapshot early exit");
9536 trans = btrfs_start_transaction(tree_root, 0);
9537 if (IS_ERR(trans)) {
9538 err = PTR_ERR(trans);
9542 trans->block_rsv = block_rsv;
9545 btrfs_release_path(path);
9549 ret = btrfs_del_root(trans, &root->root_key);
9551 btrfs_abort_transaction(trans, ret);
9556 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9557 ret = btrfs_find_root(tree_root, &root->root_key, path,
9560 btrfs_abort_transaction(trans, ret);
9563 } else if (ret > 0) {
9564 /* if we fail to delete the orphan item this time
9565 * around, it'll get picked up the next time.
9567 * The most common failure here is just -ENOENT.
9569 btrfs_del_orphan_item(trans, tree_root,
9570 root->root_key.objectid);
9574 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9575 btrfs_add_dropped_root(trans, root);
9577 free_extent_buffer(root->node);
9578 free_extent_buffer(root->commit_root);
9579 btrfs_put_fs_root(root);
9581 root_dropped = true;
9583 btrfs_end_transaction_throttle(trans);
9586 btrfs_free_path(path);
9589 * So if we need to stop dropping the snapshot for whatever reason we
9590 * need to make sure to add it back to the dead root list so that we
9591 * keep trying to do the work later. This also cleans up roots if we
9592 * don't have it in the radix (like when we recover after a power fail
9593 * or unmount) so we don't leak memory.
9595 if (!for_reloc && !root_dropped)
9596 btrfs_add_dead_root(root);
9597 if (err && err != -EAGAIN)
9598 btrfs_handle_fs_error(fs_info, err, NULL);
9603 * drop subtree rooted at tree block 'node'.
9605 * NOTE: this function will unlock and release tree block 'node'
9606 * only used by relocation code
9608 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9609 struct btrfs_root *root,
9610 struct extent_buffer *node,
9611 struct extent_buffer *parent)
9613 struct btrfs_fs_info *fs_info = root->fs_info;
9614 struct btrfs_path *path;
9615 struct walk_control *wc;
9621 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9623 path = btrfs_alloc_path();
9627 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9629 btrfs_free_path(path);
9633 btrfs_assert_tree_locked(parent);
9634 parent_level = btrfs_header_level(parent);
9635 extent_buffer_get(parent);
9636 path->nodes[parent_level] = parent;
9637 path->slots[parent_level] = btrfs_header_nritems(parent);
9639 btrfs_assert_tree_locked(node);
9640 level = btrfs_header_level(node);
9641 path->nodes[level] = node;
9642 path->slots[level] = 0;
9643 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9645 wc->refs[parent_level] = 1;
9646 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9648 wc->shared_level = -1;
9649 wc->stage = DROP_REFERENCE;
9652 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9655 wret = walk_down_tree(trans, root, path, wc);
9661 wret = walk_up_tree(trans, root, path, wc, parent_level);
9669 btrfs_free_path(path);
9673 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9679 * if restripe for this chunk_type is on pick target profile and
9680 * return, otherwise do the usual balance
9682 stripped = get_restripe_target(fs_info, flags);
9684 return extended_to_chunk(stripped);
9686 num_devices = fs_info->fs_devices->rw_devices;
9688 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9689 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9690 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9692 if (num_devices == 1) {
9693 stripped |= BTRFS_BLOCK_GROUP_DUP;
9694 stripped = flags & ~stripped;
9696 /* turn raid0 into single device chunks */
9697 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9700 /* turn mirroring into duplication */
9701 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9702 BTRFS_BLOCK_GROUP_RAID10))
9703 return stripped | BTRFS_BLOCK_GROUP_DUP;
9705 /* they already had raid on here, just return */
9706 if (flags & stripped)
9709 stripped |= BTRFS_BLOCK_GROUP_DUP;
9710 stripped = flags & ~stripped;
9712 /* switch duplicated blocks with raid1 */
9713 if (flags & BTRFS_BLOCK_GROUP_DUP)
9714 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9716 /* this is drive concat, leave it alone */
9722 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9724 struct btrfs_space_info *sinfo = cache->space_info;
9727 u64 min_allocable_bytes;
9731 * We need some metadata space and system metadata space for
9732 * allocating chunks in some corner cases until we force to set
9733 * it to be readonly.
9736 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9738 min_allocable_bytes = SZ_1M;
9740 min_allocable_bytes = 0;
9742 spin_lock(&sinfo->lock);
9743 spin_lock(&cache->lock);
9751 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9752 cache->bytes_super - btrfs_block_group_used(&cache->item);
9753 sinfo_used = btrfs_space_info_used(sinfo, true);
9755 if (sinfo_used + num_bytes + min_allocable_bytes <=
9756 sinfo->total_bytes) {
9757 sinfo->bytes_readonly += num_bytes;
9759 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9763 spin_unlock(&cache->lock);
9764 spin_unlock(&sinfo->lock);
9765 if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) {
9766 btrfs_info(cache->fs_info,
9767 "unable to make block group %llu ro",
9768 cache->key.objectid);
9769 btrfs_info(cache->fs_info,
9770 "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu",
9771 sinfo_used, num_bytes, min_allocable_bytes);
9772 dump_space_info(cache->fs_info, cache->space_info, 0, 0);
9777 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9780 struct btrfs_fs_info *fs_info = cache->fs_info;
9781 struct btrfs_trans_handle *trans;
9786 trans = btrfs_join_transaction(fs_info->extent_root);
9788 return PTR_ERR(trans);
9791 * we're not allowed to set block groups readonly after the dirty
9792 * block groups cache has started writing. If it already started,
9793 * back off and let this transaction commit
9795 mutex_lock(&fs_info->ro_block_group_mutex);
9796 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9797 u64 transid = trans->transid;
9799 mutex_unlock(&fs_info->ro_block_group_mutex);
9800 btrfs_end_transaction(trans);
9802 ret = btrfs_wait_for_commit(fs_info, transid);
9809 * if we are changing raid levels, try to allocate a corresponding
9810 * block group with the new raid level.
