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_batch(&space_info->total_bytes_pinned, num_bytes,
759 BTRFS_TOTAL_BYTES_PINNED_BATCH);
763 * after adding space to the filesystem, we need to clear the full flags
764 * on all the space infos.
766 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
768 struct list_head *head = &info->space_info;
769 struct btrfs_space_info *found;
772 list_for_each_entry_rcu(found, head, list)
777 /* simple helper to search for an existing data extent at a given offset */
778 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
781 struct btrfs_key key;
782 struct btrfs_path *path;
784 path = btrfs_alloc_path();
788 key.objectid = start;
790 key.type = BTRFS_EXTENT_ITEM_KEY;
791 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
792 btrfs_free_path(path);
797 * helper function to lookup reference count and flags of a tree block.
799 * the head node for delayed ref is used to store the sum of all the
800 * reference count modifications queued up in the rbtree. the head
801 * node may also store the extent flags to set. This way you can check
802 * to see what the reference count and extent flags would be if all of
803 * the delayed refs are not processed.
805 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
806 struct btrfs_fs_info *fs_info, u64 bytenr,
807 u64 offset, int metadata, u64 *refs, u64 *flags)
809 struct btrfs_delayed_ref_head *head;
810 struct btrfs_delayed_ref_root *delayed_refs;
811 struct btrfs_path *path;
812 struct btrfs_extent_item *ei;
813 struct extent_buffer *leaf;
814 struct btrfs_key key;
821 * If we don't have skinny metadata, don't bother doing anything
824 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
825 offset = fs_info->nodesize;
829 path = btrfs_alloc_path();
834 path->skip_locking = 1;
835 path->search_commit_root = 1;
839 key.objectid = bytenr;
842 key.type = BTRFS_METADATA_ITEM_KEY;
844 key.type = BTRFS_EXTENT_ITEM_KEY;
846 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
850 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
851 if (path->slots[0]) {
853 btrfs_item_key_to_cpu(path->nodes[0], &key,
855 if (key.objectid == bytenr &&
856 key.type == BTRFS_EXTENT_ITEM_KEY &&
857 key.offset == fs_info->nodesize)
863 leaf = path->nodes[0];
864 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
865 if (item_size >= sizeof(*ei)) {
866 ei = btrfs_item_ptr(leaf, path->slots[0],
867 struct btrfs_extent_item);
868 num_refs = btrfs_extent_refs(leaf, ei);
869 extent_flags = btrfs_extent_flags(leaf, ei);
872 btrfs_print_v0_err(fs_info);
874 btrfs_abort_transaction(trans, ret);
876 btrfs_handle_fs_error(fs_info, ret, NULL);
881 BUG_ON(num_refs == 0);
891 delayed_refs = &trans->transaction->delayed_refs;
892 spin_lock(&delayed_refs->lock);
893 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
895 if (!mutex_trylock(&head->mutex)) {
896 refcount_inc(&head->refs);
897 spin_unlock(&delayed_refs->lock);
899 btrfs_release_path(path);
902 * Mutex was contended, block until it's released and try
905 mutex_lock(&head->mutex);
906 mutex_unlock(&head->mutex);
907 btrfs_put_delayed_ref_head(head);
910 spin_lock(&head->lock);
911 if (head->extent_op && head->extent_op->update_flags)
912 extent_flags |= head->extent_op->flags_to_set;
914 BUG_ON(num_refs == 0);
916 num_refs += head->ref_mod;
917 spin_unlock(&head->lock);
918 mutex_unlock(&head->mutex);
920 spin_unlock(&delayed_refs->lock);
922 WARN_ON(num_refs == 0);
926 *flags = extent_flags;
928 btrfs_free_path(path);
933 * Back reference rules. Back refs have three main goals:
935 * 1) differentiate between all holders of references to an extent so that
936 * when a reference is dropped we can make sure it was a valid reference
937 * before freeing the extent.
939 * 2) Provide enough information to quickly find the holders of an extent
940 * if we notice a given block is corrupted or bad.
942 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
943 * maintenance. This is actually the same as #2, but with a slightly
944 * different use case.
