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,
54 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_delayed_ref_node *node, u64 parent,
56 u64 root_objectid, u64 owner_objectid,
57 u64 owner_offset, int refs_to_drop,
58 struct btrfs_delayed_extent_op *extra_op);
59 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
60 struct extent_buffer *leaf,
61 struct btrfs_extent_item *ei);
62 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
63 u64 parent, u64 root_objectid,
64 u64 flags, u64 owner, u64 offset,
65 struct btrfs_key *ins, int ref_mod);
66 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
67 struct btrfs_delayed_ref_node *node,
68 struct btrfs_delayed_extent_op *extent_op);
69 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
71 static int find_next_key(struct btrfs_path *path, int level,
72 struct btrfs_key *key);
73 static void dump_space_info(struct btrfs_fs_info *fs_info,
74 struct btrfs_space_info *info, u64 bytes,
75 int dump_block_groups);
76 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
78 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
79 struct btrfs_space_info *space_info,
81 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
82 struct btrfs_space_info *space_info,
86 block_group_cache_done(struct btrfs_block_group_cache *cache)
89 return cache->cached == BTRFS_CACHE_FINISHED ||
90 cache->cached == BTRFS_CACHE_ERROR;
93 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
95 return (cache->flags & bits) == bits;
98 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
100 atomic_inc(&cache->count);
103 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
105 if (atomic_dec_and_test(&cache->count)) {
106 WARN_ON(cache->pinned > 0);
107 WARN_ON(cache->reserved > 0);
110 * If not empty, someone is still holding mutex of
111 * full_stripe_lock, which can only be released by caller.
112 * And it will definitely cause use-after-free when caller
113 * tries to release full stripe lock.
115 * No better way to resolve, but only to warn.
117 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
118 kfree(cache->free_space_ctl);
124 * this adds the block group to the fs_info rb tree for the block group
127 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
128 struct btrfs_block_group_cache *block_group)
131 struct rb_node *parent = NULL;
132 struct btrfs_block_group_cache *cache;
134 spin_lock(&info->block_group_cache_lock);
135 p = &info->block_group_cache_tree.rb_node;
139 cache = rb_entry(parent, struct btrfs_block_group_cache,
141 if (block_group->key.objectid < cache->key.objectid) {
143 } else if (block_group->key.objectid > cache->key.objectid) {
146 spin_unlock(&info->block_group_cache_lock);
151 rb_link_node(&block_group->cache_node, parent, p);
152 rb_insert_color(&block_group->cache_node,
153 &info->block_group_cache_tree);
155 if (info->first_logical_byte > block_group->key.objectid)
156 info->first_logical_byte = block_group->key.objectid;
158 spin_unlock(&info->block_group_cache_lock);
164 * This will return the block group at or after bytenr if contains is 0, else
165 * it will return the block group that contains the bytenr
167 static struct btrfs_block_group_cache *
168 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
171 struct btrfs_block_group_cache *cache, *ret = NULL;
175 spin_lock(&info->block_group_cache_lock);
176 n = info->block_group_cache_tree.rb_node;
179 cache = rb_entry(n, struct btrfs_block_group_cache,
181 end = cache->key.objectid + cache->key.offset - 1;
182 start = cache->key.objectid;
184 if (bytenr < start) {
185 if (!contains && (!ret || start < ret->key.objectid))
188 } else if (bytenr > start) {
189 if (contains && bytenr <= end) {
200 btrfs_get_block_group(ret);
201 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
202 info->first_logical_byte = ret->key.objectid;
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE);
215 set_extent_bits(&fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE);
220 static void free_excluded_extents(struct btrfs_block_group_cache *cache)
222 struct btrfs_fs_info *fs_info = cache->fs_info;
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE);
230 clear_extent_bits(&fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE);
234 static int exclude_super_stripes(struct btrfs_block_group_cache *cache)
236 struct btrfs_fs_info *fs_info = cache->fs_info;
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(fs_info, cache->key.objectid,
251 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
252 bytenr = btrfs_sb_offset(i);
253 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
254 bytenr, &logical, &nr, &stripe_len);
261 if (logical[nr] > cache->key.objectid +
265 if (logical[nr] + stripe_len <= cache->key.objectid)
269 if (start < cache->key.objectid) {
270 start = cache->key.objectid;
271 len = (logical[nr] + stripe_len) - start;
273 len = min_t(u64, stripe_len,
274 cache->key.objectid +
275 cache->key.offset - start);
278 cache->bytes_super += len;
279 ret = add_excluded_extent(fs_info, start, len);
291 static struct btrfs_caching_control *
292 get_caching_control(struct btrfs_block_group_cache *cache)
294 struct btrfs_caching_control *ctl;
296 spin_lock(&cache->lock);
297 if (!cache->caching_ctl) {
298 spin_unlock(&cache->lock);
302 ctl = cache->caching_ctl;
303 refcount_inc(&ctl->count);
304 spin_unlock(&cache->lock);
308 static void put_caching_control(struct btrfs_caching_control *ctl)
310 if (refcount_dec_and_test(&ctl->count))
314 #ifdef CONFIG_BTRFS_DEBUG
315 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
317 struct btrfs_fs_info *fs_info = block_group->fs_info;
318 u64 start = block_group->key.objectid;
319 u64 len = block_group->key.offset;
320 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
321 fs_info->nodesize : fs_info->sectorsize;
322 u64 step = chunk << 1;
324 while (len > chunk) {
325 btrfs_remove_free_space(block_group, start, chunk);
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info = block_group->fs_info;
344 u64 extent_start, extent_end, size, total_added = 0;
347 while (start < end) {
348 ret = find_first_extent_bit(info->pinned_extents, start,
349 &extent_start, &extent_end,
350 EXTENT_DIRTY | EXTENT_UPTODATE,
355 if (extent_start <= start) {
356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
360 ret = btrfs_add_free_space(block_group, start,
362 BUG_ON(ret); /* -ENOMEM or logic error */
363 start = extent_end + 1;
372 ret = btrfs_add_free_space(block_group, start, size);
373 BUG_ON(ret); /* -ENOMEM or logic error */
379 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
381 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
382 struct btrfs_fs_info *fs_info = block_group->fs_info;
383 struct btrfs_root *extent_root = fs_info->extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
393 path = btrfs_alloc_path();
397 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
399 #ifdef CONFIG_BTRFS_DEBUG
401 * If we're fragmenting we don't want to make anybody think we can
402 * allocate from this block group until we've had a chance to fragment
405 if (btrfs_should_fragment_free_space(block_group))
409 * We don't want to deadlock with somebody trying to allocate a new
410 * extent for the extent root while also trying to search the extent
411 * root to add free space. So we skip locking and search the commit
412 * root, since its read-only
414 path->skip_locking = 1;
415 path->search_commit_root = 1;
416 path->reada = READA_FORWARD;
420 key.type = BTRFS_EXTENT_ITEM_KEY;
423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
431 if (btrfs_fs_closing(fs_info) > 1) {
436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
439 ret = find_next_key(path, 0, &key);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info->commit_root_sem)) {
446 caching_ctl->progress = last;
447 btrfs_release_path(path);
448 up_read(&fs_info->commit_root_sem);
449 mutex_unlock(&caching_ctl->mutex);
451 mutex_lock(&caching_ctl->mutex);
452 down_read(&fs_info->commit_root_sem);
456 ret = btrfs_next_leaf(extent_root, path);
461 leaf = path->nodes[0];
462 nritems = btrfs_header_nritems(leaf);
466 if (key.objectid < last) {
469 key.type = BTRFS_EXTENT_ITEM_KEY;
472 caching_ctl->progress = last;
473 btrfs_release_path(path);
477 if (key.objectid < block_group->key.objectid) {
482 if (key.objectid >= block_group->key.objectid +
483 block_group->key.offset)
486 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
487 key.type == BTRFS_METADATA_ITEM_KEY) {
488 total_found += add_new_free_space(block_group, last,
490 if (key.type == BTRFS_METADATA_ITEM_KEY)
491 last = key.objectid +
494 last = key.objectid + key.offset;
496 if (total_found > CACHING_CTL_WAKE_UP) {
499 wake_up(&caching_ctl->wait);
506 total_found += add_new_free_space(block_group, last,
507 block_group->key.objectid +
508 block_group->key.offset);
509 caching_ctl->progress = (u64)-1;
512 btrfs_free_path(path);
516 static noinline void caching_thread(struct btrfs_work *work)
518 struct btrfs_block_group_cache *block_group;
519 struct btrfs_fs_info *fs_info;
520 struct btrfs_caching_control *caching_ctl;
523 caching_ctl = container_of(work, struct btrfs_caching_control, work);
524 block_group = caching_ctl->block_group;
525 fs_info = block_group->fs_info;
527 mutex_lock(&caching_ctl->mutex);
528 down_read(&fs_info->commit_root_sem);
530 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
531 ret = load_free_space_tree(caching_ctl);
533 ret = load_extent_tree_free(caching_ctl);
535 spin_lock(&block_group->lock);
536 block_group->caching_ctl = NULL;
537 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
538 spin_unlock(&block_group->lock);
540 #ifdef CONFIG_BTRFS_DEBUG
541 if (btrfs_should_fragment_free_space(block_group)) {
544 spin_lock(&block_group->space_info->lock);
545 spin_lock(&block_group->lock);
546 bytes_used = block_group->key.offset -
547 btrfs_block_group_used(&block_group->item);
548 block_group->space_info->bytes_used += bytes_used >> 1;
549 spin_unlock(&block_group->lock);
550 spin_unlock(&block_group->space_info->lock);
551 fragment_free_space(block_group);
555 caching_ctl->progress = (u64)-1;
557 up_read(&fs_info->commit_root_sem);
558 free_excluded_extents(block_group);
559 mutex_unlock(&caching_ctl->mutex);
561 wake_up(&caching_ctl->wait);
563 put_caching_control(caching_ctl);
564 btrfs_put_block_group(block_group);
567 static int cache_block_group(struct btrfs_block_group_cache *cache,
571 struct btrfs_fs_info *fs_info = cache->fs_info;
572 struct btrfs_caching_control *caching_ctl;
575 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
579 INIT_LIST_HEAD(&caching_ctl->list);
580 mutex_init(&caching_ctl->mutex);
581 init_waitqueue_head(&caching_ctl->wait);
582 caching_ctl->block_group = cache;
583 caching_ctl->progress = cache->key.objectid;
584 refcount_set(&caching_ctl->count, 1);
585 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
586 caching_thread, NULL, NULL);
588 spin_lock(&cache->lock);
590 * This should be a rare occasion, but this could happen I think in the
591 * case where one thread starts to load the space cache info, and then
592 * some other thread starts a transaction commit which tries to do an
593 * allocation while the other thread is still loading the space cache
594 * info. The previous loop should have kept us from choosing this block
595 * group, but if we've moved to the state where we will wait on caching
596 * block groups we need to first check if we're doing a fast load here,
597 * so we can wait for it to finish, otherwise we could end up allocating
598 * from a block group who's cache gets evicted for one reason or
601 while (cache->cached == BTRFS_CACHE_FAST) {
602 struct btrfs_caching_control *ctl;
604 ctl = cache->caching_ctl;
605 refcount_inc(&ctl->count);
606 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
607 spin_unlock(&cache->lock);
611 finish_wait(&ctl->wait, &wait);
612 put_caching_control(ctl);
613 spin_lock(&cache->lock);
616 if (cache->cached != BTRFS_CACHE_NO) {
617 spin_unlock(&cache->lock);
621 WARN_ON(cache->caching_ctl);
622 cache->caching_ctl = caching_ctl;
623 cache->cached = BTRFS_CACHE_FAST;
624 spin_unlock(&cache->lock);
626 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
627 mutex_lock(&caching_ctl->mutex);
628 ret = load_free_space_cache(fs_info, cache);
630 spin_lock(&cache->lock);
632 cache->caching_ctl = NULL;
633 cache->cached = BTRFS_CACHE_FINISHED;
634 cache->last_byte_to_unpin = (u64)-1;
635 caching_ctl->progress = (u64)-1;
637 if (load_cache_only) {
638 cache->caching_ctl = NULL;
639 cache->cached = BTRFS_CACHE_NO;
641 cache->cached = BTRFS_CACHE_STARTED;
642 cache->has_caching_ctl = 1;
645 spin_unlock(&cache->lock);
646 #ifdef CONFIG_BTRFS_DEBUG
648 btrfs_should_fragment_free_space(cache)) {
651 spin_lock(&cache->space_info->lock);
652 spin_lock(&cache->lock);
653 bytes_used = cache->key.offset -
654 btrfs_block_group_used(&cache->item);
655 cache->space_info->bytes_used += bytes_used >> 1;
656 spin_unlock(&cache->lock);
657 spin_unlock(&cache->space_info->lock);
658 fragment_free_space(cache);
661 mutex_unlock(&caching_ctl->mutex);
663 wake_up(&caching_ctl->wait);
665 put_caching_control(caching_ctl);
666 free_excluded_extents(cache);
671 * We're either using the free space tree or no caching at all.
672 * Set cached to the appropriate value and wakeup any waiters.
674 spin_lock(&cache->lock);
675 if (load_cache_only) {
676 cache->caching_ctl = NULL;
677 cache->cached = BTRFS_CACHE_NO;
679 cache->cached = BTRFS_CACHE_STARTED;
680 cache->has_caching_ctl = 1;
682 spin_unlock(&cache->lock);
683 wake_up(&caching_ctl->wait);
686 if (load_cache_only) {
687 put_caching_control(caching_ctl);
691 down_write(&fs_info->commit_root_sem);
692 refcount_inc(&caching_ctl->count);
693 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
694 up_write(&fs_info->commit_root_sem);
696 btrfs_get_block_group(cache);
698 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
704 * return the block group that starts at or after bytenr
706 static struct btrfs_block_group_cache *
707 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
709 return block_group_cache_tree_search(info, bytenr, 0);
713 * return the block group that contains the given bytenr
715 struct btrfs_block_group_cache *btrfs_lookup_block_group(
716 struct btrfs_fs_info *info,
719 return block_group_cache_tree_search(info, bytenr, 1);
722 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
725 struct list_head *head = &info->space_info;
726 struct btrfs_space_info *found;
728 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
731 list_for_each_entry_rcu(found, head, list) {
732 if (found->flags & flags) {
741 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
742 bool metadata, u64 root_objectid)
744 struct btrfs_space_info *space_info;
748 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
749 flags = BTRFS_BLOCK_GROUP_SYSTEM;
751 flags = BTRFS_BLOCK_GROUP_METADATA;
753 flags = BTRFS_BLOCK_GROUP_DATA;
756 space_info = __find_space_info(fs_info, flags);
758 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
762 * after adding space to the filesystem, we need to clear the full flags
763 * on all the space infos.
765 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
767 struct list_head *head = &info->space_info;
768 struct btrfs_space_info *found;
771 list_for_each_entry_rcu(found, head, list)
776 /* simple helper to search for an existing data extent at a given offset */
777 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
780 struct btrfs_key key;
781 struct btrfs_path *path;
783 path = btrfs_alloc_path();
787 key.objectid = start;
789 key.type = BTRFS_EXTENT_ITEM_KEY;
790 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
791 btrfs_free_path(path);
796 * helper function to lookup reference count and flags of a tree block.
798 * the head node for delayed ref is used to store the sum of all the
799 * reference count modifications queued up in the rbtree. the head
800 * node may also store the extent flags to set. This way you can check
801 * to see what the reference count and extent flags would be if all of
802 * the delayed refs are not processed.
804 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
805 struct btrfs_fs_info *fs_info, u64 bytenr,
806 u64 offset, int metadata, u64 *refs, u64 *flags)
808 struct btrfs_delayed_ref_head *head;
809 struct btrfs_delayed_ref_root *delayed_refs;
810 struct btrfs_path *path;
811 struct btrfs_extent_item *ei;
812 struct extent_buffer *leaf;
813 struct btrfs_key key;
820 * If we don't have skinny metadata, don't bother doing anything
823 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
824 offset = fs_info->nodesize;
828 path = btrfs_alloc_path();
833 path->skip_locking = 1;
834 path->search_commit_root = 1;
838 key.objectid = bytenr;
841 key.type = BTRFS_METADATA_ITEM_KEY;
843 key.type = BTRFS_EXTENT_ITEM_KEY;
845 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
849 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
850 if (path->slots[0]) {
852 btrfs_item_key_to_cpu(path->nodes[0], &key,
854 if (key.objectid == bytenr &&
855 key.type == BTRFS_EXTENT_ITEM_KEY &&
856 key.offset == fs_info->nodesize)
862 leaf = path->nodes[0];
863 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
864 if (item_size >= sizeof(*ei)) {
865 ei = btrfs_item_ptr(leaf, path->slots[0],
866 struct btrfs_extent_item);
867 num_refs = btrfs_extent_refs(leaf, ei);
868 extent_flags = btrfs_extent_flags(leaf, ei);
870 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
871 struct btrfs_extent_item_v0 *ei0;
872 BUG_ON(item_size != sizeof(*ei0));
873 ei0 = btrfs_item_ptr(leaf, path->slots[0],
874 struct btrfs_extent_item_v0);
875 num_refs = btrfs_extent_refs_v0(leaf, ei0);
876 /* FIXME: this isn't correct for data */
877 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
882 BUG_ON(num_refs == 0);
892 delayed_refs = &trans->transaction->delayed_refs;
893 spin_lock(&delayed_refs->lock);
894 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
896 if (!mutex_trylock(&head->mutex)) {
897 refcount_inc(&head->refs);
898 spin_unlock(&delayed_refs->lock);
900 btrfs_release_path(path);
903 * Mutex was contended, block until it's released and try
906 mutex_lock(&head->mutex);
907 mutex_unlock(&head->mutex);
908 btrfs_put_delayed_ref_head(head);
911 spin_lock(&head->lock);
912 if (head->extent_op && head->extent_op->update_flags)
913 extent_flags |= head->extent_op->flags_to_set;
915 BUG_ON(num_refs == 0);
917 num_refs += head->ref_mod;
918 spin_unlock(&head->lock);
919 mutex_unlock(&head->mutex);
921 spin_unlock(&delayed_refs->lock);
923 WARN_ON(num_refs == 0);
927 *flags = extent_flags;
929 btrfs_free_path(path);
934 * Back reference rules. Back refs have three main goals:
936 * 1) differentiate between all holders of references to an extent so that
937 * when a reference is dropped we can make sure it was a valid reference
938 * before freeing the extent.
940 * 2) Provide enough information to quickly find the holders of an extent
941 * if we notice a given block is corrupted or bad.
943 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
944 * maintenance. This is actually the same as #2, but with a slightly
945 * different use case.
947 * There are two kinds of back refs. The implicit back refs is optimized
948 * for pointers in non-shared tree blocks. For a given pointer in a block,
949 * back refs of this kind provide information about the block's owner tree
950 * and the pointer's key. These information allow us to find the block by
951 * b-tree searching. The full back refs is for pointers in tree blocks not
952 * referenced by their owner trees. The location of tree block is recorded
953 * in the back refs. Actually the full back refs is generic, and can be
954 * used in all cases the implicit back refs is used. The major shortcoming
955 * of the full back refs is its overhead. Every time a tree block gets
956 * COWed, we have to update back refs entry for all pointers in it.
958 * For a newly allocated tree block, we use implicit back refs for
959 * pointers in it. This means most tree related operations only involve
960 * implicit back refs. For a tree block created in old transaction, the
961 * only way to drop a reference to it is COW it. So we can detect the
962 * event that tree block loses its owner tree's reference and do the
963 * back refs conversion.
965 * When a tree block is COWed through a tree, there are four cases:
967 * The reference count of the block is one and the tree is the block's
968 * owner tree. Nothing to do in this case.
970 * The reference count of the block is one and the tree is not the
971 * block's owner tree. In this case, full back refs is used for pointers
972 * in the block. Remove these full back refs, add implicit back refs for
973 * every pointers in the new block.
975 * The reference count of the block is greater than one and the tree is
976 * the block's owner tree. In this case, implicit back refs is used for
977 * pointers in the block. Add full back refs for every pointers in the
978 * block, increase lower level extents' reference counts. The original
979 * implicit back refs are entailed to the new block.
981 * The reference count of the block is greater than one and the tree is
982 * not the block's owner tree. Add implicit back refs for every pointer in
983 * the new block, increase lower level extents' reference count.
985 * Back Reference Key composing:
987 * The key objectid corresponds to the first byte in the extent,
988 * The key type is used to differentiate between types of back refs.
989 * There are different meanings of the key offset for different types
992 * File extents can be referenced by:
994 * - multiple snapshots, subvolumes, or different generations in one subvol
995 * - different files inside a single subvolume
996 * - different offsets inside a file (bookend extents in file.c)
998 * The extent ref structure for the implicit back refs has fields for:
1000 * - Objectid of the subvolume root
1001 * - objectid of the file holding the reference
1002 * - original offset in the file
1003 * - how many bookend extents
1005 * The key offset for the implicit back refs is hash of the first
1008 * The extent ref structure for the full back refs has field for:
1010 * - number of pointers in the tree leaf
1012 * The key offset for the implicit back refs is the first byte of
1015 * When a file extent is allocated, The implicit back refs is used.
1016 * the fields are filled in:
1018 * (root_key.objectid, inode objectid, offset in file, 1)
1020 * When a file extent is removed file truncation, we find the
1021 * corresponding implicit back refs and check the following fields:
1023 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1025 * Btree extents can be referenced by:
1027 * - Different subvolumes
1029 * Both the implicit back refs and the full back refs for tree blocks
1030 * only consist of key. The key offset for the implicit back refs is
1031 * objectid of block's owner tree. The key offset for the full back refs
1032 * is the first byte of parent block.
1034 * When implicit back refs is used, information about the lowest key and
1035 * level of the tree block are required. These information are stored in
1036 * tree block info structure.
1039 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1040 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1041 struct btrfs_fs_info *fs_info,
1042 struct btrfs_path *path,
1043 u64 owner, u32 extra_size)
1045 struct btrfs_root *root = fs_info->extent_root;
1046 struct btrfs_extent_item *item;
1047 struct btrfs_extent_item_v0 *ei0;
1048 struct btrfs_extent_ref_v0 *ref0;
1049 struct btrfs_tree_block_info *bi;
1050 struct extent_buffer *leaf;
1051 struct btrfs_key key;
1052 struct btrfs_key found_key;
1053 u32 new_size = sizeof(*item);
1057 leaf = path->nodes[0];
1058 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1060 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1061 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1062 struct btrfs_extent_item_v0);
1063 refs = btrfs_extent_refs_v0(leaf, ei0);
1065 if (owner == (u64)-1) {
1067 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1068 ret = btrfs_next_leaf(root, path);
1071 BUG_ON(ret > 0); /* Corruption */
1072 leaf = path->nodes[0];
1074 btrfs_item_key_to_cpu(leaf, &found_key,
1076 BUG_ON(key.objectid != found_key.objectid);
1077 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1081 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1082 struct btrfs_extent_ref_v0);
1083 owner = btrfs_ref_objectid_v0(leaf, ref0);
1087 btrfs_release_path(path);
1089 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1090 new_size += sizeof(*bi);
1092 new_size -= sizeof(*ei0);
1093 ret = btrfs_search_slot(trans, root, &key, path,
1094 new_size + extra_size, 1);
1097 BUG_ON(ret); /* Corruption */
1099 btrfs_extend_item(fs_info, path, new_size);
1101 leaf = path->nodes[0];
1102 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1103 btrfs_set_extent_refs(leaf, item, refs);
1104 /* FIXME: get real generation */
1105 btrfs_set_extent_generation(leaf, item, 0);
1106 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1107 btrfs_set_extent_flags(leaf, item,
1108 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1109 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1110 bi = (struct btrfs_tree_block_info *)(item + 1);
1111 /* FIXME: get first key of the block */
1112 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1113 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1115 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1117 btrfs_mark_buffer_dirty(leaf);
1123 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1124 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1125 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1127 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1128 struct btrfs_extent_inline_ref *iref,
1129 enum btrfs_inline_ref_type is_data)
1131 int type = btrfs_extent_inline_ref_type(eb, iref);
1132 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1134 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1135 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1136 type == BTRFS_SHARED_DATA_REF_KEY ||
1137 type == BTRFS_EXTENT_DATA_REF_KEY) {
1138 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1139 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1141 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1142 ASSERT(eb->fs_info);
1144 * Every shared one has parent tree
1145 * block, which must be aligned to
1149 IS_ALIGNED(offset, eb->fs_info->nodesize))
1152 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1153 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1155 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1156 ASSERT(eb->fs_info);
1158 * Every shared one has parent tree
1159 * block, which must be aligned to
1163 IS_ALIGNED(offset, eb->fs_info->nodesize))
1167 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1172 btrfs_print_leaf((struct extent_buffer *)eb);
1173 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1177 return BTRFS_REF_TYPE_INVALID;
1180 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1182 u32 high_crc = ~(u32)0;
1183 u32 low_crc = ~(u32)0;
1186 lenum = cpu_to_le64(root_objectid);
1187 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1188 lenum = cpu_to_le64(owner);
1189 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1190 lenum = cpu_to_le64(offset);
1191 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1193 return ((u64)high_crc << 31) ^ (u64)low_crc;
1196 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1197 struct btrfs_extent_data_ref *ref)
1199 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1200 btrfs_extent_data_ref_objectid(leaf, ref),
1201 btrfs_extent_data_ref_offset(leaf, ref));
1204 static int match_extent_data_ref(struct extent_buffer *leaf,
1205 struct btrfs_extent_data_ref *ref,
1206 u64 root_objectid, u64 owner, u64 offset)
1208 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1209 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1210 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1215 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1216 struct btrfs_path *path,
1217 u64 bytenr, u64 parent,
1219 u64 owner, u64 offset)
1221 struct btrfs_root *root = trans->fs_info->extent_root;
1222 struct btrfs_key key;
1223 struct btrfs_extent_data_ref *ref;
1224 struct extent_buffer *leaf;
1230 key.objectid = bytenr;
1232 key.type = BTRFS_SHARED_DATA_REF_KEY;
1233 key.offset = parent;
1235 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1236 key.offset = hash_extent_data_ref(root_objectid,
1241 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1250 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1251 key.type = BTRFS_EXTENT_REF_V0_KEY;
1252 btrfs_release_path(path);
1253 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1264 leaf = path->nodes[0];
1265 nritems = btrfs_header_nritems(leaf);
1267 if (path->slots[0] >= nritems) {
1268 ret = btrfs_next_leaf(root, path);
1274 leaf = path->nodes[0];
1275 nritems = btrfs_header_nritems(leaf);
1279 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1280 if (key.objectid != bytenr ||
1281 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1284 ref = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_data_ref);
1287 if (match_extent_data_ref(leaf, ref, root_objectid,
1290 btrfs_release_path(path);
1302 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1303 struct btrfs_path *path,
1304 u64 bytenr, u64 parent,
1305 u64 root_objectid, u64 owner,
1306 u64 offset, int refs_to_add)
1308 struct btrfs_root *root = trans->fs_info->extent_root;
1309 struct btrfs_key key;
1310 struct extent_buffer *leaf;
1315 key.objectid = bytenr;
1317 key.type = BTRFS_SHARED_DATA_REF_KEY;
1318 key.offset = parent;
1319 size = sizeof(struct btrfs_shared_data_ref);
1321 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1322 key.offset = hash_extent_data_ref(root_objectid,
1324 size = sizeof(struct btrfs_extent_data_ref);
1327 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1328 if (ret && ret != -EEXIST)
1331 leaf = path->nodes[0];
1333 struct btrfs_shared_data_ref *ref;
1334 ref = btrfs_item_ptr(leaf, path->slots[0],
1335 struct btrfs_shared_data_ref);
1337 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1339 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1340 num_refs += refs_to_add;
1341 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1344 struct btrfs_extent_data_ref *ref;
1345 while (ret == -EEXIST) {
1346 ref = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_extent_data_ref);
1348 if (match_extent_data_ref(leaf, ref, root_objectid,
1351 btrfs_release_path(path);
1353 ret = btrfs_insert_empty_item(trans, root, path, &key,
1355 if (ret && ret != -EEXIST)
1358 leaf = path->nodes[0];
1360 ref = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_extent_data_ref);
1363 btrfs_set_extent_data_ref_root(leaf, ref,
1365 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1366 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1367 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1369 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1370 num_refs += refs_to_add;
1371 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1374 btrfs_mark_buffer_dirty(leaf);
1377 btrfs_release_path(path);
1381 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1382 struct btrfs_path *path,
1383 int refs_to_drop, int *last_ref)
1385 struct btrfs_key key;
1386 struct btrfs_extent_data_ref *ref1 = NULL;
1387 struct btrfs_shared_data_ref *ref2 = NULL;
1388 struct extent_buffer *leaf;
1392 leaf = path->nodes[0];
1393 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1395 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1396 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1397 struct btrfs_extent_data_ref);
1398 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1399 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1400 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1401 struct btrfs_shared_data_ref);
1402 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1404 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1405 struct btrfs_extent_ref_v0 *ref0;
1406 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1407 struct btrfs_extent_ref_v0);
1408 num_refs = btrfs_ref_count_v0(leaf, ref0);
1414 BUG_ON(num_refs < refs_to_drop);
1415 num_refs -= refs_to_drop;
1417 if (num_refs == 0) {
1418 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1421 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1422 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1423 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1424 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1425 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 struct btrfs_extent_ref_v0 *ref0;
1428 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1429 struct btrfs_extent_ref_v0);
1430 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1433 btrfs_mark_buffer_dirty(leaf);
1438 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1439 struct btrfs_extent_inline_ref *iref)
1441 struct btrfs_key key;
1442 struct extent_buffer *leaf;
1443 struct btrfs_extent_data_ref *ref1;
1444 struct btrfs_shared_data_ref *ref2;
1448 leaf = path->nodes[0];
1449 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1452 * If type is invalid, we should have bailed out earlier than
1455 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1456 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1457 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1458 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1459 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1461 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1462 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1464 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1465 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1466 struct btrfs_extent_data_ref);
1467 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1468 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1469 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1470 struct btrfs_shared_data_ref);
1471 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1472 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1473 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1474 struct btrfs_extent_ref_v0 *ref0;
1475 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1476 struct btrfs_extent_ref_v0);
1477 num_refs = btrfs_ref_count_v0(leaf, ref0);
1485 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1486 struct btrfs_path *path,
1487 u64 bytenr, u64 parent,
1490 struct btrfs_root *root = trans->fs_info->extent_root;
1491 struct btrfs_key key;
1494 key.objectid = bytenr;
1496 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1497 key.offset = parent;
1499 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1500 key.offset = root_objectid;
1503 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1506 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1507 if (ret == -ENOENT && parent) {
1508 btrfs_release_path(path);
1509 key.type = BTRFS_EXTENT_REF_V0_KEY;
1510 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1518 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1519 struct btrfs_path *path,
1520 u64 bytenr, u64 parent,
1523 struct btrfs_key key;
1526 key.objectid = bytenr;
1528 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1529 key.offset = parent;
1531 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1532 key.offset = root_objectid;
1535 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1537 btrfs_release_path(path);
1541 static inline int extent_ref_type(u64 parent, u64 owner)
1544 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1546 type = BTRFS_SHARED_BLOCK_REF_KEY;
1548 type = BTRFS_TREE_BLOCK_REF_KEY;
1551 type = BTRFS_SHARED_DATA_REF_KEY;
1553 type = BTRFS_EXTENT_DATA_REF_KEY;
1558 static int find_next_key(struct btrfs_path *path, int level,
1559 struct btrfs_key *key)
1562 for (; level < BTRFS_MAX_LEVEL; level++) {
1563 if (!path->nodes[level])
1565 if (path->slots[level] + 1 >=
1566 btrfs_header_nritems(path->nodes[level]))
1569 btrfs_item_key_to_cpu(path->nodes[level], key,
1570 path->slots[level] + 1);
1572 btrfs_node_key_to_cpu(path->nodes[level], key,
1573 path->slots[level] + 1);
1580 * look for inline back ref. if back ref is found, *ref_ret is set
1581 * to the address of inline back ref, and 0 is returned.