9812 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9813 if (alloc_flags != cache->flags) {
9814 ret = do_chunk_alloc(trans, alloc_flags,
9817 * ENOSPC is allowed here, we may have enough space
9818 * already allocated at the new raid level to
9827 ret = inc_block_group_ro(cache, 0);
9830 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9831 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9834 ret = inc_block_group_ro(cache, 0);
9836 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9837 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9838 mutex_lock(&fs_info->chunk_mutex);
9839 check_system_chunk(trans, alloc_flags);
9840 mutex_unlock(&fs_info->chunk_mutex);
9842 mutex_unlock(&fs_info->ro_block_group_mutex);
9844 btrfs_end_transaction(trans);
9848 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9850 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9852 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9856 * helper to account the unused space of all the readonly block group in the
9857 * space_info. takes mirrors into account.
9859 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9861 struct btrfs_block_group_cache *block_group;
9865 /* It's df, we don't care if it's racy */
9866 if (list_empty(&sinfo->ro_bgs))
9869 spin_lock(&sinfo->lock);
9870 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9871 spin_lock(&block_group->lock);
9873 if (!block_group->ro) {
9874 spin_unlock(&block_group->lock);
9878 factor = btrfs_bg_type_to_factor(block_group->flags);
9879 free_bytes += (block_group->key.offset -
9880 btrfs_block_group_used(&block_group->item)) *
9883 spin_unlock(&block_group->lock);
9885 spin_unlock(&sinfo->lock);
9890 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9892 struct btrfs_space_info *sinfo = cache->space_info;
9897 spin_lock(&sinfo->lock);
9898 spin_lock(&cache->lock);
9900 num_bytes = cache->key.offset - cache->reserved -
9901 cache->pinned - cache->bytes_super -
9902 btrfs_block_group_used(&cache->item);
9903 sinfo->bytes_readonly -= num_bytes;
9904 list_del_init(&cache->ro_list);
9906 spin_unlock(&cache->lock);
9907 spin_unlock(&sinfo->lock);
9911 * Checks to see if it's even possible to relocate this block group.
9913 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9914 * ok to go ahead and try.
9916 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9918 struct btrfs_root *root = fs_info->extent_root;
9919 struct btrfs_block_group_cache *block_group;
9920 struct btrfs_space_info *space_info;
9921 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9922 struct btrfs_device *device;
9923 struct btrfs_trans_handle *trans;
9933 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9935 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9937 /* odd, couldn't find the block group, leave it alone */
9941 "can't find block group for bytenr %llu",
9946 min_free = btrfs_block_group_used(&block_group->item);
9948 /* no bytes used, we're good */
9952 space_info = block_group->space_info;
9953 spin_lock(&space_info->lock);
9955 full = space_info->full;
9958 * if this is the last block group we have in this space, we can't
9959 * relocate it unless we're able to allocate a new chunk below.
9961 * Otherwise, we need to make sure we have room in the space to handle
9962 * all of the extents from this block group. If we can, we're good
9964 if ((space_info->total_bytes != block_group->key.offset) &&
9965 (btrfs_space_info_used(space_info, false) + min_free <
9966 space_info->total_bytes)) {
9967 spin_unlock(&space_info->lock);
9970 spin_unlock(&space_info->lock);
9973 * ok we don't have enough space, but maybe we have free space on our
9974 * devices to allocate new chunks for relocation, so loop through our
9975 * alloc devices and guess if we have enough space. if this block
9976 * group is going to be restriped, run checks against the target
9977 * profile instead of the current one.
9989 target = get_restripe_target(fs_info, block_group->flags);
9991 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9994 * this is just a balance, so if we were marked as full
9995 * we know there is no space for a new chunk
10000 "no space to alloc new chunk for block group %llu",
10001 block_group->key.objectid);
10005 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10008 if (index == BTRFS_RAID_RAID10) {
10012 } else if (index == BTRFS_RAID_RAID1) {
10014 } else if (index == BTRFS_RAID_DUP) {
10015 /* Multiply by 2 */
10017 } else if (index == BTRFS_RAID_RAID0) {
10018 dev_min = fs_devices->rw_devices;
10019 min_free = div64_u64(min_free, dev_min);
10022 /* We need to do this so that we can look at pending chunks */
10023 trans = btrfs_join_transaction(root);
10024 if (IS_ERR(trans)) {
10025 ret = PTR_ERR(trans);
10029 mutex_lock(&fs_info->chunk_mutex);
10030 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
10034 * check to make sure we can actually find a chunk with enough
10035 * space to fit our block group in.
10037 if (device->total_bytes > device->bytes_used + min_free &&
10038 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
10039 ret = find_free_dev_extent(trans, device, min_free,
10040 &dev_offset, NULL);
10044 if (dev_nr >= dev_min)
10050 if (debug && ret == -1)
10051 btrfs_warn(fs_info,
10052 "no space to allocate a new chunk for block group %llu",
10053 block_group->key.objectid);
10054 mutex_unlock(&fs_info->chunk_mutex);
10055 btrfs_end_transaction(trans);
10057 btrfs_put_block_group(block_group);
10061 static int find_first_block_group(struct btrfs_fs_info *fs_info,
10062 struct btrfs_path *path,
10063 struct btrfs_key *key)
10065 struct btrfs_root *root = fs_info->extent_root;
10067 struct btrfs_key found_key;
10068 struct extent_buffer *leaf;
10069 struct btrfs_block_group_item bg;
10073 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
10078 slot = path->slots[0];
10079 leaf = path->nodes[0];
10080 if (slot >= btrfs_header_nritems(leaf)) {
10081 ret = btrfs_next_leaf(root, path);
10088 btrfs_item_key_to_cpu(leaf, &found_key, slot);
10090 if (found_key.objectid >= key->objectid &&
10091 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
10092 struct extent_map_tree *em_tree;
10093 struct extent_map *em;
10095 em_tree = &root->fs_info->mapping_tree.map_tree;
10096 read_lock(&em_tree->lock);
10097 em = lookup_extent_mapping(em_tree, found_key.objectid,
10099 read_unlock(&em_tree->lock);
10102 "logical %llu len %llu found bg but no related chunk",
10103 found_key.objectid, found_key.offset);
10105 } else if (em->start != found_key.objectid ||
10106 em->len != found_key.offset) {
10108 "block group %llu len %llu mismatch with chunk %llu len %llu",
10109 found_key.objectid, found_key.offset,
10110 em->start, em->len);
10113 read_extent_buffer(leaf, &bg,
10114 btrfs_item_ptr_offset(leaf, slot),
10116 flags = btrfs_block_group_flags(&bg) &
10117 BTRFS_BLOCK_GROUP_TYPE_MASK;
10119 if (flags != (em->map_lookup->type &
10120 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10122 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
10123 found_key.objectid,
10124 found_key.offset, flags,
10125 (BTRFS_BLOCK_GROUP_TYPE_MASK &
10126 em->map_lookup->type));
10132 free_extent_map(em);
10141 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
10143 struct btrfs_block_group_cache *block_group;
10147 struct inode *inode;
10149 block_group = btrfs_lookup_first_block_group(info, last);
10150 while (block_group) {
10151 wait_block_group_cache_done(block_group);
10152 spin_lock(&block_group->lock);
10153 if (block_group->iref)
10155 spin_unlock(&block_group->lock);
10156 block_group = next_block_group(info, block_group);
10158 if (!block_group) {
10165 inode = block_group->inode;
10166 block_group->iref = 0;
10167 block_group->inode = NULL;
10168 spin_unlock(&block_group->lock);
10169 ASSERT(block_group->io_ctl.inode == NULL);
10171 last = block_group->key.objectid + block_group->key.offset;
10172 btrfs_put_block_group(block_group);
10177 * Must be called only after stopping all workers, since we could have block
10178 * group caching kthreads running, and therefore they could race with us if we
10179 * freed the block groups before stopping them.
