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 int cleanup_ref_head(struct btrfs_trans_handle *trans,
2447 struct btrfs_delayed_ref_head *head)
2450 struct btrfs_fs_info *fs_info = trans->fs_info;
2451 struct btrfs_delayed_ref_root *delayed_refs;
2454 delayed_refs = &trans->transaction->delayed_refs;
2456 ret = cleanup_extent_op(trans, head);
2458 unselect_delayed_ref_head(delayed_refs, head);
2459 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2466 * Need to drop our head ref lock and re-acquire the delayed ref lock
2467 * and then re-check to make sure nobody got added.
2469 spin_unlock(&head->lock);
2470 spin_lock(&delayed_refs->lock);
2471 spin_lock(&head->lock);
2472 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2473 spin_unlock(&head->lock);
2474 spin_unlock(&delayed_refs->lock);
2477 delayed_refs->num_heads--;
2478 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
2479 RB_CLEAR_NODE(&head->href_node);
2480 spin_unlock(&head->lock);
2481 spin_unlock(&delayed_refs->lock);
2482 atomic_dec(&delayed_refs->num_entries);
2484 trace_run_delayed_ref_head(fs_info, head, 0);
2486 if (head->total_ref_mod < 0) {
2487 struct btrfs_space_info *space_info;
2491 flags = BTRFS_BLOCK_GROUP_DATA;
2492 else if (head->is_system)
2493 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2495 flags = BTRFS_BLOCK_GROUP_METADATA;
2496 space_info = __find_space_info(fs_info, flags);
2498 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2500 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2502 if (head->is_data) {
2503 spin_lock(&delayed_refs->lock);
2504 delayed_refs->pending_csums -= head->num_bytes;
2505 spin_unlock(&delayed_refs->lock);
2509 if (head->must_insert_reserved) {
2510 btrfs_pin_extent(fs_info, head->bytenr,
2511 head->num_bytes, 1);
2512 if (head->is_data) {
2513 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2518 /* Also free its reserved qgroup space */
2519 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2520 head->qgroup_reserved);
2521 btrfs_delayed_ref_unlock(head);
2522 btrfs_put_delayed_ref_head(head);
2526 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2527 struct btrfs_trans_handle *trans)
2529 struct btrfs_delayed_ref_root *delayed_refs =
2530 &trans->transaction->delayed_refs;
2531 struct btrfs_delayed_ref_head *head = NULL;
2534 spin_lock(&delayed_refs->lock);
2535 head = btrfs_select_ref_head(delayed_refs);
2537 spin_unlock(&delayed_refs->lock);
2542 * Grab the lock that says we are going to process all the refs for
2545 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2546 spin_unlock(&delayed_refs->lock);
2549 * We may have dropped the spin lock to get the head mutex lock, and
2550 * that might have given someone else time to free the head. If that's
2551 * true, it has been removed from our list and we can move on.
2554 head = ERR_PTR(-EAGAIN);
2559 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2560 struct btrfs_delayed_ref_head *locked_ref,
2561 unsigned long *run_refs)
2563 struct btrfs_fs_info *fs_info = trans->fs_info;
2564 struct btrfs_delayed_ref_root *delayed_refs;
2565 struct btrfs_delayed_extent_op *extent_op;
2566 struct btrfs_delayed_ref_node *ref;
2567 int must_insert_reserved = 0;
2570 delayed_refs = &trans->transaction->delayed_refs;
2572 lockdep_assert_held(&locked_ref->mutex);
2573 lockdep_assert_held(&locked_ref->lock);
2575 while ((ref = select_delayed_ref(locked_ref))) {
2577 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2578 spin_unlock(&locked_ref->lock);
2579 unselect_delayed_ref_head(delayed_refs, locked_ref);
2585 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2586 RB_CLEAR_NODE(&ref->ref_node);
2587 if (!list_empty(&ref->add_list))
2588 list_del(&ref->add_list);
2590 * When we play the delayed ref, also correct the ref_mod on
2593 switch (ref->action) {
2594 case BTRFS_ADD_DELAYED_REF:
2595 case BTRFS_ADD_DELAYED_EXTENT:
2596 locked_ref->ref_mod -= ref->ref_mod;
2598 case BTRFS_DROP_DELAYED_REF:
2599 locked_ref->ref_mod += ref->ref_mod;
2604 atomic_dec(&delayed_refs->num_entries);
2607 * Record the must_insert_reserved flag before we drop the
2610 must_insert_reserved = locked_ref->must_insert_reserved;
2611 locked_ref->must_insert_reserved = 0;
2613 extent_op = locked_ref->extent_op;
2614 locked_ref->extent_op = NULL;
2615 spin_unlock(&locked_ref->lock);
2617 ret = run_one_delayed_ref(trans, ref, extent_op,
2618 must_insert_reserved);
2620 btrfs_free_delayed_extent_op(extent_op);
2622 unselect_delayed_ref_head(delayed_refs, locked_ref);
2623 btrfs_put_delayed_ref(ref);
2624 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2629 btrfs_put_delayed_ref(ref);
2632 spin_lock(&locked_ref->lock);
2633 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2640 * Returns 0 on success or if called with an already aborted transaction.
2641 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2643 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2646 struct btrfs_fs_info *fs_info = trans->fs_info;
2647 struct btrfs_delayed_ref_root *delayed_refs;
2648 struct btrfs_delayed_ref_head *locked_ref = NULL;
2649 ktime_t start = ktime_get();
2651 unsigned long count = 0;
2652 unsigned long actual_count = 0;
2654 delayed_refs = &trans->transaction->delayed_refs;
2657 locked_ref = btrfs_obtain_ref_head(trans);
2658 if (IS_ERR_OR_NULL(locked_ref)) {
2659 if (PTR_ERR(locked_ref) == -EAGAIN) {
2668 * We need to try and merge add/drops of the same ref since we
2669 * can run into issues with relocate dropping the implicit ref
2670 * and then it being added back again before the drop can
2671 * finish. If we merged anything we need to re-loop so we can
2673 * Or we can get node references of the same type that weren't
2674 * merged when created due to bumps in the tree mod seq, and
2675 * we need to merge them to prevent adding an inline extent
2676 * backref before dropping it (triggering a BUG_ON at
2677 * insert_inline_extent_backref()).
2679 spin_lock(&locked_ref->lock);
2680 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2682 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2684 if (ret < 0 && ret != -EAGAIN) {
2686 * Error, btrfs_run_delayed_refs_for_head already
2687 * unlocked everything so just bail out
2692 * Success, perform the usual cleanup of a processed
2695 ret = cleanup_ref_head(trans, locked_ref);
2697 /* We dropped our lock, we need to loop. */
2706 * Either success case or btrfs_run_delayed_refs_for_head
2707 * returned -EAGAIN, meaning we need to select another head
2712 } while ((nr != -1 && count < nr) || locked_ref);
2715 * We don't want to include ref heads since we can have empty ref heads
2716 * and those will drastically skew our runtime down since we just do
2717 * accounting, no actual extent tree updates.
2719 if (actual_count > 0) {
2720 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2724 * We weigh the current average higher than our current runtime
2725 * to avoid large swings in the average.
2727 spin_lock(&delayed_refs->lock);
2728 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2729 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2730 spin_unlock(&delayed_refs->lock);
2735 #ifdef SCRAMBLE_DELAYED_REFS
2737 * Normally delayed refs get processed in ascending bytenr order. This
2738 * correlates in most cases to the order added. To expose dependencies on this
2739 * order, we start to process the tree in the middle instead of the beginning
2741 static u64 find_middle(struct rb_root *root)
2743 struct rb_node *n = root->rb_node;
2744 struct btrfs_delayed_ref_node *entry;
2747 u64 first = 0, last = 0;
2751 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2752 first = entry->bytenr;
2756 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2757 last = entry->bytenr;
2762 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2763 WARN_ON(!entry->in_tree);
2765 middle = entry->bytenr;
2778 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2782 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2783 sizeof(struct btrfs_extent_inline_ref));
2784 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2785 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2788 * We don't ever fill up leaves all the way so multiply by 2 just to be
2789 * closer to what we're really going to want to use.
2791 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2795 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2796 * would require to store the csums for that many bytes.
2798 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2801 u64 num_csums_per_leaf;
2804 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2805 num_csums_per_leaf = div64_u64(csum_size,
2806 (u64)btrfs_super_csum_size(fs_info->super_copy));
2807 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2808 num_csums += num_csums_per_leaf - 1;
2809 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2813 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans)
2815 struct btrfs_fs_info *fs_info = trans->fs_info;
2816 struct btrfs_block_rsv *global_rsv;
2817 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2818 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2819 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2820 u64 num_bytes, num_dirty_bgs_bytes;
2823 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2824 num_heads = heads_to_leaves(fs_info, num_heads);
2826 num_bytes += (num_heads - 1) * fs_info->nodesize;
2828 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2830 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2832 global_rsv = &fs_info->global_block_rsv;
2835 * If we can't allocate any more chunks lets make sure we have _lots_ of
2836 * wiggle room since running delayed refs can create more delayed refs.
2838 if (global_rsv->space_info->full) {
2839 num_dirty_bgs_bytes <<= 1;
2843 spin_lock(&global_rsv->lock);
2844 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2846 spin_unlock(&global_rsv->lock);
2850 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2853 atomic_read(&trans->transaction->delayed_refs.num_entries);
2858 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2859 val = num_entries * avg_runtime;
2860 if (val >= NSEC_PER_SEC)
2862 if (val >= NSEC_PER_SEC / 2)
2865 return btrfs_check_space_for_delayed_refs(trans);
2868 struct async_delayed_refs {
2869 struct btrfs_root *root;
2874 struct completion wait;
2875 struct btrfs_work work;
2878 static inline struct async_delayed_refs *
2879 to_async_delayed_refs(struct btrfs_work *work)
2881 return container_of(work, struct async_delayed_refs, work);
2884 static void delayed_ref_async_start(struct btrfs_work *work)
2886 struct async_delayed_refs *async = to_async_delayed_refs(work);
2887 struct btrfs_trans_handle *trans;
2888 struct btrfs_fs_info *fs_info = async->root->fs_info;
2891 /* if the commit is already started, we don't need to wait here */
2892 if (btrfs_transaction_blocked(fs_info))
2895 trans = btrfs_join_transaction(async->root);
2896 if (IS_ERR(trans)) {
2897 async->error = PTR_ERR(trans);
2902 * trans->sync means that when we call end_transaction, we won't
2903 * wait on delayed refs
2907 /* Don't bother flushing if we got into a different transaction */
2908 if (trans->transid > async->transid)
2911 ret = btrfs_run_delayed_refs(trans, async->count);
2915 ret = btrfs_end_transaction(trans);
2916 if (ret && !async->error)
2920 complete(&async->wait);
2925 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2926 unsigned long count, u64 transid, int wait)
2928 struct async_delayed_refs *async;
2931 async = kmalloc(sizeof(*async), GFP_NOFS);
2935 async->root = fs_info->tree_root;
2936 async->count = count;
2938 async->transid = transid;
2943 init_completion(&async->wait);
2945 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2946 delayed_ref_async_start, NULL, NULL);
2948 btrfs_queue_work(fs_info->extent_workers, &async->work);
2951 wait_for_completion(&async->wait);
2960 * this starts processing the delayed reference count updates and
2961 * extent insertions we have queued up so far. count can be
2962 * 0, which means to process everything in the tree at the start
2963 * of the run (but not newly added entries), or it can be some target
2964 * number you'd like to process.
2966 * Returns 0 on success or if called with an aborted transaction
2967 * Returns <0 on error and aborts the transaction
2969 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2970 unsigned long count)
2972 struct btrfs_fs_info *fs_info = trans->fs_info;
2973 struct rb_node *node;
2974 struct btrfs_delayed_ref_root *delayed_refs;
2975 struct btrfs_delayed_ref_head *head;
2977 int run_all = count == (unsigned long)-1;
2979 /* We'll clean this up in btrfs_cleanup_transaction */
2983 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2986 delayed_refs = &trans->transaction->delayed_refs;
2988 count = atomic_read(&delayed_refs->num_entries) * 2;
2991 #ifdef SCRAMBLE_DELAYED_REFS
2992 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2994 ret = __btrfs_run_delayed_refs(trans, count);
2996 btrfs_abort_transaction(trans, ret);
3001 if (!list_empty(&trans->new_bgs))
3002 btrfs_create_pending_block_groups(trans);
3004 spin_lock(&delayed_refs->lock);
3005 node = rb_first_cached(&delayed_refs->href_root);
3007 spin_unlock(&delayed_refs->lock);
3010 head = rb_entry(node, struct btrfs_delayed_ref_head,
3012 refcount_inc(&head->refs);
3013 spin_unlock(&delayed_refs->lock);
3015 /* Mutex was contended, block until it's released and retry. */
3016 mutex_lock(&head->mutex);
3017 mutex_unlock(&head->mutex);
3019 btrfs_put_delayed_ref_head(head);
3027 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3028 struct btrfs_fs_info *fs_info,
3029 u64 bytenr, u64 num_bytes, u64 flags,
3030 int level, int is_data)
3032 struct btrfs_delayed_extent_op *extent_op;
3035 extent_op = btrfs_alloc_delayed_extent_op();
3039 extent_op->flags_to_set = flags;
3040 extent_op->update_flags = true;
3041 extent_op->update_key = false;
3042 extent_op->is_data = is_data ? true : false;
3043 extent_op->level = level;
3045 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3046 num_bytes, extent_op);
3048 btrfs_free_delayed_extent_op(extent_op);
3052 static noinline int check_delayed_ref(struct btrfs_root *root,
3053 struct btrfs_path *path,
3054 u64 objectid, u64 offset, u64 bytenr)
3056 struct btrfs_delayed_ref_head *head;
3057 struct btrfs_delayed_ref_node *ref;
3058 struct btrfs_delayed_data_ref *data_ref;
3059 struct btrfs_delayed_ref_root *delayed_refs;
3060 struct btrfs_transaction *cur_trans;
3061 struct rb_node *node;
3064 spin_lock(&root->fs_info->trans_lock);
3065 cur_trans = root->fs_info->running_transaction;
3067 refcount_inc(&cur_trans->use_count);
3068 spin_unlock(&root->fs_info->trans_lock);
3072 delayed_refs = &cur_trans->delayed_refs;
3073 spin_lock(&delayed_refs->lock);
3074 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3076 spin_unlock(&delayed_refs->lock);
3077 btrfs_put_transaction(cur_trans);
3081 if (!mutex_trylock(&head->mutex)) {
3082 refcount_inc(&head->refs);
3083 spin_unlock(&delayed_refs->lock);
3085 btrfs_release_path(path);
3088 * Mutex was contended, block until it's released and let
3091 mutex_lock(&head->mutex);
3092 mutex_unlock(&head->mutex);
3093 btrfs_put_delayed_ref_head(head);
3094 btrfs_put_transaction(cur_trans);
3097 spin_unlock(&delayed_refs->lock);
3099 spin_lock(&head->lock);
3101 * XXX: We should replace this with a proper search function in the
3104 for (node = rb_first_cached(&head->ref_tree); node;
3105 node = rb_next(node)) {
3106 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3107 /* If it's a shared ref we know a cross reference exists */
3108 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3113 data_ref = btrfs_delayed_node_to_data_ref(ref);
3116 * If our ref doesn't match the one we're currently looking at
3117 * then we have a cross reference.
3119 if (data_ref->root != root->root_key.objectid ||
3120 data_ref->objectid != objectid ||
3121 data_ref->offset != offset) {
3126 spin_unlock(&head->lock);
3127 mutex_unlock(&head->mutex);
3128 btrfs_put_transaction(cur_trans);
3132 static noinline int check_committed_ref(struct btrfs_root *root,
3133 struct btrfs_path *path,
3134 u64 objectid, u64 offset, u64 bytenr)
3136 struct btrfs_fs_info *fs_info = root->fs_info;
3137 struct btrfs_root *extent_root = fs_info->extent_root;
3138 struct extent_buffer *leaf;
3139 struct btrfs_extent_data_ref *ref;
3140 struct btrfs_extent_inline_ref *iref;
3141 struct btrfs_extent_item *ei;
3142 struct btrfs_key key;
3147 key.objectid = bytenr;
3148 key.offset = (u64)-1;
3149 key.type = BTRFS_EXTENT_ITEM_KEY;
3151 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3154 BUG_ON(ret == 0); /* Corruption */
3157 if (path->slots[0] == 0)
3161 leaf = path->nodes[0];
3162 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3164 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3168 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3169 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3171 if (item_size != sizeof(*ei) +
3172 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3175 if (btrfs_extent_generation(leaf, ei) <=
3176 btrfs_root_last_snapshot(&root->root_item))
3179 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3181 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3182 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3185 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3186 if (btrfs_extent_refs(leaf, ei) !=
3187 btrfs_extent_data_ref_count(leaf, ref) ||
3188 btrfs_extent_data_ref_root(leaf, ref) !=
3189 root->root_key.objectid ||
3190 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3191 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3199 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3202 struct btrfs_path *path;
3205 path = btrfs_alloc_path();
3210 ret = check_committed_ref(root, path, objectid,
3212 if (ret && ret != -ENOENT)
3215 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3216 } while (ret == -EAGAIN);
3219 btrfs_free_path(path);
3220 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3225 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3226 struct btrfs_root *root,
3227 struct extent_buffer *buf,
3228 int full_backref, int inc)
3230 struct btrfs_fs_info *fs_info = root->fs_info;
3236 struct btrfs_key key;
3237 struct btrfs_file_extent_item *fi;
3241 int (*process_func)(struct btrfs_trans_handle *,
3242 struct btrfs_root *,
3243 u64, u64, u64, u64, u64, u64);
3246 if (btrfs_is_testing(fs_info))
3249 ref_root = btrfs_header_owner(buf);
3250 nritems = btrfs_header_nritems(buf);
3251 level = btrfs_header_level(buf);
3253 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3257 process_func = btrfs_inc_extent_ref;
3259 process_func = btrfs_free_extent;
3262 parent = buf->start;
3266 for (i = 0; i < nritems; i++) {
3268 btrfs_item_key_to_cpu(buf, &key, i);
3269 if (key.type != BTRFS_EXTENT_DATA_KEY)
3271 fi = btrfs_item_ptr(buf, i,
3272 struct btrfs_file_extent_item);
3273 if (btrfs_file_extent_type(buf, fi) ==
3274 BTRFS_FILE_EXTENT_INLINE)
3276 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3280 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3281 key.offset -= btrfs_file_extent_offset(buf, fi);
3282 ret = process_func(trans, root, bytenr, num_bytes,
3283 parent, ref_root, key.objectid,
3288 bytenr = btrfs_node_blockptr(buf, i);
3289 num_bytes = fs_info->nodesize;
3290 ret = process_func(trans, root, bytenr, num_bytes,
3291 parent, ref_root, level - 1, 0);
3301 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3302 struct extent_buffer *buf, int full_backref)
3304 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3307 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3308 struct extent_buffer *buf, int full_backref)
3310 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3313 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3314 struct btrfs_fs_info *fs_info,
3315 struct btrfs_path *path,
3316 struct btrfs_block_group_cache *cache)
3319 struct btrfs_root *extent_root = fs_info->extent_root;
3321 struct extent_buffer *leaf;
3323 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3330 leaf = path->nodes[0];
3331 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3332 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3333 btrfs_mark_buffer_dirty(leaf);
3335 btrfs_release_path(path);
3340 static struct btrfs_block_group_cache *
3341 next_block_group(struct btrfs_fs_info *fs_info,
3342 struct btrfs_block_group_cache *cache)
3344 struct rb_node *node;
3346 spin_lock(&fs_info->block_group_cache_lock);
3348 /* If our block group was removed, we need a full search. */
3349 if (RB_EMPTY_NODE(&cache->cache_node)) {
3350 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3352 spin_unlock(&fs_info->block_group_cache_lock);
3353 btrfs_put_block_group(cache);
3354 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3356 node = rb_next(&cache->cache_node);
3357 btrfs_put_block_group(cache);
3359 cache = rb_entry(node, struct btrfs_block_group_cache,
3361 btrfs_get_block_group(cache);
3364 spin_unlock(&fs_info->block_group_cache_lock);
3368 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3369 struct btrfs_trans_handle *trans,
3370 struct btrfs_path *path)
3372 struct btrfs_fs_info *fs_info = block_group->fs_info;
3373 struct btrfs_root *root = fs_info->tree_root;
3374 struct inode *inode = NULL;
3375 struct extent_changeset *data_reserved = NULL;
3377 int dcs = BTRFS_DC_ERROR;
3383 * If this block group is smaller than 100 megs don't bother caching the
3386 if (block_group->key.offset < (100 * SZ_1M)) {
3387 spin_lock(&block_group->lock);
3388 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3389 spin_unlock(&block_group->lock);
3396 inode = lookup_free_space_inode(fs_info, block_group, path);
3397 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3398 ret = PTR_ERR(inode);
3399 btrfs_release_path(path);
3403 if (IS_ERR(inode)) {
3407 if (block_group->ro)
3410 ret = create_free_space_inode(fs_info, trans, block_group,
3418 * We want to set the generation to 0, that way if anything goes wrong
3419 * from here on out we know not to trust this cache when we load up next
3422 BTRFS_I(inode)->generation = 0;
3423 ret = btrfs_update_inode(trans, root, inode);
3426 * So theoretically we could recover from this, simply set the
3427 * super cache generation to 0 so we know to invalidate the
3428 * cache, but then we'd have to keep track of the block groups
3429 * that fail this way so we know we _have_ to reset this cache
3430 * before the next commit or risk reading stale cache. So to
3431 * limit our exposure to horrible edge cases lets just abort the
3432 * transaction, this only happens in really bad situations
3435 btrfs_abort_transaction(trans, ret);
3440 /* We've already setup this transaction, go ahead and exit */
3441 if (block_group->cache_generation == trans->transid &&
3442 i_size_read(inode)) {
3443 dcs = BTRFS_DC_SETUP;
3447 if (i_size_read(inode) > 0) {
3448 ret = btrfs_check_trunc_cache_free_space(fs_info,
3449 &fs_info->global_block_rsv);
3453 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3458 spin_lock(&block_group->lock);
3459 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3460 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3462 * don't bother trying to write stuff out _if_
3463 * a) we're not cached,
3464 * b) we're with nospace_cache mount option,
3465 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3467 dcs = BTRFS_DC_WRITTEN;
3468 spin_unlock(&block_group->lock);
3471 spin_unlock(&block_group->lock);
3474 * We hit an ENOSPC when setting up the cache in this transaction, just
3475 * skip doing the setup, we've already cleared the cache so we're safe.
3477 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3483 * Try to preallocate enough space based on how big the block group is.
3484 * Keep in mind this has to include any pinned space which could end up
3485 * taking up quite a bit since it's not folded into the other space
3488 num_pages = div_u64(block_group->key.offset, SZ_256M);
3493 num_pages *= PAGE_SIZE;
3495 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3499 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3500 num_pages, num_pages,
3503 * Our cache requires contiguous chunks so that we don't modify a bunch
3504 * of metadata or split extents when writing the cache out, which means
3505 * we can enospc if we are heavily fragmented in addition to just normal
3506 * out of space conditions. So if we hit this just skip setting up any
3507 * other block groups for this transaction, maybe we'll unpin enough
3508 * space the next time around.
3511 dcs = BTRFS_DC_SETUP;
3512 else if (ret == -ENOSPC)
3513 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3518 btrfs_release_path(path);
3520 spin_lock(&block_group->lock);
3521 if (!ret && dcs == BTRFS_DC_SETUP)
3522 block_group->cache_generation = trans->transid;
3523 block_group->disk_cache_state = dcs;
3524 spin_unlock(&block_group->lock);
3526 extent_changeset_free(data_reserved);
3530 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3531 struct btrfs_fs_info *fs_info)
3533 struct btrfs_block_group_cache *cache, *tmp;
3534 struct btrfs_transaction *cur_trans = trans->transaction;
3535 struct btrfs_path *path;
3537 if (list_empty(&cur_trans->dirty_bgs) ||
3538 !btrfs_test_opt(fs_info, SPACE_CACHE))
3541 path = btrfs_alloc_path();
3545 /* Could add new block groups, use _safe just in case */
3546 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3548 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3549 cache_save_setup(cache, trans, path);
3552 btrfs_free_path(path);
3557 * transaction commit does final block group cache writeback during a
3558 * critical section where nothing is allowed to change the FS. This is
3559 * required in order for the cache to actually match the block group,
3560 * but can introduce a lot of latency into the commit.
3562 * So, btrfs_start_dirty_block_groups is here to kick off block group
3563 * cache IO. There's a chance we'll have to redo some of it if the
3564 * block group changes again during the commit, but it greatly reduces
3565 * the commit latency by getting rid of the easy block groups while
3566 * we're still allowing others to join the commit.
3568 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3570 struct btrfs_fs_info *fs_info = trans->fs_info;
3571 struct btrfs_block_group_cache *cache;
3572 struct btrfs_transaction *cur_trans = trans->transaction;
3575 struct btrfs_path *path = NULL;
3577 struct list_head *io = &cur_trans->io_bgs;
3578 int num_started = 0;
3581 spin_lock(&cur_trans->dirty_bgs_lock);
3582 if (list_empty(&cur_trans->dirty_bgs)) {
3583 spin_unlock(&cur_trans->dirty_bgs_lock);
3586 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3587 spin_unlock(&cur_trans->dirty_bgs_lock);
3591 * make sure all the block groups on our dirty list actually
3594 btrfs_create_pending_block_groups(trans);
3597 path = btrfs_alloc_path();
3603 * cache_write_mutex is here only to save us from balance or automatic
3604 * removal of empty block groups deleting this block group while we are
3605 * writing out the cache
3607 mutex_lock(&trans->transaction->cache_write_mutex);
3608 while (!list_empty(&dirty)) {
3609 cache = list_first_entry(&dirty,
3610 struct btrfs_block_group_cache,
3613 * this can happen if something re-dirties a block
3614 * group that is already under IO. Just wait for it to
3615 * finish and then do it all again
3617 if (!list_empty(&cache->io_list)) {
3618 list_del_init(&cache->io_list);
3619 btrfs_wait_cache_io(trans, cache, path);
3620 btrfs_put_block_group(cache);
3625 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3626 * if it should update the cache_state. Don't delete
3627 * until after we wait.