1583 * if back ref isn't found, *ref_ret is set to the address where it
1584 * should be inserted, and -ENOENT is returned.
1586 * if insert is true and there are too many inline back refs, the path
1587 * points to the extent item, and -EAGAIN is returned.
1589 * NOTE: inline back refs are ordered in the same way that back ref
1590 * items in the tree are ordered.
1592 static noinline_for_stack
1593 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1594 struct btrfs_path *path,
1595 struct btrfs_extent_inline_ref **ref_ret,
1596 u64 bytenr, u64 num_bytes,
1597 u64 parent, u64 root_objectid,
1598 u64 owner, u64 offset, int insert)
1600 struct btrfs_fs_info *fs_info = trans->fs_info;
1601 struct btrfs_root *root = fs_info->extent_root;
1602 struct btrfs_key key;
1603 struct extent_buffer *leaf;
1604 struct btrfs_extent_item *ei;
1605 struct btrfs_extent_inline_ref *iref;
1615 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1618 key.objectid = bytenr;
1619 key.type = BTRFS_EXTENT_ITEM_KEY;
1620 key.offset = num_bytes;
1622 want = extent_ref_type(parent, owner);
1624 extra_size = btrfs_extent_inline_ref_size(want);
1625 path->keep_locks = 1;
1630 * Owner is our level, so we can just add one to get the level for the
1631 * block we are interested in.
1633 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1634 key.type = BTRFS_METADATA_ITEM_KEY;
1639 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1646 * We may be a newly converted file system which still has the old fat
1647 * extent entries for metadata, so try and see if we have one of those.
1649 if (ret > 0 && skinny_metadata) {
1650 skinny_metadata = false;
1651 if (path->slots[0]) {
1653 btrfs_item_key_to_cpu(path->nodes[0], &key,
1655 if (key.objectid == bytenr &&
1656 key.type == BTRFS_EXTENT_ITEM_KEY &&
1657 key.offset == num_bytes)
1661 key.objectid = bytenr;
1662 key.type = BTRFS_EXTENT_ITEM_KEY;
1663 key.offset = num_bytes;
1664 btrfs_release_path(path);
1669 if (ret && !insert) {
1672 } else if (WARN_ON(ret)) {
1677 leaf = path->nodes[0];
1678 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1679 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1680 if (item_size < sizeof(*ei)) {
1685 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1691 leaf = path->nodes[0];
1692 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1695 BUG_ON(item_size < sizeof(*ei));
1697 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1698 flags = btrfs_extent_flags(leaf, ei);
1700 ptr = (unsigned long)(ei + 1);
1701 end = (unsigned long)ei + item_size;
1703 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1704 ptr += sizeof(struct btrfs_tree_block_info);
1708 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1709 needed = BTRFS_REF_TYPE_DATA;
1711 needed = BTRFS_REF_TYPE_BLOCK;
1719 iref = (struct btrfs_extent_inline_ref *)ptr;
1720 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1721 if (type == BTRFS_REF_TYPE_INVALID) {
1729 ptr += btrfs_extent_inline_ref_size(type);
1733 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1734 struct btrfs_extent_data_ref *dref;
1735 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1736 if (match_extent_data_ref(leaf, dref, root_objectid,
1741 if (hash_extent_data_ref_item(leaf, dref) <
1742 hash_extent_data_ref(root_objectid, owner, offset))
1746 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1748 if (parent == ref_offset) {
1752 if (ref_offset < parent)
1755 if (root_objectid == ref_offset) {
1759 if (ref_offset < root_objectid)
1763 ptr += btrfs_extent_inline_ref_size(type);
1765 if (err == -ENOENT && insert) {
1766 if (item_size + extra_size >=
1767 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1772 * To add new inline back ref, we have to make sure
1773 * there is no corresponding back ref item.
1774 * For simplicity, we just do not add new inline back
1775 * ref if there is any kind of item for this block
1777 if (find_next_key(path, 0, &key) == 0 &&
1778 key.objectid == bytenr &&
1779 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1784 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1787 path->keep_locks = 0;
1788 btrfs_unlock_up_safe(path, 1);
1794 * helper to add new inline back ref
1796 static noinline_for_stack
1797 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1798 struct btrfs_path *path,
1799 struct btrfs_extent_inline_ref *iref,
1800 u64 parent, u64 root_objectid,
1801 u64 owner, u64 offset, int refs_to_add,
1802 struct btrfs_delayed_extent_op *extent_op)
1804 struct extent_buffer *leaf;
1805 struct btrfs_extent_item *ei;
1808 unsigned long item_offset;
1813 leaf = path->nodes[0];
1814 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1815 item_offset = (unsigned long)iref - (unsigned long)ei;
1817 type = extent_ref_type(parent, owner);
1818 size = btrfs_extent_inline_ref_size(type);
1820 btrfs_extend_item(fs_info, path, size);
1822 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1823 refs = btrfs_extent_refs(leaf, ei);
1824 refs += refs_to_add;
1825 btrfs_set_extent_refs(leaf, ei, refs);
1827 __run_delayed_extent_op(extent_op, leaf, ei);
1829 ptr = (unsigned long)ei + item_offset;
1830 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1831 if (ptr < end - size)
1832 memmove_extent_buffer(leaf, ptr + size, ptr,
1835 iref = (struct btrfs_extent_inline_ref *)ptr;
1836 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1837 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1838 struct btrfs_extent_data_ref *dref;
1839 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1840 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1841 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1842 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1843 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1844 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1845 struct btrfs_shared_data_ref *sref;
1846 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1847 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1848 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1849 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1850 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1852 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1854 btrfs_mark_buffer_dirty(leaf);
1857 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1858 struct btrfs_path *path,
1859 struct btrfs_extent_inline_ref **ref_ret,
1860 u64 bytenr, u64 num_bytes, u64 parent,
1861 u64 root_objectid, u64 owner, u64 offset)
1865 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1866 num_bytes, parent, root_objectid,
1871 btrfs_release_path(path);
1874 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1875 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1878 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1879 root_objectid, owner, offset);
1885 * helper to update/remove inline back ref
1887 static noinline_for_stack
1888 void update_inline_extent_backref(struct btrfs_path *path,
1889 struct btrfs_extent_inline_ref *iref,
1891 struct btrfs_delayed_extent_op *extent_op,
1894 struct extent_buffer *leaf = path->nodes[0];
1895 struct btrfs_fs_info *fs_info = leaf->fs_info;
1896 struct btrfs_extent_item *ei;
1897 struct btrfs_extent_data_ref *dref = NULL;
1898 struct btrfs_shared_data_ref *sref = NULL;
1906 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1907 refs = btrfs_extent_refs(leaf, ei);
1908 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1909 refs += refs_to_mod;
1910 btrfs_set_extent_refs(leaf, ei, refs);
1912 __run_delayed_extent_op(extent_op, leaf, ei);
1915 * If type is invalid, we should have bailed out after
1916 * lookup_inline_extent_backref().
1918 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1919 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1921 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1922 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1923 refs = btrfs_extent_data_ref_count(leaf, dref);
1924 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1925 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1926 refs = btrfs_shared_data_ref_count(leaf, sref);
1929 BUG_ON(refs_to_mod != -1);
1932 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1933 refs += refs_to_mod;
1936 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1937 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1939 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1942 size = btrfs_extent_inline_ref_size(type);
1943 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1944 ptr = (unsigned long)iref;
1945 end = (unsigned long)ei + item_size;
1946 if (ptr + size < end)
1947 memmove_extent_buffer(leaf, ptr, ptr + size,
1950 btrfs_truncate_item(fs_info, path, item_size, 1);
1952 btrfs_mark_buffer_dirty(leaf);
1955 static noinline_for_stack
1956 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1957 struct btrfs_path *path,
1958 u64 bytenr, u64 num_bytes, u64 parent,
1959 u64 root_objectid, u64 owner,
1960 u64 offset, int refs_to_add,
1961 struct btrfs_delayed_extent_op *extent_op)
1963 struct btrfs_extent_inline_ref *iref;
1966 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1967 num_bytes, parent, root_objectid,
1970 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1971 update_inline_extent_backref(path, iref, refs_to_add,
1973 } else if (ret == -ENOENT) {
1974 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1975 root_objectid, owner, offset,
1976 refs_to_add, extent_op);
1982 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1983 struct btrfs_path *path,
1984 u64 bytenr, u64 parent, u64 root_objectid,
1985 u64 owner, u64 offset, int refs_to_add)
1988 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1989 BUG_ON(refs_to_add != 1);
1990 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1993 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1994 root_objectid, owner, offset,
2000 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2001 struct btrfs_path *path,
2002 struct btrfs_extent_inline_ref *iref,
2003 int refs_to_drop, int is_data, int *last_ref)
2007 BUG_ON(!is_data && refs_to_drop != 1);
2009 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
2011 } else if (is_data) {
2012 ret = remove_extent_data_ref(trans, path, refs_to_drop,
2016 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
2021 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2022 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2023 u64 *discarded_bytes)
2026 u64 bytes_left, end;
2027 u64 aligned_start = ALIGN(start, 1 << 9);
2029 if (WARN_ON(start != aligned_start)) {
2030 len -= aligned_start - start;
2031 len = round_down(len, 1 << 9);
2032 start = aligned_start;
2035 *discarded_bytes = 0;
2043 /* Skip any superblocks on this device. */
2044 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2045 u64 sb_start = btrfs_sb_offset(j);
2046 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2047 u64 size = sb_start - start;
2049 if (!in_range(sb_start, start, bytes_left) &&
2050 !in_range(sb_end, start, bytes_left) &&
2051 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2055 * Superblock spans beginning of range. Adjust start and
2058 if (sb_start <= start) {
2059 start += sb_end - start;
2064 bytes_left = end - start;
2069 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2072 *discarded_bytes += size;
2073 else if (ret != -EOPNOTSUPP)
2082 bytes_left = end - start;
2086 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2089 *discarded_bytes += bytes_left;
2094 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2095 u64 num_bytes, u64 *actual_bytes)
2098 u64 discarded_bytes = 0;
2099 struct btrfs_bio *bbio = NULL;
2103 * Avoid races with device replace and make sure our bbio has devices
2104 * associated to its stripes that don't go away while we are discarding.
2106 btrfs_bio_counter_inc_blocked(fs_info);
2107 /* Tell the block device(s) that the sectors can be discarded */
2108 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2110 /* Error condition is -ENOMEM */
2112 struct btrfs_bio_stripe *stripe = bbio->stripes;
2116 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2118 struct request_queue *req_q;
2120 if (!stripe->dev->bdev) {
2121 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2124 req_q = bdev_get_queue(stripe->dev->bdev);
2125 if (!blk_queue_discard(req_q))
2128 ret = btrfs_issue_discard(stripe->dev->bdev,
2133 discarded_bytes += bytes;
2134 else if (ret != -EOPNOTSUPP)
2135 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2138 * Just in case we get back EOPNOTSUPP for some reason,
2139 * just ignore the return value so we don't screw up
2140 * people calling discard_extent.
2144 btrfs_put_bbio(bbio);
2146 btrfs_bio_counter_dec(fs_info);
2149 *actual_bytes = discarded_bytes;
2152 if (ret == -EOPNOTSUPP)
2157 /* Can return -ENOMEM */
2158 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2159 struct btrfs_root *root,
2160 u64 bytenr, u64 num_bytes, u64 parent,
2161 u64 root_objectid, u64 owner, u64 offset)
2163 struct btrfs_fs_info *fs_info = root->fs_info;
2164 int old_ref_mod, new_ref_mod;
2167 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2168 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2170 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2171 owner, offset, BTRFS_ADD_DELAYED_REF);
2173 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2174 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
2176 root_objectid, (int)owner,
2177 BTRFS_ADD_DELAYED_REF, NULL,
2178 &old_ref_mod, &new_ref_mod);
2180 ret = btrfs_add_delayed_data_ref(trans, bytenr,
2182 root_objectid, owner, offset,
2183 0, BTRFS_ADD_DELAYED_REF,
2184 &old_ref_mod, &new_ref_mod);
2187 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2188 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2190 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2197 * __btrfs_inc_extent_ref - insert backreference for a given extent
2199 * @trans: Handle of transaction
2201 * @node: The delayed ref node used to get the bytenr/length for
2202 * extent whose references are incremented.
2204 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2205 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2206 * bytenr of the parent block. Since new extents are always
2207 * created with indirect references, this will only be the case
2208 * when relocating a shared extent. In that case, root_objectid
2209 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2212 * @root_objectid: The id of the root where this modification has originated,
2213 * this can be either one of the well-known metadata trees or
2214 * the subvolume id which references this extent.
2216 * @owner: For data extents it is the inode number of the owning file.
2217 * For metadata extents this parameter holds the level in the
2218 * tree of the extent.
2220 * @offset: For metadata extents the offset is ignored and is currently
2221 * always passed as 0. For data extents it is the fileoffset
2222 * this extent belongs to.
2224 * @refs_to_add Number of references to add
2226 * @extent_op Pointer to a structure, holding information necessary when
2227 * updating a tree block's flags
2230 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2231 struct btrfs_delayed_ref_node *node,
2232 u64 parent, u64 root_objectid,
2233 u64 owner, u64 offset, int refs_to_add,
2234 struct btrfs_delayed_extent_op *extent_op)
2236 struct btrfs_path *path;
2237 struct extent_buffer *leaf;
2238 struct btrfs_extent_item *item;
2239 struct btrfs_key key;
2240 u64 bytenr = node->bytenr;
2241 u64 num_bytes = node->num_bytes;
2245 path = btrfs_alloc_path();
2249 path->reada = READA_FORWARD;
2250 path->leave_spinning = 1;
2251 /* this will setup the path even if it fails to insert the back ref */
2252 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2253 parent, root_objectid, owner,
2254 offset, refs_to_add, extent_op);
2255 if ((ret < 0 && ret != -EAGAIN) || !ret)
2259 * Ok we had -EAGAIN which means we didn't have space to insert and
2260 * inline extent ref, so just update the reference count and add a
2263 leaf = path->nodes[0];
2264 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2265 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2266 refs = btrfs_extent_refs(leaf, item);
2267 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2269 __run_delayed_extent_op(extent_op, leaf, item);
2271 btrfs_mark_buffer_dirty(leaf);
2272 btrfs_release_path(path);
2274 path->reada = READA_FORWARD;
2275 path->leave_spinning = 1;
2276 /* now insert the actual backref */
2277 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2278 owner, offset, refs_to_add);
2280 btrfs_abort_transaction(trans, ret);
2282 btrfs_free_path(path);
2286 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2287 struct btrfs_delayed_ref_node *node,
2288 struct btrfs_delayed_extent_op *extent_op,
2289 int insert_reserved)
2292 struct btrfs_delayed_data_ref *ref;
2293 struct btrfs_key ins;
2298 ins.objectid = node->bytenr;
2299 ins.offset = node->num_bytes;
2300 ins.type = BTRFS_EXTENT_ITEM_KEY;
2302 ref = btrfs_delayed_node_to_data_ref(node);
2303 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2305 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2306 parent = ref->parent;
2307 ref_root = ref->root;
2309 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2311 flags |= extent_op->flags_to_set;
2312 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2313 flags, ref->objectid,
2316 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2317 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2318 ref->objectid, ref->offset,
2319 node->ref_mod, extent_op);
2320 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2321 ret = __btrfs_free_extent(trans, node, parent,
2322 ref_root, ref->objectid,
2323 ref->offset, node->ref_mod,
2331 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2332 struct extent_buffer *leaf,
2333 struct btrfs_extent_item *ei)
2335 u64 flags = btrfs_extent_flags(leaf, ei);
2336 if (extent_op->update_flags) {
2337 flags |= extent_op->flags_to_set;
2338 btrfs_set_extent_flags(leaf, ei, flags);
2341 if (extent_op->update_key) {
2342 struct btrfs_tree_block_info *bi;
2343 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2344 bi = (struct btrfs_tree_block_info *)(ei + 1);
2345 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2349 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2350 struct btrfs_delayed_ref_head *head,
2351 struct btrfs_delayed_extent_op *extent_op)
2353 struct btrfs_fs_info *fs_info = trans->fs_info;
2354 struct btrfs_key key;
2355 struct btrfs_path *path;
2356 struct btrfs_extent_item *ei;
2357 struct extent_buffer *leaf;
2361 int metadata = !extent_op->is_data;
2366 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2369 path = btrfs_alloc_path();
2373 key.objectid = head->bytenr;
2376 key.type = BTRFS_METADATA_ITEM_KEY;
2377 key.offset = extent_op->level;
2379 key.type = BTRFS_EXTENT_ITEM_KEY;
2380 key.offset = head->num_bytes;
2384 path->reada = READA_FORWARD;
2385 path->leave_spinning = 1;
2386 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2393 if (path->slots[0] > 0) {
2395 btrfs_item_key_to_cpu(path->nodes[0], &key,
2397 if (key.objectid == head->bytenr &&
2398 key.type == BTRFS_EXTENT_ITEM_KEY &&
2399 key.offset == head->num_bytes)
2403 btrfs_release_path(path);
2406 key.objectid = head->bytenr;
2407 key.offset = head->num_bytes;
2408 key.type = BTRFS_EXTENT_ITEM_KEY;
2417 leaf = path->nodes[0];
2418 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420 if (item_size < sizeof(*ei)) {
2421 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2426 leaf = path->nodes[0];
2427 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2430 BUG_ON(item_size < sizeof(*ei));
2431 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2432 __run_delayed_extent_op(extent_op, leaf, ei);
2434 btrfs_mark_buffer_dirty(leaf);
2436 btrfs_free_path(path);
2440 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2441 struct btrfs_delayed_ref_node *node,
2442 struct btrfs_delayed_extent_op *extent_op,
2443 int insert_reserved)
2446 struct btrfs_delayed_tree_ref *ref;
2450 ref = btrfs_delayed_node_to_tree_ref(node);
2451 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2453 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2454 parent = ref->parent;
2455 ref_root = ref->root;
2457 if (node->ref_mod != 1) {
2458 btrfs_err(trans->fs_info,
2459 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2460 node->bytenr, node->ref_mod, node->action, ref_root,
2464 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2465 BUG_ON(!extent_op || !extent_op->update_flags);
2466 ret = alloc_reserved_tree_block(trans, node, extent_op);
2467 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2468 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2469 ref->level, 0, 1, extent_op);
2470 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2471 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2472 ref->level, 0, 1, extent_op);
2479 /* helper function to actually process a single delayed ref entry */
2480 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2481 struct btrfs_delayed_ref_node *node,
2482 struct btrfs_delayed_extent_op *extent_op,
2483 int insert_reserved)
2487 if (trans->aborted) {
2488 if (insert_reserved)
2489 btrfs_pin_extent(trans->fs_info, node->bytenr,
2490 node->num_bytes, 1);
2494 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2495 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2496 ret = run_delayed_tree_ref(trans, node, extent_op,
2498 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2499 node->type == BTRFS_SHARED_DATA_REF_KEY)
2500 ret = run_delayed_data_ref(trans, node, extent_op,
2507 static inline struct btrfs_delayed_ref_node *
2508 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2510 struct btrfs_delayed_ref_node *ref;
2512 if (RB_EMPTY_ROOT(&head->ref_tree))
2516 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2517 * This is to prevent a ref count from going down to zero, which deletes
2518 * the extent item from the extent tree, when there still are references
2519 * to add, which would fail because they would not find the extent item.
2521 if (!list_empty(&head->ref_add_list))
2522 return list_first_entry(&head->ref_add_list,
2523 struct btrfs_delayed_ref_node, add_list);
2525 ref = rb_entry(rb_first(&head->ref_tree),
2526 struct btrfs_delayed_ref_node, ref_node);
2527 ASSERT(list_empty(&ref->add_list));
2531 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2532 struct btrfs_delayed_ref_head *head)
2534 spin_lock(&delayed_refs->lock);
2535 head->processing = 0;
2536 delayed_refs->num_heads_ready++;
2537 spin_unlock(&delayed_refs->lock);
2538 btrfs_delayed_ref_unlock(head);
2541 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2542 struct btrfs_delayed_ref_head *head)
2544 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2549 head->extent_op = NULL;
2550 if (head->must_insert_reserved) {
2551 btrfs_free_delayed_extent_op(extent_op);
2554 spin_unlock(&head->lock);
2555 ret = run_delayed_extent_op(trans, head, extent_op);
2556 btrfs_free_delayed_extent_op(extent_op);
2557 return ret ? ret : 1;
2560 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2561 struct btrfs_delayed_ref_head *head)
2564 struct btrfs_fs_info *fs_info = trans->fs_info;
2565 struct btrfs_delayed_ref_root *delayed_refs;
2568 delayed_refs = &trans->transaction->delayed_refs;
2570 ret = cleanup_extent_op(trans, head);
2572 unselect_delayed_ref_head(delayed_refs, head);
2573 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2580 * Need to drop our head ref lock and re-acquire the delayed ref lock
2581 * and then re-check to make sure nobody got added.
2583 spin_unlock(&head->lock);
2584 spin_lock(&delayed_refs->lock);
2585 spin_lock(&head->lock);
2586 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2587 spin_unlock(&head->lock);
2588 spin_unlock(&delayed_refs->lock);
2591 delayed_refs->num_heads--;
2592 rb_erase(&head->href_node, &delayed_refs->href_root);
2593 RB_CLEAR_NODE(&head->href_node);
2594 spin_unlock(&head->lock);
2595 spin_unlock(&delayed_refs->lock);
2596 atomic_dec(&delayed_refs->num_entries);
2598 trace_run_delayed_ref_head(fs_info, head, 0);
2600 if (head->total_ref_mod < 0) {
2601 struct btrfs_space_info *space_info;
2605 flags = BTRFS_BLOCK_GROUP_DATA;
2606 else if (head->is_system)
2607 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2609 flags = BTRFS_BLOCK_GROUP_METADATA;
2610 space_info = __find_space_info(fs_info, flags);
2612 percpu_counter_add(&space_info->total_bytes_pinned,
2615 if (head->is_data) {
2616 spin_lock(&delayed_refs->lock);
2617 delayed_refs->pending_csums -= head->num_bytes;
2618 spin_unlock(&delayed_refs->lock);
2622 if (head->must_insert_reserved) {
2623 btrfs_pin_extent(fs_info, head->bytenr,
2624 head->num_bytes, 1);
2625 if (head->is_data) {
2626 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2631 /* Also free its reserved qgroup space */
2632 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2633 head->qgroup_reserved);
2634 btrfs_delayed_ref_unlock(head);
2635 btrfs_put_delayed_ref_head(head);
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_node *ref;
2649 struct btrfs_delayed_ref_head *locked_ref = NULL;
2650 struct btrfs_delayed_extent_op *extent_op;
2651 ktime_t start = ktime_get();
2653 unsigned long count = 0;
2654 unsigned long actual_count = 0;
2655 int must_insert_reserved = 0;
2657 delayed_refs = &trans->transaction->delayed_refs;
2663 spin_lock(&delayed_refs->lock);
2664 locked_ref = btrfs_select_ref_head(trans);
2666 spin_unlock(&delayed_refs->lock);
2670 /* grab the lock that says we are going to process
2671 * all the refs for this head */
2672 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2673 spin_unlock(&delayed_refs->lock);
2675 * we may have dropped the spin lock to get the head
2676 * mutex lock, and that might have given someone else
2677 * time to free the head. If that's true, it has been
2678 * removed from our list and we can move on.
2680 if (ret == -EAGAIN) {
2688 * We need to try and merge add/drops of the same ref since we
2689 * can run into issues with relocate dropping the implicit ref
2690 * and then it being added back again before the drop can
2691 * finish. If we merged anything we need to re-loop so we can
2693 * Or we can get node references of the same type that weren't
2694 * merged when created due to bumps in the tree mod seq, and
2695 * we need to merge them to prevent adding an inline extent
2696 * backref before dropping it (triggering a BUG_ON at
2697 * insert_inline_extent_backref()).
2699 spin_lock(&locked_ref->lock);
2700 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2702 ref = select_delayed_ref(locked_ref);
2704 if (ref && ref->seq &&
2705 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2706 spin_unlock(&locked_ref->lock);
2707 unselect_delayed_ref_head(delayed_refs, locked_ref);
2715 * We're done processing refs in this ref_head, clean everything
2716 * up and move on to the next ref_head.
2719 ret = cleanup_ref_head(trans, locked_ref);
2721 /* We dropped our lock, we need to loop. */
2734 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2735 RB_CLEAR_NODE(&ref->ref_node);
2736 if (!list_empty(&ref->add_list))
2737 list_del(&ref->add_list);
2739 * When we play the delayed ref, also correct the ref_mod on
2742 switch (ref->action) {
2743 case BTRFS_ADD_DELAYED_REF:
2744 case BTRFS_ADD_DELAYED_EXTENT:
2745 locked_ref->ref_mod -= ref->ref_mod;
2747 case BTRFS_DROP_DELAYED_REF:
2748 locked_ref->ref_mod += ref->ref_mod;
2753 atomic_dec(&delayed_refs->num_entries);
2756 * Record the must-insert_reserved flag before we drop the spin
2759 must_insert_reserved = locked_ref->must_insert_reserved;
2760 locked_ref->must_insert_reserved = 0;
2762 extent_op = locked_ref->extent_op;
2763 locked_ref->extent_op = NULL;
2764 spin_unlock(&locked_ref->lock);
2766 ret = run_one_delayed_ref(trans, ref, extent_op,
2767 must_insert_reserved);
2769 btrfs_free_delayed_extent_op(extent_op);
2771 unselect_delayed_ref_head(delayed_refs, locked_ref);
2772 btrfs_put_delayed_ref(ref);
2773 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2778 btrfs_put_delayed_ref(ref);
2784 * We don't want to include ref heads since we can have empty ref heads
2785 * and those will drastically skew our runtime down since we just do
2786 * accounting, no actual extent tree updates.
2788 if (actual_count > 0) {
2789 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2793 * We weigh the current average higher than our current runtime
2794 * to avoid large swings in the average.
2796 spin_lock(&delayed_refs->lock);
2797 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2798 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2799 spin_unlock(&delayed_refs->lock);
2804 #ifdef SCRAMBLE_DELAYED_REFS
2806 * Normally delayed refs get processed in ascending bytenr order. This
2807 * correlates in most cases to the order added. To expose dependencies on this
2808 * order, we start to process the tree in the middle instead of the beginning
2810 static u64 find_middle(struct rb_root *root)
2812 struct rb_node *n = root->rb_node;
2813 struct btrfs_delayed_ref_node *entry;
2816 u64 first = 0, last = 0;
2820 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2821 first = entry->bytenr;
2825 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2826 last = entry->bytenr;
2831 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2832 WARN_ON(!entry->in_tree);
2834 middle = entry->bytenr;
2847 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2851 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2852 sizeof(struct btrfs_extent_inline_ref));
2853 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2854 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2857 * We don't ever fill up leaves all the way so multiply by 2 just to be
2858 * closer to what we're really going to want to use.
2860 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2864 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2865 * would require to store the csums for that many bytes.
2867 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2870 u64 num_csums_per_leaf;
2873 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2874 num_csums_per_leaf = div64_u64(csum_size,
2875 (u64)btrfs_super_csum_size(fs_info->super_copy));
2876 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2877 num_csums += num_csums_per_leaf - 1;
2878 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2882 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2883 struct btrfs_fs_info *fs_info)
2885 struct btrfs_block_rsv *global_rsv;
2886 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2887 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2888 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2889 u64 num_bytes, num_dirty_bgs_bytes;
2892 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2893 num_heads = heads_to_leaves(fs_info, num_heads);
2895 num_bytes += (num_heads - 1) * fs_info->nodesize;
2897 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2899 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2901 global_rsv = &fs_info->global_block_rsv;
2904 * If we can't allocate any more chunks lets make sure we have _lots_ of
2905 * wiggle room since running delayed refs can create more delayed refs.
2907 if (global_rsv->space_info->full) {
2908 num_dirty_bgs_bytes <<= 1;
2912 spin_lock(&global_rsv->lock);
2913 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2915 spin_unlock(&global_rsv->lock);
2919 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2920 struct btrfs_fs_info *fs_info)
2923 atomic_read(&trans->transaction->delayed_refs.num_entries);
2928 avg_runtime = fs_info->avg_delayed_ref_runtime;
2929 val = num_entries * avg_runtime;
2930 if (val >= NSEC_PER_SEC)
2932 if (val >= NSEC_PER_SEC / 2)
2935 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2938 struct async_delayed_refs {
2939 struct btrfs_root *root;
2944 struct completion wait;
2945 struct btrfs_work work;
2948 static inline struct async_delayed_refs *
2949 to_async_delayed_refs(struct btrfs_work *work)
2951 return container_of(work, struct async_delayed_refs, work);
2954 static void delayed_ref_async_start(struct btrfs_work *work)
2956 struct async_delayed_refs *async = to_async_delayed_refs(work);
2957 struct btrfs_trans_handle *trans;
2958 struct btrfs_fs_info *fs_info = async->root->fs_info;
2961 /* if the commit is already started, we don't need to wait here */
2962 if (btrfs_transaction_blocked(fs_info))
2965 trans = btrfs_join_transaction(async->root);
2966 if (IS_ERR(trans)) {
2967 async->error = PTR_ERR(trans);
2972 * trans->sync means that when we call end_transaction, we won't
2973 * wait on delayed refs
2977 /* Don't bother flushing if we got into a different transaction */
2978 if (trans->transid > async->transid)
2981 ret = btrfs_run_delayed_refs(trans, async->count);
2985 ret = btrfs_end_transaction(trans);
2986 if (ret && !async->error)
2990 complete(&async->wait);
2995 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2996 unsigned long count, u64 transid, int wait)
2998 struct async_delayed_refs *async;
3001 async = kmalloc(sizeof(*async), GFP_NOFS);
3005 async->root = fs_info->tree_root;
3006 async->count = count;
3008 async->transid = transid;
3013 init_completion(&async->wait);
3015 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3016 delayed_ref_async_start, NULL, NULL);
3018 btrfs_queue_work(fs_info->extent_workers, &async->work);
3021 wait_for_completion(&async->wait);
3030 * this starts processing the delayed reference count updates and
3031 * extent insertions we have queued up so far. count can be
3032 * 0, which means to process everything in the tree at the start
3033 * of the run (but not newly added entries), or it can be some target
3034 * number you'd like to process.