10181 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10183 struct btrfs_block_group_cache *block_group;
10184 struct btrfs_space_info *space_info;
10185 struct btrfs_caching_control *caching_ctl;
10188 down_write(&info->commit_root_sem);
10189 while (!list_empty(&info->caching_block_groups)) {
10190 caching_ctl = list_entry(info->caching_block_groups.next,
10191 struct btrfs_caching_control, list);
10192 list_del(&caching_ctl->list);
10193 put_caching_control(caching_ctl);
10195 up_write(&info->commit_root_sem);
10197 spin_lock(&info->unused_bgs_lock);
10198 while (!list_empty(&info->unused_bgs)) {
10199 block_group = list_first_entry(&info->unused_bgs,
10200 struct btrfs_block_group_cache,
10202 list_del_init(&block_group->bg_list);
10203 btrfs_put_block_group(block_group);
10205 spin_unlock(&info->unused_bgs_lock);
10207 spin_lock(&info->block_group_cache_lock);
10208 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10209 block_group = rb_entry(n, struct btrfs_block_group_cache,
10211 rb_erase(&block_group->cache_node,
10212 &info->block_group_cache_tree);
10213 RB_CLEAR_NODE(&block_group->cache_node);
10214 spin_unlock(&info->block_group_cache_lock);
10216 down_write(&block_group->space_info->groups_sem);
10217 list_del(&block_group->list);
10218 up_write(&block_group->space_info->groups_sem);
10221 * We haven't cached this block group, which means we could
10222 * possibly have excluded extents on this block group.
10224 if (block_group->cached == BTRFS_CACHE_NO ||
10225 block_group->cached == BTRFS_CACHE_ERROR)
10226 free_excluded_extents(block_group);
10228 btrfs_remove_free_space_cache(block_group);
10229 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10230 ASSERT(list_empty(&block_group->dirty_list));
10231 ASSERT(list_empty(&block_group->io_list));
10232 ASSERT(list_empty(&block_group->bg_list));
10233 ASSERT(atomic_read(&block_group->count) == 1);
10234 btrfs_put_block_group(block_group);
10236 spin_lock(&info->block_group_cache_lock);
10238 spin_unlock(&info->block_group_cache_lock);
10240 /* now that all the block groups are freed, go through and
10241 * free all the space_info structs. This is only called during
10242 * the final stages of unmount, and so we know nobody is
10243 * using them. We call synchronize_rcu() once before we start,
10244 * just to be on the safe side.
10248 release_global_block_rsv(info);
10250 while (!list_empty(&info->space_info)) {
10253 space_info = list_entry(info->space_info.next,
10254 struct btrfs_space_info,
10258 * Do not hide this behind enospc_debug, this is actually
10259 * important and indicates a real bug if this happens.
10261 if (WARN_ON(space_info->bytes_pinned > 0 ||
10262 space_info->bytes_reserved > 0 ||
10263 space_info->bytes_may_use > 0))
10264 dump_space_info(info, space_info, 0, 0);
10265 list_del(&space_info->list);
10266 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10267 struct kobject *kobj;
10268 kobj = space_info->block_group_kobjs[i];
10269 space_info->block_group_kobjs[i] = NULL;
10275 kobject_del(&space_info->kobj);
10276 kobject_put(&space_info->kobj);
10281 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10282 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10284 struct btrfs_space_info *space_info;
10285 struct raid_kobject *rkobj;
10290 spin_lock(&fs_info->pending_raid_kobjs_lock);
10291 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10292 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10294 list_for_each_entry(rkobj, &list, list) {
10295 space_info = __find_space_info(fs_info, rkobj->flags);
10296 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10298 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10299 "%s", get_raid_name(index));
10301 kobject_put(&rkobj->kobj);
10306 btrfs_warn(fs_info,
10307 "failed to add kobject for block cache, ignoring");
10310 static void link_block_group(struct btrfs_block_group_cache *cache)
10312 struct btrfs_space_info *space_info = cache->space_info;
10313 struct btrfs_fs_info *fs_info = cache->fs_info;
10314 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10315 bool first = false;
10317 down_write(&space_info->groups_sem);
10318 if (list_empty(&space_info->block_groups[index]))
10320 list_add_tail(&cache->list, &space_info->block_groups[index]);
10321 up_write(&space_info->groups_sem);
10324 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10326 btrfs_warn(cache->fs_info,
10327 "couldn't alloc memory for raid level kobject");
10330 rkobj->flags = cache->flags;
10331 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10333 spin_lock(&fs_info->pending_raid_kobjs_lock);
10334 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10335 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10336 space_info->block_group_kobjs[index] = &rkobj->kobj;
10340 static struct btrfs_block_group_cache *
10341 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10342 u64 start, u64 size)
10344 struct btrfs_block_group_cache *cache;
10346 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10350 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10352 if (!cache->free_space_ctl) {
10357 cache->key.objectid = start;
10358 cache->key.offset = size;
10359 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10361 cache->fs_info = fs_info;
10362 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10363 set_free_space_tree_thresholds(cache);
10365 atomic_set(&cache->count, 1);
10366 spin_lock_init(&cache->lock);
10367 init_rwsem(&cache->data_rwsem);
10368 INIT_LIST_HEAD(&cache->list);
10369 INIT_LIST_HEAD(&cache->cluster_list);
10370 INIT_LIST_HEAD(&cache->bg_list);
10371 INIT_LIST_HEAD(&cache->ro_list);
10372 INIT_LIST_HEAD(&cache->dirty_list);
10373 INIT_LIST_HEAD(&cache->io_list);
10374 btrfs_init_free_space_ctl(cache);
10375 atomic_set(&cache->trimming, 0);
10376 mutex_init(&cache->free_space_lock);
10377 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10384 * Iterate all chunks and verify that each of them has the corresponding block
10387 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10389 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10390 struct extent_map *em;
10391 struct btrfs_block_group_cache *bg;
10396 read_lock(&map_tree->map_tree.lock);
10398 * lookup_extent_mapping will return the first extent map
10399 * intersecting the range, so setting @len to 1 is enough to
10400 * get the first chunk.
