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 requried,
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 int cleanup_extent_op(struct btrfs_trans_handle *trans,
2428 struct btrfs_delayed_ref_head *head)
2430 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2435 head->extent_op = NULL;
2436 if (head->must_insert_reserved) {
2437 btrfs_free_delayed_extent_op(extent_op);
2440 spin_unlock(&head->lock);
2441 ret = run_delayed_extent_op(trans, head, extent_op);
2442 btrfs_free_delayed_extent_op(extent_op);
2443 return ret ? ret : 1;
2446 static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
2447 struct btrfs_delayed_ref_head *head)
2449 struct btrfs_fs_info *fs_info = trans->fs_info;
2450 struct btrfs_delayed_ref_root *delayed_refs =
2451 &trans->transaction->delayed_refs;
2453 if (head->total_ref_mod < 0) {
2454 struct btrfs_space_info *space_info;
2458 flags = BTRFS_BLOCK_GROUP_DATA;
2459 else if (head->is_system)
2460 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2462 flags = BTRFS_BLOCK_GROUP_METADATA;
2463 space_info = __find_space_info(fs_info, flags);
2465 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2467 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2469 if (head->is_data) {
2470 spin_lock(&delayed_refs->lock);
2471 delayed_refs->pending_csums -= head->num_bytes;
2472 spin_unlock(&delayed_refs->lock);
2476 /* Also free its reserved qgroup space */
2477 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2478 head->qgroup_reserved);
2481 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2482 struct btrfs_delayed_ref_head *head)
2485 struct btrfs_fs_info *fs_info = trans->fs_info;
2486 struct btrfs_delayed_ref_root *delayed_refs;
2489 delayed_refs = &trans->transaction->delayed_refs;
2491 ret = cleanup_extent_op(trans, head);
2493 unselect_delayed_ref_head(delayed_refs, head);
2494 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2501 * Need to drop our head ref lock and re-acquire the delayed ref lock
2502 * and then re-check to make sure nobody got added.
2504 spin_unlock(&head->lock);
2505 spin_lock(&delayed_refs->lock);
2506 spin_lock(&head->lock);
2507 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2508 spin_unlock(&head->lock);
2509 spin_unlock(&delayed_refs->lock);
2512 btrfs_delete_ref_head(delayed_refs, head);
2513 spin_unlock(&head->lock);
2514 spin_unlock(&delayed_refs->lock);
2516 if (head->must_insert_reserved) {
2517 btrfs_pin_extent(fs_info, head->bytenr,
2518 head->num_bytes, 1);
2519 if (head->is_data) {
2520 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2525 cleanup_ref_head_accounting(trans, head);
2527 trace_run_delayed_ref_head(fs_info, head, 0);
2528 btrfs_delayed_ref_unlock(head);
2529 btrfs_put_delayed_ref_head(head);
2533 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2534 struct btrfs_trans_handle *trans)
2536 struct btrfs_delayed_ref_root *delayed_refs =
2537 &trans->transaction->delayed_refs;
2538 struct btrfs_delayed_ref_head *head = NULL;
2541 spin_lock(&delayed_refs->lock);
2542 head = btrfs_select_ref_head(delayed_refs);
2544 spin_unlock(&delayed_refs->lock);
2549 * Grab the lock that says we are going to process all the refs for
2552 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2553 spin_unlock(&delayed_refs->lock);
2556 * We may have dropped the spin lock to get the head mutex lock, and
2557 * that might have given someone else time to free the head. If that's
2558 * true, it has been removed from our list and we can move on.
2561 head = ERR_PTR(-EAGAIN);
2566 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2567 struct btrfs_delayed_ref_head *locked_ref,
2568 unsigned long *run_refs)
2570 struct btrfs_fs_info *fs_info = trans->fs_info;
2571 struct btrfs_delayed_ref_root *delayed_refs;
2572 struct btrfs_delayed_extent_op *extent_op;
2573 struct btrfs_delayed_ref_node *ref;
2574 int must_insert_reserved = 0;
2577 delayed_refs = &trans->transaction->delayed_refs;
2579 lockdep_assert_held(&locked_ref->mutex);
2580 lockdep_assert_held(&locked_ref->lock);
2582 while ((ref = select_delayed_ref(locked_ref))) {
2584 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2585 spin_unlock(&locked_ref->lock);
2586 unselect_delayed_ref_head(delayed_refs, locked_ref);
2592 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2593 RB_CLEAR_NODE(&ref->ref_node);
2594 if (!list_empty(&ref->add_list))
2595 list_del(&ref->add_list);
2597 * When we play the delayed ref, also correct the ref_mod on
2600 switch (ref->action) {
2601 case BTRFS_ADD_DELAYED_REF:
2602 case BTRFS_ADD_DELAYED_EXTENT:
2603 locked_ref->ref_mod -= ref->ref_mod;
2605 case BTRFS_DROP_DELAYED_REF:
2606 locked_ref->ref_mod += ref->ref_mod;
2611 atomic_dec(&delayed_refs->num_entries);
2614 * Record the must_insert_reserved flag before we drop the
2617 must_insert_reserved = locked_ref->must_insert_reserved;
2618 locked_ref->must_insert_reserved = 0;
2620 extent_op = locked_ref->extent_op;
2621 locked_ref->extent_op = NULL;
2622 spin_unlock(&locked_ref->lock);
2624 ret = run_one_delayed_ref(trans, ref, extent_op,
2625 must_insert_reserved);
2627 btrfs_free_delayed_extent_op(extent_op);
2629 unselect_delayed_ref_head(delayed_refs, locked_ref);
2630 btrfs_put_delayed_ref(ref);
2631 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2636 btrfs_put_delayed_ref(ref);
2639 spin_lock(&locked_ref->lock);
2640 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2647 * Returns 0 on success or if called with an already aborted transaction.
2648 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2650 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2653 struct btrfs_fs_info *fs_info = trans->fs_info;
2654 struct btrfs_delayed_ref_root *delayed_refs;
2655 struct btrfs_delayed_ref_head *locked_ref = NULL;
2656 ktime_t start = ktime_get();
2658 unsigned long count = 0;
2659 unsigned long actual_count = 0;
2661 delayed_refs = &trans->transaction->delayed_refs;
2664 locked_ref = btrfs_obtain_ref_head(trans);
2665 if (IS_ERR_OR_NULL(locked_ref)) {
2666 if (PTR_ERR(locked_ref) == -EAGAIN) {
2675 * We need to try and merge add/drops of the same ref since we
2676 * can run into issues with relocate dropping the implicit ref
2677 * and then it being added back again before the drop can
2678 * finish. If we merged anything we need to re-loop so we can
2680 * Or we can get node references of the same type that weren't
2681 * merged when created due to bumps in the tree mod seq, and
2682 * we need to merge them to prevent adding an inline extent
2683 * backref before dropping it (triggering a BUG_ON at
2684 * insert_inline_extent_backref()).
2686 spin_lock(&locked_ref->lock);
2687 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2689 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2691 if (ret < 0 && ret != -EAGAIN) {
2693 * Error, btrfs_run_delayed_refs_for_head already
2694 * unlocked everything so just bail out
2699 * Success, perform the usual cleanup of a processed
2702 ret = cleanup_ref_head(trans, locked_ref);
2704 /* We dropped our lock, we need to loop. */
2713 * Either success case or btrfs_run_delayed_refs_for_head
2714 * returned -EAGAIN, meaning we need to select another head
2719 } while ((nr != -1 && count < nr) || locked_ref);
2722 * We don't want to include ref heads since we can have empty ref heads
2723 * and those will drastically skew our runtime down since we just do
2724 * accounting, no actual extent tree updates.
2726 if (actual_count > 0) {
2727 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2731 * We weigh the current average higher than our current runtime
2732 * to avoid large swings in the average.
2734 spin_lock(&delayed_refs->lock);
2735 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2736 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2737 spin_unlock(&delayed_refs->lock);
2742 #ifdef SCRAMBLE_DELAYED_REFS
2744 * Normally delayed refs get processed in ascending bytenr order. This
2745 * correlates in most cases to the order added. To expose dependencies on this
2746 * order, we start to process the tree in the middle instead of the beginning
2748 static u64 find_middle(struct rb_root *root)
2750 struct rb_node *n = root->rb_node;
2751 struct btrfs_delayed_ref_node *entry;
2754 u64 first = 0, last = 0;
2758 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2759 first = entry->bytenr;
2763 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2764 last = entry->bytenr;
2769 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2770 WARN_ON(!entry->in_tree);
2772 middle = entry->bytenr;
2785 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2789 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2790 sizeof(struct btrfs_extent_inline_ref));
2791 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2792 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2795 * We don't ever fill up leaves all the way so multiply by 2 just to be
2796 * closer to what we're really going to want to use.
2798 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2802 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2803 * would require to store the csums for that many bytes.
2805 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2808 u64 num_csums_per_leaf;
2811 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2812 num_csums_per_leaf = div64_u64(csum_size,
2813 (u64)btrfs_super_csum_size(fs_info->super_copy));
2814 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2815 num_csums += num_csums_per_leaf - 1;
2816 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2820 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans)
2822 struct btrfs_fs_info *fs_info = trans->fs_info;
2823 struct btrfs_block_rsv *global_rsv;
2824 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2825 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2826 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2827 u64 num_bytes, num_dirty_bgs_bytes;
2830 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2831 num_heads = heads_to_leaves(fs_info, num_heads);
2833 num_bytes += (num_heads - 1) * fs_info->nodesize;
2835 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2837 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2839 global_rsv = &fs_info->global_block_rsv;
2842 * If we can't allocate any more chunks lets make sure we have _lots_ of
2843 * wiggle room since running delayed refs can create more delayed refs.
2845 if (global_rsv->space_info->full) {
2846 num_dirty_bgs_bytes <<= 1;
2850 spin_lock(&global_rsv->lock);
2851 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2853 spin_unlock(&global_rsv->lock);
2857 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2860 atomic_read(&trans->transaction->delayed_refs.num_entries);
2865 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2866 val = num_entries * avg_runtime;
2867 if (val >= NSEC_PER_SEC)
2869 if (val >= NSEC_PER_SEC / 2)
2872 return btrfs_check_space_for_delayed_refs(trans);
2875 struct async_delayed_refs {
2876 struct btrfs_root *root;
2881 struct completion wait;
2882 struct btrfs_work work;
2885 static inline struct async_delayed_refs *
2886 to_async_delayed_refs(struct btrfs_work *work)
2888 return container_of(work, struct async_delayed_refs, work);
2891 static void delayed_ref_async_start(struct btrfs_work *work)
2893 struct async_delayed_refs *async = to_async_delayed_refs(work);
2894 struct btrfs_trans_handle *trans;
2895 struct btrfs_fs_info *fs_info = async->root->fs_info;
2898 /* if the commit is already started, we don't need to wait here */
2899 if (btrfs_transaction_blocked(fs_info))
2902 trans = btrfs_join_transaction(async->root);
2903 if (IS_ERR(trans)) {
2904 async->error = PTR_ERR(trans);
2909 * trans->sync means that when we call end_transaction, we won't
2910 * wait on delayed refs
2914 /* Don't bother flushing if we got into a different transaction */
2915 if (trans->transid > async->transid)
2918 ret = btrfs_run_delayed_refs(trans, async->count);
2922 ret = btrfs_end_transaction(trans);
2923 if (ret && !async->error)
2927 complete(&async->wait);
2932 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2933 unsigned long count, u64 transid, int wait)
2935 struct async_delayed_refs *async;
2938 async = kmalloc(sizeof(*async), GFP_NOFS);
2942 async->root = fs_info->tree_root;
2943 async->count = count;
2945 async->transid = transid;
2950 init_completion(&async->wait);
2952 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2953 delayed_ref_async_start, NULL, NULL);
2955 btrfs_queue_work(fs_info->extent_workers, &async->work);
2958 wait_for_completion(&async->wait);
2967 * this starts processing the delayed reference count updates and
2968 * extent insertions we have queued up so far. count can be
2969 * 0, which means to process everything in the tree at the start
2970 * of the run (but not newly added entries), or it can be some target
2971 * number you'd like to process.
2973 * Returns 0 on success or if called with an aborted transaction
2974 * Returns <0 on error and aborts the transaction
2976 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2977 unsigned long count)
2979 struct btrfs_fs_info *fs_info = trans->fs_info;
2980 struct rb_node *node;
2981 struct btrfs_delayed_ref_root *delayed_refs;
2982 struct btrfs_delayed_ref_head *head;
2984 int run_all = count == (unsigned long)-1;
2986 /* We'll clean this up in btrfs_cleanup_transaction */
2990 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2993 delayed_refs = &trans->transaction->delayed_refs;
2995 count = atomic_read(&delayed_refs->num_entries) * 2;
2998 #ifdef SCRAMBLE_DELAYED_REFS
2999 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3001 ret = __btrfs_run_delayed_refs(trans, count);
3003 btrfs_abort_transaction(trans, ret);
3008 if (!list_empty(&trans->new_bgs))
3009 btrfs_create_pending_block_groups(trans);
3011 spin_lock(&delayed_refs->lock);
3012 node = rb_first_cached(&delayed_refs->href_root);
3014 spin_unlock(&delayed_refs->lock);
3017 head = rb_entry(node, struct btrfs_delayed_ref_head,
3019 refcount_inc(&head->refs);
3020 spin_unlock(&delayed_refs->lock);
3022 /* Mutex was contended, block until it's released and retry. */
3023 mutex_lock(&head->mutex);
3024 mutex_unlock(&head->mutex);
3026 btrfs_put_delayed_ref_head(head);
3034 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3035 struct btrfs_fs_info *fs_info,
3036 u64 bytenr, u64 num_bytes, u64 flags,
3037 int level, int is_data)
3039 struct btrfs_delayed_extent_op *extent_op;
3042 extent_op = btrfs_alloc_delayed_extent_op();
3046 extent_op->flags_to_set = flags;
3047 extent_op->update_flags = true;
3048 extent_op->update_key = false;
3049 extent_op->is_data = is_data ? true : false;
3050 extent_op->level = level;
3052 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3053 num_bytes, extent_op);
3055 btrfs_free_delayed_extent_op(extent_op);
3059 static noinline int check_delayed_ref(struct btrfs_root *root,
3060 struct btrfs_path *path,
3061 u64 objectid, u64 offset, u64 bytenr)
3063 struct btrfs_delayed_ref_head *head;
3064 struct btrfs_delayed_ref_node *ref;
3065 struct btrfs_delayed_data_ref *data_ref;
3066 struct btrfs_delayed_ref_root *delayed_refs;
3067 struct btrfs_transaction *cur_trans;
3068 struct rb_node *node;
3071 spin_lock(&root->fs_info->trans_lock);
3072 cur_trans = root->fs_info->running_transaction;
3074 refcount_inc(&cur_trans->use_count);
3075 spin_unlock(&root->fs_info->trans_lock);
3079 delayed_refs = &cur_trans->delayed_refs;
3080 spin_lock(&delayed_refs->lock);
3081 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3083 spin_unlock(&delayed_refs->lock);
3084 btrfs_put_transaction(cur_trans);
3088 if (!mutex_trylock(&head->mutex)) {
3089 refcount_inc(&head->refs);
3090 spin_unlock(&delayed_refs->lock);
3092 btrfs_release_path(path);
3095 * Mutex was contended, block until it's released and let
3098 mutex_lock(&head->mutex);
3099 mutex_unlock(&head->mutex);
3100 btrfs_put_delayed_ref_head(head);
3101 btrfs_put_transaction(cur_trans);
3104 spin_unlock(&delayed_refs->lock);
3106 spin_lock(&head->lock);
3108 * XXX: We should replace this with a proper search function in the
3111 for (node = rb_first_cached(&head->ref_tree); node;
3112 node = rb_next(node)) {
3113 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3114 /* If it's a shared ref we know a cross reference exists */
3115 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3120 data_ref = btrfs_delayed_node_to_data_ref(ref);
3123 * If our ref doesn't match the one we're currently looking at
3124 * then we have a cross reference.
3126 if (data_ref->root != root->root_key.objectid ||
3127 data_ref->objectid != objectid ||
3128 data_ref->offset != offset) {
3133 spin_unlock(&head->lock);
3134 mutex_unlock(&head->mutex);
3135 btrfs_put_transaction(cur_trans);
3139 static noinline int check_committed_ref(struct btrfs_root *root,
3140 struct btrfs_path *path,
3141 u64 objectid, u64 offset, u64 bytenr)
3143 struct btrfs_fs_info *fs_info = root->fs_info;
3144 struct btrfs_root *extent_root = fs_info->extent_root;
3145 struct extent_buffer *leaf;
3146 struct btrfs_extent_data_ref *ref;
3147 struct btrfs_extent_inline_ref *iref;
3148 struct btrfs_extent_item *ei;
3149 struct btrfs_key key;
3154 key.objectid = bytenr;
3155 key.offset = (u64)-1;
3156 key.type = BTRFS_EXTENT_ITEM_KEY;
3158 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3161 BUG_ON(ret == 0); /* Corruption */
3164 if (path->slots[0] == 0)
3168 leaf = path->nodes[0];
3169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3171 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3175 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3176 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3178 if (item_size != sizeof(*ei) +
3179 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3182 if (btrfs_extent_generation(leaf, ei) <=
3183 btrfs_root_last_snapshot(&root->root_item))
3186 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3188 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3189 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3192 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3193 if (btrfs_extent_refs(leaf, ei) !=
3194 btrfs_extent_data_ref_count(leaf, ref) ||
3195 btrfs_extent_data_ref_root(leaf, ref) !=
3196 root->root_key.objectid ||
3197 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3198 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3206 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3209 struct btrfs_path *path;
3212 path = btrfs_alloc_path();
3217 ret = check_committed_ref(root, path, objectid,
3219 if (ret && ret != -ENOENT)
3222 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3223 } while (ret == -EAGAIN);
3226 btrfs_free_path(path);
3227 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3232 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3233 struct btrfs_root *root,
3234 struct extent_buffer *buf,
3235 int full_backref, int inc)
3237 struct btrfs_fs_info *fs_info = root->fs_info;
3243 struct btrfs_key key;
3244 struct btrfs_file_extent_item *fi;
3248 int (*process_func)(struct btrfs_trans_handle *,
3249 struct btrfs_root *,
3250 u64, u64, u64, u64, u64, u64);
3253 if (btrfs_is_testing(fs_info))
3256 ref_root = btrfs_header_owner(buf);
3257 nritems = btrfs_header_nritems(buf);
3258 level = btrfs_header_level(buf);
3260 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3264 process_func = btrfs_inc_extent_ref;
3266 process_func = btrfs_free_extent;
3269 parent = buf->start;
3273 for (i = 0; i < nritems; i++) {
3275 btrfs_item_key_to_cpu(buf, &key, i);
3276 if (key.type != BTRFS_EXTENT_DATA_KEY)
3278 fi = btrfs_item_ptr(buf, i,
3279 struct btrfs_file_extent_item);
3280 if (btrfs_file_extent_type(buf, fi) ==
3281 BTRFS_FILE_EXTENT_INLINE)
3283 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3287 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3288 key.offset -= btrfs_file_extent_offset(buf, fi);
3289 ret = process_func(trans, root, bytenr, num_bytes,
3290 parent, ref_root, key.objectid,
3295 bytenr = btrfs_node_blockptr(buf, i);
3296 num_bytes = fs_info->nodesize;
3297 ret = process_func(trans, root, bytenr, num_bytes,
3298 parent, ref_root, level - 1, 0);
3308 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3309 struct extent_buffer *buf, int full_backref)
3311 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3314 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3315 struct extent_buffer *buf, int full_backref)
3317 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3320 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3321 struct btrfs_fs_info *fs_info,
3322 struct btrfs_path *path,
3323 struct btrfs_block_group_cache *cache)
3326 struct btrfs_root *extent_root = fs_info->extent_root;
3328 struct extent_buffer *leaf;
3330 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3337 leaf = path->nodes[0];
3338 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3339 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3340 btrfs_mark_buffer_dirty(leaf);
3342 btrfs_release_path(path);
3347 static struct btrfs_block_group_cache *
3348 next_block_group(struct btrfs_fs_info *fs_info,
3349 struct btrfs_block_group_cache *cache)
3351 struct rb_node *node;
3353 spin_lock(&fs_info->block_group_cache_lock);
3355 /* If our block group was removed, we need a full search. */
3356 if (RB_EMPTY_NODE(&cache->cache_node)) {
3357 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3359 spin_unlock(&fs_info->block_group_cache_lock);
3360 btrfs_put_block_group(cache);
3361 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3363 node = rb_next(&cache->cache_node);
3364 btrfs_put_block_group(cache);
3366 cache = rb_entry(node, struct btrfs_block_group_cache,
3368 btrfs_get_block_group(cache);
3371 spin_unlock(&fs_info->block_group_cache_lock);
3375 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3376 struct btrfs_trans_handle *trans,
3377 struct btrfs_path *path)
3379 struct btrfs_fs_info *fs_info = block_group->fs_info;
3380 struct btrfs_root *root = fs_info->tree_root;
3381 struct inode *inode = NULL;
3382 struct extent_changeset *data_reserved = NULL;
3384 int dcs = BTRFS_DC_ERROR;
3390 * If this block group is smaller than 100 megs don't bother caching the
3393 if (block_group->key.offset < (100 * SZ_1M)) {
3394 spin_lock(&block_group->lock);
3395 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3396 spin_unlock(&block_group->lock);
3403 inode = lookup_free_space_inode(fs_info, block_group, path);
3404 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3405 ret = PTR_ERR(inode);
3406 btrfs_release_path(path);
3410 if (IS_ERR(inode)) {
3414 if (block_group->ro)
3417 ret = create_free_space_inode(fs_info, trans, block_group,
3425 * We want to set the generation to 0, that way if anything goes wrong
3426 * from here on out we know not to trust this cache when we load up next
3429 BTRFS_I(inode)->generation = 0;
3430 ret = btrfs_update_inode(trans, root, inode);
3433 * So theoretically we could recover from this, simply set the
3434 * super cache generation to 0 so we know to invalidate the
3435 * cache, but then we'd have to keep track of the block groups
3436 * that fail this way so we know we _have_ to reset this cache
3437 * before the next commit or risk reading stale cache. So to
3438 * limit our exposure to horrible edge cases lets just abort the
3439 * transaction, this only happens in really bad situations
3442 btrfs_abort_transaction(trans, ret);
3447 /* We've already setup this transaction, go ahead and exit */
3448 if (block_group->cache_generation == trans->transid &&
3449 i_size_read(inode)) {
3450 dcs = BTRFS_DC_SETUP;
3454 if (i_size_read(inode) > 0) {
3455 ret = btrfs_check_trunc_cache_free_space(fs_info,
3456 &fs_info->global_block_rsv);
3460 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3465 spin_lock(&block_group->lock);
3466 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3467 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3469 * don't bother trying to write stuff out _if_
3470 * a) we're not cached,
3471 * b) we're with nospace_cache mount option,
3472 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3474 dcs = BTRFS_DC_WRITTEN;
3475 spin_unlock(&block_group->lock);
3478 spin_unlock(&block_group->lock);
3481 * We hit an ENOSPC when setting up the cache in this transaction, just
3482 * skip doing the setup, we've already cleared the cache so we're safe.
3484 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3490 * Try to preallocate enough space based on how big the block group is.
3491 * Keep in mind this has to include any pinned space which could end up
3492 * taking up quite a bit since it's not folded into the other space
3495 num_pages = div_u64(block_group->key.offset, SZ_256M);
3500 num_pages *= PAGE_SIZE;
3502 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3506 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3507 num_pages, num_pages,
3510 * Our cache requires contiguous chunks so that we don't modify a bunch
3511 * of metadata or split extents when writing the cache out, which means
3512 * we can enospc if we are heavily fragmented in addition to just normal
3513 * out of space conditions. So if we hit this just skip setting up any
3514 * other block groups for this transaction, maybe we'll unpin enough
3515 * space the next time around.
3518 dcs = BTRFS_DC_SETUP;
3519 else if (ret == -ENOSPC)
3520 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3525 btrfs_release_path(path);
3527 spin_lock(&block_group->lock);
3528 if (!ret && dcs == BTRFS_DC_SETUP)
3529 block_group->cache_generation = trans->transid;
3530 block_group->disk_cache_state = dcs;
3531 spin_unlock(&block_group->lock);
3533 extent_changeset_free(data_reserved);
3537 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3538 struct btrfs_fs_info *fs_info)
3540 struct btrfs_block_group_cache *cache, *tmp;
3541 struct btrfs_transaction *cur_trans = trans->transaction;
3542 struct btrfs_path *path;
3544 if (list_empty(&cur_trans->dirty_bgs) ||
3545 !btrfs_test_opt(fs_info, SPACE_CACHE))
3548 path = btrfs_alloc_path();
3552 /* Could add new block groups, use _safe just in case */
3553 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3555 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3556 cache_save_setup(cache, trans, path);
3559 btrfs_free_path(path);
3564 * transaction commit does final block group cache writeback during a
3565 * critical section where nothing is allowed to change the FS. This is
3566 * required in order for the cache to actually match the block group,
3567 * but can introduce a lot of latency into the commit.
3569 * So, btrfs_start_dirty_block_groups is here to kick off block group
3570 * cache IO. There's a chance we'll have to redo some of it if the
3571 * block group changes again during the commit, but it greatly reduces
3572 * the commit latency by getting rid of the easy block groups while
3573 * we're still allowing others to join the commit.