3629 * Since we're not running in the commit critical section
3630 * we need the dirty_bgs_lock to protect from update_block_group
3632 spin_lock(&cur_trans->dirty_bgs_lock);
3633 list_del_init(&cache->dirty_list);
3634 spin_unlock(&cur_trans->dirty_bgs_lock);
3638 cache_save_setup(cache, trans, path);
3640 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3641 cache->io_ctl.inode = NULL;
3642 ret = btrfs_write_out_cache(fs_info, trans,
3644 if (ret == 0 && cache->io_ctl.inode) {
3649 * The cache_write_mutex is protecting the
3650 * io_list, also refer to the definition of
3651 * btrfs_transaction::io_bgs for more details
3653 list_add_tail(&cache->io_list, io);
3656 * if we failed to write the cache, the
3657 * generation will be bad and life goes on
3663 ret = write_one_cache_group(trans, fs_info,
3666 * Our block group might still be attached to the list
3667 * of new block groups in the transaction handle of some
3668 * other task (struct btrfs_trans_handle->new_bgs). This
3669 * means its block group item isn't yet in the extent
3670 * tree. If this happens ignore the error, as we will
3671 * try again later in the critical section of the
3672 * transaction commit.
3674 if (ret == -ENOENT) {
3676 spin_lock(&cur_trans->dirty_bgs_lock);
3677 if (list_empty(&cache->dirty_list)) {
3678 list_add_tail(&cache->dirty_list,
3679 &cur_trans->dirty_bgs);
3680 btrfs_get_block_group(cache);
3682 spin_unlock(&cur_trans->dirty_bgs_lock);
3684 btrfs_abort_transaction(trans, ret);
3688 /* if its not on the io list, we need to put the block group */
3690 btrfs_put_block_group(cache);
3696 * Avoid blocking other tasks for too long. It might even save
3697 * us from writing caches for block groups that are going to be
3700 mutex_unlock(&trans->transaction->cache_write_mutex);
3701 mutex_lock(&trans->transaction->cache_write_mutex);
3703 mutex_unlock(&trans->transaction->cache_write_mutex);
3706 * go through delayed refs for all the stuff we've just kicked off
3707 * and then loop back (just once)
3709 ret = btrfs_run_delayed_refs(trans, 0);
3710 if (!ret && loops == 0) {
3712 spin_lock(&cur_trans->dirty_bgs_lock);
3713 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3715 * dirty_bgs_lock protects us from concurrent block group
3716 * deletes too (not just cache_write_mutex).
3718 if (!list_empty(&dirty)) {
3719 spin_unlock(&cur_trans->dirty_bgs_lock);
3722 spin_unlock(&cur_trans->dirty_bgs_lock);
3723 } else if (ret < 0) {
3724 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3727 btrfs_free_path(path);
3731 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3732 struct btrfs_fs_info *fs_info)
3734 struct btrfs_block_group_cache *cache;
3735 struct btrfs_transaction *cur_trans = trans->transaction;
3738 struct btrfs_path *path;
3739 struct list_head *io = &cur_trans->io_bgs;
3740 int num_started = 0;
3742 path = btrfs_alloc_path();
3747 * Even though we are in the critical section of the transaction commit,
3748 * we can still have concurrent tasks adding elements to this
3749 * transaction's list of dirty block groups. These tasks correspond to
3750 * endio free space workers started when writeback finishes for a
3751 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3752 * allocate new block groups as a result of COWing nodes of the root
3753 * tree when updating the free space inode. The writeback for the space
3754 * caches is triggered by an earlier call to
3755 * btrfs_start_dirty_block_groups() and iterations of the following
3757 * Also we want to do the cache_save_setup first and then run the
3758 * delayed refs to make sure we have the best chance at doing this all
3761 spin_lock(&cur_trans->dirty_bgs_lock);
3762 while (!list_empty(&cur_trans->dirty_bgs)) {
3763 cache = list_first_entry(&cur_trans->dirty_bgs,
3764 struct btrfs_block_group_cache,
3768 * this can happen if cache_save_setup re-dirties a block
3769 * group that is already under IO. Just wait for it to
3770 * finish and then do it all again
3772 if (!list_empty(&cache->io_list)) {
3773 spin_unlock(&cur_trans->dirty_bgs_lock);
3774 list_del_init(&cache->io_list);
3775 btrfs_wait_cache_io(trans, cache, path);
3776 btrfs_put_block_group(cache);
3777 spin_lock(&cur_trans->dirty_bgs_lock);
3781 * don't remove from the dirty list until after we've waited
3784 list_del_init(&cache->dirty_list);
3785 spin_unlock(&cur_trans->dirty_bgs_lock);
3788 cache_save_setup(cache, trans, path);
3791 ret = btrfs_run_delayed_refs(trans,
3792 (unsigned long) -1);
3794 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3795 cache->io_ctl.inode = NULL;
3796 ret = btrfs_write_out_cache(fs_info, trans,
3798 if (ret == 0 && cache->io_ctl.inode) {
3801 list_add_tail(&cache->io_list, io);
3804 * if we failed to write the cache, the
3805 * generation will be bad and life goes on
3811 ret = write_one_cache_group(trans, fs_info,
3814 * One of the free space endio workers might have
3815 * created a new block group while updating a free space
3816 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3817 * and hasn't released its transaction handle yet, in
3818 * which case the new block group is still attached to
3819 * its transaction handle and its creation has not
3820 * finished yet (no block group item in the extent tree
3821 * yet, etc). If this is the case, wait for all free
3822 * space endio workers to finish and retry. This is a
3823 * a very rare case so no need for a more efficient and
3826 if (ret == -ENOENT) {
3827 wait_event(cur_trans->writer_wait,
3828 atomic_read(&cur_trans->num_writers) == 1);
3829 ret = write_one_cache_group(trans, fs_info,
3833 btrfs_abort_transaction(trans, ret);
3836 /* if its not on the io list, we need to put the block group */
3838 btrfs_put_block_group(cache);
3839 spin_lock(&cur_trans->dirty_bgs_lock);
3841 spin_unlock(&cur_trans->dirty_bgs_lock);
3844 * Refer to the definition of io_bgs member for details why it's safe
3845 * to use it without any locking
3847 while (!list_empty(io)) {
3848 cache = list_first_entry(io, struct btrfs_block_group_cache,
3850 list_del_init(&cache->io_list);
3851 btrfs_wait_cache_io(trans, cache, path);
3852 btrfs_put_block_group(cache);
3855 btrfs_free_path(path);
3859 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3861 struct btrfs_block_group_cache *block_group;
3864 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3865 if (!block_group || block_group->ro)
3868 btrfs_put_block_group(block_group);
3872 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3874 struct btrfs_block_group_cache *bg;
3877 bg = btrfs_lookup_block_group(fs_info, bytenr);
3881 spin_lock(&bg->lock);
3885 atomic_inc(&bg->nocow_writers);
3886 spin_unlock(&bg->lock);
3888 /* no put on block group, done by btrfs_dec_nocow_writers */
3890 btrfs_put_block_group(bg);
3896 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3898 struct btrfs_block_group_cache *bg;
3900 bg = btrfs_lookup_block_group(fs_info, bytenr);
3902 if (atomic_dec_and_test(&bg->nocow_writers))
3903 wake_up_var(&bg->nocow_writers);
3905 * Once for our lookup and once for the lookup done by a previous call
3906 * to btrfs_inc_nocow_writers()
3908 btrfs_put_block_group(bg);
3909 btrfs_put_block_group(bg);
3912 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3914 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3917 static const char *alloc_name(u64 flags)
3920 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3922 case BTRFS_BLOCK_GROUP_METADATA:
3924 case BTRFS_BLOCK_GROUP_DATA:
3926 case BTRFS_BLOCK_GROUP_SYSTEM:
3930 return "invalid-combination";
3934 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3937 struct btrfs_space_info *space_info;
3941 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3945 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3952 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3953 INIT_LIST_HEAD(&space_info->block_groups[i]);
3954 init_rwsem(&space_info->groups_sem);
3955 spin_lock_init(&space_info->lock);
3956 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3957 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3958 init_waitqueue_head(&space_info->wait);
3959 INIT_LIST_HEAD(&space_info->ro_bgs);
3960 INIT_LIST_HEAD(&space_info->tickets);
3961 INIT_LIST_HEAD(&space_info->priority_tickets);
3963 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3964 info->space_info_kobj, "%s",
3965 alloc_name(space_info->flags));
3967 percpu_counter_destroy(&space_info->total_bytes_pinned);
3972 list_add_rcu(&space_info->list, &info->space_info);
3973 if (flags & BTRFS_BLOCK_GROUP_DATA)
3974 info->data_sinfo = space_info;
3979 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3980 u64 total_bytes, u64 bytes_used,
3982 struct btrfs_space_info **space_info)
3984 struct btrfs_space_info *found;
3987 factor = btrfs_bg_type_to_factor(flags);
3989 found = __find_space_info(info, flags);
3991 spin_lock(&found->lock);
3992 found->total_bytes += total_bytes;
3993 found->disk_total += total_bytes * factor;
3994 found->bytes_used += bytes_used;
3995 found->disk_used += bytes_used * factor;
3996 found->bytes_readonly += bytes_readonly;
3997 if (total_bytes > 0)
3999 space_info_add_new_bytes(info, found, total_bytes -
4000 bytes_used - bytes_readonly);
4001 spin_unlock(&found->lock);
4002 *space_info = found;
4005 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4007 u64 extra_flags = chunk_to_extended(flags) &
4008 BTRFS_EXTENDED_PROFILE_MASK;
4010 write_seqlock(&fs_info->profiles_lock);
4011 if (flags & BTRFS_BLOCK_GROUP_DATA)
4012 fs_info->avail_data_alloc_bits |= extra_flags;
4013 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4014 fs_info->avail_metadata_alloc_bits |= extra_flags;
4015 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4016 fs_info->avail_system_alloc_bits |= extra_flags;
4017 write_sequnlock(&fs_info->profiles_lock);
4021 * returns target flags in extended format or 0 if restripe for this
4022 * chunk_type is not in progress
4024 * should be called with balance_lock held
4026 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4028 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4034 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4035 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4036 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4037 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4038 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4039 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4040 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4041 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4042 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4049 * @flags: available profiles in extended format (see ctree.h)
4051 * Returns reduced profile in chunk format. If profile changing is in
4052 * progress (either running or paused) picks the target profile (if it's
4053 * already available), otherwise falls back to plain reducing.
4055 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4057 u64 num_devices = fs_info->fs_devices->rw_devices;
4063 * see if restripe for this chunk_type is in progress, if so
4064 * try to reduce to the target profile
4066 spin_lock(&fs_info->balance_lock);
4067 target = get_restripe_target(fs_info, flags);
4069 /* pick target profile only if it's already available */
4070 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4071 spin_unlock(&fs_info->balance_lock);
4072 return extended_to_chunk(target);
4075 spin_unlock(&fs_info->balance_lock);
4077 /* First, mask out the RAID levels which aren't possible */
4078 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4079 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4080 allowed |= btrfs_raid_array[raid_type].bg_flag;
4084 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4085 allowed = BTRFS_BLOCK_GROUP_RAID6;
4086 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4087 allowed = BTRFS_BLOCK_GROUP_RAID5;
4088 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4089 allowed = BTRFS_BLOCK_GROUP_RAID10;
4090 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4091 allowed = BTRFS_BLOCK_GROUP_RAID1;
4092 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4093 allowed = BTRFS_BLOCK_GROUP_RAID0;
4095 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4097 return extended_to_chunk(flags | allowed);
4100 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4107 seq = read_seqbegin(&fs_info->profiles_lock);
4109 if (flags & BTRFS_BLOCK_GROUP_DATA)
4110 flags |= fs_info->avail_data_alloc_bits;
4111 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4112 flags |= fs_info->avail_system_alloc_bits;
4113 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4114 flags |= fs_info->avail_metadata_alloc_bits;
4115 } while (read_seqretry(&fs_info->profiles_lock, seq));
4117 return btrfs_reduce_alloc_profile(fs_info, flags);
4120 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4122 struct btrfs_fs_info *fs_info = root->fs_info;
4127 flags = BTRFS_BLOCK_GROUP_DATA;
4128 else if (root == fs_info->chunk_root)
4129 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4131 flags = BTRFS_BLOCK_GROUP_METADATA;
4133 ret = get_alloc_profile(fs_info, flags);
4137 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4139 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4142 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4144 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4147 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4149 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4152 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4153 bool may_use_included)
4156 return s_info->bytes_used + s_info->bytes_reserved +
4157 s_info->bytes_pinned + s_info->bytes_readonly +
4158 (may_use_included ? s_info->bytes_may_use : 0);
4161 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4163 struct btrfs_root *root = inode->root;
4164 struct btrfs_fs_info *fs_info = root->fs_info;
4165 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4168 int need_commit = 2;
4169 int have_pinned_space;
4171 /* make sure bytes are sectorsize aligned */
4172 bytes = ALIGN(bytes, fs_info->sectorsize);
4174 if (btrfs_is_free_space_inode(inode)) {
4176 ASSERT(current->journal_info);
4180 /* make sure we have enough space to handle the data first */
4181 spin_lock(&data_sinfo->lock);
4182 used = btrfs_space_info_used(data_sinfo, true);
4184 if (used + bytes > data_sinfo->total_bytes) {
4185 struct btrfs_trans_handle *trans;
4188 * if we don't have enough free bytes in this space then we need
4189 * to alloc a new chunk.
4191 if (!data_sinfo->full) {
4194 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4195 spin_unlock(&data_sinfo->lock);
4197 alloc_target = btrfs_data_alloc_profile(fs_info);
4199 * It is ugly that we don't call nolock join
4200 * transaction for the free space inode case here.
4201 * But it is safe because we only do the data space
4202 * reservation for the free space cache in the
4203 * transaction context, the common join transaction
4204 * just increase the counter of the current transaction
4205 * handler, doesn't try to acquire the trans_lock of
4208 trans = btrfs_join_transaction(root);
4210 return PTR_ERR(trans);
4212 ret = do_chunk_alloc(trans, alloc_target,
4213 CHUNK_ALLOC_NO_FORCE);
4214 btrfs_end_transaction(trans);
4219 have_pinned_space = 1;
4228 * If we don't have enough pinned space to deal with this
4229 * allocation, and no removed chunk in current transaction,
4230 * don't bother committing the transaction.
4232 have_pinned_space = __percpu_counter_compare(
4233 &data_sinfo->total_bytes_pinned,
4234 used + bytes - data_sinfo->total_bytes,
4235 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4236 spin_unlock(&data_sinfo->lock);
4238 /* commit the current transaction and try again */
4243 if (need_commit > 0) {
4244 btrfs_start_delalloc_roots(fs_info, -1);
4245 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4249 trans = btrfs_join_transaction(root);
4251 return PTR_ERR(trans);
4252 if (have_pinned_space >= 0 ||
4253 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4254 &trans->transaction->flags) ||
4256 ret = btrfs_commit_transaction(trans);
4260 * The cleaner kthread might still be doing iput
4261 * operations. Wait for it to finish so that
4262 * more space is released.
4264 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4265 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4268 btrfs_end_transaction(trans);
4272 trace_btrfs_space_reservation(fs_info,
4273 "space_info:enospc",
4274 data_sinfo->flags, bytes, 1);
4277 update_bytes_may_use(data_sinfo, bytes);
4278 trace_btrfs_space_reservation(fs_info, "space_info",
4279 data_sinfo->flags, bytes, 1);
4280 spin_unlock(&data_sinfo->lock);
4285 int btrfs_check_data_free_space(struct inode *inode,
4286 struct extent_changeset **reserved, u64 start, u64 len)
4288 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4291 /* align the range */
4292 len = round_up(start + len, fs_info->sectorsize) -
4293 round_down(start, fs_info->sectorsize);
4294 start = round_down(start, fs_info->sectorsize);
4296 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4300 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4301 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4303 btrfs_free_reserved_data_space_noquota(inode, start, len);
4310 * Called if we need to clear a data reservation for this inode
4311 * Normally in a error case.
4313 * This one will *NOT* use accurate qgroup reserved space API, just for case
4314 * which we can't sleep and is sure it won't affect qgroup reserved space.
4315 * Like clear_bit_hook().
4317 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4320 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4321 struct btrfs_space_info *data_sinfo;
4323 /* Make sure the range is aligned to sectorsize */
4324 len = round_up(start + len, fs_info->sectorsize) -
4325 round_down(start, fs_info->sectorsize);
4326 start = round_down(start, fs_info->sectorsize);
4328 data_sinfo = fs_info->data_sinfo;
4329 spin_lock(&data_sinfo->lock);
4330 update_bytes_may_use(data_sinfo, -len);
4331 trace_btrfs_space_reservation(fs_info, "space_info",
4332 data_sinfo->flags, len, 0);
4333 spin_unlock(&data_sinfo->lock);
4337 * Called if we need to clear a data reservation for this inode
4338 * Normally in a error case.
4340 * This one will handle the per-inode data rsv map for accurate reserved
4343 void btrfs_free_reserved_data_space(struct inode *inode,
4344 struct extent_changeset *reserved, u64 start, u64 len)
4346 struct btrfs_root *root = BTRFS_I(inode)->root;
4348 /* Make sure the range is aligned to sectorsize */
4349 len = round_up(start + len, root->fs_info->sectorsize) -
4350 round_down(start, root->fs_info->sectorsize);
4351 start = round_down(start, root->fs_info->sectorsize);
4353 btrfs_free_reserved_data_space_noquota(inode, start, len);
4354 btrfs_qgroup_free_data(inode, reserved, start, len);
4357 static void force_metadata_allocation(struct btrfs_fs_info *info)
4359 struct list_head *head = &info->space_info;
4360 struct btrfs_space_info *found;
4363 list_for_each_entry_rcu(found, head, list) {
4364 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4365 found->force_alloc = CHUNK_ALLOC_FORCE;
4370 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4372 return (global->size << 1);
4375 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4376 struct btrfs_space_info *sinfo, int force)
4378 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4379 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4382 if (force == CHUNK_ALLOC_FORCE)
4386 * We need to take into account the global rsv because for all intents
4387 * and purposes it's used space. Don't worry about locking the
4388 * global_rsv, it doesn't change except when the transaction commits.
4390 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4391 bytes_used += calc_global_rsv_need_space(global_rsv);
4394 * in limited mode, we want to have some free space up to
4395 * about 1% of the FS size.
4397 if (force == CHUNK_ALLOC_LIMITED) {
4398 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4399 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4401 if (sinfo->total_bytes - bytes_used < thresh)
4405 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4410 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4414 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4415 BTRFS_BLOCK_GROUP_RAID0 |
4416 BTRFS_BLOCK_GROUP_RAID5 |
4417 BTRFS_BLOCK_GROUP_RAID6))
4418 num_dev = fs_info->fs_devices->rw_devices;
4419 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4422 num_dev = 1; /* DUP or single */
4428 * If @is_allocation is true, reserve space in the system space info necessary
4429 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4432 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4434 struct btrfs_fs_info *fs_info = trans->fs_info;
4435 struct btrfs_space_info *info;
4442 * Needed because we can end up allocating a system chunk and for an
4443 * atomic and race free space reservation in the chunk block reserve.
4445 lockdep_assert_held(&fs_info->chunk_mutex);
4447 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4448 spin_lock(&info->lock);
4449 left = info->total_bytes - btrfs_space_info_used(info, true);
4450 spin_unlock(&info->lock);
4452 num_devs = get_profile_num_devs(fs_info, type);
4454 /* num_devs device items to update and 1 chunk item to add or remove */
4455 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4456 btrfs_calc_trans_metadata_size(fs_info, 1);
4458 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4459 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4460 left, thresh, type);
4461 dump_space_info(fs_info, info, 0, 0);
4464 if (left < thresh) {
4465 u64 flags = btrfs_system_alloc_profile(fs_info);
4468 * Ignore failure to create system chunk. We might end up not
4469 * needing it, as we might not need to COW all nodes/leafs from
4470 * the paths we visit in the chunk tree (they were already COWed
4471 * or created in the current transaction for example).
4473 ret = btrfs_alloc_chunk(trans, flags);
4477 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4478 &fs_info->chunk_block_rsv,
4479 thresh, BTRFS_RESERVE_NO_FLUSH);
4481 trans->chunk_bytes_reserved += thresh;
4486 * If force is CHUNK_ALLOC_FORCE:
4487 * - return 1 if it successfully allocates a chunk,
4488 * - return errors including -ENOSPC otherwise.
4489 * If force is NOT CHUNK_ALLOC_FORCE:
4490 * - return 0 if it doesn't need to allocate a new chunk,
4491 * - return 1 if it successfully allocates a chunk,
4492 * - return errors including -ENOSPC otherwise.
4494 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4497 struct btrfs_fs_info *fs_info = trans->fs_info;
4498 struct btrfs_space_info *space_info;
4499 bool wait_for_alloc = false;
4500 bool should_alloc = false;
4503 /* Don't re-enter if we're already allocating a chunk */
4504 if (trans->allocating_chunk)
4507 space_info = __find_space_info(fs_info, flags);
4511 spin_lock(&space_info->lock);
4512 if (force < space_info->force_alloc)
4513 force = space_info->force_alloc;
4514 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4515 if (space_info->full) {
4516 /* No more free physical space */
4521 spin_unlock(&space_info->lock);
4523 } else if (!should_alloc) {
4524 spin_unlock(&space_info->lock);
4526 } else if (space_info->chunk_alloc) {
4528 * Someone is already allocating, so we need to block
4529 * until this someone is finished and then loop to
4530 * recheck if we should continue with our allocation
4533 wait_for_alloc = true;
4534 spin_unlock(&space_info->lock);
4535 mutex_lock(&fs_info->chunk_mutex);
4536 mutex_unlock(&fs_info->chunk_mutex);
4538 /* Proceed with allocation */
4539 space_info->chunk_alloc = 1;
4540 wait_for_alloc = false;
4541 spin_unlock(&space_info->lock);
4545 } while (wait_for_alloc);
4547 mutex_lock(&fs_info->chunk_mutex);
4548 trans->allocating_chunk = true;
4551 * If we have mixed data/metadata chunks we want to make sure we keep
4552 * allocating mixed chunks instead of individual chunks.
4554 if (btrfs_mixed_space_info(space_info))
4555 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4558 * if we're doing a data chunk, go ahead and make sure that
4559 * we keep a reasonable number of metadata chunks allocated in the
4562 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4563 fs_info->data_chunk_allocations++;
4564 if (!(fs_info->data_chunk_allocations %
4565 fs_info->metadata_ratio))
4566 force_metadata_allocation(fs_info);
4570 * Check if we have enough space in SYSTEM chunk because we may need
4571 * to update devices.
4573 check_system_chunk(trans, flags);
4575 ret = btrfs_alloc_chunk(trans, flags);
4576 trans->allocating_chunk = false;
4578 spin_lock(&space_info->lock);
4581 space_info->full = 1;
4586 space_info->max_extent_size = 0;
4589 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4591 space_info->chunk_alloc = 0;
4592 spin_unlock(&space_info->lock);
4593 mutex_unlock(&fs_info->chunk_mutex);
4595 * When we allocate a new chunk we reserve space in the chunk block
4596 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4597 * add new nodes/leafs to it if we end up needing to do it when
4598 * inserting the chunk item and updating device items as part of the
4599 * second phase of chunk allocation, performed by
4600 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4601 * large number of new block groups to create in our transaction
4602 * handle's new_bgs list to avoid exhausting the chunk block reserve
4603 * in extreme cases - like having a single transaction create many new
4604 * block groups when starting to write out the free space caches of all
4605 * the block groups that were made dirty during the lifetime of the
4608 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4609 btrfs_create_pending_block_groups(trans);
4614 static int can_overcommit(struct btrfs_fs_info *fs_info,
4615 struct btrfs_space_info *space_info, u64 bytes,
4616 enum btrfs_reserve_flush_enum flush,
4619 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4626 /* Don't overcommit when in mixed mode. */
4627 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4631 profile = btrfs_system_alloc_profile(fs_info);
4633 profile = btrfs_metadata_alloc_profile(fs_info);
4635 used = btrfs_space_info_used(space_info, false);
4638 * We only want to allow over committing if we have lots of actual space
4639 * free, but if we don't have enough space to handle the global reserve
4640 * space then we could end up having a real enospc problem when trying
4641 * to allocate a chunk or some other such important allocation.
4643 spin_lock(&global_rsv->lock);
4644 space_size = calc_global_rsv_need_space(global_rsv);
4645 spin_unlock(&global_rsv->lock);
4646 if (used + space_size >= space_info->total_bytes)
4649 used += space_info->bytes_may_use;
4651 avail = atomic64_read(&fs_info->free_chunk_space);
4654 * If we have dup, raid1 or raid10 then only half of the free
4655 * space is actually useable. For raid56, the space info used
4656 * doesn't include the parity drive, so we don't have to
4659 factor = btrfs_bg_type_to_factor(profile);
4660 avail = div_u64(avail, factor);
4663 * If we aren't flushing all things, let us overcommit up to
4664 * 1/2th of the space. If we can flush, don't let us overcommit
4665 * too much, let it overcommit up to 1/8 of the space.