10402 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10403 read_unlock(&map_tree->map_tree.lock);
10407 bg = btrfs_lookup_block_group(fs_info, em->start);
10410 "chunk start=%llu len=%llu doesn't have corresponding block group",
10411 em->start, em->len);
10413 free_extent_map(em);
10416 if (bg->key.objectid != em->start ||
10417 bg->key.offset != em->len ||
10418 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10419 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10421 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10422 em->start, em->len,
10423 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10424 bg->key.objectid, bg->key.offset,
10425 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10427 free_extent_map(em);
10428 btrfs_put_block_group(bg);
10431 start = em->start + em->len;
10432 free_extent_map(em);
10433 btrfs_put_block_group(bg);
10438 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10440 struct btrfs_path *path;
10442 struct btrfs_block_group_cache *cache;
10443 struct btrfs_space_info *space_info;
10444 struct btrfs_key key;
10445 struct btrfs_key found_key;
10446 struct extent_buffer *leaf;
10447 int need_clear = 0;
10452 feature = btrfs_super_incompat_flags(info->super_copy);
10453 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10457 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10458 path = btrfs_alloc_path();
10461 path->reada = READA_FORWARD;
10463 cache_gen = btrfs_super_cache_generation(info->super_copy);
10464 if (btrfs_test_opt(info, SPACE_CACHE) &&
10465 btrfs_super_generation(info->super_copy) != cache_gen)
10467 if (btrfs_test_opt(info, CLEAR_CACHE))
10471 ret = find_first_block_group(info, path, &key);
10477 leaf = path->nodes[0];
10478 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10480 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10489 * When we mount with old space cache, we need to
10490 * set BTRFS_DC_CLEAR and set dirty flag.
10492 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10493 * truncate the old free space cache inode and
10495 * b) Setting 'dirty flag' makes sure that we flush
10496 * the new space cache info onto disk.
10498 if (btrfs_test_opt(info, SPACE_CACHE))
10499 cache->disk_cache_state = BTRFS_DC_CLEAR;
10502 read_extent_buffer(leaf, &cache->item,
10503 btrfs_item_ptr_offset(leaf, path->slots[0]),
10504 sizeof(cache->item));
10505 cache->flags = btrfs_block_group_flags(&cache->item);
10507 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10508 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10510 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10511 cache->key.objectid);
10516 key.objectid = found_key.objectid + found_key.offset;
10517 btrfs_release_path(path);
10520 * We need to exclude the super stripes now so that the space
10521 * info has super bytes accounted for, otherwise we'll think
10522 * we have more space than we actually do.
10524 ret = exclude_super_stripes(cache);
10527 * We may have excluded something, so call this just in
10530 free_excluded_extents(cache);
10531 btrfs_put_block_group(cache);
10536 * check for two cases, either we are full, and therefore
10537 * don't need to bother with the caching work since we won't
10538 * find any space, or we are empty, and we can just add all
10539 * the space in and be done with it. This saves us _a_lot_ of
10540 * time, particularly in the full case.
10542 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10543 cache->last_byte_to_unpin = (u64)-1;
10544 cache->cached = BTRFS_CACHE_FINISHED;
10545 free_excluded_extents(cache);
10546 } else if (btrfs_block_group_used(&cache->item) == 0) {
10547 cache->last_byte_to_unpin = (u64)-1;
10548 cache->cached = BTRFS_CACHE_FINISHED;
10549 add_new_free_space(cache, found_key.objectid,
10550 found_key.objectid +
10552 free_excluded_extents(cache);
10555 ret = btrfs_add_block_group_cache(info, cache);
10557 btrfs_remove_free_space_cache(cache);
10558 btrfs_put_block_group(cache);
10562 trace_btrfs_add_block_group(info, cache, 0);
10563 update_space_info(info, cache->flags, found_key.offset,
10564 btrfs_block_group_used(&cache->item),
10565 cache->bytes_super, &space_info);
10567 cache->space_info = space_info;
10569 link_block_group(cache);
10571 set_avail_alloc_bits(info, cache->flags);
10572 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10573 inc_block_group_ro(cache, 1);
10574 } else if (btrfs_block_group_used(&cache->item) == 0) {
10575 ASSERT(list_empty(&cache->bg_list));
10576 btrfs_mark_bg_unused(cache);
10580 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10581 if (!(get_alloc_profile(info, space_info->flags) &
10582 (BTRFS_BLOCK_GROUP_RAID10 |
10583 BTRFS_BLOCK_GROUP_RAID1 |
10584 BTRFS_BLOCK_GROUP_RAID5 |
10585 BTRFS_BLOCK_GROUP_RAID6 |
10586 BTRFS_BLOCK_GROUP_DUP)))
10589 * avoid allocating from un-mirrored block group if there are
10590 * mirrored block groups.