946 * There are two kinds of back refs. The implicit back refs is optimized
947 * for pointers in non-shared tree blocks. For a given pointer in a block,
948 * back refs of this kind provide information about the block's owner tree
949 * and the pointer's key. These information allow us to find the block by
950 * b-tree searching. The full back refs is for pointers in tree blocks not
951 * referenced by their owner trees. The location of tree block is recorded
952 * in the back refs. Actually the full back refs is generic, and can be
953 * used in all cases the implicit back refs is used. The major shortcoming
954 * of the full back refs is its overhead. Every time a tree block gets
955 * COWed, we have to update back refs entry for all pointers in it.
957 * For a newly allocated tree block, we use implicit back refs for
958 * pointers in it. This means most tree related operations only involve
959 * implicit back refs. For a tree block created in old transaction, the
960 * only way to drop a reference to it is COW it. So we can detect the
961 * event that tree block loses its owner tree's reference and do the
962 * back refs conversion.
964 * When a tree block is COWed through a tree, there are four cases:
966 * The reference count of the block is one and the tree is the block's
967 * owner tree. Nothing to do in this case.
969 * The reference count of the block is one and the tree is not the
970 * block's owner tree. In this case, full back refs is used for pointers
971 * in the block. Remove these full back refs, add implicit back refs for
972 * every pointers in the new block.
974 * The reference count of the block is greater than one and the tree is
975 * the block's owner tree. In this case, implicit back refs is used for
976 * pointers in the block. Add full back refs for every pointers in the
977 * block, increase lower level extents' reference counts. The original
978 * implicit back refs are entailed to the new block.
980 * The reference count of the block is greater than one and the tree is
981 * not the block's owner tree. Add implicit back refs for every pointer in
982 * the new block, increase lower level extents' reference count.
984 * Back Reference Key composing:
986 * The key objectid corresponds to the first byte in the extent,
987 * The key type is used to differentiate between types of back refs.
988 * There are different meanings of the key offset for different types
991 * File extents can be referenced by:
993 * - multiple snapshots, subvolumes, or different generations in one subvol
994 * - different files inside a single subvolume
995 * - different offsets inside a file (bookend extents in file.c)
997 * The extent ref structure for the implicit back refs has fields for:
999 * - Objectid of the subvolume root
1000 * - objectid of the file holding the reference
1001 * - original offset in the file
1002 * - how many bookend extents
1004 * The key offset for the implicit back refs is hash of the first
1007 * The extent ref structure for the full back refs has field for:
1009 * - number of pointers in the tree leaf
1011 * The key offset for the implicit back refs is the first byte of
1014 * When a file extent is allocated, The implicit back refs is used.
1015 * the fields are filled in:
1017 * (root_key.objectid, inode objectid, offset in file, 1)
1019 * When a file extent is removed file truncation, we find the
1020 * corresponding implicit back refs and check the following fields:
1022 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1024 * Btree extents can be referenced by:
1026 * - Different subvolumes
1028 * Both the implicit back refs and the full back refs for tree blocks
1029 * only consist of key. The key offset for the implicit back refs is
1030 * objectid of block's owner tree. The key offset for the full back refs
1031 * is the first byte of parent block.
1033 * When implicit back refs is used, information about the lowest key and
1034 * level of the tree block are required. These information are stored in
1035 * tree block info structure.