4667 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4672 if (used + bytes < space_info->total_bytes + avail)
4677 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4678 unsigned long nr_pages, int nr_items)
4680 struct super_block *sb = fs_info->sb;
4682 if (down_read_trylock(&sb->s_umount)) {
4683 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4684 up_read(&sb->s_umount);
4687 * We needn't worry the filesystem going from r/w to r/o though
4688 * we don't acquire ->s_umount mutex, because the filesystem
4689 * should guarantee the delalloc inodes list be empty after
4690 * the filesystem is readonly(all dirty pages are written to
4693 btrfs_start_delalloc_roots(fs_info, nr_items);
4694 if (!current->journal_info)
4695 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4699 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4705 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4706 nr = div64_u64(to_reclaim, bytes);
4712 #define EXTENT_SIZE_PER_ITEM SZ_256K
4715 * shrink metadata reservation for delalloc
4717 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4718 u64 orig, bool wait_ordered)
4720 struct btrfs_space_info *space_info;
4721 struct btrfs_trans_handle *trans;
4726 unsigned long nr_pages;
4729 /* Calc the number of the pages we need flush for space reservation */
4730 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4731 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4733 trans = (struct btrfs_trans_handle *)current->journal_info;
4734 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4736 delalloc_bytes = percpu_counter_sum_positive(
4737 &fs_info->delalloc_bytes);
4738 if (delalloc_bytes == 0) {
4742 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4747 while (delalloc_bytes && loops < 3) {
4748 max_reclaim = min(delalloc_bytes, to_reclaim);
4749 nr_pages = max_reclaim >> PAGE_SHIFT;
4750 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4752 * We need to wait for the async pages to actually start before
4755 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4759 if (max_reclaim <= nr_pages)
4762 max_reclaim -= nr_pages;
4764 wait_event(fs_info->async_submit_wait,
4765 atomic_read(&fs_info->async_delalloc_pages) <=
4768 spin_lock(&space_info->lock);
4769 if (list_empty(&space_info->tickets) &&
4770 list_empty(&space_info->priority_tickets)) {
4771 spin_unlock(&space_info->lock);
4774 spin_unlock(&space_info->lock);
4777 if (wait_ordered && !trans) {
4778 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4780 time_left = schedule_timeout_killable(1);
4784 delalloc_bytes = percpu_counter_sum_positive(
4785 &fs_info->delalloc_bytes);
4789 struct reserve_ticket {
4792 struct list_head list;
4793 wait_queue_head_t wait;
4797 * maybe_commit_transaction - possibly commit the transaction if its ok to
4798 * @root - the root we're allocating for
4799 * @bytes - the number of bytes we want to reserve
4800 * @force - force the commit
4802 * This will check to make sure that committing the transaction will actually
4803 * get us somewhere and then commit the transaction if it does. Otherwise it
4804 * will return -ENOSPC.
4806 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4807 struct btrfs_space_info *space_info)
4809 struct reserve_ticket *ticket = NULL;
4810 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4811 struct btrfs_trans_handle *trans;
4814 trans = (struct btrfs_trans_handle *)current->journal_info;
4818 spin_lock(&space_info->lock);
4819 if (!list_empty(&space_info->priority_tickets))
4820 ticket = list_first_entry(&space_info->priority_tickets,
4821 struct reserve_ticket, list);
4822 else if (!list_empty(&space_info->tickets))
4823 ticket = list_first_entry(&space_info->tickets,
4824 struct reserve_ticket, list);
4825 bytes = (ticket) ? ticket->bytes : 0;
4826 spin_unlock(&space_info->lock);
4831 /* See if there is enough pinned space to make this reservation */
4832 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4834 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4838 * See if there is some space in the delayed insertion reservation for
4841 if (space_info != delayed_rsv->space_info)
4844 spin_lock(&delayed_rsv->lock);
4845 if (delayed_rsv->size > bytes)
4848 bytes -= delayed_rsv->size;
4849 spin_unlock(&delayed_rsv->lock);
4851 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4853 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4858 trans = btrfs_join_transaction(fs_info->extent_root);
4862 return btrfs_commit_transaction(trans);
4866 * Try to flush some data based on policy set by @state. This is only advisory
4867 * and may fail for various reasons. The caller is supposed to examine the
4868 * state of @space_info to detect the outcome.
4870 static void flush_space(struct btrfs_fs_info *fs_info,
4871 struct btrfs_space_info *space_info, u64 num_bytes,
4874 struct btrfs_root *root = fs_info->extent_root;
4875 struct btrfs_trans_handle *trans;
4880 case FLUSH_DELAYED_ITEMS_NR:
4881 case FLUSH_DELAYED_ITEMS:
4882 if (state == FLUSH_DELAYED_ITEMS_NR)
4883 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4887 trans = btrfs_join_transaction(root);
4888 if (IS_ERR(trans)) {
4889 ret = PTR_ERR(trans);
4892 ret = btrfs_run_delayed_items_nr(trans, nr);
4893 btrfs_end_transaction(trans);
4895 case FLUSH_DELALLOC:
4896 case FLUSH_DELALLOC_WAIT:
4897 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4898 state == FLUSH_DELALLOC_WAIT);
4901 trans = btrfs_join_transaction(root);
4902 if (IS_ERR(trans)) {
4903 ret = PTR_ERR(trans);
4906 ret = do_chunk_alloc(trans,
4907 btrfs_metadata_alloc_profile(fs_info),
4908 CHUNK_ALLOC_NO_FORCE);
4909 btrfs_end_transaction(trans);
4910 if (ret > 0 || ret == -ENOSPC)
4914 ret = may_commit_transaction(fs_info, space_info);
4921 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4927 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4928 struct btrfs_space_info *space_info,
4931 struct reserve_ticket *ticket;
4936 list_for_each_entry(ticket, &space_info->tickets, list)
4937 to_reclaim += ticket->bytes;
4938 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4939 to_reclaim += ticket->bytes;
4943 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4944 if (can_overcommit(fs_info, space_info, to_reclaim,
4945 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4948 used = btrfs_space_info_used(space_info, true);
4950 if (can_overcommit(fs_info, space_info, SZ_1M,
4951 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4952 expected = div_factor_fine(space_info->total_bytes, 95);
4954 expected = div_factor_fine(space_info->total_bytes, 90);
4956 if (used > expected)
4957 to_reclaim = used - expected;
4960 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4961 space_info->bytes_reserved);
4965 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
4966 struct btrfs_space_info *space_info,
4967 u64 used, bool system_chunk)
4969 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4971 /* If we're just plain full then async reclaim just slows us down. */
4972 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4975 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4979 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4980 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4983 static void wake_all_tickets(struct list_head *head)
4985 struct reserve_ticket *ticket;
4987 while (!list_empty(head)) {
4988 ticket = list_first_entry(head, struct reserve_ticket, list);
4989 list_del_init(&ticket->list);
4990 ticket->error = -ENOSPC;
4991 wake_up(&ticket->wait);
4996 * This is for normal flushers, we can wait all goddamned day if we want to. We
4997 * will loop and continuously try to flush as long as we are making progress.
4998 * We count progress as clearing off tickets each time we have to loop.
5000 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5002 struct btrfs_fs_info *fs_info;
5003 struct btrfs_space_info *space_info;
5006 int commit_cycles = 0;
5007 u64 last_tickets_id;
5009 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5010 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5012 spin_lock(&space_info->lock);
5013 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5016 space_info->flush = 0;
5017 spin_unlock(&space_info->lock);
5020 last_tickets_id = space_info->tickets_id;
5021 spin_unlock(&space_info->lock);
5023 flush_state = FLUSH_DELAYED_ITEMS_NR;
5025 flush_space(fs_info, space_info, to_reclaim, flush_state);
5026 spin_lock(&space_info->lock);
5027 if (list_empty(&space_info->tickets)) {
5028 space_info->flush = 0;
5029 spin_unlock(&space_info->lock);
5032 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5035 if (last_tickets_id == space_info->tickets_id) {
5038 last_tickets_id = space_info->tickets_id;
5039 flush_state = FLUSH_DELAYED_ITEMS_NR;
5044 if (flush_state > COMMIT_TRANS) {
5046 if (commit_cycles > 2) {
5047 wake_all_tickets(&space_info->tickets);
5048 space_info->flush = 0;
5050 flush_state = FLUSH_DELAYED_ITEMS_NR;
5053 spin_unlock(&space_info->lock);
5054 } while (flush_state <= COMMIT_TRANS);
5057 void btrfs_init_async_reclaim_work(struct work_struct *work)
5059 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5062 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5063 struct btrfs_space_info *space_info,
5064 struct reserve_ticket *ticket)
5067 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5069 spin_lock(&space_info->lock);
5070 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5073 spin_unlock(&space_info->lock);
5076 spin_unlock(&space_info->lock);
5079 flush_space(fs_info, space_info, to_reclaim, flush_state);
5081 spin_lock(&space_info->lock);
5082 if (ticket->bytes == 0) {
5083 spin_unlock(&space_info->lock);
5086 spin_unlock(&space_info->lock);
5089 * Priority flushers can't wait on delalloc without
5092 if (flush_state == FLUSH_DELALLOC ||
5093 flush_state == FLUSH_DELALLOC_WAIT)
5094 flush_state = ALLOC_CHUNK;
5095 } while (flush_state < COMMIT_TRANS);
5098 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5099 struct btrfs_space_info *space_info,
5100 struct reserve_ticket *ticket, u64 orig_bytes)
5106 spin_lock(&space_info->lock);
5107 while (ticket->bytes > 0 && ticket->error == 0) {
5108 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5113 spin_unlock(&space_info->lock);
5117 finish_wait(&ticket->wait, &wait);
5118 spin_lock(&space_info->lock);
5121 ret = ticket->error;
5122 if (!list_empty(&ticket->list))
5123 list_del_init(&ticket->list);
5124 if (ticket->bytes && ticket->bytes < orig_bytes) {
5125 u64 num_bytes = orig_bytes - ticket->bytes;
5126 update_bytes_may_use(space_info, -num_bytes);
5127 trace_btrfs_space_reservation(fs_info, "space_info",
5128 space_info->flags, num_bytes, 0);
5130 spin_unlock(&space_info->lock);
5136 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5137 * @root - the root we're allocating for
5138 * @space_info - the space info we want to allocate from
5139 * @orig_bytes - the number of bytes we want
5140 * @flush - whether or not we can flush to make our reservation
5142 * This will reserve orig_bytes number of bytes from the space info associated
5143 * with the block_rsv. If there is not enough space it will make an attempt to
5144 * flush out space to make room. It will do this by flushing delalloc if
5145 * possible or committing the transaction. If flush is 0 then no attempts to
5146 * regain reservations will be made and this will fail if there is not enough
5149 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5150 struct btrfs_space_info *space_info,
5152 enum btrfs_reserve_flush_enum flush,
5155 struct reserve_ticket ticket;
5160 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5162 spin_lock(&space_info->lock);
5164 used = btrfs_space_info_used(space_info, true);
5167 * If we have enough space then hooray, make our reservation and carry
5168 * on. If not see if we can overcommit, and if we can, hooray carry on.
5169 * If not things get more complicated.
5171 if (used + orig_bytes <= space_info->total_bytes) {
5172 update_bytes_may_use(space_info, orig_bytes);
5173 trace_btrfs_space_reservation(fs_info, "space_info",
5174 space_info->flags, orig_bytes, 1);
5176 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5178 update_bytes_may_use(space_info, orig_bytes);
5179 trace_btrfs_space_reservation(fs_info, "space_info",
5180 space_info->flags, orig_bytes, 1);
5185 * If we couldn't make a reservation then setup our reservation ticket
5186 * and kick the async worker if it's not already running.
5188 * If we are a priority flusher then we just need to add our ticket to
5189 * the list and we will do our own flushing further down.
5191 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5192 ticket.bytes = orig_bytes;
5194 init_waitqueue_head(&ticket.wait);
5195 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5196 list_add_tail(&ticket.list, &space_info->tickets);
5197 if (!space_info->flush) {
5198 space_info->flush = 1;
5199 trace_btrfs_trigger_flush(fs_info,
5203 queue_work(system_unbound_wq,
5204 &fs_info->async_reclaim_work);
5207 list_add_tail(&ticket.list,
5208 &space_info->priority_tickets);
5210 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5213 * We will do the space reservation dance during log replay,
5214 * which means we won't have fs_info->fs_root set, so don't do
5215 * the async reclaim as we will panic.
5217 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5218 need_do_async_reclaim(fs_info, space_info,
5219 used, system_chunk) &&
5220 !work_busy(&fs_info->async_reclaim_work)) {
5221 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5222 orig_bytes, flush, "preempt");
5223 queue_work(system_unbound_wq,
5224 &fs_info->async_reclaim_work);
5227 spin_unlock(&space_info->lock);
5228 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5231 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5232 return wait_reserve_ticket(fs_info, space_info, &ticket,
5236 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5237 spin_lock(&space_info->lock);
5239 if (ticket.bytes < orig_bytes) {
5240 u64 num_bytes = orig_bytes - ticket.bytes;
5241 update_bytes_may_use(space_info, -num_bytes);
5242 trace_btrfs_space_reservation(fs_info, "space_info",
5247 list_del_init(&ticket.list);
5250 spin_unlock(&space_info->lock);
5251 ASSERT(list_empty(&ticket.list));
5256 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5257 * @root - the root we're allocating for
5258 * @block_rsv - the block_rsv we're allocating for
5259 * @orig_bytes - the number of bytes we want
5260 * @flush - whether or not we can flush to make our reservation
5262 * This will reserve orgi_bytes number of bytes from the space info associated
5263 * with the block_rsv. If there is not enough space it will make an attempt to
5264 * flush out space to make room. It will do this by flushing delalloc if
5265 * possible or committing the transaction. If flush is 0 then no attempts to
5266 * regain reservations will be made and this will fail if there is not enough
5269 static int reserve_metadata_bytes(struct btrfs_root *root,
5270 struct btrfs_block_rsv *block_rsv,
5272 enum btrfs_reserve_flush_enum flush)
5274 struct btrfs_fs_info *fs_info = root->fs_info;
5275 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5277 bool system_chunk = (root == fs_info->chunk_root);
5279 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5280 orig_bytes, flush, system_chunk);
5281 if (ret == -ENOSPC &&
5282 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5283 if (block_rsv != global_rsv &&
5284 !block_rsv_use_bytes(global_rsv, orig_bytes))
5287 if (ret == -ENOSPC) {
5288 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5289 block_rsv->space_info->flags,
5292 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5293 dump_space_info(fs_info, block_rsv->space_info,
5299 static struct btrfs_block_rsv *get_block_rsv(
5300 const struct btrfs_trans_handle *trans,
5301 const struct btrfs_root *root)
5303 struct btrfs_fs_info *fs_info = root->fs_info;
5304 struct btrfs_block_rsv *block_rsv = NULL;
5306 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5307 (root == fs_info->csum_root && trans->adding_csums) ||
5308 (root == fs_info->uuid_root))
5309 block_rsv = trans->block_rsv;
5312 block_rsv = root->block_rsv;
5315 block_rsv = &fs_info->empty_block_rsv;
5320 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5324 spin_lock(&block_rsv->lock);
5325 if (block_rsv->reserved >= num_bytes) {
5326 block_rsv->reserved -= num_bytes;
5327 if (block_rsv->reserved < block_rsv->size)
5328 block_rsv->full = 0;
5331 spin_unlock(&block_rsv->lock);
5335 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5336 u64 num_bytes, bool update_size)
5338 spin_lock(&block_rsv->lock);
5339 block_rsv->reserved += num_bytes;
5341 block_rsv->size += num_bytes;
5342 else if (block_rsv->reserved >= block_rsv->size)
5343 block_rsv->full = 1;
5344 spin_unlock(&block_rsv->lock);
5347 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5348 struct btrfs_block_rsv *dest, u64 num_bytes,
5351 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5354 if (global_rsv->space_info != dest->space_info)
5357 spin_lock(&global_rsv->lock);
5358 min_bytes = div_factor(global_rsv->size, min_factor);
5359 if (global_rsv->reserved < min_bytes + num_bytes) {
5360 spin_unlock(&global_rsv->lock);
5363 global_rsv->reserved -= num_bytes;
5364 if (global_rsv->reserved < global_rsv->size)
5365 global_rsv->full = 0;
5366 spin_unlock(&global_rsv->lock);
5368 block_rsv_add_bytes(dest, num_bytes, true);
5373 * This is for space we already have accounted in space_info->bytes_may_use, so
5374 * basically when we're returning space from block_rsv's.
5376 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5377 struct btrfs_space_info *space_info,
5380 struct reserve_ticket *ticket;
5381 struct list_head *head;
5383 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5384 bool check_overcommit = false;
5386 spin_lock(&space_info->lock);
5387 head = &space_info->priority_tickets;
5390 * If we are over our limit then we need to check and see if we can
5391 * overcommit, and if we can't then we just need to free up our space
5392 * and not satisfy any requests.
5394 used = btrfs_space_info_used(space_info, true);
5395 if (used - num_bytes >= space_info->total_bytes)
5396 check_overcommit = true;
5398 while (!list_empty(head) && num_bytes) {
5399 ticket = list_first_entry(head, struct reserve_ticket,
5402 * We use 0 bytes because this space is already reserved, so
5403 * adding the ticket space would be a double count.
5405 if (check_overcommit &&
5406 !can_overcommit(fs_info, space_info, 0, flush, false))
5408 if (num_bytes >= ticket->bytes) {
5409 list_del_init(&ticket->list);
5410 num_bytes -= ticket->bytes;
5412 space_info->tickets_id++;
5413 wake_up(&ticket->wait);
5415 ticket->bytes -= num_bytes;
5420 if (num_bytes && head == &space_info->priority_tickets) {
5421 head = &space_info->tickets;
5422 flush = BTRFS_RESERVE_FLUSH_ALL;
5425 update_bytes_may_use(space_info, -num_bytes);
5426 trace_btrfs_space_reservation(fs_info, "space_info",
5427 space_info->flags, num_bytes, 0);
5428 spin_unlock(&space_info->lock);
5432 * This is for newly allocated space that isn't accounted in
5433 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5434 * we use this helper.
5436 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5437 struct btrfs_space_info *space_info,
5440 struct reserve_ticket *ticket;
5441 struct list_head *head = &space_info->priority_tickets;
5444 while (!list_empty(head) && num_bytes) {
5445 ticket = list_first_entry(head, struct reserve_ticket,
5447 if (num_bytes >= ticket->bytes) {
5448 trace_btrfs_space_reservation(fs_info, "space_info",
5451 list_del_init(&ticket->list);
5452 num_bytes -= ticket->bytes;
5453 update_bytes_may_use(space_info, ticket->bytes);
5455 space_info->tickets_id++;
5456 wake_up(&ticket->wait);
5458 trace_btrfs_space_reservation(fs_info, "space_info",
5461 update_bytes_may_use(space_info, num_bytes);
5462 ticket->bytes -= num_bytes;
5467 if (num_bytes && head == &space_info->priority_tickets) {
5468 head = &space_info->tickets;
5473 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5474 struct btrfs_block_rsv *block_rsv,
5475 struct btrfs_block_rsv *dest, u64 num_bytes,
5476 u64 *qgroup_to_release_ret)
5478 struct btrfs_space_info *space_info = block_rsv->space_info;
5479 u64 qgroup_to_release = 0;
5482 spin_lock(&block_rsv->lock);
5483 if (num_bytes == (u64)-1) {
5484 num_bytes = block_rsv->size;
5485 qgroup_to_release = block_rsv->qgroup_rsv_size;
5487 block_rsv->size -= num_bytes;
5488 if (block_rsv->reserved >= block_rsv->size) {
5489 num_bytes = block_rsv->reserved - block_rsv->size;
5490 block_rsv->reserved = block_rsv->size;
5491 block_rsv->full = 1;
5495 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5496 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5497 block_rsv->qgroup_rsv_size;
5498 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5500 qgroup_to_release = 0;
5502 spin_unlock(&block_rsv->lock);
5505 if (num_bytes > 0) {
5507 spin_lock(&dest->lock);
5511 bytes_to_add = dest->size - dest->reserved;
5512 bytes_to_add = min(num_bytes, bytes_to_add);
5513 dest->reserved += bytes_to_add;
5514 if (dest->reserved >= dest->size)
5516 num_bytes -= bytes_to_add;
5518 spin_unlock(&dest->lock);
5521 space_info_add_old_bytes(fs_info, space_info,
5524 if (qgroup_to_release_ret)
5525 *qgroup_to_release_ret = qgroup_to_release;
5529 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5530 struct btrfs_block_rsv *dst, u64 num_bytes,
5535 ret = block_rsv_use_bytes(src, num_bytes);
5539 block_rsv_add_bytes(dst, num_bytes, update_size);
5543 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5545 memset(rsv, 0, sizeof(*rsv));
5546 spin_lock_init(&rsv->lock);
5550 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5551 struct btrfs_block_rsv *rsv,
5552 unsigned short type)
5554 btrfs_init_block_rsv(rsv, type);
5555 rsv->space_info = __find_space_info(fs_info,
5556 BTRFS_BLOCK_GROUP_METADATA);
5559 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5560 unsigned short type)
5562 struct btrfs_block_rsv *block_rsv;
5564 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5568 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5572 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5573 struct btrfs_block_rsv *rsv)
5577 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5581 int btrfs_block_rsv_add(struct btrfs_root *root,
5582 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5583 enum btrfs_reserve_flush_enum flush)
5590 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5592 block_rsv_add_bytes(block_rsv, num_bytes, true);
5597 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5605 spin_lock(&block_rsv->lock);
5606 num_bytes = div_factor(block_rsv->size, min_factor);
5607 if (block_rsv->reserved >= num_bytes)
5609 spin_unlock(&block_rsv->lock);
5614 int btrfs_block_rsv_refill(struct btrfs_root *root,
5615 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5616 enum btrfs_reserve_flush_enum flush)
5624 spin_lock(&block_rsv->lock);
5625 num_bytes = min_reserved;
5626 if (block_rsv->reserved >= num_bytes)
5629 num_bytes -= block_rsv->reserved;
5630 spin_unlock(&block_rsv->lock);
5635 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5637 block_rsv_add_bytes(block_rsv, num_bytes, false);
5645 * btrfs_inode_rsv_refill - refill the inode block rsv.
5646 * @inode - the inode we are refilling.
5647 * @flush - the flusing restriction.
5649 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5650 * block_rsv->size as the minimum size. We'll either refill the missing amount
5651 * or return if we already have enough space. This will also handle the resreve
5652 * tracepoint for the reserved amount.
5654 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5655 enum btrfs_reserve_flush_enum flush)
5657 struct btrfs_root *root = inode->root;
5658 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5660 u64 qgroup_num_bytes = 0;
5663 spin_lock(&block_rsv->lock);
5664 if (block_rsv->reserved < block_rsv->size)
5665 num_bytes = block_rsv->size - block_rsv->reserved;
5666 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5667 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5668 block_rsv->qgroup_rsv_reserved;
5669 spin_unlock(&block_rsv->lock);
5674 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5677 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5679 block_rsv_add_bytes(block_rsv, num_bytes, false);
5680 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5681 btrfs_ino(inode), num_bytes, 1);
5683 /* Don't forget to increase qgroup_rsv_reserved */
5684 spin_lock(&block_rsv->lock);
5685 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5686 spin_unlock(&block_rsv->lock);
5688 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5693 * btrfs_inode_rsv_release - release any excessive reservation.
5694 * @inode - the inode we need to release from.
5695 * @qgroup_free - free or convert qgroup meta.
5696 * Unlike normal operation, qgroup meta reservation needs to know if we are
5697 * freeing qgroup reservation or just converting it into per-trans. Normally
5698 * @qgroup_free is true for error handling, and false for normal release.
5700 * This is the same as btrfs_block_rsv_release, except that it handles the
5701 * tracepoint for the reservation.
5703 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5705 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5706 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5707 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5709 u64 qgroup_to_release = 0;
5712 * Since we statically set the block_rsv->size we just want to say we
5713 * are releasing 0 bytes, and then we'll just get the reservation over
5716 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5717 &qgroup_to_release);
5719 trace_btrfs_space_reservation(fs_info, "delalloc",
5720 btrfs_ino(inode), released, 0);
5722 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5724 btrfs_qgroup_convert_reserved_meta(inode->root,
5728 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5729 struct btrfs_block_rsv *block_rsv,
5732 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5734 if (global_rsv == block_rsv ||
5735 block_rsv->space_info != global_rsv->space_info)
5737 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5740 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5742 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5743 struct btrfs_space_info *sinfo = block_rsv->space_info;
5747 * The global block rsv is based on the size of the extent tree, the
5748 * checksum tree and the root tree. If the fs is empty we want to set
5749 * it to a minimal amount for safety.