10592 list_for_each_entry(cache,
10593 &space_info->block_groups[BTRFS_RAID_RAID0],
10595 inc_block_group_ro(cache, 1);
10596 list_for_each_entry(cache,
10597 &space_info->block_groups[BTRFS_RAID_SINGLE],
10599 inc_block_group_ro(cache, 1);
10602 btrfs_add_raid_kobjects(info);
10603 init_global_block_rsv(info);
10604 ret = check_chunk_block_group_mappings(info);
10606 btrfs_free_path(path);
10610 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10612 struct btrfs_fs_info *fs_info = trans->fs_info;
10613 struct btrfs_block_group_cache *block_group;
10614 struct btrfs_root *extent_root = fs_info->extent_root;
10615 struct btrfs_block_group_item item;
10616 struct btrfs_key key;
10619 if (!trans->can_flush_pending_bgs)
10622 while (!list_empty(&trans->new_bgs)) {
10623 block_group = list_first_entry(&trans->new_bgs,
10624 struct btrfs_block_group_cache,
10629 spin_lock(&block_group->lock);
10630 memcpy(&item, &block_group->item, sizeof(item));
10631 memcpy(&key, &block_group->key, sizeof(key));
10632 spin_unlock(&block_group->lock);
10634 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10637 btrfs_abort_transaction(trans, ret);
10638 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10640 btrfs_abort_transaction(trans, ret);
10641 add_block_group_free_space(trans, block_group);
10642 /* already aborted the transaction if it failed. */
10644 btrfs_delayed_refs_rsv_release(fs_info, 1);
10645 list_del_init(&block_group->bg_list);
10647 btrfs_trans_release_chunk_metadata(trans);
10650 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10651 u64 type, u64 chunk_offset, u64 size)
10653 struct btrfs_fs_info *fs_info = trans->fs_info;
10654 struct btrfs_block_group_cache *cache;
10657 btrfs_set_log_full_commit(fs_info, trans);
10659 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10663 btrfs_set_block_group_used(&cache->item, bytes_used);
10664 btrfs_set_block_group_chunk_objectid(&cache->item,
10665 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10666 btrfs_set_block_group_flags(&cache->item, type);
10668 cache->flags = type;
10669 cache->last_byte_to_unpin = (u64)-1;
10670 cache->cached = BTRFS_CACHE_FINISHED;
10671 cache->needs_free_space = 1;
10672 ret = exclude_super_stripes(cache);
10675 * We may have excluded something, so call this just in
10678 free_excluded_extents(cache);
10679 btrfs_put_block_group(cache);
10683 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10685 free_excluded_extents(cache);
10687 #ifdef CONFIG_BTRFS_DEBUG
10688 if (btrfs_should_fragment_free_space(cache)) {
10689 u64 new_bytes_used = size - bytes_used;
10691 bytes_used += new_bytes_used >> 1;
10692 fragment_free_space(cache);
10696 * Ensure the corresponding space_info object is created and
10697 * assigned to our block group. We want our bg to be added to the rbtree
10698 * with its ->space_info set.
10700 cache->space_info = __find_space_info(fs_info, cache->flags);
10701 ASSERT(cache->space_info);
10703 ret = btrfs_add_block_group_cache(fs_info, cache);
10705 btrfs_remove_free_space_cache(cache);
10706 btrfs_put_block_group(cache);
10711 * Now that our block group has its ->space_info set and is inserted in
10712 * the rbtree, update the space info's counters.
10714 trace_btrfs_add_block_group(fs_info, cache, 1);
10715 update_space_info(fs_info, cache->flags, size, bytes_used,
10716 cache->bytes_super, &cache->space_info);
10717 update_global_block_rsv(fs_info);
10719 link_block_group(cache);
10721 list_add_tail(&cache->bg_list, &trans->new_bgs);
10722 trans->delayed_ref_updates++;
10723 btrfs_update_delayed_refs_rsv(trans);
10725 set_avail_alloc_bits(fs_info, type);
10729 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10731 u64 extra_flags = chunk_to_extended(flags) &
10732 BTRFS_EXTENDED_PROFILE_MASK;
10734 write_seqlock(&fs_info->profiles_lock);
10735 if (flags & BTRFS_BLOCK_GROUP_DATA)
10736 fs_info->avail_data_alloc_bits &= ~extra_flags;
10737 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10738 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10739 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10740 fs_info->avail_system_alloc_bits &= ~extra_flags;
10741 write_sequnlock(&fs_info->profiles_lock);
10744 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10745 u64 group_start, struct extent_map *em)
10747 struct btrfs_fs_info *fs_info = trans->fs_info;
10748 struct btrfs_root *root = fs_info->extent_root;
10749 struct btrfs_path *path;
10750 struct btrfs_block_group_cache *block_group;
10751 struct btrfs_free_cluster *cluster;
10752 struct btrfs_root *tree_root = fs_info->tree_root;
10753 struct btrfs_key key;
10754 struct inode *inode;
10755 struct kobject *kobj = NULL;
10759 struct btrfs_caching_control *caching_ctl = NULL;
10761 bool remove_rsv = false;
10763 block_group = btrfs_lookup_block_group(fs_info, group_start);
10764 BUG_ON(!block_group);
10765 BUG_ON(!block_group->ro);
10767 trace_btrfs_remove_block_group(block_group);
10769 * Free the reserved super bytes from this block group before
10772 free_excluded_extents(block_group);
10773 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10774 block_group->key.offset);
10776 memcpy(&key, &block_group->key, sizeof(key));
10777 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10778 factor = btrfs_bg_type_to_factor(block_group->flags);
10780 /* make sure this block group isn't part of an allocation cluster */
10781 cluster = &fs_info->data_alloc_cluster;
10782 spin_lock(&cluster->refill_lock);
10783 btrfs_return_cluster_to_free_space(block_group, cluster);
10784 spin_unlock(&cluster->refill_lock);
10787 * make sure this block group isn't part of a metadata
10788 * allocation cluster
10790 cluster = &fs_info->meta_alloc_cluster;
10791 spin_lock(&cluster->refill_lock);
10792 btrfs_return_cluster_to_free_space(block_group, cluster);
10793 spin_unlock(&cluster->refill_lock);
10795 path = btrfs_alloc_path();
10802 * get the inode first so any iput calls done for the io_list
10803 * aren't the final iput (no unlinks allowed now)
10805 inode = lookup_free_space_inode(fs_info, block_group, path);
10807 mutex_lock(&trans->transaction->cache_write_mutex);
10809 * Make sure our free space cache IO is done before removing the
10812 spin_lock(&trans->transaction->dirty_bgs_lock);