3575 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3577 struct btrfs_fs_info *fs_info = trans->fs_info;
3578 struct btrfs_block_group_cache *cache;
3579 struct btrfs_transaction *cur_trans = trans->transaction;
3582 struct btrfs_path *path = NULL;
3584 struct list_head *io = &cur_trans->io_bgs;
3585 int num_started = 0;
3588 spin_lock(&cur_trans->dirty_bgs_lock);
3589 if (list_empty(&cur_trans->dirty_bgs)) {
3590 spin_unlock(&cur_trans->dirty_bgs_lock);
3593 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3594 spin_unlock(&cur_trans->dirty_bgs_lock);
3598 * make sure all the block groups on our dirty list actually
3601 btrfs_create_pending_block_groups(trans);
3604 path = btrfs_alloc_path();
3610 * cache_write_mutex is here only to save us from balance or automatic
3611 * removal of empty block groups deleting this block group while we are
3612 * writing out the cache
3614 mutex_lock(&trans->transaction->cache_write_mutex);
3615 while (!list_empty(&dirty)) {
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);
3689 spin_unlock(&cur_trans->dirty_bgs_lock);
3691 btrfs_abort_transaction(trans, ret);
3695 /* if its not on the io list, we need to put the block group */
3697 btrfs_put_block_group(cache);
3703 * Avoid blocking other tasks for too long. It might even save
3704 * us from writing caches for block groups that are going to be
3707 mutex_unlock(&trans->transaction->cache_write_mutex);
3708 mutex_lock(&trans->transaction->cache_write_mutex);
3710 mutex_unlock(&trans->transaction->cache_write_mutex);
3713 * go through delayed refs for all the stuff we've just kicked off
3714 * and then loop back (just once)
3716 ret = btrfs_run_delayed_refs(trans, 0);
3717 if (!ret && loops == 0) {
3719 spin_lock(&cur_trans->dirty_bgs_lock);
3720 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3722 * dirty_bgs_lock protects us from concurrent block group
3723 * deletes too (not just cache_write_mutex).
3725 if (!list_empty(&dirty)) {
3726 spin_unlock(&cur_trans->dirty_bgs_lock);
3729 spin_unlock(&cur_trans->dirty_bgs_lock);
3730 } else if (ret < 0) {
3731 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3734 btrfs_free_path(path);
3738 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3739 struct btrfs_fs_info *fs_info)
3741 struct btrfs_block_group_cache *cache;
3742 struct btrfs_transaction *cur_trans = trans->transaction;
3745 struct btrfs_path *path;
3746 struct list_head *io = &cur_trans->io_bgs;
3747 int num_started = 0;
3749 path = btrfs_alloc_path();
3754 * Even though we are in the critical section of the transaction commit,
3755 * we can still have concurrent tasks adding elements to this
3756 * transaction's list of dirty block groups. These tasks correspond to
3757 * endio free space workers started when writeback finishes for a
3758 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3759 * allocate new block groups as a result of COWing nodes of the root
3760 * tree when updating the free space inode. The writeback for the space
3761 * caches is triggered by an earlier call to
3762 * btrfs_start_dirty_block_groups() and iterations of the following
3764 * Also we want to do the cache_save_setup first and then run the
3765 * delayed refs to make sure we have the best chance at doing this all
3768 spin_lock(&cur_trans->dirty_bgs_lock);
3769 while (!list_empty(&cur_trans->dirty_bgs)) {
3770 cache = list_first_entry(&cur_trans->dirty_bgs,
3771 struct btrfs_block_group_cache,
3775 * this can happen if cache_save_setup re-dirties a block
3776 * group that is already under IO. Just wait for it to
3777 * finish and then do it all again
3779 if (!list_empty(&cache->io_list)) {
3780 spin_unlock(&cur_trans->dirty_bgs_lock);
3781 list_del_init(&cache->io_list);
3782 btrfs_wait_cache_io(trans, cache, path);
3783 btrfs_put_block_group(cache);
3784 spin_lock(&cur_trans->dirty_bgs_lock);
3788 * don't remove from the dirty list until after we've waited
3791 list_del_init(&cache->dirty_list);
3792 spin_unlock(&cur_trans->dirty_bgs_lock);
3795 cache_save_setup(cache, trans, path);
3798 ret = btrfs_run_delayed_refs(trans,
3799 (unsigned long) -1);
3801 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3802 cache->io_ctl.inode = NULL;
3803 ret = btrfs_write_out_cache(fs_info, trans,
3805 if (ret == 0 && cache->io_ctl.inode) {
3808 list_add_tail(&cache->io_list, io);
3811 * if we failed to write the cache, the
3812 * generation will be bad and life goes on
3818 ret = write_one_cache_group(trans, fs_info,
3821 * One of the free space endio workers might have
3822 * created a new block group while updating a free space
3823 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3824 * and hasn't released its transaction handle yet, in
3825 * which case the new block group is still attached to
3826 * its transaction handle and its creation has not
3827 * finished yet (no block group item in the extent tree
3828 * yet, etc). If this is the case, wait for all free
3829 * space endio workers to finish and retry. This is a
3830 * a very rare case so no need for a more efficient and
3833 if (ret == -ENOENT) {
3834 wait_event(cur_trans->writer_wait,
3835 atomic_read(&cur_trans->num_writers) == 1);
3836 ret = write_one_cache_group(trans, fs_info,
3840 btrfs_abort_transaction(trans, ret);
3843 /* if its not on the io list, we need to put the block group */
3845 btrfs_put_block_group(cache);
3846 spin_lock(&cur_trans->dirty_bgs_lock);
3848 spin_unlock(&cur_trans->dirty_bgs_lock);
3851 * Refer to the definition of io_bgs member for details why it's safe
3852 * to use it without any locking
3854 while (!list_empty(io)) {
3855 cache = list_first_entry(io, struct btrfs_block_group_cache,
3857 list_del_init(&cache->io_list);
3858 btrfs_wait_cache_io(trans, cache, path);
3859 btrfs_put_block_group(cache);
3862 btrfs_free_path(path);
3866 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3868 struct btrfs_block_group_cache *block_group;
3871 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3872 if (!block_group || block_group->ro)
3875 btrfs_put_block_group(block_group);
3879 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3881 struct btrfs_block_group_cache *bg;
3884 bg = btrfs_lookup_block_group(fs_info, bytenr);
3888 spin_lock(&bg->lock);
3892 atomic_inc(&bg->nocow_writers);
3893 spin_unlock(&bg->lock);
3895 /* no put on block group, done by btrfs_dec_nocow_writers */
3897 btrfs_put_block_group(bg);
3903 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3905 struct btrfs_block_group_cache *bg;
3907 bg = btrfs_lookup_block_group(fs_info, bytenr);
3909 if (atomic_dec_and_test(&bg->nocow_writers))
3910 wake_up_var(&bg->nocow_writers);
3912 * Once for our lookup and once for the lookup done by a previous call
3913 * to btrfs_inc_nocow_writers()
3915 btrfs_put_block_group(bg);
3916 btrfs_put_block_group(bg);
3919 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3921 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3924 static const char *alloc_name(u64 flags)
3927 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3929 case BTRFS_BLOCK_GROUP_METADATA:
3931 case BTRFS_BLOCK_GROUP_DATA:
3933 case BTRFS_BLOCK_GROUP_SYSTEM:
3937 return "invalid-combination";
3941 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3944 struct btrfs_space_info *space_info;
3948 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3952 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3959 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3960 INIT_LIST_HEAD(&space_info->block_groups[i]);
3961 init_rwsem(&space_info->groups_sem);
3962 spin_lock_init(&space_info->lock);
3963 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3964 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3965 init_waitqueue_head(&space_info->wait);
3966 INIT_LIST_HEAD(&space_info->ro_bgs);
3967 INIT_LIST_HEAD(&space_info->tickets);
3968 INIT_LIST_HEAD(&space_info->priority_tickets);
3970 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3971 info->space_info_kobj, "%s",
3972 alloc_name(space_info->flags));
3974 percpu_counter_destroy(&space_info->total_bytes_pinned);
3979 list_add_rcu(&space_info->list, &info->space_info);
3980 if (flags & BTRFS_BLOCK_GROUP_DATA)
3981 info->data_sinfo = space_info;
3986 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3987 u64 total_bytes, u64 bytes_used,
3989 struct btrfs_space_info **space_info)
3991 struct btrfs_space_info *found;
3994 factor = btrfs_bg_type_to_factor(flags);
3996 found = __find_space_info(info, flags);
3998 spin_lock(&found->lock);
3999 found->total_bytes += total_bytes;
4000 found->disk_total += total_bytes * factor;
4001 found->bytes_used += bytes_used;
4002 found->disk_used += bytes_used * factor;
4003 found->bytes_readonly += bytes_readonly;
4004 if (total_bytes > 0)
4006 space_info_add_new_bytes(info, found, total_bytes -
4007 bytes_used - bytes_readonly);
4008 spin_unlock(&found->lock);
4009 *space_info = found;
4012 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4014 u64 extra_flags = chunk_to_extended(flags) &
4015 BTRFS_EXTENDED_PROFILE_MASK;
4017 write_seqlock(&fs_info->profiles_lock);
4018 if (flags & BTRFS_BLOCK_GROUP_DATA)
4019 fs_info->avail_data_alloc_bits |= extra_flags;
4020 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4021 fs_info->avail_metadata_alloc_bits |= extra_flags;
4022 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4023 fs_info->avail_system_alloc_bits |= extra_flags;
4024 write_sequnlock(&fs_info->profiles_lock);
4028 * returns target flags in extended format or 0 if restripe for this
4029 * chunk_type is not in progress
4031 * should be called with balance_lock held
4033 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4035 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4041 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4042 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4043 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4044 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4045 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4046 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4047 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4048 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4049 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4056 * @flags: available profiles in extended format (see ctree.h)
4058 * Returns reduced profile in chunk format. If profile changing is in
4059 * progress (either running or paused) picks the target profile (if it's
4060 * already available), otherwise falls back to plain reducing.
4062 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4064 u64 num_devices = fs_info->fs_devices->rw_devices;
4070 * see if restripe for this chunk_type is in progress, if so
4071 * try to reduce to the target profile
4073 spin_lock(&fs_info->balance_lock);
4074 target = get_restripe_target(fs_info, flags);
4076 /* pick target profile only if it's already available */
4077 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4078 spin_unlock(&fs_info->balance_lock);
4079 return extended_to_chunk(target);
4082 spin_unlock(&fs_info->balance_lock);
4084 /* First, mask out the RAID levels which aren't possible */
4085 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4086 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4087 allowed |= btrfs_raid_array[raid_type].bg_flag;
4091 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4092 allowed = BTRFS_BLOCK_GROUP_RAID6;
4093 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4094 allowed = BTRFS_BLOCK_GROUP_RAID5;
4095 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4096 allowed = BTRFS_BLOCK_GROUP_RAID10;
4097 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4098 allowed = BTRFS_BLOCK_GROUP_RAID1;
4099 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4100 allowed = BTRFS_BLOCK_GROUP_RAID0;
4102 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4104 return extended_to_chunk(flags | allowed);
4107 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4114 seq = read_seqbegin(&fs_info->profiles_lock);
4116 if (flags & BTRFS_BLOCK_GROUP_DATA)
4117 flags |= fs_info->avail_data_alloc_bits;
4118 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4119 flags |= fs_info->avail_system_alloc_bits;
4120 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4121 flags |= fs_info->avail_metadata_alloc_bits;
4122 } while (read_seqretry(&fs_info->profiles_lock, seq));
4124 return btrfs_reduce_alloc_profile(fs_info, flags);
4127 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4129 struct btrfs_fs_info *fs_info = root->fs_info;
4134 flags = BTRFS_BLOCK_GROUP_DATA;
4135 else if (root == fs_info->chunk_root)
4136 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4138 flags = BTRFS_BLOCK_GROUP_METADATA;
4140 ret = get_alloc_profile(fs_info, flags);
4144 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4146 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4149 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4151 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4154 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4156 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4159 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4160 bool may_use_included)
4163 return s_info->bytes_used + s_info->bytes_reserved +
4164 s_info->bytes_pinned + s_info->bytes_readonly +
4165 (may_use_included ? s_info->bytes_may_use : 0);
4168 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4170 struct btrfs_root *root = inode->root;
4171 struct btrfs_fs_info *fs_info = root->fs_info;
4172 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4175 int need_commit = 2;
4176 int have_pinned_space;
4178 /* make sure bytes are sectorsize aligned */
4179 bytes = ALIGN(bytes, fs_info->sectorsize);
4181 if (btrfs_is_free_space_inode(inode)) {
4183 ASSERT(current->journal_info);
4187 /* make sure we have enough space to handle the data first */
4188 spin_lock(&data_sinfo->lock);
4189 used = btrfs_space_info_used(data_sinfo, true);
4191 if (used + bytes > data_sinfo->total_bytes) {
4192 struct btrfs_trans_handle *trans;
4195 * if we don't have enough free bytes in this space then we need
4196 * to alloc a new chunk.
4198 if (!data_sinfo->full) {
4201 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4202 spin_unlock(&data_sinfo->lock);
4204 alloc_target = btrfs_data_alloc_profile(fs_info);
4206 * It is ugly that we don't call nolock join
4207 * transaction for the free space inode case here.
4208 * But it is safe because we only do the data space
4209 * reservation for the free space cache in the
4210 * transaction context, the common join transaction
4211 * just increase the counter of the current transaction
4212 * handler, doesn't try to acquire the trans_lock of
4215 trans = btrfs_join_transaction(root);
4217 return PTR_ERR(trans);
4219 ret = do_chunk_alloc(trans, alloc_target,
4220 CHUNK_ALLOC_NO_FORCE);
4221 btrfs_end_transaction(trans);
4226 have_pinned_space = 1;
4235 * If we don't have enough pinned space to deal with this
4236 * allocation, and no removed chunk in current transaction,
4237 * don't bother committing the transaction.
4239 have_pinned_space = __percpu_counter_compare(
4240 &data_sinfo->total_bytes_pinned,
4241 used + bytes - data_sinfo->total_bytes,
4242 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4243 spin_unlock(&data_sinfo->lock);
4245 /* commit the current transaction and try again */
4250 if (need_commit > 0) {
4251 btrfs_start_delalloc_roots(fs_info, -1);
4252 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4256 trans = btrfs_join_transaction(root);
4258 return PTR_ERR(trans);
4259 if (have_pinned_space >= 0 ||
4260 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4261 &trans->transaction->flags) ||
4263 ret = btrfs_commit_transaction(trans);
4267 * The cleaner kthread might still be doing iput
4268 * operations. Wait for it to finish so that
4269 * more space is released.
4271 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4272 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4275 btrfs_end_transaction(trans);
4279 trace_btrfs_space_reservation(fs_info,
4280 "space_info:enospc",
4281 data_sinfo->flags, bytes, 1);
4284 update_bytes_may_use(data_sinfo, bytes);
4285 trace_btrfs_space_reservation(fs_info, "space_info",
4286 data_sinfo->flags, bytes, 1);
4287 spin_unlock(&data_sinfo->lock);
4292 int btrfs_check_data_free_space(struct inode *inode,
4293 struct extent_changeset **reserved, u64 start, u64 len)
4295 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4298 /* align the range */
4299 len = round_up(start + len, fs_info->sectorsize) -
4300 round_down(start, fs_info->sectorsize);
4301 start = round_down(start, fs_info->sectorsize);
4303 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4307 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4308 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4310 btrfs_free_reserved_data_space_noquota(inode, start, len);
4317 * Called if we need to clear a data reservation for this inode
4318 * Normally in a error case.
4320 * This one will *NOT* use accurate qgroup reserved space API, just for case
4321 * which we can't sleep and is sure it won't affect qgroup reserved space.
4322 * Like clear_bit_hook().
4324 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4327 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4328 struct btrfs_space_info *data_sinfo;
4330 /* Make sure the range is aligned to sectorsize */
4331 len = round_up(start + len, fs_info->sectorsize) -
4332 round_down(start, fs_info->sectorsize);
4333 start = round_down(start, fs_info->sectorsize);
4335 data_sinfo = fs_info->data_sinfo;
4336 spin_lock(&data_sinfo->lock);
4337 update_bytes_may_use(data_sinfo, -len);
4338 trace_btrfs_space_reservation(fs_info, "space_info",
4339 data_sinfo->flags, len, 0);
4340 spin_unlock(&data_sinfo->lock);
4344 * Called if we need to clear a data reservation for this inode
4345 * Normally in a error case.
4347 * This one will handle the per-inode data rsv map for accurate reserved
4350 void btrfs_free_reserved_data_space(struct inode *inode,
4351 struct extent_changeset *reserved, u64 start, u64 len)
4353 struct btrfs_root *root = BTRFS_I(inode)->root;
4355 /* Make sure the range is aligned to sectorsize */
4356 len = round_up(start + len, root->fs_info->sectorsize) -
4357 round_down(start, root->fs_info->sectorsize);
4358 start = round_down(start, root->fs_info->sectorsize);
4360 btrfs_free_reserved_data_space_noquota(inode, start, len);
4361 btrfs_qgroup_free_data(inode, reserved, start, len);
4364 static void force_metadata_allocation(struct btrfs_fs_info *info)
4366 struct list_head *head = &info->space_info;
4367 struct btrfs_space_info *found;
4370 list_for_each_entry_rcu(found, head, list) {
4371 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4372 found->force_alloc = CHUNK_ALLOC_FORCE;
4377 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4379 return (global->size << 1);
4382 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4383 struct btrfs_space_info *sinfo, int force)
4385 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4386 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4389 if (force == CHUNK_ALLOC_FORCE)
4393 * We need to take into account the global rsv because for all intents
4394 * and purposes it's used space. Don't worry about locking the
4395 * global_rsv, it doesn't change except when the transaction commits.
4397 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4398 bytes_used += calc_global_rsv_need_space(global_rsv);
4401 * in limited mode, we want to have some free space up to
4402 * about 1% of the FS size.
4404 if (force == CHUNK_ALLOC_LIMITED) {
4405 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4406 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4408 if (sinfo->total_bytes - bytes_used < thresh)
4412 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4417 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4421 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4422 BTRFS_BLOCK_GROUP_RAID0 |
4423 BTRFS_BLOCK_GROUP_RAID5 |
4424 BTRFS_BLOCK_GROUP_RAID6))
4425 num_dev = fs_info->fs_devices->rw_devices;
4426 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4429 num_dev = 1; /* DUP or single */
4435 * If @is_allocation is true, reserve space in the system space info necessary
4436 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4439 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4441 struct btrfs_fs_info *fs_info = trans->fs_info;
4442 struct btrfs_space_info *info;
4449 * Needed because we can end up allocating a system chunk and for an
4450 * atomic and race free space reservation in the chunk block reserve.
4452 lockdep_assert_held(&fs_info->chunk_mutex);
4454 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4455 spin_lock(&info->lock);
4456 left = info->total_bytes - btrfs_space_info_used(info, true);
4457 spin_unlock(&info->lock);
4459 num_devs = get_profile_num_devs(fs_info, type);
4461 /* num_devs device items to update and 1 chunk item to add or remove */
4462 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4463 btrfs_calc_trans_metadata_size(fs_info, 1);
4465 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4466 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4467 left, thresh, type);
4468 dump_space_info(fs_info, info, 0, 0);
4471 if (left < thresh) {
4472 u64 flags = btrfs_system_alloc_profile(fs_info);
4475 * Ignore failure to create system chunk. We might end up not
4476 * needing it, as we might not need to COW all nodes/leafs from
4477 * the paths we visit in the chunk tree (they were already COWed
4478 * or created in the current transaction for example).
4480 ret = btrfs_alloc_chunk(trans, flags);
4484 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4485 &fs_info->chunk_block_rsv,
4486 thresh, BTRFS_RESERVE_NO_FLUSH);
4488 trans->chunk_bytes_reserved += thresh;
4493 * If force is CHUNK_ALLOC_FORCE:
4494 * - return 1 if it successfully allocates a chunk,
4495 * - return errors including -ENOSPC otherwise.
4496 * If force is NOT CHUNK_ALLOC_FORCE:
4497 * - return 0 if it doesn't need to allocate a new chunk,
4498 * - return 1 if it successfully allocates a chunk,
4499 * - return errors including -ENOSPC otherwise.
4501 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4504 struct btrfs_fs_info *fs_info = trans->fs_info;
4505 struct btrfs_space_info *space_info;
4506 bool wait_for_alloc = false;
4507 bool should_alloc = false;
4510 /* Don't re-enter if we're already allocating a chunk */
4511 if (trans->allocating_chunk)
4514 space_info = __find_space_info(fs_info, flags);
4518 spin_lock(&space_info->lock);
4519 if (force < space_info->force_alloc)
4520 force = space_info->force_alloc;
4521 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4522 if (space_info->full) {
4523 /* No more free physical space */
4528 spin_unlock(&space_info->lock);
4530 } else if (!should_alloc) {
4531 spin_unlock(&space_info->lock);
4533 } else if (space_info->chunk_alloc) {
4535 * Someone is already allocating, so we need to block
4536 * until this someone is finished and then loop to
4537 * recheck if we should continue with our allocation
4540 wait_for_alloc = true;
4541 spin_unlock(&space_info->lock);
4542 mutex_lock(&fs_info->chunk_mutex);
4543 mutex_unlock(&fs_info->chunk_mutex);
4545 /* Proceed with allocation */
4546 space_info->chunk_alloc = 1;
4547 wait_for_alloc = false;
4548 spin_unlock(&space_info->lock);
4552 } while (wait_for_alloc);
4554 mutex_lock(&fs_info->chunk_mutex);
4555 trans->allocating_chunk = true;
4558 * If we have mixed data/metadata chunks we want to make sure we keep
4559 * allocating mixed chunks instead of individual chunks.
4561 if (btrfs_mixed_space_info(space_info))
4562 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4565 * if we're doing a data chunk, go ahead and make sure that
4566 * we keep a reasonable number of metadata chunks allocated in the
4569 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4570 fs_info->data_chunk_allocations++;
4571 if (!(fs_info->data_chunk_allocations %
4572 fs_info->metadata_ratio))
4573 force_metadata_allocation(fs_info);
4577 * Check if we have enough space in SYSTEM chunk because we may need
4578 * to update devices.
4580 check_system_chunk(trans, flags);
4582 ret = btrfs_alloc_chunk(trans, flags);
4583 trans->allocating_chunk = false;
4585 spin_lock(&space_info->lock);
4588 space_info->full = 1;
4593 space_info->max_extent_size = 0;
4596 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4598 space_info->chunk_alloc = 0;
4599 spin_unlock(&space_info->lock);
4600 mutex_unlock(&fs_info->chunk_mutex);
4602 * When we allocate a new chunk we reserve space in the chunk block
4603 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4604 * add new nodes/leafs to it if we end up needing to do it when
4605 * inserting the chunk item and updating device items as part of the
4606 * second phase of chunk allocation, performed by
4607 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4608 * large number of new block groups to create in our transaction
4609 * handle's new_bgs list to avoid exhausting the chunk block reserve
4610 * in extreme cases - like having a single transaction create many new
4611 * block groups when starting to write out the free space caches of all
4612 * the block groups that were made dirty during the lifetime of the
4615 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4616 btrfs_create_pending_block_groups(trans);
4621 static int can_overcommit(struct btrfs_fs_info *fs_info,
4622 struct btrfs_space_info *space_info, u64 bytes,
4623 enum btrfs_reserve_flush_enum flush,
4626 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4633 /* Don't overcommit when in mixed mode. */
4634 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4638 profile = btrfs_system_alloc_profile(fs_info);
4640 profile = btrfs_metadata_alloc_profile(fs_info);
4642 used = btrfs_space_info_used(space_info, false);
4645 * We only want to allow over committing if we have lots of actual space
4646 * free, but if we don't have enough space to handle the global reserve
4647 * space then we could end up having a real enospc problem when trying
4648 * to allocate a chunk or some other such important allocation.
4650 spin_lock(&global_rsv->lock);
4651 space_size = calc_global_rsv_need_space(global_rsv);
4652 spin_unlock(&global_rsv->lock);
4653 if (used + space_size >= space_info->total_bytes)
4656 used += space_info->bytes_may_use;
4658 avail = atomic64_read(&fs_info->free_chunk_space);
4661 * If we have dup, raid1 or raid10 then only half of the free
4662 * space is actually useable. For raid56, the space info used
4663 * doesn't include the parity drive, so we don't have to
4666 factor = btrfs_bg_type_to_factor(profile);
4667 avail = div_u64(avail, factor);
4670 * If we aren't flushing all things, let us overcommit up to
4671 * 1/2th of the space. If we can flush, don't let us overcommit
4672 * too much, let it overcommit up to 1/8 of the space.