5751 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5752 btrfs_root_used(&fs_info->csum_root->root_item) +
5753 btrfs_root_used(&fs_info->tree_root->root_item);
5754 num_bytes = max_t(u64, num_bytes, SZ_16M);
5756 spin_lock(&sinfo->lock);
5757 spin_lock(&block_rsv->lock);
5759 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5761 if (block_rsv->reserved < block_rsv->size) {
5762 num_bytes = btrfs_space_info_used(sinfo, true);
5763 if (sinfo->total_bytes > num_bytes) {
5764 num_bytes = sinfo->total_bytes - num_bytes;
5765 num_bytes = min(num_bytes,
5766 block_rsv->size - block_rsv->reserved);
5767 block_rsv->reserved += num_bytes;
5768 update_bytes_may_use(sinfo, num_bytes);
5769 trace_btrfs_space_reservation(fs_info, "space_info",
5770 sinfo->flags, num_bytes,
5773 } else if (block_rsv->reserved > block_rsv->size) {
5774 num_bytes = block_rsv->reserved - block_rsv->size;
5775 update_bytes_may_use(sinfo, -num_bytes);
5776 trace_btrfs_space_reservation(fs_info, "space_info",
5777 sinfo->flags, num_bytes, 0);
5778 block_rsv->reserved = block_rsv->size;
5781 if (block_rsv->reserved == block_rsv->size)
5782 block_rsv->full = 1;
5784 block_rsv->full = 0;
5786 spin_unlock(&block_rsv->lock);
5787 spin_unlock(&sinfo->lock);
5790 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5792 struct btrfs_space_info *space_info;
5794 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5795 fs_info->chunk_block_rsv.space_info = space_info;
5797 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5798 fs_info->global_block_rsv.space_info = space_info;
5799 fs_info->trans_block_rsv.space_info = space_info;
5800 fs_info->empty_block_rsv.space_info = space_info;
5801 fs_info->delayed_block_rsv.space_info = space_info;
5803 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5804 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5805 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5806 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5807 if (fs_info->quota_root)
5808 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5809 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5811 update_global_block_rsv(fs_info);
5814 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5816 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5818 WARN_ON(fs_info->trans_block_rsv.size > 0);
5819 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5820 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5821 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5822 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5823 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5828 * To be called after all the new block groups attached to the transaction
5829 * handle have been created (btrfs_create_pending_block_groups()).
5831 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5833 struct btrfs_fs_info *fs_info = trans->fs_info;
5835 if (!trans->chunk_bytes_reserved)
5838 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5840 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5841 trans->chunk_bytes_reserved, NULL);
5842 trans->chunk_bytes_reserved = 0;
5846 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5847 * root: the root of the parent directory
5848 * rsv: block reservation
5849 * items: the number of items that we need do reservation
5850 * use_global_rsv: allow fallback to the global block reservation
5852 * This function is used to reserve the space for snapshot/subvolume
5853 * creation and deletion. Those operations are different with the
5854 * common file/directory operations, they change two fs/file trees
5855 * and root tree, the number of items that the qgroup reserves is
5856 * different with the free space reservation. So we can not use
5857 * the space reservation mechanism in start_transaction().
5859 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5860 struct btrfs_block_rsv *rsv, int items,
5861 bool use_global_rsv)
5863 u64 qgroup_num_bytes = 0;
5866 struct btrfs_fs_info *fs_info = root->fs_info;
5867 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5869 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5870 /* One for parent inode, two for dir entries */
5871 qgroup_num_bytes = 3 * fs_info->nodesize;
5872 ret = btrfs_qgroup_reserve_meta_prealloc(root,
5873 qgroup_num_bytes, true);
5878 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5879 rsv->space_info = __find_space_info(fs_info,
5880 BTRFS_BLOCK_GROUP_METADATA);
5881 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5882 BTRFS_RESERVE_FLUSH_ALL);
5884 if (ret == -ENOSPC && use_global_rsv)
5885 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
5887 if (ret && qgroup_num_bytes)
5888 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5893 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5894 struct btrfs_block_rsv *rsv)
5896 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5899 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5900 struct btrfs_inode *inode)
5902 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5903 u64 reserve_size = 0;
5904 u64 qgroup_rsv_size = 0;
5906 unsigned outstanding_extents;
5908 lockdep_assert_held(&inode->lock);
5909 outstanding_extents = inode->outstanding_extents;
5910 if (outstanding_extents)
5911 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
5912 outstanding_extents + 1);
5913 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
5915 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
5918 * For qgroup rsv, the calculation is very simple:
5919 * account one nodesize for each outstanding extent
5921 * This is overestimating in most cases.
5923 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
5925 spin_lock(&block_rsv->lock);
5926 block_rsv->size = reserve_size;
5927 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
5928 spin_unlock(&block_rsv->lock);
5931 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5933 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5934 unsigned nr_extents;
5935 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5937 bool delalloc_lock = true;
5939 /* If we are a free space inode we need to not flush since we will be in
5940 * the middle of a transaction commit. We also don't need the delalloc
5941 * mutex since we won't race with anybody. We need this mostly to make
5942 * lockdep shut its filthy mouth.
5944 * If we have a transaction open (can happen if we call truncate_block
5945 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5947 if (btrfs_is_free_space_inode(inode)) {
5948 flush = BTRFS_RESERVE_NO_FLUSH;
5949 delalloc_lock = false;
5951 if (current->journal_info)
5952 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5954 if (btrfs_transaction_in_commit(fs_info))
5955 schedule_timeout(1);
5959 mutex_lock(&inode->delalloc_mutex);
5961 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5963 /* Add our new extents and calculate the new rsv size. */
5964 spin_lock(&inode->lock);
5965 nr_extents = count_max_extents(num_bytes);
5966 btrfs_mod_outstanding_extents(inode, nr_extents);
5967 inode->csum_bytes += num_bytes;
5968 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5969 spin_unlock(&inode->lock);
5971 ret = btrfs_inode_rsv_refill(inode, flush);
5976 mutex_unlock(&inode->delalloc_mutex);
5980 spin_lock(&inode->lock);
5981 nr_extents = count_max_extents(num_bytes);
5982 btrfs_mod_outstanding_extents(inode, -nr_extents);
5983 inode->csum_bytes -= num_bytes;
5984 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5985 spin_unlock(&inode->lock);
5987 btrfs_inode_rsv_release(inode, true);
5989 mutex_unlock(&inode->delalloc_mutex);
5994 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5995 * @inode: the inode to release the reservation for.
5996 * @num_bytes: the number of bytes we are releasing.
5997 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
5999 * This will release the metadata reservation for an inode. This can be called
6000 * once we complete IO for a given set of bytes to release their metadata
6001 * reservations, or on error for the same reason.
6003 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6006 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6008 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6009 spin_lock(&inode->lock);
6010 inode->csum_bytes -= num_bytes;
6011 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6012 spin_unlock(&inode->lock);
6014 if (btrfs_is_testing(fs_info))
6017 btrfs_inode_rsv_release(inode, qgroup_free);
6021 * btrfs_delalloc_release_extents - release our outstanding_extents
6022 * @inode: the inode to balance the reservation for.
6023 * @num_bytes: the number of bytes we originally reserved with
6024 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6026 * When we reserve space we increase outstanding_extents for the extents we may
6027 * add. Once we've set the range as delalloc or created our ordered extents we
6028 * have outstanding_extents to track the real usage, so we use this to free our
6029 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6030 * with btrfs_delalloc_reserve_metadata.
6032 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6035 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6036 unsigned num_extents;
6038 spin_lock(&inode->lock);
6039 num_extents = count_max_extents(num_bytes);
6040 btrfs_mod_outstanding_extents(inode, -num_extents);
6041 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6042 spin_unlock(&inode->lock);
6044 if (btrfs_is_testing(fs_info))
6047 btrfs_inode_rsv_release(inode, qgroup_free);
6051 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6053 * @inode: inode we're writing to
6054 * @start: start range we are writing to
6055 * @len: how long the range we are writing to
6056 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6057 * current reservation.
6059 * This will do the following things
6061 * o reserve space in data space info for num bytes
6062 * and reserve precious corresponding qgroup space
6063 * (Done in check_data_free_space)
6065 * o reserve space for metadata space, based on the number of outstanding
6066 * extents and how much csums will be needed
6067 * also reserve metadata space in a per root over-reserve method.
6068 * o add to the inodes->delalloc_bytes
6069 * o add it to the fs_info's delalloc inodes list.
6070 * (Above 3 all done in delalloc_reserve_metadata)
6072 * Return 0 for success
6073 * Return <0 for error(-ENOSPC or -EQUOT)
6075 int btrfs_delalloc_reserve_space(struct inode *inode,
6076 struct extent_changeset **reserved, u64 start, u64 len)
6080 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6083 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6085 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6090 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6091 * @inode: inode we're releasing space for
6092 * @start: start position of the space already reserved
6093 * @len: the len of the space already reserved
6094 * @release_bytes: the len of the space we consumed or didn't use
6096 * This function will release the metadata space that was not used and will
6097 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6098 * list if there are no delalloc bytes left.
6099 * Also it will handle the qgroup reserved space.
6101 void btrfs_delalloc_release_space(struct inode *inode,
6102 struct extent_changeset *reserved,
6103 u64 start, u64 len, bool qgroup_free)
6105 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6106 btrfs_free_reserved_data_space(inode, reserved, start, len);
6109 static int update_block_group(struct btrfs_trans_handle *trans,
6110 struct btrfs_fs_info *info, u64 bytenr,
6111 u64 num_bytes, int alloc)
6113 struct btrfs_block_group_cache *cache = NULL;
6114 u64 total = num_bytes;
6119 /* block accounting for super block */
6120 spin_lock(&info->delalloc_root_lock);
6121 old_val = btrfs_super_bytes_used(info->super_copy);
6123 old_val += num_bytes;
6125 old_val -= num_bytes;
6126 btrfs_set_super_bytes_used(info->super_copy, old_val);
6127 spin_unlock(&info->delalloc_root_lock);
6130 cache = btrfs_lookup_block_group(info, bytenr);
6133 factor = btrfs_bg_type_to_factor(cache->flags);
6136 * If this block group has free space cache written out, we
6137 * need to make sure to load it if we are removing space. This
6138 * is because we need the unpinning stage to actually add the
6139 * space back to the block group, otherwise we will leak space.
6141 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6142 cache_block_group(cache, 1);
6144 byte_in_group = bytenr - cache->key.objectid;
6145 WARN_ON(byte_in_group > cache->key.offset);
6147 spin_lock(&cache->space_info->lock);
6148 spin_lock(&cache->lock);
6150 if (btrfs_test_opt(info, SPACE_CACHE) &&
6151 cache->disk_cache_state < BTRFS_DC_CLEAR)
6152 cache->disk_cache_state = BTRFS_DC_CLEAR;
6154 old_val = btrfs_block_group_used(&cache->item);
6155 num_bytes = min(total, cache->key.offset - byte_in_group);
6157 old_val += num_bytes;
6158 btrfs_set_block_group_used(&cache->item, old_val);
6159 cache->reserved -= num_bytes;
6160 cache->space_info->bytes_reserved -= num_bytes;
6161 cache->space_info->bytes_used += num_bytes;
6162 cache->space_info->disk_used += num_bytes * factor;
6163 spin_unlock(&cache->lock);
6164 spin_unlock(&cache->space_info->lock);
6166 old_val -= num_bytes;
6167 btrfs_set_block_group_used(&cache->item, old_val);
6168 cache->pinned += num_bytes;
6169 update_bytes_pinned(cache->space_info, num_bytes);
6170 cache->space_info->bytes_used -= num_bytes;
6171 cache->space_info->disk_used -= num_bytes * factor;
6172 spin_unlock(&cache->lock);
6173 spin_unlock(&cache->space_info->lock);
6175 trace_btrfs_space_reservation(info, "pinned",
6176 cache->space_info->flags,
6178 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6180 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6181 set_extent_dirty(info->pinned_extents,
6182 bytenr, bytenr + num_bytes - 1,
6183 GFP_NOFS | __GFP_NOFAIL);
6186 spin_lock(&trans->transaction->dirty_bgs_lock);
6187 if (list_empty(&cache->dirty_list)) {
6188 list_add_tail(&cache->dirty_list,
6189 &trans->transaction->dirty_bgs);
6190 trans->transaction->num_dirty_bgs++;
6191 btrfs_get_block_group(cache);
6193 spin_unlock(&trans->transaction->dirty_bgs_lock);
6196 * No longer have used bytes in this block group, queue it for
6197 * deletion. We do this after adding the block group to the
6198 * dirty list to avoid races between cleaner kthread and space
6201 if (!alloc && old_val == 0)
6202 btrfs_mark_bg_unused(cache);
6204 btrfs_put_block_group(cache);
6206 bytenr += num_bytes;
6211 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6213 struct btrfs_block_group_cache *cache;
6216 spin_lock(&fs_info->block_group_cache_lock);
6217 bytenr = fs_info->first_logical_byte;
6218 spin_unlock(&fs_info->block_group_cache_lock);
6220 if (bytenr < (u64)-1)
6223 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6227 bytenr = cache->key.objectid;
6228 btrfs_put_block_group(cache);
6233 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6234 struct btrfs_block_group_cache *cache,
6235 u64 bytenr, u64 num_bytes, int reserved)
6237 spin_lock(&cache->space_info->lock);
6238 spin_lock(&cache->lock);
6239 cache->pinned += num_bytes;
6240 update_bytes_pinned(cache->space_info, num_bytes);
6242 cache->reserved -= num_bytes;
6243 cache->space_info->bytes_reserved -= num_bytes;
6245 spin_unlock(&cache->lock);
6246 spin_unlock(&cache->space_info->lock);
6248 trace_btrfs_space_reservation(fs_info, "pinned",
6249 cache->space_info->flags, num_bytes, 1);
6250 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6251 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6252 set_extent_dirty(fs_info->pinned_extents, bytenr,
6253 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6258 * this function must be called within transaction
6260 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6261 u64 bytenr, u64 num_bytes, int reserved)
6263 struct btrfs_block_group_cache *cache;
6265 cache = btrfs_lookup_block_group(fs_info, bytenr);
6266 BUG_ON(!cache); /* Logic error */
6268 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6270 btrfs_put_block_group(cache);
6275 * this function must be called within transaction
6277 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6278 u64 bytenr, u64 num_bytes)
6280 struct btrfs_block_group_cache *cache;
6283 cache = btrfs_lookup_block_group(fs_info, bytenr);
6288 * pull in the free space cache (if any) so that our pin
6289 * removes the free space from the cache. We have load_only set
6290 * to one because the slow code to read in the free extents does check
6291 * the pinned extents.
6293 cache_block_group(cache, 1);
6295 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6297 /* remove us from the free space cache (if we're there at all) */
6298 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6299 btrfs_put_block_group(cache);
6303 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6304 u64 start, u64 num_bytes)
6307 struct btrfs_block_group_cache *block_group;
6308 struct btrfs_caching_control *caching_ctl;
6310 block_group = btrfs_lookup_block_group(fs_info, start);
6314 cache_block_group(block_group, 0);
6315 caching_ctl = get_caching_control(block_group);
6319 BUG_ON(!block_group_cache_done(block_group));
6320 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6322 mutex_lock(&caching_ctl->mutex);
6324 if (start >= caching_ctl->progress) {
6325 ret = add_excluded_extent(fs_info, start, num_bytes);
6326 } else if (start + num_bytes <= caching_ctl->progress) {
6327 ret = btrfs_remove_free_space(block_group,
6330 num_bytes = caching_ctl->progress - start;
6331 ret = btrfs_remove_free_space(block_group,
6336 num_bytes = (start + num_bytes) -
6337 caching_ctl->progress;
6338 start = caching_ctl->progress;
6339 ret = add_excluded_extent(fs_info, start, num_bytes);
6342 mutex_unlock(&caching_ctl->mutex);
6343 put_caching_control(caching_ctl);
6345 btrfs_put_block_group(block_group);
6349 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6350 struct extent_buffer *eb)
6352 struct btrfs_file_extent_item *item;
6353 struct btrfs_key key;
6358 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6361 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6362 btrfs_item_key_to_cpu(eb, &key, i);
6363 if (key.type != BTRFS_EXTENT_DATA_KEY)
6365 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6366 found_type = btrfs_file_extent_type(eb, item);
6367 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6369 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6371 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6372 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6373 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6382 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6384 atomic_inc(&bg->reservations);
6387 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6390 struct btrfs_block_group_cache *bg;
6392 bg = btrfs_lookup_block_group(fs_info, start);
6394 if (atomic_dec_and_test(&bg->reservations))
6395 wake_up_var(&bg->reservations);
6396 btrfs_put_block_group(bg);
6399 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6401 struct btrfs_space_info *space_info = bg->space_info;
6405 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6409 * Our block group is read only but before we set it to read only,
6410 * some task might have had allocated an extent from it already, but it
6411 * has not yet created a respective ordered extent (and added it to a
6412 * root's list of ordered extents).
6413 * Therefore wait for any task currently allocating extents, since the
6414 * block group's reservations counter is incremented while a read lock
6415 * on the groups' semaphore is held and decremented after releasing
6416 * the read access on that semaphore and creating the ordered extent.
6418 down_write(&space_info->groups_sem);
6419 up_write(&space_info->groups_sem);
6421 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6425 * btrfs_add_reserved_bytes - update the block_group and space info counters
6426 * @cache: The cache we are manipulating
6427 * @ram_bytes: The number of bytes of file content, and will be same to
6428 * @num_bytes except for the compress path.
6429 * @num_bytes: The number of bytes in question
6430 * @delalloc: The blocks are allocated for the delalloc write
6432 * This is called by the allocator when it reserves space. If this is a
6433 * reservation and the block group has become read only we cannot make the
6434 * reservation and return -EAGAIN, otherwise this function always succeeds.
6436 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6437 u64 ram_bytes, u64 num_bytes, int delalloc)
6439 struct btrfs_space_info *space_info = cache->space_info;
6442 spin_lock(&space_info->lock);
6443 spin_lock(&cache->lock);
6447 cache->reserved += num_bytes;
6448 space_info->bytes_reserved += num_bytes;
6449 update_bytes_may_use(space_info, -ram_bytes);
6451 cache->delalloc_bytes += num_bytes;
6453 spin_unlock(&cache->lock);
6454 spin_unlock(&space_info->lock);
6459 * btrfs_free_reserved_bytes - update the block_group and space info counters
6460 * @cache: The cache we are manipulating
6461 * @num_bytes: The number of bytes in question
6462 * @delalloc: The blocks are allocated for the delalloc write
6464 * This is called by somebody who is freeing space that was never actually used
6465 * on disk. For example if you reserve some space for a new leaf in transaction
6466 * A and before transaction A commits you free that leaf, you call this with
6467 * reserve set to 0 in order to clear the reservation.
6470 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6471 u64 num_bytes, int delalloc)
6473 struct btrfs_space_info *space_info = cache->space_info;
6475 spin_lock(&space_info->lock);
6476 spin_lock(&cache->lock);
6478 space_info->bytes_readonly += num_bytes;
6479 cache->reserved -= num_bytes;
6480 space_info->bytes_reserved -= num_bytes;
6481 space_info->max_extent_size = 0;
6484 cache->delalloc_bytes -= num_bytes;
6485 spin_unlock(&cache->lock);
6486 spin_unlock(&space_info->lock);
6488 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6490 struct btrfs_caching_control *next;
6491 struct btrfs_caching_control *caching_ctl;
6492 struct btrfs_block_group_cache *cache;
6494 down_write(&fs_info->commit_root_sem);
6496 list_for_each_entry_safe(caching_ctl, next,
6497 &fs_info->caching_block_groups, list) {
6498 cache = caching_ctl->block_group;
6499 if (block_group_cache_done(cache)) {
6500 cache->last_byte_to_unpin = (u64)-1;
6501 list_del_init(&caching_ctl->list);
6502 put_caching_control(caching_ctl);
6504 cache->last_byte_to_unpin = caching_ctl->progress;
6508 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6509 fs_info->pinned_extents = &fs_info->freed_extents[1];
6511 fs_info->pinned_extents = &fs_info->freed_extents[0];
6513 up_write(&fs_info->commit_root_sem);
6515 update_global_block_rsv(fs_info);
6519 * Returns the free cluster for the given space info and sets empty_cluster to
6520 * what it should be based on the mount options.
6522 static struct btrfs_free_cluster *
6523 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6524 struct btrfs_space_info *space_info, u64 *empty_cluster)
6526 struct btrfs_free_cluster *ret = NULL;
6529 if (btrfs_mixed_space_info(space_info))
6532 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6533 ret = &fs_info->meta_alloc_cluster;
6534 if (btrfs_test_opt(fs_info, SSD))
6535 *empty_cluster = SZ_2M;
6537 *empty_cluster = SZ_64K;
6538 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6539 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6540 *empty_cluster = SZ_2M;
6541 ret = &fs_info->data_alloc_cluster;
6547 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6549 const bool return_free_space)
6551 struct btrfs_block_group_cache *cache = NULL;
6552 struct btrfs_space_info *space_info;
6553 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6554 struct btrfs_free_cluster *cluster = NULL;
6556 u64 total_unpinned = 0;
6557 u64 empty_cluster = 0;
6560 while (start <= end) {
6563 start >= cache->key.objectid + cache->key.offset) {
6565 btrfs_put_block_group(cache);
6567 cache = btrfs_lookup_block_group(fs_info, start);
6568 BUG_ON(!cache); /* Logic error */
6570 cluster = fetch_cluster_info(fs_info,
6573 empty_cluster <<= 1;
6576 len = cache->key.objectid + cache->key.offset - start;
6577 len = min(len, end + 1 - start);
6579 if (start < cache->last_byte_to_unpin) {
6580 len = min(len, cache->last_byte_to_unpin - start);
6581 if (return_free_space)
6582 btrfs_add_free_space(cache, start, len);
6586 total_unpinned += len;
6587 space_info = cache->space_info;
6590 * If this space cluster has been marked as fragmented and we've
6591 * unpinned enough in this block group to potentially allow a
6592 * cluster to be created inside of it go ahead and clear the
6595 if (cluster && cluster->fragmented &&
6596 total_unpinned > empty_cluster) {
6597 spin_lock(&cluster->lock);
6598 cluster->fragmented = 0;
6599 spin_unlock(&cluster->lock);
6602 spin_lock(&space_info->lock);
6603 spin_lock(&cache->lock);
6604 cache->pinned -= len;
6605 update_bytes_pinned(space_info, -len);
6607 trace_btrfs_space_reservation(fs_info, "pinned",
6608 space_info->flags, len, 0);
6609 space_info->max_extent_size = 0;
6610 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6611 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6613 space_info->bytes_readonly += len;
6616 spin_unlock(&cache->lock);
6617 if (!readonly && return_free_space &&
6618 global_rsv->space_info == space_info) {
6621 spin_lock(&global_rsv->lock);
6622 if (!global_rsv->full) {
6623 to_add = min(len, global_rsv->size -
6624 global_rsv->reserved);
6625 global_rsv->reserved += to_add;
6626 update_bytes_may_use(space_info, to_add);
6627 if (global_rsv->reserved >= global_rsv->size)
6628 global_rsv->full = 1;
6629 trace_btrfs_space_reservation(fs_info,
6635 spin_unlock(&global_rsv->lock);
6636 /* Add to any tickets we may have */
6638 space_info_add_new_bytes(fs_info, space_info,
6641 spin_unlock(&space_info->lock);
6645 btrfs_put_block_group(cache);
6649 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6651 struct btrfs_fs_info *fs_info = trans->fs_info;
6652 struct btrfs_block_group_cache *block_group, *tmp;
6653 struct list_head *deleted_bgs;
6654 struct extent_io_tree *unpin;
6659 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6660 unpin = &fs_info->freed_extents[1];
6662 unpin = &fs_info->freed_extents[0];
6664 while (!trans->aborted) {
6665 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6666 ret = find_first_extent_bit(unpin, 0, &start, &end,
6667 EXTENT_DIRTY, NULL);
6669 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6673 if (btrfs_test_opt(fs_info, DISCARD))
6674 ret = btrfs_discard_extent(fs_info, start,
6675 end + 1 - start, NULL);
6677 clear_extent_dirty(unpin, start, end);
6678 unpin_extent_range(fs_info, start, end, true);
6679 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6684 * Transaction is finished. We don't need the lock anymore. We
6685 * do need to clean up the block groups in case of a transaction
6688 deleted_bgs = &trans->transaction->deleted_bgs;
6689 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6693 if (!trans->aborted)
6694 ret = btrfs_discard_extent(fs_info,
6695 block_group->key.objectid,
6696 block_group->key.offset,
6699 list_del_init(&block_group->bg_list);
6700 btrfs_put_block_group_trimming(block_group);
6701 btrfs_put_block_group(block_group);
6704 const char *errstr = btrfs_decode_error(ret);
6706 "discard failed while removing blockgroup: errno=%d %s",
6714 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6715 struct btrfs_delayed_ref_node *node, u64 parent,
6716 u64 root_objectid, u64 owner_objectid,
6717 u64 owner_offset, int refs_to_drop,
6718 struct btrfs_delayed_extent_op *extent_op)
6720 struct btrfs_fs_info *info = trans->fs_info;
6721 struct btrfs_key key;
6722 struct btrfs_path *path;
6723 struct btrfs_root *extent_root = info->extent_root;
6724 struct extent_buffer *leaf;
6725 struct btrfs_extent_item *ei;
6726 struct btrfs_extent_inline_ref *iref;
6729 int extent_slot = 0;
6730 int found_extent = 0;
6734 u64 bytenr = node->bytenr;
6735 u64 num_bytes = node->num_bytes;
6737 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6739 path = btrfs_alloc_path();
6743 path->reada = READA_FORWARD;
6744 path->leave_spinning = 1;
6746 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6747 BUG_ON(!is_data && refs_to_drop != 1);
6750 skinny_metadata = false;
6752 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6753 parent, root_objectid, owner_objectid,
6756 extent_slot = path->slots[0];
6757 while (extent_slot >= 0) {
6758 btrfs_item_key_to_cpu(path->nodes[0], &key,
6760 if (key.objectid != bytenr)
6762 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6763 key.offset == num_bytes) {
6767 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6768 key.offset == owner_objectid) {
6772 if (path->slots[0] - extent_slot > 5)
6777 if (!found_extent) {
6779 ret = remove_extent_backref(trans, path, NULL,
6781 is_data, &last_ref);
6783 btrfs_abort_transaction(trans, ret);
6786 btrfs_release_path(path);
6787 path->leave_spinning = 1;
6789 key.objectid = bytenr;
6790 key.type = BTRFS_EXTENT_ITEM_KEY;
6791 key.offset = num_bytes;
6793 if (!is_data && skinny_metadata) {
6794 key.type = BTRFS_METADATA_ITEM_KEY;
6795 key.offset = owner_objectid;
6798 ret = btrfs_search_slot(trans, extent_root,
6800 if (ret > 0 && skinny_metadata && path->slots[0]) {
6802 * Couldn't find our skinny metadata item,
6803 * see if we have ye olde extent item.