10813 if (!list_empty(&block_group->io_list)) {
10814 list_del_init(&block_group->io_list);
10816 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10818 spin_unlock(&trans->transaction->dirty_bgs_lock);
10819 btrfs_wait_cache_io(trans, block_group, path);
10820 btrfs_put_block_group(block_group);
10821 spin_lock(&trans->transaction->dirty_bgs_lock);
10824 if (!list_empty(&block_group->dirty_list)) {
10825 list_del_init(&block_group->dirty_list);
10827 btrfs_put_block_group(block_group);
10829 spin_unlock(&trans->transaction->dirty_bgs_lock);
10830 mutex_unlock(&trans->transaction->cache_write_mutex);
10832 if (!IS_ERR(inode)) {
10833 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10835 btrfs_add_delayed_iput(inode);
10838 clear_nlink(inode);
10839 /* One for the block groups ref */
10840 spin_lock(&block_group->lock);
10841 if (block_group->iref) {
10842 block_group->iref = 0;
10843 block_group->inode = NULL;
10844 spin_unlock(&block_group->lock);
10847 spin_unlock(&block_group->lock);
10849 /* One for our lookup ref */
10850 btrfs_add_delayed_iput(inode);
10853 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10854 key.offset = block_group->key.objectid;
10857 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10861 btrfs_release_path(path);
10863 ret = btrfs_del_item(trans, tree_root, path);
10866 btrfs_release_path(path);
10869 spin_lock(&fs_info->block_group_cache_lock);
10870 rb_erase(&block_group->cache_node,
10871 &fs_info->block_group_cache_tree);
10872 RB_CLEAR_NODE(&block_group->cache_node);
10874 if (fs_info->first_logical_byte == block_group->key.objectid)
10875 fs_info->first_logical_byte = (u64)-1;
10876 spin_unlock(&fs_info->block_group_cache_lock);
10878 down_write(&block_group->space_info->groups_sem);
10880 * we must use list_del_init so people can check to see if they
10881 * are still on the list after taking the semaphore
10883 list_del_init(&block_group->list);
10884 if (list_empty(&block_group->space_info->block_groups[index])) {
10885 kobj = block_group->space_info->block_group_kobjs[index];
10886 block_group->space_info->block_group_kobjs[index] = NULL;
10887 clear_avail_alloc_bits(fs_info, block_group->flags);
10889 up_write(&block_group->space_info->groups_sem);
10895 if (block_group->has_caching_ctl)
10896 caching_ctl = get_caching_control(block_group);
10897 if (block_group->cached == BTRFS_CACHE_STARTED)
10898 wait_block_group_cache_done(block_group);
10899 if (block_group->has_caching_ctl) {
10900 down_write(&fs_info->commit_root_sem);
10901 if (!caching_ctl) {
10902 struct btrfs_caching_control *ctl;
10904 list_for_each_entry(ctl,
10905 &fs_info->caching_block_groups, list)
10906 if (ctl->block_group == block_group) {
10908 refcount_inc(&caching_ctl->count);
10913 list_del_init(&caching_ctl->list);
10914 up_write(&fs_info->commit_root_sem);
10916 /* Once for the caching bgs list and once for us. */
10917 put_caching_control(caching_ctl);
10918 put_caching_control(caching_ctl);
10922 spin_lock(&trans->transaction->dirty_bgs_lock);
10923 WARN_ON(!list_empty(&block_group->dirty_list));
10924 WARN_ON(!list_empty(&block_group->io_list));
10925 spin_unlock(&trans->transaction->dirty_bgs_lock);
10927 btrfs_remove_free_space_cache(block_group);
10929 spin_lock(&block_group->space_info->lock);
10930 list_del_init(&block_group->ro_list);
10932 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10933 WARN_ON(block_group->space_info->total_bytes
10934 < block_group->key.offset);
10935 WARN_ON(block_group->space_info->bytes_readonly
10936 < block_group->key.offset);
10937 WARN_ON(block_group->space_info->disk_total
10938 < block_group->key.offset * factor);
10940 block_group->space_info->total_bytes -= block_group->key.offset;
10941 block_group->space_info->bytes_readonly -= block_group->key.offset;
10942 block_group->space_info->disk_total -= block_group->key.offset * factor;
10944 spin_unlock(&block_group->space_info->lock);
10946 memcpy(&key, &block_group->key, sizeof(key));
10948 mutex_lock(&fs_info->chunk_mutex);
10949 if (!list_empty(&em->list)) {
10950 /* We're in the transaction->pending_chunks list. */
10951 free_extent_map(em);
10953 spin_lock(&block_group->lock);
10954 block_group->removed = 1;
10956 * At this point trimming can't start on this block group, because we
10957 * removed the block group from the tree fs_info->block_group_cache_tree
10958 * so no one can't find it anymore and even if someone already got this
10959 * block group before we removed it from the rbtree, they have already
10960 * incremented block_group->trimming - if they didn't, they won't find
10961 * any free space entries because we already removed them all when we
10962 * called btrfs_remove_free_space_cache().
10964 * And we must not remove the extent map from the fs_info->mapping_tree
10965 * to prevent the same logical address range and physical device space
10966 * ranges from being reused for a new block group. This is because our
10967 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10968 * completely transactionless, so while it is trimming a range the
10969 * currently running transaction might finish and a new one start,
10970 * allowing for new block groups to be created that can reuse the same
10971 * physical device locations unless we take this special care.
10973 * There may also be an implicit trim operation if the file system
10974 * is mounted with -odiscard. The same protections must remain
10975 * in place until the extents have been discarded completely when
10976 * the transaction commit has completed.
10978 remove_em = (atomic_read(&block_group->trimming) == 0);
10980 * Make sure a trimmer task always sees the em in the pinned_chunks list
10981 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10982 * before checking block_group->removed).
10986 * Our em might be in trans->transaction->pending_chunks which
10987 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10988 * and so is the fs_info->pinned_chunks list.
10990 * So at this point we must be holding the chunk_mutex to avoid
10991 * any races with chunk allocation (more specifically at
10992 * volumes.c:contains_pending_extent()), to ensure it always
10993 * sees the em, either in the pending_chunks list or in the
10994 * pinned_chunks list.