1039 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1040 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1041 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1043 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1044 struct btrfs_extent_inline_ref *iref,
1045 enum btrfs_inline_ref_type is_data)
1047 int type = btrfs_extent_inline_ref_type(eb, iref);
1048 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1050 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1051 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1052 type == BTRFS_SHARED_DATA_REF_KEY ||
1053 type == BTRFS_EXTENT_DATA_REF_KEY) {
1054 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1055 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1057 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1058 ASSERT(eb->fs_info);
1060 * Every shared one has parent tree
1061 * block, which must be aligned to
1065 IS_ALIGNED(offset, eb->fs_info->nodesize))
1068 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1069 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1071 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1072 ASSERT(eb->fs_info);
1074 * Every shared one has parent tree
1075 * block, which must be aligned to
1079 IS_ALIGNED(offset, eb->fs_info->nodesize))
1083 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1088 btrfs_print_leaf((struct extent_buffer *)eb);
1089 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1093 return BTRFS_REF_TYPE_INVALID;
1096 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1098 u32 high_crc = ~(u32)0;
1099 u32 low_crc = ~(u32)0;
1102 lenum = cpu_to_le64(root_objectid);
1103 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1104 lenum = cpu_to_le64(owner);
1105 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1106 lenum = cpu_to_le64(offset);
1107 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1109 return ((u64)high_crc << 31) ^ (u64)low_crc;
1112 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1113 struct btrfs_extent_data_ref *ref)
1115 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1116 btrfs_extent_data_ref_objectid(leaf, ref),
1117 btrfs_extent_data_ref_offset(leaf, ref));
1120 static int match_extent_data_ref(struct extent_buffer *leaf,
1121 struct btrfs_extent_data_ref *ref,
1122 u64 root_objectid, u64 owner, u64 offset)
1124 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1125 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1126 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1131 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1132 struct btrfs_path *path,
1133 u64 bytenr, u64 parent,
1135 u64 owner, u64 offset)
1137 struct btrfs_root *root = trans->fs_info->extent_root;
1138 struct btrfs_key key;
1139 struct btrfs_extent_data_ref *ref;
1140 struct extent_buffer *leaf;
1146 key.objectid = bytenr;
1148 key.type = BTRFS_SHARED_DATA_REF_KEY;
1149 key.offset = parent;
1151 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1152 key.offset = hash_extent_data_ref(root_objectid,
1157 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1169 leaf = path->nodes[0];
1170 nritems = btrfs_header_nritems(leaf);
1172 if (path->slots[0] >= nritems) {
1173 ret = btrfs_next_leaf(root, path);
1179 leaf = path->nodes[0];
1180 nritems = btrfs_header_nritems(leaf);
1184 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1185 if (key.objectid != bytenr ||
1186 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1189 ref = btrfs_item_ptr(leaf, path->slots[0],
1190 struct btrfs_extent_data_ref);
1192 if (match_extent_data_ref(leaf, ref, root_objectid,
1195 btrfs_release_path(path);
1207 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1208 struct btrfs_path *path,
1209 u64 bytenr, u64 parent,
1210 u64 root_objectid, u64 owner,
1211 u64 offset, int refs_to_add)
1213 struct btrfs_root *root = trans->fs_info->extent_root;
1214 struct btrfs_key key;
1215 struct extent_buffer *leaf;
1220 key.objectid = bytenr;
1222 key.type = BTRFS_SHARED_DATA_REF_KEY;
1223 key.offset = parent;
1224 size = sizeof(struct btrfs_shared_data_ref);
1226 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1227 key.offset = hash_extent_data_ref(root_objectid,
1229 size = sizeof(struct btrfs_extent_data_ref);
1232 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1233 if (ret && ret != -EEXIST)
1236 leaf = path->nodes[0];
1238 struct btrfs_shared_data_ref *ref;
1239 ref = btrfs_item_ptr(leaf, path->slots[0],
1240 struct btrfs_shared_data_ref);
1242 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1244 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1245 num_refs += refs_to_add;
1246 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1249 struct btrfs_extent_data_ref *ref;
1250 while (ret == -EEXIST) {
1251 ref = btrfs_item_ptr(leaf, path->slots[0],
1252 struct btrfs_extent_data_ref);
1253 if (match_extent_data_ref(leaf, ref, root_objectid,
1256 btrfs_release_path(path);
1258 ret = btrfs_insert_empty_item(trans, root, path, &key,
1260 if (ret && ret != -EEXIST)
1263 leaf = path->nodes[0];
1265 ref = btrfs_item_ptr(leaf, path->slots[0],
1266 struct btrfs_extent_data_ref);
1268 btrfs_set_extent_data_ref_root(leaf, ref,
1270 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1271 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1272 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1274 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1275 num_refs += refs_to_add;
1276 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1279 btrfs_mark_buffer_dirty(leaf);
1282 btrfs_release_path(path);
1286 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1287 struct btrfs_path *path,
1288 int refs_to_drop, int *last_ref)
1290 struct btrfs_key key;