6806 btrfs_item_key_to_cpu(path->nodes[0], &key,
6808 if (key.objectid == bytenr &&
6809 key.type == BTRFS_EXTENT_ITEM_KEY &&
6810 key.offset == num_bytes)
6814 if (ret > 0 && skinny_metadata) {
6815 skinny_metadata = false;
6816 key.objectid = bytenr;
6817 key.type = BTRFS_EXTENT_ITEM_KEY;
6818 key.offset = num_bytes;
6819 btrfs_release_path(path);
6820 ret = btrfs_search_slot(trans, extent_root,
6826 "umm, got %d back from search, was looking for %llu",
6829 btrfs_print_leaf(path->nodes[0]);
6832 btrfs_abort_transaction(trans, ret);
6835 extent_slot = path->slots[0];
6837 } else if (WARN_ON(ret == -ENOENT)) {
6838 btrfs_print_leaf(path->nodes[0]);
6840 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6841 bytenr, parent, root_objectid, owner_objectid,
6843 btrfs_abort_transaction(trans, ret);
6846 btrfs_abort_transaction(trans, ret);
6850 leaf = path->nodes[0];
6851 item_size = btrfs_item_size_nr(leaf, extent_slot);
6852 if (unlikely(item_size < sizeof(*ei))) {
6854 btrfs_print_v0_err(info);
6855 btrfs_abort_transaction(trans, ret);
6858 ei = btrfs_item_ptr(leaf, extent_slot,
6859 struct btrfs_extent_item);
6860 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6861 key.type == BTRFS_EXTENT_ITEM_KEY) {
6862 struct btrfs_tree_block_info *bi;
6863 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6864 bi = (struct btrfs_tree_block_info *)(ei + 1);
6865 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6868 refs = btrfs_extent_refs(leaf, ei);
6869 if (refs < refs_to_drop) {
6871 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
6872 refs_to_drop, refs, bytenr);
6874 btrfs_abort_transaction(trans, ret);
6877 refs -= refs_to_drop;
6881 __run_delayed_extent_op(extent_op, leaf, ei);
6883 * In the case of inline back ref, reference count will
6884 * be updated by remove_extent_backref
6887 BUG_ON(!found_extent);
6889 btrfs_set_extent_refs(leaf, ei, refs);
6890 btrfs_mark_buffer_dirty(leaf);
6893 ret = remove_extent_backref(trans, path, iref,
6894 refs_to_drop, is_data,
6897 btrfs_abort_transaction(trans, ret);
6903 BUG_ON(is_data && refs_to_drop !=
6904 extent_data_ref_count(path, iref));
6906 BUG_ON(path->slots[0] != extent_slot);
6908 BUG_ON(path->slots[0] != extent_slot + 1);
6909 path->slots[0] = extent_slot;
6915 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6918 btrfs_abort_transaction(trans, ret);
6921 btrfs_release_path(path);
6924 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
6926 btrfs_abort_transaction(trans, ret);
6931 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
6933 btrfs_abort_transaction(trans, ret);
6937 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
6939 btrfs_abort_transaction(trans, ret);
6943 btrfs_release_path(path);
6946 btrfs_free_path(path);
6951 * when we free an block, it is possible (and likely) that we free the last
6952 * delayed ref for that extent as well. This searches the delayed ref tree for
6953 * a given extent, and if there are no other delayed refs to be processed, it
6954 * removes it from the tree.
6956 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6959 struct btrfs_delayed_ref_head *head;
6960 struct btrfs_delayed_ref_root *delayed_refs;
6963 delayed_refs = &trans->transaction->delayed_refs;
6964 spin_lock(&delayed_refs->lock);
6965 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
6967 goto out_delayed_unlock;
6969 spin_lock(&head->lock);
6970 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
6973 if (head->extent_op) {
6974 if (!head->must_insert_reserved)
6976 btrfs_free_delayed_extent_op(head->extent_op);
6977 head->extent_op = NULL;
6981 * waiting for the lock here would deadlock. If someone else has it
6982 * locked they are already in the process of dropping it anyway
6984 if (!mutex_trylock(&head->mutex))
6988 * at this point we have a head with no other entries. Go
6989 * ahead and process it.
6991 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
6992 RB_CLEAR_NODE(&head->href_node);
6993 atomic_dec(&delayed_refs->num_entries);
6996 * we don't take a ref on the node because we're removing it from the
6997 * tree, so we just steal the ref the tree was holding.
6999 delayed_refs->num_heads--;
7000 if (head->processing == 0)
7001 delayed_refs->num_heads_ready--;
7002 head->processing = 0;
7003 spin_unlock(&head->lock);
7004 spin_unlock(&delayed_refs->lock);
7006 BUG_ON(head->extent_op);
7007 if (head->must_insert_reserved)
7010 mutex_unlock(&head->mutex);
7011 btrfs_put_delayed_ref_head(head);
7014 spin_unlock(&head->lock);
7017 spin_unlock(&delayed_refs->lock);
7021 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7022 struct btrfs_root *root,
7023 struct extent_buffer *buf,
7024 u64 parent, int last_ref)
7026 struct btrfs_fs_info *fs_info = root->fs_info;
7030 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7031 int old_ref_mod, new_ref_mod;
7033 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7034 root->root_key.objectid,
7035 btrfs_header_level(buf), 0,
7036 BTRFS_DROP_DELAYED_REF);
7037 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7039 root->root_key.objectid,
7040 btrfs_header_level(buf),
7041 BTRFS_DROP_DELAYED_REF, NULL,
7042 &old_ref_mod, &new_ref_mod);
7043 BUG_ON(ret); /* -ENOMEM */
7044 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7047 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7048 struct btrfs_block_group_cache *cache;
7050 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7051 ret = check_ref_cleanup(trans, buf->start);
7057 cache = btrfs_lookup_block_group(fs_info, buf->start);
7059 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7060 pin_down_extent(fs_info, cache, buf->start,
7062 btrfs_put_block_group(cache);
7066 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7068 btrfs_add_free_space(cache, buf->start, buf->len);
7069 btrfs_free_reserved_bytes(cache, buf->len, 0);
7070 btrfs_put_block_group(cache);
7071 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7075 add_pinned_bytes(fs_info, buf->len, true,
7076 root->root_key.objectid);
7080 * Deleting the buffer, clear the corrupt flag since it doesn't
7083 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7087 /* Can return -ENOMEM */
7088 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7089 struct btrfs_root *root,
7090 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7091 u64 owner, u64 offset)
7093 struct btrfs_fs_info *fs_info = root->fs_info;
7094 int old_ref_mod, new_ref_mod;
7097 if (btrfs_is_testing(fs_info))
7100 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7101 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7102 root_objectid, owner, offset,
7103 BTRFS_DROP_DELAYED_REF);
7106 * tree log blocks never actually go into the extent allocation
7107 * tree, just update pinning info and exit early.
7109 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7110 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7111 /* unlocks the pinned mutex */
7112 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7113 old_ref_mod = new_ref_mod = 0;
7115 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7116 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7118 root_objectid, (int)owner,
7119 BTRFS_DROP_DELAYED_REF, NULL,
7120 &old_ref_mod, &new_ref_mod);
7122 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7124 root_objectid, owner, offset,
7125 0, BTRFS_DROP_DELAYED_REF,
7126 &old_ref_mod, &new_ref_mod);
7129 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7130 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7132 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7139 * when we wait for progress in the block group caching, its because
7140 * our allocation attempt failed at least once. So, we must sleep
7141 * and let some progress happen before we try again.
7143 * This function will sleep at least once waiting for new free space to
7144 * show up, and then it will check the block group free space numbers
7145 * for our min num_bytes. Another option is to have it go ahead
7146 * and look in the rbtree for a free extent of a given size, but this
7149 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7150 * any of the information in this block group.
7152 static noinline void
7153 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7156 struct btrfs_caching_control *caching_ctl;
7158 caching_ctl = get_caching_control(cache);
7162 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7163 (cache->free_space_ctl->free_space >= num_bytes));
7165 put_caching_control(caching_ctl);
7169 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7171 struct btrfs_caching_control *caching_ctl;
7174 caching_ctl = get_caching_control(cache);
7176 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7178 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7179 if (cache->cached == BTRFS_CACHE_ERROR)
7181 put_caching_control(caching_ctl);
7185 enum btrfs_loop_type {
7186 LOOP_CACHING_NOWAIT = 0,
7187 LOOP_CACHING_WAIT = 1,
7188 LOOP_ALLOC_CHUNK = 2,
7189 LOOP_NO_EMPTY_SIZE = 3,
7193 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7197 down_read(&cache->data_rwsem);
7201 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7204 btrfs_get_block_group(cache);
7206 down_read(&cache->data_rwsem);
7209 static struct btrfs_block_group_cache *
7210 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7211 struct btrfs_free_cluster *cluster,
7214 struct btrfs_block_group_cache *used_bg = NULL;
7216 spin_lock(&cluster->refill_lock);
7218 used_bg = cluster->block_group;
7222 if (used_bg == block_group)
7225 btrfs_get_block_group(used_bg);
7230 if (down_read_trylock(&used_bg->data_rwsem))
7233 spin_unlock(&cluster->refill_lock);
7235 /* We should only have one-level nested. */
7236 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7238 spin_lock(&cluster->refill_lock);
7239 if (used_bg == cluster->block_group)
7242 up_read(&used_bg->data_rwsem);
7243 btrfs_put_block_group(used_bg);
7248 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7252 up_read(&cache->data_rwsem);
7253 btrfs_put_block_group(cache);
7257 * Structure used internally for find_free_extent() function. Wraps needed
7260 struct find_free_extent_ctl {
7261 /* Basic allocation info */
7268 /* Where to start the search inside the bg */
7271 /* For clustered allocation */
7274 bool have_caching_bg;
7275 bool orig_have_caching_bg;
7277 /* RAID index, converted from flags */
7281 * Current loop number, check find_free_extent_update_loop() for details
7286 * Whether we're refilling a cluster, if true we need to re-search
7287 * current block group but don't try to refill the cluster again.
7289 bool retry_clustered;
7292 * Whether we're updating free space cache, if true we need to re-search
7293 * current block group but don't try updating free space cache again.
7295 bool retry_unclustered;
7297 /* If current block group is cached */
7300 /* Max contiguous hole found */
7301 u64 max_extent_size;
7303 /* Total free space from free space cache, not always contiguous */
7304 u64 total_free_space;
7312 * Helper function for find_free_extent().
7314 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7315 * Return -EAGAIN to inform caller that we need to re-search this block group
7316 * Return >0 to inform caller that we find nothing
7317 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7319 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7320 struct btrfs_free_cluster *last_ptr,
7321 struct find_free_extent_ctl *ffe_ctl,
7322 struct btrfs_block_group_cache **cluster_bg_ret)
7324 struct btrfs_fs_info *fs_info = bg->fs_info;
7325 struct btrfs_block_group_cache *cluster_bg;
7326 u64 aligned_cluster;
7330 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7332 goto refill_cluster;
7333 if (cluster_bg != bg && (cluster_bg->ro ||
7334 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7335 goto release_cluster;
7337 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7338 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7339 &ffe_ctl->max_extent_size);
7341 /* We have a block, we're done */
7342 spin_unlock(&last_ptr->refill_lock);
7343 trace_btrfs_reserve_extent_cluster(cluster_bg,
7344 ffe_ctl->search_start, ffe_ctl->num_bytes);
7345 *cluster_bg_ret = cluster_bg;
7346 ffe_ctl->found_offset = offset;
7349 WARN_ON(last_ptr->block_group != cluster_bg);
7353 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7354 * lets just skip it and let the allocator find whatever block it can
7355 * find. If we reach this point, we will have tried the cluster
7356 * allocator plenty of times and not have found anything, so we are
7357 * likely way too fragmented for the clustering stuff to find anything.
7359 * However, if the cluster is taken from the current block group,
7360 * release the cluster first, so that we stand a better chance of
7361 * succeeding in the unclustered allocation.
7363 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7364 spin_unlock(&last_ptr->refill_lock);
7365 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7369 /* This cluster didn't work out, free it and start over */
7370 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7372 if (cluster_bg != bg)
7373 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7376 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7377 spin_unlock(&last_ptr->refill_lock);
7381 aligned_cluster = max_t(u64,
7382 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7383 bg->full_stripe_len);
7384 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7385 ffe_ctl->search_start, ffe_ctl->num_bytes,
7388 /* Now pull our allocation out of this cluster */
7389 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7390 ffe_ctl->num_bytes, ffe_ctl->search_start,
7391 &ffe_ctl->max_extent_size);
7393 /* We found one, proceed */
7394 spin_unlock(&last_ptr->refill_lock);
7395 trace_btrfs_reserve_extent_cluster(bg,
7396 ffe_ctl->search_start,
7397 ffe_ctl->num_bytes);
7398 ffe_ctl->found_offset = offset;
7401 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7402 !ffe_ctl->retry_clustered) {
7403 spin_unlock(&last_ptr->refill_lock);
7405 ffe_ctl->retry_clustered = true;
7406 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7407 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7411 * At this point we either didn't find a cluster or we weren't able to
7412 * allocate a block from our cluster. Free the cluster we've been
7413 * trying to use, and go to the next block group.
7415 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7416 spin_unlock(&last_ptr->refill_lock);
7421 * Return >0 to inform caller that we find nothing
7422 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7423 * Return -EAGAIN to inform caller that we need to re-search this block group
7425 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7426 struct btrfs_free_cluster *last_ptr,
7427 struct find_free_extent_ctl *ffe_ctl)
7432 * We are doing an unclustered allocation, set the fragmented flag so
7433 * we don't bother trying to setup a cluster again until we get more
7436 if (unlikely(last_ptr)) {
7437 spin_lock(&last_ptr->lock);
7438 last_ptr->fragmented = 1;
7439 spin_unlock(&last_ptr->lock);
7441 if (ffe_ctl->cached) {
7442 struct btrfs_free_space_ctl *free_space_ctl;
7444 free_space_ctl = bg->free_space_ctl;
7445 spin_lock(&free_space_ctl->tree_lock);
7446 if (free_space_ctl->free_space <
7447 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7448 ffe_ctl->empty_size) {
7449 ffe_ctl->total_free_space = max_t(u64,
7450 ffe_ctl->total_free_space,
7451 free_space_ctl->free_space);
7452 spin_unlock(&free_space_ctl->tree_lock);
7455 spin_unlock(&free_space_ctl->tree_lock);
7458 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7459 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7460 &ffe_ctl->max_extent_size);
7463 * If we didn't find a chunk, and we haven't failed on this block group
7464 * before, and this block group is in the middle of caching and we are
7465 * ok with waiting, then go ahead and wait for progress to be made, and
7466 * set @retry_unclustered to true.
7468 * If @retry_unclustered is true then we've already waited on this
7469 * block group once and should move on to the next block group.
7471 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7472 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7473 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7474 ffe_ctl->empty_size);
7475 ffe_ctl->retry_unclustered = true;
7477 } else if (!offset) {
7480 ffe_ctl->found_offset = offset;
7485 * Return >0 means caller needs to re-search for free extent
7486 * Return 0 means we have the needed free extent.
7487 * Return <0 means we failed to locate any free extent.
7489 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7490 struct btrfs_free_cluster *last_ptr,
7491 struct btrfs_key *ins,
7492 struct find_free_extent_ctl *ffe_ctl,
7493 int full_search, bool use_cluster)
7495 struct btrfs_root *root = fs_info->extent_root;
7498 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7499 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7500 ffe_ctl->orig_have_caching_bg = true;
7502 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7503 ffe_ctl->have_caching_bg)
7506 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7509 if (ins->objectid) {
7510 if (!use_cluster && last_ptr) {
7511 spin_lock(&last_ptr->lock);
7512 last_ptr->window_start = ins->objectid;
7513 spin_unlock(&last_ptr->lock);
7519 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7520 * caching kthreads as we move along
7521 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7522 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7523 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7526 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7528 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7530 * We want to skip the LOOP_CACHING_WAIT step if we
7531 * don't have any uncached bgs and we've already done a
7532 * full search through.
7534 if (ffe_ctl->orig_have_caching_bg || !full_search)
7535 ffe_ctl->loop = LOOP_CACHING_WAIT;
7537 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7542 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7543 struct btrfs_trans_handle *trans;
7546 trans = current->journal_info;
7550 trans = btrfs_join_transaction(root);
7552 if (IS_ERR(trans)) {
7553 ret = PTR_ERR(trans);
7557 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7561 * If we can't allocate a new chunk we've already looped
7562 * through at least once, move on to the NO_EMPTY_SIZE
7566 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7568 /* Do not bail out on ENOSPC since we can do more. */
7569 if (ret < 0 && ret != -ENOSPC)
7570 btrfs_abort_transaction(trans, ret);
7574 btrfs_end_transaction(trans);
7579 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7581 * Don't loop again if we already have no empty_size and
7584 if (ffe_ctl->empty_size == 0 &&
7585 ffe_ctl->empty_cluster == 0)
7587 ffe_ctl->empty_size = 0;
7588 ffe_ctl->empty_cluster = 0;
7596 * walks the btree of allocated extents and find a hole of a given size.
7597 * The key ins is changed to record the hole:
7598 * ins->objectid == start position
7599 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7600 * ins->offset == the size of the hole.
7601 * Any available blocks before search_start are skipped.
7603 * If there is no suitable free space, we will record the max size of
7604 * the free space extent currently.
7606 * The overall logic and call chain:
7608 * find_free_extent()
7609 * |- Iterate through all block groups
7610 * | |- Get a valid block group
7611 * | |- Try to do clustered allocation in that block group
7612 * | |- Try to do unclustered allocation in that block group
7613 * | |- Check if the result is valid
7614 * | | |- If valid, then exit
7615 * | |- Jump to next block group
7617 * |- Push harder to find free extents
7618 * |- If not found, re-iterate all block groups
7620 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7621 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7622 u64 hint_byte, struct btrfs_key *ins,
7623 u64 flags, int delalloc)
7626 struct btrfs_free_cluster *last_ptr = NULL;
7627 struct btrfs_block_group_cache *block_group = NULL;
7628 struct find_free_extent_ctl ffe_ctl = {0};
7629 struct btrfs_space_info *space_info;
7630 bool use_cluster = true;
7631 bool full_search = false;
7633 WARN_ON(num_bytes < fs_info->sectorsize);
7635 ffe_ctl.ram_bytes = ram_bytes;
7636 ffe_ctl.num_bytes = num_bytes;
7637 ffe_ctl.empty_size = empty_size;
7638 ffe_ctl.flags = flags;
7639 ffe_ctl.search_start = 0;
7640 ffe_ctl.retry_clustered = false;
7641 ffe_ctl.retry_unclustered = false;
7642 ffe_ctl.delalloc = delalloc;
7643 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7644 ffe_ctl.have_caching_bg = false;
7645 ffe_ctl.orig_have_caching_bg = false;
7646 ffe_ctl.found_offset = 0;
7648 ins->type = BTRFS_EXTENT_ITEM_KEY;
7652 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7654 space_info = __find_space_info(fs_info, flags);
7656 btrfs_err(fs_info, "No space info for %llu", flags);
7661 * If our free space is heavily fragmented we may not be able to make
7662 * big contiguous allocations, so instead of doing the expensive search
7663 * for free space, simply return ENOSPC with our max_extent_size so we
7664 * can go ahead and search for a more manageable chunk.
7666 * If our max_extent_size is large enough for our allocation simply
7667 * disable clustering since we will likely not be able to find enough
7668 * space to create a cluster and induce latency trying.
7670 if (unlikely(space_info->max_extent_size)) {
7671 spin_lock(&space_info->lock);
7672 if (space_info->max_extent_size &&
7673 num_bytes > space_info->max_extent_size) {
7674 ins->offset = space_info->max_extent_size;
7675 spin_unlock(&space_info->lock);
7677 } else if (space_info->max_extent_size) {
7678 use_cluster = false;
7680 spin_unlock(&space_info->lock);
7683 last_ptr = fetch_cluster_info(fs_info, space_info,
7684 &ffe_ctl.empty_cluster);
7686 spin_lock(&last_ptr->lock);
7687 if (last_ptr->block_group)
7688 hint_byte = last_ptr->window_start;
7689 if (last_ptr->fragmented) {
7691 * We still set window_start so we can keep track of the
7692 * last place we found an allocation to try and save
7695 hint_byte = last_ptr->window_start;
7696 use_cluster = false;
7698 spin_unlock(&last_ptr->lock);
7701 ffe_ctl.search_start = max(ffe_ctl.search_start,
7702 first_logical_byte(fs_info, 0));
7703 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7704 if (ffe_ctl.search_start == hint_byte) {
7705 block_group = btrfs_lookup_block_group(fs_info,
7706 ffe_ctl.search_start);
7708 * we don't want to use the block group if it doesn't match our
7709 * allocation bits, or if its not cached.
7711 * However if we are re-searching with an ideal block group
7712 * picked out then we don't care that the block group is cached.
7714 if (block_group && block_group_bits(block_group, flags) &&
7715 block_group->cached != BTRFS_CACHE_NO) {
7716 down_read(&space_info->groups_sem);
7717 if (list_empty(&block_group->list) ||
7720 * someone is removing this block group,
7721 * we can't jump into the have_block_group
7722 * target because our list pointers are not
7725 btrfs_put_block_group(block_group);
7726 up_read(&space_info->groups_sem);
7728 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7729 block_group->flags);
7730 btrfs_lock_block_group(block_group, delalloc);
7731 goto have_block_group;
7733 } else if (block_group) {
7734 btrfs_put_block_group(block_group);
7738 ffe_ctl.have_caching_bg = false;
7739 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7742 down_read(&space_info->groups_sem);
7743 list_for_each_entry(block_group,
7744 &space_info->block_groups[ffe_ctl.index], list) {
7745 /* If the block group is read-only, we can skip it entirely. */
7746 if (unlikely(block_group->ro))
7749 btrfs_grab_block_group(block_group, delalloc);
7750 ffe_ctl.search_start = block_group->key.objectid;
7753 * this can happen if we end up cycling through all the
7754 * raid types, but we want to make sure we only allocate
7755 * for the proper type.
7757 if (!block_group_bits(block_group, flags)) {
7758 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7759 BTRFS_BLOCK_GROUP_RAID1 |
7760 BTRFS_BLOCK_GROUP_RAID5 |
7761 BTRFS_BLOCK_GROUP_RAID6 |
7762 BTRFS_BLOCK_GROUP_RAID10;
7765 * if they asked for extra copies and this block group
7766 * doesn't provide them, bail. This does allow us to
7767 * fill raid0 from raid1.
7769 if ((flags & extra) && !(block_group->flags & extra))
7774 ffe_ctl.cached = block_group_cache_done(block_group);
7775 if (unlikely(!ffe_ctl.cached)) {
7776 ffe_ctl.have_caching_bg = true;
7777 ret = cache_block_group(block_group, 0);
7782 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7786 * Ok we want to try and use the cluster allocator, so
7789 if (last_ptr && use_cluster) {
7790 struct btrfs_block_group_cache *cluster_bg = NULL;
7792 ret = find_free_extent_clustered(block_group, last_ptr,
7793 &ffe_ctl, &cluster_bg);
7796 if (cluster_bg && cluster_bg != block_group) {
7797 btrfs_release_block_group(block_group,
7799 block_group = cluster_bg;
7802 } else if (ret == -EAGAIN) {
7803 goto have_block_group;
7804 } else if (ret > 0) {
7807 /* ret == -ENOENT case falls through */
7810 ret = find_free_extent_unclustered(block_group, last_ptr,
7813 goto have_block_group;
7816 /* ret == 0 case falls through */
7818 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
7819 fs_info->stripesize);
7821 /* move on to the next group */
7822 if (ffe_ctl.search_start + num_bytes >
7823 block_group->key.objectid + block_group->key.offset) {
7824 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7829 if (ffe_ctl.found_offset < ffe_ctl.search_start)
7830 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7831 ffe_ctl.search_start - ffe_ctl.found_offset);
7833 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7834 num_bytes, delalloc);
7835 if (ret == -EAGAIN) {
7836 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
7840 btrfs_inc_block_group_reservations(block_group);
7842 /* we are all good, lets return */
7843 ins->objectid = ffe_ctl.search_start;
7844 ins->offset = num_bytes;
7846 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
7848 btrfs_release_block_group(block_group, delalloc);
7851 ffe_ctl.retry_clustered = false;
7852 ffe_ctl.retry_unclustered = false;
7853 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7855 btrfs_release_block_group(block_group, delalloc);
7858 up_read(&space_info->groups_sem);
7860 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
7861 full_search, use_cluster);
7865 if (ret == -ENOSPC) {
7867 * Use ffe_ctl->total_free_space as fallback if we can't find
7868 * any contiguous hole.