3036 * Returns 0 on success or if called with an aborted transaction
3037 * Returns <0 on error and aborts the transaction
3039 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3040 unsigned long count)
3042 struct btrfs_fs_info *fs_info = trans->fs_info;
3043 struct rb_node *node;
3044 struct btrfs_delayed_ref_root *delayed_refs;
3045 struct btrfs_delayed_ref_head *head;
3047 int run_all = count == (unsigned long)-1;
3048 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3050 /* We'll clean this up in btrfs_cleanup_transaction */
3054 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3057 delayed_refs = &trans->transaction->delayed_refs;
3059 count = atomic_read(&delayed_refs->num_entries) * 2;
3062 #ifdef SCRAMBLE_DELAYED_REFS
3063 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3065 trans->can_flush_pending_bgs = false;
3066 ret = __btrfs_run_delayed_refs(trans, count);
3068 btrfs_abort_transaction(trans, ret);
3073 if (!list_empty(&trans->new_bgs))
3074 btrfs_create_pending_block_groups(trans);
3076 spin_lock(&delayed_refs->lock);
3077 node = rb_first(&delayed_refs->href_root);
3079 spin_unlock(&delayed_refs->lock);
3082 head = rb_entry(node, struct btrfs_delayed_ref_head,
3084 refcount_inc(&head->refs);
3085 spin_unlock(&delayed_refs->lock);
3087 /* Mutex was contended, block until it's released and retry. */
3088 mutex_lock(&head->mutex);
3089 mutex_unlock(&head->mutex);
3091 btrfs_put_delayed_ref_head(head);
3096 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3100 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3101 struct btrfs_fs_info *fs_info,
3102 u64 bytenr, u64 num_bytes, u64 flags,
3103 int level, int is_data)
3105 struct btrfs_delayed_extent_op *extent_op;
3108 extent_op = btrfs_alloc_delayed_extent_op();
3112 extent_op->flags_to_set = flags;
3113 extent_op->update_flags = true;
3114 extent_op->update_key = false;
3115 extent_op->is_data = is_data ? true : false;
3116 extent_op->level = level;
3118 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3119 num_bytes, extent_op);
3121 btrfs_free_delayed_extent_op(extent_op);
3125 static noinline int check_delayed_ref(struct btrfs_root *root,
3126 struct btrfs_path *path,
3127 u64 objectid, u64 offset, u64 bytenr)
3129 struct btrfs_delayed_ref_head *head;
3130 struct btrfs_delayed_ref_node *ref;
3131 struct btrfs_delayed_data_ref *data_ref;
3132 struct btrfs_delayed_ref_root *delayed_refs;
3133 struct btrfs_transaction *cur_trans;
3134 struct rb_node *node;
3137 spin_lock(&root->fs_info->trans_lock);
3138 cur_trans = root->fs_info->running_transaction;
3140 refcount_inc(&cur_trans->use_count);
3141 spin_unlock(&root->fs_info->trans_lock);
3145 delayed_refs = &cur_trans->delayed_refs;
3146 spin_lock(&delayed_refs->lock);
3147 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3149 spin_unlock(&delayed_refs->lock);
3150 btrfs_put_transaction(cur_trans);
3154 if (!mutex_trylock(&head->mutex)) {
3155 refcount_inc(&head->refs);
3156 spin_unlock(&delayed_refs->lock);
3158 btrfs_release_path(path);
3161 * Mutex was contended, block until it's released and let
3164 mutex_lock(&head->mutex);
3165 mutex_unlock(&head->mutex);
3166 btrfs_put_delayed_ref_head(head);
3167 btrfs_put_transaction(cur_trans);
3170 spin_unlock(&delayed_refs->lock);
3172 spin_lock(&head->lock);
3174 * XXX: We should replace this with a proper search function in the
3177 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3178 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3179 /* If it's a shared ref we know a cross reference exists */
3180 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3185 data_ref = btrfs_delayed_node_to_data_ref(ref);
3188 * If our ref doesn't match the one we're currently looking at
3189 * then we have a cross reference.
3191 if (data_ref->root != root->root_key.objectid ||
3192 data_ref->objectid != objectid ||
3193 data_ref->offset != offset) {
3198 spin_unlock(&head->lock);
3199 mutex_unlock(&head->mutex);
3200 btrfs_put_transaction(cur_trans);
3204 static noinline int check_committed_ref(struct btrfs_root *root,
3205 struct btrfs_path *path,
3206 u64 objectid, u64 offset, u64 bytenr)
3208 struct btrfs_fs_info *fs_info = root->fs_info;
3209 struct btrfs_root *extent_root = fs_info->extent_root;
3210 struct extent_buffer *leaf;
3211 struct btrfs_extent_data_ref *ref;
3212 struct btrfs_extent_inline_ref *iref;
3213 struct btrfs_extent_item *ei;
3214 struct btrfs_key key;
3219 key.objectid = bytenr;
3220 key.offset = (u64)-1;
3221 key.type = BTRFS_EXTENT_ITEM_KEY;
3223 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3226 BUG_ON(ret == 0); /* Corruption */
3229 if (path->slots[0] == 0)
3233 leaf = path->nodes[0];
3234 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3236 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3240 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3241 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3242 if (item_size < sizeof(*ei)) {
3243 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3247 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3249 if (item_size != sizeof(*ei) +
3250 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3253 if (btrfs_extent_generation(leaf, ei) <=
3254 btrfs_root_last_snapshot(&root->root_item))
3257 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3259 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3260 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3263 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3264 if (btrfs_extent_refs(leaf, ei) !=
3265 btrfs_extent_data_ref_count(leaf, ref) ||
3266 btrfs_extent_data_ref_root(leaf, ref) !=
3267 root->root_key.objectid ||
3268 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3269 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3277 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3280 struct btrfs_path *path;
3284 path = btrfs_alloc_path();
3289 ret = check_committed_ref(root, path, objectid,
3291 if (ret && ret != -ENOENT)
3294 ret2 = check_delayed_ref(root, path, objectid,
3296 } while (ret2 == -EAGAIN);
3298 if (ret2 && ret2 != -ENOENT) {
3303 if (ret != -ENOENT || ret2 != -ENOENT)
3306 btrfs_free_path(path);
3307 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3312 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3313 struct btrfs_root *root,
3314 struct extent_buffer *buf,
3315 int full_backref, int inc)
3317 struct btrfs_fs_info *fs_info = root->fs_info;
3323 struct btrfs_key key;
3324 struct btrfs_file_extent_item *fi;
3328 int (*process_func)(struct btrfs_trans_handle *,
3329 struct btrfs_root *,
3330 u64, u64, u64, u64, u64, u64);
3333 if (btrfs_is_testing(fs_info))
3336 ref_root = btrfs_header_owner(buf);
3337 nritems = btrfs_header_nritems(buf);
3338 level = btrfs_header_level(buf);
3340 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3344 process_func = btrfs_inc_extent_ref;
3346 process_func = btrfs_free_extent;
3349 parent = buf->start;
3353 for (i = 0; i < nritems; i++) {
3355 btrfs_item_key_to_cpu(buf, &key, i);
3356 if (key.type != BTRFS_EXTENT_DATA_KEY)
3358 fi = btrfs_item_ptr(buf, i,
3359 struct btrfs_file_extent_item);
3360 if (btrfs_file_extent_type(buf, fi) ==
3361 BTRFS_FILE_EXTENT_INLINE)
3363 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3367 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3368 key.offset -= btrfs_file_extent_offset(buf, fi);
3369 ret = process_func(trans, root, bytenr, num_bytes,
3370 parent, ref_root, key.objectid,
3375 bytenr = btrfs_node_blockptr(buf, i);
3376 num_bytes = fs_info->nodesize;
3377 ret = process_func(trans, root, bytenr, num_bytes,
3378 parent, ref_root, level - 1, 0);
3388 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3389 struct extent_buffer *buf, int full_backref)
3391 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3394 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3395 struct extent_buffer *buf, int full_backref)
3397 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3400 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3401 struct btrfs_fs_info *fs_info,
3402 struct btrfs_path *path,
3403 struct btrfs_block_group_cache *cache)
3406 struct btrfs_root *extent_root = fs_info->extent_root;
3408 struct extent_buffer *leaf;
3410 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3417 leaf = path->nodes[0];
3418 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3419 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3420 btrfs_mark_buffer_dirty(leaf);
3422 btrfs_release_path(path);
3427 static struct btrfs_block_group_cache *
3428 next_block_group(struct btrfs_fs_info *fs_info,
3429 struct btrfs_block_group_cache *cache)
3431 struct rb_node *node;
3433 spin_lock(&fs_info->block_group_cache_lock);
3435 /* If our block group was removed, we need a full search. */
3436 if (RB_EMPTY_NODE(&cache->cache_node)) {
3437 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3439 spin_unlock(&fs_info->block_group_cache_lock);
3440 btrfs_put_block_group(cache);
3441 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3443 node = rb_next(&cache->cache_node);
3444 btrfs_put_block_group(cache);
3446 cache = rb_entry(node, struct btrfs_block_group_cache,
3448 btrfs_get_block_group(cache);
3451 spin_unlock(&fs_info->block_group_cache_lock);
3455 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3456 struct btrfs_trans_handle *trans,
3457 struct btrfs_path *path)
3459 struct btrfs_fs_info *fs_info = block_group->fs_info;
3460 struct btrfs_root *root = fs_info->tree_root;
3461 struct inode *inode = NULL;
3462 struct extent_changeset *data_reserved = NULL;
3464 int dcs = BTRFS_DC_ERROR;
3470 * If this block group is smaller than 100 megs don't bother caching the
3473 if (block_group->key.offset < (100 * SZ_1M)) {
3474 spin_lock(&block_group->lock);
3475 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3476 spin_unlock(&block_group->lock);
3483 inode = lookup_free_space_inode(fs_info, block_group, path);
3484 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3485 ret = PTR_ERR(inode);
3486 btrfs_release_path(path);
3490 if (IS_ERR(inode)) {
3494 if (block_group->ro)
3497 ret = create_free_space_inode(fs_info, trans, block_group,
3505 * We want to set the generation to 0, that way if anything goes wrong
3506 * from here on out we know not to trust this cache when we load up next
3509 BTRFS_I(inode)->generation = 0;
3510 ret = btrfs_update_inode(trans, root, inode);
3513 * So theoretically we could recover from this, simply set the
3514 * super cache generation to 0 so we know to invalidate the
3515 * cache, but then we'd have to keep track of the block groups
3516 * that fail this way so we know we _have_ to reset this cache
3517 * before the next commit or risk reading stale cache. So to
3518 * limit our exposure to horrible edge cases lets just abort the
3519 * transaction, this only happens in really bad situations
3522 btrfs_abort_transaction(trans, ret);
3527 /* We've already setup this transaction, go ahead and exit */
3528 if (block_group->cache_generation == trans->transid &&
3529 i_size_read(inode)) {
3530 dcs = BTRFS_DC_SETUP;
3534 if (i_size_read(inode) > 0) {
3535 ret = btrfs_check_trunc_cache_free_space(fs_info,
3536 &fs_info->global_block_rsv);
3540 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3545 spin_lock(&block_group->lock);
3546 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3547 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3549 * don't bother trying to write stuff out _if_
3550 * a) we're not cached,
3551 * b) we're with nospace_cache mount option,
3552 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3554 dcs = BTRFS_DC_WRITTEN;
3555 spin_unlock(&block_group->lock);
3558 spin_unlock(&block_group->lock);
3561 * We hit an ENOSPC when setting up the cache in this transaction, just
3562 * skip doing the setup, we've already cleared the cache so we're safe.
3564 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3570 * Try to preallocate enough space based on how big the block group is.
3571 * Keep in mind this has to include any pinned space which could end up
3572 * taking up quite a bit since it's not folded into the other space
3575 num_pages = div_u64(block_group->key.offset, SZ_256M);
3580 num_pages *= PAGE_SIZE;
3582 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3586 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3587 num_pages, num_pages,
3590 * Our cache requires contiguous chunks so that we don't modify a bunch
3591 * of metadata or split extents when writing the cache out, which means
3592 * we can enospc if we are heavily fragmented in addition to just normal
3593 * out of space conditions. So if we hit this just skip setting up any
3594 * other block groups for this transaction, maybe we'll unpin enough
3595 * space the next time around.
3598 dcs = BTRFS_DC_SETUP;
3599 else if (ret == -ENOSPC)
3600 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3605 btrfs_release_path(path);
3607 spin_lock(&block_group->lock);
3608 if (!ret && dcs == BTRFS_DC_SETUP)
3609 block_group->cache_generation = trans->transid;
3610 block_group->disk_cache_state = dcs;
3611 spin_unlock(&block_group->lock);
3613 extent_changeset_free(data_reserved);
3617 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3618 struct btrfs_fs_info *fs_info)
3620 struct btrfs_block_group_cache *cache, *tmp;
3621 struct btrfs_transaction *cur_trans = trans->transaction;
3622 struct btrfs_path *path;
3624 if (list_empty(&cur_trans->dirty_bgs) ||
3625 !btrfs_test_opt(fs_info, SPACE_CACHE))
3628 path = btrfs_alloc_path();
3632 /* Could add new block groups, use _safe just in case */
3633 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3635 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3636 cache_save_setup(cache, trans, path);
3639 btrfs_free_path(path);
3644 * transaction commit does final block group cache writeback during a
3645 * critical section where nothing is allowed to change the FS. This is
3646 * required in order for the cache to actually match the block group,
3647 * but can introduce a lot of latency into the commit.
3649 * So, btrfs_start_dirty_block_groups is here to kick off block group
3650 * cache IO. There's a chance we'll have to redo some of it if the
3651 * block group changes again during the commit, but it greatly reduces
3652 * the commit latency by getting rid of the easy block groups while
3653 * we're still allowing others to join the commit.
3655 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3657 struct btrfs_fs_info *fs_info = trans->fs_info;
3658 struct btrfs_block_group_cache *cache;
3659 struct btrfs_transaction *cur_trans = trans->transaction;
3662 struct btrfs_path *path = NULL;
3664 struct list_head *io = &cur_trans->io_bgs;
3665 int num_started = 0;
3668 spin_lock(&cur_trans->dirty_bgs_lock);
3669 if (list_empty(&cur_trans->dirty_bgs)) {
3670 spin_unlock(&cur_trans->dirty_bgs_lock);
3673 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3674 spin_unlock(&cur_trans->dirty_bgs_lock);
3678 * make sure all the block groups on our dirty list actually
3681 btrfs_create_pending_block_groups(trans);
3684 path = btrfs_alloc_path();
3690 * cache_write_mutex is here only to save us from balance or automatic
3691 * removal of empty block groups deleting this block group while we are
3692 * writing out the cache
3694 mutex_lock(&trans->transaction->cache_write_mutex);
3695 while (!list_empty(&dirty)) {
3696 cache = list_first_entry(&dirty,
3697 struct btrfs_block_group_cache,
3700 * this can happen if something re-dirties a block
3701 * group that is already under IO. Just wait for it to
3702 * finish and then do it all again
3704 if (!list_empty(&cache->io_list)) {
3705 list_del_init(&cache->io_list);
3706 btrfs_wait_cache_io(trans, cache, path);
3707 btrfs_put_block_group(cache);
3712 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3713 * if it should update the cache_state. Don't delete
3714 * until after we wait.
3716 * Since we're not running in the commit critical section
3717 * we need the dirty_bgs_lock to protect from update_block_group
3719 spin_lock(&cur_trans->dirty_bgs_lock);
3720 list_del_init(&cache->dirty_list);
3721 spin_unlock(&cur_trans->dirty_bgs_lock);
3725 cache_save_setup(cache, trans, path);
3727 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3728 cache->io_ctl.inode = NULL;
3729 ret = btrfs_write_out_cache(fs_info, trans,
3731 if (ret == 0 && cache->io_ctl.inode) {
3736 * The cache_write_mutex is protecting the
3737 * io_list, also refer to the definition of
3738 * btrfs_transaction::io_bgs for more details
3740 list_add_tail(&cache->io_list, io);
3743 * if we failed to write the cache, the
3744 * generation will be bad and life goes on
3750 ret = write_one_cache_group(trans, fs_info,
3753 * Our block group might still be attached to the list
3754 * of new block groups in the transaction handle of some
3755 * other task (struct btrfs_trans_handle->new_bgs). This
3756 * means its block group item isn't yet in the extent
3757 * tree. If this happens ignore the error, as we will
3758 * try again later in the critical section of the
3759 * transaction commit.
3761 if (ret == -ENOENT) {
3763 spin_lock(&cur_trans->dirty_bgs_lock);
3764 if (list_empty(&cache->dirty_list)) {
3765 list_add_tail(&cache->dirty_list,
3766 &cur_trans->dirty_bgs);
3767 btrfs_get_block_group(cache);
3769 spin_unlock(&cur_trans->dirty_bgs_lock);
3771 btrfs_abort_transaction(trans, ret);
3775 /* if its not on the io list, we need to put the block group */
3777 btrfs_put_block_group(cache);
3783 * Avoid blocking other tasks for too long. It might even save
3784 * us from writing caches for block groups that are going to be
3787 mutex_unlock(&trans->transaction->cache_write_mutex);
3788 mutex_lock(&trans->transaction->cache_write_mutex);
3790 mutex_unlock(&trans->transaction->cache_write_mutex);
3793 * go through delayed refs for all the stuff we've just kicked off
3794 * and then loop back (just once)
3796 ret = btrfs_run_delayed_refs(trans, 0);
3797 if (!ret && loops == 0) {
3799 spin_lock(&cur_trans->dirty_bgs_lock);
3800 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3802 * dirty_bgs_lock protects us from concurrent block group
3803 * deletes too (not just cache_write_mutex).
3805 if (!list_empty(&dirty)) {
3806 spin_unlock(&cur_trans->dirty_bgs_lock);
3809 spin_unlock(&cur_trans->dirty_bgs_lock);
3810 } else if (ret < 0) {
3811 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3814 btrfs_free_path(path);
3818 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3819 struct btrfs_fs_info *fs_info)
3821 struct btrfs_block_group_cache *cache;
3822 struct btrfs_transaction *cur_trans = trans->transaction;
3825 struct btrfs_path *path;
3826 struct list_head *io = &cur_trans->io_bgs;
3827 int num_started = 0;
3829 path = btrfs_alloc_path();
3834 * Even though we are in the critical section of the transaction commit,
3835 * we can still have concurrent tasks adding elements to this
3836 * transaction's list of dirty block groups. These tasks correspond to
3837 * endio free space workers started when writeback finishes for a
3838 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3839 * allocate new block groups as a result of COWing nodes of the root
3840 * tree when updating the free space inode. The writeback for the space
3841 * caches is triggered by an earlier call to
3842 * btrfs_start_dirty_block_groups() and iterations of the following
3844 * Also we want to do the cache_save_setup first and then run the
3845 * delayed refs to make sure we have the best chance at doing this all
3848 spin_lock(&cur_trans->dirty_bgs_lock);
3849 while (!list_empty(&cur_trans->dirty_bgs)) {
3850 cache = list_first_entry(&cur_trans->dirty_bgs,
3851 struct btrfs_block_group_cache,
3855 * this can happen if cache_save_setup re-dirties a block
3856 * group that is already under IO. Just wait for it to
3857 * finish and then do it all again
3859 if (!list_empty(&cache->io_list)) {
3860 spin_unlock(&cur_trans->dirty_bgs_lock);
3861 list_del_init(&cache->io_list);
3862 btrfs_wait_cache_io(trans, cache, path);
3863 btrfs_put_block_group(cache);
3864 spin_lock(&cur_trans->dirty_bgs_lock);
3868 * don't remove from the dirty list until after we've waited
3871 list_del_init(&cache->dirty_list);
3872 spin_unlock(&cur_trans->dirty_bgs_lock);
3875 cache_save_setup(cache, trans, path);
3878 ret = btrfs_run_delayed_refs(trans,
3879 (unsigned long) -1);
3881 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3882 cache->io_ctl.inode = NULL;
3883 ret = btrfs_write_out_cache(fs_info, trans,
3885 if (ret == 0 && cache->io_ctl.inode) {
3888 list_add_tail(&cache->io_list, io);
3891 * if we failed to write the cache, the
3892 * generation will be bad and life goes on
3898 ret = write_one_cache_group(trans, fs_info,
3901 * One of the free space endio workers might have
3902 * created a new block group while updating a free space
3903 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3904 * and hasn't released its transaction handle yet, in
3905 * which case the new block group is still attached to
3906 * its transaction handle and its creation has not
3907 * finished yet (no block group item in the extent tree
3908 * yet, etc). If this is the case, wait for all free
3909 * space endio workers to finish and retry. This is a
3910 * a very rare case so no need for a more efficient and
3913 if (ret == -ENOENT) {
3914 wait_event(cur_trans->writer_wait,
3915 atomic_read(&cur_trans->num_writers) == 1);
3916 ret = write_one_cache_group(trans, fs_info,
3920 btrfs_abort_transaction(trans, ret);
3923 /* if its not on the io list, we need to put the block group */
3925 btrfs_put_block_group(cache);
3926 spin_lock(&cur_trans->dirty_bgs_lock);
3928 spin_unlock(&cur_trans->dirty_bgs_lock);
3931 * Refer to the definition of io_bgs member for details why it's safe
3932 * to use it without any locking
3934 while (!list_empty(io)) {
3935 cache = list_first_entry(io, struct btrfs_block_group_cache,
3937 list_del_init(&cache->io_list);
3938 btrfs_wait_cache_io(trans, cache, path);
3939 btrfs_put_block_group(cache);
3942 btrfs_free_path(path);
3946 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3948 struct btrfs_block_group_cache *block_group;
3951 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3952 if (!block_group || block_group->ro)
3955 btrfs_put_block_group(block_group);
3959 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3961 struct btrfs_block_group_cache *bg;
3964 bg = btrfs_lookup_block_group(fs_info, bytenr);
3968 spin_lock(&bg->lock);
3972 atomic_inc(&bg->nocow_writers);
3973 spin_unlock(&bg->lock);
3975 /* no put on block group, done by btrfs_dec_nocow_writers */
3977 btrfs_put_block_group(bg);
3983 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3985 struct btrfs_block_group_cache *bg;
3987 bg = btrfs_lookup_block_group(fs_info, bytenr);
3989 if (atomic_dec_and_test(&bg->nocow_writers))
3990 wake_up_var(&bg->nocow_writers);
3992 * Once for our lookup and once for the lookup done by a previous call
3993 * to btrfs_inc_nocow_writers()
3995 btrfs_put_block_group(bg);
3996 btrfs_put_block_group(bg);
3999 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4001 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
4004 static const char *alloc_name(u64 flags)
4007 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4009 case BTRFS_BLOCK_GROUP_METADATA:
4011 case BTRFS_BLOCK_GROUP_DATA:
4013 case BTRFS_BLOCK_GROUP_SYSTEM:
4017 return "invalid-combination";
4021 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
4024 struct btrfs_space_info *space_info;
4028 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4032 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4039 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4040 INIT_LIST_HEAD(&space_info->block_groups[i]);
4041 init_rwsem(&space_info->groups_sem);
4042 spin_lock_init(&space_info->lock);
4043 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4044 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4045 init_waitqueue_head(&space_info->wait);
4046 INIT_LIST_HEAD(&space_info->ro_bgs);
4047 INIT_LIST_HEAD(&space_info->tickets);
4048 INIT_LIST_HEAD(&space_info->priority_tickets);
4050 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4051 info->space_info_kobj, "%s",
4052 alloc_name(space_info->flags));
4054 percpu_counter_destroy(&space_info->total_bytes_pinned);
4059 list_add_rcu(&space_info->list, &info->space_info);
4060 if (flags & BTRFS_BLOCK_GROUP_DATA)
4061 info->data_sinfo = space_info;
4066 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4067 u64 total_bytes, u64 bytes_used,
4069 struct btrfs_space_info **space_info)
4071 struct btrfs_space_info *found;
4074 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4075 BTRFS_BLOCK_GROUP_RAID10))
4080 found = __find_space_info(info, flags);
4082 spin_lock(&found->lock);
4083 found->total_bytes += total_bytes;
4084 found->disk_total += total_bytes * factor;
4085 found->bytes_used += bytes_used;
4086 found->disk_used += bytes_used * factor;
4087 found->bytes_readonly += bytes_readonly;
4088 if (total_bytes > 0)
4090 space_info_add_new_bytes(info, found, total_bytes -
4091 bytes_used - bytes_readonly);
4092 spin_unlock(&found->lock);
4093 *space_info = found;
4096 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4098 u64 extra_flags = chunk_to_extended(flags) &
4099 BTRFS_EXTENDED_PROFILE_MASK;
4101 write_seqlock(&fs_info->profiles_lock);
4102 if (flags & BTRFS_BLOCK_GROUP_DATA)
4103 fs_info->avail_data_alloc_bits |= extra_flags;
4104 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4105 fs_info->avail_metadata_alloc_bits |= extra_flags;
4106 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4107 fs_info->avail_system_alloc_bits |= extra_flags;
4108 write_sequnlock(&fs_info->profiles_lock);
4112 * returns target flags in extended format or 0 if restripe for this
4113 * chunk_type is not in progress
4115 * should be called with balance_lock held
4117 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4119 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4125 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4126 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4127 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4128 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4129 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4130 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4131 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4132 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4133 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4140 * @flags: available profiles in extended format (see ctree.h)
4142 * Returns reduced profile in chunk format. If profile changing is in
4143 * progress (either running or paused) picks the target profile (if it's
4144 * already available), otherwise falls back to plain reducing.
4146 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4148 u64 num_devices = fs_info->fs_devices->rw_devices;
4154 * see if restripe for this chunk_type is in progress, if so
4155 * try to reduce to the target profile
4157 spin_lock(&fs_info->balance_lock);
4158 target = get_restripe_target(fs_info, flags);
4160 /* pick target profile only if it's already available */
4161 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4162 spin_unlock(&fs_info->balance_lock);
4163 return extended_to_chunk(target);
4166 spin_unlock(&fs_info->balance_lock);
4168 /* First, mask out the RAID levels which aren't possible */
4169 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4170 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4171 allowed |= btrfs_raid_array[raid_type].bg_flag;
4175 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4176 allowed = BTRFS_BLOCK_GROUP_RAID6;
4177 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4178 allowed = BTRFS_BLOCK_GROUP_RAID5;
4179 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4180 allowed = BTRFS_BLOCK_GROUP_RAID10;
4181 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4182 allowed = BTRFS_BLOCK_GROUP_RAID1;
4183 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4184 allowed = BTRFS_BLOCK_GROUP_RAID0;
4186 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4188 return extended_to_chunk(flags | allowed);
4191 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4198 seq = read_seqbegin(&fs_info->profiles_lock);
4200 if (flags & BTRFS_BLOCK_GROUP_DATA)
4201 flags |= fs_info->avail_data_alloc_bits;
4202 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4203 flags |= fs_info->avail_system_alloc_bits;
4204 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4205 flags |= fs_info->avail_metadata_alloc_bits;
4206 } while (read_seqretry(&fs_info->profiles_lock, seq));
4208 return btrfs_reduce_alloc_profile(fs_info, flags);
4211 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4213 struct btrfs_fs_info *fs_info = root->fs_info;
4218 flags = BTRFS_BLOCK_GROUP_DATA;
4219 else if (root == fs_info->chunk_root)
4220 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4222 flags = BTRFS_BLOCK_GROUP_METADATA;
4224 ret = get_alloc_profile(fs_info, flags);
4228 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4230 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4233 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4235 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4238 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4240 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4243 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4244 bool may_use_included)
4247 return s_info->bytes_used + s_info->bytes_reserved +
4248 s_info->bytes_pinned + s_info->bytes_readonly +
4249 (may_use_included ? s_info->bytes_may_use : 0);
4252 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4254 struct btrfs_root *root = inode->root;
4255 struct btrfs_fs_info *fs_info = root->fs_info;
4256 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4259 int need_commit = 2;
4260 int have_pinned_space;
4262 /* make sure bytes are sectorsize aligned */
4263 bytes = ALIGN(bytes, fs_info->sectorsize);
4265 if (btrfs_is_free_space_inode(inode)) {
4267 ASSERT(current->journal_info);
4271 /* make sure we have enough space to handle the data first */
4272 spin_lock(&data_sinfo->lock);
4273 used = btrfs_space_info_used(data_sinfo, true);
4275 if (used + bytes > data_sinfo->total_bytes) {
4276 struct btrfs_trans_handle *trans;
4279 * if we don't have enough free bytes in this space then we need
4280 * to alloc a new chunk.
4282 if (!data_sinfo->full) {
4285 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4286 spin_unlock(&data_sinfo->lock);
4288 alloc_target = btrfs_data_alloc_profile(fs_info);
4290 * It is ugly that we don't call nolock join
4291 * transaction for the free space inode case here.
4292 * But it is safe because we only do the data space
4293 * reservation for the free space cache in the
4294 * transaction context, the common join transaction
4295 * just increase the counter of the current transaction
4296 * handler, doesn't try to acquire the trans_lock of
4299 trans = btrfs_join_transaction(root);
4301 return PTR_ERR(trans);
4303 ret = do_chunk_alloc(trans, alloc_target,
4304 CHUNK_ALLOC_NO_FORCE);
4305 btrfs_end_transaction(trans);
4310 have_pinned_space = 1;
4319 * If we don't have enough pinned space to deal with this
4320 * allocation, and no removed chunk in current transaction,
4321 * don't bother committing the transaction.
4323 have_pinned_space = percpu_counter_compare(
4324 &data_sinfo->total_bytes_pinned,
4325 used + bytes - data_sinfo->total_bytes);
4326 spin_unlock(&data_sinfo->lock);
4328 /* commit the current transaction and try again */
4333 if (need_commit > 0) {
4334 btrfs_start_delalloc_roots(fs_info, -1);
4335 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4339 trans = btrfs_join_transaction(root);
4341 return PTR_ERR(trans);
4342 if (have_pinned_space >= 0 ||
4343 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4344 &trans->transaction->flags) ||
4346 ret = btrfs_commit_transaction(trans);
4350 * The cleaner kthread might still be doing iput
4351 * operations. Wait for it to finish so that
4352 * more space is released.
4354 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4355 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4358 btrfs_end_transaction(trans);
4362 trace_btrfs_space_reservation(fs_info,
4363 "space_info:enospc",
4364 data_sinfo->flags, bytes, 1);
4367 data_sinfo->bytes_may_use += bytes;
4368 trace_btrfs_space_reservation(fs_info, "space_info",
4369 data_sinfo->flags, bytes, 1);
4370 spin_unlock(&data_sinfo->lock);
4375 int btrfs_check_data_free_space(struct inode *inode,
4376 struct extent_changeset **reserved, u64 start, u64 len)
4378 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4381 /* align the range */
4382 len = round_up(start + len, fs_info->sectorsize) -
4383 round_down(start, fs_info->sectorsize);
4384 start = round_down(start, fs_info->sectorsize);
4386 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4390 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4391 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4393 btrfs_free_reserved_data_space_noquota(inode, start, len);
4400 * Called if we need to clear a data reservation for this inode
4401 * Normally in a error case.
4403 * This one will *NOT* use accurate qgroup reserved space API, just for case
4404 * which we can't sleep and is sure it won't affect qgroup reserved space.
4405 * Like clear_bit_hook().
4407 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4410 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4411 struct btrfs_space_info *data_sinfo;
4413 /* Make sure the range is aligned to sectorsize */
4414 len = round_up(start + len, fs_info->sectorsize) -
4415 round_down(start, fs_info->sectorsize);
4416 start = round_down(start, fs_info->sectorsize);
4418 data_sinfo = fs_info->data_sinfo;
4419 spin_lock(&data_sinfo->lock);
4420 if (WARN_ON(data_sinfo->bytes_may_use < len))
4421 data_sinfo->bytes_may_use = 0;
4423 data_sinfo->bytes_may_use -= len;
4424 trace_btrfs_space_reservation(fs_info, "space_info",
4425 data_sinfo->flags, len, 0);
4426 spin_unlock(&data_sinfo->lock);
4430 * Called if we need to clear a data reservation for this inode
4431 * Normally in a error case.