1291 struct btrfs_extent_data_ref *ref1 = NULL;
1292 struct btrfs_shared_data_ref *ref2 = NULL;
1293 struct extent_buffer *leaf;
1297 leaf = path->nodes[0];
1298 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1300 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1301 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_data_ref);
1303 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1304 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1305 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_shared_data_ref);
1307 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1308 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
1309 btrfs_print_v0_err(trans->fs_info);
1310 btrfs_abort_transaction(trans, -EINVAL);
1316 BUG_ON(num_refs < refs_to_drop);
1317 num_refs -= refs_to_drop;
1319 if (num_refs == 0) {
1320 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1323 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1324 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1325 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1326 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1327 btrfs_mark_buffer_dirty(leaf);
1332 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1333 struct btrfs_extent_inline_ref *iref)
1335 struct btrfs_key key;
1336 struct extent_buffer *leaf;
1337 struct btrfs_extent_data_ref *ref1;
1338 struct btrfs_shared_data_ref *ref2;
1342 leaf = path->nodes[0];
1343 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1345 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
1348 * If type is invalid, we should have bailed out earlier than
1351 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1352 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1353 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1354 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1355 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1357 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1358 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1360 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1361 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1362 struct btrfs_extent_data_ref);
1363 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1364 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1365 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_shared_data_ref);
1367 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1374 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1375 struct btrfs_path *path,
1376 u64 bytenr, u64 parent,
1379 struct btrfs_root *root = trans->fs_info->extent_root;
1380 struct btrfs_key key;
1383 key.objectid = bytenr;
1385 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1386 key.offset = parent;
1388 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1389 key.offset = root_objectid;
1392 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1398 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1399 struct btrfs_path *path,
1400 u64 bytenr, u64 parent,
1403 struct btrfs_key key;
1406 key.objectid = bytenr;
1408 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1409 key.offset = parent;
1411 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1412 key.offset = root_objectid;
1415 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1417 btrfs_release_path(path);
1421 static inline int extent_ref_type(u64 parent, u64 owner)
1424 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1426 type = BTRFS_SHARED_BLOCK_REF_KEY;
1428 type = BTRFS_TREE_BLOCK_REF_KEY;
1431 type = BTRFS_SHARED_DATA_REF_KEY;
1433 type = BTRFS_EXTENT_DATA_REF_KEY;
1438 static int find_next_key(struct btrfs_path *path, int level,
1439 struct btrfs_key *key)
1442 for (; level < BTRFS_MAX_LEVEL; level++) {
1443 if (!path->nodes[level])
1445 if (path->slots[level] + 1 >=
1446 btrfs_header_nritems(path->nodes[level]))
1449 btrfs_item_key_to_cpu(path->nodes[level], key,
1450 path->slots[level] + 1);
1452 btrfs_node_key_to_cpu(path->nodes[level], key,
1453 path->slots[level] + 1);
1460 * look for inline back ref. if back ref is found, *ref_ret is set
1461 * to the address of inline back ref, and 0 is returned.
1463 * if back ref isn't found, *ref_ret is set to the address where it
1464 * should be inserted, and -ENOENT is returned.
1466 * if insert is true and there are too many inline back refs, the path
1467 * points to the extent item, and -EAGAIN is returned.
1469 * NOTE: inline back refs are ordered in the same way that back ref
1470 * items in the tree are ordered.
1472 static noinline_for_stack
1473 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1474 struct btrfs_path *path,
1475 struct btrfs_extent_inline_ref **ref_ret,
1476 u64 bytenr, u64 num_bytes,
1477 u64 parent, u64 root_objectid,
1478 u64 owner, u64 offset, int insert)
1480 struct btrfs_fs_info *fs_info = trans->fs_info;
1481 struct btrfs_root *root = fs_info->extent_root;
1482 struct btrfs_key key;
1483 struct extent_buffer *leaf;
1484 struct btrfs_extent_item *ei;
1485 struct btrfs_extent_inline_ref *iref;
1495 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1498 key.objectid = bytenr;
1499 key.type = BTRFS_EXTENT_ITEM_KEY;
1500 key.offset = num_bytes;
1502 want = extent_ref_type(parent, owner);
1504 extra_size = btrfs_extent_inline_ref_size(want);
1505 path->keep_locks = 1;
1510 * Owner is our level, so we can just add one to get the level for the
1511 * block we are interested in.
1513 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1514 key.type = BTRFS_METADATA_ITEM_KEY;
1519 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1526 * We may be a newly converted file system which still has the old fat
1527 * extent entries for metadata, so try and see if we have one of those.