7870 if (!ffe_ctl.max_extent_size)
7871 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
7872 spin_lock(&space_info->lock);
7873 space_info->max_extent_size = ffe_ctl.max_extent_size;
7874 spin_unlock(&space_info->lock);
7875 ins->offset = ffe_ctl.max_extent_size;
7880 static void dump_space_info(struct btrfs_fs_info *fs_info,
7881 struct btrfs_space_info *info, u64 bytes,
7882 int dump_block_groups)
7884 struct btrfs_block_group_cache *cache;
7887 spin_lock(&info->lock);
7888 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7890 info->total_bytes - btrfs_space_info_used(info, true),
7891 info->full ? "" : "not ");
7893 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7894 info->total_bytes, info->bytes_used, info->bytes_pinned,
7895 info->bytes_reserved, info->bytes_may_use,
7896 info->bytes_readonly);
7897 spin_unlock(&info->lock);
7899 if (!dump_block_groups)
7902 down_read(&info->groups_sem);
7904 list_for_each_entry(cache, &info->block_groups[index], list) {
7905 spin_lock(&cache->lock);
7907 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7908 cache->key.objectid, cache->key.offset,
7909 btrfs_block_group_used(&cache->item), cache->pinned,
7910 cache->reserved, cache->ro ? "[readonly]" : "");
7911 btrfs_dump_free_space(cache, bytes);
7912 spin_unlock(&cache->lock);
7914 if (++index < BTRFS_NR_RAID_TYPES)
7916 up_read(&info->groups_sem);
7920 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7921 * hole that is at least as big as @num_bytes.
7923 * @root - The root that will contain this extent
7925 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7926 * is used for accounting purposes. This value differs
7927 * from @num_bytes only in the case of compressed extents.
7929 * @num_bytes - Number of bytes to allocate on-disk.
7931 * @min_alloc_size - Indicates the minimum amount of space that the
7932 * allocator should try to satisfy. In some cases
7933 * @num_bytes may be larger than what is required and if
7934 * the filesystem is fragmented then allocation fails.
7935 * However, the presence of @min_alloc_size gives a
7936 * chance to try and satisfy the smaller allocation.
7938 * @empty_size - A hint that you plan on doing more COW. This is the
7939 * size in bytes the allocator should try to find free
7940 * next to the block it returns. This is just a hint and
7941 * may be ignored by the allocator.
7943 * @hint_byte - Hint to the allocator to start searching above the byte
7944 * address passed. It might be ignored.
7946 * @ins - This key is modified to record the found hole. It will
7947 * have the following values:
7948 * ins->objectid == start position
7949 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7950 * ins->offset == the size of the hole.
7952 * @is_data - Boolean flag indicating whether an extent is
7953 * allocated for data (true) or metadata (false)
7955 * @delalloc - Boolean flag indicating whether this allocation is for
7956 * delalloc or not. If 'true' data_rwsem of block groups
7957 * is going to be acquired.
7960 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7961 * case -ENOSPC is returned then @ins->offset will contain the size of the
7962 * largest available hole the allocator managed to find.
7964 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7965 u64 num_bytes, u64 min_alloc_size,
7966 u64 empty_size, u64 hint_byte,
7967 struct btrfs_key *ins, int is_data, int delalloc)
7969 struct btrfs_fs_info *fs_info = root->fs_info;
7970 bool final_tried = num_bytes == min_alloc_size;
7974 flags = get_alloc_profile_by_root(root, is_data);
7976 WARN_ON(num_bytes < fs_info->sectorsize);
7977 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7978 hint_byte, ins, flags, delalloc);
7979 if (!ret && !is_data) {
7980 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7981 } else if (ret == -ENOSPC) {
7982 if (!final_tried && ins->offset) {
7983 num_bytes = min(num_bytes >> 1, ins->offset);
7984 num_bytes = round_down(num_bytes,
7985 fs_info->sectorsize);
7986 num_bytes = max(num_bytes, min_alloc_size);
7987 ram_bytes = num_bytes;
7988 if (num_bytes == min_alloc_size)
7991 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7992 struct btrfs_space_info *sinfo;
7994 sinfo = __find_space_info(fs_info, flags);
7996 "allocation failed flags %llu, wanted %llu",
7999 dump_space_info(fs_info, sinfo, num_bytes, 1);
8006 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8008 int pin, int delalloc)
8010 struct btrfs_block_group_cache *cache;
8013 cache = btrfs_lookup_block_group(fs_info, start);
8015 btrfs_err(fs_info, "Unable to find block group for %llu",
8021 pin_down_extent(fs_info, cache, start, len, 1);
8023 if (btrfs_test_opt(fs_info, DISCARD))
8024 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8025 btrfs_add_free_space(cache, start, len);
8026 btrfs_free_reserved_bytes(cache, len, delalloc);
8027 trace_btrfs_reserved_extent_free(fs_info, start, len);
8030 btrfs_put_block_group(cache);
8034 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8035 u64 start, u64 len, int delalloc)
8037 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8040 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8043 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8046 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8047 u64 parent, u64 root_objectid,
8048 u64 flags, u64 owner, u64 offset,
8049 struct btrfs_key *ins, int ref_mod)
8051 struct btrfs_fs_info *fs_info = trans->fs_info;
8053 struct btrfs_extent_item *extent_item;
8054 struct btrfs_extent_inline_ref *iref;
8055 struct btrfs_path *path;
8056 struct extent_buffer *leaf;
8061 type = BTRFS_SHARED_DATA_REF_KEY;
8063 type = BTRFS_EXTENT_DATA_REF_KEY;
8065 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8067 path = btrfs_alloc_path();
8071 path->leave_spinning = 1;
8072 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8075 btrfs_free_path(path);
8079 leaf = path->nodes[0];
8080 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8081 struct btrfs_extent_item);
8082 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8083 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8084 btrfs_set_extent_flags(leaf, extent_item,
8085 flags | BTRFS_EXTENT_FLAG_DATA);
8087 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8088 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8090 struct btrfs_shared_data_ref *ref;
8091 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8092 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8093 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8095 struct btrfs_extent_data_ref *ref;
8096 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8097 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8098 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8099 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8100 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8103 btrfs_mark_buffer_dirty(path->nodes[0]);
8104 btrfs_free_path(path);
8106 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8110 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8111 if (ret) { /* -ENOENT, logic error */
8112 btrfs_err(fs_info, "update block group failed for %llu %llu",
8113 ins->objectid, ins->offset);
8116 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8120 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8121 struct btrfs_delayed_ref_node *node,
8122 struct btrfs_delayed_extent_op *extent_op)
8124 struct btrfs_fs_info *fs_info = trans->fs_info;
8126 struct btrfs_extent_item *extent_item;
8127 struct btrfs_key extent_key;
8128 struct btrfs_tree_block_info *block_info;
8129 struct btrfs_extent_inline_ref *iref;
8130 struct btrfs_path *path;
8131 struct extent_buffer *leaf;
8132 struct btrfs_delayed_tree_ref *ref;
8133 u32 size = sizeof(*extent_item) + sizeof(*iref);
8135 u64 flags = extent_op->flags_to_set;
8136 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8138 ref = btrfs_delayed_node_to_tree_ref(node);
8140 extent_key.objectid = node->bytenr;
8141 if (skinny_metadata) {
8142 extent_key.offset = ref->level;
8143 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8144 num_bytes = fs_info->nodesize;
8146 extent_key.offset = node->num_bytes;
8147 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8148 size += sizeof(*block_info);
8149 num_bytes = node->num_bytes;
8152 path = btrfs_alloc_path();
8156 path->leave_spinning = 1;
8157 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8160 btrfs_free_path(path);
8164 leaf = path->nodes[0];
8165 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8166 struct btrfs_extent_item);
8167 btrfs_set_extent_refs(leaf, extent_item, 1);
8168 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8169 btrfs_set_extent_flags(leaf, extent_item,
8170 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8172 if (skinny_metadata) {
8173 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8175 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8176 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8177 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8178 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8181 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8182 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8183 btrfs_set_extent_inline_ref_type(leaf, iref,
8184 BTRFS_SHARED_BLOCK_REF_KEY);
8185 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8187 btrfs_set_extent_inline_ref_type(leaf, iref,
8188 BTRFS_TREE_BLOCK_REF_KEY);
8189 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8192 btrfs_mark_buffer_dirty(leaf);
8193 btrfs_free_path(path);
8195 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8200 ret = update_block_group(trans, fs_info, extent_key.objectid,
8201 fs_info->nodesize, 1);
8202 if (ret) { /* -ENOENT, logic error */
8203 btrfs_err(fs_info, "update block group failed for %llu %llu",
8204 extent_key.objectid, extent_key.offset);
8208 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8213 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8214 struct btrfs_root *root, u64 owner,
8215 u64 offset, u64 ram_bytes,
8216 struct btrfs_key *ins)
8220 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8222 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8223 root->root_key.objectid, owner, offset,
8224 BTRFS_ADD_DELAYED_EXTENT);
8226 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8228 root->root_key.objectid, owner,
8230 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8235 * this is used by the tree logging recovery code. It records that
8236 * an extent has been allocated and makes sure to clear the free
8237 * space cache bits as well
8239 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8240 u64 root_objectid, u64 owner, u64 offset,
8241 struct btrfs_key *ins)
8243 struct btrfs_fs_info *fs_info = trans->fs_info;
8245 struct btrfs_block_group_cache *block_group;
8246 struct btrfs_space_info *space_info;
8249 * Mixed block groups will exclude before processing the log so we only
8250 * need to do the exclude dance if this fs isn't mixed.
8252 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8253 ret = __exclude_logged_extent(fs_info, ins->objectid,
8259 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8263 space_info = block_group->space_info;
8264 spin_lock(&space_info->lock);
8265 spin_lock(&block_group->lock);
8266 space_info->bytes_reserved += ins->offset;
8267 block_group->reserved += ins->offset;
8268 spin_unlock(&block_group->lock);
8269 spin_unlock(&space_info->lock);
8271 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8273 btrfs_put_block_group(block_group);
8277 static struct extent_buffer *
8278 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8279 u64 bytenr, int level, u64 owner)
8281 struct btrfs_fs_info *fs_info = root->fs_info;
8282 struct extent_buffer *buf;
8284 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8289 * Extra safety check in case the extent tree is corrupted and extent
8290 * allocator chooses to use a tree block which is already used and
8293 if (buf->lock_owner == current->pid) {
8294 btrfs_err_rl(fs_info,
8295 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8296 buf->start, btrfs_header_owner(buf), current->pid);
8297 free_extent_buffer(buf);
8298 return ERR_PTR(-EUCLEAN);
8301 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8302 btrfs_tree_lock(buf);
8303 clean_tree_block(fs_info, buf);
8304 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8306 btrfs_set_lock_blocking(buf);
8307 set_extent_buffer_uptodate(buf);
8309 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8310 btrfs_set_header_level(buf, level);
8311 btrfs_set_header_bytenr(buf, buf->start);
8312 btrfs_set_header_generation(buf, trans->transid);
8313 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8314 btrfs_set_header_owner(buf, owner);
8315 write_extent_buffer_fsid(buf, fs_info->fsid);
8316 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8317 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8318 buf->log_index = root->log_transid % 2;
8320 * we allow two log transactions at a time, use different
8321 * EXENT bit to differentiate dirty pages.
8323 if (buf->log_index == 0)
8324 set_extent_dirty(&root->dirty_log_pages, buf->start,
8325 buf->start + buf->len - 1, GFP_NOFS);
8327 set_extent_new(&root->dirty_log_pages, buf->start,
8328 buf->start + buf->len - 1);
8330 buf->log_index = -1;
8331 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8332 buf->start + buf->len - 1, GFP_NOFS);
8334 trans->dirty = true;
8335 /* this returns a buffer locked for blocking */
8339 static struct btrfs_block_rsv *
8340 use_block_rsv(struct btrfs_trans_handle *trans,
8341 struct btrfs_root *root, u32 blocksize)
8343 struct btrfs_fs_info *fs_info = root->fs_info;
8344 struct btrfs_block_rsv *block_rsv;
8345 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8347 bool global_updated = false;
8349 block_rsv = get_block_rsv(trans, root);
8351 if (unlikely(block_rsv->size == 0))
8354 ret = block_rsv_use_bytes(block_rsv, blocksize);
8358 if (block_rsv->failfast)
8359 return ERR_PTR(ret);
8361 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8362 global_updated = true;
8363 update_global_block_rsv(fs_info);
8367 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8368 static DEFINE_RATELIMIT_STATE(_rs,
8369 DEFAULT_RATELIMIT_INTERVAL * 10,
8370 /*DEFAULT_RATELIMIT_BURST*/ 1);
8371 if (__ratelimit(&_rs))
8373 "BTRFS: block rsv returned %d\n", ret);
8376 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8377 BTRFS_RESERVE_NO_FLUSH);
8381 * If we couldn't reserve metadata bytes try and use some from
8382 * the global reserve if its space type is the same as the global
8385 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8386 block_rsv->space_info == global_rsv->space_info) {
8387 ret = block_rsv_use_bytes(global_rsv, blocksize);
8391 return ERR_PTR(ret);
8394 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8395 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8397 block_rsv_add_bytes(block_rsv, blocksize, false);
8398 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8402 * finds a free extent and does all the dirty work required for allocation
8403 * returns the tree buffer or an ERR_PTR on error.
8405 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8406 struct btrfs_root *root,
8407 u64 parent, u64 root_objectid,
8408 const struct btrfs_disk_key *key,
8409 int level, u64 hint,
8412 struct btrfs_fs_info *fs_info = root->fs_info;
8413 struct btrfs_key ins;
8414 struct btrfs_block_rsv *block_rsv;
8415 struct extent_buffer *buf;
8416 struct btrfs_delayed_extent_op *extent_op;
8419 u32 blocksize = fs_info->nodesize;
8420 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8422 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8423 if (btrfs_is_testing(fs_info)) {
8424 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8425 level, root_objectid);
8427 root->alloc_bytenr += blocksize;
8432 block_rsv = use_block_rsv(trans, root, blocksize);
8433 if (IS_ERR(block_rsv))
8434 return ERR_CAST(block_rsv);
8436 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8437 empty_size, hint, &ins, 0, 0);
8441 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8445 goto out_free_reserved;
8448 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8450 parent = ins.objectid;
8451 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8455 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8456 extent_op = btrfs_alloc_delayed_extent_op();
8462 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8464 memset(&extent_op->key, 0, sizeof(extent_op->key));
8465 extent_op->flags_to_set = flags;
8466 extent_op->update_key = skinny_metadata ? false : true;
8467 extent_op->update_flags = true;
8468 extent_op->is_data = false;
8469 extent_op->level = level;
8471 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8472 root_objectid, level, 0,
8473 BTRFS_ADD_DELAYED_EXTENT);
8474 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8476 root_objectid, level,
8477 BTRFS_ADD_DELAYED_EXTENT,
8478 extent_op, NULL, NULL);
8480 goto out_free_delayed;
8485 btrfs_free_delayed_extent_op(extent_op);
8487 free_extent_buffer(buf);
8489 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8491 unuse_block_rsv(fs_info, block_rsv, blocksize);
8492 return ERR_PTR(ret);
8495 struct walk_control {
8496 u64 refs[BTRFS_MAX_LEVEL];
8497 u64 flags[BTRFS_MAX_LEVEL];
8498 struct btrfs_key update_progress;
8508 #define DROP_REFERENCE 1
8509 #define UPDATE_BACKREF 2
8511 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8512 struct btrfs_root *root,
8513 struct walk_control *wc,
8514 struct btrfs_path *path)
8516 struct btrfs_fs_info *fs_info = root->fs_info;
8522 struct btrfs_key key;
8523 struct extent_buffer *eb;
8528 if (path->slots[wc->level] < wc->reada_slot) {
8529 wc->reada_count = wc->reada_count * 2 / 3;
8530 wc->reada_count = max(wc->reada_count, 2);
8532 wc->reada_count = wc->reada_count * 3 / 2;
8533 wc->reada_count = min_t(int, wc->reada_count,
8534 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8537 eb = path->nodes[wc->level];
8538 nritems = btrfs_header_nritems(eb);
8540 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8541 if (nread >= wc->reada_count)
8545 bytenr = btrfs_node_blockptr(eb, slot);
8546 generation = btrfs_node_ptr_generation(eb, slot);
8548 if (slot == path->slots[wc->level])
8551 if (wc->stage == UPDATE_BACKREF &&
8552 generation <= root->root_key.offset)
8555 /* We don't lock the tree block, it's OK to be racy here */
8556 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8557 wc->level - 1, 1, &refs,
8559 /* We don't care about errors in readahead. */
8564 if (wc->stage == DROP_REFERENCE) {
8568 if (wc->level == 1 &&
8569 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8571 if (!wc->update_ref ||
8572 generation <= root->root_key.offset)
8574 btrfs_node_key_to_cpu(eb, &key, slot);
8575 ret = btrfs_comp_cpu_keys(&key,
8576 &wc->update_progress);
8580 if (wc->level == 1 &&
8581 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8585 readahead_tree_block(fs_info, bytenr);
8588 wc->reada_slot = slot;
8592 * helper to process tree block while walking down the tree.
8594 * when wc->stage == UPDATE_BACKREF, this function updates
8595 * back refs for pointers in the block.
8597 * NOTE: return value 1 means we should stop walking down.
8599 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8600 struct btrfs_root *root,
8601 struct btrfs_path *path,
8602 struct walk_control *wc, int lookup_info)
8604 struct btrfs_fs_info *fs_info = root->fs_info;
8605 int level = wc->level;
8606 struct extent_buffer *eb = path->nodes[level];
8607 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8610 if (wc->stage == UPDATE_BACKREF &&
8611 btrfs_header_owner(eb) != root->root_key.objectid)
8615 * when reference count of tree block is 1, it won't increase
8616 * again. once full backref flag is set, we never clear it.
8619 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8620 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8621 BUG_ON(!path->locks[level]);
8622 ret = btrfs_lookup_extent_info(trans, fs_info,
8623 eb->start, level, 1,
8626 BUG_ON(ret == -ENOMEM);
8629 BUG_ON(wc->refs[level] == 0);
8632 if (wc->stage == DROP_REFERENCE) {
8633 if (wc->refs[level] > 1)
8636 if (path->locks[level] && !wc->keep_locks) {
8637 btrfs_tree_unlock_rw(eb, path->locks[level]);
8638 path->locks[level] = 0;
8643 /* wc->stage == UPDATE_BACKREF */
8644 if (!(wc->flags[level] & flag)) {
8645 BUG_ON(!path->locks[level]);
8646 ret = btrfs_inc_ref(trans, root, eb, 1);
8647 BUG_ON(ret); /* -ENOMEM */
8648 ret = btrfs_dec_ref(trans, root, eb, 0);
8649 BUG_ON(ret); /* -ENOMEM */
8650 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8652 btrfs_header_level(eb), 0);
8653 BUG_ON(ret); /* -ENOMEM */
8654 wc->flags[level] |= flag;
8658 * the block is shared by multiple trees, so it's not good to
8659 * keep the tree lock
8661 if (path->locks[level] && level > 0) {
8662 btrfs_tree_unlock_rw(eb, path->locks[level]);
8663 path->locks[level] = 0;
8669 * helper to process tree block pointer.
8671 * when wc->stage == DROP_REFERENCE, this function checks
8672 * reference count of the block pointed to. if the block
8673 * is shared and we need update back refs for the subtree
8674 * rooted at the block, this function changes wc->stage to
8675 * UPDATE_BACKREF. if the block is shared and there is no
8676 * need to update back, this function drops the reference
8679 * NOTE: return value 1 means we should stop walking down.
8681 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8682 struct btrfs_root *root,
8683 struct btrfs_path *path,
8684 struct walk_control *wc, int *lookup_info)
8686 struct btrfs_fs_info *fs_info = root->fs_info;
8691 struct btrfs_key key;
8692 struct btrfs_key first_key;
8693 struct extent_buffer *next;
8694 int level = wc->level;
8697 bool need_account = false;
8699 generation = btrfs_node_ptr_generation(path->nodes[level],
8700 path->slots[level]);
8702 * if the lower level block was created before the snapshot
8703 * was created, we know there is no need to update back refs
8706 if (wc->stage == UPDATE_BACKREF &&
8707 generation <= root->root_key.offset) {
8712 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8713 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8714 path->slots[level]);
8715 blocksize = fs_info->nodesize;
8717 next = find_extent_buffer(fs_info, bytenr);
8719 next = btrfs_find_create_tree_block(fs_info, bytenr);
8721 return PTR_ERR(next);
8723 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8727 btrfs_tree_lock(next);
8728 btrfs_set_lock_blocking(next);
8730 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8731 &wc->refs[level - 1],
8732 &wc->flags[level - 1]);
8736 if (unlikely(wc->refs[level - 1] == 0)) {
8737 btrfs_err(fs_info, "Missing references.");
8743 if (wc->stage == DROP_REFERENCE) {
8744 if (wc->refs[level - 1] > 1) {
8745 need_account = true;
8747 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8750 if (!wc->update_ref ||
8751 generation <= root->root_key.offset)
8754 btrfs_node_key_to_cpu(path->nodes[level], &key,
8755 path->slots[level]);
8756 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8760 wc->stage = UPDATE_BACKREF;
8761 wc->shared_level = level - 1;
8765 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8769 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8770 btrfs_tree_unlock(next);
8771 free_extent_buffer(next);
8777 if (reada && level == 1)
8778 reada_walk_down(trans, root, wc, path);
8779 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8782 return PTR_ERR(next);
8783 } else if (!extent_buffer_uptodate(next)) {
8784 free_extent_buffer(next);
8787 btrfs_tree_lock(next);
8788 btrfs_set_lock_blocking(next);
8792 ASSERT(level == btrfs_header_level(next));
8793 if (level != btrfs_header_level(next)) {
8794 btrfs_err(root->fs_info, "mismatched level");
8798 path->nodes[level] = next;
8799 path->slots[level] = 0;
8800 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8806 wc->refs[level - 1] = 0;
8807 wc->flags[level - 1] = 0;
8808 if (wc->stage == DROP_REFERENCE) {
8809 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8810 parent = path->nodes[level]->start;
8812 ASSERT(root->root_key.objectid ==
8813 btrfs_header_owner(path->nodes[level]));
8814 if (root->root_key.objectid !=
8815 btrfs_header_owner(path->nodes[level])) {
8816 btrfs_err(root->fs_info,
8817 "mismatched block owner");
8825 * Reloc tree doesn't contribute to qgroup numbers, and we have
8826 * already accounted them at merge time (replace_path),
8827 * thus we could skip expensive subtree trace here.
8829 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
8831 ret = btrfs_qgroup_trace_subtree(trans, next,
8832 generation, level - 1);
8834 btrfs_err_rl(fs_info,
8835 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8839 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8840 parent, root->root_key.objectid,
8850 btrfs_tree_unlock(next);
8851 free_extent_buffer(next);
8857 * helper to process tree block while walking up the tree.
8859 * when wc->stage == DROP_REFERENCE, this function drops
8860 * reference count on the block.
8862 * when wc->stage == UPDATE_BACKREF, this function changes
8863 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8864 * to UPDATE_BACKREF previously while processing the block.
8866 * NOTE: return value 1 means we should stop walking up.
8868 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8869 struct btrfs_root *root,
8870 struct btrfs_path *path,
8871 struct walk_control *wc)
8873 struct btrfs_fs_info *fs_info = root->fs_info;
8875 int level = wc->level;
8876 struct extent_buffer *eb = path->nodes[level];
8879 if (wc->stage == UPDATE_BACKREF) {
8880 BUG_ON(wc->shared_level < level);
8881 if (level < wc->shared_level)
8884 ret = find_next_key(path, level + 1, &wc->update_progress);
8888 wc->stage = DROP_REFERENCE;
8889 wc->shared_level = -1;
8890 path->slots[level] = 0;
8893 * check reference count again if the block isn't locked.
8894 * we should start walking down the tree again if reference
8897 if (!path->locks[level]) {
8899 btrfs_tree_lock(eb);
8900 btrfs_set_lock_blocking(eb);
8901 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8903 ret = btrfs_lookup_extent_info(trans, fs_info,
8904 eb->start, level, 1,
8908 btrfs_tree_unlock_rw(eb, path->locks[level]);
8909 path->locks[level] = 0;
8912 BUG_ON(wc->refs[level] == 0);
8913 if (wc->refs[level] == 1) {
8914 btrfs_tree_unlock_rw(eb, path->locks[level]);
8915 path->locks[level] = 0;
8921 /* wc->stage == DROP_REFERENCE */
8922 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8924 if (wc->refs[level] == 1) {
8926 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8927 ret = btrfs_dec_ref(trans, root, eb, 1);
8929 ret = btrfs_dec_ref(trans, root, eb, 0);
8930 BUG_ON(ret); /* -ENOMEM */
8931 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
8933 btrfs_err_rl(fs_info,
8934 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8938 /* make block locked assertion in clean_tree_block happy */
8939 if (!path->locks[level] &&
8940 btrfs_header_generation(eb) == trans->transid) {
8941 btrfs_tree_lock(eb);
8942 btrfs_set_lock_blocking(eb);
8943 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8945 clean_tree_block(fs_info, eb);
8948 if (eb == root->node) {
8949 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8951 else if (root->root_key.objectid != btrfs_header_owner(eb))
8952 goto owner_mismatch;
8954 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8955 parent = path->nodes[level + 1]->start;
8956 else if (root->root_key.objectid !=
8957 btrfs_header_owner(path->nodes[level + 1]))
8958 goto owner_mismatch;
8961 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8963 wc->refs[level] = 0;
8964 wc->flags[level] = 0;
8968 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
8969 btrfs_header_owner(eb), root->root_key.objectid);
8973 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8974 struct btrfs_root *root,
8975 struct btrfs_path *path,
8976 struct walk_control *wc)
8978 int level = wc->level;
8979 int lookup_info = 1;
8982 while (level >= 0) {
8983 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8990 if (path->slots[level] >=
8991 btrfs_header_nritems(path->nodes[level]))
8994 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8996 path->slots[level]++;
9005 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9006 struct btrfs_root *root,
9007 struct btrfs_path *path,
9008 struct walk_control *wc, int max_level)
9010 int level = wc->level;
9013 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9014 while (level < max_level && path->nodes[level]) {
9016 if (path->slots[level] + 1 <
9017 btrfs_header_nritems(path->nodes[level])) {
9018 path->slots[level]++;
9021 ret = walk_up_proc(trans, root, path, wc);
9027 if (path->locks[level]) {
9028 btrfs_tree_unlock_rw(path->nodes[level],
9029 path->locks[level]);
9030 path->locks[level] = 0;
9032 free_extent_buffer(path->nodes[level]);
9033 path->nodes[level] = NULL;
9041 * drop a subvolume tree.