4433 * This one will handle the per-inode data rsv map for accurate reserved
4436 void btrfs_free_reserved_data_space(struct inode *inode,
4437 struct extent_changeset *reserved, u64 start, u64 len)
4439 struct btrfs_root *root = BTRFS_I(inode)->root;
4441 /* Make sure the range is aligned to sectorsize */
4442 len = round_up(start + len, root->fs_info->sectorsize) -
4443 round_down(start, root->fs_info->sectorsize);
4444 start = round_down(start, root->fs_info->sectorsize);
4446 btrfs_free_reserved_data_space_noquota(inode, start, len);
4447 btrfs_qgroup_free_data(inode, reserved, start, len);
4450 static void force_metadata_allocation(struct btrfs_fs_info *info)
4452 struct list_head *head = &info->space_info;
4453 struct btrfs_space_info *found;
4456 list_for_each_entry_rcu(found, head, list) {
4457 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4458 found->force_alloc = CHUNK_ALLOC_FORCE;
4463 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4465 return (global->size << 1);
4468 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4469 struct btrfs_space_info *sinfo, int force)
4471 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4472 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4475 if (force == CHUNK_ALLOC_FORCE)
4479 * We need to take into account the global rsv because for all intents
4480 * and purposes it's used space. Don't worry about locking the
4481 * global_rsv, it doesn't change except when the transaction commits.
4483 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4484 bytes_used += calc_global_rsv_need_space(global_rsv);
4487 * in limited mode, we want to have some free space up to
4488 * about 1% of the FS size.
4490 if (force == CHUNK_ALLOC_LIMITED) {
4491 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4492 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4494 if (sinfo->total_bytes - bytes_used < thresh)
4498 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4503 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4507 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4508 BTRFS_BLOCK_GROUP_RAID0 |
4509 BTRFS_BLOCK_GROUP_RAID5 |
4510 BTRFS_BLOCK_GROUP_RAID6))
4511 num_dev = fs_info->fs_devices->rw_devices;
4512 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4515 num_dev = 1; /* DUP or single */
4521 * If @is_allocation is true, reserve space in the system space info necessary
4522 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4525 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4527 struct btrfs_fs_info *fs_info = trans->fs_info;
4528 struct btrfs_space_info *info;
4535 * Needed because we can end up allocating a system chunk and for an
4536 * atomic and race free space reservation in the chunk block reserve.
4538 lockdep_assert_held(&fs_info->chunk_mutex);
4540 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4541 spin_lock(&info->lock);
4542 left = info->total_bytes - btrfs_space_info_used(info, true);
4543 spin_unlock(&info->lock);
4545 num_devs = get_profile_num_devs(fs_info, type);
4547 /* num_devs device items to update and 1 chunk item to add or remove */
4548 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4549 btrfs_calc_trans_metadata_size(fs_info, 1);
4551 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4552 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4553 left, thresh, type);
4554 dump_space_info(fs_info, info, 0, 0);
4557 if (left < thresh) {
4558 u64 flags = btrfs_system_alloc_profile(fs_info);
4561 * Ignore failure to create system chunk. We might end up not
4562 * needing it, as we might not need to COW all nodes/leafs from
4563 * the paths we visit in the chunk tree (they were already COWed
4564 * or created in the current transaction for example).
4566 ret = btrfs_alloc_chunk(trans, flags);
4570 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4571 &fs_info->chunk_block_rsv,
4572 thresh, BTRFS_RESERVE_NO_FLUSH);
4574 trans->chunk_bytes_reserved += thresh;
4579 * If force is CHUNK_ALLOC_FORCE:
4580 * - return 1 if it successfully allocates a chunk,
4581 * - return errors including -ENOSPC otherwise.
4582 * If force is NOT CHUNK_ALLOC_FORCE:
4583 * - return 0 if it doesn't need to allocate a new chunk,
4584 * - return 1 if it successfully allocates a chunk,
4585 * - return errors including -ENOSPC otherwise.
4587 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4590 struct btrfs_fs_info *fs_info = trans->fs_info;
4591 struct btrfs_space_info *space_info;
4592 int wait_for_alloc = 0;
4595 /* Don't re-enter if we're already allocating a chunk */
4596 if (trans->allocating_chunk)
4599 space_info = __find_space_info(fs_info, flags);
4603 spin_lock(&space_info->lock);
4604 if (force < space_info->force_alloc)
4605 force = space_info->force_alloc;
4606 if (space_info->full) {
4607 if (should_alloc_chunk(fs_info, space_info, force))
4611 spin_unlock(&space_info->lock);
4615 if (!should_alloc_chunk(fs_info, space_info, force)) {
4616 spin_unlock(&space_info->lock);
4618 } else if (space_info->chunk_alloc) {
4621 space_info->chunk_alloc = 1;
4624 spin_unlock(&space_info->lock);
4626 mutex_lock(&fs_info->chunk_mutex);
4629 * The chunk_mutex is held throughout the entirety of a chunk
4630 * allocation, so once we've acquired the chunk_mutex we know that the
4631 * other guy is done and we need to recheck and see if we should
4634 if (wait_for_alloc) {
4635 mutex_unlock(&fs_info->chunk_mutex);
4641 trans->allocating_chunk = true;
4644 * If we have mixed data/metadata chunks we want to make sure we keep
4645 * allocating mixed chunks instead of individual chunks.
4647 if (btrfs_mixed_space_info(space_info))
4648 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4651 * if we're doing a data chunk, go ahead and make sure that
4652 * we keep a reasonable number of metadata chunks allocated in the
4655 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4656 fs_info->data_chunk_allocations++;
4657 if (!(fs_info->data_chunk_allocations %
4658 fs_info->metadata_ratio))
4659 force_metadata_allocation(fs_info);
4663 * Check if we have enough space in SYSTEM chunk because we may need
4664 * to update devices.
4666 check_system_chunk(trans, flags);
4668 ret = btrfs_alloc_chunk(trans, flags);
4669 trans->allocating_chunk = false;
4671 spin_lock(&space_info->lock);
4674 space_info->full = 1;
4681 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4683 space_info->chunk_alloc = 0;
4684 spin_unlock(&space_info->lock);
4685 mutex_unlock(&fs_info->chunk_mutex);
4687 * When we allocate a new chunk we reserve space in the chunk block
4688 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4689 * add new nodes/leafs to it if we end up needing to do it when
4690 * inserting the chunk item and updating device items as part of the
4691 * second phase of chunk allocation, performed by
4692 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4693 * large number of new block groups to create in our transaction
4694 * handle's new_bgs list to avoid exhausting the chunk block reserve
4695 * in extreme cases - like having a single transaction create many new
4696 * block groups when starting to write out the free space caches of all
4697 * the block groups that were made dirty during the lifetime of the
4700 if (trans->can_flush_pending_bgs &&
4701 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4702 btrfs_create_pending_block_groups(trans);
4703 btrfs_trans_release_chunk_metadata(trans);
4708 static int can_overcommit(struct btrfs_fs_info *fs_info,
4709 struct btrfs_space_info *space_info, u64 bytes,
4710 enum btrfs_reserve_flush_enum flush,
4713 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4719 /* Don't overcommit when in mixed mode. */
4720 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4724 profile = btrfs_system_alloc_profile(fs_info);
4726 profile = btrfs_metadata_alloc_profile(fs_info);
4728 used = btrfs_space_info_used(space_info, false);
4731 * We only want to allow over committing if we have lots of actual space
4732 * free, but if we don't have enough space to handle the global reserve
4733 * space then we could end up having a real enospc problem when trying
4734 * to allocate a chunk or some other such important allocation.
4736 spin_lock(&global_rsv->lock);
4737 space_size = calc_global_rsv_need_space(global_rsv);
4738 spin_unlock(&global_rsv->lock);
4739 if (used + space_size >= space_info->total_bytes)
4742 used += space_info->bytes_may_use;
4744 avail = atomic64_read(&fs_info->free_chunk_space);
4747 * If we have dup, raid1 or raid10 then only half of the free
4748 * space is actually useable. For raid56, the space info used
4749 * doesn't include the parity drive, so we don't have to
4752 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4753 BTRFS_BLOCK_GROUP_RAID1 |
4754 BTRFS_BLOCK_GROUP_RAID10))
4758 * If we aren't flushing all things, let us overcommit up to
4759 * 1/2th of the space. If we can flush, don't let us overcommit
4760 * too much, let it overcommit up to 1/8 of the space.
4762 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4767 if (used + bytes < space_info->total_bytes + avail)
4772 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4773 unsigned long nr_pages, int nr_items)
4775 struct super_block *sb = fs_info->sb;
4777 if (down_read_trylock(&sb->s_umount)) {
4778 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4779 up_read(&sb->s_umount);
4782 * We needn't worry the filesystem going from r/w to r/o though
4783 * we don't acquire ->s_umount mutex, because the filesystem
4784 * should guarantee the delalloc inodes list be empty after
4785 * the filesystem is readonly(all dirty pages are written to
4788 btrfs_start_delalloc_roots(fs_info, nr_items);
4789 if (!current->journal_info)
4790 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4794 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4800 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4801 nr = div64_u64(to_reclaim, bytes);
4807 #define EXTENT_SIZE_PER_ITEM SZ_256K
4810 * shrink metadata reservation for delalloc
4812 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4813 u64 orig, bool wait_ordered)
4815 struct btrfs_space_info *space_info;
4816 struct btrfs_trans_handle *trans;
4821 unsigned long nr_pages;
4824 /* Calc the number of the pages we need flush for space reservation */
4825 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4826 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4828 trans = (struct btrfs_trans_handle *)current->journal_info;
4829 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4831 delalloc_bytes = percpu_counter_sum_positive(
4832 &fs_info->delalloc_bytes);
4833 if (delalloc_bytes == 0) {
4837 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4842 while (delalloc_bytes && loops < 3) {
4843 max_reclaim = min(delalloc_bytes, to_reclaim);
4844 nr_pages = max_reclaim >> PAGE_SHIFT;
4845 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4847 * We need to wait for the async pages to actually start before
4850 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4854 if (max_reclaim <= nr_pages)
4857 max_reclaim -= nr_pages;
4859 wait_event(fs_info->async_submit_wait,
4860 atomic_read(&fs_info->async_delalloc_pages) <=
4863 spin_lock(&space_info->lock);
4864 if (list_empty(&space_info->tickets) &&
4865 list_empty(&space_info->priority_tickets)) {
4866 spin_unlock(&space_info->lock);
4869 spin_unlock(&space_info->lock);
4872 if (wait_ordered && !trans) {
4873 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4875 time_left = schedule_timeout_killable(1);
4879 delalloc_bytes = percpu_counter_sum_positive(
4880 &fs_info->delalloc_bytes);
4884 struct reserve_ticket {
4887 struct list_head list;
4888 wait_queue_head_t wait;
4892 * maybe_commit_transaction - possibly commit the transaction if its ok to
4893 * @root - the root we're allocating for
4894 * @bytes - the number of bytes we want to reserve
4895 * @force - force the commit
4897 * This will check to make sure that committing the transaction will actually
4898 * get us somewhere and then commit the transaction if it does. Otherwise it
4899 * will return -ENOSPC.
4901 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4902 struct btrfs_space_info *space_info)
4904 struct reserve_ticket *ticket = NULL;
4905 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4906 struct btrfs_trans_handle *trans;
4909 trans = (struct btrfs_trans_handle *)current->journal_info;
4913 spin_lock(&space_info->lock);
4914 if (!list_empty(&space_info->priority_tickets))
4915 ticket = list_first_entry(&space_info->priority_tickets,
4916 struct reserve_ticket, list);
4917 else if (!list_empty(&space_info->tickets))
4918 ticket = list_first_entry(&space_info->tickets,
4919 struct reserve_ticket, list);
4920 bytes = (ticket) ? ticket->bytes : 0;
4921 spin_unlock(&space_info->lock);
4926 /* See if there is enough pinned space to make this reservation */
4927 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4932 * See if there is some space in the delayed insertion reservation for
4935 if (space_info != delayed_rsv->space_info)
4938 spin_lock(&delayed_rsv->lock);
4939 if (delayed_rsv->size > bytes)
4942 bytes -= delayed_rsv->size;
4943 spin_unlock(&delayed_rsv->lock);
4945 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4951 trans = btrfs_join_transaction(fs_info->extent_root);
4955 return btrfs_commit_transaction(trans);
4959 * Try to flush some data based on policy set by @state. This is only advisory
4960 * and may fail for various reasons. The caller is supposed to examine the
4961 * state of @space_info to detect the outcome.
4963 static void flush_space(struct btrfs_fs_info *fs_info,
4964 struct btrfs_space_info *space_info, u64 num_bytes,
4967 struct btrfs_root *root = fs_info->extent_root;
4968 struct btrfs_trans_handle *trans;
4973 case FLUSH_DELAYED_ITEMS_NR:
4974 case FLUSH_DELAYED_ITEMS:
4975 if (state == FLUSH_DELAYED_ITEMS_NR)
4976 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4980 trans = btrfs_join_transaction(root);
4981 if (IS_ERR(trans)) {
4982 ret = PTR_ERR(trans);
4985 ret = btrfs_run_delayed_items_nr(trans, nr);
4986 btrfs_end_transaction(trans);
4988 case FLUSH_DELALLOC:
4989 case FLUSH_DELALLOC_WAIT:
4990 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4991 state == FLUSH_DELALLOC_WAIT);
4994 trans = btrfs_join_transaction(root);
4995 if (IS_ERR(trans)) {
4996 ret = PTR_ERR(trans);
4999 ret = do_chunk_alloc(trans,
5000 btrfs_metadata_alloc_profile(fs_info),
5001 CHUNK_ALLOC_NO_FORCE);
5002 btrfs_end_transaction(trans);
5003 if (ret > 0 || ret == -ENOSPC)
5007 ret = may_commit_transaction(fs_info, space_info);
5014 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5020 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5021 struct btrfs_space_info *space_info,
5024 struct reserve_ticket *ticket;
5029 list_for_each_entry(ticket, &space_info->tickets, list)
5030 to_reclaim += ticket->bytes;
5031 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5032 to_reclaim += ticket->bytes;
5036 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5037 if (can_overcommit(fs_info, space_info, to_reclaim,
5038 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5041 used = btrfs_space_info_used(space_info, true);
5043 if (can_overcommit(fs_info, space_info, SZ_1M,
5044 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5045 expected = div_factor_fine(space_info->total_bytes, 95);
5047 expected = div_factor_fine(space_info->total_bytes, 90);
5049 if (used > expected)
5050 to_reclaim = used - expected;
5053 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5054 space_info->bytes_reserved);
5058 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5059 struct btrfs_space_info *space_info,
5060 u64 used, bool system_chunk)
5062 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5064 /* If we're just plain full then async reclaim just slows us down. */
5065 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5068 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5072 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5073 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5076 static void wake_all_tickets(struct list_head *head)
5078 struct reserve_ticket *ticket;
5080 while (!list_empty(head)) {
5081 ticket = list_first_entry(head, struct reserve_ticket, list);
5082 list_del_init(&ticket->list);
5083 ticket->error = -ENOSPC;
5084 wake_up(&ticket->wait);
5089 * This is for normal flushers, we can wait all goddamned day if we want to. We
5090 * will loop and continuously try to flush as long as we are making progress.
5091 * We count progress as clearing off tickets each time we have to loop.
5093 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5095 struct btrfs_fs_info *fs_info;
5096 struct btrfs_space_info *space_info;
5099 int commit_cycles = 0;
5100 u64 last_tickets_id;
5102 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5103 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5105 spin_lock(&space_info->lock);
5106 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5109 space_info->flush = 0;
5110 spin_unlock(&space_info->lock);
5113 last_tickets_id = space_info->tickets_id;
5114 spin_unlock(&space_info->lock);
5116 flush_state = FLUSH_DELAYED_ITEMS_NR;
5118 flush_space(fs_info, space_info, to_reclaim, flush_state);
5119 spin_lock(&space_info->lock);
5120 if (list_empty(&space_info->tickets)) {
5121 space_info->flush = 0;
5122 spin_unlock(&space_info->lock);
5125 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5128 if (last_tickets_id == space_info->tickets_id) {
5131 last_tickets_id = space_info->tickets_id;
5132 flush_state = FLUSH_DELAYED_ITEMS_NR;
5137 if (flush_state > COMMIT_TRANS) {
5139 if (commit_cycles > 2) {
5140 wake_all_tickets(&space_info->tickets);
5141 space_info->flush = 0;
5143 flush_state = FLUSH_DELAYED_ITEMS_NR;
5146 spin_unlock(&space_info->lock);
5147 } while (flush_state <= COMMIT_TRANS);
5150 void btrfs_init_async_reclaim_work(struct work_struct *work)
5152 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5155 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5156 struct btrfs_space_info *space_info,
5157 struct reserve_ticket *ticket)
5160 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5162 spin_lock(&space_info->lock);
5163 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5166 spin_unlock(&space_info->lock);
5169 spin_unlock(&space_info->lock);
5172 flush_space(fs_info, space_info, to_reclaim, flush_state);
5174 spin_lock(&space_info->lock);
5175 if (ticket->bytes == 0) {
5176 spin_unlock(&space_info->lock);
5179 spin_unlock(&space_info->lock);
5182 * Priority flushers can't wait on delalloc without
5185 if (flush_state == FLUSH_DELALLOC ||
5186 flush_state == FLUSH_DELALLOC_WAIT)
5187 flush_state = ALLOC_CHUNK;
5188 } while (flush_state < COMMIT_TRANS);
5191 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5192 struct btrfs_space_info *space_info,
5193 struct reserve_ticket *ticket, u64 orig_bytes)
5199 spin_lock(&space_info->lock);
5200 while (ticket->bytes > 0 && ticket->error == 0) {
5201 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5206 spin_unlock(&space_info->lock);
5210 finish_wait(&ticket->wait, &wait);
5211 spin_lock(&space_info->lock);
5214 ret = ticket->error;
5215 if (!list_empty(&ticket->list))
5216 list_del_init(&ticket->list);
5217 if (ticket->bytes && ticket->bytes < orig_bytes) {
5218 u64 num_bytes = orig_bytes - ticket->bytes;
5219 space_info->bytes_may_use -= num_bytes;
5220 trace_btrfs_space_reservation(fs_info, "space_info",
5221 space_info->flags, num_bytes, 0);
5223 spin_unlock(&space_info->lock);
5229 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5230 * @root - the root we're allocating for
5231 * @space_info - the space info we want to allocate from
5232 * @orig_bytes - the number of bytes we want
5233 * @flush - whether or not we can flush to make our reservation
5235 * This will reserve orig_bytes number of bytes from the space info associated
5236 * with the block_rsv. If there is not enough space it will make an attempt to
5237 * flush out space to make room. It will do this by flushing delalloc if
5238 * possible or committing the transaction. If flush is 0 then no attempts to
5239 * regain reservations will be made and this will fail if there is not enough
5242 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5243 struct btrfs_space_info *space_info,
5245 enum btrfs_reserve_flush_enum flush,
5248 struct reserve_ticket ticket;
5253 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5255 spin_lock(&space_info->lock);
5257 used = btrfs_space_info_used(space_info, true);
5260 * If we have enough space then hooray, make our reservation and carry
5261 * on. If not see if we can overcommit, and if we can, hooray carry on.
5262 * If not things get more complicated.
5264 if (used + orig_bytes <= space_info->total_bytes) {
5265 space_info->bytes_may_use += orig_bytes;
5266 trace_btrfs_space_reservation(fs_info, "space_info",
5267 space_info->flags, orig_bytes, 1);
5269 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5271 space_info->bytes_may_use += orig_bytes;
5272 trace_btrfs_space_reservation(fs_info, "space_info",
5273 space_info->flags, orig_bytes, 1);
5278 * If we couldn't make a reservation then setup our reservation ticket
5279 * and kick the async worker if it's not already running.
5281 * If we are a priority flusher then we just need to add our ticket to
5282 * the list and we will do our own flushing further down.
5284 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5285 ticket.bytes = orig_bytes;
5287 init_waitqueue_head(&ticket.wait);
5288 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5289 list_add_tail(&ticket.list, &space_info->tickets);
5290 if (!space_info->flush) {
5291 space_info->flush = 1;
5292 trace_btrfs_trigger_flush(fs_info,
5296 queue_work(system_unbound_wq,
5297 &fs_info->async_reclaim_work);
5300 list_add_tail(&ticket.list,
5301 &space_info->priority_tickets);
5303 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5306 * We will do the space reservation dance during log replay,
5307 * which means we won't have fs_info->fs_root set, so don't do
5308 * the async reclaim as we will panic.
5310 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5311 need_do_async_reclaim(fs_info, space_info,
5312 used, system_chunk) &&
5313 !work_busy(&fs_info->async_reclaim_work)) {
5314 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5315 orig_bytes, flush, "preempt");
5316 queue_work(system_unbound_wq,
5317 &fs_info->async_reclaim_work);
5320 spin_unlock(&space_info->lock);
5321 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5324 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5325 return wait_reserve_ticket(fs_info, space_info, &ticket,
5329 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5330 spin_lock(&space_info->lock);
5332 if (ticket.bytes < orig_bytes) {
5333 u64 num_bytes = orig_bytes - ticket.bytes;
5334 space_info->bytes_may_use -= num_bytes;
5335 trace_btrfs_space_reservation(fs_info, "space_info",
5340 list_del_init(&ticket.list);
5343 spin_unlock(&space_info->lock);
5344 ASSERT(list_empty(&ticket.list));
5349 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5350 * @root - the root we're allocating for
5351 * @block_rsv - the block_rsv we're allocating for
5352 * @orig_bytes - the number of bytes we want
5353 * @flush - whether or not we can flush to make our reservation
5355 * This will reserve orgi_bytes number of bytes from the space info associated
5356 * with the block_rsv. If there is not enough space it will make an attempt to
5357 * flush out space to make room. It will do this by flushing delalloc if
5358 * possible or committing the transaction. If flush is 0 then no attempts to
5359 * regain reservations will be made and this will fail if there is not enough
5362 static int reserve_metadata_bytes(struct btrfs_root *root,
5363 struct btrfs_block_rsv *block_rsv,
5365 enum btrfs_reserve_flush_enum flush)
5367 struct btrfs_fs_info *fs_info = root->fs_info;
5368 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5370 bool system_chunk = (root == fs_info->chunk_root);
5372 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5373 orig_bytes, flush, system_chunk);
5374 if (ret == -ENOSPC &&
5375 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5376 if (block_rsv != global_rsv &&
5377 !block_rsv_use_bytes(global_rsv, orig_bytes))
5380 if (ret == -ENOSPC) {
5381 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5382 block_rsv->space_info->flags,
5385 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5386 dump_space_info(fs_info, block_rsv->space_info,
5392 static struct btrfs_block_rsv *get_block_rsv(
5393 const struct btrfs_trans_handle *trans,
5394 const struct btrfs_root *root)
5396 struct btrfs_fs_info *fs_info = root->fs_info;
5397 struct btrfs_block_rsv *block_rsv = NULL;
5399 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5400 (root == fs_info->csum_root && trans->adding_csums) ||
5401 (root == fs_info->uuid_root))
5402 block_rsv = trans->block_rsv;
5405 block_rsv = root->block_rsv;
5408 block_rsv = &fs_info->empty_block_rsv;
5413 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5417 spin_lock(&block_rsv->lock);
5418 if (block_rsv->reserved >= num_bytes) {
5419 block_rsv->reserved -= num_bytes;
5420 if (block_rsv->reserved < block_rsv->size)
5421 block_rsv->full = 0;
5424 spin_unlock(&block_rsv->lock);
5428 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5429 u64 num_bytes, int update_size)
5431 spin_lock(&block_rsv->lock);
5432 block_rsv->reserved += num_bytes;
5434 block_rsv->size += num_bytes;
5435 else if (block_rsv->reserved >= block_rsv->size)
5436 block_rsv->full = 1;
5437 spin_unlock(&block_rsv->lock);
5440 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5441 struct btrfs_block_rsv *dest, u64 num_bytes,
5444 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5447 if (global_rsv->space_info != dest->space_info)
5450 spin_lock(&global_rsv->lock);
5451 min_bytes = div_factor(global_rsv->size, min_factor);
5452 if (global_rsv->reserved < min_bytes + num_bytes) {
5453 spin_unlock(&global_rsv->lock);
5456 global_rsv->reserved -= num_bytes;
5457 if (global_rsv->reserved < global_rsv->size)
5458 global_rsv->full = 0;
5459 spin_unlock(&global_rsv->lock);
5461 block_rsv_add_bytes(dest, num_bytes, 1);
5466 * This is for space we already have accounted in space_info->bytes_may_use, so
5467 * basically when we're returning space from block_rsv's.
5469 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5470 struct btrfs_space_info *space_info,
5473 struct reserve_ticket *ticket;
5474 struct list_head *head;
5476 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5477 bool check_overcommit = false;
5479 spin_lock(&space_info->lock);
5480 head = &space_info->priority_tickets;
5483 * If we are over our limit then we need to check and see if we can
5484 * overcommit, and if we can't then we just need to free up our space
5485 * and not satisfy any requests.
5487 used = btrfs_space_info_used(space_info, true);
5488 if (used - num_bytes >= space_info->total_bytes)
5489 check_overcommit = true;
5491 while (!list_empty(head) && num_bytes) {
5492 ticket = list_first_entry(head, struct reserve_ticket,
5495 * We use 0 bytes because this space is already reserved, so
5496 * adding the ticket space would be a double count.
5498 if (check_overcommit &&
5499 !can_overcommit(fs_info, space_info, 0, flush, false))
5501 if (num_bytes >= ticket->bytes) {
5502 list_del_init(&ticket->list);
5503 num_bytes -= ticket->bytes;
5505 space_info->tickets_id++;
5506 wake_up(&ticket->wait);
5508 ticket->bytes -= num_bytes;
5513 if (num_bytes && head == &space_info->priority_tickets) {
5514 head = &space_info->tickets;
5515 flush = BTRFS_RESERVE_FLUSH_ALL;
5518 space_info->bytes_may_use -= num_bytes;
5519 trace_btrfs_space_reservation(fs_info, "space_info",
5520 space_info->flags, num_bytes, 0);
5521 spin_unlock(&space_info->lock);
5525 * This is for newly allocated space that isn't accounted in
5526 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5527 * we use this helper.
5529 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5530 struct btrfs_space_info *space_info,
5533 struct reserve_ticket *ticket;
5534 struct list_head *head = &space_info->priority_tickets;
5537 while (!list_empty(head) && num_bytes) {
5538 ticket = list_first_entry(head, struct reserve_ticket,
5540 if (num_bytes >= ticket->bytes) {
5541 trace_btrfs_space_reservation(fs_info, "space_info",
5544 list_del_init(&ticket->list);
5545 num_bytes -= ticket->bytes;
5546 space_info->bytes_may_use += ticket->bytes;
5548 space_info->tickets_id++;
5549 wake_up(&ticket->wait);
5551 trace_btrfs_space_reservation(fs_info, "space_info",
5554 space_info->bytes_may_use += num_bytes;
5555 ticket->bytes -= num_bytes;
5560 if (num_bytes && head == &space_info->priority_tickets) {
5561 head = &space_info->tickets;
5566 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5567 struct btrfs_block_rsv *block_rsv,
5568 struct btrfs_block_rsv *dest, u64 num_bytes,
5569 u64 *qgroup_to_release_ret)
5571 struct btrfs_space_info *space_info = block_rsv->space_info;
5572 u64 qgroup_to_release = 0;
5575 spin_lock(&block_rsv->lock);
5576 if (num_bytes == (u64)-1) {
5577 num_bytes = block_rsv->size;
5578 qgroup_to_release = block_rsv->qgroup_rsv_size;
5580 block_rsv->size -= num_bytes;
5581 if (block_rsv->reserved >= block_rsv->size) {
5582 num_bytes = block_rsv->reserved - block_rsv->size;
5583 block_rsv->reserved = block_rsv->size;
5584 block_rsv->full = 1;
5588 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5589 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5590 block_rsv->qgroup_rsv_size;
5591 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5593 qgroup_to_release = 0;
5595 spin_unlock(&block_rsv->lock);
5598 if (num_bytes > 0) {
5600 spin_lock(&dest->lock);
5604 bytes_to_add = dest->size - dest->reserved;
5605 bytes_to_add = min(num_bytes, bytes_to_add);
5606 dest->reserved += bytes_to_add;
5607 if (dest->reserved >= dest->size)
5609 num_bytes -= bytes_to_add;
5611 spin_unlock(&dest->lock);
5614 space_info_add_old_bytes(fs_info, space_info,
5617 if (qgroup_to_release_ret)
5618 *qgroup_to_release_ret = qgroup_to_release;
5622 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5623 struct btrfs_block_rsv *dst, u64 num_bytes,
5628 ret = block_rsv_use_bytes(src, num_bytes);
5632 block_rsv_add_bytes(dst, num_bytes, update_size);
5636 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5638 memset(rsv, 0, sizeof(*rsv));
5639 spin_lock_init(&rsv->lock);
5643 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5644 struct btrfs_block_rsv *rsv,
5645 unsigned short type)
5647 btrfs_init_block_rsv(rsv, type);
5648 rsv->space_info = __find_space_info(fs_info,
5649 BTRFS_BLOCK_GROUP_METADATA);
5652 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5653 unsigned short type)
5655 struct btrfs_block_rsv *block_rsv;
5657 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5661 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5665 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5666 struct btrfs_block_rsv *rsv)
5670 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5674 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5679 int btrfs_block_rsv_add(struct btrfs_root *root,
5680 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5681 enum btrfs_reserve_flush_enum flush)
5688 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5690 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5697 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5705 spin_lock(&block_rsv->lock);
5706 num_bytes = div_factor(block_rsv->size, min_factor);
5707 if (block_rsv->reserved >= num_bytes)
5709 spin_unlock(&block_rsv->lock);
5714 int btrfs_block_rsv_refill(struct btrfs_root *root,
5715 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5716 enum btrfs_reserve_flush_enum flush)
5724 spin_lock(&block_rsv->lock);
5725 num_bytes = min_reserved;
5726 if (block_rsv->reserved >= num_bytes)
5729 num_bytes -= block_rsv->reserved;
5730 spin_unlock(&block_rsv->lock);
5735 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5737 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5745 * btrfs_inode_rsv_refill - refill the inode block rsv.
5746 * @inode - the inode we are refilling.
5747 * @flush - the flusing restriction.
5749 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5750 * block_rsv->size as the minimum size. We'll either refill the missing amount
5751 * or return if we already have enough space. This will also handle the resreve
5752 * tracepoint for the reserved amount.
5754 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5755 enum btrfs_reserve_flush_enum flush)
5757 struct btrfs_root *root = inode->root;
5758 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5760 u64 qgroup_num_bytes = 0;
5763 spin_lock(&block_rsv->lock);
5764 if (block_rsv->reserved < block_rsv->size)
5765 num_bytes = block_rsv->size - block_rsv->reserved;
5766 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5767 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5768 block_rsv->qgroup_rsv_reserved;
5769 spin_unlock(&block_rsv->lock);
5774 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5777 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5779 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5780 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5781 btrfs_ino(inode), num_bytes, 1);
5783 /* Don't forget to increase qgroup_rsv_reserved */
5784 spin_lock(&block_rsv->lock);
5785 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5786 spin_unlock(&block_rsv->lock);
5788 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5793 * btrfs_inode_rsv_release - release any excessive reservation.