10996 list_move_tail(&em->list, &fs_info->pinned_chunks);
10998 spin_unlock(&block_group->lock);
11001 struct extent_map_tree *em_tree;
11003 em_tree = &fs_info->mapping_tree.map_tree;
11004 write_lock(&em_tree->lock);
11006 * The em might be in the pending_chunks list, so make sure the
11007 * chunk mutex is locked, since remove_extent_mapping() will
11008 * delete us from that list.
11010 remove_extent_mapping(em_tree, em);
11011 write_unlock(&em_tree->lock);
11012 /* once for the tree */
11013 free_extent_map(em);
11016 mutex_unlock(&fs_info->chunk_mutex);
11018 ret = remove_block_group_free_space(trans, block_group);
11022 btrfs_put_block_group(block_group);
11023 btrfs_put_block_group(block_group);
11025 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
11031 ret = btrfs_del_item(trans, root, path);
11034 btrfs_delayed_refs_rsv_release(fs_info, 1);
11035 btrfs_free_path(path);
11039 struct btrfs_trans_handle *
11040 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
11041 const u64 chunk_offset)
11043 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
11044 struct extent_map *em;
11045 struct map_lookup *map;
11046 unsigned int num_items;
11048 read_lock(&em_tree->lock);
11049 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
11050 read_unlock(&em_tree->lock);
11051 ASSERT(em && em->start == chunk_offset);
11054 * We need to reserve 3 + N units from the metadata space info in order
11055 * to remove a block group (done at btrfs_remove_chunk() and at
11056 * btrfs_remove_block_group()), which are used for:
11058 * 1 unit for adding the free space inode's orphan (located in the tree
11060 * 1 unit for deleting the block group item (located in the extent
11062 * 1 unit for deleting the free space item (located in tree of tree
11064 * N units for deleting N device extent items corresponding to each
11065 * stripe (located in the device tree).
11067 * In order to remove a block group we also need to reserve units in the
11068 * system space info in order to update the chunk tree (update one or
11069 * more device items and remove one chunk item), but this is done at
11070 * btrfs_remove_chunk() through a call to check_system_chunk().
11072 map = em->map_lookup;
11073 num_items = 3 + map->num_stripes;
11074 free_extent_map(em);
11076 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
11081 * Process the unused_bgs list and remove any that don't have any allocated
11082 * space inside of them.
11084 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
11086 struct btrfs_block_group_cache *block_group;
11087 struct btrfs_space_info *space_info;
11088 struct btrfs_trans_handle *trans;
11091 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
11094 spin_lock(&fs_info->unused_bgs_lock);
11095 while (!list_empty(&fs_info->unused_bgs)) {
11099 block_group = list_first_entry(&fs_info->unused_bgs,
11100 struct btrfs_block_group_cache,
11102 list_del_init(&block_group->bg_list);
11104 space_info = block_group->space_info;
11106 if (ret || btrfs_mixed_space_info(space_info)) {
11107 btrfs_put_block_group(block_group);
11110 spin_unlock(&fs_info->unused_bgs_lock);
11112 mutex_lock(&fs_info->delete_unused_bgs_mutex);
11114 /* Don't want to race with allocators so take the groups_sem */
11115 down_write(&space_info->groups_sem);
11116 spin_lock(&block_group->lock);
11117 if (block_group->reserved || block_group->pinned ||
11118 btrfs_block_group_used(&block_group->item) ||
11120 list_is_singular(&block_group->list)) {
11122 * We want to bail if we made new allocations or have
11123 * outstanding allocations in this block group. We do
11124 * the ro check in case balance is currently acting on
11125 * this block group.
11127 trace_btrfs_skip_unused_block_group(block_group);
11128 spin_unlock(&block_group->lock);
11129 up_write(&space_info->groups_sem);
11132 spin_unlock(&block_group->lock);
11134 /* We don't want to force the issue, only flip if it's ok. */
11135 ret = inc_block_group_ro(block_group, 0);
11136 up_write(&space_info->groups_sem);
11143 * Want to do this before we do anything else so we can recover
11144 * properly if we fail to join the transaction.
11146 trans = btrfs_start_trans_remove_block_group(fs_info,
11147 block_group->key.objectid);
11148 if (IS_ERR(trans)) {
11149 btrfs_dec_block_group_ro(block_group);
11150 ret = PTR_ERR(trans);
11155 * We could have pending pinned extents for this block group,
11156 * just delete them, we don't care about them anymore.
11158 start = block_group->key.objectid;
11159 end = start + block_group->key.offset - 1;
11161 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11162 * btrfs_finish_extent_commit(). If we are at transaction N,
11163 * another task might be running finish_extent_commit() for the
11164 * previous transaction N - 1, and have seen a range belonging
11165 * to the block group in freed_extents[] before we were able to
11166 * clear the whole block group range from freed_extents[]. This
11167 * means that task can lookup for the block group after we
11168 * unpinned it from freed_extents[] and removed it, leading to
11169 * a BUG_ON() at btrfs_unpin_extent_range().
11171 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11172 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11175 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11176 btrfs_dec_block_group_ro(block_group);
11179 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11182 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11183 btrfs_dec_block_group_ro(block_group);
11186 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11188 /* Reset pinned so btrfs_put_block_group doesn't complain */
11189 spin_lock(&space_info->lock);
11190 spin_lock(&block_group->lock);
11192 update_bytes_pinned(space_info, -block_group->pinned);
11193 space_info->bytes_readonly += block_group->pinned;
11194 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11195 -block_group->pinned,
11196 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11197 block_group->pinned = 0;
11199 spin_unlock(&block_group->lock);
11200 spin_unlock(&space_info->lock);
11202 /* DISCARD can flip during remount */
11203 trimming = btrfs_test_opt(fs_info, DISCARD);
11205 /* Implicit trim during transaction commit. */
11207 btrfs_get_block_group_trimming(block_group);
11210 * Btrfs_remove_chunk will abort the transaction if things go
11213 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11217 btrfs_put_block_group_trimming(block_group);
11222 * If we're not mounted with -odiscard, we can just forget
11223 * about this block group. Otherwise we'll need to wait
11224 * until transaction commit to do the actual discard.
11227 spin_lock(&fs_info->unused_bgs_lock);
11229 * A concurrent scrub might have added us to the list
11230 * fs_info->unused_bgs, so use a list_move operation
11231 * to add the block group to the deleted_bgs list.