1529 if (ret > 0 && skinny_metadata) {
1530 skinny_metadata = false;
1531 if (path->slots[0]) {
1533 btrfs_item_key_to_cpu(path->nodes[0], &key,
1535 if (key.objectid == bytenr &&
1536 key.type == BTRFS_EXTENT_ITEM_KEY &&
1537 key.offset == num_bytes)
1541 key.objectid = bytenr;
1542 key.type = BTRFS_EXTENT_ITEM_KEY;
1543 key.offset = num_bytes;
1544 btrfs_release_path(path);
1549 if (ret && !insert) {
1552 } else if (WARN_ON(ret)) {
1557 leaf = path->nodes[0];
1558 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1559 if (unlikely(item_size < sizeof(*ei))) {
1561 btrfs_print_v0_err(fs_info);
1562 btrfs_abort_transaction(trans, err);
1566 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1567 flags = btrfs_extent_flags(leaf, ei);
1569 ptr = (unsigned long)(ei + 1);
1570 end = (unsigned long)ei + item_size;
1572 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1573 ptr += sizeof(struct btrfs_tree_block_info);
1577 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1578 needed = BTRFS_REF_TYPE_DATA;
1580 needed = BTRFS_REF_TYPE_BLOCK;
1588 iref = (struct btrfs_extent_inline_ref *)ptr;
1589 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1590 if (type == BTRFS_REF_TYPE_INVALID) {
1598 ptr += btrfs_extent_inline_ref_size(type);
1602 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1603 struct btrfs_extent_data_ref *dref;
1604 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1605 if (match_extent_data_ref(leaf, dref, root_objectid,
1610 if (hash_extent_data_ref_item(leaf, dref) <
1611 hash_extent_data_ref(root_objectid, owner, offset))
1615 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1617 if (parent == ref_offset) {
1621 if (ref_offset < parent)
1624 if (root_objectid == ref_offset) {
1628 if (ref_offset < root_objectid)
1632 ptr += btrfs_extent_inline_ref_size(type);
1634 if (err == -ENOENT && insert) {
1635 if (item_size + extra_size >=
1636 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1641 * To add new inline back ref, we have to make sure
1642 * there is no corresponding back ref item.
1643 * For simplicity, we just do not add new inline back
1644 * ref if there is any kind of item for this block
1646 if (find_next_key(path, 0, &key) == 0 &&
1647 key.objectid == bytenr &&
1648 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1653 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1656 path->keep_locks = 0;
1657 btrfs_unlock_up_safe(path, 1);
1663 * helper to add new inline back ref
1665 static noinline_for_stack
1666 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1667 struct btrfs_path *path,
1668 struct btrfs_extent_inline_ref *iref,
1669 u64 parent, u64 root_objectid,
1670 u64 owner, u64 offset, int refs_to_add,
1671 struct btrfs_delayed_extent_op *extent_op)
1673 struct extent_buffer *leaf;
1674 struct btrfs_extent_item *ei;
1677 unsigned long item_offset;
1682 leaf = path->nodes[0];
1683 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1684 item_offset = (unsigned long)iref - (unsigned long)ei;
1686 type = extent_ref_type(parent, owner);
1687 size = btrfs_extent_inline_ref_size(type);
1689 btrfs_extend_item(fs_info, path, size);
1691 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1692 refs = btrfs_extent_refs(leaf, ei);
1693 refs += refs_to_add;
1694 btrfs_set_extent_refs(leaf, ei, refs);
1696 __run_delayed_extent_op(extent_op, leaf, ei);
1698 ptr = (unsigned long)ei + item_offset;
1699 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1700 if (ptr < end - size)
1701 memmove_extent_buffer(leaf, ptr + size, ptr,
1704 iref = (struct btrfs_extent_inline_ref *)ptr;
1705 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1706 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1707 struct btrfs_extent_data_ref *dref;
1708 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1709 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1710 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1711 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1712 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1713 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1714 struct btrfs_shared_data_ref *sref;
1715 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1716 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1719 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1723 btrfs_mark_buffer_dirty(leaf);
1726 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1727 struct btrfs_path *path,
1728 struct btrfs_extent_inline_ref **ref_ret,
1729 u64 bytenr, u64 num_bytes, u64 parent,
1730 u64 root_objectid, u64 owner, u64 offset)
1734 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1735 num_bytes, parent, root_objectid,
1740 btrfs_release_path(path);
1743 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1744 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1747 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1748 root_objectid, owner, offset);
1754 * helper to update/remove inline back ref
1756 static noinline_for_stack
1757 void update_inline_extent_backref(struct btrfs_path *path,
1758 struct btrfs_extent_inline_ref *iref,
1760 struct btrfs_delayed_extent_op *extent_op,
1763 struct extent_buffer *leaf = path->nodes[0];
1764 struct btrfs_fs_info *fs_info = leaf->fs_info;
1765 struct btrfs_extent_item *ei;
1766 struct btrfs_extent_data_ref *dref = NULL;
1767 struct btrfs_shared_data_ref *sref = NULL;
1775 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1776 refs = btrfs_extent_refs(leaf, ei);
1777 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1778 refs += refs_to_mod;
1779 btrfs_set_extent_refs(leaf, ei, refs);
1781 __run_delayed_extent_op(extent_op, leaf, ei);
1784 * If type is invalid, we should have bailed out after
1785 * lookup_inline_extent_backref().