5794 * @inode - the inode we need to release from.
5795 * @qgroup_free - free or convert qgroup meta.
5796 * Unlike normal operation, qgroup meta reservation needs to know if we are
5797 * freeing qgroup reservation or just converting it into per-trans. Normally
5798 * @qgroup_free is true for error handling, and false for normal release.
5800 * This is the same as btrfs_block_rsv_release, except that it handles the
5801 * tracepoint for the reservation.
5803 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5805 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5806 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5807 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5809 u64 qgroup_to_release = 0;
5812 * Since we statically set the block_rsv->size we just want to say we
5813 * are releasing 0 bytes, and then we'll just get the reservation over
5816 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5817 &qgroup_to_release);
5819 trace_btrfs_space_reservation(fs_info, "delalloc",
5820 btrfs_ino(inode), released, 0);
5822 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5824 btrfs_qgroup_convert_reserved_meta(inode->root,
5828 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5829 struct btrfs_block_rsv *block_rsv,
5832 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5834 if (global_rsv == block_rsv ||
5835 block_rsv->space_info != global_rsv->space_info)
5837 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5840 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5842 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5843 struct btrfs_space_info *sinfo = block_rsv->space_info;
5847 * The global block rsv is based on the size of the extent tree, the
5848 * checksum tree and the root tree. If the fs is empty we want to set
5849 * it to a minimal amount for safety.
5851 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5852 btrfs_root_used(&fs_info->csum_root->root_item) +
5853 btrfs_root_used(&fs_info->tree_root->root_item);
5854 num_bytes = max_t(u64, num_bytes, SZ_16M);
5856 spin_lock(&sinfo->lock);
5857 spin_lock(&block_rsv->lock);
5859 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5861 if (block_rsv->reserved < block_rsv->size) {
5862 num_bytes = btrfs_space_info_used(sinfo, true);
5863 if (sinfo->total_bytes > num_bytes) {
5864 num_bytes = sinfo->total_bytes - num_bytes;
5865 num_bytes = min(num_bytes,
5866 block_rsv->size - block_rsv->reserved);
5867 block_rsv->reserved += num_bytes;
5868 sinfo->bytes_may_use += num_bytes;
5869 trace_btrfs_space_reservation(fs_info, "space_info",
5870 sinfo->flags, num_bytes,
5873 } else if (block_rsv->reserved > block_rsv->size) {
5874 num_bytes = block_rsv->reserved - block_rsv->size;
5875 sinfo->bytes_may_use -= num_bytes;
5876 trace_btrfs_space_reservation(fs_info, "space_info",
5877 sinfo->flags, num_bytes, 0);
5878 block_rsv->reserved = block_rsv->size;
5881 if (block_rsv->reserved == block_rsv->size)
5882 block_rsv->full = 1;
5884 block_rsv->full = 0;
5886 spin_unlock(&block_rsv->lock);
5887 spin_unlock(&sinfo->lock);
5890 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5892 struct btrfs_space_info *space_info;
5894 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5895 fs_info->chunk_block_rsv.space_info = space_info;
5897 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5898 fs_info->global_block_rsv.space_info = space_info;
5899 fs_info->trans_block_rsv.space_info = space_info;
5900 fs_info->empty_block_rsv.space_info = space_info;
5901 fs_info->delayed_block_rsv.space_info = space_info;
5903 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5904 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5905 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5906 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5907 if (fs_info->quota_root)
5908 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5909 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5911 update_global_block_rsv(fs_info);
5914 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5916 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5918 WARN_ON(fs_info->trans_block_rsv.size > 0);
5919 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5920 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5921 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5922 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5923 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5928 * To be called after all the new block groups attached to the transaction
5929 * handle have been created (btrfs_create_pending_block_groups()).
5931 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5933 struct btrfs_fs_info *fs_info = trans->fs_info;
5935 if (!trans->chunk_bytes_reserved)
5938 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5940 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5941 trans->chunk_bytes_reserved, NULL);
5942 trans->chunk_bytes_reserved = 0;
5946 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5947 * root: the root of the parent directory
5948 * rsv: block reservation
5949 * items: the number of items that we need do reservation
5950 * qgroup_reserved: used to return the reserved size in qgroup
5952 * This function is used to reserve the space for snapshot/subvolume
5953 * creation and deletion. Those operations are different with the
5954 * common file/directory operations, they change two fs/file trees
5955 * and root tree, the number of items that the qgroup reserves is
5956 * different with the free space reservation. So we can not use
5957 * the space reservation mechanism in start_transaction().
5959 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5960 struct btrfs_block_rsv *rsv,
5962 bool use_global_rsv)
5966 struct btrfs_fs_info *fs_info = root->fs_info;
5967 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5969 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5970 /* One for parent inode, two for dir entries */
5971 num_bytes = 3 * fs_info->nodesize;
5972 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5979 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5980 rsv->space_info = __find_space_info(fs_info,
5981 BTRFS_BLOCK_GROUP_METADATA);
5982 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5983 BTRFS_RESERVE_FLUSH_ALL);
5985 if (ret == -ENOSPC && use_global_rsv)
5986 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5988 if (ret && num_bytes)
5989 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
5994 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5995 struct btrfs_block_rsv *rsv)
5997 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6000 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6001 struct btrfs_inode *inode)
6003 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6004 u64 reserve_size = 0;
6005 u64 qgroup_rsv_size = 0;
6007 unsigned outstanding_extents;
6009 lockdep_assert_held(&inode->lock);
6010 outstanding_extents = inode->outstanding_extents;
6011 if (outstanding_extents)
6012 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6013 outstanding_extents + 1);
6014 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6016 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6019 * For qgroup rsv, the calculation is very simple:
6020 * account one nodesize for each outstanding extent
6022 * This is overestimating in most cases.
6024 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6026 spin_lock(&block_rsv->lock);
6027 block_rsv->size = reserve_size;
6028 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6029 spin_unlock(&block_rsv->lock);
6032 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6034 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6035 unsigned nr_extents;
6036 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6038 bool delalloc_lock = true;
6040 /* If we are a free space inode we need to not flush since we will be in
6041 * the middle of a transaction commit. We also don't need the delalloc
6042 * mutex since we won't race with anybody. We need this mostly to make
6043 * lockdep shut its filthy mouth.
6045 * If we have a transaction open (can happen if we call truncate_block
6046 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6048 if (btrfs_is_free_space_inode(inode)) {
6049 flush = BTRFS_RESERVE_NO_FLUSH;
6050 delalloc_lock = false;
6052 if (current->journal_info)
6053 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6055 if (btrfs_transaction_in_commit(fs_info))
6056 schedule_timeout(1);
6060 mutex_lock(&inode->delalloc_mutex);
6062 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6064 /* Add our new extents and calculate the new rsv size. */
6065 spin_lock(&inode->lock);
6066 nr_extents = count_max_extents(num_bytes);
6067 btrfs_mod_outstanding_extents(inode, nr_extents);
6068 inode->csum_bytes += num_bytes;
6069 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6070 spin_unlock(&inode->lock);
6072 ret = btrfs_inode_rsv_refill(inode, flush);
6077 mutex_unlock(&inode->delalloc_mutex);
6081 spin_lock(&inode->lock);
6082 nr_extents = count_max_extents(num_bytes);
6083 btrfs_mod_outstanding_extents(inode, -nr_extents);
6084 inode->csum_bytes -= num_bytes;
6085 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6086 spin_unlock(&inode->lock);
6088 btrfs_inode_rsv_release(inode, true);
6090 mutex_unlock(&inode->delalloc_mutex);
6095 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6096 * @inode: the inode to release the reservation for.
6097 * @num_bytes: the number of bytes we are releasing.
6098 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6100 * This will release the metadata reservation for an inode. This can be called
6101 * once we complete IO for a given set of bytes to release their metadata
6102 * reservations, or on error for the same reason.
6104 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6107 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6109 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6110 spin_lock(&inode->lock);
6111 inode->csum_bytes -= num_bytes;
6112 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6113 spin_unlock(&inode->lock);
6115 if (btrfs_is_testing(fs_info))
6118 btrfs_inode_rsv_release(inode, qgroup_free);
6122 * btrfs_delalloc_release_extents - release our outstanding_extents
6123 * @inode: the inode to balance the reservation for.
6124 * @num_bytes: the number of bytes we originally reserved with
6125 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6127 * When we reserve space we increase outstanding_extents for the extents we may
6128 * add. Once we've set the range as delalloc or created our ordered extents we
6129 * have outstanding_extents to track the real usage, so we use this to free our
6130 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6131 * with btrfs_delalloc_reserve_metadata.
6133 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6136 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6137 unsigned num_extents;
6139 spin_lock(&inode->lock);
6140 num_extents = count_max_extents(num_bytes);
6141 btrfs_mod_outstanding_extents(inode, -num_extents);
6142 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6143 spin_unlock(&inode->lock);
6145 if (btrfs_is_testing(fs_info))
6148 btrfs_inode_rsv_release(inode, qgroup_free);
6152 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6154 * @inode: inode we're writing to
6155 * @start: start range we are writing to
6156 * @len: how long the range we are writing to
6157 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6158 * current reservation.
6160 * This will do the following things
6162 * o reserve space in data space info for num bytes
6163 * and reserve precious corresponding qgroup space
6164 * (Done in check_data_free_space)
6166 * o reserve space for metadata space, based on the number of outstanding
6167 * extents and how much csums will be needed
6168 * also reserve metadata space in a per root over-reserve method.
6169 * o add to the inodes->delalloc_bytes
6170 * o add it to the fs_info's delalloc inodes list.
6171 * (Above 3 all done in delalloc_reserve_metadata)
6173 * Return 0 for success
6174 * Return <0 for error(-ENOSPC or -EQUOT)
6176 int btrfs_delalloc_reserve_space(struct inode *inode,
6177 struct extent_changeset **reserved, u64 start, u64 len)
6181 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6184 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6186 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6191 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6192 * @inode: inode we're releasing space for
6193 * @start: start position of the space already reserved
6194 * @len: the len of the space already reserved
6195 * @release_bytes: the len of the space we consumed or didn't use
6197 * This function will release the metadata space that was not used and will
6198 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6199 * list if there are no delalloc bytes left.
6200 * Also it will handle the qgroup reserved space.
6202 void btrfs_delalloc_release_space(struct inode *inode,
6203 struct extent_changeset *reserved,
6204 u64 start, u64 len, bool qgroup_free)
6206 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6207 btrfs_free_reserved_data_space(inode, reserved, start, len);
6210 static int update_block_group(struct btrfs_trans_handle *trans,
6211 struct btrfs_fs_info *info, u64 bytenr,
6212 u64 num_bytes, int alloc)
6214 struct btrfs_block_group_cache *cache = NULL;
6215 u64 total = num_bytes;
6220 /* block accounting for super block */
6221 spin_lock(&info->delalloc_root_lock);
6222 old_val = btrfs_super_bytes_used(info->super_copy);
6224 old_val += num_bytes;
6226 old_val -= num_bytes;
6227 btrfs_set_super_bytes_used(info->super_copy, old_val);
6228 spin_unlock(&info->delalloc_root_lock);
6231 cache = btrfs_lookup_block_group(info, bytenr);
6234 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6235 BTRFS_BLOCK_GROUP_RAID1 |
6236 BTRFS_BLOCK_GROUP_RAID10))
6241 * If this block group has free space cache written out, we
6242 * need to make sure to load it if we are removing space. This
6243 * is because we need the unpinning stage to actually add the
6244 * space back to the block group, otherwise we will leak space.
6246 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6247 cache_block_group(cache, 1);
6249 byte_in_group = bytenr - cache->key.objectid;
6250 WARN_ON(byte_in_group > cache->key.offset);
6252 spin_lock(&cache->space_info->lock);
6253 spin_lock(&cache->lock);
6255 if (btrfs_test_opt(info, SPACE_CACHE) &&
6256 cache->disk_cache_state < BTRFS_DC_CLEAR)
6257 cache->disk_cache_state = BTRFS_DC_CLEAR;
6259 old_val = btrfs_block_group_used(&cache->item);
6260 num_bytes = min(total, cache->key.offset - byte_in_group);
6262 old_val += num_bytes;
6263 btrfs_set_block_group_used(&cache->item, old_val);
6264 cache->reserved -= num_bytes;
6265 cache->space_info->bytes_reserved -= num_bytes;
6266 cache->space_info->bytes_used += num_bytes;
6267 cache->space_info->disk_used += num_bytes * factor;
6268 spin_unlock(&cache->lock);
6269 spin_unlock(&cache->space_info->lock);
6271 old_val -= num_bytes;
6272 btrfs_set_block_group_used(&cache->item, old_val);
6273 cache->pinned += num_bytes;
6274 cache->space_info->bytes_pinned += num_bytes;
6275 cache->space_info->bytes_used -= num_bytes;
6276 cache->space_info->disk_used -= num_bytes * factor;
6277 spin_unlock(&cache->lock);
6278 spin_unlock(&cache->space_info->lock);
6280 trace_btrfs_space_reservation(info, "pinned",
6281 cache->space_info->flags,
6283 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6285 set_extent_dirty(info->pinned_extents,
6286 bytenr, bytenr + num_bytes - 1,
6287 GFP_NOFS | __GFP_NOFAIL);
6290 spin_lock(&trans->transaction->dirty_bgs_lock);
6291 if (list_empty(&cache->dirty_list)) {
6292 list_add_tail(&cache->dirty_list,
6293 &trans->transaction->dirty_bgs);
6294 trans->transaction->num_dirty_bgs++;
6295 btrfs_get_block_group(cache);
6297 spin_unlock(&trans->transaction->dirty_bgs_lock);
6300 * No longer have used bytes in this block group, queue it for
6301 * deletion. We do this after adding the block group to the
6302 * dirty list to avoid races between cleaner kthread and space
6305 if (!alloc && old_val == 0) {
6306 spin_lock(&info->unused_bgs_lock);
6307 if (list_empty(&cache->bg_list)) {
6308 btrfs_get_block_group(cache);
6309 trace_btrfs_add_unused_block_group(cache);
6310 list_add_tail(&cache->bg_list,
6313 spin_unlock(&info->unused_bgs_lock);
6316 btrfs_put_block_group(cache);
6318 bytenr += num_bytes;
6323 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6325 struct btrfs_block_group_cache *cache;
6328 spin_lock(&fs_info->block_group_cache_lock);
6329 bytenr = fs_info->first_logical_byte;
6330 spin_unlock(&fs_info->block_group_cache_lock);
6332 if (bytenr < (u64)-1)
6335 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6339 bytenr = cache->key.objectid;
6340 btrfs_put_block_group(cache);
6345 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6346 struct btrfs_block_group_cache *cache,
6347 u64 bytenr, u64 num_bytes, int reserved)
6349 spin_lock(&cache->space_info->lock);
6350 spin_lock(&cache->lock);
6351 cache->pinned += num_bytes;
6352 cache->space_info->bytes_pinned += num_bytes;
6354 cache->reserved -= num_bytes;
6355 cache->space_info->bytes_reserved -= num_bytes;
6357 spin_unlock(&cache->lock);
6358 spin_unlock(&cache->space_info->lock);
6360 trace_btrfs_space_reservation(fs_info, "pinned",
6361 cache->space_info->flags, num_bytes, 1);
6362 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6363 set_extent_dirty(fs_info->pinned_extents, bytenr,
6364 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6369 * this function must be called within transaction
6371 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6372 u64 bytenr, u64 num_bytes, int reserved)
6374 struct btrfs_block_group_cache *cache;
6376 cache = btrfs_lookup_block_group(fs_info, bytenr);
6377 BUG_ON(!cache); /* Logic error */
6379 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6381 btrfs_put_block_group(cache);
6386 * this function must be called within transaction
6388 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6389 u64 bytenr, u64 num_bytes)
6391 struct btrfs_block_group_cache *cache;
6394 cache = btrfs_lookup_block_group(fs_info, bytenr);
6399 * pull in the free space cache (if any) so that our pin
6400 * removes the free space from the cache. We have load_only set
6401 * to one because the slow code to read in the free extents does check
6402 * the pinned extents.
6404 cache_block_group(cache, 1);
6406 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6408 /* remove us from the free space cache (if we're there at all) */
6409 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6410 btrfs_put_block_group(cache);
6414 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6415 u64 start, u64 num_bytes)
6418 struct btrfs_block_group_cache *block_group;
6419 struct btrfs_caching_control *caching_ctl;
6421 block_group = btrfs_lookup_block_group(fs_info, start);
6425 cache_block_group(block_group, 0);
6426 caching_ctl = get_caching_control(block_group);
6430 BUG_ON(!block_group_cache_done(block_group));
6431 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6433 mutex_lock(&caching_ctl->mutex);
6435 if (start >= caching_ctl->progress) {
6436 ret = add_excluded_extent(fs_info, start, num_bytes);
6437 } else if (start + num_bytes <= caching_ctl->progress) {
6438 ret = btrfs_remove_free_space(block_group,
6441 num_bytes = caching_ctl->progress - start;
6442 ret = btrfs_remove_free_space(block_group,
6447 num_bytes = (start + num_bytes) -
6448 caching_ctl->progress;
6449 start = caching_ctl->progress;
6450 ret = add_excluded_extent(fs_info, start, num_bytes);
6453 mutex_unlock(&caching_ctl->mutex);
6454 put_caching_control(caching_ctl);
6456 btrfs_put_block_group(block_group);
6460 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6461 struct extent_buffer *eb)
6463 struct btrfs_file_extent_item *item;
6464 struct btrfs_key key;
6469 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6472 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6473 btrfs_item_key_to_cpu(eb, &key, i);
6474 if (key.type != BTRFS_EXTENT_DATA_KEY)
6476 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6477 found_type = btrfs_file_extent_type(eb, item);
6478 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6480 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6482 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6483 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6484 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6493 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6495 atomic_inc(&bg->reservations);
6498 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6501 struct btrfs_block_group_cache *bg;
6503 bg = btrfs_lookup_block_group(fs_info, start);
6505 if (atomic_dec_and_test(&bg->reservations))
6506 wake_up_var(&bg->reservations);
6507 btrfs_put_block_group(bg);
6510 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6512 struct btrfs_space_info *space_info = bg->space_info;
6516 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6520 * Our block group is read only but before we set it to read only,
6521 * some task might have had allocated an extent from it already, but it
6522 * has not yet created a respective ordered extent (and added it to a
6523 * root's list of ordered extents).
6524 * Therefore wait for any task currently allocating extents, since the
6525 * block group's reservations counter is incremented while a read lock
6526 * on the groups' semaphore is held and decremented after releasing
6527 * the read access on that semaphore and creating the ordered extent.
6529 down_write(&space_info->groups_sem);
6530 up_write(&space_info->groups_sem);
6532 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6536 * btrfs_add_reserved_bytes - update the block_group and space info counters
6537 * @cache: The cache we are manipulating
6538 * @ram_bytes: The number of bytes of file content, and will be same to
6539 * @num_bytes except for the compress path.
6540 * @num_bytes: The number of bytes in question
6541 * @delalloc: The blocks are allocated for the delalloc write
6543 * This is called by the allocator when it reserves space. If this is a
6544 * reservation and the block group has become read only we cannot make the
6545 * reservation and return -EAGAIN, otherwise this function always succeeds.
6547 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6548 u64 ram_bytes, u64 num_bytes, int delalloc)
6550 struct btrfs_space_info *space_info = cache->space_info;
6553 spin_lock(&space_info->lock);
6554 spin_lock(&cache->lock);
6558 cache->reserved += num_bytes;
6559 space_info->bytes_reserved += num_bytes;
6561 trace_btrfs_space_reservation(cache->fs_info,
6562 "space_info", space_info->flags,
6564 space_info->bytes_may_use -= ram_bytes;
6566 cache->delalloc_bytes += num_bytes;
6568 spin_unlock(&cache->lock);
6569 spin_unlock(&space_info->lock);
6574 * btrfs_free_reserved_bytes - update the block_group and space info counters
6575 * @cache: The cache we are manipulating
6576 * @num_bytes: The number of bytes in question
6577 * @delalloc: The blocks are allocated for the delalloc write
6579 * This is called by somebody who is freeing space that was never actually used
6580 * on disk. For example if you reserve some space for a new leaf in transaction
6581 * A and before transaction A commits you free that leaf, you call this with
6582 * reserve set to 0 in order to clear the reservation.
6585 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6586 u64 num_bytes, int delalloc)
6588 struct btrfs_space_info *space_info = cache->space_info;
6591 spin_lock(&space_info->lock);
6592 spin_lock(&cache->lock);
6594 space_info->bytes_readonly += num_bytes;
6595 cache->reserved -= num_bytes;
6596 space_info->bytes_reserved -= num_bytes;
6599 cache->delalloc_bytes -= num_bytes;
6600 spin_unlock(&cache->lock);
6601 spin_unlock(&space_info->lock);
6604 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6606 struct btrfs_caching_control *next;
6607 struct btrfs_caching_control *caching_ctl;
6608 struct btrfs_block_group_cache *cache;
6610 down_write(&fs_info->commit_root_sem);
6612 list_for_each_entry_safe(caching_ctl, next,
6613 &fs_info->caching_block_groups, list) {
6614 cache = caching_ctl->block_group;
6615 if (block_group_cache_done(cache)) {
6616 cache->last_byte_to_unpin = (u64)-1;
6617 list_del_init(&caching_ctl->list);
6618 put_caching_control(caching_ctl);
6620 cache->last_byte_to_unpin = caching_ctl->progress;
6624 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6625 fs_info->pinned_extents = &fs_info->freed_extents[1];
6627 fs_info->pinned_extents = &fs_info->freed_extents[0];
6629 up_write(&fs_info->commit_root_sem);
6631 update_global_block_rsv(fs_info);
6635 * Returns the free cluster for the given space info and sets empty_cluster to
6636 * what it should be based on the mount options.
6638 static struct btrfs_free_cluster *
6639 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6640 struct btrfs_space_info *space_info, u64 *empty_cluster)
6642 struct btrfs_free_cluster *ret = NULL;
6645 if (btrfs_mixed_space_info(space_info))
6648 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6649 ret = &fs_info->meta_alloc_cluster;
6650 if (btrfs_test_opt(fs_info, SSD))
6651 *empty_cluster = SZ_2M;
6653 *empty_cluster = SZ_64K;
6654 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6655 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6656 *empty_cluster = SZ_2M;
6657 ret = &fs_info->data_alloc_cluster;
6663 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6665 const bool return_free_space)
6667 struct btrfs_block_group_cache *cache = NULL;
6668 struct btrfs_space_info *space_info;
6669 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6670 struct btrfs_free_cluster *cluster = NULL;
6672 u64 total_unpinned = 0;
6673 u64 empty_cluster = 0;
6676 while (start <= end) {
6679 start >= cache->key.objectid + cache->key.offset) {
6681 btrfs_put_block_group(cache);
6683 cache = btrfs_lookup_block_group(fs_info, start);
6684 BUG_ON(!cache); /* Logic error */
6686 cluster = fetch_cluster_info(fs_info,
6689 empty_cluster <<= 1;
6692 len = cache->key.objectid + cache->key.offset - start;
6693 len = min(len, end + 1 - start);
6695 if (start < cache->last_byte_to_unpin) {
6696 len = min(len, cache->last_byte_to_unpin - start);
6697 if (return_free_space)
6698 btrfs_add_free_space(cache, start, len);
6702 total_unpinned += len;
6703 space_info = cache->space_info;
6706 * If this space cluster has been marked as fragmented and we've
6707 * unpinned enough in this block group to potentially allow a
6708 * cluster to be created inside of it go ahead and clear the
6711 if (cluster && cluster->fragmented &&
6712 total_unpinned > empty_cluster) {
6713 spin_lock(&cluster->lock);
6714 cluster->fragmented = 0;
6715 spin_unlock(&cluster->lock);
6718 spin_lock(&space_info->lock);
6719 spin_lock(&cache->lock);
6720 cache->pinned -= len;
6721 space_info->bytes_pinned -= len;
6723 trace_btrfs_space_reservation(fs_info, "pinned",
6724 space_info->flags, len, 0);
6725 space_info->max_extent_size = 0;
6726 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6728 space_info->bytes_readonly += len;
6731 spin_unlock(&cache->lock);
6732 if (!readonly && return_free_space &&
6733 global_rsv->space_info == space_info) {
6736 spin_lock(&global_rsv->lock);
6737 if (!global_rsv->full) {
6738 to_add = min(len, global_rsv->size -
6739 global_rsv->reserved);
6740 global_rsv->reserved += to_add;
6741 space_info->bytes_may_use += to_add;
6742 if (global_rsv->reserved >= global_rsv->size)
6743 global_rsv->full = 1;
6744 trace_btrfs_space_reservation(fs_info,
6750 spin_unlock(&global_rsv->lock);
6751 /* Add to any tickets we may have */
6753 space_info_add_new_bytes(fs_info, space_info,
6756 spin_unlock(&space_info->lock);
6760 btrfs_put_block_group(cache);
6764 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6766 struct btrfs_fs_info *fs_info = trans->fs_info;
6767 struct btrfs_block_group_cache *block_group, *tmp;
6768 struct list_head *deleted_bgs;
6769 struct extent_io_tree *unpin;
6774 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6775 unpin = &fs_info->freed_extents[1];
6777 unpin = &fs_info->freed_extents[0];
6779 while (!trans->aborted) {
6780 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6781 ret = find_first_extent_bit(unpin, 0, &start, &end,
6782 EXTENT_DIRTY, NULL);
6784 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6788 if (btrfs_test_opt(fs_info, DISCARD))
6789 ret = btrfs_discard_extent(fs_info, start,
6790 end + 1 - start, NULL);
6792 clear_extent_dirty(unpin, start, end);
6793 unpin_extent_range(fs_info, start, end, true);
6794 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6799 * Transaction is finished. We don't need the lock anymore. We
6800 * do need to clean up the block groups in case of a transaction
6803 deleted_bgs = &trans->transaction->deleted_bgs;
6804 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6808 if (!trans->aborted)
6809 ret = btrfs_discard_extent(fs_info,
6810 block_group->key.objectid,
6811 block_group->key.offset,
6814 list_del_init(&block_group->bg_list);
6815 btrfs_put_block_group_trimming(block_group);
6816 btrfs_put_block_group(block_group);
6819 const char *errstr = btrfs_decode_error(ret);
6821 "discard failed while removing blockgroup: errno=%d %s",
6829 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6830 struct btrfs_delayed_ref_node *node, u64 parent,
6831 u64 root_objectid, u64 owner_objectid,
6832 u64 owner_offset, int refs_to_drop,
6833 struct btrfs_delayed_extent_op *extent_op)
6835 struct btrfs_fs_info *info = trans->fs_info;
6836 struct btrfs_key key;
6837 struct btrfs_path *path;
6838 struct btrfs_root *extent_root = info->extent_root;
6839 struct extent_buffer *leaf;
6840 struct btrfs_extent_item *ei;
6841 struct btrfs_extent_inline_ref *iref;
6844 int extent_slot = 0;
6845 int found_extent = 0;
6849 u64 bytenr = node->bytenr;
6850 u64 num_bytes = node->num_bytes;
6852 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6854 path = btrfs_alloc_path();
6858 path->reada = READA_FORWARD;
6859 path->leave_spinning = 1;
6861 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6862 BUG_ON(!is_data && refs_to_drop != 1);
6865 skinny_metadata = false;
6867 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6868 parent, root_objectid, owner_objectid,
6871 extent_slot = path->slots[0];
6872 while (extent_slot >= 0) {
6873 btrfs_item_key_to_cpu(path->nodes[0], &key,
6875 if (key.objectid != bytenr)
6877 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6878 key.offset == num_bytes) {
6882 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6883 key.offset == owner_objectid) {
6887 if (path->slots[0] - extent_slot > 5)
6891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6892 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6893 if (found_extent && item_size < sizeof(*ei))
6896 if (!found_extent) {
6898 ret = remove_extent_backref(trans, path, NULL,
6900 is_data, &last_ref);
6902 btrfs_abort_transaction(trans, ret);
6905 btrfs_release_path(path);
6906 path->leave_spinning = 1;
6908 key.objectid = bytenr;
6909 key.type = BTRFS_EXTENT_ITEM_KEY;
6910 key.offset = num_bytes;
6912 if (!is_data && skinny_metadata) {
6913 key.type = BTRFS_METADATA_ITEM_KEY;
6914 key.offset = owner_objectid;
6917 ret = btrfs_search_slot(trans, extent_root,
6919 if (ret > 0 && skinny_metadata && path->slots[0]) {
6921 * Couldn't find our skinny metadata item,
6922 * see if we have ye olde extent item.
6925 btrfs_item_key_to_cpu(path->nodes[0], &key,
6927 if (key.objectid == bytenr &&
6928 key.type == BTRFS_EXTENT_ITEM_KEY &&
6929 key.offset == num_bytes)
6933 if (ret > 0 && skinny_metadata) {
6934 skinny_metadata = false;
6935 key.objectid = bytenr;
6936 key.type = BTRFS_EXTENT_ITEM_KEY;
6937 key.offset = num_bytes;
6938 btrfs_release_path(path);
6939 ret = btrfs_search_slot(trans, extent_root,
6945 "umm, got %d back from search, was looking for %llu",
6948 btrfs_print_leaf(path->nodes[0]);
6951 btrfs_abort_transaction(trans, ret);
6954 extent_slot = path->slots[0];
6956 } else if (WARN_ON(ret == -ENOENT)) {
6957 btrfs_print_leaf(path->nodes[0]);
6959 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6960 bytenr, parent, root_objectid, owner_objectid,
6962 btrfs_abort_transaction(trans, ret);
6965 btrfs_abort_transaction(trans, ret);
6969 leaf = path->nodes[0];
6970 item_size = btrfs_item_size_nr(leaf, extent_slot);
6971 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6972 if (item_size < sizeof(*ei)) {
6973 BUG_ON(found_extent || extent_slot != path->slots[0]);
6974 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6977 btrfs_abort_transaction(trans, ret);
6981 btrfs_release_path(path);
6982 path->leave_spinning = 1;
6984 key.objectid = bytenr;
6985 key.type = BTRFS_EXTENT_ITEM_KEY;
6986 key.offset = num_bytes;
6988 ret = btrfs_search_slot(trans, extent_root, &key, path,
6992 "umm, got %d back from search, was looking for %llu",
6994 btrfs_print_leaf(path->nodes[0]);
6997 btrfs_abort_transaction(trans, ret);
7001 extent_slot = path->slots[0];
7002 leaf = path->nodes[0];
7003 item_size = btrfs_item_size_nr(leaf, extent_slot);
7006 BUG_ON(item_size < sizeof(*ei));
7007 ei = btrfs_item_ptr(leaf, extent_slot,
7008 struct btrfs_extent_item);
7009 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7010 key.type == BTRFS_EXTENT_ITEM_KEY) {
7011 struct btrfs_tree_block_info *bi;
7012 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7013 bi = (struct btrfs_tree_block_info *)(ei + 1);
7014 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7017 refs = btrfs_extent_refs(leaf, ei);
7018 if (refs < refs_to_drop) {
7020 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7021 refs_to_drop, refs, bytenr);
7023 btrfs_abort_transaction(trans, ret);
7026 refs -= refs_to_drop;
7030 __run_delayed_extent_op(extent_op, leaf, ei);
7032 * In the case of inline back ref, reference count will
7033 * be updated by remove_extent_backref
7036 BUG_ON(!found_extent);
7038 btrfs_set_extent_refs(leaf, ei, refs);
7039 btrfs_mark_buffer_dirty(leaf);
7042 ret = remove_extent_backref(trans, path, iref,
7043 refs_to_drop, is_data,
7046 btrfs_abort_transaction(trans, ret);
7052 BUG_ON(is_data && refs_to_drop !=
7053 extent_data_ref_count(path, iref));
7055 BUG_ON(path->slots[0] != extent_slot);
7057 BUG_ON(path->slots[0] != extent_slot + 1);
7058 path->slots[0] = extent_slot;
7064 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7067 btrfs_abort_transaction(trans, ret);
7070 btrfs_release_path(path);
7073 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7075 btrfs_abort_transaction(trans, ret);
7080 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7082 btrfs_abort_transaction(trans, ret);
7086 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7088 btrfs_abort_transaction(trans, ret);
7092 btrfs_release_path(path);
7095 btrfs_free_path(path);
7100 * when we free an block, it is possible (and likely) that we free the last
7101 * delayed ref for that extent as well. This searches the delayed ref tree for
7102 * a given extent, and if there are no other delayed refs to be processed, it
7103 * removes it from the tree.