4674 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4679 if (used + bytes < space_info->total_bytes + avail)
4684 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4685 unsigned long nr_pages, int nr_items)
4687 struct super_block *sb = fs_info->sb;
4689 if (down_read_trylock(&sb->s_umount)) {
4690 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4691 up_read(&sb->s_umount);
4694 * We needn't worry the filesystem going from r/w to r/o though
4695 * we don't acquire ->s_umount mutex, because the filesystem
4696 * should guarantee the delalloc inodes list be empty after
4697 * the filesystem is readonly(all dirty pages are written to
4700 btrfs_start_delalloc_roots(fs_info, nr_items);
4701 if (!current->journal_info)
4702 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4706 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4712 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4713 nr = div64_u64(to_reclaim, bytes);
4719 #define EXTENT_SIZE_PER_ITEM SZ_256K
4722 * shrink metadata reservation for delalloc
4724 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4725 u64 orig, bool wait_ordered)
4727 struct btrfs_space_info *space_info;
4728 struct btrfs_trans_handle *trans;
4733 unsigned long nr_pages;
4736 /* Calc the number of the pages we need flush for space reservation */
4737 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4738 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4740 trans = (struct btrfs_trans_handle *)current->journal_info;
4741 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4743 delalloc_bytes = percpu_counter_sum_positive(
4744 &fs_info->delalloc_bytes);
4745 if (delalloc_bytes == 0) {
4749 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4754 while (delalloc_bytes && loops < 3) {
4755 max_reclaim = min(delalloc_bytes, to_reclaim);
4756 nr_pages = max_reclaim >> PAGE_SHIFT;
4757 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4759 * We need to wait for the async pages to actually start before
4762 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4766 if (max_reclaim <= nr_pages)
4769 max_reclaim -= nr_pages;
4771 wait_event(fs_info->async_submit_wait,
4772 atomic_read(&fs_info->async_delalloc_pages) <=
4775 spin_lock(&space_info->lock);
4776 if (list_empty(&space_info->tickets) &&
4777 list_empty(&space_info->priority_tickets)) {
4778 spin_unlock(&space_info->lock);
4781 spin_unlock(&space_info->lock);
4784 if (wait_ordered && !trans) {
4785 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4787 time_left = schedule_timeout_killable(1);
4791 delalloc_bytes = percpu_counter_sum_positive(
4792 &fs_info->delalloc_bytes);
4796 struct reserve_ticket {
4799 struct list_head list;
4800 wait_queue_head_t wait;
4804 * maybe_commit_transaction - possibly commit the transaction if its ok to
4805 * @root - the root we're allocating for
4806 * @bytes - the number of bytes we want to reserve
4807 * @force - force the commit
4809 * This will check to make sure that committing the transaction will actually
4810 * get us somewhere and then commit the transaction if it does. Otherwise it
4811 * will return -ENOSPC.
4813 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4814 struct btrfs_space_info *space_info)
4816 struct reserve_ticket *ticket = NULL;
4817 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4818 struct btrfs_trans_handle *trans;
4821 trans = (struct btrfs_trans_handle *)current->journal_info;
4825 spin_lock(&space_info->lock);
4826 if (!list_empty(&space_info->priority_tickets))
4827 ticket = list_first_entry(&space_info->priority_tickets,
4828 struct reserve_ticket, list);
4829 else if (!list_empty(&space_info->tickets))
4830 ticket = list_first_entry(&space_info->tickets,
4831 struct reserve_ticket, list);
4832 bytes = (ticket) ? ticket->bytes : 0;
4833 spin_unlock(&space_info->lock);
4838 /* See if there is enough pinned space to make this reservation */
4839 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4841 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4845 * See if there is some space in the delayed insertion reservation for
4848 if (space_info != delayed_rsv->space_info)
4851 spin_lock(&delayed_rsv->lock);
4852 if (delayed_rsv->size > bytes)
4855 bytes -= delayed_rsv->size;
4856 spin_unlock(&delayed_rsv->lock);
4858 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4860 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4865 trans = btrfs_join_transaction(fs_info->extent_root);
4869 return btrfs_commit_transaction(trans);
4873 * Try to flush some data based on policy set by @state. This is only advisory
4874 * and may fail for various reasons. The caller is supposed to examine the
4875 * state of @space_info to detect the outcome.
4877 static void flush_space(struct btrfs_fs_info *fs_info,
4878 struct btrfs_space_info *space_info, u64 num_bytes,
4881 struct btrfs_root *root = fs_info->extent_root;
4882 struct btrfs_trans_handle *trans;
4887 case FLUSH_DELAYED_ITEMS_NR:
4888 case FLUSH_DELAYED_ITEMS:
4889 if (state == FLUSH_DELAYED_ITEMS_NR)
4890 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4894 trans = btrfs_join_transaction(root);
4895 if (IS_ERR(trans)) {
4896 ret = PTR_ERR(trans);
4899 ret = btrfs_run_delayed_items_nr(trans, nr);
4900 btrfs_end_transaction(trans);
4902 case FLUSH_DELALLOC:
4903 case FLUSH_DELALLOC_WAIT:
4904 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4905 state == FLUSH_DELALLOC_WAIT);
4908 trans = btrfs_join_transaction(root);
4909 if (IS_ERR(trans)) {
4910 ret = PTR_ERR(trans);
4913 ret = do_chunk_alloc(trans,
4914 btrfs_metadata_alloc_profile(fs_info),
4915 CHUNK_ALLOC_NO_FORCE);
4916 btrfs_end_transaction(trans);
4917 if (ret > 0 || ret == -ENOSPC)
4921 ret = may_commit_transaction(fs_info, space_info);
4928 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4934 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4935 struct btrfs_space_info *space_info,
4938 struct reserve_ticket *ticket;
4943 list_for_each_entry(ticket, &space_info->tickets, list)
4944 to_reclaim += ticket->bytes;
4945 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4946 to_reclaim += ticket->bytes;
4950 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4951 if (can_overcommit(fs_info, space_info, to_reclaim,
4952 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4955 used = btrfs_space_info_used(space_info, true);
4957 if (can_overcommit(fs_info, space_info, SZ_1M,
4958 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4959 expected = div_factor_fine(space_info->total_bytes, 95);
4961 expected = div_factor_fine(space_info->total_bytes, 90);
4963 if (used > expected)
4964 to_reclaim = used - expected;
4967 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4968 space_info->bytes_reserved);
4972 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
4973 struct btrfs_space_info *space_info,
4974 u64 used, bool system_chunk)
4976 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4978 /* If we're just plain full then async reclaim just slows us down. */
4979 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4982 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4986 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4987 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4990 static void wake_all_tickets(struct list_head *head)
4992 struct reserve_ticket *ticket;
4994 while (!list_empty(head)) {
4995 ticket = list_first_entry(head, struct reserve_ticket, list);
4996 list_del_init(&ticket->list);
4997 ticket->error = -ENOSPC;
4998 wake_up(&ticket->wait);
5003 * This is for normal flushers, we can wait all goddamned day if we want to. We
5004 * will loop and continuously try to flush as long as we are making progress.
5005 * We count progress as clearing off tickets each time we have to loop.
5007 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5009 struct btrfs_fs_info *fs_info;
5010 struct btrfs_space_info *space_info;
5013 int commit_cycles = 0;
5014 u64 last_tickets_id;
5016 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5017 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5019 spin_lock(&space_info->lock);
5020 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5023 space_info->flush = 0;
5024 spin_unlock(&space_info->lock);
5027 last_tickets_id = space_info->tickets_id;
5028 spin_unlock(&space_info->lock);
5030 flush_state = FLUSH_DELAYED_ITEMS_NR;
5032 flush_space(fs_info, space_info, to_reclaim, flush_state);
5033 spin_lock(&space_info->lock);
5034 if (list_empty(&space_info->tickets)) {
5035 space_info->flush = 0;
5036 spin_unlock(&space_info->lock);
5039 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5042 if (last_tickets_id == space_info->tickets_id) {
5045 last_tickets_id = space_info->tickets_id;
5046 flush_state = FLUSH_DELAYED_ITEMS_NR;
5051 if (flush_state > COMMIT_TRANS) {
5053 if (commit_cycles > 2) {
5054 wake_all_tickets(&space_info->tickets);
5055 space_info->flush = 0;
5057 flush_state = FLUSH_DELAYED_ITEMS_NR;
5060 spin_unlock(&space_info->lock);
5061 } while (flush_state <= COMMIT_TRANS);
5064 void btrfs_init_async_reclaim_work(struct work_struct *work)
5066 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5069 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5070 struct btrfs_space_info *space_info,
5071 struct reserve_ticket *ticket)
5074 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5076 spin_lock(&space_info->lock);
5077 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5080 spin_unlock(&space_info->lock);
5083 spin_unlock(&space_info->lock);
5086 flush_space(fs_info, space_info, to_reclaim, flush_state);
5088 spin_lock(&space_info->lock);
5089 if (ticket->bytes == 0) {
5090 spin_unlock(&space_info->lock);
5093 spin_unlock(&space_info->lock);
5096 * Priority flushers can't wait on delalloc without
5099 if (flush_state == FLUSH_DELALLOC ||
5100 flush_state == FLUSH_DELALLOC_WAIT)
5101 flush_state = ALLOC_CHUNK;
5102 } while (flush_state < COMMIT_TRANS);
5105 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5106 struct btrfs_space_info *space_info,
5107 struct reserve_ticket *ticket, u64 orig_bytes)
5113 spin_lock(&space_info->lock);
5114 while (ticket->bytes > 0 && ticket->error == 0) {
5115 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5120 spin_unlock(&space_info->lock);
5124 finish_wait(&ticket->wait, &wait);
5125 spin_lock(&space_info->lock);
5128 ret = ticket->error;
5129 if (!list_empty(&ticket->list))
5130 list_del_init(&ticket->list);
5131 if (ticket->bytes && ticket->bytes < orig_bytes) {
5132 u64 num_bytes = orig_bytes - ticket->bytes;
5133 update_bytes_may_use(space_info, -num_bytes);
5134 trace_btrfs_space_reservation(fs_info, "space_info",
5135 space_info->flags, num_bytes, 0);
5137 spin_unlock(&space_info->lock);
5143 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5144 * @root - the root we're allocating for
5145 * @space_info - the space info we want to allocate from
5146 * @orig_bytes - the number of bytes we want
5147 * @flush - whether or not we can flush to make our reservation
5149 * This will reserve orig_bytes number of bytes from the space info associated
5150 * with the block_rsv. If there is not enough space it will make an attempt to
5151 * flush out space to make room. It will do this by flushing delalloc if
5152 * possible or committing the transaction. If flush is 0 then no attempts to
5153 * regain reservations will be made and this will fail if there is not enough
5156 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5157 struct btrfs_space_info *space_info,
5159 enum btrfs_reserve_flush_enum flush,
5162 struct reserve_ticket ticket;
5167 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5169 spin_lock(&space_info->lock);
5171 used = btrfs_space_info_used(space_info, true);
5174 * If we have enough space then hooray, make our reservation and carry
5175 * on. If not see if we can overcommit, and if we can, hooray carry on.
5176 * If not things get more complicated.
5178 if (used + orig_bytes <= space_info->total_bytes) {
5179 update_bytes_may_use(space_info, orig_bytes);
5180 trace_btrfs_space_reservation(fs_info, "space_info",
5181 space_info->flags, orig_bytes, 1);
5183 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5185 update_bytes_may_use(space_info, orig_bytes);
5186 trace_btrfs_space_reservation(fs_info, "space_info",
5187 space_info->flags, orig_bytes, 1);
5192 * If we couldn't make a reservation then setup our reservation ticket
5193 * and kick the async worker if it's not already running.
5195 * If we are a priority flusher then we just need to add our ticket to
5196 * the list and we will do our own flushing further down.
5198 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5199 ticket.bytes = orig_bytes;
5201 init_waitqueue_head(&ticket.wait);
5202 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5203 list_add_tail(&ticket.list, &space_info->tickets);
5204 if (!space_info->flush) {
5205 space_info->flush = 1;
5206 trace_btrfs_trigger_flush(fs_info,
5210 queue_work(system_unbound_wq,
5211 &fs_info->async_reclaim_work);
5214 list_add_tail(&ticket.list,
5215 &space_info->priority_tickets);
5217 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5220 * We will do the space reservation dance during log replay,
5221 * which means we won't have fs_info->fs_root set, so don't do
5222 * the async reclaim as we will panic.
5224 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5225 need_do_async_reclaim(fs_info, space_info,
5226 used, system_chunk) &&
5227 !work_busy(&fs_info->async_reclaim_work)) {
5228 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5229 orig_bytes, flush, "preempt");
5230 queue_work(system_unbound_wq,
5231 &fs_info->async_reclaim_work);
5234 spin_unlock(&space_info->lock);
5235 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5238 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5239 return wait_reserve_ticket(fs_info, space_info, &ticket,
5243 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5244 spin_lock(&space_info->lock);
5246 if (ticket.bytes < orig_bytes) {
5247 u64 num_bytes = orig_bytes - ticket.bytes;
5248 update_bytes_may_use(space_info, -num_bytes);
5249 trace_btrfs_space_reservation(fs_info, "space_info",
5254 list_del_init(&ticket.list);
5257 spin_unlock(&space_info->lock);
5258 ASSERT(list_empty(&ticket.list));
5263 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5264 * @root - the root we're allocating for
5265 * @block_rsv - the block_rsv we're allocating for
5266 * @orig_bytes - the number of bytes we want
5267 * @flush - whether or not we can flush to make our reservation
5269 * This will reserve orgi_bytes number of bytes from the space info associated
5270 * with the block_rsv. If there is not enough space it will make an attempt to
5271 * flush out space to make room. It will do this by flushing delalloc if
5272 * possible or committing the transaction. If flush is 0 then no attempts to
5273 * regain reservations will be made and this will fail if there is not enough
5276 static int reserve_metadata_bytes(struct btrfs_root *root,
5277 struct btrfs_block_rsv *block_rsv,
5279 enum btrfs_reserve_flush_enum flush)
5281 struct btrfs_fs_info *fs_info = root->fs_info;
5282 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5284 bool system_chunk = (root == fs_info->chunk_root);
5286 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5287 orig_bytes, flush, system_chunk);
5288 if (ret == -ENOSPC &&
5289 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5290 if (block_rsv != global_rsv &&
5291 !block_rsv_use_bytes(global_rsv, orig_bytes))
5294 if (ret == -ENOSPC) {
5295 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5296 block_rsv->space_info->flags,
5299 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5300 dump_space_info(fs_info, block_rsv->space_info,
5306 static struct btrfs_block_rsv *get_block_rsv(
5307 const struct btrfs_trans_handle *trans,
5308 const struct btrfs_root *root)
5310 struct btrfs_fs_info *fs_info = root->fs_info;
5311 struct btrfs_block_rsv *block_rsv = NULL;
5313 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5314 (root == fs_info->csum_root && trans->adding_csums) ||
5315 (root == fs_info->uuid_root))
5316 block_rsv = trans->block_rsv;
5319 block_rsv = root->block_rsv;
5322 block_rsv = &fs_info->empty_block_rsv;
5327 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5331 spin_lock(&block_rsv->lock);
5332 if (block_rsv->reserved >= num_bytes) {
5333 block_rsv->reserved -= num_bytes;
5334 if (block_rsv->reserved < block_rsv->size)
5335 block_rsv->full = 0;
5338 spin_unlock(&block_rsv->lock);
5342 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5343 u64 num_bytes, bool update_size)
5345 spin_lock(&block_rsv->lock);
5346 block_rsv->reserved += num_bytes;
5348 block_rsv->size += num_bytes;
5349 else if (block_rsv->reserved >= block_rsv->size)
5350 block_rsv->full = 1;
5351 spin_unlock(&block_rsv->lock);
5354 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5355 struct btrfs_block_rsv *dest, u64 num_bytes,
5358 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5361 if (global_rsv->space_info != dest->space_info)
5364 spin_lock(&global_rsv->lock);
5365 min_bytes = div_factor(global_rsv->size, min_factor);
5366 if (global_rsv->reserved < min_bytes + num_bytes) {
5367 spin_unlock(&global_rsv->lock);
5370 global_rsv->reserved -= num_bytes;
5371 if (global_rsv->reserved < global_rsv->size)
5372 global_rsv->full = 0;
5373 spin_unlock(&global_rsv->lock);
5375 block_rsv_add_bytes(dest, num_bytes, true);
5380 * This is for space we already have accounted in space_info->bytes_may_use, so
5381 * basically when we're returning space from block_rsv's.
5383 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5384 struct btrfs_space_info *space_info,
5387 struct reserve_ticket *ticket;
5388 struct list_head *head;
5390 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5391 bool check_overcommit = false;
5393 spin_lock(&space_info->lock);
5394 head = &space_info->priority_tickets;
5397 * If we are over our limit then we need to check and see if we can
5398 * overcommit, and if we can't then we just need to free up our space
5399 * and not satisfy any requests.
5401 used = btrfs_space_info_used(space_info, true);
5402 if (used - num_bytes >= space_info->total_bytes)
5403 check_overcommit = true;
5405 while (!list_empty(head) && num_bytes) {
5406 ticket = list_first_entry(head, struct reserve_ticket,
5409 * We use 0 bytes because this space is already reserved, so
5410 * adding the ticket space would be a double count.
5412 if (check_overcommit &&
5413 !can_overcommit(fs_info, space_info, 0, flush, false))
5415 if (num_bytes >= ticket->bytes) {
5416 list_del_init(&ticket->list);
5417 num_bytes -= ticket->bytes;
5419 space_info->tickets_id++;
5420 wake_up(&ticket->wait);
5422 ticket->bytes -= num_bytes;
5427 if (num_bytes && head == &space_info->priority_tickets) {
5428 head = &space_info->tickets;
5429 flush = BTRFS_RESERVE_FLUSH_ALL;
5432 update_bytes_may_use(space_info, -num_bytes);
5433 trace_btrfs_space_reservation(fs_info, "space_info",
5434 space_info->flags, num_bytes, 0);
5435 spin_unlock(&space_info->lock);
5439 * This is for newly allocated space that isn't accounted in
5440 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5441 * we use this helper.
5443 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5444 struct btrfs_space_info *space_info,
5447 struct reserve_ticket *ticket;
5448 struct list_head *head = &space_info->priority_tickets;
5451 while (!list_empty(head) && num_bytes) {
5452 ticket = list_first_entry(head, struct reserve_ticket,
5454 if (num_bytes >= ticket->bytes) {
5455 trace_btrfs_space_reservation(fs_info, "space_info",
5458 list_del_init(&ticket->list);
5459 num_bytes -= ticket->bytes;
5460 update_bytes_may_use(space_info, ticket->bytes);
5462 space_info->tickets_id++;
5463 wake_up(&ticket->wait);
5465 trace_btrfs_space_reservation(fs_info, "space_info",
5468 update_bytes_may_use(space_info, num_bytes);
5469 ticket->bytes -= num_bytes;
5474 if (num_bytes && head == &space_info->priority_tickets) {
5475 head = &space_info->tickets;
5480 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5481 struct btrfs_block_rsv *block_rsv,
5482 struct btrfs_block_rsv *dest, u64 num_bytes,
5483 u64 *qgroup_to_release_ret)
5485 struct btrfs_space_info *space_info = block_rsv->space_info;
5486 u64 qgroup_to_release = 0;
5489 spin_lock(&block_rsv->lock);
5490 if (num_bytes == (u64)-1) {
5491 num_bytes = block_rsv->size;
5492 qgroup_to_release = block_rsv->qgroup_rsv_size;
5494 block_rsv->size -= num_bytes;
5495 if (block_rsv->reserved >= block_rsv->size) {
5496 num_bytes = block_rsv->reserved - block_rsv->size;
5497 block_rsv->reserved = block_rsv->size;
5498 block_rsv->full = 1;
5502 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5503 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5504 block_rsv->qgroup_rsv_size;
5505 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5507 qgroup_to_release = 0;
5509 spin_unlock(&block_rsv->lock);
5512 if (num_bytes > 0) {
5514 spin_lock(&dest->lock);
5518 bytes_to_add = dest->size - dest->reserved;
5519 bytes_to_add = min(num_bytes, bytes_to_add);
5520 dest->reserved += bytes_to_add;
5521 if (dest->reserved >= dest->size)
5523 num_bytes -= bytes_to_add;
5525 spin_unlock(&dest->lock);
5528 space_info_add_old_bytes(fs_info, space_info,
5531 if (qgroup_to_release_ret)
5532 *qgroup_to_release_ret = qgroup_to_release;
5536 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5537 struct btrfs_block_rsv *dst, u64 num_bytes,
5542 ret = block_rsv_use_bytes(src, num_bytes);
5546 block_rsv_add_bytes(dst, num_bytes, update_size);
5550 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5552 memset(rsv, 0, sizeof(*rsv));
5553 spin_lock_init(&rsv->lock);
5557 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5558 struct btrfs_block_rsv *rsv,
5559 unsigned short type)
5561 btrfs_init_block_rsv(rsv, type);
5562 rsv->space_info = __find_space_info(fs_info,
5563 BTRFS_BLOCK_GROUP_METADATA);
5566 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5567 unsigned short type)
5569 struct btrfs_block_rsv *block_rsv;
5571 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5575 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5579 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5580 struct btrfs_block_rsv *rsv)
5584 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5588 int btrfs_block_rsv_add(struct btrfs_root *root,
5589 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5590 enum btrfs_reserve_flush_enum flush)
5597 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5599 block_rsv_add_bytes(block_rsv, num_bytes, true);
5604 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5612 spin_lock(&block_rsv->lock);
5613 num_bytes = div_factor(block_rsv->size, min_factor);
5614 if (block_rsv->reserved >= num_bytes)
5616 spin_unlock(&block_rsv->lock);
5621 int btrfs_block_rsv_refill(struct btrfs_root *root,
5622 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5623 enum btrfs_reserve_flush_enum flush)
5631 spin_lock(&block_rsv->lock);
5632 num_bytes = min_reserved;
5633 if (block_rsv->reserved >= num_bytes)
5636 num_bytes -= block_rsv->reserved;
5637 spin_unlock(&block_rsv->lock);
5642 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5644 block_rsv_add_bytes(block_rsv, num_bytes, false);
5652 * btrfs_inode_rsv_refill - refill the inode block rsv.
5653 * @inode - the inode we are refilling.
5654 * @flush - the flusing restriction.
5656 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5657 * block_rsv->size as the minimum size. We'll either refill the missing amount
5658 * or return if we already have enough space. This will also handle the resreve
5659 * tracepoint for the reserved amount.
5661 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5662 enum btrfs_reserve_flush_enum flush)
5664 struct btrfs_root *root = inode->root;
5665 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5667 u64 qgroup_num_bytes = 0;
5670 spin_lock(&block_rsv->lock);
5671 if (block_rsv->reserved < block_rsv->size)
5672 num_bytes = block_rsv->size - block_rsv->reserved;
5673 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5674 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5675 block_rsv->qgroup_rsv_reserved;
5676 spin_unlock(&block_rsv->lock);
5681 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5684 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5686 block_rsv_add_bytes(block_rsv, num_bytes, false);
5687 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5688 btrfs_ino(inode), num_bytes, 1);
5690 /* Don't forget to increase qgroup_rsv_reserved */
5691 spin_lock(&block_rsv->lock);
5692 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5693 spin_unlock(&block_rsv->lock);
5695 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5700 * btrfs_inode_rsv_release - release any excessive reservation.
5701 * @inode - the inode we need to release from.
5702 * @qgroup_free - free or convert qgroup meta.
5703 * Unlike normal operation, qgroup meta reservation needs to know if we are
5704 * freeing qgroup reservation or just converting it into per-trans. Normally
5705 * @qgroup_free is true for error handling, and false for normal release.
5707 * This is the same as btrfs_block_rsv_release, except that it handles the
5708 * tracepoint for the reservation.
5710 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5712 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5713 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5714 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5716 u64 qgroup_to_release = 0;
5719 * Since we statically set the block_rsv->size we just want to say we
5720 * are releasing 0 bytes, and then we'll just get the reservation over
5723 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5724 &qgroup_to_release);
5726 trace_btrfs_space_reservation(fs_info, "delalloc",
5727 btrfs_ino(inode), released, 0);
5729 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5731 btrfs_qgroup_convert_reserved_meta(inode->root,
5735 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5736 struct btrfs_block_rsv *block_rsv,
5739 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5741 if (global_rsv == block_rsv ||
5742 block_rsv->space_info != global_rsv->space_info)
5744 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5747 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5749 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5750 struct btrfs_space_info *sinfo = block_rsv->space_info;
5754 * The global block rsv is based on the size of the extent tree, the
5755 * checksum tree and the root tree. If the fs is empty we want to set
5756 * it to a minimal amount for safety.
5758 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5759 btrfs_root_used(&fs_info->csum_root->root_item) +
5760 btrfs_root_used(&fs_info->tree_root->root_item);
5761 num_bytes = max_t(u64, num_bytes, SZ_16M);
5763 spin_lock(&sinfo->lock);
5764 spin_lock(&block_rsv->lock);
5766 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5768 if (block_rsv->reserved < block_rsv->size) {
5769 num_bytes = btrfs_space_info_used(sinfo, true);
5770 if (sinfo->total_bytes > num_bytes) {
5771 num_bytes = sinfo->total_bytes - num_bytes;
5772 num_bytes = min(num_bytes,
5773 block_rsv->size - block_rsv->reserved);
5774 block_rsv->reserved += num_bytes;
5775 update_bytes_may_use(sinfo, num_bytes);
5776 trace_btrfs_space_reservation(fs_info, "space_info",
5777 sinfo->flags, num_bytes,
5780 } else if (block_rsv->reserved > block_rsv->size) {
5781 num_bytes = block_rsv->reserved - block_rsv->size;
5782 update_bytes_may_use(sinfo, -num_bytes);
5783 trace_btrfs_space_reservation(fs_info, "space_info",
5784 sinfo->flags, num_bytes, 0);
5785 block_rsv->reserved = block_rsv->size;
5788 if (block_rsv->reserved == block_rsv->size)
5789 block_rsv->full = 1;
5791 block_rsv->full = 0;
5793 spin_unlock(&block_rsv->lock);
5794 spin_unlock(&sinfo->lock);
5797 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5799 struct btrfs_space_info *space_info;
5801 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5802 fs_info->chunk_block_rsv.space_info = space_info;
5804 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5805 fs_info->global_block_rsv.space_info = space_info;
5806 fs_info->trans_block_rsv.space_info = space_info;
5807 fs_info->empty_block_rsv.space_info = space_info;
5808 fs_info->delayed_block_rsv.space_info = space_info;
5810 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5811 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5812 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5813 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5814 if (fs_info->quota_root)
5815 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5816 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5818 update_global_block_rsv(fs_info);
5821 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5823 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5825 WARN_ON(fs_info->trans_block_rsv.size > 0);
5826 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5827 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5828 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5829 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5830 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5835 * To be called after all the new block groups attached to the transaction
5836 * handle have been created (btrfs_create_pending_block_groups()).