11233 list_move(&block_group->bg_list,
11234 &trans->transaction->deleted_bgs);
11235 spin_unlock(&fs_info->unused_bgs_lock);
11236 btrfs_get_block_group(block_group);
11239 btrfs_end_transaction(trans);
11241 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11242 btrfs_put_block_group(block_group);
11243 spin_lock(&fs_info->unused_bgs_lock);
11245 spin_unlock(&fs_info->unused_bgs_lock);
11248 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11250 struct btrfs_super_block *disk_super;
11256 disk_super = fs_info->super_copy;
11257 if (!btrfs_super_root(disk_super))
11260 features = btrfs_super_incompat_flags(disk_super);
11261 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11264 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11265 ret = create_space_info(fs_info, flags);
11270 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11271 ret = create_space_info(fs_info, flags);
11273 flags = BTRFS_BLOCK_GROUP_METADATA;
11274 ret = create_space_info(fs_info, flags);
11278 flags = BTRFS_BLOCK_GROUP_DATA;
11279 ret = create_space_info(fs_info, flags);
11285 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11286 u64 start, u64 end)
11288 return unpin_extent_range(fs_info, start, end, false);
11292 * It used to be that old block groups would be left around forever.
11293 * Iterating over them would be enough to trim unused space. Since we
11294 * now automatically remove them, we also need to iterate over unallocated
11297 * We don't want a transaction for this since the discard may take a
11298 * substantial amount of time. We don't require that a transaction be
11299 * running, but we do need to take a running transaction into account
11300 * to ensure that we're not discarding chunks that were released or
11301 * allocated in the current transaction.
11303 * Holding the chunks lock will prevent other threads from allocating
11304 * or releasing chunks, but it won't prevent a running transaction
11305 * from committing and releasing the memory that the pending chunks
11306 * list head uses. For that, we need to take a reference to the
11307 * transaction and hold the commit root sem. We only need to hold
11308 * it while performing the free space search since we have already
11309 * held back allocations.
11311 static int btrfs_trim_free_extents(struct btrfs_device *device,
11312 u64 minlen, u64 *trimmed)
11314 u64 start = 0, len = 0;
11319 /* Discard not supported = nothing to do. */
11320 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11323 /* Not writable = nothing to do. */
11324 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11327 /* No free space = nothing to do. */
11328 if (device->total_bytes <= device->bytes_used)
11334 struct btrfs_fs_info *fs_info = device->fs_info;
11335 struct btrfs_transaction *trans;
11338 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11342 ret = down_read_killable(&fs_info->commit_root_sem);
11344 mutex_unlock(&fs_info->chunk_mutex);
11348 spin_lock(&fs_info->trans_lock);
11349 trans = fs_info->running_transaction;
11351 refcount_inc(&trans->use_count);
11352 spin_unlock(&fs_info->trans_lock);
11355 up_read(&fs_info->commit_root_sem);
11357 ret = find_free_dev_extent_start(trans, device, minlen, start,
11360 up_read(&fs_info->commit_root_sem);
11361 btrfs_put_transaction(trans);
11365 mutex_unlock(&fs_info->chunk_mutex);
11366 if (ret == -ENOSPC)
11371 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11372 mutex_unlock(&fs_info->chunk_mutex);
11380 if (fatal_signal_pending(current)) {
11381 ret = -ERESTARTSYS;
11392 * Trim the whole filesystem by:
11393 * 1) trimming the free space in each block group
11394 * 2) trimming the unallocated space on each device
11396 * This will also continue trimming even if a block group or device encounters
11397 * an error. The return value will be the last error, or 0 if nothing bad
11400 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11402 struct btrfs_block_group_cache *cache = NULL;
11403 struct btrfs_device *device;
11404 struct list_head *devices;
11410 u64 dev_failed = 0;
11415 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11416 for (; cache; cache = next_block_group(fs_info, cache)) {
11417 if (cache->key.objectid >= (range->start + range->len)) {
11418 btrfs_put_block_group(cache);
11422 start = max(range->start, cache->key.objectid);
11423 end = min(range->start + range->len,
11424 cache->key.objectid + cache->key.offset);
11426 if (end - start >= range->minlen) {
11427 if (!block_group_cache_done(cache)) {
11428 ret = cache_block_group(cache, 0);
11434 ret = wait_block_group_cache_done(cache);
11441 ret = btrfs_trim_block_group(cache,
11447 trimmed += group_trimmed;
11457 btrfs_warn(fs_info,
11458 "failed to trim %llu block group(s), last error %d",
11459 bg_failed, bg_ret);
11460 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11461 devices = &fs_info->fs_devices->devices;
11462 list_for_each_entry(device, devices, dev_list) {
11463 ret = btrfs_trim_free_extents(device, range->minlen,
11471 trimmed += group_trimmed;
11473 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11476 btrfs_warn(fs_info,
11477 "failed to trim %llu device(s), last error %d",
11478 dev_failed, dev_ret);
11479 range->len = trimmed;
11486 * btrfs_{start,end}_write_no_snapshotting() are similar to
11487 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11488 * data into the page cache through nocow before the subvolume is snapshoted,
11489 * but flush the data into disk after the snapshot creation, or to prevent
11490 * operations while snapshotting is ongoing and that cause the snapshot to be
11491 * inconsistent (writes followed by expanding truncates for example).
11493 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11495 percpu_counter_dec(&root->subv_writers->counter);
11496 cond_wake_up(&root->subv_writers->wait);
11499 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11501 if (atomic_read(&root->will_be_snapshotted))
11504 percpu_counter_inc(&root->subv_writers->counter);
11506 * Make sure counter is updated before we check for snapshot creation.
11509 if (atomic_read(&root->will_be_snapshotted)) {
11510 btrfs_end_write_no_snapshotting(root);
11516 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11521 ret = btrfs_start_write_no_snapshotting(root);
11524 wait_var_event(&root->will_be_snapshotted,
11525 !atomic_read(&root->will_be_snapshotted));
11529 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11531 struct btrfs_fs_info *fs_info = bg->fs_info;
11533 spin_lock(&fs_info->unused_bgs_lock);
11534 if (list_empty(&bg->bg_list)) {
11535 btrfs_get_block_group(bg);
11536 trace_btrfs_add_unused_block_group(bg);
11537 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11539 spin_unlock(&fs_info->unused_bgs_lock);