7105 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7108 struct btrfs_delayed_ref_head *head;
7109 struct btrfs_delayed_ref_root *delayed_refs;
7112 delayed_refs = &trans->transaction->delayed_refs;
7113 spin_lock(&delayed_refs->lock);
7114 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7116 goto out_delayed_unlock;
7118 spin_lock(&head->lock);
7119 if (!RB_EMPTY_ROOT(&head->ref_tree))
7122 if (head->extent_op) {
7123 if (!head->must_insert_reserved)
7125 btrfs_free_delayed_extent_op(head->extent_op);
7126 head->extent_op = NULL;
7130 * waiting for the lock here would deadlock. If someone else has it
7131 * locked they are already in the process of dropping it anyway
7133 if (!mutex_trylock(&head->mutex))
7137 * at this point we have a head with no other entries. Go
7138 * ahead and process it.
7140 rb_erase(&head->href_node, &delayed_refs->href_root);
7141 RB_CLEAR_NODE(&head->href_node);
7142 atomic_dec(&delayed_refs->num_entries);
7145 * we don't take a ref on the node because we're removing it from the
7146 * tree, so we just steal the ref the tree was holding.
7148 delayed_refs->num_heads--;
7149 if (head->processing == 0)
7150 delayed_refs->num_heads_ready--;
7151 head->processing = 0;
7152 spin_unlock(&head->lock);
7153 spin_unlock(&delayed_refs->lock);
7155 BUG_ON(head->extent_op);
7156 if (head->must_insert_reserved)
7159 mutex_unlock(&head->mutex);
7160 btrfs_put_delayed_ref_head(head);
7163 spin_unlock(&head->lock);
7166 spin_unlock(&delayed_refs->lock);
7170 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7171 struct btrfs_root *root,
7172 struct extent_buffer *buf,
7173 u64 parent, int last_ref)
7175 struct btrfs_fs_info *fs_info = root->fs_info;
7179 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7180 int old_ref_mod, new_ref_mod;
7182 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7183 root->root_key.objectid,
7184 btrfs_header_level(buf), 0,
7185 BTRFS_DROP_DELAYED_REF);
7186 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7188 root->root_key.objectid,
7189 btrfs_header_level(buf),
7190 BTRFS_DROP_DELAYED_REF, NULL,
7191 &old_ref_mod, &new_ref_mod);
7192 BUG_ON(ret); /* -ENOMEM */
7193 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7196 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7197 struct btrfs_block_group_cache *cache;
7199 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7200 ret = check_ref_cleanup(trans, buf->start);
7206 cache = btrfs_lookup_block_group(fs_info, buf->start);
7208 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7209 pin_down_extent(fs_info, cache, buf->start,
7211 btrfs_put_block_group(cache);
7215 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7217 btrfs_add_free_space(cache, buf->start, buf->len);
7218 btrfs_free_reserved_bytes(cache, buf->len, 0);
7219 btrfs_put_block_group(cache);
7220 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7224 add_pinned_bytes(fs_info, buf->len, true,
7225 root->root_key.objectid);
7229 * Deleting the buffer, clear the corrupt flag since it doesn't
7232 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7236 /* Can return -ENOMEM */
7237 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7238 struct btrfs_root *root,
7239 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7240 u64 owner, u64 offset)
7242 struct btrfs_fs_info *fs_info = root->fs_info;
7243 int old_ref_mod, new_ref_mod;
7246 if (btrfs_is_testing(fs_info))
7249 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7250 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7251 root_objectid, owner, offset,
7252 BTRFS_DROP_DELAYED_REF);
7255 * tree log blocks never actually go into the extent allocation
7256 * tree, just update pinning info and exit early.
7258 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7259 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7260 /* unlocks the pinned mutex */
7261 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7262 old_ref_mod = new_ref_mod = 0;
7264 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7265 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7267 root_objectid, (int)owner,
7268 BTRFS_DROP_DELAYED_REF, NULL,
7269 &old_ref_mod, &new_ref_mod);
7271 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7273 root_objectid, owner, offset,
7274 0, BTRFS_DROP_DELAYED_REF,
7275 &old_ref_mod, &new_ref_mod);
7278 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7279 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7281 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7288 * when we wait for progress in the block group caching, its because
7289 * our allocation attempt failed at least once. So, we must sleep
7290 * and let some progress happen before we try again.
7292 * This function will sleep at least once waiting for new free space to
7293 * show up, and then it will check the block group free space numbers
7294 * for our min num_bytes. Another option is to have it go ahead
7295 * and look in the rbtree for a free extent of a given size, but this
7298 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7299 * any of the information in this block group.
7301 static noinline void
7302 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7305 struct btrfs_caching_control *caching_ctl;
7307 caching_ctl = get_caching_control(cache);
7311 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7312 (cache->free_space_ctl->free_space >= num_bytes));
7314 put_caching_control(caching_ctl);
7318 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7320 struct btrfs_caching_control *caching_ctl;
7323 caching_ctl = get_caching_control(cache);
7325 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7327 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7328 if (cache->cached == BTRFS_CACHE_ERROR)
7330 put_caching_control(caching_ctl);
7334 enum btrfs_loop_type {
7335 LOOP_CACHING_NOWAIT = 0,
7336 LOOP_CACHING_WAIT = 1,
7337 LOOP_ALLOC_CHUNK = 2,
7338 LOOP_NO_EMPTY_SIZE = 3,
7342 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7346 down_read(&cache->data_rwsem);
7350 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7353 btrfs_get_block_group(cache);
7355 down_read(&cache->data_rwsem);
7358 static struct btrfs_block_group_cache *
7359 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7360 struct btrfs_free_cluster *cluster,
7363 struct btrfs_block_group_cache *used_bg = NULL;
7365 spin_lock(&cluster->refill_lock);
7367 used_bg = cluster->block_group;
7371 if (used_bg == block_group)
7374 btrfs_get_block_group(used_bg);
7379 if (down_read_trylock(&used_bg->data_rwsem))
7382 spin_unlock(&cluster->refill_lock);
7384 /* We should only have one-level nested. */
7385 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7387 spin_lock(&cluster->refill_lock);
7388 if (used_bg == cluster->block_group)
7391 up_read(&used_bg->data_rwsem);
7392 btrfs_put_block_group(used_bg);
7397 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7401 up_read(&cache->data_rwsem);
7402 btrfs_put_block_group(cache);
7406 * walks the btree of allocated extents and find a hole of a given size.
7407 * The key ins is changed to record the hole:
7408 * ins->objectid == start position
7409 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7410 * ins->offset == the size of the hole.
7411 * Any available blocks before search_start are skipped.
7413 * If there is no suitable free space, we will record the max size of
7414 * the free space extent currently.
7416 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7417 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7418 u64 hint_byte, struct btrfs_key *ins,
7419 u64 flags, int delalloc)
7422 struct btrfs_root *root = fs_info->extent_root;
7423 struct btrfs_free_cluster *last_ptr = NULL;
7424 struct btrfs_block_group_cache *block_group = NULL;
7425 u64 search_start = 0;
7426 u64 max_extent_size = 0;
7427 u64 empty_cluster = 0;
7428 struct btrfs_space_info *space_info;
7430 int index = btrfs_bg_flags_to_raid_index(flags);
7431 bool failed_cluster_refill = false;
7432 bool failed_alloc = false;
7433 bool use_cluster = true;
7434 bool have_caching_bg = false;
7435 bool orig_have_caching_bg = false;
7436 bool full_search = false;
7438 WARN_ON(num_bytes < fs_info->sectorsize);
7439 ins->type = BTRFS_EXTENT_ITEM_KEY;
7443 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7445 space_info = __find_space_info(fs_info, flags);
7447 btrfs_err(fs_info, "No space info for %llu", flags);
7452 * If our free space is heavily fragmented we may not be able to make
7453 * big contiguous allocations, so instead of doing the expensive search
7454 * for free space, simply return ENOSPC with our max_extent_size so we
7455 * can go ahead and search for a more manageable chunk.
7457 * If our max_extent_size is large enough for our allocation simply
7458 * disable clustering since we will likely not be able to find enough
7459 * space to create a cluster and induce latency trying.
7461 if (unlikely(space_info->max_extent_size)) {
7462 spin_lock(&space_info->lock);
7463 if (space_info->max_extent_size &&
7464 num_bytes > space_info->max_extent_size) {
7465 ins->offset = space_info->max_extent_size;
7466 spin_unlock(&space_info->lock);
7468 } else if (space_info->max_extent_size) {
7469 use_cluster = false;
7471 spin_unlock(&space_info->lock);
7474 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7476 spin_lock(&last_ptr->lock);
7477 if (last_ptr->block_group)
7478 hint_byte = last_ptr->window_start;
7479 if (last_ptr->fragmented) {
7481 * We still set window_start so we can keep track of the
7482 * last place we found an allocation to try and save
7485 hint_byte = last_ptr->window_start;
7486 use_cluster = false;
7488 spin_unlock(&last_ptr->lock);
7491 search_start = max(search_start, first_logical_byte(fs_info, 0));
7492 search_start = max(search_start, hint_byte);
7493 if (search_start == hint_byte) {
7494 block_group = btrfs_lookup_block_group(fs_info, search_start);
7496 * we don't want to use the block group if it doesn't match our
7497 * allocation bits, or if its not cached.
7499 * However if we are re-searching with an ideal block group
7500 * picked out then we don't care that the block group is cached.
7502 if (block_group && block_group_bits(block_group, flags) &&
7503 block_group->cached != BTRFS_CACHE_NO) {
7504 down_read(&space_info->groups_sem);
7505 if (list_empty(&block_group->list) ||
7508 * someone is removing this block group,
7509 * we can't jump into the have_block_group
7510 * target because our list pointers are not
7513 btrfs_put_block_group(block_group);
7514 up_read(&space_info->groups_sem);
7516 index = btrfs_bg_flags_to_raid_index(
7517 block_group->flags);
7518 btrfs_lock_block_group(block_group, delalloc);
7519 goto have_block_group;
7521 } else if (block_group) {
7522 btrfs_put_block_group(block_group);
7526 have_caching_bg = false;
7527 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7529 down_read(&space_info->groups_sem);
7530 list_for_each_entry(block_group, &space_info->block_groups[index],
7535 /* If the block group is read-only, we can skip it entirely. */
7536 if (unlikely(block_group->ro))
7539 btrfs_grab_block_group(block_group, delalloc);
7540 search_start = block_group->key.objectid;
7543 * this can happen if we end up cycling through all the
7544 * raid types, but we want to make sure we only allocate
7545 * for the proper type.
7547 if (!block_group_bits(block_group, flags)) {
7548 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7549 BTRFS_BLOCK_GROUP_RAID1 |
7550 BTRFS_BLOCK_GROUP_RAID5 |
7551 BTRFS_BLOCK_GROUP_RAID6 |
7552 BTRFS_BLOCK_GROUP_RAID10;
7555 * if they asked for extra copies and this block group
7556 * doesn't provide them, bail. This does allow us to
7557 * fill raid0 from raid1.
7559 if ((flags & extra) && !(block_group->flags & extra))
7564 cached = block_group_cache_done(block_group);
7565 if (unlikely(!cached)) {
7566 have_caching_bg = true;
7567 ret = cache_block_group(block_group, 0);
7572 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7576 * Ok we want to try and use the cluster allocator, so
7579 if (last_ptr && use_cluster) {
7580 struct btrfs_block_group_cache *used_block_group;
7581 unsigned long aligned_cluster;
7583 * the refill lock keeps out other
7584 * people trying to start a new cluster
7586 used_block_group = btrfs_lock_cluster(block_group,
7589 if (!used_block_group)
7590 goto refill_cluster;
7592 if (used_block_group != block_group &&
7593 (used_block_group->ro ||
7594 !block_group_bits(used_block_group, flags)))
7595 goto release_cluster;
7597 offset = btrfs_alloc_from_cluster(used_block_group,
7600 used_block_group->key.objectid,
7603 /* we have a block, we're done */
7604 spin_unlock(&last_ptr->refill_lock);
7605 trace_btrfs_reserve_extent_cluster(
7607 search_start, num_bytes);
7608 if (used_block_group != block_group) {
7609 btrfs_release_block_group(block_group,
7611 block_group = used_block_group;
7616 WARN_ON(last_ptr->block_group != used_block_group);
7618 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7619 * set up a new clusters, so lets just skip it
7620 * and let the allocator find whatever block
7621 * it can find. If we reach this point, we
7622 * will have tried the cluster allocator
7623 * plenty of times and not have found
7624 * anything, so we are likely way too
7625 * fragmented for the clustering stuff to find
7628 * However, if the cluster is taken from the
7629 * current block group, release the cluster
7630 * first, so that we stand a better chance of
7631 * succeeding in the unclustered
7633 if (loop >= LOOP_NO_EMPTY_SIZE &&
7634 used_block_group != block_group) {
7635 spin_unlock(&last_ptr->refill_lock);
7636 btrfs_release_block_group(used_block_group,
7638 goto unclustered_alloc;
7642 * this cluster didn't work out, free it and
7645 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7647 if (used_block_group != block_group)
7648 btrfs_release_block_group(used_block_group,
7651 if (loop >= LOOP_NO_EMPTY_SIZE) {
7652 spin_unlock(&last_ptr->refill_lock);
7653 goto unclustered_alloc;
7656 aligned_cluster = max_t(unsigned long,
7657 empty_cluster + empty_size,
7658 block_group->full_stripe_len);
7660 /* allocate a cluster in this block group */
7661 ret = btrfs_find_space_cluster(fs_info, block_group,
7662 last_ptr, search_start,
7667 * now pull our allocation out of this
7670 offset = btrfs_alloc_from_cluster(block_group,
7676 /* we found one, proceed */
7677 spin_unlock(&last_ptr->refill_lock);
7678 trace_btrfs_reserve_extent_cluster(
7679 block_group, search_start,
7683 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7684 && !failed_cluster_refill) {
7685 spin_unlock(&last_ptr->refill_lock);
7687 failed_cluster_refill = true;
7688 wait_block_group_cache_progress(block_group,
7689 num_bytes + empty_cluster + empty_size);
7690 goto have_block_group;
7694 * at this point we either didn't find a cluster
7695 * or we weren't able to allocate a block from our
7696 * cluster. Free the cluster we've been trying
7697 * to use, and go to the next block group
7699 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7700 spin_unlock(&last_ptr->refill_lock);
7706 * We are doing an unclustered alloc, set the fragmented flag so
7707 * we don't bother trying to setup a cluster again until we get
7710 if (unlikely(last_ptr)) {
7711 spin_lock(&last_ptr->lock);
7712 last_ptr->fragmented = 1;
7713 spin_unlock(&last_ptr->lock);
7716 struct btrfs_free_space_ctl *ctl =
7717 block_group->free_space_ctl;
7719 spin_lock(&ctl->tree_lock);
7720 if (ctl->free_space <
7721 num_bytes + empty_cluster + empty_size) {
7722 if (ctl->free_space > max_extent_size)
7723 max_extent_size = ctl->free_space;
7724 spin_unlock(&ctl->tree_lock);
7727 spin_unlock(&ctl->tree_lock);
7730 offset = btrfs_find_space_for_alloc(block_group, search_start,
7731 num_bytes, empty_size,
7734 * If we didn't find a chunk, and we haven't failed on this
7735 * block group before, and this block group is in the middle of
7736 * caching and we are ok with waiting, then go ahead and wait
7737 * for progress to be made, and set failed_alloc to true.
7739 * If failed_alloc is true then we've already waited on this
7740 * block group once and should move on to the next block group.
7742 if (!offset && !failed_alloc && !cached &&
7743 loop > LOOP_CACHING_NOWAIT) {
7744 wait_block_group_cache_progress(block_group,
7745 num_bytes + empty_size);
7746 failed_alloc = true;
7747 goto have_block_group;
7748 } else if (!offset) {
7752 search_start = ALIGN(offset, fs_info->stripesize);
7754 /* move on to the next group */
7755 if (search_start + num_bytes >
7756 block_group->key.objectid + block_group->key.offset) {
7757 btrfs_add_free_space(block_group, offset, num_bytes);
7761 if (offset < search_start)
7762 btrfs_add_free_space(block_group, offset,
7763 search_start - offset);
7764 BUG_ON(offset > search_start);
7766 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7767 num_bytes, delalloc);
7768 if (ret == -EAGAIN) {
7769 btrfs_add_free_space(block_group, offset, num_bytes);
7772 btrfs_inc_block_group_reservations(block_group);
7774 /* we are all good, lets return */
7775 ins->objectid = search_start;
7776 ins->offset = num_bytes;
7778 trace_btrfs_reserve_extent(block_group, search_start, num_bytes);
7779 btrfs_release_block_group(block_group, delalloc);
7782 failed_cluster_refill = false;
7783 failed_alloc = false;
7784 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7786 btrfs_release_block_group(block_group, delalloc);
7789 up_read(&space_info->groups_sem);
7791 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7792 && !orig_have_caching_bg)
7793 orig_have_caching_bg = true;
7795 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7798 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7802 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7803 * caching kthreads as we move along
7804 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7805 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7806 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7809 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7811 if (loop == LOOP_CACHING_NOWAIT) {
7813 * We want to skip the LOOP_CACHING_WAIT step if we
7814 * don't have any uncached bgs and we've already done a
7815 * full search through.
7817 if (orig_have_caching_bg || !full_search)
7818 loop = LOOP_CACHING_WAIT;
7820 loop = LOOP_ALLOC_CHUNK;
7825 if (loop == LOOP_ALLOC_CHUNK) {
7826 struct btrfs_trans_handle *trans;
7829 trans = current->journal_info;
7833 trans = btrfs_join_transaction(root);
7835 if (IS_ERR(trans)) {
7836 ret = PTR_ERR(trans);
7840 ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE);
7843 * If we can't allocate a new chunk we've already looped
7844 * through at least once, move on to the NO_EMPTY_SIZE
7848 loop = LOOP_NO_EMPTY_SIZE;
7851 * Do not bail out on ENOSPC since we
7852 * can do more things.
7854 if (ret < 0 && ret != -ENOSPC)
7855 btrfs_abort_transaction(trans, ret);
7859 btrfs_end_transaction(trans);
7864 if (loop == LOOP_NO_EMPTY_SIZE) {
7866 * Don't loop again if we already have no empty_size and
7869 if (empty_size == 0 &&
7870 empty_cluster == 0) {
7879 } else if (!ins->objectid) {
7881 } else if (ins->objectid) {
7882 if (!use_cluster && last_ptr) {
7883 spin_lock(&last_ptr->lock);
7884 last_ptr->window_start = ins->objectid;
7885 spin_unlock(&last_ptr->lock);
7890 if (ret == -ENOSPC) {
7891 spin_lock(&space_info->lock);
7892 space_info->max_extent_size = max_extent_size;
7893 spin_unlock(&space_info->lock);
7894 ins->offset = max_extent_size;
7899 static void dump_space_info(struct btrfs_fs_info *fs_info,
7900 struct btrfs_space_info *info, u64 bytes,
7901 int dump_block_groups)
7903 struct btrfs_block_group_cache *cache;
7906 spin_lock(&info->lock);
7907 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7909 info->total_bytes - btrfs_space_info_used(info, true),
7910 info->full ? "" : "not ");
7912 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7913 info->total_bytes, info->bytes_used, info->bytes_pinned,
7914 info->bytes_reserved, info->bytes_may_use,
7915 info->bytes_readonly);
7916 spin_unlock(&info->lock);
7918 if (!dump_block_groups)
7921 down_read(&info->groups_sem);
7923 list_for_each_entry(cache, &info->block_groups[index], list) {
7924 spin_lock(&cache->lock);
7926 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7927 cache->key.objectid, cache->key.offset,
7928 btrfs_block_group_used(&cache->item), cache->pinned,
7929 cache->reserved, cache->ro ? "[readonly]" : "");
7930 btrfs_dump_free_space(cache, bytes);
7931 spin_unlock(&cache->lock);
7933 if (++index < BTRFS_NR_RAID_TYPES)
7935 up_read(&info->groups_sem);
7939 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7940 * hole that is at least as big as @num_bytes.
7942 * @root - The root that will contain this extent
7944 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7945 * is used for accounting purposes. This value differs
7946 * from @num_bytes only in the case of compressed extents.
7948 * @num_bytes - Number of bytes to allocate on-disk.
7950 * @min_alloc_size - Indicates the minimum amount of space that the
7951 * allocator should try to satisfy. In some cases
7952 * @num_bytes may be larger than what is required and if
7953 * the filesystem is fragmented then allocation fails.
7954 * However, the presence of @min_alloc_size gives a
7955 * chance to try and satisfy the smaller allocation.
7957 * @empty_size - A hint that you plan on doing more COW. This is the
7958 * size in bytes the allocator should try to find free
7959 * next to the block it returns. This is just a hint and
7960 * may be ignored by the allocator.
7962 * @hint_byte - Hint to the allocator to start searching above the byte
7963 * address passed. It might be ignored.
7965 * @ins - This key is modified to record the found hole. It will
7966 * have the following values:
7967 * ins->objectid == start position
7968 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7969 * ins->offset == the size of the hole.
7971 * @is_data - Boolean flag indicating whether an extent is
7972 * allocated for data (true) or metadata (false)
7974 * @delalloc - Boolean flag indicating whether this allocation is for
7975 * delalloc or not. If 'true' data_rwsem of block groups
7976 * is going to be acquired.
7979 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7980 * case -ENOSPC is returned then @ins->offset will contain the size of the
7981 * largest available hole the allocator managed to find.
7983 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7984 u64 num_bytes, u64 min_alloc_size,
7985 u64 empty_size, u64 hint_byte,
7986 struct btrfs_key *ins, int is_data, int delalloc)
7988 struct btrfs_fs_info *fs_info = root->fs_info;
7989 bool final_tried = num_bytes == min_alloc_size;
7993 flags = get_alloc_profile_by_root(root, is_data);
7995 WARN_ON(num_bytes < fs_info->sectorsize);
7996 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7997 hint_byte, ins, flags, delalloc);
7998 if (!ret && !is_data) {
7999 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8000 } else if (ret == -ENOSPC) {
8001 if (!final_tried && ins->offset) {
8002 num_bytes = min(num_bytes >> 1, ins->offset);
8003 num_bytes = round_down(num_bytes,
8004 fs_info->sectorsize);
8005 num_bytes = max(num_bytes, min_alloc_size);
8006 ram_bytes = num_bytes;
8007 if (num_bytes == min_alloc_size)
8010 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8011 struct btrfs_space_info *sinfo;
8013 sinfo = __find_space_info(fs_info, flags);
8015 "allocation failed flags %llu, wanted %llu",
8018 dump_space_info(fs_info, sinfo, num_bytes, 1);
8025 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8027 int pin, int delalloc)
8029 struct btrfs_block_group_cache *cache;
8032 cache = btrfs_lookup_block_group(fs_info, start);
8034 btrfs_err(fs_info, "Unable to find block group for %llu",
8040 pin_down_extent(fs_info, cache, start, len, 1);
8042 if (btrfs_test_opt(fs_info, DISCARD))
8043 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8044 btrfs_add_free_space(cache, start, len);
8045 btrfs_free_reserved_bytes(cache, len, delalloc);
8046 trace_btrfs_reserved_extent_free(fs_info, start, len);
8049 btrfs_put_block_group(cache);
8053 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8054 u64 start, u64 len, int delalloc)
8056 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8059 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8062 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8065 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8066 u64 parent, u64 root_objectid,
8067 u64 flags, u64 owner, u64 offset,
8068 struct btrfs_key *ins, int ref_mod)
8070 struct btrfs_fs_info *fs_info = trans->fs_info;
8072 struct btrfs_extent_item *extent_item;
8073 struct btrfs_extent_inline_ref *iref;
8074 struct btrfs_path *path;
8075 struct extent_buffer *leaf;
8080 type = BTRFS_SHARED_DATA_REF_KEY;
8082 type = BTRFS_EXTENT_DATA_REF_KEY;
8084 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8086 path = btrfs_alloc_path();
8090 path->leave_spinning = 1;
8091 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8094 btrfs_free_path(path);
8098 leaf = path->nodes[0];
8099 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8100 struct btrfs_extent_item);
8101 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8102 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8103 btrfs_set_extent_flags(leaf, extent_item,
8104 flags | BTRFS_EXTENT_FLAG_DATA);
8106 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8107 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8109 struct btrfs_shared_data_ref *ref;
8110 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8111 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8112 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8114 struct btrfs_extent_data_ref *ref;
8115 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8116 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8117 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8118 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8119 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8122 btrfs_mark_buffer_dirty(path->nodes[0]);
8123 btrfs_free_path(path);
8125 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8129 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8130 if (ret) { /* -ENOENT, logic error */
8131 btrfs_err(fs_info, "update block group failed for %llu %llu",
8132 ins->objectid, ins->offset);
8135 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8139 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8140 struct btrfs_delayed_ref_node *node,
8141 struct btrfs_delayed_extent_op *extent_op)
8143 struct btrfs_fs_info *fs_info = trans->fs_info;
8145 struct btrfs_extent_item *extent_item;
8146 struct btrfs_key extent_key;
8147 struct btrfs_tree_block_info *block_info;
8148 struct btrfs_extent_inline_ref *iref;
8149 struct btrfs_path *path;
8150 struct extent_buffer *leaf;
8151 struct btrfs_delayed_tree_ref *ref;
8152 u32 size = sizeof(*extent_item) + sizeof(*iref);
8154 u64 flags = extent_op->flags_to_set;
8155 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8157 ref = btrfs_delayed_node_to_tree_ref(node);
8159 extent_key.objectid = node->bytenr;
8160 if (skinny_metadata) {
8161 extent_key.offset = ref->level;
8162 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8163 num_bytes = fs_info->nodesize;
8165 extent_key.offset = node->num_bytes;
8166 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8167 size += sizeof(*block_info);
8168 num_bytes = node->num_bytes;
8171 path = btrfs_alloc_path();
8173 btrfs_free_and_pin_reserved_extent(fs_info,
8174 extent_key.objectid,
8179 path->leave_spinning = 1;
8180 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8183 btrfs_free_path(path);
8184 btrfs_free_and_pin_reserved_extent(fs_info,
8185 extent_key.objectid,
8190 leaf = path->nodes[0];
8191 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8192 struct btrfs_extent_item);
8193 btrfs_set_extent_refs(leaf, extent_item, 1);
8194 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8195 btrfs_set_extent_flags(leaf, extent_item,
8196 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8198 if (skinny_metadata) {
8199 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8201 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8202 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8203 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8204 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8207 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8208 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8209 btrfs_set_extent_inline_ref_type(leaf, iref,
8210 BTRFS_SHARED_BLOCK_REF_KEY);
8211 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8213 btrfs_set_extent_inline_ref_type(leaf, iref,
8214 BTRFS_TREE_BLOCK_REF_KEY);
8215 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8218 btrfs_mark_buffer_dirty(leaf);
8219 btrfs_free_path(path);
8221 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8226 ret = update_block_group(trans, fs_info, extent_key.objectid,
8227 fs_info->nodesize, 1);
8228 if (ret) { /* -ENOENT, logic error */
8229 btrfs_err(fs_info, "update block group failed for %llu %llu",
8230 extent_key.objectid, extent_key.offset);
8234 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8239 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8240 struct btrfs_root *root, u64 owner,
8241 u64 offset, u64 ram_bytes,
8242 struct btrfs_key *ins)
8246 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8248 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8249 root->root_key.objectid, owner, offset,
8250 BTRFS_ADD_DELAYED_EXTENT);
8252 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8254 root->root_key.objectid, owner,
8256 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8261 * this is used by the tree logging recovery code. It records that
8262 * an extent has been allocated and makes sure to clear the free
8263 * space cache bits as well
8265 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8266 u64 root_objectid, u64 owner, u64 offset,
8267 struct btrfs_key *ins)
8269 struct btrfs_fs_info *fs_info = trans->fs_info;
8271 struct btrfs_block_group_cache *block_group;
8272 struct btrfs_space_info *space_info;
8275 * Mixed block groups will exclude before processing the log so we only
8276 * need to do the exclude dance if this fs isn't mixed.
8278 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8279 ret = __exclude_logged_extent(fs_info, ins->objectid,
8285 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8289 space_info = block_group->space_info;
8290 spin_lock(&space_info->lock);
8291 spin_lock(&block_group->lock);
8292 space_info->bytes_reserved += ins->offset;
8293 block_group->reserved += ins->offset;
8294 spin_unlock(&block_group->lock);
8295 spin_unlock(&space_info->lock);
8297 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8299 btrfs_put_block_group(block_group);
8303 static struct extent_buffer *
8304 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8305 u64 bytenr, int level, u64 owner)
8307 struct btrfs_fs_info *fs_info = root->fs_info;
8308 struct extent_buffer *buf;
8310 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8314 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8315 btrfs_tree_lock(buf);
8316 clean_tree_block(fs_info, buf);
8317 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8319 btrfs_set_lock_blocking(buf);
8320 set_extent_buffer_uptodate(buf);
8322 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8323 btrfs_set_header_level(buf, level);
8324 btrfs_set_header_bytenr(buf, buf->start);
8325 btrfs_set_header_generation(buf, trans->transid);
8326 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8327 btrfs_set_header_owner(buf, owner);
8328 write_extent_buffer_fsid(buf, fs_info->fsid);
8329 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8330 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8331 buf->log_index = root->log_transid % 2;
8333 * we allow two log transactions at a time, use different
8334 * EXENT bit to differentiate dirty pages.