5838 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5840 struct btrfs_fs_info *fs_info = trans->fs_info;
5842 if (!trans->chunk_bytes_reserved)
5845 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5847 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5848 trans->chunk_bytes_reserved, NULL);
5849 trans->chunk_bytes_reserved = 0;
5853 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5854 * root: the root of the parent directory
5855 * rsv: block reservation
5856 * items: the number of items that we need do reservation
5857 * use_global_rsv: allow fallback to the global block reservation
5859 * This function is used to reserve the space for snapshot/subvolume
5860 * creation and deletion. Those operations are different with the
5861 * common file/directory operations, they change two fs/file trees
5862 * and root tree, the number of items that the qgroup reserves is
5863 * different with the free space reservation. So we can not use
5864 * the space reservation mechanism in start_transaction().
5866 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5867 struct btrfs_block_rsv *rsv, int items,
5868 bool use_global_rsv)
5870 u64 qgroup_num_bytes = 0;
5873 struct btrfs_fs_info *fs_info = root->fs_info;
5874 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5876 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5877 /* One for parent inode, two for dir entries */
5878 qgroup_num_bytes = 3 * fs_info->nodesize;
5879 ret = btrfs_qgroup_reserve_meta_prealloc(root,
5880 qgroup_num_bytes, true);
5885 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5886 rsv->space_info = __find_space_info(fs_info,
5887 BTRFS_BLOCK_GROUP_METADATA);
5888 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5889 BTRFS_RESERVE_FLUSH_ALL);
5891 if (ret == -ENOSPC && use_global_rsv)
5892 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
5894 if (ret && qgroup_num_bytes)
5895 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5900 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5901 struct btrfs_block_rsv *rsv)
5903 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5906 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5907 struct btrfs_inode *inode)
5909 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5910 u64 reserve_size = 0;
5911 u64 qgroup_rsv_size = 0;
5913 unsigned outstanding_extents;
5915 lockdep_assert_held(&inode->lock);
5916 outstanding_extents = inode->outstanding_extents;
5917 if (outstanding_extents)
5918 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
5919 outstanding_extents + 1);
5920 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
5922 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
5925 * For qgroup rsv, the calculation is very simple:
5926 * account one nodesize for each outstanding extent
5928 * This is overestimating in most cases.
5930 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
5932 spin_lock(&block_rsv->lock);
5933 block_rsv->size = reserve_size;
5934 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
5935 spin_unlock(&block_rsv->lock);
5938 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5940 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5941 unsigned nr_extents;
5942 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5944 bool delalloc_lock = true;
5946 /* If we are a free space inode we need to not flush since we will be in
5947 * the middle of a transaction commit. We also don't need the delalloc
5948 * mutex since we won't race with anybody. We need this mostly to make
5949 * lockdep shut its filthy mouth.
5951 * If we have a transaction open (can happen if we call truncate_block
5952 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5954 if (btrfs_is_free_space_inode(inode)) {
5955 flush = BTRFS_RESERVE_NO_FLUSH;
5956 delalloc_lock = false;
5958 if (current->journal_info)
5959 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5961 if (btrfs_transaction_in_commit(fs_info))
5962 schedule_timeout(1);
5966 mutex_lock(&inode->delalloc_mutex);
5968 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5970 /* Add our new extents and calculate the new rsv size. */
5971 spin_lock(&inode->lock);
5972 nr_extents = count_max_extents(num_bytes);
5973 btrfs_mod_outstanding_extents(inode, nr_extents);
5974 inode->csum_bytes += num_bytes;
5975 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5976 spin_unlock(&inode->lock);
5978 ret = btrfs_inode_rsv_refill(inode, flush);
5983 mutex_unlock(&inode->delalloc_mutex);
5987 spin_lock(&inode->lock);
5988 nr_extents = count_max_extents(num_bytes);
5989 btrfs_mod_outstanding_extents(inode, -nr_extents);
5990 inode->csum_bytes -= num_bytes;
5991 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5992 spin_unlock(&inode->lock);
5994 btrfs_inode_rsv_release(inode, true);
5996 mutex_unlock(&inode->delalloc_mutex);
6001 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6002 * @inode: the inode to release the reservation for.
6003 * @num_bytes: the number of bytes we are releasing.
6004 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6006 * This will release the metadata reservation for an inode. This can be called
6007 * once we complete IO for a given set of bytes to release their metadata
6008 * reservations, or on error for the same reason.
6010 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6013 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6015 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6016 spin_lock(&inode->lock);
6017 inode->csum_bytes -= num_bytes;
6018 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6019 spin_unlock(&inode->lock);
6021 if (btrfs_is_testing(fs_info))
6024 btrfs_inode_rsv_release(inode, qgroup_free);
6028 * btrfs_delalloc_release_extents - release our outstanding_extents
6029 * @inode: the inode to balance the reservation for.
6030 * @num_bytes: the number of bytes we originally reserved with
6031 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6033 * When we reserve space we increase outstanding_extents for the extents we may
6034 * add. Once we've set the range as delalloc or created our ordered extents we
6035 * have outstanding_extents to track the real usage, so we use this to free our
6036 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6037 * with btrfs_delalloc_reserve_metadata.
6039 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6042 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6043 unsigned num_extents;
6045 spin_lock(&inode->lock);
6046 num_extents = count_max_extents(num_bytes);
6047 btrfs_mod_outstanding_extents(inode, -num_extents);
6048 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6049 spin_unlock(&inode->lock);
6051 if (btrfs_is_testing(fs_info))
6054 btrfs_inode_rsv_release(inode, qgroup_free);
6058 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6060 * @inode: inode we're writing to
6061 * @start: start range we are writing to
6062 * @len: how long the range we are writing to
6063 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6064 * current reservation.
6066 * This will do the following things
6068 * o reserve space in data space info for num bytes
6069 * and reserve precious corresponding qgroup space
6070 * (Done in check_data_free_space)
6072 * o reserve space for metadata space, based on the number of outstanding
6073 * extents and how much csums will be needed
6074 * also reserve metadata space in a per root over-reserve method.
6075 * o add to the inodes->delalloc_bytes
6076 * o add it to the fs_info's delalloc inodes list.
6077 * (Above 3 all done in delalloc_reserve_metadata)
6079 * Return 0 for success
6080 * Return <0 for error(-ENOSPC or -EQUOT)
6082 int btrfs_delalloc_reserve_space(struct inode *inode,
6083 struct extent_changeset **reserved, u64 start, u64 len)
6087 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6090 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6092 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6097 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6098 * @inode: inode we're releasing space for
6099 * @start: start position of the space already reserved
6100 * @len: the len of the space already reserved
6101 * @release_bytes: the len of the space we consumed or didn't use
6103 * This function will release the metadata space that was not used and will
6104 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6105 * list if there are no delalloc bytes left.
6106 * Also it will handle the qgroup reserved space.
6108 void btrfs_delalloc_release_space(struct inode *inode,
6109 struct extent_changeset *reserved,
6110 u64 start, u64 len, bool qgroup_free)
6112 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6113 btrfs_free_reserved_data_space(inode, reserved, start, len);
6116 static int update_block_group(struct btrfs_trans_handle *trans,
6117 struct btrfs_fs_info *info, u64 bytenr,
6118 u64 num_bytes, int alloc)
6120 struct btrfs_block_group_cache *cache = NULL;
6121 u64 total = num_bytes;
6126 /* block accounting for super block */
6127 spin_lock(&info->delalloc_root_lock);
6128 old_val = btrfs_super_bytes_used(info->super_copy);
6130 old_val += num_bytes;
6132 old_val -= num_bytes;
6133 btrfs_set_super_bytes_used(info->super_copy, old_val);
6134 spin_unlock(&info->delalloc_root_lock);
6137 cache = btrfs_lookup_block_group(info, bytenr);
6140 factor = btrfs_bg_type_to_factor(cache->flags);
6143 * If this block group has free space cache written out, we
6144 * need to make sure to load it if we are removing space. This
6145 * is because we need the unpinning stage to actually add the
6146 * space back to the block group, otherwise we will leak space.
6148 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6149 cache_block_group(cache, 1);
6151 byte_in_group = bytenr - cache->key.objectid;
6152 WARN_ON(byte_in_group > cache->key.offset);
6154 spin_lock(&cache->space_info->lock);
6155 spin_lock(&cache->lock);
6157 if (btrfs_test_opt(info, SPACE_CACHE) &&
6158 cache->disk_cache_state < BTRFS_DC_CLEAR)
6159 cache->disk_cache_state = BTRFS_DC_CLEAR;
6161 old_val = btrfs_block_group_used(&cache->item);
6162 num_bytes = min(total, cache->key.offset - byte_in_group);
6164 old_val += num_bytes;
6165 btrfs_set_block_group_used(&cache->item, old_val);
6166 cache->reserved -= num_bytes;
6167 cache->space_info->bytes_reserved -= num_bytes;
6168 cache->space_info->bytes_used += num_bytes;
6169 cache->space_info->disk_used += num_bytes * factor;
6170 spin_unlock(&cache->lock);
6171 spin_unlock(&cache->space_info->lock);
6173 old_val -= num_bytes;
6174 btrfs_set_block_group_used(&cache->item, old_val);
6175 cache->pinned += num_bytes;
6176 update_bytes_pinned(cache->space_info, num_bytes);
6177 cache->space_info->bytes_used -= num_bytes;
6178 cache->space_info->disk_used -= num_bytes * factor;
6179 spin_unlock(&cache->lock);
6180 spin_unlock(&cache->space_info->lock);
6182 trace_btrfs_space_reservation(info, "pinned",
6183 cache->space_info->flags,
6185 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6187 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6188 set_extent_dirty(info->pinned_extents,
6189 bytenr, bytenr + num_bytes - 1,
6190 GFP_NOFS | __GFP_NOFAIL);
6193 spin_lock(&trans->transaction->dirty_bgs_lock);
6194 if (list_empty(&cache->dirty_list)) {
6195 list_add_tail(&cache->dirty_list,
6196 &trans->transaction->dirty_bgs);
6197 trans->transaction->num_dirty_bgs++;
6198 btrfs_get_block_group(cache);
6200 spin_unlock(&trans->transaction->dirty_bgs_lock);
6203 * No longer have used bytes in this block group, queue it for
6204 * deletion. We do this after adding the block group to the
6205 * dirty list to avoid races between cleaner kthread and space
6208 if (!alloc && old_val == 0)
6209 btrfs_mark_bg_unused(cache);
6211 btrfs_put_block_group(cache);
6213 bytenr += num_bytes;
6218 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6220 struct btrfs_block_group_cache *cache;
6223 spin_lock(&fs_info->block_group_cache_lock);
6224 bytenr = fs_info->first_logical_byte;
6225 spin_unlock(&fs_info->block_group_cache_lock);
6227 if (bytenr < (u64)-1)
6230 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6234 bytenr = cache->key.objectid;
6235 btrfs_put_block_group(cache);
6240 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6241 struct btrfs_block_group_cache *cache,
6242 u64 bytenr, u64 num_bytes, int reserved)
6244 spin_lock(&cache->space_info->lock);
6245 spin_lock(&cache->lock);
6246 cache->pinned += num_bytes;
6247 update_bytes_pinned(cache->space_info, num_bytes);
6249 cache->reserved -= num_bytes;
6250 cache->space_info->bytes_reserved -= num_bytes;
6252 spin_unlock(&cache->lock);
6253 spin_unlock(&cache->space_info->lock);
6255 trace_btrfs_space_reservation(fs_info, "pinned",
6256 cache->space_info->flags, num_bytes, 1);
6257 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6258 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6259 set_extent_dirty(fs_info->pinned_extents, bytenr,
6260 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6265 * this function must be called within transaction
6267 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6268 u64 bytenr, u64 num_bytes, int reserved)
6270 struct btrfs_block_group_cache *cache;
6272 cache = btrfs_lookup_block_group(fs_info, bytenr);
6273 BUG_ON(!cache); /* Logic error */
6275 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6277 btrfs_put_block_group(cache);
6282 * this function must be called within transaction
6284 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6285 u64 bytenr, u64 num_bytes)
6287 struct btrfs_block_group_cache *cache;
6290 cache = btrfs_lookup_block_group(fs_info, bytenr);
6295 * pull in the free space cache (if any) so that our pin
6296 * removes the free space from the cache. We have load_only set
6297 * to one because the slow code to read in the free extents does check
6298 * the pinned extents.
6300 cache_block_group(cache, 1);
6302 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6304 /* remove us from the free space cache (if we're there at all) */
6305 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6306 btrfs_put_block_group(cache);
6310 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6311 u64 start, u64 num_bytes)
6314 struct btrfs_block_group_cache *block_group;
6315 struct btrfs_caching_control *caching_ctl;
6317 block_group = btrfs_lookup_block_group(fs_info, start);
6321 cache_block_group(block_group, 0);
6322 caching_ctl = get_caching_control(block_group);
6326 BUG_ON(!block_group_cache_done(block_group));
6327 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6329 mutex_lock(&caching_ctl->mutex);
6331 if (start >= caching_ctl->progress) {
6332 ret = add_excluded_extent(fs_info, start, num_bytes);
6333 } else if (start + num_bytes <= caching_ctl->progress) {
6334 ret = btrfs_remove_free_space(block_group,
6337 num_bytes = caching_ctl->progress - start;
6338 ret = btrfs_remove_free_space(block_group,
6343 num_bytes = (start + num_bytes) -
6344 caching_ctl->progress;
6345 start = caching_ctl->progress;
6346 ret = add_excluded_extent(fs_info, start, num_bytes);
6349 mutex_unlock(&caching_ctl->mutex);
6350 put_caching_control(caching_ctl);
6352 btrfs_put_block_group(block_group);
6356 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6357 struct extent_buffer *eb)
6359 struct btrfs_file_extent_item *item;
6360 struct btrfs_key key;
6365 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6368 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6369 btrfs_item_key_to_cpu(eb, &key, i);
6370 if (key.type != BTRFS_EXTENT_DATA_KEY)
6372 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6373 found_type = btrfs_file_extent_type(eb, item);
6374 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6376 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6378 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6379 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6380 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6389 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6391 atomic_inc(&bg->reservations);
6394 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6397 struct btrfs_block_group_cache *bg;
6399 bg = btrfs_lookup_block_group(fs_info, start);
6401 if (atomic_dec_and_test(&bg->reservations))
6402 wake_up_var(&bg->reservations);
6403 btrfs_put_block_group(bg);
6406 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6408 struct btrfs_space_info *space_info = bg->space_info;
6412 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6416 * Our block group is read only but before we set it to read only,
6417 * some task might have had allocated an extent from it already, but it
6418 * has not yet created a respective ordered extent (and added it to a
6419 * root's list of ordered extents).
6420 * Therefore wait for any task currently allocating extents, since the
6421 * block group's reservations counter is incremented while a read lock
6422 * on the groups' semaphore is held and decremented after releasing
6423 * the read access on that semaphore and creating the ordered extent.
6425 down_write(&space_info->groups_sem);
6426 up_write(&space_info->groups_sem);
6428 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6432 * btrfs_add_reserved_bytes - update the block_group and space info counters
6433 * @cache: The cache we are manipulating
6434 * @ram_bytes: The number of bytes of file content, and will be same to
6435 * @num_bytes except for the compress path.
6436 * @num_bytes: The number of bytes in question
6437 * @delalloc: The blocks are allocated for the delalloc write
6439 * This is called by the allocator when it reserves space. If this is a
6440 * reservation and the block group has become read only we cannot make the
6441 * reservation and return -EAGAIN, otherwise this function always succeeds.
6443 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6444 u64 ram_bytes, u64 num_bytes, int delalloc)
6446 struct btrfs_space_info *space_info = cache->space_info;
6449 spin_lock(&space_info->lock);
6450 spin_lock(&cache->lock);
6454 cache->reserved += num_bytes;
6455 space_info->bytes_reserved += num_bytes;
6456 update_bytes_may_use(space_info, -ram_bytes);
6458 cache->delalloc_bytes += num_bytes;
6460 spin_unlock(&cache->lock);
6461 spin_unlock(&space_info->lock);
6466 * btrfs_free_reserved_bytes - update the block_group and space info counters
6467 * @cache: The cache we are manipulating
6468 * @num_bytes: The number of bytes in question
6469 * @delalloc: The blocks are allocated for the delalloc write
6471 * This is called by somebody who is freeing space that was never actually used
6472 * on disk. For example if you reserve some space for a new leaf in transaction
6473 * A and before transaction A commits you free that leaf, you call this with
6474 * reserve set to 0 in order to clear the reservation.
6477 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6478 u64 num_bytes, int delalloc)
6480 struct btrfs_space_info *space_info = cache->space_info;
6482 spin_lock(&space_info->lock);
6483 spin_lock(&cache->lock);
6485 space_info->bytes_readonly += num_bytes;
6486 cache->reserved -= num_bytes;
6487 space_info->bytes_reserved -= num_bytes;
6488 space_info->max_extent_size = 0;
6491 cache->delalloc_bytes -= num_bytes;
6492 spin_unlock(&cache->lock);
6493 spin_unlock(&space_info->lock);
6495 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6497 struct btrfs_caching_control *next;
6498 struct btrfs_caching_control *caching_ctl;
6499 struct btrfs_block_group_cache *cache;
6501 down_write(&fs_info->commit_root_sem);
6503 list_for_each_entry_safe(caching_ctl, next,
6504 &fs_info->caching_block_groups, list) {
6505 cache = caching_ctl->block_group;
6506 if (block_group_cache_done(cache)) {
6507 cache->last_byte_to_unpin = (u64)-1;
6508 list_del_init(&caching_ctl->list);
6509 put_caching_control(caching_ctl);
6511 cache->last_byte_to_unpin = caching_ctl->progress;
6515 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6516 fs_info->pinned_extents = &fs_info->freed_extents[1];
6518 fs_info->pinned_extents = &fs_info->freed_extents[0];
6520 up_write(&fs_info->commit_root_sem);
6522 update_global_block_rsv(fs_info);
6526 * Returns the free cluster for the given space info and sets empty_cluster to
6527 * what it should be based on the mount options.
6529 static struct btrfs_free_cluster *
6530 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6531 struct btrfs_space_info *space_info, u64 *empty_cluster)
6533 struct btrfs_free_cluster *ret = NULL;
6536 if (btrfs_mixed_space_info(space_info))
6539 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6540 ret = &fs_info->meta_alloc_cluster;
6541 if (btrfs_test_opt(fs_info, SSD))
6542 *empty_cluster = SZ_2M;
6544 *empty_cluster = SZ_64K;
6545 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6546 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6547 *empty_cluster = SZ_2M;
6548 ret = &fs_info->data_alloc_cluster;
6554 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6556 const bool return_free_space)
6558 struct btrfs_block_group_cache *cache = NULL;
6559 struct btrfs_space_info *space_info;
6560 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6561 struct btrfs_free_cluster *cluster = NULL;
6563 u64 total_unpinned = 0;
6564 u64 empty_cluster = 0;
6567 while (start <= end) {
6570 start >= cache->key.objectid + cache->key.offset) {
6572 btrfs_put_block_group(cache);
6574 cache = btrfs_lookup_block_group(fs_info, start);
6575 BUG_ON(!cache); /* Logic error */
6577 cluster = fetch_cluster_info(fs_info,
6580 empty_cluster <<= 1;
6583 len = cache->key.objectid + cache->key.offset - start;
6584 len = min(len, end + 1 - start);
6586 if (start < cache->last_byte_to_unpin) {
6587 len = min(len, cache->last_byte_to_unpin - start);
6588 if (return_free_space)
6589 btrfs_add_free_space(cache, start, len);
6593 total_unpinned += len;
6594 space_info = cache->space_info;
6597 * If this space cluster has been marked as fragmented and we've
6598 * unpinned enough in this block group to potentially allow a
6599 * cluster to be created inside of it go ahead and clear the
6602 if (cluster && cluster->fragmented &&
6603 total_unpinned > empty_cluster) {
6604 spin_lock(&cluster->lock);
6605 cluster->fragmented = 0;
6606 spin_unlock(&cluster->lock);
6609 spin_lock(&space_info->lock);
6610 spin_lock(&cache->lock);
6611 cache->pinned -= len;
6612 update_bytes_pinned(space_info, -len);
6614 trace_btrfs_space_reservation(fs_info, "pinned",
6615 space_info->flags, len, 0);
6616 space_info->max_extent_size = 0;
6617 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6618 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6620 space_info->bytes_readonly += len;
6623 spin_unlock(&cache->lock);
6624 if (!readonly && return_free_space &&
6625 global_rsv->space_info == space_info) {
6628 spin_lock(&global_rsv->lock);
6629 if (!global_rsv->full) {
6630 to_add = min(len, global_rsv->size -
6631 global_rsv->reserved);
6632 global_rsv->reserved += to_add;
6633 update_bytes_may_use(space_info, to_add);
6634 if (global_rsv->reserved >= global_rsv->size)
6635 global_rsv->full = 1;
6636 trace_btrfs_space_reservation(fs_info,
6642 spin_unlock(&global_rsv->lock);
6643 /* Add to any tickets we may have */
6645 space_info_add_new_bytes(fs_info, space_info,
6648 spin_unlock(&space_info->lock);
6652 btrfs_put_block_group(cache);
6656 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6658 struct btrfs_fs_info *fs_info = trans->fs_info;
6659 struct btrfs_block_group_cache *block_group, *tmp;
6660 struct list_head *deleted_bgs;
6661 struct extent_io_tree *unpin;
6666 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6667 unpin = &fs_info->freed_extents[1];
6669 unpin = &fs_info->freed_extents[0];
6671 while (!trans->aborted) {
6672 struct extent_state *cached_state = NULL;
6674 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6675 ret = find_first_extent_bit(unpin, 0, &start, &end,
6676 EXTENT_DIRTY, &cached_state);
6678 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6682 if (btrfs_test_opt(fs_info, DISCARD))
6683 ret = btrfs_discard_extent(fs_info, start,
6684 end + 1 - start, NULL);
6686 clear_extent_dirty(unpin, start, end, &cached_state);
6687 unpin_extent_range(fs_info, start, end, true);
6688 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6689 free_extent_state(cached_state);
6694 * Transaction is finished. We don't need the lock anymore. We
6695 * do need to clean up the block groups in case of a transaction
6698 deleted_bgs = &trans->transaction->deleted_bgs;
6699 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6703 if (!trans->aborted)
6704 ret = btrfs_discard_extent(fs_info,
6705 block_group->key.objectid,
6706 block_group->key.offset,
6709 list_del_init(&block_group->bg_list);
6710 btrfs_put_block_group_trimming(block_group);
6711 btrfs_put_block_group(block_group);
6714 const char *errstr = btrfs_decode_error(ret);
6716 "discard failed while removing blockgroup: errno=%d %s",
6724 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6725 struct btrfs_delayed_ref_node *node, u64 parent,
6726 u64 root_objectid, u64 owner_objectid,
6727 u64 owner_offset, int refs_to_drop,
6728 struct btrfs_delayed_extent_op *extent_op)
6730 struct btrfs_fs_info *info = trans->fs_info;
6731 struct btrfs_key key;
6732 struct btrfs_path *path;
6733 struct btrfs_root *extent_root = info->extent_root;
6734 struct extent_buffer *leaf;
6735 struct btrfs_extent_item *ei;
6736 struct btrfs_extent_inline_ref *iref;
6739 int extent_slot = 0;
6740 int found_extent = 0;
6744 u64 bytenr = node->bytenr;
6745 u64 num_bytes = node->num_bytes;
6747 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6749 path = btrfs_alloc_path();
6753 path->reada = READA_FORWARD;
6754 path->leave_spinning = 1;
6756 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6757 BUG_ON(!is_data && refs_to_drop != 1);
6760 skinny_metadata = false;
6762 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6763 parent, root_objectid, owner_objectid,
6766 extent_slot = path->slots[0];
6767 while (extent_slot >= 0) {
6768 btrfs_item_key_to_cpu(path->nodes[0], &key,
6770 if (key.objectid != bytenr)
6772 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6773 key.offset == num_bytes) {
6777 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6778 key.offset == owner_objectid) {
6782 if (path->slots[0] - extent_slot > 5)
6787 if (!found_extent) {
6789 ret = remove_extent_backref(trans, path, NULL,
6791 is_data, &last_ref);
6793 btrfs_abort_transaction(trans, ret);
6796 btrfs_release_path(path);
6797 path->leave_spinning = 1;
6799 key.objectid = bytenr;
6800 key.type = BTRFS_EXTENT_ITEM_KEY;
6801 key.offset = num_bytes;
6803 if (!is_data && skinny_metadata) {
6804 key.type = BTRFS_METADATA_ITEM_KEY;
6805 key.offset = owner_objectid;
6808 ret = btrfs_search_slot(trans, extent_root,
6810 if (ret > 0 && skinny_metadata && path->slots[0]) {
6812 * Couldn't find our skinny metadata item,
6813 * see if we have ye olde extent item.