9043 * this function traverses the tree freeing any blocks that only
9044 * referenced by the tree.
9046 * when a shared tree block is found. this function decreases its
9047 * reference count by one. if update_ref is true, this function
9048 * also make sure backrefs for the shared block and all lower level
9049 * blocks are properly updated.
9051 * If called with for_reloc == 0, may exit early with -EAGAIN
9053 int btrfs_drop_snapshot(struct btrfs_root *root,
9054 struct btrfs_block_rsv *block_rsv, int update_ref,
9057 struct btrfs_fs_info *fs_info = root->fs_info;
9058 struct btrfs_path *path;
9059 struct btrfs_trans_handle *trans;
9060 struct btrfs_root *tree_root = fs_info->tree_root;
9061 struct btrfs_root_item *root_item = &root->root_item;
9062 struct walk_control *wc;
9063 struct btrfs_key key;
9067 bool root_dropped = false;
9069 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9071 path = btrfs_alloc_path();
9077 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9079 btrfs_free_path(path);
9084 trans = btrfs_start_transaction(tree_root, 0);
9085 if (IS_ERR(trans)) {
9086 err = PTR_ERR(trans);
9091 trans->block_rsv = block_rsv;
9093 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9094 level = btrfs_header_level(root->node);
9095 path->nodes[level] = btrfs_lock_root_node(root);
9096 btrfs_set_lock_blocking(path->nodes[level]);
9097 path->slots[level] = 0;
9098 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9099 memset(&wc->update_progress, 0,
9100 sizeof(wc->update_progress));
9102 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9103 memcpy(&wc->update_progress, &key,
9104 sizeof(wc->update_progress));
9106 level = root_item->drop_level;
9108 path->lowest_level = level;
9109 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9110 path->lowest_level = 0;
9118 * unlock our path, this is safe because only this
9119 * function is allowed to delete this snapshot
9121 btrfs_unlock_up_safe(path, 0);
9123 level = btrfs_header_level(root->node);
9125 btrfs_tree_lock(path->nodes[level]);
9126 btrfs_set_lock_blocking(path->nodes[level]);
9127 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9129 ret = btrfs_lookup_extent_info(trans, fs_info,
9130 path->nodes[level]->start,
9131 level, 1, &wc->refs[level],
9137 BUG_ON(wc->refs[level] == 0);
9139 if (level == root_item->drop_level)
9142 btrfs_tree_unlock(path->nodes[level]);
9143 path->locks[level] = 0;
9144 WARN_ON(wc->refs[level] != 1);
9150 wc->shared_level = -1;
9151 wc->stage = DROP_REFERENCE;
9152 wc->update_ref = update_ref;
9154 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9158 ret = walk_down_tree(trans, root, path, wc);
9164 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9171 BUG_ON(wc->stage != DROP_REFERENCE);
9175 if (wc->stage == DROP_REFERENCE) {
9177 btrfs_node_key(path->nodes[level],
9178 &root_item->drop_progress,
9179 path->slots[level]);
9180 root_item->drop_level = level;
9183 BUG_ON(wc->level == 0);
9184 if (btrfs_should_end_transaction(trans) ||
9185 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9186 ret = btrfs_update_root(trans, tree_root,
9190 btrfs_abort_transaction(trans, ret);
9195 btrfs_end_transaction_throttle(trans);
9196 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9197 btrfs_debug(fs_info,
9198 "drop snapshot early exit");
9203 trans = btrfs_start_transaction(tree_root, 0);
9204 if (IS_ERR(trans)) {
9205 err = PTR_ERR(trans);
9209 trans->block_rsv = block_rsv;
9212 btrfs_release_path(path);
9216 ret = btrfs_del_root(trans, &root->root_key);
9218 btrfs_abort_transaction(trans, ret);
9223 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9224 ret = btrfs_find_root(tree_root, &root->root_key, path,
9227 btrfs_abort_transaction(trans, ret);
9230 } else if (ret > 0) {
9231 /* if we fail to delete the orphan item this time
9232 * around, it'll get picked up the next time.
9234 * The most common failure here is just -ENOENT.
9236 btrfs_del_orphan_item(trans, tree_root,
9237 root->root_key.objectid);
9241 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9242 btrfs_add_dropped_root(trans, root);
9244 free_extent_buffer(root->node);
9245 free_extent_buffer(root->commit_root);
9246 btrfs_put_fs_root(root);
9248 root_dropped = true;
9250 btrfs_end_transaction_throttle(trans);
9253 btrfs_free_path(path);
9256 * So if we need to stop dropping the snapshot for whatever reason we
9257 * need to make sure to add it back to the dead root list so that we
9258 * keep trying to do the work later. This also cleans up roots if we
9259 * don't have it in the radix (like when we recover after a power fail
9260 * or unmount) so we don't leak memory.
9262 if (!for_reloc && !root_dropped)
9263 btrfs_add_dead_root(root);
9264 if (err && err != -EAGAIN)
9265 btrfs_handle_fs_error(fs_info, err, NULL);
9270 * drop subtree rooted at tree block 'node'.
9272 * NOTE: this function will unlock and release tree block 'node'
9273 * only used by relocation code
9275 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9276 struct btrfs_root *root,
9277 struct extent_buffer *node,
9278 struct extent_buffer *parent)
9280 struct btrfs_fs_info *fs_info = root->fs_info;
9281 struct btrfs_path *path;
9282 struct walk_control *wc;
9288 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9290 path = btrfs_alloc_path();
9294 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9296 btrfs_free_path(path);
9300 btrfs_assert_tree_locked(parent);
9301 parent_level = btrfs_header_level(parent);
9302 extent_buffer_get(parent);
9303 path->nodes[parent_level] = parent;
9304 path->slots[parent_level] = btrfs_header_nritems(parent);
9306 btrfs_assert_tree_locked(node);
9307 level = btrfs_header_level(node);
9308 path->nodes[level] = node;
9309 path->slots[level] = 0;
9310 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9312 wc->refs[parent_level] = 1;
9313 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9315 wc->shared_level = -1;
9316 wc->stage = DROP_REFERENCE;
9319 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9322 wret = walk_down_tree(trans, root, path, wc);
9328 wret = walk_up_tree(trans, root, path, wc, parent_level);
9336 btrfs_free_path(path);
9340 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9346 * if restripe for this chunk_type is on pick target profile and
9347 * return, otherwise do the usual balance
9349 stripped = get_restripe_target(fs_info, flags);
9351 return extended_to_chunk(stripped);
9353 num_devices = fs_info->fs_devices->rw_devices;
9355 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9356 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9357 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9359 if (num_devices == 1) {
9360 stripped |= BTRFS_BLOCK_GROUP_DUP;
9361 stripped = flags & ~stripped;
9363 /* turn raid0 into single device chunks */
9364 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9367 /* turn mirroring into duplication */
9368 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9369 BTRFS_BLOCK_GROUP_RAID10))
9370 return stripped | BTRFS_BLOCK_GROUP_DUP;
9372 /* they already had raid on here, just return */
9373 if (flags & stripped)
9376 stripped |= BTRFS_BLOCK_GROUP_DUP;
9377 stripped = flags & ~stripped;
9379 /* switch duplicated blocks with raid1 */
9380 if (flags & BTRFS_BLOCK_GROUP_DUP)
9381 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9383 /* this is drive concat, leave it alone */
9389 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9391 struct btrfs_space_info *sinfo = cache->space_info;
9393 u64 min_allocable_bytes;
9397 * We need some metadata space and system metadata space for
9398 * allocating chunks in some corner cases until we force to set
9399 * it to be readonly.
9402 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9404 min_allocable_bytes = SZ_1M;
9406 min_allocable_bytes = 0;
9408 spin_lock(&sinfo->lock);
9409 spin_lock(&cache->lock);
9417 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9418 cache->bytes_super - btrfs_block_group_used(&cache->item);
9420 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9421 min_allocable_bytes <= sinfo->total_bytes) {
9422 sinfo->bytes_readonly += num_bytes;
9424 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9428 spin_unlock(&cache->lock);
9429 spin_unlock(&sinfo->lock);
9433 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9436 struct btrfs_fs_info *fs_info = cache->fs_info;
9437 struct btrfs_trans_handle *trans;
9442 trans = btrfs_join_transaction(fs_info->extent_root);
9444 return PTR_ERR(trans);
9447 * we're not allowed to set block groups readonly after the dirty
9448 * block groups cache has started writing. If it already started,
9449 * back off and let this transaction commit
9451 mutex_lock(&fs_info->ro_block_group_mutex);
9452 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9453 u64 transid = trans->transid;
9455 mutex_unlock(&fs_info->ro_block_group_mutex);
9456 btrfs_end_transaction(trans);
9458 ret = btrfs_wait_for_commit(fs_info, transid);
9465 * if we are changing raid levels, try to allocate a corresponding
9466 * block group with the new raid level.
9468 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9469 if (alloc_flags != cache->flags) {
9470 ret = do_chunk_alloc(trans, alloc_flags,
9473 * ENOSPC is allowed here, we may have enough space
9474 * already allocated at the new raid level to
9483 ret = inc_block_group_ro(cache, 0);
9486 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9487 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9490 ret = inc_block_group_ro(cache, 0);
9492 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9493 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9494 mutex_lock(&fs_info->chunk_mutex);
9495 check_system_chunk(trans, alloc_flags);
9496 mutex_unlock(&fs_info->chunk_mutex);
9498 mutex_unlock(&fs_info->ro_block_group_mutex);
9500 btrfs_end_transaction(trans);
9504 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9506 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9508 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9512 * helper to account the unused space of all the readonly block group in the
9513 * space_info. takes mirrors into account.
9515 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9517 struct btrfs_block_group_cache *block_group;
9521 /* It's df, we don't care if it's racy */
9522 if (list_empty(&sinfo->ro_bgs))
9525 spin_lock(&sinfo->lock);
9526 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9527 spin_lock(&block_group->lock);
9529 if (!block_group->ro) {
9530 spin_unlock(&block_group->lock);
9534 factor = btrfs_bg_type_to_factor(block_group->flags);
9535 free_bytes += (block_group->key.offset -
9536 btrfs_block_group_used(&block_group->item)) *
9539 spin_unlock(&block_group->lock);
9541 spin_unlock(&sinfo->lock);
9546 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9548 struct btrfs_space_info *sinfo = cache->space_info;
9553 spin_lock(&sinfo->lock);
9554 spin_lock(&cache->lock);
9556 num_bytes = cache->key.offset - cache->reserved -
9557 cache->pinned - cache->bytes_super -
9558 btrfs_block_group_used(&cache->item);
9559 sinfo->bytes_readonly -= num_bytes;
9560 list_del_init(&cache->ro_list);
9562 spin_unlock(&cache->lock);
9563 spin_unlock(&sinfo->lock);
9567 * checks to see if its even possible to relocate this block group.
9569 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9570 * ok to go ahead and try.
9572 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9574 struct btrfs_root *root = fs_info->extent_root;
9575 struct btrfs_block_group_cache *block_group;
9576 struct btrfs_space_info *space_info;
9577 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9578 struct btrfs_device *device;
9579 struct btrfs_trans_handle *trans;
9589 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9591 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9593 /* odd, couldn't find the block group, leave it alone */
9597 "can't find block group for bytenr %llu",
9602 min_free = btrfs_block_group_used(&block_group->item);
9604 /* no bytes used, we're good */
9608 space_info = block_group->space_info;
9609 spin_lock(&space_info->lock);
9611 full = space_info->full;
9614 * if this is the last block group we have in this space, we can't
9615 * relocate it unless we're able to allocate a new chunk below.
9617 * Otherwise, we need to make sure we have room in the space to handle
9618 * all of the extents from this block group. If we can, we're good
9620 if ((space_info->total_bytes != block_group->key.offset) &&
9621 (btrfs_space_info_used(space_info, false) + min_free <
9622 space_info->total_bytes)) {
9623 spin_unlock(&space_info->lock);
9626 spin_unlock(&space_info->lock);
9629 * ok we don't have enough space, but maybe we have free space on our
9630 * devices to allocate new chunks for relocation, so loop through our
9631 * alloc devices and guess if we have enough space. if this block
9632 * group is going to be restriped, run checks against the target
9633 * profile instead of the current one.
9645 target = get_restripe_target(fs_info, block_group->flags);
9647 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9650 * this is just a balance, so if we were marked as full
9651 * we know there is no space for a new chunk
9656 "no space to alloc new chunk for block group %llu",
9657 block_group->key.objectid);
9661 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9664 if (index == BTRFS_RAID_RAID10) {
9668 } else if (index == BTRFS_RAID_RAID1) {
9670 } else if (index == BTRFS_RAID_DUP) {
9673 } else if (index == BTRFS_RAID_RAID0) {
9674 dev_min = fs_devices->rw_devices;
9675 min_free = div64_u64(min_free, dev_min);
9678 /* We need to do this so that we can look at pending chunks */
9679 trans = btrfs_join_transaction(root);
9680 if (IS_ERR(trans)) {
9681 ret = PTR_ERR(trans);
9685 mutex_lock(&fs_info->chunk_mutex);
9686 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9690 * check to make sure we can actually find a chunk with enough
9691 * space to fit our block group in.
9693 if (device->total_bytes > device->bytes_used + min_free &&
9694 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9695 ret = find_free_dev_extent(trans, device, min_free,
9700 if (dev_nr >= dev_min)
9706 if (debug && ret == -1)
9708 "no space to allocate a new chunk for block group %llu",
9709 block_group->key.objectid);
9710 mutex_unlock(&fs_info->chunk_mutex);
9711 btrfs_end_transaction(trans);
9713 btrfs_put_block_group(block_group);
9717 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9718 struct btrfs_path *path,
9719 struct btrfs_key *key)
9721 struct btrfs_root *root = fs_info->extent_root;
9723 struct btrfs_key found_key;
9724 struct extent_buffer *leaf;
9725 struct btrfs_block_group_item bg;
9729 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9734 slot = path->slots[0];
9735 leaf = path->nodes[0];
9736 if (slot >= btrfs_header_nritems(leaf)) {
9737 ret = btrfs_next_leaf(root, path);
9744 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9746 if (found_key.objectid >= key->objectid &&
9747 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9748 struct extent_map_tree *em_tree;
9749 struct extent_map *em;
9751 em_tree = &root->fs_info->mapping_tree.map_tree;
9752 read_lock(&em_tree->lock);
9753 em = lookup_extent_mapping(em_tree, found_key.objectid,
9755 read_unlock(&em_tree->lock);
9758 "logical %llu len %llu found bg but no related chunk",
9759 found_key.objectid, found_key.offset);
9761 } else if (em->start != found_key.objectid ||
9762 em->len != found_key.offset) {
9764 "block group %llu len %llu mismatch with chunk %llu len %llu",
9765 found_key.objectid, found_key.offset,
9766 em->start, em->len);
9769 read_extent_buffer(leaf, &bg,
9770 btrfs_item_ptr_offset(leaf, slot),
9772 flags = btrfs_block_group_flags(&bg) &
9773 BTRFS_BLOCK_GROUP_TYPE_MASK;
9775 if (flags != (em->map_lookup->type &
9776 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9778 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9780 found_key.offset, flags,
9781 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9782 em->map_lookup->type));
9788 free_extent_map(em);
9797 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9799 struct btrfs_block_group_cache *block_group;
9803 struct inode *inode;
9805 block_group = btrfs_lookup_first_block_group(info, last);
9806 while (block_group) {
9807 wait_block_group_cache_done(block_group);
9808 spin_lock(&block_group->lock);
9809 if (block_group->iref)
9811 spin_unlock(&block_group->lock);
9812 block_group = next_block_group(info, block_group);
9821 inode = block_group->inode;
9822 block_group->iref = 0;
9823 block_group->inode = NULL;
9824 spin_unlock(&block_group->lock);
9825 ASSERT(block_group->io_ctl.inode == NULL);
9827 last = block_group->key.objectid + block_group->key.offset;
9828 btrfs_put_block_group(block_group);
9833 * Must be called only after stopping all workers, since we could have block
9834 * group caching kthreads running, and therefore they could race with us if we
9835 * freed the block groups before stopping them.
9837 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9839 struct btrfs_block_group_cache *block_group;
9840 struct btrfs_space_info *space_info;
9841 struct btrfs_caching_control *caching_ctl;
9844 down_write(&info->commit_root_sem);
9845 while (!list_empty(&info->caching_block_groups)) {
9846 caching_ctl = list_entry(info->caching_block_groups.next,
9847 struct btrfs_caching_control, list);
9848 list_del(&caching_ctl->list);
9849 put_caching_control(caching_ctl);
9851 up_write(&info->commit_root_sem);
9853 spin_lock(&info->unused_bgs_lock);
9854 while (!list_empty(&info->unused_bgs)) {
9855 block_group = list_first_entry(&info->unused_bgs,
9856 struct btrfs_block_group_cache,
9858 list_del_init(&block_group->bg_list);
9859 btrfs_put_block_group(block_group);
9861 spin_unlock(&info->unused_bgs_lock);
9863 spin_lock(&info->block_group_cache_lock);
9864 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9865 block_group = rb_entry(n, struct btrfs_block_group_cache,
9867 rb_erase(&block_group->cache_node,
9868 &info->block_group_cache_tree);
9869 RB_CLEAR_NODE(&block_group->cache_node);
9870 spin_unlock(&info->block_group_cache_lock);
9872 down_write(&block_group->space_info->groups_sem);
9873 list_del(&block_group->list);
9874 up_write(&block_group->space_info->groups_sem);
9877 * We haven't cached this block group, which means we could
9878 * possibly have excluded extents on this block group.
9880 if (block_group->cached == BTRFS_CACHE_NO ||
9881 block_group->cached == BTRFS_CACHE_ERROR)
9882 free_excluded_extents(block_group);
9884 btrfs_remove_free_space_cache(block_group);
9885 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9886 ASSERT(list_empty(&block_group->dirty_list));
9887 ASSERT(list_empty(&block_group->io_list));
9888 ASSERT(list_empty(&block_group->bg_list));
9889 ASSERT(atomic_read(&block_group->count) == 1);
9890 btrfs_put_block_group(block_group);
9892 spin_lock(&info->block_group_cache_lock);
9894 spin_unlock(&info->block_group_cache_lock);
9896 /* now that all the block groups are freed, go through and
9897 * free all the space_info structs. This is only called during
9898 * the final stages of unmount, and so we know nobody is
9899 * using them. We call synchronize_rcu() once before we start,
9900 * just to be on the safe side.
9904 release_global_block_rsv(info);
9906 while (!list_empty(&info->space_info)) {
9909 space_info = list_entry(info->space_info.next,
9910 struct btrfs_space_info,
9914 * Do not hide this behind enospc_debug, this is actually
9915 * important and indicates a real bug if this happens.
9917 if (WARN_ON(space_info->bytes_pinned > 0 ||
9918 space_info->bytes_reserved > 0 ||
9919 space_info->bytes_may_use > 0))
9920 dump_space_info(info, space_info, 0, 0);
9921 list_del(&space_info->list);
9922 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9923 struct kobject *kobj;
9924 kobj = space_info->block_group_kobjs[i];
9925 space_info->block_group_kobjs[i] = NULL;
9931 kobject_del(&space_info->kobj);
9932 kobject_put(&space_info->kobj);
9937 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9938 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9940 struct btrfs_space_info *space_info;
9941 struct raid_kobject *rkobj;
9946 spin_lock(&fs_info->pending_raid_kobjs_lock);
9947 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9948 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9950 list_for_each_entry(rkobj, &list, list) {
9951 space_info = __find_space_info(fs_info, rkobj->flags);
9952 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9954 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9955 "%s", get_raid_name(index));
9957 kobject_put(&rkobj->kobj);
9963 "failed to add kobject for block cache, ignoring");
9966 static void link_block_group(struct btrfs_block_group_cache *cache)
9968 struct btrfs_space_info *space_info = cache->space_info;
9969 struct btrfs_fs_info *fs_info = cache->fs_info;
9970 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9973 down_write(&space_info->groups_sem);
9974 if (list_empty(&space_info->block_groups[index]))
9976 list_add_tail(&cache->list, &space_info->block_groups[index]);
9977 up_write(&space_info->groups_sem);
9980 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9982 btrfs_warn(cache->fs_info,
9983 "couldn't alloc memory for raid level kobject");
9986 rkobj->flags = cache->flags;
9987 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9989 spin_lock(&fs_info->pending_raid_kobjs_lock);
9990 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9991 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9992 space_info->block_group_kobjs[index] = &rkobj->kobj;
9996 static struct btrfs_block_group_cache *
9997 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9998 u64 start, u64 size)
10000 struct btrfs_block_group_cache *cache;
10002 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10006 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10008 if (!cache->free_space_ctl) {
10013 cache->key.objectid = start;
10014 cache->key.offset = size;
10015 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10017 cache->fs_info = fs_info;
10018 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10019 set_free_space_tree_thresholds(cache);
10021 atomic_set(&cache->count, 1);
10022 spin_lock_init(&cache->lock);
10023 init_rwsem(&cache->data_rwsem);
10024 INIT_LIST_HEAD(&cache->list);
10025 INIT_LIST_HEAD(&cache->cluster_list);
10026 INIT_LIST_HEAD(&cache->bg_list);
10027 INIT_LIST_HEAD(&cache->ro_list);
10028 INIT_LIST_HEAD(&cache->dirty_list);
10029 INIT_LIST_HEAD(&cache->io_list);
10030 btrfs_init_free_space_ctl(cache);
10031 atomic_set(&cache->trimming, 0);
10032 mutex_init(&cache->free_space_lock);
10033 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10040 * Iterate all chunks and verify that each of them has the corresponding block
10043 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10045 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10046 struct extent_map *em;
10047 struct btrfs_block_group_cache *bg;
10052 read_lock(&map_tree->map_tree.lock);
10054 * lookup_extent_mapping will return the first extent map
10055 * intersecting the range, so setting @len to 1 is enough to
10056 * get the first chunk.
10058 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10059 read_unlock(&map_tree->map_tree.lock);
10063 bg = btrfs_lookup_block_group(fs_info, em->start);
10066 "chunk start=%llu len=%llu doesn't have corresponding block group",
10067 em->start, em->len);
10069 free_extent_map(em);
10072 if (bg->key.objectid != em->start ||
10073 bg->key.offset != em->len ||
10074 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10075 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10077 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10078 em->start, em->len,
10079 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10080 bg->key.objectid, bg->key.offset,
10081 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10083 free_extent_map(em);
10084 btrfs_put_block_group(bg);
10087 start = em->start + em->len;
10088 free_extent_map(em);
10089 btrfs_put_block_group(bg);
10094 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10096 struct btrfs_path *path;
10098 struct btrfs_block_group_cache *cache;
10099 struct btrfs_space_info *space_info;
10100 struct btrfs_key key;
10101 struct btrfs_key found_key;
10102 struct extent_buffer *leaf;
10103 int need_clear = 0;
10108 feature = btrfs_super_incompat_flags(info->super_copy);
10109 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10113 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10114 path = btrfs_alloc_path();
10117 path->reada = READA_FORWARD;
10119 cache_gen = btrfs_super_cache_generation(info->super_copy);
10120 if (btrfs_test_opt(info, SPACE_CACHE) &&
10121 btrfs_super_generation(info->super_copy) != cache_gen)
10123 if (btrfs_test_opt(info, CLEAR_CACHE))
10127 ret = find_first_block_group(info, path, &key);
10133 leaf = path->nodes[0];
10134 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10136 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10145 * When we mount with old space cache, we need to
10146 * set BTRFS_DC_CLEAR and set dirty flag.
10148 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10149 * truncate the old free space cache inode and
10151 * b) Setting 'dirty flag' makes sure that we flush
10152 * the new space cache info onto disk.
10154 if (btrfs_test_opt(info, SPACE_CACHE))
10155 cache->disk_cache_state = BTRFS_DC_CLEAR;
10158 read_extent_buffer(leaf, &cache->item,
10159 btrfs_item_ptr_offset(leaf, path->slots[0]),
10160 sizeof(cache->item));
10161 cache->flags = btrfs_block_group_flags(&cache->item);
10163 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10164 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10166 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10167 cache->key.objectid);
10172 key.objectid = found_key.objectid + found_key.offset;
10173 btrfs_release_path(path);
10176 * We need to exclude the super stripes now so that the space
10177 * info has super bytes accounted for, otherwise we'll think
10178 * we have more space than we actually do.
10180 ret = exclude_super_stripes(cache);
10183 * We may have excluded something, so call this just in
10186 free_excluded_extents(cache);
10187 btrfs_put_block_group(cache);
10192 * check for two cases, either we are full, and therefore
10193 * don't need to bother with the caching work since we won't
10194 * find any space, or we are empty, and we can just add all
10195 * the space in and be done with it. This saves us _alot_ of
10196 * time, particularly in the full case.