8336 if (buf->log_index == 0)
8337 set_extent_dirty(&root->dirty_log_pages, buf->start,
8338 buf->start + buf->len - 1, GFP_NOFS);
8340 set_extent_new(&root->dirty_log_pages, buf->start,
8341 buf->start + buf->len - 1);
8343 buf->log_index = -1;
8344 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8345 buf->start + buf->len - 1, GFP_NOFS);
8347 trans->dirty = true;
8348 /* this returns a buffer locked for blocking */
8352 static struct btrfs_block_rsv *
8353 use_block_rsv(struct btrfs_trans_handle *trans,
8354 struct btrfs_root *root, u32 blocksize)
8356 struct btrfs_fs_info *fs_info = root->fs_info;
8357 struct btrfs_block_rsv *block_rsv;
8358 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8360 bool global_updated = false;
8362 block_rsv = get_block_rsv(trans, root);
8364 if (unlikely(block_rsv->size == 0))
8367 ret = block_rsv_use_bytes(block_rsv, blocksize);
8371 if (block_rsv->failfast)
8372 return ERR_PTR(ret);
8374 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8375 global_updated = true;
8376 update_global_block_rsv(fs_info);
8380 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8381 static DEFINE_RATELIMIT_STATE(_rs,
8382 DEFAULT_RATELIMIT_INTERVAL * 10,
8383 /*DEFAULT_RATELIMIT_BURST*/ 1);
8384 if (__ratelimit(&_rs))
8386 "BTRFS: block rsv returned %d\n", ret);
8389 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8390 BTRFS_RESERVE_NO_FLUSH);
8394 * If we couldn't reserve metadata bytes try and use some from
8395 * the global reserve if its space type is the same as the global
8398 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8399 block_rsv->space_info == global_rsv->space_info) {
8400 ret = block_rsv_use_bytes(global_rsv, blocksize);
8404 return ERR_PTR(ret);
8407 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8408 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8410 block_rsv_add_bytes(block_rsv, blocksize, 0);
8411 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8415 * finds a free extent and does all the dirty work required for allocation
8416 * returns the tree buffer or an ERR_PTR on error.
8418 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8419 struct btrfs_root *root,
8420 u64 parent, u64 root_objectid,
8421 const struct btrfs_disk_key *key,
8422 int level, u64 hint,
8425 struct btrfs_fs_info *fs_info = root->fs_info;
8426 struct btrfs_key ins;
8427 struct btrfs_block_rsv *block_rsv;
8428 struct extent_buffer *buf;
8429 struct btrfs_delayed_extent_op *extent_op;
8432 u32 blocksize = fs_info->nodesize;
8433 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8435 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8436 if (btrfs_is_testing(fs_info)) {
8437 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8438 level, root_objectid);
8440 root->alloc_bytenr += blocksize;
8445 block_rsv = use_block_rsv(trans, root, blocksize);
8446 if (IS_ERR(block_rsv))
8447 return ERR_CAST(block_rsv);
8449 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8450 empty_size, hint, &ins, 0, 0);
8454 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8458 goto out_free_reserved;
8461 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8463 parent = ins.objectid;
8464 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8468 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8469 extent_op = btrfs_alloc_delayed_extent_op();
8475 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8477 memset(&extent_op->key, 0, sizeof(extent_op->key));
8478 extent_op->flags_to_set = flags;
8479 extent_op->update_key = skinny_metadata ? false : true;
8480 extent_op->update_flags = true;
8481 extent_op->is_data = false;
8482 extent_op->level = level;
8484 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8485 root_objectid, level, 0,
8486 BTRFS_ADD_DELAYED_EXTENT);
8487 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8489 root_objectid, level,
8490 BTRFS_ADD_DELAYED_EXTENT,
8491 extent_op, NULL, NULL);
8493 goto out_free_delayed;
8498 btrfs_free_delayed_extent_op(extent_op);
8500 free_extent_buffer(buf);
8502 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8504 unuse_block_rsv(fs_info, block_rsv, blocksize);
8505 return ERR_PTR(ret);
8508 struct walk_control {
8509 u64 refs[BTRFS_MAX_LEVEL];
8510 u64 flags[BTRFS_MAX_LEVEL];
8511 struct btrfs_key update_progress;
8522 #define DROP_REFERENCE 1
8523 #define UPDATE_BACKREF 2
8525 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8526 struct btrfs_root *root,
8527 struct walk_control *wc,
8528 struct btrfs_path *path)
8530 struct btrfs_fs_info *fs_info = root->fs_info;
8536 struct btrfs_key key;
8537 struct extent_buffer *eb;
8542 if (path->slots[wc->level] < wc->reada_slot) {
8543 wc->reada_count = wc->reada_count * 2 / 3;
8544 wc->reada_count = max(wc->reada_count, 2);
8546 wc->reada_count = wc->reada_count * 3 / 2;
8547 wc->reada_count = min_t(int, wc->reada_count,
8548 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8551 eb = path->nodes[wc->level];
8552 nritems = btrfs_header_nritems(eb);
8554 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8555 if (nread >= wc->reada_count)
8559 bytenr = btrfs_node_blockptr(eb, slot);
8560 generation = btrfs_node_ptr_generation(eb, slot);
8562 if (slot == path->slots[wc->level])
8565 if (wc->stage == UPDATE_BACKREF &&
8566 generation <= root->root_key.offset)
8569 /* We don't lock the tree block, it's OK to be racy here */
8570 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8571 wc->level - 1, 1, &refs,
8573 /* We don't care about errors in readahead. */
8578 if (wc->stage == DROP_REFERENCE) {
8582 if (wc->level == 1 &&
8583 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8585 if (!wc->update_ref ||
8586 generation <= root->root_key.offset)
8588 btrfs_node_key_to_cpu(eb, &key, slot);
8589 ret = btrfs_comp_cpu_keys(&key,
8590 &wc->update_progress);
8594 if (wc->level == 1 &&
8595 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8599 readahead_tree_block(fs_info, bytenr);
8602 wc->reada_slot = slot;
8606 * helper to process tree block while walking down the tree.
8608 * when wc->stage == UPDATE_BACKREF, this function updates
8609 * back refs for pointers in the block.
8611 * NOTE: return value 1 means we should stop walking down.
8613 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8614 struct btrfs_root *root,
8615 struct btrfs_path *path,
8616 struct walk_control *wc, int lookup_info)
8618 struct btrfs_fs_info *fs_info = root->fs_info;
8619 int level = wc->level;
8620 struct extent_buffer *eb = path->nodes[level];
8621 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8624 if (wc->stage == UPDATE_BACKREF &&
8625 btrfs_header_owner(eb) != root->root_key.objectid)
8629 * when reference count of tree block is 1, it won't increase
8630 * again. once full backref flag is set, we never clear it.
8633 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8634 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8635 BUG_ON(!path->locks[level]);
8636 ret = btrfs_lookup_extent_info(trans, fs_info,
8637 eb->start, level, 1,
8640 BUG_ON(ret == -ENOMEM);
8643 BUG_ON(wc->refs[level] == 0);
8646 if (wc->stage == DROP_REFERENCE) {
8647 if (wc->refs[level] > 1)
8650 if (path->locks[level] && !wc->keep_locks) {
8651 btrfs_tree_unlock_rw(eb, path->locks[level]);
8652 path->locks[level] = 0;
8657 /* wc->stage == UPDATE_BACKREF */
8658 if (!(wc->flags[level] & flag)) {
8659 BUG_ON(!path->locks[level]);
8660 ret = btrfs_inc_ref(trans, root, eb, 1);
8661 BUG_ON(ret); /* -ENOMEM */
8662 ret = btrfs_dec_ref(trans, root, eb, 0);
8663 BUG_ON(ret); /* -ENOMEM */
8664 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8666 btrfs_header_level(eb), 0);
8667 BUG_ON(ret); /* -ENOMEM */
8668 wc->flags[level] |= flag;
8672 * the block is shared by multiple trees, so it's not good to
8673 * keep the tree lock
8675 if (path->locks[level] && level > 0) {
8676 btrfs_tree_unlock_rw(eb, path->locks[level]);
8677 path->locks[level] = 0;
8683 * helper to process tree block pointer.
8685 * when wc->stage == DROP_REFERENCE, this function checks
8686 * reference count of the block pointed to. if the block
8687 * is shared and we need update back refs for the subtree
8688 * rooted at the block, this function changes wc->stage to
8689 * UPDATE_BACKREF. if the block is shared and there is no
8690 * need to update back, this function drops the reference
8693 * NOTE: return value 1 means we should stop walking down.
8695 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8696 struct btrfs_root *root,
8697 struct btrfs_path *path,
8698 struct walk_control *wc, int *lookup_info)
8700 struct btrfs_fs_info *fs_info = root->fs_info;
8705 struct btrfs_key key;
8706 struct btrfs_key first_key;
8707 struct extent_buffer *next;
8708 int level = wc->level;
8711 bool need_account = false;
8713 generation = btrfs_node_ptr_generation(path->nodes[level],
8714 path->slots[level]);
8716 * if the lower level block was created before the snapshot
8717 * was created, we know there is no need to update back refs
8720 if (wc->stage == UPDATE_BACKREF &&
8721 generation <= root->root_key.offset) {
8726 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8727 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8728 path->slots[level]);
8729 blocksize = fs_info->nodesize;
8731 next = find_extent_buffer(fs_info, bytenr);
8733 next = btrfs_find_create_tree_block(fs_info, bytenr);
8735 return PTR_ERR(next);
8737 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8741 btrfs_tree_lock(next);
8742 btrfs_set_lock_blocking(next);
8744 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8745 &wc->refs[level - 1],
8746 &wc->flags[level - 1]);
8750 if (unlikely(wc->refs[level - 1] == 0)) {
8751 btrfs_err(fs_info, "Missing references.");
8757 if (wc->stage == DROP_REFERENCE) {
8758 if (wc->refs[level - 1] > 1) {
8759 need_account = true;
8761 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8764 if (!wc->update_ref ||
8765 generation <= root->root_key.offset)
8768 btrfs_node_key_to_cpu(path->nodes[level], &key,
8769 path->slots[level]);
8770 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8774 wc->stage = UPDATE_BACKREF;
8775 wc->shared_level = level - 1;
8779 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8783 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8784 btrfs_tree_unlock(next);
8785 free_extent_buffer(next);
8791 if (reada && level == 1)
8792 reada_walk_down(trans, root, wc, path);
8793 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8796 return PTR_ERR(next);
8797 } else if (!extent_buffer_uptodate(next)) {
8798 free_extent_buffer(next);
8801 btrfs_tree_lock(next);
8802 btrfs_set_lock_blocking(next);
8806 ASSERT(level == btrfs_header_level(next));
8807 if (level != btrfs_header_level(next)) {
8808 btrfs_err(root->fs_info, "mismatched level");
8812 path->nodes[level] = next;
8813 path->slots[level] = 0;
8814 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8820 wc->refs[level - 1] = 0;
8821 wc->flags[level - 1] = 0;
8822 if (wc->stage == DROP_REFERENCE) {
8823 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8824 parent = path->nodes[level]->start;
8826 ASSERT(root->root_key.objectid ==
8827 btrfs_header_owner(path->nodes[level]));
8828 if (root->root_key.objectid !=
8829 btrfs_header_owner(path->nodes[level])) {
8830 btrfs_err(root->fs_info,
8831 "mismatched block owner");
8839 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8840 generation, level - 1);
8842 btrfs_err_rl(fs_info,
8843 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8847 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8848 parent, root->root_key.objectid,
8858 btrfs_tree_unlock(next);
8859 free_extent_buffer(next);
8865 * helper to process tree block while walking up the tree.
8867 * when wc->stage == DROP_REFERENCE, this function drops
8868 * reference count on the block.
8870 * when wc->stage == UPDATE_BACKREF, this function changes
8871 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8872 * to UPDATE_BACKREF previously while processing the block.
8874 * NOTE: return value 1 means we should stop walking up.
8876 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8877 struct btrfs_root *root,
8878 struct btrfs_path *path,
8879 struct walk_control *wc)
8881 struct btrfs_fs_info *fs_info = root->fs_info;
8883 int level = wc->level;
8884 struct extent_buffer *eb = path->nodes[level];
8887 if (wc->stage == UPDATE_BACKREF) {
8888 BUG_ON(wc->shared_level < level);
8889 if (level < wc->shared_level)
8892 ret = find_next_key(path, level + 1, &wc->update_progress);
8896 wc->stage = DROP_REFERENCE;
8897 wc->shared_level = -1;
8898 path->slots[level] = 0;
8901 * check reference count again if the block isn't locked.
8902 * we should start walking down the tree again if reference
8905 if (!path->locks[level]) {
8907 btrfs_tree_lock(eb);
8908 btrfs_set_lock_blocking(eb);
8909 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8911 ret = btrfs_lookup_extent_info(trans, fs_info,
8912 eb->start, level, 1,
8916 btrfs_tree_unlock_rw(eb, path->locks[level]);
8917 path->locks[level] = 0;
8920 BUG_ON(wc->refs[level] == 0);
8921 if (wc->refs[level] == 1) {
8922 btrfs_tree_unlock_rw(eb, path->locks[level]);
8923 path->locks[level] = 0;
8929 /* wc->stage == DROP_REFERENCE */
8930 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8932 if (wc->refs[level] == 1) {
8934 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8935 ret = btrfs_dec_ref(trans, root, eb, 1);
8937 ret = btrfs_dec_ref(trans, root, eb, 0);
8938 BUG_ON(ret); /* -ENOMEM */
8939 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8941 btrfs_err_rl(fs_info,
8942 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8946 /* make block locked assertion in clean_tree_block happy */
8947 if (!path->locks[level] &&
8948 btrfs_header_generation(eb) == trans->transid) {
8949 btrfs_tree_lock(eb);
8950 btrfs_set_lock_blocking(eb);
8951 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8953 clean_tree_block(fs_info, eb);
8956 if (eb == root->node) {
8957 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8960 BUG_ON(root->root_key.objectid !=
8961 btrfs_header_owner(eb));
8963 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8964 parent = path->nodes[level + 1]->start;
8966 BUG_ON(root->root_key.objectid !=
8967 btrfs_header_owner(path->nodes[level + 1]));
8970 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8972 wc->refs[level] = 0;
8973 wc->flags[level] = 0;
8977 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8978 struct btrfs_root *root,
8979 struct btrfs_path *path,
8980 struct walk_control *wc)
8982 int level = wc->level;
8983 int lookup_info = 1;
8986 while (level >= 0) {
8987 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8994 if (path->slots[level] >=
8995 btrfs_header_nritems(path->nodes[level]))
8998 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9000 path->slots[level]++;
9009 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9010 struct btrfs_root *root,
9011 struct btrfs_path *path,
9012 struct walk_control *wc, int max_level)
9014 int level = wc->level;
9017 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9018 while (level < max_level && path->nodes[level]) {
9020 if (path->slots[level] + 1 <
9021 btrfs_header_nritems(path->nodes[level])) {
9022 path->slots[level]++;
9025 ret = walk_up_proc(trans, root, path, wc);
9029 if (path->locks[level]) {
9030 btrfs_tree_unlock_rw(path->nodes[level],
9031 path->locks[level]);
9032 path->locks[level] = 0;
9034 free_extent_buffer(path->nodes[level]);
9035 path->nodes[level] = NULL;
9043 * drop a subvolume tree.
9045 * this function traverses the tree freeing any blocks that only
9046 * referenced by the tree.
9048 * when a shared tree block is found. this function decreases its
9049 * reference count by one. if update_ref is true, this function
9050 * also make sure backrefs for the shared block and all lower level
9051 * blocks are properly updated.
9053 * If called with for_reloc == 0, may exit early with -EAGAIN
9055 int btrfs_drop_snapshot(struct btrfs_root *root,
9056 struct btrfs_block_rsv *block_rsv, int update_ref,
9059 struct btrfs_fs_info *fs_info = root->fs_info;
9060 struct btrfs_path *path;
9061 struct btrfs_trans_handle *trans;
9062 struct btrfs_root *tree_root = fs_info->tree_root;
9063 struct btrfs_root_item *root_item = &root->root_item;
9064 struct walk_control *wc;
9065 struct btrfs_key key;
9069 bool root_dropped = false;
9071 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9073 path = btrfs_alloc_path();
9079 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9081 btrfs_free_path(path);
9086 trans = btrfs_start_transaction(tree_root, 0);
9087 if (IS_ERR(trans)) {
9088 err = PTR_ERR(trans);
9093 trans->block_rsv = block_rsv;
9095 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9096 level = btrfs_header_level(root->node);
9097 path->nodes[level] = btrfs_lock_root_node(root);
9098 btrfs_set_lock_blocking(path->nodes[level]);
9099 path->slots[level] = 0;
9100 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9101 memset(&wc->update_progress, 0,
9102 sizeof(wc->update_progress));
9104 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9105 memcpy(&wc->update_progress, &key,
9106 sizeof(wc->update_progress));
9108 level = root_item->drop_level;
9110 path->lowest_level = level;
9111 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9112 path->lowest_level = 0;
9120 * unlock our path, this is safe because only this
9121 * function is allowed to delete this snapshot
9123 btrfs_unlock_up_safe(path, 0);
9125 level = btrfs_header_level(root->node);
9127 btrfs_tree_lock(path->nodes[level]);
9128 btrfs_set_lock_blocking(path->nodes[level]);
9129 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9131 ret = btrfs_lookup_extent_info(trans, fs_info,
9132 path->nodes[level]->start,
9133 level, 1, &wc->refs[level],
9139 BUG_ON(wc->refs[level] == 0);
9141 if (level == root_item->drop_level)
9144 btrfs_tree_unlock(path->nodes[level]);
9145 path->locks[level] = 0;
9146 WARN_ON(wc->refs[level] != 1);
9152 wc->shared_level = -1;
9153 wc->stage = DROP_REFERENCE;
9154 wc->update_ref = update_ref;
9156 wc->for_reloc = for_reloc;
9157 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9161 ret = walk_down_tree(trans, root, path, wc);
9167 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9174 BUG_ON(wc->stage != DROP_REFERENCE);
9178 if (wc->stage == DROP_REFERENCE) {
9180 btrfs_node_key(path->nodes[level],
9181 &root_item->drop_progress,
9182 path->slots[level]);
9183 root_item->drop_level = level;
9186 BUG_ON(wc->level == 0);
9187 if (btrfs_should_end_transaction(trans) ||
9188 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9189 ret = btrfs_update_root(trans, tree_root,
9193 btrfs_abort_transaction(trans, ret);
9198 btrfs_end_transaction_throttle(trans);
9199 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9200 btrfs_debug(fs_info,
9201 "drop snapshot early exit");
9206 trans = btrfs_start_transaction(tree_root, 0);
9207 if (IS_ERR(trans)) {
9208 err = PTR_ERR(trans);
9212 trans->block_rsv = block_rsv;
9215 btrfs_release_path(path);
9219 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9221 btrfs_abort_transaction(trans, ret);
9226 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9227 ret = btrfs_find_root(tree_root, &root->root_key, path,
9230 btrfs_abort_transaction(trans, ret);
9233 } else if (ret > 0) {
9234 /* if we fail to delete the orphan item this time
9235 * around, it'll get picked up the next time.
9237 * The most common failure here is just -ENOENT.
9239 btrfs_del_orphan_item(trans, tree_root,
9240 root->root_key.objectid);
9244 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9245 btrfs_add_dropped_root(trans, root);
9247 free_extent_buffer(root->node);
9248 free_extent_buffer(root->commit_root);
9249 btrfs_put_fs_root(root);
9251 root_dropped = true;
9253 btrfs_end_transaction_throttle(trans);
9256 btrfs_free_path(path);
9259 * So if we need to stop dropping the snapshot for whatever reason we
9260 * need to make sure to add it back to the dead root list so that we
9261 * keep trying to do the work later. This also cleans up roots if we
9262 * don't have it in the radix (like when we recover after a power fail
9263 * or unmount) so we don't leak memory.
9265 if (!for_reloc && !root_dropped)
9266 btrfs_add_dead_root(root);
9267 if (err && err != -EAGAIN)
9268 btrfs_handle_fs_error(fs_info, err, NULL);
9273 * drop subtree rooted at tree block 'node'.
9275 * NOTE: this function will unlock and release tree block 'node'
9276 * only used by relocation code
9278 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9279 struct btrfs_root *root,
9280 struct extent_buffer *node,
9281 struct extent_buffer *parent)
9283 struct btrfs_fs_info *fs_info = root->fs_info;
9284 struct btrfs_path *path;
9285 struct walk_control *wc;
9291 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9293 path = btrfs_alloc_path();
9297 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9299 btrfs_free_path(path);
9303 btrfs_assert_tree_locked(parent);
9304 parent_level = btrfs_header_level(parent);
9305 extent_buffer_get(parent);
9306 path->nodes[parent_level] = parent;
9307 path->slots[parent_level] = btrfs_header_nritems(parent);
9309 btrfs_assert_tree_locked(node);
9310 level = btrfs_header_level(node);
9311 path->nodes[level] = node;
9312 path->slots[level] = 0;
9313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9315 wc->refs[parent_level] = 1;
9316 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9318 wc->shared_level = -1;
9319 wc->stage = DROP_REFERENCE;
9323 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9326 wret = walk_down_tree(trans, root, path, wc);
9332 wret = walk_up_tree(trans, root, path, wc, parent_level);
9340 btrfs_free_path(path);
9344 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9350 * if restripe for this chunk_type is on pick target profile and
9351 * return, otherwise do the usual balance
9353 stripped = get_restripe_target(fs_info, flags);
9355 return extended_to_chunk(stripped);
9357 num_devices = fs_info->fs_devices->rw_devices;
9359 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9360 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9361 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9363 if (num_devices == 1) {
9364 stripped |= BTRFS_BLOCK_GROUP_DUP;
9365 stripped = flags & ~stripped;
9367 /* turn raid0 into single device chunks */
9368 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9371 /* turn mirroring into duplication */
9372 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9373 BTRFS_BLOCK_GROUP_RAID10))
9374 return stripped | BTRFS_BLOCK_GROUP_DUP;
9376 /* they already had raid on here, just return */
9377 if (flags & stripped)
9380 stripped |= BTRFS_BLOCK_GROUP_DUP;
9381 stripped = flags & ~stripped;
9383 /* switch duplicated blocks with raid1 */
9384 if (flags & BTRFS_BLOCK_GROUP_DUP)
9385 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9387 /* this is drive concat, leave it alone */
9393 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9395 struct btrfs_space_info *sinfo = cache->space_info;
9397 u64 min_allocable_bytes;
9401 * We need some metadata space and system metadata space for
9402 * allocating chunks in some corner cases until we force to set
9403 * it to be readonly.
9406 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9408 min_allocable_bytes = SZ_1M;
9410 min_allocable_bytes = 0;
9412 spin_lock(&sinfo->lock);
9413 spin_lock(&cache->lock);
9421 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9422 cache->bytes_super - btrfs_block_group_used(&cache->item);
9424 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9425 min_allocable_bytes <= sinfo->total_bytes) {
9426 sinfo->bytes_readonly += num_bytes;
9428 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9432 spin_unlock(&cache->lock);
9433 spin_unlock(&sinfo->lock);
9437 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9440 struct btrfs_fs_info *fs_info = cache->fs_info;
9441 struct btrfs_trans_handle *trans;
9446 trans = btrfs_join_transaction(fs_info->extent_root);
9448 return PTR_ERR(trans);
9451 * we're not allowed to set block groups readonly after the dirty
9452 * block groups cache has started writing. If it already started,
9453 * back off and let this transaction commit
9455 mutex_lock(&fs_info->ro_block_group_mutex);
9456 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9457 u64 transid = trans->transid;
9459 mutex_unlock(&fs_info->ro_block_group_mutex);
9460 btrfs_end_transaction(trans);
9462 ret = btrfs_wait_for_commit(fs_info, transid);
9469 * if we are changing raid levels, try to allocate a corresponding
9470 * block group with the new raid level.
9472 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9473 if (alloc_flags != cache->flags) {
9474 ret = do_chunk_alloc(trans, alloc_flags,
9477 * ENOSPC is allowed here, we may have enough space
9478 * already allocated at the new raid level to
9487 ret = inc_block_group_ro(cache, 0);
9490 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9491 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9494 ret = inc_block_group_ro(cache, 0);
9496 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9497 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9498 mutex_lock(&fs_info->chunk_mutex);
9499 check_system_chunk(trans, alloc_flags);
9500 mutex_unlock(&fs_info->chunk_mutex);
9502 mutex_unlock(&fs_info->ro_block_group_mutex);
9504 btrfs_end_transaction(trans);
9508 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9510 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9512 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9516 * helper to account the unused space of all the readonly block group in the
9517 * space_info. takes mirrors into account.
9519 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9521 struct btrfs_block_group_cache *block_group;
9525 /* It's df, we don't care if it's racy */
9526 if (list_empty(&sinfo->ro_bgs))
9529 spin_lock(&sinfo->lock);
9530 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9531 spin_lock(&block_group->lock);
9533 if (!block_group->ro) {
9534 spin_unlock(&block_group->lock);
9538 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9539 BTRFS_BLOCK_GROUP_RAID10 |
9540 BTRFS_BLOCK_GROUP_DUP))
9545 free_bytes += (block_group->key.offset -
9546 btrfs_block_group_used(&block_group->item)) *
9549 spin_unlock(&block_group->lock);
9551 spin_unlock(&sinfo->lock);
9556 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9558 struct btrfs_space_info *sinfo = cache->space_info;
9563 spin_lock(&sinfo->lock);
9564 spin_lock(&cache->lock);
9566 num_bytes = cache->key.offset - cache->reserved -
9567 cache->pinned - cache->bytes_super -
9568 btrfs_block_group_used(&cache->item);
9569 sinfo->bytes_readonly -= num_bytes;
9570 list_del_init(&cache->ro_list);
9572 spin_unlock(&cache->lock);
9573 spin_unlock(&sinfo->lock);
9577 * checks to see if its even possible to relocate this block group.
9579 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9580 * ok to go ahead and try.
9582 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9584 struct btrfs_root *root = fs_info->extent_root;
9585 struct btrfs_block_group_cache *block_group;
9586 struct btrfs_space_info *space_info;
9587 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9588 struct btrfs_device *device;
9589 struct btrfs_trans_handle *trans;
9599 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9601 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9603 /* odd, couldn't find the block group, leave it alone */
9607 "can't find block group for bytenr %llu",
9612 min_free = btrfs_block_group_used(&block_group->item);
9614 /* no bytes used, we're good */
9618 space_info = block_group->space_info;
9619 spin_lock(&space_info->lock);
9621 full = space_info->full;
9624 * if this is the last block group we have in this space, we can't
9625 * relocate it unless we're able to allocate a new chunk below.
9627 * Otherwise, we need to make sure we have room in the space to handle
9628 * all of the extents from this block group. If we can, we're good
9630 if ((space_info->total_bytes != block_group->key.offset) &&
9631 (btrfs_space_info_used(space_info, false) + min_free <
9632 space_info->total_bytes)) {
9633 spin_unlock(&space_info->lock);
9636 spin_unlock(&space_info->lock);
9639 * ok we don't have enough space, but maybe we have free space on our
9640 * devices to allocate new chunks for relocation, so loop through our
9641 * alloc devices and guess if we have enough space. if this block
9642 * group is going to be restriped, run checks against the target
9643 * profile instead of the current one.
9655 target = get_restripe_target(fs_info, block_group->flags);
9657 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9660 * this is just a balance, so if we were marked as full
9661 * we know there is no space for a new chunk
9666 "no space to alloc new chunk for block group %llu",
9667 block_group->key.objectid);
9671 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9674 if (index == BTRFS_RAID_RAID10) {
9678 } else if (index == BTRFS_RAID_RAID1) {
9680 } else if (index == BTRFS_RAID_DUP) {
9683 } else if (index == BTRFS_RAID_RAID0) {
9684 dev_min = fs_devices->rw_devices;
9685 min_free = div64_u64(min_free, dev_min);
9688 /* We need to do this so that we can look at pending chunks */
9689 trans = btrfs_join_transaction(root);
9690 if (IS_ERR(trans)) {
9691 ret = PTR_ERR(trans);
9695 mutex_lock(&fs_info->chunk_mutex);
9696 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9700 * check to make sure we can actually find a chunk with enough
9701 * space to fit our block group in.
9703 if (device->total_bytes > device->bytes_used + min_free &&
9704 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9705 ret = find_free_dev_extent(trans, device, min_free,
9710 if (dev_nr >= dev_min)
9716 if (debug && ret == -1)
9718 "no space to allocate a new chunk for block group %llu",
9719 block_group->key.objectid);
9720 mutex_unlock(&fs_info->chunk_mutex);
9721 btrfs_end_transaction(trans);
9723 btrfs_put_block_group(block_group);
9727 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9728 struct btrfs_path *path,
9729 struct btrfs_key *key)
9731 struct btrfs_root *root = fs_info->extent_root;
9733 struct btrfs_key found_key;
9734 struct extent_buffer *leaf;
9737 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9742 slot = path->slots[0];
9743 leaf = path->nodes[0];
9744 if (slot >= btrfs_header_nritems(leaf)) {
9745 ret = btrfs_next_leaf(root, path);
9752 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9754 if (found_key.objectid >= key->objectid &&
9755 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9756 struct extent_map_tree *em_tree;
9757 struct extent_map *em;
9759 em_tree = &root->fs_info->mapping_tree.map_tree;
9760 read_lock(&em_tree->lock);
9761 em = lookup_extent_mapping(em_tree, found_key.objectid,
9763 read_unlock(&em_tree->lock);
9766 "logical %llu len %llu found bg but no related chunk",
9767 found_key.objectid, found_key.offset);
9772 free_extent_map(em);
9781 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9783 struct btrfs_block_group_cache *block_group;
9787 struct inode *inode;
9789 block_group = btrfs_lookup_first_block_group(info, last);
9790 while (block_group) {
9791 spin_lock(&block_group->lock);
9792 if (block_group->iref)
9794 spin_unlock(&block_group->lock);
9795 block_group = next_block_group(info, block_group);
9804 inode = block_group->inode;
9805 block_group->iref = 0;
9806 block_group->inode = NULL;
9807 spin_unlock(&block_group->lock);
9808 ASSERT(block_group->io_ctl.inode == NULL);
9810 last = block_group->key.objectid + block_group->key.offset;
9811 btrfs_put_block_group(block_group);
9816 * Must be called only after stopping all workers, since we could have block
9817 * group caching kthreads running, and therefore they could race with us if we
9818 * freed the block groups before stopping them.
9820 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9822 struct btrfs_block_group_cache *block_group;
9823 struct btrfs_space_info *space_info;
9824 struct btrfs_caching_control *caching_ctl;
9827 down_write(&info->commit_root_sem);
9828 while (!list_empty(&info->caching_block_groups)) {
9829 caching_ctl = list_entry(info->caching_block_groups.next,
9830 struct btrfs_caching_control, list);
9831 list_del(&caching_ctl->list);
9832 put_caching_control(caching_ctl);
9834 up_write(&info->commit_root_sem);
9836 spin_lock(&info->unused_bgs_lock);
9837 while (!list_empty(&info->unused_bgs)) {
9838 block_group = list_first_entry(&info->unused_bgs,
9839 struct btrfs_block_group_cache,
9841 list_del_init(&block_group->bg_list);
9842 btrfs_put_block_group(block_group);
9844 spin_unlock(&info->unused_bgs_lock);
9846 spin_lock(&info->block_group_cache_lock);
9847 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9848 block_group = rb_entry(n, struct btrfs_block_group_cache,
9850 rb_erase(&block_group->cache_node,
9851 &info->block_group_cache_tree);
9852 RB_CLEAR_NODE(&block_group->cache_node);
9853 spin_unlock(&info->block_group_cache_lock);
9855 down_write(&block_group->space_info->groups_sem);
9856 list_del(&block_group->list);
9857 up_write(&block_group->space_info->groups_sem);
9860 * We haven't cached this block group, which means we could
9861 * possibly have excluded extents on this block group.