6816 btrfs_item_key_to_cpu(path->nodes[0], &key,
6818 if (key.objectid == bytenr &&
6819 key.type == BTRFS_EXTENT_ITEM_KEY &&
6820 key.offset == num_bytes)
6824 if (ret > 0 && skinny_metadata) {
6825 skinny_metadata = false;
6826 key.objectid = bytenr;
6827 key.type = BTRFS_EXTENT_ITEM_KEY;
6828 key.offset = num_bytes;
6829 btrfs_release_path(path);
6830 ret = btrfs_search_slot(trans, extent_root,
6836 "umm, got %d back from search, was looking for %llu",
6839 btrfs_print_leaf(path->nodes[0]);
6842 btrfs_abort_transaction(trans, ret);
6845 extent_slot = path->slots[0];
6847 } else if (WARN_ON(ret == -ENOENT)) {
6848 btrfs_print_leaf(path->nodes[0]);
6850 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6851 bytenr, parent, root_objectid, owner_objectid,
6853 btrfs_abort_transaction(trans, ret);
6856 btrfs_abort_transaction(trans, ret);
6860 leaf = path->nodes[0];
6861 item_size = btrfs_item_size_nr(leaf, extent_slot);
6862 if (unlikely(item_size < sizeof(*ei))) {
6864 btrfs_print_v0_err(info);
6865 btrfs_abort_transaction(trans, ret);
6868 ei = btrfs_item_ptr(leaf, extent_slot,
6869 struct btrfs_extent_item);
6870 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6871 key.type == BTRFS_EXTENT_ITEM_KEY) {
6872 struct btrfs_tree_block_info *bi;
6873 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6874 bi = (struct btrfs_tree_block_info *)(ei + 1);
6875 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6878 refs = btrfs_extent_refs(leaf, ei);
6879 if (refs < refs_to_drop) {
6881 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
6882 refs_to_drop, refs, bytenr);
6884 btrfs_abort_transaction(trans, ret);
6887 refs -= refs_to_drop;
6891 __run_delayed_extent_op(extent_op, leaf, ei);
6893 * In the case of inline back ref, reference count will
6894 * be updated by remove_extent_backref
6897 BUG_ON(!found_extent);
6899 btrfs_set_extent_refs(leaf, ei, refs);
6900 btrfs_mark_buffer_dirty(leaf);
6903 ret = remove_extent_backref(trans, path, iref,
6904 refs_to_drop, is_data,
6907 btrfs_abort_transaction(trans, ret);
6913 BUG_ON(is_data && refs_to_drop !=
6914 extent_data_ref_count(path, iref));
6916 BUG_ON(path->slots[0] != extent_slot);
6918 BUG_ON(path->slots[0] != extent_slot + 1);
6919 path->slots[0] = extent_slot;
6925 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6928 btrfs_abort_transaction(trans, ret);
6931 btrfs_release_path(path);
6934 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
6936 btrfs_abort_transaction(trans, ret);
6941 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
6943 btrfs_abort_transaction(trans, ret);
6947 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
6949 btrfs_abort_transaction(trans, ret);
6953 btrfs_release_path(path);
6956 btrfs_free_path(path);
6961 * when we free an block, it is possible (and likely) that we free the last
6962 * delayed ref for that extent as well. This searches the delayed ref tree for
6963 * a given extent, and if there are no other delayed refs to be processed, it
6964 * removes it from the tree.
6966 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6969 struct btrfs_delayed_ref_head *head;
6970 struct btrfs_delayed_ref_root *delayed_refs;
6973 delayed_refs = &trans->transaction->delayed_refs;
6974 spin_lock(&delayed_refs->lock);
6975 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
6977 goto out_delayed_unlock;
6979 spin_lock(&head->lock);
6980 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
6983 if (head->extent_op) {
6984 if (!head->must_insert_reserved)
6986 btrfs_free_delayed_extent_op(head->extent_op);
6987 head->extent_op = NULL;
6991 * waiting for the lock here would deadlock. If someone else has it
6992 * locked they are already in the process of dropping it anyway
6994 if (!mutex_trylock(&head->mutex))
6997 btrfs_delete_ref_head(delayed_refs, head);
6998 head->processing = 0;
7000 spin_unlock(&head->lock);
7001 spin_unlock(&delayed_refs->lock);
7003 BUG_ON(head->extent_op);
7004 if (head->must_insert_reserved)
7007 cleanup_ref_head_accounting(trans, head);
7008 mutex_unlock(&head->mutex);
7009 btrfs_put_delayed_ref_head(head);
7012 spin_unlock(&head->lock);
7015 spin_unlock(&delayed_refs->lock);
7019 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7020 struct btrfs_root *root,
7021 struct extent_buffer *buf,
7022 u64 parent, int last_ref)
7024 struct btrfs_fs_info *fs_info = root->fs_info;
7028 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7029 int old_ref_mod, new_ref_mod;
7031 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7032 root->root_key.objectid,
7033 btrfs_header_level(buf), 0,
7034 BTRFS_DROP_DELAYED_REF);
7035 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7037 root->root_key.objectid,
7038 btrfs_header_level(buf),
7039 BTRFS_DROP_DELAYED_REF, NULL,
7040 &old_ref_mod, &new_ref_mod);
7041 BUG_ON(ret); /* -ENOMEM */
7042 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7045 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7046 struct btrfs_block_group_cache *cache;
7048 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7049 ret = check_ref_cleanup(trans, buf->start);
7055 cache = btrfs_lookup_block_group(fs_info, buf->start);
7057 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7058 pin_down_extent(fs_info, cache, buf->start,
7060 btrfs_put_block_group(cache);
7064 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7066 btrfs_add_free_space(cache, buf->start, buf->len);
7067 btrfs_free_reserved_bytes(cache, buf->len, 0);
7068 btrfs_put_block_group(cache);
7069 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7073 add_pinned_bytes(fs_info, buf->len, true,
7074 root->root_key.objectid);
7078 * Deleting the buffer, clear the corrupt flag since it doesn't
7081 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7085 /* Can return -ENOMEM */
7086 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7087 struct btrfs_root *root,
7088 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7089 u64 owner, u64 offset)
7091 struct btrfs_fs_info *fs_info = root->fs_info;
7092 int old_ref_mod, new_ref_mod;
7095 if (btrfs_is_testing(fs_info))
7098 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7099 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7100 root_objectid, owner, offset,
7101 BTRFS_DROP_DELAYED_REF);
7104 * tree log blocks never actually go into the extent allocation
7105 * tree, just update pinning info and exit early.
7107 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7108 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7109 /* unlocks the pinned mutex */
7110 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7111 old_ref_mod = new_ref_mod = 0;
7113 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7114 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7116 root_objectid, (int)owner,
7117 BTRFS_DROP_DELAYED_REF, NULL,
7118 &old_ref_mod, &new_ref_mod);
7120 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7122 root_objectid, owner, offset,
7123 0, BTRFS_DROP_DELAYED_REF,
7124 &old_ref_mod, &new_ref_mod);
7127 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7128 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7130 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7137 * when we wait for progress in the block group caching, its because
7138 * our allocation attempt failed at least once. So, we must sleep
7139 * and let some progress happen before we try again.
7141 * This function will sleep at least once waiting for new free space to
7142 * show up, and then it will check the block group free space numbers
7143 * for our min num_bytes. Another option is to have it go ahead
7144 * and look in the rbtree for a free extent of a given size, but this
7147 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7148 * any of the information in this block group.
7150 static noinline void
7151 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7154 struct btrfs_caching_control *caching_ctl;
7156 caching_ctl = get_caching_control(cache);
7160 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7161 (cache->free_space_ctl->free_space >= num_bytes));
7163 put_caching_control(caching_ctl);
7167 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7169 struct btrfs_caching_control *caching_ctl;
7172 caching_ctl = get_caching_control(cache);
7174 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7176 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7177 if (cache->cached == BTRFS_CACHE_ERROR)
7179 put_caching_control(caching_ctl);
7183 enum btrfs_loop_type {
7184 LOOP_CACHING_NOWAIT = 0,
7185 LOOP_CACHING_WAIT = 1,
7186 LOOP_ALLOC_CHUNK = 2,
7187 LOOP_NO_EMPTY_SIZE = 3,
7191 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7195 down_read(&cache->data_rwsem);
7199 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7202 btrfs_get_block_group(cache);
7204 down_read(&cache->data_rwsem);
7207 static struct btrfs_block_group_cache *
7208 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7209 struct btrfs_free_cluster *cluster,
7212 struct btrfs_block_group_cache *used_bg = NULL;
7214 spin_lock(&cluster->refill_lock);
7216 used_bg = cluster->block_group;
7220 if (used_bg == block_group)
7223 btrfs_get_block_group(used_bg);
7228 if (down_read_trylock(&used_bg->data_rwsem))
7231 spin_unlock(&cluster->refill_lock);
7233 /* We should only have one-level nested. */
7234 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7236 spin_lock(&cluster->refill_lock);
7237 if (used_bg == cluster->block_group)
7240 up_read(&used_bg->data_rwsem);
7241 btrfs_put_block_group(used_bg);
7246 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7250 up_read(&cache->data_rwsem);
7251 btrfs_put_block_group(cache);
7255 * Structure used internally for find_free_extent() function. Wraps needed
7258 struct find_free_extent_ctl {
7259 /* Basic allocation info */
7266 /* Where to start the search inside the bg */
7269 /* For clustered allocation */
7272 bool have_caching_bg;
7273 bool orig_have_caching_bg;
7275 /* RAID index, converted from flags */
7279 * Current loop number, check find_free_extent_update_loop() for details
7284 * Whether we're refilling a cluster, if true we need to re-search
7285 * current block group but don't try to refill the cluster again.
7287 bool retry_clustered;
7290 * Whether we're updating free space cache, if true we need to re-search
7291 * current block group but don't try updating free space cache again.
7293 bool retry_unclustered;
7295 /* If current block group is cached */
7298 /* Max contiguous hole found */
7299 u64 max_extent_size;
7301 /* Total free space from free space cache, not always contiguous */
7302 u64 total_free_space;
7310 * Helper function for find_free_extent().
7312 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7313 * Return -EAGAIN to inform caller that we need to re-search this block group
7314 * Return >0 to inform caller that we find nothing
7315 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7317 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7318 struct btrfs_free_cluster *last_ptr,
7319 struct find_free_extent_ctl *ffe_ctl,
7320 struct btrfs_block_group_cache **cluster_bg_ret)
7322 struct btrfs_fs_info *fs_info = bg->fs_info;
7323 struct btrfs_block_group_cache *cluster_bg;
7324 u64 aligned_cluster;
7328 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7330 goto refill_cluster;
7331 if (cluster_bg != bg && (cluster_bg->ro ||
7332 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7333 goto release_cluster;
7335 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7336 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7337 &ffe_ctl->max_extent_size);
7339 /* We have a block, we're done */
7340 spin_unlock(&last_ptr->refill_lock);
7341 trace_btrfs_reserve_extent_cluster(cluster_bg,
7342 ffe_ctl->search_start, ffe_ctl->num_bytes);
7343 *cluster_bg_ret = cluster_bg;
7344 ffe_ctl->found_offset = offset;
7347 WARN_ON(last_ptr->block_group != cluster_bg);
7351 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7352 * lets just skip it and let the allocator find whatever block it can
7353 * find. If we reach this point, we will have tried the cluster
7354 * allocator plenty of times and not have found anything, so we are
7355 * likely way too fragmented for the clustering stuff to find anything.
7357 * However, if the cluster is taken from the current block group,
7358 * release the cluster first, so that we stand a better chance of
7359 * succeeding in the unclustered allocation.
7361 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7362 spin_unlock(&last_ptr->refill_lock);
7363 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7367 /* This cluster didn't work out, free it and start over */
7368 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7370 if (cluster_bg != bg)
7371 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7374 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7375 spin_unlock(&last_ptr->refill_lock);
7379 aligned_cluster = max_t(u64,
7380 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7381 bg->full_stripe_len);
7382 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7383 ffe_ctl->search_start, ffe_ctl->num_bytes,
7386 /* Now pull our allocation out of this cluster */
7387 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7388 ffe_ctl->num_bytes, ffe_ctl->search_start,
7389 &ffe_ctl->max_extent_size);
7391 /* We found one, proceed */
7392 spin_unlock(&last_ptr->refill_lock);
7393 trace_btrfs_reserve_extent_cluster(bg,
7394 ffe_ctl->search_start,
7395 ffe_ctl->num_bytes);
7396 ffe_ctl->found_offset = offset;
7399 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7400 !ffe_ctl->retry_clustered) {
7401 spin_unlock(&last_ptr->refill_lock);
7403 ffe_ctl->retry_clustered = true;
7404 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7405 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7409 * At this point we either didn't find a cluster or we weren't able to
7410 * allocate a block from our cluster. Free the cluster we've been
7411 * trying to use, and go to the next block group.
7413 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7414 spin_unlock(&last_ptr->refill_lock);
7419 * Return >0 to inform caller that we find nothing
7420 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7421 * Return -EAGAIN to inform caller that we need to re-search this block group
7423 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7424 struct btrfs_free_cluster *last_ptr,
7425 struct find_free_extent_ctl *ffe_ctl)
7430 * We are doing an unclustered allocation, set the fragmented flag so
7431 * we don't bother trying to setup a cluster again until we get more
7434 if (unlikely(last_ptr)) {
7435 spin_lock(&last_ptr->lock);
7436 last_ptr->fragmented = 1;
7437 spin_unlock(&last_ptr->lock);
7439 if (ffe_ctl->cached) {
7440 struct btrfs_free_space_ctl *free_space_ctl;
7442 free_space_ctl = bg->free_space_ctl;
7443 spin_lock(&free_space_ctl->tree_lock);
7444 if (free_space_ctl->free_space <
7445 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7446 ffe_ctl->empty_size) {
7447 ffe_ctl->total_free_space = max_t(u64,
7448 ffe_ctl->total_free_space,
7449 free_space_ctl->free_space);
7450 spin_unlock(&free_space_ctl->tree_lock);
7453 spin_unlock(&free_space_ctl->tree_lock);
7456 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7457 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7458 &ffe_ctl->max_extent_size);
7461 * If we didn't find a chunk, and we haven't failed on this block group
7462 * before, and this block group is in the middle of caching and we are
7463 * ok with waiting, then go ahead and wait for progress to be made, and
7464 * set @retry_unclustered to true.
7466 * If @retry_unclustered is true then we've already waited on this
7467 * block group once and should move on to the next block group.
7469 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7470 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7471 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7472 ffe_ctl->empty_size);
7473 ffe_ctl->retry_unclustered = true;
7475 } else if (!offset) {
7478 ffe_ctl->found_offset = offset;
7483 * Return >0 means caller needs to re-search for free extent
7484 * Return 0 means we have the needed free extent.
7485 * Return <0 means we failed to locate any free extent.
7487 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7488 struct btrfs_free_cluster *last_ptr,
7489 struct btrfs_key *ins,
7490 struct find_free_extent_ctl *ffe_ctl,
7491 int full_search, bool use_cluster)
7493 struct btrfs_root *root = fs_info->extent_root;
7496 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7497 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7498 ffe_ctl->orig_have_caching_bg = true;
7500 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7501 ffe_ctl->have_caching_bg)
7504 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7507 if (ins->objectid) {
7508 if (!use_cluster && last_ptr) {
7509 spin_lock(&last_ptr->lock);
7510 last_ptr->window_start = ins->objectid;
7511 spin_unlock(&last_ptr->lock);
7517 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7518 * caching kthreads as we move along
7519 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7520 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7521 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7524 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7526 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7528 * We want to skip the LOOP_CACHING_WAIT step if we
7529 * don't have any uncached bgs and we've already done a
7530 * full search through.
7532 if (ffe_ctl->orig_have_caching_bg || !full_search)
7533 ffe_ctl->loop = LOOP_CACHING_WAIT;
7535 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7540 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7541 struct btrfs_trans_handle *trans;
7544 trans = current->journal_info;
7548 trans = btrfs_join_transaction(root);
7550 if (IS_ERR(trans)) {
7551 ret = PTR_ERR(trans);
7555 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7559 * If we can't allocate a new chunk we've already looped
7560 * through at least once, move on to the NO_EMPTY_SIZE
7564 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7566 /* Do not bail out on ENOSPC since we can do more. */
7567 if (ret < 0 && ret != -ENOSPC)
7568 btrfs_abort_transaction(trans, ret);
7572 btrfs_end_transaction(trans);
7577 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7579 * Don't loop again if we already have no empty_size and
7582 if (ffe_ctl->empty_size == 0 &&
7583 ffe_ctl->empty_cluster == 0)
7585 ffe_ctl->empty_size = 0;
7586 ffe_ctl->empty_cluster = 0;
7594 * walks the btree of allocated extents and find a hole of a given size.
7595 * The key ins is changed to record the hole:
7596 * ins->objectid == start position
7597 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7598 * ins->offset == the size of the hole.
7599 * Any available blocks before search_start are skipped.
7601 * If there is no suitable free space, we will record the max size of
7602 * the free space extent currently.
7604 * The overall logic and call chain:
7606 * find_free_extent()
7607 * |- Iterate through all block groups
7608 * | |- Get a valid block group
7609 * | |- Try to do clustered allocation in that block group
7610 * | |- Try to do unclustered allocation in that block group
7611 * | |- Check if the result is valid
7612 * | | |- If valid, then exit
7613 * | |- Jump to next block group
7615 * |- Push harder to find free extents
7616 * |- If not found, re-iterate all block groups
7618 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7619 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7620 u64 hint_byte, struct btrfs_key *ins,
7621 u64 flags, int delalloc)
7624 struct btrfs_free_cluster *last_ptr = NULL;
7625 struct btrfs_block_group_cache *block_group = NULL;
7626 struct find_free_extent_ctl ffe_ctl = {0};
7627 struct btrfs_space_info *space_info;
7628 bool use_cluster = true;
7629 bool full_search = false;
7631 WARN_ON(num_bytes < fs_info->sectorsize);
7633 ffe_ctl.ram_bytes = ram_bytes;
7634 ffe_ctl.num_bytes = num_bytes;
7635 ffe_ctl.empty_size = empty_size;
7636 ffe_ctl.flags = flags;
7637 ffe_ctl.search_start = 0;
7638 ffe_ctl.retry_clustered = false;
7639 ffe_ctl.retry_unclustered = false;
7640 ffe_ctl.delalloc = delalloc;
7641 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7642 ffe_ctl.have_caching_bg = false;
7643 ffe_ctl.orig_have_caching_bg = false;
7644 ffe_ctl.found_offset = 0;
7646 ins->type = BTRFS_EXTENT_ITEM_KEY;
7650 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7652 space_info = __find_space_info(fs_info, flags);
7654 btrfs_err(fs_info, "No space info for %llu", flags);
7659 * If our free space is heavily fragmented we may not be able to make
7660 * big contiguous allocations, so instead of doing the expensive search
7661 * for free space, simply return ENOSPC with our max_extent_size so we
7662 * can go ahead and search for a more manageable chunk.
7664 * If our max_extent_size is large enough for our allocation simply
7665 * disable clustering since we will likely not be able to find enough
7666 * space to create a cluster and induce latency trying.
7668 if (unlikely(space_info->max_extent_size)) {
7669 spin_lock(&space_info->lock);
7670 if (space_info->max_extent_size &&
7671 num_bytes > space_info->max_extent_size) {
7672 ins->offset = space_info->max_extent_size;
7673 spin_unlock(&space_info->lock);
7675 } else if (space_info->max_extent_size) {
7676 use_cluster = false;
7678 spin_unlock(&space_info->lock);
7681 last_ptr = fetch_cluster_info(fs_info, space_info,
7682 &ffe_ctl.empty_cluster);
7684 spin_lock(&last_ptr->lock);
7685 if (last_ptr->block_group)
7686 hint_byte = last_ptr->window_start;
7687 if (last_ptr->fragmented) {
7689 * We still set window_start so we can keep track of the
7690 * last place we found an allocation to try and save
7693 hint_byte = last_ptr->window_start;
7694 use_cluster = false;
7696 spin_unlock(&last_ptr->lock);
7699 ffe_ctl.search_start = max(ffe_ctl.search_start,
7700 first_logical_byte(fs_info, 0));
7701 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7702 if (ffe_ctl.search_start == hint_byte) {
7703 block_group = btrfs_lookup_block_group(fs_info,
7704 ffe_ctl.search_start);
7706 * we don't want to use the block group if it doesn't match our
7707 * allocation bits, or if its not cached.
7709 * However if we are re-searching with an ideal block group
7710 * picked out then we don't care that the block group is cached.
7712 if (block_group && block_group_bits(block_group, flags) &&
7713 block_group->cached != BTRFS_CACHE_NO) {
7714 down_read(&space_info->groups_sem);
7715 if (list_empty(&block_group->list) ||
7718 * someone is removing this block group,
7719 * we can't jump into the have_block_group
7720 * target because our list pointers are not
7723 btrfs_put_block_group(block_group);
7724 up_read(&space_info->groups_sem);
7726 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7727 block_group->flags);
7728 btrfs_lock_block_group(block_group, delalloc);
7729 goto have_block_group;
7731 } else if (block_group) {
7732 btrfs_put_block_group(block_group);
7736 ffe_ctl.have_caching_bg = false;
7737 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7740 down_read(&space_info->groups_sem);
7741 list_for_each_entry(block_group,
7742 &space_info->block_groups[ffe_ctl.index], list) {
7743 /* If the block group is read-only, we can skip it entirely. */
7744 if (unlikely(block_group->ro))
7747 btrfs_grab_block_group(block_group, delalloc);
7748 ffe_ctl.search_start = block_group->key.objectid;
7751 * this can happen if we end up cycling through all the
7752 * raid types, but we want to make sure we only allocate
7753 * for the proper type.
7755 if (!block_group_bits(block_group, flags)) {
7756 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7757 BTRFS_BLOCK_GROUP_RAID1 |
7758 BTRFS_BLOCK_GROUP_RAID5 |
7759 BTRFS_BLOCK_GROUP_RAID6 |
7760 BTRFS_BLOCK_GROUP_RAID10;
7763 * if they asked for extra copies and this block group
7764 * doesn't provide them, bail. This does allow us to
7765 * fill raid0 from raid1.
7767 if ((flags & extra) && !(block_group->flags & extra))
7772 ffe_ctl.cached = block_group_cache_done(block_group);
7773 if (unlikely(!ffe_ctl.cached)) {
7774 ffe_ctl.have_caching_bg = true;
7775 ret = cache_block_group(block_group, 0);
7780 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7784 * Ok we want to try and use the cluster allocator, so
7787 if (last_ptr && use_cluster) {
7788 struct btrfs_block_group_cache *cluster_bg = NULL;
7790 ret = find_free_extent_clustered(block_group, last_ptr,
7791 &ffe_ctl, &cluster_bg);
7794 if (cluster_bg && cluster_bg != block_group) {
7795 btrfs_release_block_group(block_group,
7797 block_group = cluster_bg;
7800 } else if (ret == -EAGAIN) {
7801 goto have_block_group;
7802 } else if (ret > 0) {
7805 /* ret == -ENOENT case falls through */
7808 ret = find_free_extent_unclustered(block_group, last_ptr,
7811 goto have_block_group;
7814 /* ret == 0 case falls through */
7816 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
7817 fs_info->stripesize);
7819 /* move on to the next group */
7820 if (ffe_ctl.search_start + num_bytes >
7821 block_group->key.objectid + block_group->key.offset) {
7822 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7827 if (ffe_ctl.found_offset < ffe_ctl.search_start)
7828 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7829 ffe_ctl.search_start - ffe_ctl.found_offset);
7831 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7832 num_bytes, delalloc);
7833 if (ret == -EAGAIN) {
7834 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7838 btrfs_inc_block_group_reservations(block_group);
7840 /* we are all good, lets return */
7841 ins->objectid = ffe_ctl.search_start;
7842 ins->offset = num_bytes;
7844 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
7846 btrfs_release_block_group(block_group, delalloc);
7849 ffe_ctl.retry_clustered = false;
7850 ffe_ctl.retry_unclustered = false;
7851 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7853 btrfs_release_block_group(block_group, delalloc);
7856 up_read(&space_info->groups_sem);
7858 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
7859 full_search, use_cluster);
7863 if (ret == -ENOSPC) {
7865 * Use ffe_ctl->total_free_space as fallback if we can't find
7866 * any contiguous hole.
7868 if (!ffe_ctl.max_extent_size)
7869 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
7870 spin_lock(&space_info->lock);
7871 space_info->max_extent_size = ffe_ctl.max_extent_size;
7872 spin_unlock(&space_info->lock);
7873 ins->offset = ffe_ctl.max_extent_size;
7878 static void dump_space_info(struct btrfs_fs_info *fs_info,
7879 struct btrfs_space_info *info, u64 bytes,
7880 int dump_block_groups)
7882 struct btrfs_block_group_cache *cache;
7885 spin_lock(&info->lock);
7886 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7888 info->total_bytes - btrfs_space_info_used(info, true),
7889 info->full ? "" : "not ");
7891 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7892 info->total_bytes, info->bytes_used, info->bytes_pinned,
7893 info->bytes_reserved, info->bytes_may_use,
7894 info->bytes_readonly);
7895 spin_unlock(&info->lock);
7897 if (!dump_block_groups)
7900 down_read(&info->groups_sem);
7902 list_for_each_entry(cache, &info->block_groups[index], list) {
7903 spin_lock(&cache->lock);
7905 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7906 cache->key.objectid, cache->key.offset,
7907 btrfs_block_group_used(&cache->item), cache->pinned,
7908 cache->reserved, cache->ro ? "[readonly]" : "");
7909 btrfs_dump_free_space(cache, bytes);
7910 spin_unlock(&cache->lock);
7912 if (++index < BTRFS_NR_RAID_TYPES)
7914 up_read(&info->groups_sem);
7918 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7919 * hole that is at least as big as @num_bytes.