10198 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10199 cache->last_byte_to_unpin = (u64)-1;
10200 cache->cached = BTRFS_CACHE_FINISHED;
10201 free_excluded_extents(cache);
10202 } else if (btrfs_block_group_used(&cache->item) == 0) {
10203 cache->last_byte_to_unpin = (u64)-1;
10204 cache->cached = BTRFS_CACHE_FINISHED;
10205 add_new_free_space(cache, found_key.objectid,
10206 found_key.objectid +
10208 free_excluded_extents(cache);
10211 ret = btrfs_add_block_group_cache(info, cache);
10213 btrfs_remove_free_space_cache(cache);
10214 btrfs_put_block_group(cache);
10218 trace_btrfs_add_block_group(info, cache, 0);
10219 update_space_info(info, cache->flags, found_key.offset,
10220 btrfs_block_group_used(&cache->item),
10221 cache->bytes_super, &space_info);
10223 cache->space_info = space_info;
10225 link_block_group(cache);
10227 set_avail_alloc_bits(info, cache->flags);
10228 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10229 inc_block_group_ro(cache, 1);
10230 } else if (btrfs_block_group_used(&cache->item) == 0) {
10231 ASSERT(list_empty(&cache->bg_list));
10232 btrfs_mark_bg_unused(cache);
10236 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10237 if (!(get_alloc_profile(info, space_info->flags) &
10238 (BTRFS_BLOCK_GROUP_RAID10 |
10239 BTRFS_BLOCK_GROUP_RAID1 |
10240 BTRFS_BLOCK_GROUP_RAID5 |
10241 BTRFS_BLOCK_GROUP_RAID6 |
10242 BTRFS_BLOCK_GROUP_DUP)))
10245 * avoid allocating from un-mirrored block group if there are
10246 * mirrored block groups.
10248 list_for_each_entry(cache,
10249 &space_info->block_groups[BTRFS_RAID_RAID0],
10251 inc_block_group_ro(cache, 1);
10252 list_for_each_entry(cache,
10253 &space_info->block_groups[BTRFS_RAID_SINGLE],
10255 inc_block_group_ro(cache, 1);
10258 btrfs_add_raid_kobjects(info);
10259 init_global_block_rsv(info);
10260 ret = check_chunk_block_group_mappings(info);
10262 btrfs_free_path(path);
10266 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10268 struct btrfs_fs_info *fs_info = trans->fs_info;
10269 struct btrfs_block_group_cache *block_group;
10270 struct btrfs_root *extent_root = fs_info->extent_root;
10271 struct btrfs_block_group_item item;
10272 struct btrfs_key key;
10275 if (!trans->can_flush_pending_bgs)
10278 while (!list_empty(&trans->new_bgs)) {
10279 block_group = list_first_entry(&trans->new_bgs,
10280 struct btrfs_block_group_cache,
10285 spin_lock(&block_group->lock);
10286 memcpy(&item, &block_group->item, sizeof(item));
10287 memcpy(&key, &block_group->key, sizeof(key));
10288 spin_unlock(&block_group->lock);
10290 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10293 btrfs_abort_transaction(trans, ret);
10294 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10296 btrfs_abort_transaction(trans, ret);
10297 add_block_group_free_space(trans, block_group);
10298 /* already aborted the transaction if it failed. */
10300 list_del_init(&block_group->bg_list);
10302 btrfs_trans_release_chunk_metadata(trans);
10305 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10306 u64 type, u64 chunk_offset, u64 size)
10308 struct btrfs_fs_info *fs_info = trans->fs_info;
10309 struct btrfs_block_group_cache *cache;
10312 btrfs_set_log_full_commit(fs_info, trans);
10314 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10318 btrfs_set_block_group_used(&cache->item, bytes_used);
10319 btrfs_set_block_group_chunk_objectid(&cache->item,
10320 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10321 btrfs_set_block_group_flags(&cache->item, type);
10323 cache->flags = type;
10324 cache->last_byte_to_unpin = (u64)-1;
10325 cache->cached = BTRFS_CACHE_FINISHED;
10326 cache->needs_free_space = 1;
10327 ret = exclude_super_stripes(cache);
10330 * We may have excluded something, so call this just in
10333 free_excluded_extents(cache);
10334 btrfs_put_block_group(cache);
10338 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10340 free_excluded_extents(cache);
10342 #ifdef CONFIG_BTRFS_DEBUG
10343 if (btrfs_should_fragment_free_space(cache)) {
10344 u64 new_bytes_used = size - bytes_used;
10346 bytes_used += new_bytes_used >> 1;
10347 fragment_free_space(cache);
10351 * Ensure the corresponding space_info object is created and
10352 * assigned to our block group. We want our bg to be added to the rbtree
10353 * with its ->space_info set.
10355 cache->space_info = __find_space_info(fs_info, cache->flags);
10356 ASSERT(cache->space_info);
10358 ret = btrfs_add_block_group_cache(fs_info, cache);
10360 btrfs_remove_free_space_cache(cache);
10361 btrfs_put_block_group(cache);
10366 * Now that our block group has its ->space_info set and is inserted in
10367 * the rbtree, update the space info's counters.
10369 trace_btrfs_add_block_group(fs_info, cache, 1);
10370 update_space_info(fs_info, cache->flags, size, bytes_used,
10371 cache->bytes_super, &cache->space_info);
10372 update_global_block_rsv(fs_info);
10374 link_block_group(cache);
10376 list_add_tail(&cache->bg_list, &trans->new_bgs);
10378 set_avail_alloc_bits(fs_info, type);
10382 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10384 u64 extra_flags = chunk_to_extended(flags) &
10385 BTRFS_EXTENDED_PROFILE_MASK;
10387 write_seqlock(&fs_info->profiles_lock);
10388 if (flags & BTRFS_BLOCK_GROUP_DATA)
10389 fs_info->avail_data_alloc_bits &= ~extra_flags;
10390 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10391 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10392 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10393 fs_info->avail_system_alloc_bits &= ~extra_flags;
10394 write_sequnlock(&fs_info->profiles_lock);
10397 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10398 u64 group_start, struct extent_map *em)
10400 struct btrfs_fs_info *fs_info = trans->fs_info;
10401 struct btrfs_root *root = fs_info->extent_root;
10402 struct btrfs_path *path;
10403 struct btrfs_block_group_cache *block_group;
10404 struct btrfs_free_cluster *cluster;
10405 struct btrfs_root *tree_root = fs_info->tree_root;
10406 struct btrfs_key key;
10407 struct inode *inode;
10408 struct kobject *kobj = NULL;
10412 struct btrfs_caching_control *caching_ctl = NULL;
10415 block_group = btrfs_lookup_block_group(fs_info, group_start);
10416 BUG_ON(!block_group);
10417 BUG_ON(!block_group->ro);
10419 trace_btrfs_remove_block_group(block_group);
10421 * Free the reserved super bytes from this block group before
10424 free_excluded_extents(block_group);
10425 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10426 block_group->key.offset);
10428 memcpy(&key, &block_group->key, sizeof(key));
10429 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10430 factor = btrfs_bg_type_to_factor(block_group->flags);
10432 /* make sure this block group isn't part of an allocation cluster */
10433 cluster = &fs_info->data_alloc_cluster;
10434 spin_lock(&cluster->refill_lock);
10435 btrfs_return_cluster_to_free_space(block_group, cluster);
10436 spin_unlock(&cluster->refill_lock);
10439 * make sure this block group isn't part of a metadata
10440 * allocation cluster
10442 cluster = &fs_info->meta_alloc_cluster;
10443 spin_lock(&cluster->refill_lock);
10444 btrfs_return_cluster_to_free_space(block_group, cluster);
10445 spin_unlock(&cluster->refill_lock);
10447 path = btrfs_alloc_path();
10454 * get the inode first so any iput calls done for the io_list
10455 * aren't the final iput (no unlinks allowed now)
10457 inode = lookup_free_space_inode(fs_info, block_group, path);
10459 mutex_lock(&trans->transaction->cache_write_mutex);
10461 * make sure our free spache cache IO is done before remove the
10464 spin_lock(&trans->transaction->dirty_bgs_lock);
10465 if (!list_empty(&block_group->io_list)) {
10466 list_del_init(&block_group->io_list);
10468 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10470 spin_unlock(&trans->transaction->dirty_bgs_lock);
10471 btrfs_wait_cache_io(trans, block_group, path);
10472 btrfs_put_block_group(block_group);
10473 spin_lock(&trans->transaction->dirty_bgs_lock);
10476 if (!list_empty(&block_group->dirty_list)) {
10477 list_del_init(&block_group->dirty_list);
10478 btrfs_put_block_group(block_group);
10480 spin_unlock(&trans->transaction->dirty_bgs_lock);
10481 mutex_unlock(&trans->transaction->cache_write_mutex);
10483 if (!IS_ERR(inode)) {
10484 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10486 btrfs_add_delayed_iput(inode);
10489 clear_nlink(inode);
10490 /* One for the block groups ref */
10491 spin_lock(&block_group->lock);
10492 if (block_group->iref) {
10493 block_group->iref = 0;
10494 block_group->inode = NULL;
10495 spin_unlock(&block_group->lock);
10498 spin_unlock(&block_group->lock);
10500 /* One for our lookup ref */
10501 btrfs_add_delayed_iput(inode);
10504 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10505 key.offset = block_group->key.objectid;
10508 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10512 btrfs_release_path(path);
10514 ret = btrfs_del_item(trans, tree_root, path);
10517 btrfs_release_path(path);
10520 spin_lock(&fs_info->block_group_cache_lock);
10521 rb_erase(&block_group->cache_node,
10522 &fs_info->block_group_cache_tree);
10523 RB_CLEAR_NODE(&block_group->cache_node);
10525 if (fs_info->first_logical_byte == block_group->key.objectid)
10526 fs_info->first_logical_byte = (u64)-1;
10527 spin_unlock(&fs_info->block_group_cache_lock);
10529 down_write(&block_group->space_info->groups_sem);
10531 * we must use list_del_init so people can check to see if they
10532 * are still on the list after taking the semaphore
10534 list_del_init(&block_group->list);
10535 if (list_empty(&block_group->space_info->block_groups[index])) {
10536 kobj = block_group->space_info->block_group_kobjs[index];
10537 block_group->space_info->block_group_kobjs[index] = NULL;
10538 clear_avail_alloc_bits(fs_info, block_group->flags);
10540 up_write(&block_group->space_info->groups_sem);
10546 if (block_group->has_caching_ctl)
10547 caching_ctl = get_caching_control(block_group);
10548 if (block_group->cached == BTRFS_CACHE_STARTED)
10549 wait_block_group_cache_done(block_group);
10550 if (block_group->has_caching_ctl) {
10551 down_write(&fs_info->commit_root_sem);
10552 if (!caching_ctl) {
10553 struct btrfs_caching_control *ctl;
10555 list_for_each_entry(ctl,
10556 &fs_info->caching_block_groups, list)
10557 if (ctl->block_group == block_group) {
10559 refcount_inc(&caching_ctl->count);
10564 list_del_init(&caching_ctl->list);
10565 up_write(&fs_info->commit_root_sem);
10567 /* Once for the caching bgs list and once for us. */
10568 put_caching_control(caching_ctl);
10569 put_caching_control(caching_ctl);
10573 spin_lock(&trans->transaction->dirty_bgs_lock);
10574 if (!list_empty(&block_group->dirty_list)) {
10577 if (!list_empty(&block_group->io_list)) {
10580 spin_unlock(&trans->transaction->dirty_bgs_lock);
10581 btrfs_remove_free_space_cache(block_group);
10583 spin_lock(&block_group->space_info->lock);
10584 list_del_init(&block_group->ro_list);
10586 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10587 WARN_ON(block_group->space_info->total_bytes
10588 < block_group->key.offset);
10589 WARN_ON(block_group->space_info->bytes_readonly
10590 < block_group->key.offset);
10591 WARN_ON(block_group->space_info->disk_total
10592 < block_group->key.offset * factor);
10594 block_group->space_info->total_bytes -= block_group->key.offset;
10595 block_group->space_info->bytes_readonly -= block_group->key.offset;
10596 block_group->space_info->disk_total -= block_group->key.offset * factor;
10598 spin_unlock(&block_group->space_info->lock);
10600 memcpy(&key, &block_group->key, sizeof(key));
10602 mutex_lock(&fs_info->chunk_mutex);
10603 if (!list_empty(&em->list)) {
10604 /* We're in the transaction->pending_chunks list. */
10605 free_extent_map(em);
10607 spin_lock(&block_group->lock);
10608 block_group->removed = 1;
10610 * At this point trimming can't start on this block group, because we
10611 * removed the block group from the tree fs_info->block_group_cache_tree
10612 * so no one can't find it anymore and even if someone already got this
10613 * block group before we removed it from the rbtree, they have already
10614 * incremented block_group->trimming - if they didn't, they won't find
10615 * any free space entries because we already removed them all when we
10616 * called btrfs_remove_free_space_cache().
10618 * And we must not remove the extent map from the fs_info->mapping_tree
10619 * to prevent the same logical address range and physical device space
10620 * ranges from being reused for a new block group. This is because our
10621 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10622 * completely transactionless, so while it is trimming a range the
10623 * currently running transaction might finish and a new one start,
10624 * allowing for new block groups to be created that can reuse the same
10625 * physical device locations unless we take this special care.
10627 * There may also be an implicit trim operation if the file system
10628 * is mounted with -odiscard. The same protections must remain
10629 * in place until the extents have been discarded completely when
10630 * the transaction commit has completed.
10632 remove_em = (atomic_read(&block_group->trimming) == 0);
10634 * Make sure a trimmer task always sees the em in the pinned_chunks list
10635 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10636 * before checking block_group->removed).
10640 * Our em might be in trans->transaction->pending_chunks which
10641 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10642 * and so is the fs_info->pinned_chunks list.
10644 * So at this point we must be holding the chunk_mutex to avoid
10645 * any races with chunk allocation (more specifically at
10646 * volumes.c:contains_pending_extent()), to ensure it always
10647 * sees the em, either in the pending_chunks list or in the
10648 * pinned_chunks list.
10650 list_move_tail(&em->list, &fs_info->pinned_chunks);
10652 spin_unlock(&block_group->lock);
10655 struct extent_map_tree *em_tree;
10657 em_tree = &fs_info->mapping_tree.map_tree;
10658 write_lock(&em_tree->lock);
10660 * The em might be in the pending_chunks list, so make sure the
10661 * chunk mutex is locked, since remove_extent_mapping() will
10662 * delete us from that list.
10664 remove_extent_mapping(em_tree, em);
10665 write_unlock(&em_tree->lock);
10666 /* once for the tree */
10667 free_extent_map(em);
10670 mutex_unlock(&fs_info->chunk_mutex);
10672 ret = remove_block_group_free_space(trans, block_group);
10676 btrfs_put_block_group(block_group);
10677 btrfs_put_block_group(block_group);
10679 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10685 ret = btrfs_del_item(trans, root, path);
10687 btrfs_free_path(path);
10691 struct btrfs_trans_handle *
10692 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10693 const u64 chunk_offset)
10695 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10696 struct extent_map *em;
10697 struct map_lookup *map;
10698 unsigned int num_items;
10700 read_lock(&em_tree->lock);
10701 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10702 read_unlock(&em_tree->lock);
10703 ASSERT(em && em->start == chunk_offset);
10706 * We need to reserve 3 + N units from the metadata space info in order
10707 * to remove a block group (done at btrfs_remove_chunk() and at
10708 * btrfs_remove_block_group()), which are used for:
10710 * 1 unit for adding the free space inode's orphan (located in the tree
10712 * 1 unit for deleting the block group item (located in the extent
10714 * 1 unit for deleting the free space item (located in tree of tree
10716 * N units for deleting N device extent items corresponding to each
10717 * stripe (located in the device tree).
10719 * In order to remove a block group we also need to reserve units in the
10720 * system space info in order to update the chunk tree (update one or
10721 * more device items and remove one chunk item), but this is done at
10722 * btrfs_remove_chunk() through a call to check_system_chunk().
10724 map = em->map_lookup;
10725 num_items = 3 + map->num_stripes;
10726 free_extent_map(em);
10728 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10733 * Process the unused_bgs list and remove any that don't have any allocated
10734 * space inside of them.
10736 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10738 struct btrfs_block_group_cache *block_group;
10739 struct btrfs_space_info *space_info;
10740 struct btrfs_trans_handle *trans;
10743 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10746 spin_lock(&fs_info->unused_bgs_lock);
10747 while (!list_empty(&fs_info->unused_bgs)) {
10751 block_group = list_first_entry(&fs_info->unused_bgs,
10752 struct btrfs_block_group_cache,
10754 list_del_init(&block_group->bg_list);
10756 space_info = block_group->space_info;
10758 if (ret || btrfs_mixed_space_info(space_info)) {
10759 btrfs_put_block_group(block_group);
10762 spin_unlock(&fs_info->unused_bgs_lock);
10764 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10766 /* Don't want to race with allocators so take the groups_sem */
10767 down_write(&space_info->groups_sem);
10768 spin_lock(&block_group->lock);
10769 if (block_group->reserved || block_group->pinned ||
10770 btrfs_block_group_used(&block_group->item) ||
10772 list_is_singular(&block_group->list)) {
10774 * We want to bail if we made new allocations or have
10775 * outstanding allocations in this block group. We do
10776 * the ro check in case balance is currently acting on
10777 * this block group.
10779 trace_btrfs_skip_unused_block_group(block_group);
10780 spin_unlock(&block_group->lock);
10781 up_write(&space_info->groups_sem);
10784 spin_unlock(&block_group->lock);
10786 /* We don't want to force the issue, only flip if it's ok. */
10787 ret = inc_block_group_ro(block_group, 0);
10788 up_write(&space_info->groups_sem);
10795 * Want to do this before we do anything else so we can recover
10796 * properly if we fail to join the transaction.
10798 trans = btrfs_start_trans_remove_block_group(fs_info,
10799 block_group->key.objectid);
10800 if (IS_ERR(trans)) {
10801 btrfs_dec_block_group_ro(block_group);
10802 ret = PTR_ERR(trans);
10807 * We could have pending pinned extents for this block group,
10808 * just delete them, we don't care about them anymore.
10810 start = block_group->key.objectid;
10811 end = start + block_group->key.offset - 1;
10813 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10814 * btrfs_finish_extent_commit(). If we are at transaction N,
10815 * another task might be running finish_extent_commit() for the
10816 * previous transaction N - 1, and have seen a range belonging
10817 * to the block group in freed_extents[] before we were able to
10818 * clear the whole block group range from freed_extents[]. This
10819 * means that task can lookup for the block group after we
10820 * unpinned it from freed_extents[] and removed it, leading to
10821 * a BUG_ON() at btrfs_unpin_extent_range().
10823 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10824 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10827 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10828 btrfs_dec_block_group_ro(block_group);
10831 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10834 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10835 btrfs_dec_block_group_ro(block_group);
10838 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10840 /* Reset pinned so btrfs_put_block_group doesn't complain */
10841 spin_lock(&space_info->lock);
10842 spin_lock(&block_group->lock);
10844 update_bytes_pinned(space_info, -block_group->pinned);
10845 space_info->bytes_readonly += block_group->pinned;
10846 percpu_counter_add_batch(&space_info->total_bytes_pinned,
10847 -block_group->pinned,
10848 BTRFS_TOTAL_BYTES_PINNED_BATCH);
10849 block_group->pinned = 0;
10851 spin_unlock(&block_group->lock);
10852 spin_unlock(&space_info->lock);
10854 /* DISCARD can flip during remount */
10855 trimming = btrfs_test_opt(fs_info, DISCARD);
10857 /* Implicit trim during transaction commit. */
10859 btrfs_get_block_group_trimming(block_group);
10862 * Btrfs_remove_chunk will abort the transaction if things go
10865 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
10869 btrfs_put_block_group_trimming(block_group);
10874 * If we're not mounted with -odiscard, we can just forget
10875 * about this block group. Otherwise we'll need to wait
10876 * until transaction commit to do the actual discard.
10879 spin_lock(&fs_info->unused_bgs_lock);
10881 * A concurrent scrub might have added us to the list
10882 * fs_info->unused_bgs, so use a list_move operation
10883 * to add the block group to the deleted_bgs list.
10885 list_move(&block_group->bg_list,
10886 &trans->transaction->deleted_bgs);
10887 spin_unlock(&fs_info->unused_bgs_lock);
10888 btrfs_get_block_group(block_group);
10891 btrfs_end_transaction(trans);
10893 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10894 btrfs_put_block_group(block_group);
10895 spin_lock(&fs_info->unused_bgs_lock);
10897 spin_unlock(&fs_info->unused_bgs_lock);
10900 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10902 struct btrfs_super_block *disk_super;
10908 disk_super = fs_info->super_copy;
10909 if (!btrfs_super_root(disk_super))
10912 features = btrfs_super_incompat_flags(disk_super);
10913 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10916 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10917 ret = create_space_info(fs_info, flags);
10922 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10923 ret = create_space_info(fs_info, flags);
10925 flags = BTRFS_BLOCK_GROUP_METADATA;
10926 ret = create_space_info(fs_info, flags);
10930 flags = BTRFS_BLOCK_GROUP_DATA;
10931 ret = create_space_info(fs_info, flags);
10937 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10938 u64 start, u64 end)
10940 return unpin_extent_range(fs_info, start, end, false);
10944 * It used to be that old block groups would be left around forever.
10945 * Iterating over them would be enough to trim unused space. Since we
10946 * now automatically remove them, we also need to iterate over unallocated
10949 * We don't want a transaction for this since the discard may take a
10950 * substantial amount of time. We don't require that a transaction be
10951 * running, but we do need to take a running transaction into account
10952 * to ensure that we're not discarding chunks that were released or
10953 * allocated in the current transaction.
10955 * Holding the chunks lock will prevent other threads from allocating
10956 * or releasing chunks, but it won't prevent a running transaction
10957 * from committing and releasing the memory that the pending chunks
10958 * list head uses. For that, we need to take a reference to the
10959 * transaction and hold the commit root sem. We only need to hold
10960 * it while performing the free space search since we have already
10961 * held back allocations.
10963 static int btrfs_trim_free_extents(struct btrfs_device *device,
10964 u64 minlen, u64 *trimmed)
10966 u64 start = 0, len = 0;
10971 /* Discard not supported = nothing to do. */
10972 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
10975 /* Not writeable = nothing to do. */
10976 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10979 /* No free space = nothing to do. */
10980 if (device->total_bytes <= device->bytes_used)
10986 struct btrfs_fs_info *fs_info = device->fs_info;
10987 struct btrfs_transaction *trans;
10990 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10994 ret = down_read_killable(&fs_info->commit_root_sem);
10996 mutex_unlock(&fs_info->chunk_mutex);
11000 spin_lock(&fs_info->trans_lock);
11001 trans = fs_info->running_transaction;
11003 refcount_inc(&trans->use_count);
11004 spin_unlock(&fs_info->trans_lock);
11007 up_read(&fs_info->commit_root_sem);
11009 ret = find_free_dev_extent_start(trans, device, minlen, start,
11012 up_read(&fs_info->commit_root_sem);
11013 btrfs_put_transaction(trans);
11017 mutex_unlock(&fs_info->chunk_mutex);
11018 if (ret == -ENOSPC)
11023 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11024 mutex_unlock(&fs_info->chunk_mutex);
11032 if (fatal_signal_pending(current)) {
11033 ret = -ERESTARTSYS;
11044 * Trim the whole filesystem by:
11045 * 1) trimming the free space in each block group
11046 * 2) trimming the unallocated space on each device
11048 * This will also continue trimming even if a block group or device encounters
11049 * an error. The return value will be the last error, or 0 if nothing bad
11052 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11054 struct btrfs_block_group_cache *cache = NULL;
11055 struct btrfs_device *device;
11056 struct list_head *devices;
11062 u64 dev_failed = 0;
11067 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11068 for (; cache; cache = next_block_group(fs_info, cache)) {
11069 if (cache->key.objectid >= (range->start + range->len)) {
11070 btrfs_put_block_group(cache);
11074 start = max(range->start, cache->key.objectid);
11075 end = min(range->start + range->len,
11076 cache->key.objectid + cache->key.offset);
11078 if (end - start >= range->minlen) {
11079 if (!block_group_cache_done(cache)) {
11080 ret = cache_block_group(cache, 0);
11086 ret = wait_block_group_cache_done(cache);
11093 ret = btrfs_trim_block_group(cache,
11099 trimmed += group_trimmed;
11109 btrfs_warn(fs_info,
11110 "failed to trim %llu block group(s), last error %d",
11111 bg_failed, bg_ret);
11112 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11113 devices = &fs_info->fs_devices->devices;
11114 list_for_each_entry(device, devices, dev_list) {
11115 ret = btrfs_trim_free_extents(device, range->minlen,
11123 trimmed += group_trimmed;
11125 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11128 btrfs_warn(fs_info,
11129 "failed to trim %llu device(s), last error %d",
11130 dev_failed, dev_ret);
11131 range->len = trimmed;
11138 * btrfs_{start,end}_write_no_snapshotting() are similar to
11139 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11140 * data into the page cache through nocow before the subvolume is snapshoted,
11141 * but flush the data into disk after the snapshot creation, or to prevent
11142 * operations while snapshotting is ongoing and that cause the snapshot to be
11143 * inconsistent (writes followed by expanding truncates for example).
11145 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11147 percpu_counter_dec(&root->subv_writers->counter);
11148 cond_wake_up(&root->subv_writers->wait);
11151 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11153 if (atomic_read(&root->will_be_snapshotted))
11156 percpu_counter_inc(&root->subv_writers->counter);
11158 * Make sure counter is updated before we check for snapshot creation.
11161 if (atomic_read(&root->will_be_snapshotted)) {
11162 btrfs_end_write_no_snapshotting(root);
11168 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11173 ret = btrfs_start_write_no_snapshotting(root);
11176 wait_var_event(&root->will_be_snapshotted,
11177 !atomic_read(&root->will_be_snapshotted));
11181 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11183 struct btrfs_fs_info *fs_info = bg->fs_info;
11185 spin_lock(&fs_info->unused_bgs_lock);
11186 if (list_empty(&bg->bg_list)) {
11187 btrfs_get_block_group(bg);
11188 trace_btrfs_add_unused_block_group(bg);
11189 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11191 spin_unlock(&fs_info->unused_bgs_lock);