9863 if (block_group->cached == BTRFS_CACHE_NO ||
9864 block_group->cached == BTRFS_CACHE_ERROR)
9865 free_excluded_extents(block_group);
9867 btrfs_remove_free_space_cache(block_group);
9868 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9869 ASSERT(list_empty(&block_group->dirty_list));
9870 ASSERT(list_empty(&block_group->io_list));
9871 ASSERT(list_empty(&block_group->bg_list));
9872 ASSERT(atomic_read(&block_group->count) == 1);
9873 btrfs_put_block_group(block_group);
9875 spin_lock(&info->block_group_cache_lock);
9877 spin_unlock(&info->block_group_cache_lock);
9879 /* now that all the block groups are freed, go through and
9880 * free all the space_info structs. This is only called during
9881 * the final stages of unmount, and so we know nobody is
9882 * using them. We call synchronize_rcu() once before we start,
9883 * just to be on the safe side.
9887 release_global_block_rsv(info);
9889 while (!list_empty(&info->space_info)) {
9892 space_info = list_entry(info->space_info.next,
9893 struct btrfs_space_info,
9897 * Do not hide this behind enospc_debug, this is actually
9898 * important and indicates a real bug if this happens.
9900 if (WARN_ON(space_info->bytes_pinned > 0 ||
9901 space_info->bytes_reserved > 0 ||
9902 space_info->bytes_may_use > 0))
9903 dump_space_info(info, space_info, 0, 0);
9904 list_del(&space_info->list);
9905 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9906 struct kobject *kobj;
9907 kobj = space_info->block_group_kobjs[i];
9908 space_info->block_group_kobjs[i] = NULL;
9914 kobject_del(&space_info->kobj);
9915 kobject_put(&space_info->kobj);
9920 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9921 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9923 struct btrfs_space_info *space_info;
9924 struct raid_kobject *rkobj;
9929 spin_lock(&fs_info->pending_raid_kobjs_lock);
9930 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9931 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9933 list_for_each_entry(rkobj, &list, list) {
9934 space_info = __find_space_info(fs_info, rkobj->flags);
9935 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9937 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9938 "%s", get_raid_name(index));
9940 kobject_put(&rkobj->kobj);
9946 "failed to add kobject for block cache, ignoring");
9949 static void link_block_group(struct btrfs_block_group_cache *cache)
9951 struct btrfs_space_info *space_info = cache->space_info;
9952 struct btrfs_fs_info *fs_info = cache->fs_info;
9953 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9956 down_write(&space_info->groups_sem);
9957 if (list_empty(&space_info->block_groups[index]))
9959 list_add_tail(&cache->list, &space_info->block_groups[index]);
9960 up_write(&space_info->groups_sem);
9963 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9965 btrfs_warn(cache->fs_info,
9966 "couldn't alloc memory for raid level kobject");
9969 rkobj->flags = cache->flags;
9970 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9972 spin_lock(&fs_info->pending_raid_kobjs_lock);
9973 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9974 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9975 space_info->block_group_kobjs[index] = &rkobj->kobj;
9979 static struct btrfs_block_group_cache *
9980 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9981 u64 start, u64 size)
9983 struct btrfs_block_group_cache *cache;
9985 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9989 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9991 if (!cache->free_space_ctl) {
9996 cache->key.objectid = start;
9997 cache->key.offset = size;
9998 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10000 cache->fs_info = fs_info;
10001 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10002 set_free_space_tree_thresholds(cache);
10004 atomic_set(&cache->count, 1);
10005 spin_lock_init(&cache->lock);
10006 init_rwsem(&cache->data_rwsem);
10007 INIT_LIST_HEAD(&cache->list);
10008 INIT_LIST_HEAD(&cache->cluster_list);
10009 INIT_LIST_HEAD(&cache->bg_list);
10010 INIT_LIST_HEAD(&cache->ro_list);
10011 INIT_LIST_HEAD(&cache->dirty_list);
10012 INIT_LIST_HEAD(&cache->io_list);
10013 btrfs_init_free_space_ctl(cache);
10014 atomic_set(&cache->trimming, 0);
10015 mutex_init(&cache->free_space_lock);
10016 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10021 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10023 struct btrfs_path *path;
10025 struct btrfs_block_group_cache *cache;
10026 struct btrfs_space_info *space_info;
10027 struct btrfs_key key;
10028 struct btrfs_key found_key;
10029 struct extent_buffer *leaf;
10030 int need_clear = 0;
10035 feature = btrfs_super_incompat_flags(info->super_copy);
10036 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10040 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10041 path = btrfs_alloc_path();
10044 path->reada = READA_FORWARD;
10046 cache_gen = btrfs_super_cache_generation(info->super_copy);
10047 if (btrfs_test_opt(info, SPACE_CACHE) &&
10048 btrfs_super_generation(info->super_copy) != cache_gen)
10050 if (btrfs_test_opt(info, CLEAR_CACHE))
10054 ret = find_first_block_group(info, path, &key);
10060 leaf = path->nodes[0];
10061 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10063 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10072 * When we mount with old space cache, we need to
10073 * set BTRFS_DC_CLEAR and set dirty flag.
10075 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10076 * truncate the old free space cache inode and
10078 * b) Setting 'dirty flag' makes sure that we flush
10079 * the new space cache info onto disk.
10081 if (btrfs_test_opt(info, SPACE_CACHE))
10082 cache->disk_cache_state = BTRFS_DC_CLEAR;
10085 read_extent_buffer(leaf, &cache->item,
10086 btrfs_item_ptr_offset(leaf, path->slots[0]),
10087 sizeof(cache->item));
10088 cache->flags = btrfs_block_group_flags(&cache->item);
10090 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10091 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10093 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10094 cache->key.objectid);
10099 key.objectid = found_key.objectid + found_key.offset;
10100 btrfs_release_path(path);
10103 * We need to exclude the super stripes now so that the space
10104 * info has super bytes accounted for, otherwise we'll think
10105 * we have more space than we actually do.
10107 ret = exclude_super_stripes(cache);
10110 * We may have excluded something, so call this just in
10113 free_excluded_extents(cache);
10114 btrfs_put_block_group(cache);
10119 * check for two cases, either we are full, and therefore
10120 * don't need to bother with the caching work since we won't
10121 * find any space, or we are empty, and we can just add all
10122 * the space in and be done with it. This saves us _alot_ of
10123 * time, particularly in the full case.
10125 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10126 cache->last_byte_to_unpin = (u64)-1;
10127 cache->cached = BTRFS_CACHE_FINISHED;
10128 free_excluded_extents(cache);
10129 } else if (btrfs_block_group_used(&cache->item) == 0) {
10130 cache->last_byte_to_unpin = (u64)-1;
10131 cache->cached = BTRFS_CACHE_FINISHED;
10132 add_new_free_space(cache, found_key.objectid,
10133 found_key.objectid +
10135 free_excluded_extents(cache);
10138 ret = btrfs_add_block_group_cache(info, cache);
10140 btrfs_remove_free_space_cache(cache);
10141 btrfs_put_block_group(cache);
10145 trace_btrfs_add_block_group(info, cache, 0);
10146 update_space_info(info, cache->flags, found_key.offset,
10147 btrfs_block_group_used(&cache->item),
10148 cache->bytes_super, &space_info);
10150 cache->space_info = space_info;
10152 link_block_group(cache);
10154 set_avail_alloc_bits(info, cache->flags);
10155 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10156 inc_block_group_ro(cache, 1);
10157 } else if (btrfs_block_group_used(&cache->item) == 0) {
10158 spin_lock(&info->unused_bgs_lock);
10159 /* Should always be true but just in case. */
10160 if (list_empty(&cache->bg_list)) {
10161 btrfs_get_block_group(cache);
10162 trace_btrfs_add_unused_block_group(cache);
10163 list_add_tail(&cache->bg_list,
10164 &info->unused_bgs);
10166 spin_unlock(&info->unused_bgs_lock);
10170 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10171 if (!(get_alloc_profile(info, space_info->flags) &
10172 (BTRFS_BLOCK_GROUP_RAID10 |
10173 BTRFS_BLOCK_GROUP_RAID1 |
10174 BTRFS_BLOCK_GROUP_RAID5 |
10175 BTRFS_BLOCK_GROUP_RAID6 |
10176 BTRFS_BLOCK_GROUP_DUP)))
10179 * avoid allocating from un-mirrored block group if there are
10180 * mirrored block groups.
10182 list_for_each_entry(cache,
10183 &space_info->block_groups[BTRFS_RAID_RAID0],
10185 inc_block_group_ro(cache, 1);
10186 list_for_each_entry(cache,
10187 &space_info->block_groups[BTRFS_RAID_SINGLE],
10189 inc_block_group_ro(cache, 1);
10192 btrfs_add_raid_kobjects(info);
10193 init_global_block_rsv(info);
10196 btrfs_free_path(path);
10200 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10202 struct btrfs_fs_info *fs_info = trans->fs_info;
10203 struct btrfs_block_group_cache *block_group, *tmp;
10204 struct btrfs_root *extent_root = fs_info->extent_root;
10205 struct btrfs_block_group_item item;
10206 struct btrfs_key key;
10208 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10210 trans->can_flush_pending_bgs = false;
10211 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10215 spin_lock(&block_group->lock);
10216 memcpy(&item, &block_group->item, sizeof(item));
10217 memcpy(&key, &block_group->key, sizeof(key));
10218 spin_unlock(&block_group->lock);
10220 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10223 btrfs_abort_transaction(trans, ret);
10224 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10227 btrfs_abort_transaction(trans, ret);
10228 add_block_group_free_space(trans, block_group);
10229 /* already aborted the transaction if it failed. */
10231 list_del_init(&block_group->bg_list);
10233 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10236 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10237 u64 type, u64 chunk_offset, u64 size)
10239 struct btrfs_fs_info *fs_info = trans->fs_info;
10240 struct btrfs_block_group_cache *cache;
10243 btrfs_set_log_full_commit(fs_info, trans);
10245 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10249 btrfs_set_block_group_used(&cache->item, bytes_used);
10250 btrfs_set_block_group_chunk_objectid(&cache->item,
10251 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10252 btrfs_set_block_group_flags(&cache->item, type);
10254 cache->flags = type;
10255 cache->last_byte_to_unpin = (u64)-1;
10256 cache->cached = BTRFS_CACHE_FINISHED;
10257 cache->needs_free_space = 1;
10258 ret = exclude_super_stripes(cache);
10261 * We may have excluded something, so call this just in
10264 free_excluded_extents(cache);
10265 btrfs_put_block_group(cache);
10269 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10271 free_excluded_extents(cache);
10273 #ifdef CONFIG_BTRFS_DEBUG
10274 if (btrfs_should_fragment_free_space(cache)) {
10275 u64 new_bytes_used = size - bytes_used;
10277 bytes_used += new_bytes_used >> 1;
10278 fragment_free_space(cache);
10282 * Ensure the corresponding space_info object is created and
10283 * assigned to our block group. We want our bg to be added to the rbtree
10284 * with its ->space_info set.
10286 cache->space_info = __find_space_info(fs_info, cache->flags);
10287 ASSERT(cache->space_info);
10289 ret = btrfs_add_block_group_cache(fs_info, cache);
10291 btrfs_remove_free_space_cache(cache);
10292 btrfs_put_block_group(cache);
10297 * Now that our block group has its ->space_info set and is inserted in
10298 * the rbtree, update the space info's counters.
10300 trace_btrfs_add_block_group(fs_info, cache, 1);
10301 update_space_info(fs_info, cache->flags, size, bytes_used,
10302 cache->bytes_super, &cache->space_info);
10303 update_global_block_rsv(fs_info);
10305 link_block_group(cache);
10307 list_add_tail(&cache->bg_list, &trans->new_bgs);
10309 set_avail_alloc_bits(fs_info, type);
10313 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10315 u64 extra_flags = chunk_to_extended(flags) &
10316 BTRFS_EXTENDED_PROFILE_MASK;
10318 write_seqlock(&fs_info->profiles_lock);
10319 if (flags & BTRFS_BLOCK_GROUP_DATA)
10320 fs_info->avail_data_alloc_bits &= ~extra_flags;
10321 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10322 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10323 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10324 fs_info->avail_system_alloc_bits &= ~extra_flags;
10325 write_sequnlock(&fs_info->profiles_lock);
10328 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10329 u64 group_start, struct extent_map *em)
10331 struct btrfs_fs_info *fs_info = trans->fs_info;
10332 struct btrfs_root *root = fs_info->extent_root;
10333 struct btrfs_path *path;
10334 struct btrfs_block_group_cache *block_group;
10335 struct btrfs_free_cluster *cluster;
10336 struct btrfs_root *tree_root = fs_info->tree_root;
10337 struct btrfs_key key;
10338 struct inode *inode;
10339 struct kobject *kobj = NULL;
10343 struct btrfs_caching_control *caching_ctl = NULL;
10346 block_group = btrfs_lookup_block_group(fs_info, group_start);
10347 BUG_ON(!block_group);
10348 BUG_ON(!block_group->ro);
10350 trace_btrfs_remove_block_group(block_group);
10352 * Free the reserved super bytes from this block group before
10355 free_excluded_extents(block_group);
10356 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10357 block_group->key.offset);
10359 memcpy(&key, &block_group->key, sizeof(key));
10360 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10361 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10362 BTRFS_BLOCK_GROUP_RAID1 |
10363 BTRFS_BLOCK_GROUP_RAID10))
10368 /* make sure this block group isn't part of an allocation cluster */
10369 cluster = &fs_info->data_alloc_cluster;
10370 spin_lock(&cluster->refill_lock);
10371 btrfs_return_cluster_to_free_space(block_group, cluster);
10372 spin_unlock(&cluster->refill_lock);
10375 * make sure this block group isn't part of a metadata
10376 * allocation cluster
10378 cluster = &fs_info->meta_alloc_cluster;
10379 spin_lock(&cluster->refill_lock);
10380 btrfs_return_cluster_to_free_space(block_group, cluster);
10381 spin_unlock(&cluster->refill_lock);
10383 path = btrfs_alloc_path();
10390 * get the inode first so any iput calls done for the io_list
10391 * aren't the final iput (no unlinks allowed now)
10393 inode = lookup_free_space_inode(fs_info, block_group, path);
10395 mutex_lock(&trans->transaction->cache_write_mutex);
10397 * make sure our free spache cache IO is done before remove the
10400 spin_lock(&trans->transaction->dirty_bgs_lock);
10401 if (!list_empty(&block_group->io_list)) {
10402 list_del_init(&block_group->io_list);
10404 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10406 spin_unlock(&trans->transaction->dirty_bgs_lock);
10407 btrfs_wait_cache_io(trans, block_group, path);
10408 btrfs_put_block_group(block_group);
10409 spin_lock(&trans->transaction->dirty_bgs_lock);
10412 if (!list_empty(&block_group->dirty_list)) {
10413 list_del_init(&block_group->dirty_list);
10414 btrfs_put_block_group(block_group);
10416 spin_unlock(&trans->transaction->dirty_bgs_lock);
10417 mutex_unlock(&trans->transaction->cache_write_mutex);
10419 if (!IS_ERR(inode)) {
10420 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10422 btrfs_add_delayed_iput(inode);
10425 clear_nlink(inode);
10426 /* One for the block groups ref */
10427 spin_lock(&block_group->lock);
10428 if (block_group->iref) {
10429 block_group->iref = 0;
10430 block_group->inode = NULL;
10431 spin_unlock(&block_group->lock);
10434 spin_unlock(&block_group->lock);
10436 /* One for our lookup ref */
10437 btrfs_add_delayed_iput(inode);
10440 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10441 key.offset = block_group->key.objectid;
10444 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10448 btrfs_release_path(path);
10450 ret = btrfs_del_item(trans, tree_root, path);
10453 btrfs_release_path(path);
10456 spin_lock(&fs_info->block_group_cache_lock);
10457 rb_erase(&block_group->cache_node,
10458 &fs_info->block_group_cache_tree);
10459 RB_CLEAR_NODE(&block_group->cache_node);
10461 if (fs_info->first_logical_byte == block_group->key.objectid)
10462 fs_info->first_logical_byte = (u64)-1;
10463 spin_unlock(&fs_info->block_group_cache_lock);
10465 down_write(&block_group->space_info->groups_sem);
10467 * we must use list_del_init so people can check to see if they
10468 * are still on the list after taking the semaphore
10470 list_del_init(&block_group->list);
10471 if (list_empty(&block_group->space_info->block_groups[index])) {
10472 kobj = block_group->space_info->block_group_kobjs[index];
10473 block_group->space_info->block_group_kobjs[index] = NULL;
10474 clear_avail_alloc_bits(fs_info, block_group->flags);
10476 up_write(&block_group->space_info->groups_sem);
10482 if (block_group->has_caching_ctl)
10483 caching_ctl = get_caching_control(block_group);
10484 if (block_group->cached == BTRFS_CACHE_STARTED)
10485 wait_block_group_cache_done(block_group);
10486 if (block_group->has_caching_ctl) {
10487 down_write(&fs_info->commit_root_sem);
10488 if (!caching_ctl) {
10489 struct btrfs_caching_control *ctl;
10491 list_for_each_entry(ctl,
10492 &fs_info->caching_block_groups, list)
10493 if (ctl->block_group == block_group) {
10495 refcount_inc(&caching_ctl->count);
10500 list_del_init(&caching_ctl->list);
10501 up_write(&fs_info->commit_root_sem);
10503 /* Once for the caching bgs list and once for us. */
10504 put_caching_control(caching_ctl);
10505 put_caching_control(caching_ctl);
10509 spin_lock(&trans->transaction->dirty_bgs_lock);
10510 if (!list_empty(&block_group->dirty_list)) {
10513 if (!list_empty(&block_group->io_list)) {
10516 spin_unlock(&trans->transaction->dirty_bgs_lock);
10517 btrfs_remove_free_space_cache(block_group);
10519 spin_lock(&block_group->space_info->lock);
10520 list_del_init(&block_group->ro_list);
10522 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10523 WARN_ON(block_group->space_info->total_bytes
10524 < block_group->key.offset);
10525 WARN_ON(block_group->space_info->bytes_readonly
10526 < block_group->key.offset);
10527 WARN_ON(block_group->space_info->disk_total
10528 < block_group->key.offset * factor);
10530 block_group->space_info->total_bytes -= block_group->key.offset;
10531 block_group->space_info->bytes_readonly -= block_group->key.offset;
10532 block_group->space_info->disk_total -= block_group->key.offset * factor;
10534 spin_unlock(&block_group->space_info->lock);
10536 memcpy(&key, &block_group->key, sizeof(key));
10538 mutex_lock(&fs_info->chunk_mutex);
10539 if (!list_empty(&em->list)) {
10540 /* We're in the transaction->pending_chunks list. */
10541 free_extent_map(em);
10543 spin_lock(&block_group->lock);
10544 block_group->removed = 1;
10546 * At this point trimming can't start on this block group, because we
10547 * removed the block group from the tree fs_info->block_group_cache_tree
10548 * so no one can't find it anymore and even if someone already got this
10549 * block group before we removed it from the rbtree, they have already
10550 * incremented block_group->trimming - if they didn't, they won't find
10551 * any free space entries because we already removed them all when we
10552 * called btrfs_remove_free_space_cache().
10554 * And we must not remove the extent map from the fs_info->mapping_tree
10555 * to prevent the same logical address range and physical device space
10556 * ranges from being reused for a new block group. This is because our
10557 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10558 * completely transactionless, so while it is trimming a range the
10559 * currently running transaction might finish and a new one start,
10560 * allowing for new block groups to be created that can reuse the same
10561 * physical device locations unless we take this special care.
10563 * There may also be an implicit trim operation if the file system
10564 * is mounted with -odiscard. The same protections must remain
10565 * in place until the extents have been discarded completely when
10566 * the transaction commit has completed.
10568 remove_em = (atomic_read(&block_group->trimming) == 0);
10570 * Make sure a trimmer task always sees the em in the pinned_chunks list
10571 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10572 * before checking block_group->removed).
10576 * Our em might be in trans->transaction->pending_chunks which
10577 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10578 * and so is the fs_info->pinned_chunks list.
10580 * So at this point we must be holding the chunk_mutex to avoid
10581 * any races with chunk allocation (more specifically at
10582 * volumes.c:contains_pending_extent()), to ensure it always
10583 * sees the em, either in the pending_chunks list or in the
10584 * pinned_chunks list.
10586 list_move_tail(&em->list, &fs_info->pinned_chunks);
10588 spin_unlock(&block_group->lock);
10591 struct extent_map_tree *em_tree;
10593 em_tree = &fs_info->mapping_tree.map_tree;
10594 write_lock(&em_tree->lock);
10596 * The em might be in the pending_chunks list, so make sure the
10597 * chunk mutex is locked, since remove_extent_mapping() will
10598 * delete us from that list.
10600 remove_extent_mapping(em_tree, em);
10601 write_unlock(&em_tree->lock);
10602 /* once for the tree */
10603 free_extent_map(em);
10606 mutex_unlock(&fs_info->chunk_mutex);
10608 ret = remove_block_group_free_space(trans, block_group);
10612 btrfs_put_block_group(block_group);
10613 btrfs_put_block_group(block_group);
10615 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10621 ret = btrfs_del_item(trans, root, path);
10623 btrfs_free_path(path);
10627 struct btrfs_trans_handle *
10628 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10629 const u64 chunk_offset)
10631 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10632 struct extent_map *em;
10633 struct map_lookup *map;
10634 unsigned int num_items;
10636 read_lock(&em_tree->lock);
10637 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10638 read_unlock(&em_tree->lock);
10639 ASSERT(em && em->start == chunk_offset);
10642 * We need to reserve 3 + N units from the metadata space info in order
10643 * to remove a block group (done at btrfs_remove_chunk() and at
10644 * btrfs_remove_block_group()), which are used for:
10646 * 1 unit for adding the free space inode's orphan (located in the tree
10648 * 1 unit for deleting the block group item (located in the extent
10650 * 1 unit for deleting the free space item (located in tree of tree
10652 * N units for deleting N device extent items corresponding to each
10653 * stripe (located in the device tree).
10655 * In order to remove a block group we also need to reserve units in the
10656 * system space info in order to update the chunk tree (update one or
10657 * more device items and remove one chunk item), but this is done at
10658 * btrfs_remove_chunk() through a call to check_system_chunk().
10660 map = em->map_lookup;
10661 num_items = 3 + map->num_stripes;
10662 free_extent_map(em);
10664 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10669 * Process the unused_bgs list and remove any that don't have any allocated
10670 * space inside of them.
10672 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10674 struct btrfs_block_group_cache *block_group;
10675 struct btrfs_space_info *space_info;
10676 struct btrfs_trans_handle *trans;
10679 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10682 spin_lock(&fs_info->unused_bgs_lock);
10683 while (!list_empty(&fs_info->unused_bgs)) {
10687 block_group = list_first_entry(&fs_info->unused_bgs,
10688 struct btrfs_block_group_cache,
10690 list_del_init(&block_group->bg_list);
10692 space_info = block_group->space_info;
10694 if (ret || btrfs_mixed_space_info(space_info)) {
10695 btrfs_put_block_group(block_group);
10698 spin_unlock(&fs_info->unused_bgs_lock);
10700 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10702 /* Don't want to race with allocators so take the groups_sem */
10703 down_write(&space_info->groups_sem);
10704 spin_lock(&block_group->lock);
10705 if (block_group->reserved || block_group->pinned ||
10706 btrfs_block_group_used(&block_group->item) ||
10708 list_is_singular(&block_group->list)) {
10710 * We want to bail if we made new allocations or have
10711 * outstanding allocations in this block group. We do
10712 * the ro check in case balance is currently acting on
10713 * this block group.
10715 trace_btrfs_skip_unused_block_group(block_group);
10716 spin_unlock(&block_group->lock);
10717 up_write(&space_info->groups_sem);
10720 spin_unlock(&block_group->lock);
10722 /* We don't want to force the issue, only flip if it's ok. */
10723 ret = inc_block_group_ro(block_group, 0);
10724 up_write(&space_info->groups_sem);
10731 * Want to do this before we do anything else so we can recover
10732 * properly if we fail to join the transaction.
10734 trans = btrfs_start_trans_remove_block_group(fs_info,
10735 block_group->key.objectid);
10736 if (IS_ERR(trans)) {
10737 btrfs_dec_block_group_ro(block_group);
10738 ret = PTR_ERR(trans);
10743 * We could have pending pinned extents for this block group,
10744 * just delete them, we don't care about them anymore.
10746 start = block_group->key.objectid;
10747 end = start + block_group->key.offset - 1;
10749 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10750 * btrfs_finish_extent_commit(). If we are at transaction N,
10751 * another task might be running finish_extent_commit() for the
10752 * previous transaction N - 1, and have seen a range belonging
10753 * to the block group in freed_extents[] before we were able to
10754 * clear the whole block group range from freed_extents[]. This
10755 * means that task can lookup for the block group after we
10756 * unpinned it from freed_extents[] and removed it, leading to
10757 * a BUG_ON() at btrfs_unpin_extent_range().
10759 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10760 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10763 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10764 btrfs_dec_block_group_ro(block_group);
10767 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10770 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10771 btrfs_dec_block_group_ro(block_group);
10774 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10776 /* Reset pinned so btrfs_put_block_group doesn't complain */
10777 spin_lock(&space_info->lock);
10778 spin_lock(&block_group->lock);
10780 space_info->bytes_pinned -= block_group->pinned;
10781 space_info->bytes_readonly += block_group->pinned;
10782 percpu_counter_add(&space_info->total_bytes_pinned,
10783 -block_group->pinned);
10784 block_group->pinned = 0;
10786 spin_unlock(&block_group->lock);
10787 spin_unlock(&space_info->lock);
10789 /* DISCARD can flip during remount */
10790 trimming = btrfs_test_opt(fs_info, DISCARD);
10792 /* Implicit trim during transaction commit. */
10794 btrfs_get_block_group_trimming(block_group);
10797 * Btrfs_remove_chunk will abort the transaction if things go
10800 ret = btrfs_remove_chunk(trans, fs_info,
10801 block_group->key.objectid);
10805 btrfs_put_block_group_trimming(block_group);
10810 * If we're not mounted with -odiscard, we can just forget
10811 * about this block group. Otherwise we'll need to wait
10812 * until transaction commit to do the actual discard.
10815 spin_lock(&fs_info->unused_bgs_lock);
10817 * A concurrent scrub might have added us to the list
10818 * fs_info->unused_bgs, so use a list_move operation
10819 * to add the block group to the deleted_bgs list.
10821 list_move(&block_group->bg_list,
10822 &trans->transaction->deleted_bgs);
10823 spin_unlock(&fs_info->unused_bgs_lock);
10824 btrfs_get_block_group(block_group);
10827 btrfs_end_transaction(trans);
10829 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10830 btrfs_put_block_group(block_group);
10831 spin_lock(&fs_info->unused_bgs_lock);
10833 spin_unlock(&fs_info->unused_bgs_lock);
10836 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10838 struct btrfs_super_block *disk_super;
10844 disk_super = fs_info->super_copy;
10845 if (!btrfs_super_root(disk_super))
10848 features = btrfs_super_incompat_flags(disk_super);
10849 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10852 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10853 ret = create_space_info(fs_info, flags);
10858 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10859 ret = create_space_info(fs_info, flags);
10861 flags = BTRFS_BLOCK_GROUP_METADATA;
10862 ret = create_space_info(fs_info, flags);
10866 flags = BTRFS_BLOCK_GROUP_DATA;
10867 ret = create_space_info(fs_info, flags);
10873 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10874 u64 start, u64 end)
10876 return unpin_extent_range(fs_info, start, end, false);
10880 * It used to be that old block groups would be left around forever.
10881 * Iterating over them would be enough to trim unused space. Since we
10882 * now automatically remove them, we also need to iterate over unallocated
10885 * We don't want a transaction for this since the discard may take a
10886 * substantial amount of time. We don't require that a transaction be
10887 * running, but we do need to take a running transaction into account
10888 * to ensure that we're not discarding chunks that were released in
10889 * the current transaction.
10891 * Holding the chunks lock will prevent other threads from allocating
10892 * or releasing chunks, but it won't prevent a running transaction
10893 * from committing and releasing the memory that the pending chunks
10894 * list head uses. For that, we need to take a reference to the
10897 static int btrfs_trim_free_extents(struct btrfs_device *device,
10898 u64 minlen, u64 *trimmed)
10900 u64 start = 0, len = 0;
10905 /* Not writeable = nothing to do. */
10906 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10909 /* No free space = nothing to do. */
10910 if (device->total_bytes <= device->bytes_used)
10916 struct btrfs_fs_info *fs_info = device->fs_info;
10917 struct btrfs_transaction *trans;
10920 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10924 down_read(&fs_info->commit_root_sem);
10926 spin_lock(&fs_info->trans_lock);
10927 trans = fs_info->running_transaction;
10929 refcount_inc(&trans->use_count);
10930 spin_unlock(&fs_info->trans_lock);
10932 ret = find_free_dev_extent_start(trans, device, minlen, start,
10935 btrfs_put_transaction(trans);
10938 up_read(&fs_info->commit_root_sem);
10939 mutex_unlock(&fs_info->chunk_mutex);
10940 if (ret == -ENOSPC)
10945 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10946 up_read(&fs_info->commit_root_sem);
10947 mutex_unlock(&fs_info->chunk_mutex);
10955 if (fatal_signal_pending(current)) {
10956 ret = -ERESTARTSYS;
10966 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10968 struct btrfs_block_group_cache *cache = NULL;
10969 struct btrfs_device *device;
10970 struct list_head *devices;
10975 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10979 * try to trim all FS space, our block group may start from non-zero.
10981 if (range->len == total_bytes)
10982 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10984 cache = btrfs_lookup_block_group(fs_info, range->start);
10987 if (cache->key.objectid >= (range->start + range->len)) {
10988 btrfs_put_block_group(cache);
10992 start = max(range->start, cache->key.objectid);
10993 end = min(range->start + range->len,
10994 cache->key.objectid + cache->key.offset);
10996 if (end - start >= range->minlen) {
10997 if (!block_group_cache_done(cache)) {
10998 ret = cache_block_group(cache, 0);
11000 btrfs_put_block_group(cache);
11003 ret = wait_block_group_cache_done(cache);
11005 btrfs_put_block_group(cache);
11009 ret = btrfs_trim_block_group(cache,
11015 trimmed += group_trimmed;
11017 btrfs_put_block_group(cache);
11022 cache = next_block_group(fs_info, cache);
11025 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11026 devices = &fs_info->fs_devices->alloc_list;
11027 list_for_each_entry(device, devices, dev_alloc_list) {
11028 ret = btrfs_trim_free_extents(device, range->minlen,
11033 trimmed += group_trimmed;
11035 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11037 range->len = trimmed;
11042 * btrfs_{start,end}_write_no_snapshotting() are similar to
11043 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11044 * data into the page cache through nocow before the subvolume is snapshoted,
11045 * but flush the data into disk after the snapshot creation, or to prevent
11046 * operations while snapshotting is ongoing and that cause the snapshot to be
11047 * inconsistent (writes followed by expanding truncates for example).
11049 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11051 percpu_counter_dec(&root->subv_writers->counter);
11052 cond_wake_up(&root->subv_writers->wait);
11055 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11057 if (atomic_read(&root->will_be_snapshotted))
11060 percpu_counter_inc(&root->subv_writers->counter);
11062 * Make sure counter is updated before we check for snapshot creation.
11065 if (atomic_read(&root->will_be_snapshotted)) {
11066 btrfs_end_write_no_snapshotting(root);
11072 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11077 ret = btrfs_start_write_no_snapshotting(root);
11080 wait_var_event(&root->will_be_snapshotted,
11081 !atomic_read(&root->will_be_snapshotted));