7921 * @root - The root that will contain this extent
7923 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7924 * is used for accounting purposes. This value differs
7925 * from @num_bytes only in the case of compressed extents.
7927 * @num_bytes - Number of bytes to allocate on-disk.
7929 * @min_alloc_size - Indicates the minimum amount of space that the
7930 * allocator should try to satisfy. In some cases
7931 * @num_bytes may be larger than what is required and if
7932 * the filesystem is fragmented then allocation fails.
7933 * However, the presence of @min_alloc_size gives a
7934 * chance to try and satisfy the smaller allocation.
7936 * @empty_size - A hint that you plan on doing more COW. This is the
7937 * size in bytes the allocator should try to find free
7938 * next to the block it returns. This is just a hint and
7939 * may be ignored by the allocator.
7941 * @hint_byte - Hint to the allocator to start searching above the byte
7942 * address passed. It might be ignored.
7944 * @ins - This key is modified to record the found hole. It will
7945 * have the following values:
7946 * ins->objectid == start position
7947 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7948 * ins->offset == the size of the hole.
7950 * @is_data - Boolean flag indicating whether an extent is
7951 * allocated for data (true) or metadata (false)
7953 * @delalloc - Boolean flag indicating whether this allocation is for
7954 * delalloc or not. If 'true' data_rwsem of block groups
7955 * is going to be acquired.
7958 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7959 * case -ENOSPC is returned then @ins->offset will contain the size of the
7960 * largest available hole the allocator managed to find.
7962 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7963 u64 num_bytes, u64 min_alloc_size,
7964 u64 empty_size, u64 hint_byte,
7965 struct btrfs_key *ins, int is_data, int delalloc)
7967 struct btrfs_fs_info *fs_info = root->fs_info;
7968 bool final_tried = num_bytes == min_alloc_size;
7972 flags = get_alloc_profile_by_root(root, is_data);
7974 WARN_ON(num_bytes < fs_info->sectorsize);
7975 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7976 hint_byte, ins, flags, delalloc);
7977 if (!ret && !is_data) {
7978 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7979 } else if (ret == -ENOSPC) {
7980 if (!final_tried && ins->offset) {
7981 num_bytes = min(num_bytes >> 1, ins->offset);
7982 num_bytes = round_down(num_bytes,
7983 fs_info->sectorsize);
7984 num_bytes = max(num_bytes, min_alloc_size);
7985 ram_bytes = num_bytes;
7986 if (num_bytes == min_alloc_size)
7989 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7990 struct btrfs_space_info *sinfo;
7992 sinfo = __find_space_info(fs_info, flags);
7994 "allocation failed flags %llu, wanted %llu",
7997 dump_space_info(fs_info, sinfo, num_bytes, 1);
8004 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8006 int pin, int delalloc)
8008 struct btrfs_block_group_cache *cache;
8011 cache = btrfs_lookup_block_group(fs_info, start);
8013 btrfs_err(fs_info, "Unable to find block group for %llu",
8019 pin_down_extent(fs_info, cache, start, len, 1);
8021 if (btrfs_test_opt(fs_info, DISCARD))
8022 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8023 btrfs_add_free_space(cache, start, len);
8024 btrfs_free_reserved_bytes(cache, len, delalloc);
8025 trace_btrfs_reserved_extent_free(fs_info, start, len);
8028 btrfs_put_block_group(cache);
8032 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8033 u64 start, u64 len, int delalloc)
8035 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8038 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8041 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8044 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8045 u64 parent, u64 root_objectid,
8046 u64 flags, u64 owner, u64 offset,
8047 struct btrfs_key *ins, int ref_mod)
8049 struct btrfs_fs_info *fs_info = trans->fs_info;
8051 struct btrfs_extent_item *extent_item;
8052 struct btrfs_extent_inline_ref *iref;
8053 struct btrfs_path *path;
8054 struct extent_buffer *leaf;
8059 type = BTRFS_SHARED_DATA_REF_KEY;
8061 type = BTRFS_EXTENT_DATA_REF_KEY;
8063 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8065 path = btrfs_alloc_path();
8069 path->leave_spinning = 1;
8070 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8073 btrfs_free_path(path);
8077 leaf = path->nodes[0];
8078 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8079 struct btrfs_extent_item);
8080 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8081 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8082 btrfs_set_extent_flags(leaf, extent_item,
8083 flags | BTRFS_EXTENT_FLAG_DATA);
8085 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8086 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8088 struct btrfs_shared_data_ref *ref;
8089 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8090 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8091 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8093 struct btrfs_extent_data_ref *ref;
8094 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8095 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8096 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8097 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8098 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8101 btrfs_mark_buffer_dirty(path->nodes[0]);
8102 btrfs_free_path(path);
8104 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8108 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8109 if (ret) { /* -ENOENT, logic error */
8110 btrfs_err(fs_info, "update block group failed for %llu %llu",
8111 ins->objectid, ins->offset);
8114 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8118 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8119 struct btrfs_delayed_ref_node *node,
8120 struct btrfs_delayed_extent_op *extent_op)
8122 struct btrfs_fs_info *fs_info = trans->fs_info;
8124 struct btrfs_extent_item *extent_item;
8125 struct btrfs_key extent_key;
8126 struct btrfs_tree_block_info *block_info;
8127 struct btrfs_extent_inline_ref *iref;
8128 struct btrfs_path *path;
8129 struct extent_buffer *leaf;
8130 struct btrfs_delayed_tree_ref *ref;
8131 u32 size = sizeof(*extent_item) + sizeof(*iref);
8133 u64 flags = extent_op->flags_to_set;
8134 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8136 ref = btrfs_delayed_node_to_tree_ref(node);
8138 extent_key.objectid = node->bytenr;
8139 if (skinny_metadata) {
8140 extent_key.offset = ref->level;
8141 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8142 num_bytes = fs_info->nodesize;
8144 extent_key.offset = node->num_bytes;
8145 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8146 size += sizeof(*block_info);
8147 num_bytes = node->num_bytes;
8150 path = btrfs_alloc_path();
8154 path->leave_spinning = 1;
8155 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8158 btrfs_free_path(path);
8162 leaf = path->nodes[0];
8163 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8164 struct btrfs_extent_item);
8165 btrfs_set_extent_refs(leaf, extent_item, 1);
8166 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8167 btrfs_set_extent_flags(leaf, extent_item,
8168 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8170 if (skinny_metadata) {
8171 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8173 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8174 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8175 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8176 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8179 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8180 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8181 btrfs_set_extent_inline_ref_type(leaf, iref,
8182 BTRFS_SHARED_BLOCK_REF_KEY);
8183 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8185 btrfs_set_extent_inline_ref_type(leaf, iref,
8186 BTRFS_TREE_BLOCK_REF_KEY);
8187 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8190 btrfs_mark_buffer_dirty(leaf);
8191 btrfs_free_path(path);
8193 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8198 ret = update_block_group(trans, fs_info, extent_key.objectid,
8199 fs_info->nodesize, 1);
8200 if (ret) { /* -ENOENT, logic error */
8201 btrfs_err(fs_info, "update block group failed for %llu %llu",
8202 extent_key.objectid, extent_key.offset);
8206 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8211 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8212 struct btrfs_root *root, u64 owner,
8213 u64 offset, u64 ram_bytes,
8214 struct btrfs_key *ins)
8218 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8220 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8221 root->root_key.objectid, owner, offset,
8222 BTRFS_ADD_DELAYED_EXTENT);
8224 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8226 root->root_key.objectid, owner,
8228 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8233 * this is used by the tree logging recovery code. It records that
8234 * an extent has been allocated and makes sure to clear the free
8235 * space cache bits as well
8237 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8238 u64 root_objectid, u64 owner, u64 offset,
8239 struct btrfs_key *ins)
8241 struct btrfs_fs_info *fs_info = trans->fs_info;
8243 struct btrfs_block_group_cache *block_group;
8244 struct btrfs_space_info *space_info;
8247 * Mixed block groups will exclude before processing the log so we only
8248 * need to do the exclude dance if this fs isn't mixed.
8250 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8251 ret = __exclude_logged_extent(fs_info, ins->objectid,
8257 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8261 space_info = block_group->space_info;
8262 spin_lock(&space_info->lock);
8263 spin_lock(&block_group->lock);
8264 space_info->bytes_reserved += ins->offset;
8265 block_group->reserved += ins->offset;
8266 spin_unlock(&block_group->lock);
8267 spin_unlock(&space_info->lock);
8269 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8271 btrfs_put_block_group(block_group);
8275 static struct extent_buffer *
8276 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8277 u64 bytenr, int level, u64 owner)
8279 struct btrfs_fs_info *fs_info = root->fs_info;
8280 struct extent_buffer *buf;
8282 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8287 * Extra safety check in case the extent tree is corrupted and extent
8288 * allocator chooses to use a tree block which is already used and
8291 if (buf->lock_owner == current->pid) {
8292 btrfs_err_rl(fs_info,
8293 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8294 buf->start, btrfs_header_owner(buf), current->pid);
8295 free_extent_buffer(buf);
8296 return ERR_PTR(-EUCLEAN);
8299 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8300 btrfs_tree_lock(buf);
8301 clean_tree_block(fs_info, buf);
8302 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8304 btrfs_set_lock_blocking(buf);
8305 set_extent_buffer_uptodate(buf);
8307 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8308 btrfs_set_header_level(buf, level);
8309 btrfs_set_header_bytenr(buf, buf->start);
8310 btrfs_set_header_generation(buf, trans->transid);
8311 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8312 btrfs_set_header_owner(buf, owner);
8313 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8314 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8315 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8316 buf->log_index = root->log_transid % 2;
8318 * we allow two log transactions at a time, use different
8319 * EXENT bit to differentiate dirty pages.
8321 if (buf->log_index == 0)
8322 set_extent_dirty(&root->dirty_log_pages, buf->start,
8323 buf->start + buf->len - 1, GFP_NOFS);
8325 set_extent_new(&root->dirty_log_pages, buf->start,
8326 buf->start + buf->len - 1);
8328 buf->log_index = -1;
8329 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8330 buf->start + buf->len - 1, GFP_NOFS);
8332 trans->dirty = true;
8333 /* this returns a buffer locked for blocking */
8337 static struct btrfs_block_rsv *
8338 use_block_rsv(struct btrfs_trans_handle *trans,
8339 struct btrfs_root *root, u32 blocksize)
8341 struct btrfs_fs_info *fs_info = root->fs_info;
8342 struct btrfs_block_rsv *block_rsv;
8343 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8345 bool global_updated = false;
8347 block_rsv = get_block_rsv(trans, root);
8349 if (unlikely(block_rsv->size == 0))
8352 ret = block_rsv_use_bytes(block_rsv, blocksize);
8356 if (block_rsv->failfast)
8357 return ERR_PTR(ret);
8359 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8360 global_updated = true;
8361 update_global_block_rsv(fs_info);
8365 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8366 static DEFINE_RATELIMIT_STATE(_rs,
8367 DEFAULT_RATELIMIT_INTERVAL * 10,
8368 /*DEFAULT_RATELIMIT_BURST*/ 1);
8369 if (__ratelimit(&_rs))
8371 "BTRFS: block rsv returned %d\n", ret);
8374 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8375 BTRFS_RESERVE_NO_FLUSH);
8379 * If we couldn't reserve metadata bytes try and use some from
8380 * the global reserve if its space type is the same as the global
8383 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8384 block_rsv->space_info == global_rsv->space_info) {
8385 ret = block_rsv_use_bytes(global_rsv, blocksize);
8389 return ERR_PTR(ret);
8392 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8393 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8395 block_rsv_add_bytes(block_rsv, blocksize, false);
8396 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8400 * finds a free extent and does all the dirty work required for allocation
8401 * returns the tree buffer or an ERR_PTR on error.
8403 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8404 struct btrfs_root *root,
8405 u64 parent, u64 root_objectid,
8406 const struct btrfs_disk_key *key,
8407 int level, u64 hint,
8410 struct btrfs_fs_info *fs_info = root->fs_info;
8411 struct btrfs_key ins;
8412 struct btrfs_block_rsv *block_rsv;
8413 struct extent_buffer *buf;
8414 struct btrfs_delayed_extent_op *extent_op;
8417 u32 blocksize = fs_info->nodesize;
8418 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8420 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8421 if (btrfs_is_testing(fs_info)) {
8422 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8423 level, root_objectid);
8425 root->alloc_bytenr += blocksize;
8430 block_rsv = use_block_rsv(trans, root, blocksize);
8431 if (IS_ERR(block_rsv))
8432 return ERR_CAST(block_rsv);
8434 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8435 empty_size, hint, &ins, 0, 0);
8439 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8443 goto out_free_reserved;
8446 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8448 parent = ins.objectid;
8449 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8453 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8454 extent_op = btrfs_alloc_delayed_extent_op();
8460 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8462 memset(&extent_op->key, 0, sizeof(extent_op->key));
8463 extent_op->flags_to_set = flags;
8464 extent_op->update_key = skinny_metadata ? false : true;
8465 extent_op->update_flags = true;
8466 extent_op->is_data = false;
8467 extent_op->level = level;
8469 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8470 root_objectid, level, 0,
8471 BTRFS_ADD_DELAYED_EXTENT);
8472 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8474 root_objectid, level,
8475 BTRFS_ADD_DELAYED_EXTENT,
8476 extent_op, NULL, NULL);
8478 goto out_free_delayed;
8483 btrfs_free_delayed_extent_op(extent_op);
8485 free_extent_buffer(buf);
8487 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8489 unuse_block_rsv(fs_info, block_rsv, blocksize);
8490 return ERR_PTR(ret);
8493 struct walk_control {
8494 u64 refs[BTRFS_MAX_LEVEL];
8495 u64 flags[BTRFS_MAX_LEVEL];
8496 struct btrfs_key update_progress;
8506 #define DROP_REFERENCE 1
8507 #define UPDATE_BACKREF 2
8509 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8510 struct btrfs_root *root,
8511 struct walk_control *wc,
8512 struct btrfs_path *path)
8514 struct btrfs_fs_info *fs_info = root->fs_info;
8520 struct btrfs_key key;
8521 struct extent_buffer *eb;
8526 if (path->slots[wc->level] < wc->reada_slot) {
8527 wc->reada_count = wc->reada_count * 2 / 3;
8528 wc->reada_count = max(wc->reada_count, 2);
8530 wc->reada_count = wc->reada_count * 3 / 2;
8531 wc->reada_count = min_t(int, wc->reada_count,
8532 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8535 eb = path->nodes[wc->level];
8536 nritems = btrfs_header_nritems(eb);
8538 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8539 if (nread >= wc->reada_count)
8543 bytenr = btrfs_node_blockptr(eb, slot);
8544 generation = btrfs_node_ptr_generation(eb, slot);
8546 if (slot == path->slots[wc->level])
8549 if (wc->stage == UPDATE_BACKREF &&
8550 generation <= root->root_key.offset)
8553 /* We don't lock the tree block, it's OK to be racy here */
8554 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8555 wc->level - 1, 1, &refs,
8557 /* We don't care about errors in readahead. */
8562 if (wc->stage == DROP_REFERENCE) {
8566 if (wc->level == 1 &&
8567 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8569 if (!wc->update_ref ||
8570 generation <= root->root_key.offset)
8572 btrfs_node_key_to_cpu(eb, &key, slot);
8573 ret = btrfs_comp_cpu_keys(&key,
8574 &wc->update_progress);
8578 if (wc->level == 1 &&
8579 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8583 readahead_tree_block(fs_info, bytenr);
8586 wc->reada_slot = slot;
8590 * helper to process tree block while walking down the tree.
8592 * when wc->stage == UPDATE_BACKREF, this function updates
8593 * back refs for pointers in the block.
8595 * NOTE: return value 1 means we should stop walking down.
8597 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8598 struct btrfs_root *root,
8599 struct btrfs_path *path,
8600 struct walk_control *wc, int lookup_info)
8602 struct btrfs_fs_info *fs_info = root->fs_info;
8603 int level = wc->level;
8604 struct extent_buffer *eb = path->nodes[level];
8605 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8608 if (wc->stage == UPDATE_BACKREF &&
8609 btrfs_header_owner(eb) != root->root_key.objectid)
8613 * when reference count of tree block is 1, it won't increase
8614 * again. once full backref flag is set, we never clear it.
8617 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8618 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8619 BUG_ON(!path->locks[level]);
8620 ret = btrfs_lookup_extent_info(trans, fs_info,
8621 eb->start, level, 1,
8624 BUG_ON(ret == -ENOMEM);
8627 BUG_ON(wc->refs[level] == 0);
8630 if (wc->stage == DROP_REFERENCE) {
8631 if (wc->refs[level] > 1)
8634 if (path->locks[level] && !wc->keep_locks) {
8635 btrfs_tree_unlock_rw(eb, path->locks[level]);
8636 path->locks[level] = 0;
8641 /* wc->stage == UPDATE_BACKREF */
8642 if (!(wc->flags[level] & flag)) {
8643 BUG_ON(!path->locks[level]);
8644 ret = btrfs_inc_ref(trans, root, eb, 1);
8645 BUG_ON(ret); /* -ENOMEM */
8646 ret = btrfs_dec_ref(trans, root, eb, 0);
8647 BUG_ON(ret); /* -ENOMEM */
8648 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8650 btrfs_header_level(eb), 0);
8651 BUG_ON(ret); /* -ENOMEM */
8652 wc->flags[level] |= flag;
8656 * the block is shared by multiple trees, so it's not good to
8657 * keep the tree lock
8659 if (path->locks[level] && level > 0) {
8660 btrfs_tree_unlock_rw(eb, path->locks[level]);
8661 path->locks[level] = 0;
8667 * helper to process tree block pointer.
8669 * when wc->stage == DROP_REFERENCE, this function checks
8670 * reference count of the block pointed to. if the block
8671 * is shared and we need update back refs for the subtree
8672 * rooted at the block, this function changes wc->stage to
8673 * UPDATE_BACKREF. if the block is shared and there is no
8674 * need to update back, this function drops the reference
8677 * NOTE: return value 1 means we should stop walking down.
8679 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8680 struct btrfs_root *root,
8681 struct btrfs_path *path,
8682 struct walk_control *wc, int *lookup_info)
8684 struct btrfs_fs_info *fs_info = root->fs_info;
8689 struct btrfs_key key;
8690 struct btrfs_key first_key;
8691 struct extent_buffer *next;
8692 int level = wc->level;
8695 bool need_account = false;
8697 generation = btrfs_node_ptr_generation(path->nodes[level],
8698 path->slots[level]);
8700 * if the lower level block was created before the snapshot
8701 * was created, we know there is no need to update back refs
8704 if (wc->stage == UPDATE_BACKREF &&
8705 generation <= root->root_key.offset) {
8710 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8711 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8712 path->slots[level]);
8713 blocksize = fs_info->nodesize;
8715 next = find_extent_buffer(fs_info, bytenr);
8717 next = btrfs_find_create_tree_block(fs_info, bytenr);
8719 return PTR_ERR(next);
8721 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8725 btrfs_tree_lock(next);
8726 btrfs_set_lock_blocking(next);
8728 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8729 &wc->refs[level - 1],
8730 &wc->flags[level - 1]);
8734 if (unlikely(wc->refs[level - 1] == 0)) {
8735 btrfs_err(fs_info, "Missing references.");
8741 if (wc->stage == DROP_REFERENCE) {
8742 if (wc->refs[level - 1] > 1) {
8743 need_account = true;
8745 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8748 if (!wc->update_ref ||
8749 generation <= root->root_key.offset)
8752 btrfs_node_key_to_cpu(path->nodes[level], &key,
8753 path->slots[level]);
8754 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8758 wc->stage = UPDATE_BACKREF;
8759 wc->shared_level = level - 1;
8763 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8767 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8768 btrfs_tree_unlock(next);
8769 free_extent_buffer(next);
8775 if (reada && level == 1)
8776 reada_walk_down(trans, root, wc, path);
8777 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8780 return PTR_ERR(next);
8781 } else if (!extent_buffer_uptodate(next)) {
8782 free_extent_buffer(next);
8785 btrfs_tree_lock(next);
8786 btrfs_set_lock_blocking(next);
8790 ASSERT(level == btrfs_header_level(next));
8791 if (level != btrfs_header_level(next)) {
8792 btrfs_err(root->fs_info, "mismatched level");
8796 path->nodes[level] = next;
8797 path->slots[level] = 0;
8798 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8804 wc->refs[level - 1] = 0;
8805 wc->flags[level - 1] = 0;
8806 if (wc->stage == DROP_REFERENCE) {
8807 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8808 parent = path->nodes[level]->start;
8810 ASSERT(root->root_key.objectid ==
8811 btrfs_header_owner(path->nodes[level]));
8812 if (root->root_key.objectid !=
8813 btrfs_header_owner(path->nodes[level])) {
8814 btrfs_err(root->fs_info,
8815 "mismatched block owner");
8823 * Reloc tree doesn't contribute to qgroup numbers, and we have
8824 * already accounted them at merge time (replace_path),
8825 * thus we could skip expensive subtree trace here.
8827 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
8829 ret = btrfs_qgroup_trace_subtree(trans, next,
8830 generation, level - 1);
8832 btrfs_err_rl(fs_info,
8833 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8837 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8838 parent, root->root_key.objectid,
8848 btrfs_tree_unlock(next);
8849 free_extent_buffer(next);
8855 * helper to process tree block while walking up the tree.
8857 * when wc->stage == DROP_REFERENCE, this function drops
8858 * reference count on the block.
8860 * when wc->stage == UPDATE_BACKREF, this function changes
8861 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8862 * to UPDATE_BACKREF previously while processing the block.
8864 * NOTE: return value 1 means we should stop walking up.
8866 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8867 struct btrfs_root *root,
8868 struct btrfs_path *path,
8869 struct walk_control *wc)
8871 struct btrfs_fs_info *fs_info = root->fs_info;
8873 int level = wc->level;
8874 struct extent_buffer *eb = path->nodes[level];
8877 if (wc->stage == UPDATE_BACKREF) {
8878 BUG_ON(wc->shared_level < level);
8879 if (level < wc->shared_level)
8882 ret = find_next_key(path, level + 1, &wc->update_progress);
8886 wc->stage = DROP_REFERENCE;
8887 wc->shared_level = -1;
8888 path->slots[level] = 0;
8891 * check reference count again if the block isn't locked.
8892 * we should start walking down the tree again if reference
8895 if (!path->locks[level]) {
8897 btrfs_tree_lock(eb);
8898 btrfs_set_lock_blocking(eb);
8899 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8901 ret = btrfs_lookup_extent_info(trans, fs_info,
8902 eb->start, level, 1,
8906 btrfs_tree_unlock_rw(eb, path->locks[level]);
8907 path->locks[level] = 0;
8910 BUG_ON(wc->refs[level] == 0);
8911 if (wc->refs[level] == 1) {
8912 btrfs_tree_unlock_rw(eb, path->locks[level]);
8913 path->locks[level] = 0;
8919 /* wc->stage == DROP_REFERENCE */
8920 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8922 if (wc->refs[level] == 1) {
8924 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8925 ret = btrfs_dec_ref(trans, root, eb, 1);
8927 ret = btrfs_dec_ref(trans, root, eb, 0);
8928 BUG_ON(ret); /* -ENOMEM */
8929 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
8931 btrfs_err_rl(fs_info,
8932 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8936 /* make block locked assertion in clean_tree_block happy */
8937 if (!path->locks[level] &&
8938 btrfs_header_generation(eb) == trans->transid) {
8939 btrfs_tree_lock(eb);
8940 btrfs_set_lock_blocking(eb);
8941 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8943 clean_tree_block(fs_info, eb);
8946 if (eb == root->node) {
8947 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8949 else if (root->root_key.objectid != btrfs_header_owner(eb))
8950 goto owner_mismatch;
8952 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8953 parent = path->nodes[level + 1]->start;
8954 else if (root->root_key.objectid !=
8955 btrfs_header_owner(path->nodes[level + 1]))
8956 goto owner_mismatch;
8959 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8961 wc->refs[level] = 0;
8962 wc->flags[level] = 0;
8966 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
8967 btrfs_header_owner(eb), root->root_key.objectid);
8971 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8972 struct btrfs_root *root,
8973 struct btrfs_path *path,
8974 struct walk_control *wc)
8976 int level = wc->level;
8977 int lookup_info = 1;
8980 while (level >= 0) {
8981 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8988 if (path->slots[level] >=
8989 btrfs_header_nritems(path->nodes[level]))
8992 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8994 path->slots[level]++;
9003 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9004 struct btrfs_root *root,
9005 struct btrfs_path *path,
9006 struct walk_control *wc, int max_level)
9008 int level = wc->level;
9011 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9012 while (level < max_level && path->nodes[level]) {
9014 if (path->slots[level] + 1 <
9015 btrfs_header_nritems(path->nodes[level])) {
9016 path->slots[level]++;
9019 ret = walk_up_proc(trans, root, path, wc);
9025 if (path->locks[level]) {
9026 btrfs_tree_unlock_rw(path->nodes[level],
9027 path->locks[level]);
9028 path->locks[level] = 0;
9030 free_extent_buffer(path->nodes[level]);
9031 path->nodes[level] = NULL;
9039 * drop a subvolume tree.
9041 * this function traverses the tree freeing any blocks that only
9042 * referenced by the tree.
9044 * when a shared tree block is found. this function decreases its
9045 * reference count by one. if update_ref is true, this function
9046 * also make sure backrefs for the shared block and all lower level
9047 * blocks are properly updated.
9049 * If called with for_reloc == 0, may exit early with -EAGAIN
9051 int btrfs_drop_snapshot(struct btrfs_root *root,
9052 struct btrfs_block_rsv *block_rsv, int update_ref,
9055 struct btrfs_fs_info *fs_info = root->fs_info;
9056 struct btrfs_path *path;
9057 struct btrfs_trans_handle *trans;
9058 struct btrfs_root *tree_root = fs_info->tree_root;
9059 struct btrfs_root_item *root_item = &root->root_item;
9060 struct walk_control *wc;
9061 struct btrfs_key key;
9065 bool root_dropped = false;
9067 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9069 path = btrfs_alloc_path();
9075 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9077 btrfs_free_path(path);
9082 trans = btrfs_start_transaction(tree_root, 0);
9083 if (IS_ERR(trans)) {
9084 err = PTR_ERR(trans);
9089 trans->block_rsv = block_rsv;
9091 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9092 level = btrfs_header_level(root->node);
9093 path->nodes[level] = btrfs_lock_root_node(root);
9094 btrfs_set_lock_blocking(path->nodes[level]);
9095 path->slots[level] = 0;
9096 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9097 memset(&wc->update_progress, 0,
9098 sizeof(wc->update_progress));
9100 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9101 memcpy(&wc->update_progress, &key,
9102 sizeof(wc->update_progress));
9104 level = root_item->drop_level;
9106 path->lowest_level = level;
9107 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9108 path->lowest_level = 0;
9116 * unlock our path, this is safe because only this
9117 * function is allowed to delete this snapshot
9119 btrfs_unlock_up_safe(path, 0);
9121 level = btrfs_header_level(root->node);
9123 btrfs_tree_lock(path->nodes[level]);
9124 btrfs_set_lock_blocking(path->nodes[level]);
9125 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9127 ret = btrfs_lookup_extent_info(trans, fs_info,
9128 path->nodes[level]->start,
9129 level, 1, &wc->refs[level],
9135 BUG_ON(wc->refs[level] == 0);
9137 if (level == root_item->drop_level)
9140 btrfs_tree_unlock(path->nodes[level]);
9141 path->locks[level] = 0;
9142 WARN_ON(wc->refs[level] != 1);
9148 wc->shared_level = -1;
9149 wc->stage = DROP_REFERENCE;
9150 wc->update_ref = update_ref;
9152 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9156 ret = walk_down_tree(trans, root, path, wc);
9162 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9169 BUG_ON(wc->stage != DROP_REFERENCE);
9173 if (wc->stage == DROP_REFERENCE) {
9175 btrfs_node_key(path->nodes[level],
9176 &root_item->drop_progress,
9177 path->slots[level]);
9178 root_item->drop_level = level;
9181 BUG_ON(wc->level == 0);
9182 if (btrfs_should_end_transaction(trans) ||
9183 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9184 ret = btrfs_update_root(trans, tree_root,
9188 btrfs_abort_transaction(trans, ret);
9193 btrfs_end_transaction_throttle(trans);
9194 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9195 btrfs_debug(fs_info,
9196 "drop snapshot early exit");
9201 trans = btrfs_start_transaction(tree_root, 0);
9202 if (IS_ERR(trans)) {
9203 err = PTR_ERR(trans);
9207 trans->block_rsv = block_rsv;
9210 btrfs_release_path(path);
9214 ret = btrfs_del_root(trans, &root->root_key);
9216 btrfs_abort_transaction(trans, ret);
9221 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9222 ret = btrfs_find_root(tree_root, &root->root_key, path,
9225 btrfs_abort_transaction(trans, ret);
9228 } else if (ret > 0) {
9229 /* if we fail to delete the orphan item this time
9230 * around, it'll get picked up the next time.
9232 * The most common failure here is just -ENOENT.
9234 btrfs_del_orphan_item(trans, tree_root,
9235 root->root_key.objectid);
9239 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9240 btrfs_add_dropped_root(trans, root);
9242 free_extent_buffer(root->node);
9243 free_extent_buffer(root->commit_root);
9244 btrfs_put_fs_root(root);
9246 root_dropped = true;
9248 btrfs_end_transaction_throttle(trans);
9251 btrfs_free_path(path);
9254 * So if we need to stop dropping the snapshot for whatever reason we
9255 * need to make sure to add it back to the dead root list so that we
9256 * keep trying to do the work later. This also cleans up roots if we
9257 * don't have it in the radix (like when we recover after a power fail
9258 * or unmount) so we don't leak memory.
9260 if (!for_reloc && !root_dropped)
9261 btrfs_add_dead_root(root);
9262 if (err && err != -EAGAIN)
9263 btrfs_handle_fs_error(fs_info, err, NULL);
9268 * drop subtree rooted at tree block 'node'.
9270 * NOTE: this function will unlock and release tree block 'node'
9271 * only used by relocation code
9273 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9274 struct btrfs_root *root,
9275 struct extent_buffer *node,
9276 struct extent_buffer *parent)
9278 struct btrfs_fs_info *fs_info = root->fs_info;
9279 struct btrfs_path *path;
9280 struct walk_control *wc;
9286 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9288 path = btrfs_alloc_path();
9292 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9294 btrfs_free_path(path);
9298 btrfs_assert_tree_locked(parent);
9299 parent_level = btrfs_header_level(parent);
9300 extent_buffer_get(parent);
9301 path->nodes[parent_level] = parent;
9302 path->slots[parent_level] = btrfs_header_nritems(parent);
9304 btrfs_assert_tree_locked(node);
9305 level = btrfs_header_level(node);
9306 path->nodes[level] = node;
9307 path->slots[level] = 0;
9308 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9310 wc->refs[parent_level] = 1;
9311 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9313 wc->shared_level = -1;
9314 wc->stage = DROP_REFERENCE;
9317 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9320 wret = walk_down_tree(trans, root, path, wc);
9326 wret = walk_up_tree(trans, root, path, wc, parent_level);
9334 btrfs_free_path(path);
9338 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9344 * if restripe for this chunk_type is on pick target profile and
9345 * return, otherwise do the usual balance
9347 stripped = get_restripe_target(fs_info, flags);
9349 return extended_to_chunk(stripped);
9351 num_devices = fs_info->fs_devices->rw_devices;
9353 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9354 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9355 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9357 if (num_devices == 1) {
9358 stripped |= BTRFS_BLOCK_GROUP_DUP;
9359 stripped = flags & ~stripped;
9361 /* turn raid0 into single device chunks */
9362 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9365 /* turn mirroring into duplication */
9366 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9367 BTRFS_BLOCK_GROUP_RAID10))
9368 return stripped | BTRFS_BLOCK_GROUP_DUP;
9370 /* they already had raid on here, just return */
9371 if (flags & stripped)
9374 stripped |= BTRFS_BLOCK_GROUP_DUP;
9375 stripped = flags & ~stripped;
9377 /* switch duplicated blocks with raid1 */
9378 if (flags & BTRFS_BLOCK_GROUP_DUP)
9379 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9381 /* this is drive concat, leave it alone */
9387 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9389 struct btrfs_space_info *sinfo = cache->space_info;
9391 u64 min_allocable_bytes;
9395 * We need some metadata space and system metadata space for
9396 * allocating chunks in some corner cases until we force to set
9397 * it to be readonly.
9400 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9402 min_allocable_bytes = SZ_1M;
9404 min_allocable_bytes = 0;
9406 spin_lock(&sinfo->lock);
9407 spin_lock(&cache->lock);
9415 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9416 cache->bytes_super - btrfs_block_group_used(&cache->item);
9418 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9419 min_allocable_bytes <= sinfo->total_bytes) {
9420 sinfo->bytes_readonly += num_bytes;
9422 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9426 spin_unlock(&cache->lock);
9427 spin_unlock(&sinfo->lock);
9431 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9434 struct btrfs_fs_info *fs_info = cache->fs_info;
9435 struct btrfs_trans_handle *trans;
9440 trans = btrfs_join_transaction(fs_info->extent_root);
9442 return PTR_ERR(trans);
9445 * we're not allowed to set block groups readonly after the dirty
9446 * block groups cache has started writing. If it already started,
9447 * back off and let this transaction commit
9449 mutex_lock(&fs_info->ro_block_group_mutex);
9450 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9451 u64 transid = trans->transid;
9453 mutex_unlock(&fs_info->ro_block_group_mutex);
9454 btrfs_end_transaction(trans);
9456 ret = btrfs_wait_for_commit(fs_info, transid);
9463 * if we are changing raid levels, try to allocate a corresponding
9464 * block group with the new raid level.
9466 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9467 if (alloc_flags != cache->flags) {
9468 ret = do_chunk_alloc(trans, alloc_flags,
9471 * ENOSPC is allowed here, we may have enough space
9472 * already allocated at the new raid level to
9481 ret = inc_block_group_ro(cache, 0);
9484 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9485 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9488 ret = inc_block_group_ro(cache, 0);
9490 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9491 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9492 mutex_lock(&fs_info->chunk_mutex);
9493 check_system_chunk(trans, alloc_flags);
9494 mutex_unlock(&fs_info->chunk_mutex);
9496 mutex_unlock(&fs_info->ro_block_group_mutex);
9498 btrfs_end_transaction(trans);
9502 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9504 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9506 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9510 * helper to account the unused space of all the readonly block group in the
9511 * space_info. takes mirrors into account.
9513 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9515 struct btrfs_block_group_cache *block_group;
9519 /* It's df, we don't care if it's racy */
9520 if (list_empty(&sinfo->ro_bgs))
9523 spin_lock(&sinfo->lock);
9524 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9525 spin_lock(&block_group->lock);
9527 if (!block_group->ro) {
9528 spin_unlock(&block_group->lock);
9532 factor = btrfs_bg_type_to_factor(block_group->flags);
9533 free_bytes += (block_group->key.offset -
9534 btrfs_block_group_used(&block_group->item)) *
9537 spin_unlock(&block_group->lock);
9539 spin_unlock(&sinfo->lock);
9544 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9546 struct btrfs_space_info *sinfo = cache->space_info;
9551 spin_lock(&sinfo->lock);
9552 spin_lock(&cache->lock);
9554 num_bytes = cache->key.offset - cache->reserved -
9555 cache->pinned - cache->bytes_super -
9556 btrfs_block_group_used(&cache->item);
9557 sinfo->bytes_readonly -= num_bytes;
9558 list_del_init(&cache->ro_list);
9560 spin_unlock(&cache->lock);
9561 spin_unlock(&sinfo->lock);
9565 * checks to see if its even possible to relocate this block group.
9567 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9568 * ok to go ahead and try.
9570 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9572 struct btrfs_root *root = fs_info->extent_root;
9573 struct btrfs_block_group_cache *block_group;
9574 struct btrfs_space_info *space_info;
9575 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9576 struct btrfs_device *device;
9577 struct btrfs_trans_handle *trans;
9587 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9589 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9591 /* odd, couldn't find the block group, leave it alone */
9595 "can't find block group for bytenr %llu",
9600 min_free = btrfs_block_group_used(&block_group->item);
9602 /* no bytes used, we're good */
9606 space_info = block_group->space_info;
9607 spin_lock(&space_info->lock);
9609 full = space_info->full;
9612 * if this is the last block group we have in this space, we can't
9613 * relocate it unless we're able to allocate a new chunk below.
9615 * Otherwise, we need to make sure we have room in the space to handle
9616 * all of the extents from this block group. If we can, we're good
9618 if ((space_info->total_bytes != block_group->key.offset) &&
9619 (btrfs_space_info_used(space_info, false) + min_free <
9620 space_info->total_bytes)) {
9621 spin_unlock(&space_info->lock);
9624 spin_unlock(&space_info->lock);
9627 * ok we don't have enough space, but maybe we have free space on our
9628 * devices to allocate new chunks for relocation, so loop through our
9629 * alloc devices and guess if we have enough space. if this block
9630 * group is going to be restriped, run checks against the target
9631 * profile instead of the current one.
9643 target = get_restripe_target(fs_info, block_group->flags);
9645 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9648 * this is just a balance, so if we were marked as full
9649 * we know there is no space for a new chunk
9654 "no space to alloc new chunk for block group %llu",
9655 block_group->key.objectid);
9659 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9662 if (index == BTRFS_RAID_RAID10) {
9666 } else if (index == BTRFS_RAID_RAID1) {
9668 } else if (index == BTRFS_RAID_DUP) {
9671 } else if (index == BTRFS_RAID_RAID0) {
9672 dev_min = fs_devices->rw_devices;
9673 min_free = div64_u64(min_free, dev_min);
9676 /* We need to do this so that we can look at pending chunks */
9677 trans = btrfs_join_transaction(root);
9678 if (IS_ERR(trans)) {
9679 ret = PTR_ERR(trans);
9683 mutex_lock(&fs_info->chunk_mutex);
9684 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9688 * check to make sure we can actually find a chunk with enough
9689 * space to fit our block group in.
9691 if (device->total_bytes > device->bytes_used + min_free &&
9692 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9693 ret = find_free_dev_extent(trans, device, min_free,
9698 if (dev_nr >= dev_min)
9704 if (debug && ret == -1)
9706 "no space to allocate a new chunk for block group %llu",
9707 block_group->key.objectid);
9708 mutex_unlock(&fs_info->chunk_mutex);
9709 btrfs_end_transaction(trans);
9711 btrfs_put_block_group(block_group);
9715 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9716 struct btrfs_path *path,
9717 struct btrfs_key *key)
9719 struct btrfs_root *root = fs_info->extent_root;
9721 struct btrfs_key found_key;
9722 struct extent_buffer *leaf;
9723 struct btrfs_block_group_item bg;
9727 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9732 slot = path->slots[0];
9733 leaf = path->nodes[0];
9734 if (slot >= btrfs_header_nritems(leaf)) {
9735 ret = btrfs_next_leaf(root, path);
9742 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9744 if (found_key.objectid >= key->objectid &&
9745 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9746 struct extent_map_tree *em_tree;
9747 struct extent_map *em;
9749 em_tree = &root->fs_info->mapping_tree.map_tree;
9750 read_lock(&em_tree->lock);
9751 em = lookup_extent_mapping(em_tree, found_key.objectid,
9753 read_unlock(&em_tree->lock);
9756 "logical %llu len %llu found bg but no related chunk",
9757 found_key.objectid, found_key.offset);
9759 } else if (em->start != found_key.objectid ||
9760 em->len != found_key.offset) {
9762 "block group %llu len %llu mismatch with chunk %llu len %llu",
9763 found_key.objectid, found_key.offset,
9764 em->start, em->len);
9767 read_extent_buffer(leaf, &bg,
9768 btrfs_item_ptr_offset(leaf, slot),
9770 flags = btrfs_block_group_flags(&bg) &
9771 BTRFS_BLOCK_GROUP_TYPE_MASK;
9773 if (flags != (em->map_lookup->type &
9774 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9776 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9778 found_key.offset, flags,
9779 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9780 em->map_lookup->type));
9786 free_extent_map(em);
9795 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9797 struct btrfs_block_group_cache *block_group;
9801 struct inode *inode;
9803 block_group = btrfs_lookup_first_block_group(info, last);
9804 while (block_group) {
9805 wait_block_group_cache_done(block_group);
9806 spin_lock(&block_group->lock);
9807 if (block_group->iref)
9809 spin_unlock(&block_group->lock);
9810 block_group = next_block_group(info, block_group);
9819 inode = block_group->inode;
9820 block_group->iref = 0;
9821 block_group->inode = NULL;
9822 spin_unlock(&block_group->lock);
9823 ASSERT(block_group->io_ctl.inode == NULL);
9825 last = block_group->key.objectid + block_group->key.offset;
9826 btrfs_put_block_group(block_group);
9831 * Must be called only after stopping all workers, since we could have block
9832 * group caching kthreads running, and therefore they could race with us if we
9833 * freed the block groups before stopping them.
9835 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9837 struct btrfs_block_group_cache *block_group;
9838 struct btrfs_space_info *space_info;
9839 struct btrfs_caching_control *caching_ctl;
9842 down_write(&info->commit_root_sem);
9843 while (!list_empty(&info->caching_block_groups)) {
9844 caching_ctl = list_entry(info->caching_block_groups.next,
9845 struct btrfs_caching_control, list);
9846 list_del(&caching_ctl->list);
9847 put_caching_control(caching_ctl);
9849 up_write(&info->commit_root_sem);
9851 spin_lock(&info->unused_bgs_lock);
9852 while (!list_empty(&info->unused_bgs)) {
9853 block_group = list_first_entry(&info->unused_bgs,
9854 struct btrfs_block_group_cache,
9856 list_del_init(&block_group->bg_list);
9857 btrfs_put_block_group(block_group);
9859 spin_unlock(&info->unused_bgs_lock);
9861 spin_lock(&info->block_group_cache_lock);
9862 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9863 block_group = rb_entry(n, struct btrfs_block_group_cache,
9865 rb_erase(&block_group->cache_node,
9866 &info->block_group_cache_tree);
9867 RB_CLEAR_NODE(&block_group->cache_node);
9868 spin_unlock(&info->block_group_cache_lock);
9870 down_write(&block_group->space_info->groups_sem);
9871 list_del(&block_group->list);
9872 up_write(&block_group->space_info->groups_sem);
9875 * We haven't cached this block group, which means we could
9876 * possibly have excluded extents on this block group.
9878 if (block_group->cached == BTRFS_CACHE_NO ||
9879 block_group->cached == BTRFS_CACHE_ERROR)
9880 free_excluded_extents(block_group);
9882 btrfs_remove_free_space_cache(block_group);
9883 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9884 ASSERT(list_empty(&block_group->dirty_list));
9885 ASSERT(list_empty(&block_group->io_list));
9886 ASSERT(list_empty(&block_group->bg_list));
9887 ASSERT(atomic_read(&block_group->count) == 1);
9888 btrfs_put_block_group(block_group);
9890 spin_lock(&info->block_group_cache_lock);
9892 spin_unlock(&info->block_group_cache_lock);
9894 /* now that all the block groups are freed, go through and
9895 * free all the space_info structs. This is only called during
9896 * the final stages of unmount, and so we know nobody is
9897 * using them. We call synchronize_rcu() once before we start,
9898 * just to be on the safe side.
9902 release_global_block_rsv(info);
9904 while (!list_empty(&info->space_info)) {
9907 space_info = list_entry(info->space_info.next,
9908 struct btrfs_space_info,
9912 * Do not hide this behind enospc_debug, this is actually
9913 * important and indicates a real bug if this happens.
9915 if (WARN_ON(space_info->bytes_pinned > 0 ||
9916 space_info->bytes_reserved > 0 ||
9917 space_info->bytes_may_use > 0))
9918 dump_space_info(info, space_info, 0, 0);
9919 list_del(&space_info->list);
9920 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9921 struct kobject *kobj;
9922 kobj = space_info->block_group_kobjs[i];
9923 space_info->block_group_kobjs[i] = NULL;
9929 kobject_del(&space_info->kobj);
9930 kobject_put(&space_info->kobj);
9935 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9936 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9938 struct btrfs_space_info *space_info;
9939 struct raid_kobject *rkobj;
9944 spin_lock(&fs_info->pending_raid_kobjs_lock);
9945 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9946 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9948 list_for_each_entry(rkobj, &list, list) {
9949 space_info = __find_space_info(fs_info, rkobj->flags);
9950 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9952 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9953 "%s", get_raid_name(index));
9955 kobject_put(&rkobj->kobj);
9961 "failed to add kobject for block cache, ignoring");
9964 static void link_block_group(struct btrfs_block_group_cache *cache)
9966 struct btrfs_space_info *space_info = cache->space_info;
9967 struct btrfs_fs_info *fs_info = cache->fs_info;
9968 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9971 down_write(&space_info->groups_sem);
9972 if (list_empty(&space_info->block_groups[index]))
9974 list_add_tail(&cache->list, &space_info->block_groups[index]);
9975 up_write(&space_info->groups_sem);
9978 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9980 btrfs_warn(cache->fs_info,
9981 "couldn't alloc memory for raid level kobject");
9984 rkobj->flags = cache->flags;
9985 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9987 spin_lock(&fs_info->pending_raid_kobjs_lock);
9988 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9989 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9990 space_info->block_group_kobjs[index] = &rkobj->kobj;
9994 static struct btrfs_block_group_cache *
9995 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9996 u64 start, u64 size)
9998 struct btrfs_block_group_cache *cache;
10000 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10004 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10006 if (!cache->free_space_ctl) {
10011 cache->key.objectid = start;
10012 cache->key.offset = size;
10013 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10015 cache->fs_info = fs_info;
10016 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10017 set_free_space_tree_thresholds(cache);
10019 atomic_set(&cache->count, 1);
10020 spin_lock_init(&cache->lock);
10021 init_rwsem(&cache->data_rwsem);
10022 INIT_LIST_HEAD(&cache->list);
10023 INIT_LIST_HEAD(&cache->cluster_list);
10024 INIT_LIST_HEAD(&cache->bg_list);
10025 INIT_LIST_HEAD(&cache->ro_list);
10026 INIT_LIST_HEAD(&cache->dirty_list);
10027 INIT_LIST_HEAD(&cache->io_list);
10028 btrfs_init_free_space_ctl(cache);
10029 atomic_set(&cache->trimming, 0);
10030 mutex_init(&cache->free_space_lock);
10031 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10038 * Iterate all chunks and verify that each of them has the corresponding block
10041 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10043 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10044 struct extent_map *em;
10045 struct btrfs_block_group_cache *bg;
10050 read_lock(&map_tree->map_tree.lock);
10052 * lookup_extent_mapping will return the first extent map
10053 * intersecting the range, so setting @len to 1 is enough to
10054 * get the first chunk.
10056 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10057 read_unlock(&map_tree->map_tree.lock);
10061 bg = btrfs_lookup_block_group(fs_info, em->start);
10064 "chunk start=%llu len=%llu doesn't have corresponding block group",
10065 em->start, em->len);
10067 free_extent_map(em);
10070 if (bg->key.objectid != em->start ||
10071 bg->key.offset != em->len ||
10072 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10073 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10075 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10076 em->start, em->len,
10077 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10078 bg->key.objectid, bg->key.offset,
10079 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10081 free_extent_map(em);
10082 btrfs_put_block_group(bg);
10085 start = em->start + em->len;
10086 free_extent_map(em);
10087 btrfs_put_block_group(bg);
10092 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10094 struct btrfs_path *path;
10096 struct btrfs_block_group_cache *cache;
10097 struct btrfs_space_info *space_info;
10098 struct btrfs_key key;
10099 struct btrfs_key found_key;
10100 struct extent_buffer *leaf;
10101 int need_clear = 0;
10106 feature = btrfs_super_incompat_flags(info->super_copy);
10107 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10111 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10112 path = btrfs_alloc_path();
10115 path->reada = READA_FORWARD;
10117 cache_gen = btrfs_super_cache_generation(info->super_copy);
10118 if (btrfs_test_opt(info, SPACE_CACHE) &&
10119 btrfs_super_generation(info->super_copy) != cache_gen)
10121 if (btrfs_test_opt(info, CLEAR_CACHE))
10125 ret = find_first_block_group(info, path, &key);
10131 leaf = path->nodes[0];
10132 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10134 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10143 * When we mount with old space cache, we need to
10144 * set BTRFS_DC_CLEAR and set dirty flag.
10146 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10147 * truncate the old free space cache inode and
10149 * b) Setting 'dirty flag' makes sure that we flush
10150 * the new space cache info onto disk.
10152 if (btrfs_test_opt(info, SPACE_CACHE))
10153 cache->disk_cache_state = BTRFS_DC_CLEAR;
10156 read_extent_buffer(leaf, &cache->item,
10157 btrfs_item_ptr_offset(leaf, path->slots[0]),
10158 sizeof(cache->item));
10159 cache->flags = btrfs_block_group_flags(&cache->item);
10161 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10162 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10164 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10165 cache->key.objectid);
10170 key.objectid = found_key.objectid + found_key.offset;
10171 btrfs_release_path(path);
10174 * We need to exclude the super stripes now so that the space
10175 * info has super bytes accounted for, otherwise we'll think
10176 * we have more space than we actually do.
10178 ret = exclude_super_stripes(cache);
10181 * We may have excluded something, so call this just in
10184 free_excluded_extents(cache);
10185 btrfs_put_block_group(cache);
10190 * check for two cases, either we are full, and therefore
10191 * don't need to bother with the caching work since we won't
10192 * find any space, or we are empty, and we can just add all
10193 * the space in and be done with it. This saves us _alot_ of
10194 * time, particularly in the full case.
10196 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10197 cache->last_byte_to_unpin = (u64)-1;
10198 cache->cached = BTRFS_CACHE_FINISHED;
10199 free_excluded_extents(cache);
10200 } else if (btrfs_block_group_used(&cache->item) == 0) {
10201 cache->last_byte_to_unpin = (u64)-1;
10202 cache->cached = BTRFS_CACHE_FINISHED;
10203 add_new_free_space(cache, found_key.objectid,
10204 found_key.objectid +
10206 free_excluded_extents(cache);
10209 ret = btrfs_add_block_group_cache(info, cache);
10211 btrfs_remove_free_space_cache(cache);
10212 btrfs_put_block_group(cache);
10216 trace_btrfs_add_block_group(info, cache, 0);
10217 update_space_info(info, cache->flags, found_key.offset,
10218 btrfs_block_group_used(&cache->item),
10219 cache->bytes_super, &space_info);
10221 cache->space_info = space_info;
10223 link_block_group(cache);
10225 set_avail_alloc_bits(info, cache->flags);
10226 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10227 inc_block_group_ro(cache, 1);
10228 } else if (btrfs_block_group_used(&cache->item) == 0) {
10229 ASSERT(list_empty(&cache->bg_list));
10230 btrfs_mark_bg_unused(cache);
10234 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10235 if (!(get_alloc_profile(info, space_info->flags) &
10236 (BTRFS_BLOCK_GROUP_RAID10 |
10237 BTRFS_BLOCK_GROUP_RAID1 |
10238 BTRFS_BLOCK_GROUP_RAID5 |
10239 BTRFS_BLOCK_GROUP_RAID6 |
10240 BTRFS_BLOCK_GROUP_DUP)))
10243 * avoid allocating from un-mirrored block group if there are
10244 * mirrored block groups.
10246 list_for_each_entry(cache,
10247 &space_info->block_groups[BTRFS_RAID_RAID0],
10249 inc_block_group_ro(cache, 1);
10250 list_for_each_entry(cache,
10251 &space_info->block_groups[BTRFS_RAID_SINGLE],
10253 inc_block_group_ro(cache, 1);
10256 btrfs_add_raid_kobjects(info);
10257 init_global_block_rsv(info);
10258 ret = check_chunk_block_group_mappings(info);
10260 btrfs_free_path(path);
10264 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10266 struct btrfs_fs_info *fs_info = trans->fs_info;
10267 struct btrfs_block_group_cache *block_group;
10268 struct btrfs_root *extent_root = fs_info->extent_root;
10269 struct btrfs_block_group_item item;
10270 struct btrfs_key key;
10273 if (!trans->can_flush_pending_bgs)
10276 while (!list_empty(&trans->new_bgs)) {
10277 block_group = list_first_entry(&trans->new_bgs,
10278 struct btrfs_block_group_cache,
10283 spin_lock(&block_group->lock);
10284 memcpy(&item, &block_group->item, sizeof(item));
10285 memcpy(&key, &block_group->key, sizeof(key));
10286 spin_unlock(&block_group->lock);
10288 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10291 btrfs_abort_transaction(trans, ret);
10292 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10294 btrfs_abort_transaction(trans, ret);
10295 add_block_group_free_space(trans, block_group);
10296 /* already aborted the transaction if it failed. */
10298 list_del_init(&block_group->bg_list);
10300 btrfs_trans_release_chunk_metadata(trans);
10303 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10304 u64 type, u64 chunk_offset, u64 size)
10306 struct btrfs_fs_info *fs_info = trans->fs_info;
10307 struct btrfs_block_group_cache *cache;
10310 btrfs_set_log_full_commit(fs_info, trans);
10312 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10316 btrfs_set_block_group_used(&cache->item, bytes_used);
10317 btrfs_set_block_group_chunk_objectid(&cache->item,
10318 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10319 btrfs_set_block_group_flags(&cache->item, type);
10321 cache->flags = type;
10322 cache->last_byte_to_unpin = (u64)-1;
10323 cache->cached = BTRFS_CACHE_FINISHED;
10324 cache->needs_free_space = 1;
10325 ret = exclude_super_stripes(cache);
10328 * We may have excluded something, so call this just in
10331 free_excluded_extents(cache);
10332 btrfs_put_block_group(cache);
10336 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10338 free_excluded_extents(cache);
10340 #ifdef CONFIG_BTRFS_DEBUG
10341 if (btrfs_should_fragment_free_space(cache)) {
10342 u64 new_bytes_used = size - bytes_used;
10344 bytes_used += new_bytes_used >> 1;
10345 fragment_free_space(cache);
10349 * Ensure the corresponding space_info object is created and
10350 * assigned to our block group. We want our bg to be added to the rbtree
10351 * with its ->space_info set.
10353 cache->space_info = __find_space_info(fs_info, cache->flags);
10354 ASSERT(cache->space_info);
10356 ret = btrfs_add_block_group_cache(fs_info, cache);
10358 btrfs_remove_free_space_cache(cache);
10359 btrfs_put_block_group(cache);
10364 * Now that our block group has its ->space_info set and is inserted in
10365 * the rbtree, update the space info's counters.
10367 trace_btrfs_add_block_group(fs_info, cache, 1);
10368 update_space_info(fs_info, cache->flags, size, bytes_used,
10369 cache->bytes_super, &cache->space_info);
10370 update_global_block_rsv(fs_info);
10372 link_block_group(cache);
10374 list_add_tail(&cache->bg_list, &trans->new_bgs);
10376 set_avail_alloc_bits(fs_info, type);
10380 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10382 u64 extra_flags = chunk_to_extended(flags) &
10383 BTRFS_EXTENDED_PROFILE_MASK;
10385 write_seqlock(&fs_info->profiles_lock);
10386 if (flags & BTRFS_BLOCK_GROUP_DATA)
10387 fs_info->avail_data_alloc_bits &= ~extra_flags;
10388 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10389 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10390 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10391 fs_info->avail_system_alloc_bits &= ~extra_flags;
10392 write_sequnlock(&fs_info->profiles_lock);
10395 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10396 u64 group_start, struct extent_map *em)
10398 struct btrfs_fs_info *fs_info = trans->fs_info;
10399 struct btrfs_root *root = fs_info->extent_root;
10400 struct btrfs_path *path;
10401 struct btrfs_block_group_cache *block_group;
10402 struct btrfs_free_cluster *cluster;
10403 struct btrfs_root *tree_root = fs_info->tree_root;
10404 struct btrfs_key key;
10405 struct inode *inode;
10406 struct kobject *kobj = NULL;
10410 struct btrfs_caching_control *caching_ctl = NULL;
10413 block_group = btrfs_lookup_block_group(fs_info, group_start);
10414 BUG_ON(!block_group);
10415 BUG_ON(!block_group->ro);
10417 trace_btrfs_remove_block_group(block_group);
10419 * Free the reserved super bytes from this block group before
10422 free_excluded_extents(block_group);
10423 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10424 block_group->key.offset);
10426 memcpy(&key, &block_group->key, sizeof(key));
10427 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10428 factor = btrfs_bg_type_to_factor(block_group->flags);
10430 /* make sure this block group isn't part of an allocation cluster */
10431 cluster = &fs_info->data_alloc_cluster;
10432 spin_lock(&cluster->refill_lock);
10433 btrfs_return_cluster_to_free_space(block_group, cluster);
10434 spin_unlock(&cluster->refill_lock);
10437 * make sure this block group isn't part of a metadata
10438 * allocation cluster
10440 cluster = &fs_info->meta_alloc_cluster;
10441 spin_lock(&cluster->refill_lock);
10442 btrfs_return_cluster_to_free_space(block_group, cluster);
10443 spin_unlock(&cluster->refill_lock);
10445 path = btrfs_alloc_path();
10452 * get the inode first so any iput calls done for the io_list
10453 * aren't the final iput (no unlinks allowed now)
10455 inode = lookup_free_space_inode(fs_info, block_group, path);
10457 mutex_lock(&trans->transaction->cache_write_mutex);
10459 * make sure our free spache cache IO is done before remove the
10462 spin_lock(&trans->transaction->dirty_bgs_lock);
10463 if (!list_empty(&block_group->io_list)) {
10464 list_del_init(&block_group->io_list);
10466 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10468 spin_unlock(&trans->transaction->dirty_bgs_lock);
10469 btrfs_wait_cache_io(trans, block_group, path);
10470 btrfs_put_block_group(block_group);
10471 spin_lock(&trans->transaction->dirty_bgs_lock);
10474 if (!list_empty(&block_group->dirty_list)) {
10475 list_del_init(&block_group->dirty_list);
10476 btrfs_put_block_group(block_group);
10478 spin_unlock(&trans->transaction->dirty_bgs_lock);
10479 mutex_unlock(&trans->transaction->cache_write_mutex);
10481 if (!IS_ERR(inode)) {
10482 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10484 btrfs_add_delayed_iput(inode);
10487 clear_nlink(inode);
10488 /* One for the block groups ref */
10489 spin_lock(&block_group->lock);
10490 if (block_group->iref) {
10491 block_group->iref = 0;
10492 block_group->inode = NULL;
10493 spin_unlock(&block_group->lock);
10496 spin_unlock(&block_group->lock);
10498 /* One for our lookup ref */
10499 btrfs_add_delayed_iput(inode);
10502 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10503 key.offset = block_group->key.objectid;
10506 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10510 btrfs_release_path(path);
10512 ret = btrfs_del_item(trans, tree_root, path);
10515 btrfs_release_path(path);
10518 spin_lock(&fs_info->block_group_cache_lock);
10519 rb_erase(&block_group->cache_node,
10520 &fs_info->block_group_cache_tree);
10521 RB_CLEAR_NODE(&block_group->cache_node);
10523 if (fs_info->first_logical_byte == block_group->key.objectid)
10524 fs_info->first_logical_byte = (u64)-1;
10525 spin_unlock(&fs_info->block_group_cache_lock);
10527 down_write(&block_group->space_info->groups_sem);
10529 * we must use list_del_init so people can check to see if they
10530 * are still on the list after taking the semaphore
10532 list_del_init(&block_group->list);
10533 if (list_empty(&block_group->space_info->block_groups[index])) {
10534 kobj = block_group->space_info->block_group_kobjs[index];
10535 block_group->space_info->block_group_kobjs[index] = NULL;
10536 clear_avail_alloc_bits(fs_info, block_group->flags);
10538 up_write(&block_group->space_info->groups_sem);
10544 if (block_group->has_caching_ctl)
10545 caching_ctl = get_caching_control(block_group);
10546 if (block_group->cached == BTRFS_CACHE_STARTED)
10547 wait_block_group_cache_done(block_group);
10548 if (block_group->has_caching_ctl) {
10549 down_write(&fs_info->commit_root_sem);
10550 if (!caching_ctl) {
10551 struct btrfs_caching_control *ctl;
10553 list_for_each_entry(ctl,
10554 &fs_info->caching_block_groups, list)
10555 if (ctl->block_group == block_group) {
10557 refcount_inc(&caching_ctl->count);
10562 list_del_init(&caching_ctl->list);
10563 up_write(&fs_info->commit_root_sem);
10565 /* Once for the caching bgs list and once for us. */
10566 put_caching_control(caching_ctl);
10567 put_caching_control(caching_ctl);
10571 spin_lock(&trans->transaction->dirty_bgs_lock);
10572 if (!list_empty(&block_group->dirty_list)) {
10575 if (!list_empty(&block_group->io_list)) {
10578 spin_unlock(&trans->transaction->dirty_bgs_lock);
10579 btrfs_remove_free_space_cache(block_group);
10581 spin_lock(&block_group->space_info->lock);
10582 list_del_init(&block_group->ro_list);
10584 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10585 WARN_ON(block_group->space_info->total_bytes
10586 < block_group->key.offset);
10587 WARN_ON(block_group->space_info->bytes_readonly
10588 < block_group->key.offset);
10589 WARN_ON(block_group->space_info->disk_total
10590 < block_group->key.offset * factor);
10592 block_group->space_info->total_bytes -= block_group->key.offset;
10593 block_group->space_info->bytes_readonly -= block_group->key.offset;
10594 block_group->space_info->disk_total -= block_group->key.offset * factor;
10596 spin_unlock(&block_group->space_info->lock);
10598 memcpy(&key, &block_group->key, sizeof(key));
10600 mutex_lock(&fs_info->chunk_mutex);
10601 if (!list_empty(&em->list)) {
10602 /* We're in the transaction->pending_chunks list. */
10603 free_extent_map(em);
10605 spin_lock(&block_group->lock);
10606 block_group->removed = 1;
10608 * At this point trimming can't start on this block group, because we
10609 * removed the block group from the tree fs_info->block_group_cache_tree
10610 * so no one can't find it anymore and even if someone already got this
10611 * block group before we removed it from the rbtree, they have already
10612 * incremented block_group->trimming - if they didn't, they won't find
10613 * any free space entries because we already removed them all when we
10614 * called btrfs_remove_free_space_cache().
10616 * And we must not remove the extent map from the fs_info->mapping_tree
10617 * to prevent the same logical address range and physical device space
10618 * ranges from being reused for a new block group. This is because our
10619 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10620 * completely transactionless, so while it is trimming a range the
10621 * currently running transaction might finish and a new one start,
10622 * allowing for new block groups to be created that can reuse the same
10623 * physical device locations unless we take this special care.
10625 * There may also be an implicit trim operation if the file system
10626 * is mounted with -odiscard. The same protections must remain
10627 * in place until the extents have been discarded completely when
10628 * the transaction commit has completed.
10630 remove_em = (atomic_read(&block_group->trimming) == 0);
10632 * Make sure a trimmer task always sees the em in the pinned_chunks list
10633 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10634 * before checking block_group->removed).
10638 * Our em might be in trans->transaction->pending_chunks which
10639 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10640 * and so is the fs_info->pinned_chunks list.
10642 * So at this point we must be holding the chunk_mutex to avoid
10643 * any races with chunk allocation (more specifically at
10644 * volumes.c:contains_pending_extent()), to ensure it always
10645 * sees the em, either in the pending_chunks list or in the
10646 * pinned_chunks list.
10648 list_move_tail(&em->list, &fs_info->pinned_chunks);
10650 spin_unlock(&block_group->lock);
10653 struct extent_map_tree *em_tree;
10655 em_tree = &fs_info->mapping_tree.map_tree;
10656 write_lock(&em_tree->lock);
10658 * The em might be in the pending_chunks list, so make sure the
10659 * chunk mutex is locked, since remove_extent_mapping() will
10660 * delete us from that list.
10662 remove_extent_mapping(em_tree, em);
10663 write_unlock(&em_tree->lock);
10664 /* once for the tree */
10665 free_extent_map(em);
10668 mutex_unlock(&fs_info->chunk_mutex);
10670 ret = remove_block_group_free_space(trans, block_group);
10674 btrfs_put_block_group(block_group);
10675 btrfs_put_block_group(block_group);
10677 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10683 ret = btrfs_del_item(trans, root, path);
10685 btrfs_free_path(path);
10689 struct btrfs_trans_handle *
10690 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10691 const u64 chunk_offset)
10693 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10694 struct extent_map *em;
10695 struct map_lookup *map;
10696 unsigned int num_items;
10698 read_lock(&em_tree->lock);
10699 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10700 read_unlock(&em_tree->lock);
10701 ASSERT(em && em->start == chunk_offset);
10704 * We need to reserve 3 + N units from the metadata space info in order
10705 * to remove a block group (done at btrfs_remove_chunk() and at
10706 * btrfs_remove_block_group()), which are used for:
10708 * 1 unit for adding the free space inode's orphan (located in the tree
10710 * 1 unit for deleting the block group item (located in the extent
10712 * 1 unit for deleting the free space item (located in tree of tree
10714 * N units for deleting N device extent items corresponding to each
10715 * stripe (located in the device tree).
10717 * In order to remove a block group we also need to reserve units in the
10718 * system space info in order to update the chunk tree (update one or
10719 * more device items and remove one chunk item), but this is done at
10720 * btrfs_remove_chunk() through a call to check_system_chunk().
10722 map = em->map_lookup;
10723 num_items = 3 + map->num_stripes;
10724 free_extent_map(em);
10726 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10731 * Process the unused_bgs list and remove any that don't have any allocated
10732 * space inside of them.
10734 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10736 struct btrfs_block_group_cache *block_group;
10737 struct btrfs_space_info *space_info;
10738 struct btrfs_trans_handle *trans;
10741 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10744 spin_lock(&fs_info->unused_bgs_lock);
10745 while (!list_empty(&fs_info->unused_bgs)) {
10749 block_group = list_first_entry(&fs_info->unused_bgs,
10750 struct btrfs_block_group_cache,
10752 list_del_init(&block_group->bg_list);
10754 space_info = block_group->space_info;
10756 if (ret || btrfs_mixed_space_info(space_info)) {
10757 btrfs_put_block_group(block_group);
10760 spin_unlock(&fs_info->unused_bgs_lock);
10762 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10764 /* Don't want to race with allocators so take the groups_sem */
10765 down_write(&space_info->groups_sem);
10766 spin_lock(&block_group->lock);
10767 if (block_group->reserved || block_group->pinned ||
10768 btrfs_block_group_used(&block_group->item) ||
10770 list_is_singular(&block_group->list)) {
10772 * We want to bail if we made new allocations or have
10773 * outstanding allocations in this block group. We do
10774 * the ro check in case balance is currently acting on
10775 * this block group.
10777 trace_btrfs_skip_unused_block_group(block_group);
10778 spin_unlock(&block_group->lock);
10779 up_write(&space_info->groups_sem);
10782 spin_unlock(&block_group->lock);
10784 /* We don't want to force the issue, only flip if it's ok. */
10785 ret = inc_block_group_ro(block_group, 0);
10786 up_write(&space_info->groups_sem);
10793 * Want to do this before we do anything else so we can recover
10794 * properly if we fail to join the transaction.
10796 trans = btrfs_start_trans_remove_block_group(fs_info,
10797 block_group->key.objectid);
10798 if (IS_ERR(trans)) {
10799 btrfs_dec_block_group_ro(block_group);
10800 ret = PTR_ERR(trans);
10805 * We could have pending pinned extents for this block group,
10806 * just delete them, we don't care about them anymore.
10808 start = block_group->key.objectid;
10809 end = start + block_group->key.offset - 1;
10811 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10812 * btrfs_finish_extent_commit(). If we are at transaction N,
10813 * another task might be running finish_extent_commit() for the
10814 * previous transaction N - 1, and have seen a range belonging
10815 * to the block group in freed_extents[] before we were able to
10816 * clear the whole block group range from freed_extents[]. This
10817 * means that task can lookup for the block group after we
10818 * unpinned it from freed_extents[] and removed it, leading to
10819 * a BUG_ON() at btrfs_unpin_extent_range().
10821 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10822 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10825 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10826 btrfs_dec_block_group_ro(block_group);
10829 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10832 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10833 btrfs_dec_block_group_ro(block_group);
10836 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10838 /* Reset pinned so btrfs_put_block_group doesn't complain */
10839 spin_lock(&space_info->lock);
10840 spin_lock(&block_group->lock);
10842 update_bytes_pinned(space_info, -block_group->pinned);
10843 space_info->bytes_readonly += block_group->pinned;
10844 percpu_counter_add_batch(&space_info->total_bytes_pinned,
10845 -block_group->pinned,
10846 BTRFS_TOTAL_BYTES_PINNED_BATCH);
10847 block_group->pinned = 0;
10849 spin_unlock(&block_group->lock);
10850 spin_unlock(&space_info->lock);
10852 /* DISCARD can flip during remount */
10853 trimming = btrfs_test_opt(fs_info, DISCARD);
10855 /* Implicit trim during transaction commit. */
10857 btrfs_get_block_group_trimming(block_group);
10860 * Btrfs_remove_chunk will abort the transaction if things go
10863 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
10867 btrfs_put_block_group_trimming(block_group);
10872 * If we're not mounted with -odiscard, we can just forget
10873 * about this block group. Otherwise we'll need to wait
10874 * until transaction commit to do the actual discard.
10877 spin_lock(&fs_info->unused_bgs_lock);
10879 * A concurrent scrub might have added us to the list
10880 * fs_info->unused_bgs, so use a list_move operation
10881 * to add the block group to the deleted_bgs list.
10883 list_move(&block_group->bg_list,
10884 &trans->transaction->deleted_bgs);
10885 spin_unlock(&fs_info->unused_bgs_lock);
10886 btrfs_get_block_group(block_group);
10889 btrfs_end_transaction(trans);
10891 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10892 btrfs_put_block_group(block_group);
10893 spin_lock(&fs_info->unused_bgs_lock);
10895 spin_unlock(&fs_info->unused_bgs_lock);
10898 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10900 struct btrfs_super_block *disk_super;
10906 disk_super = fs_info->super_copy;
10907 if (!btrfs_super_root(disk_super))
10910 features = btrfs_super_incompat_flags(disk_super);
10911 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10914 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10915 ret = create_space_info(fs_info, flags);
10920 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10921 ret = create_space_info(fs_info, flags);
10923 flags = BTRFS_BLOCK_GROUP_METADATA;
10924 ret = create_space_info(fs_info, flags);
10928 flags = BTRFS_BLOCK_GROUP_DATA;
10929 ret = create_space_info(fs_info, flags);
10935 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10936 u64 start, u64 end)
10938 return unpin_extent_range(fs_info, start, end, false);
10942 * It used to be that old block groups would be left around forever.
10943 * Iterating over them would be enough to trim unused space. Since we
10944 * now automatically remove them, we also need to iterate over unallocated
10947 * We don't want a transaction for this since the discard may take a
10948 * substantial amount of time. We don't require that a transaction be
10949 * running, but we do need to take a running transaction into account
10950 * to ensure that we're not discarding chunks that were released or
10951 * allocated in the current transaction.
10953 * Holding the chunks lock will prevent other threads from allocating
10954 * or releasing chunks, but it won't prevent a running transaction
10955 * from committing and releasing the memory that the pending chunks
10956 * list head uses. For that, we need to take a reference to the
10957 * transaction and hold the commit root sem. We only need to hold
10958 * it while performing the free space search since we have already
10959 * held back allocations.
10961 static int btrfs_trim_free_extents(struct btrfs_device *device,
10962 u64 minlen, u64 *trimmed)
10964 u64 start = 0, len = 0;
10969 /* Discard not supported = nothing to do. */
10970 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
10973 /* Not writeable = nothing to do. */
10974 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10977 /* No free space = nothing to do. */
10978 if (device->total_bytes <= device->bytes_used)
10984 struct btrfs_fs_info *fs_info = device->fs_info;
10985 struct btrfs_transaction *trans;
10988 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10992 ret = down_read_killable(&fs_info->commit_root_sem);
10994 mutex_unlock(&fs_info->chunk_mutex);
10998 spin_lock(&fs_info->trans_lock);
10999 trans = fs_info->running_transaction;
11001 refcount_inc(&trans->use_count);
11002 spin_unlock(&fs_info->trans_lock);
11005 up_read(&fs_info->commit_root_sem);
11007 ret = find_free_dev_extent_start(trans, device, minlen, start,
11010 up_read(&fs_info->commit_root_sem);
11011 btrfs_put_transaction(trans);
11015 mutex_unlock(&fs_info->chunk_mutex);
11016 if (ret == -ENOSPC)
11021 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11022 mutex_unlock(&fs_info->chunk_mutex);
11030 if (fatal_signal_pending(current)) {
11031 ret = -ERESTARTSYS;
11042 * Trim the whole filesystem by:
11043 * 1) trimming the free space in each block group
11044 * 2) trimming the unallocated space on each device
11046 * This will also continue trimming even if a block group or device encounters
11047 * an error. The return value will be the last error, or 0 if nothing bad
11050 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11052 struct btrfs_block_group_cache *cache = NULL;
11053 struct btrfs_device *device;
11054 struct list_head *devices;
11060 u64 dev_failed = 0;
11065 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11066 for (; cache; cache = next_block_group(fs_info, cache)) {
11067 if (cache->key.objectid >= (range->start + range->len)) {
11068 btrfs_put_block_group(cache);
11072 start = max(range->start, cache->key.objectid);
11073 end = min(range->start + range->len,
11074 cache->key.objectid + cache->key.offset);
11076 if (end - start >= range->minlen) {
11077 if (!block_group_cache_done(cache)) {
11078 ret = cache_block_group(cache, 0);
11084 ret = wait_block_group_cache_done(cache);
11091 ret = btrfs_trim_block_group(cache,
11097 trimmed += group_trimmed;
11107 btrfs_warn(fs_info,
11108 "failed to trim %llu block group(s), last error %d",
11109 bg_failed, bg_ret);
11110 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11111 devices = &fs_info->fs_devices->devices;
11112 list_for_each_entry(device, devices, dev_list) {
11113 ret = btrfs_trim_free_extents(device, range->minlen,
11121 trimmed += group_trimmed;
11123 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11126 btrfs_warn(fs_info,
11127 "failed to trim %llu device(s), last error %d",
11128 dev_failed, dev_ret);
11129 range->len = trimmed;
11136 * btrfs_{start,end}_write_no_snapshotting() are similar to
11137 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11138 * data into the page cache through nocow before the subvolume is snapshoted,
11139 * but flush the data into disk after the snapshot creation, or to prevent
11140 * operations while snapshotting is ongoing and that cause the snapshot to be
11141 * inconsistent (writes followed by expanding truncates for example).
11143 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11145 percpu_counter_dec(&root->subv_writers->counter);
11146 cond_wake_up(&root->subv_writers->wait);
11149 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11151 if (atomic_read(&root->will_be_snapshotted))
11154 percpu_counter_inc(&root->subv_writers->counter);
11156 * Make sure counter is updated before we check for snapshot creation.
11159 if (atomic_read(&root->will_be_snapshotted)) {
11160 btrfs_end_write_no_snapshotting(root);
11166 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11171 ret = btrfs_start_write_no_snapshotting(root);
11174 wait_var_event(&root->will_be_snapshotted,
11175 !atomic_read(&root->will_be_snapshotted));
11179 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11181 struct btrfs_fs_info *fs_info = bg->fs_info;
11183 spin_lock(&fs_info->unused_bgs_lock);
11184 if (list_empty(&bg->bg_list)) {
11185 btrfs_get_block_group(bg);
11186 trace_btrfs_add_unused_block_group(bg);
11187 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11189 spin_unlock(&fs_info->unused_bgs_lock);