1787 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1788 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1811 size = btrfs_extent_inline_ref_size(type);
1812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1813 ptr = (unsigned long)iref;
1814 end = (unsigned long)ei + item_size;
1815 if (ptr + size < end)
1816 memmove_extent_buffer(leaf, ptr, ptr + size,
1819 btrfs_truncate_item(fs_info, path, item_size, 1);
1821 btrfs_mark_buffer_dirty(leaf);
1824 static noinline_for_stack
1825 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1836 num_bytes, parent, root_objectid,
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(path, iref, refs_to_add,
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_path *path,
1853 u64 bytenr, u64 parent, u64 root_objectid,
1854 u64 owner, u64 offset, int refs_to_add)
1857 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1858 BUG_ON(refs_to_add != 1);
1859 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1862 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1863 root_objectid, owner, offset,
1869 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1870 struct btrfs_path *path,
1871 struct btrfs_extent_inline_ref *iref,
1872 int refs_to_drop, int is_data, int *last_ref)
1876 BUG_ON(!is_data && refs_to_drop != 1);
1878 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1880 } else if (is_data) {
1881 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1885 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1890 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1891 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1892 u64 *discarded_bytes)
1895 u64 bytes_left, end;
1896 u64 aligned_start = ALIGN(start, 1 << 9);
1898 if (WARN_ON(start != aligned_start)) {
1899 len -= aligned_start - start;
1900 len = round_down(len, 1 << 9);
1901 start = aligned_start;
1904 *discarded_bytes = 0;
1912 /* Skip any superblocks on this device. */
1913 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1914 u64 sb_start = btrfs_sb_offset(j);
1915 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1916 u64 size = sb_start - start;
1918 if (!in_range(sb_start, start, bytes_left) &&
1919 !in_range(sb_end, start, bytes_left) &&
1920 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1924 * Superblock spans beginning of range. Adjust start and
1927 if (sb_start <= start) {
1928 start += sb_end - start;
1933 bytes_left = end - start;
1938 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1941 *discarded_bytes += size;
1942 else if (ret != -EOPNOTSUPP)
1951 bytes_left = end - start;
1955 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1958 *discarded_bytes += bytes_left;
1963 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1964 u64 num_bytes, u64 *actual_bytes)
1967 u64 discarded_bytes = 0;
1968 struct btrfs_bio *bbio = NULL;
1972 * Avoid races with device replace and make sure our bbio has devices
1973 * associated to its stripes that don't go away while we are discarding.
1975 btrfs_bio_counter_inc_blocked(fs_info);
1976 /* Tell the block device(s) that the sectors can be discarded */
1977 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
1979 /* Error condition is -ENOMEM */
1981 struct btrfs_bio_stripe *stripe = bbio->stripes;
1985 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1987 struct request_queue *req_q;
1989 if (!stripe->dev->bdev) {
1990 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1993 req_q = bdev_get_queue(stripe->dev->bdev);
1994 if (!blk_queue_discard(req_q))
1997 ret = btrfs_issue_discard(stripe->dev->bdev,
2002 discarded_bytes += bytes;
2003 else if (ret != -EOPNOTSUPP)
2004 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2007 * Just in case we get back EOPNOTSUPP for some reason,
2008 * just ignore the return value so we don't screw up
2009 * people calling discard_extent.
2013 btrfs_put_bbio(bbio);
2015 btrfs_bio_counter_dec(fs_info);
2018 *actual_bytes = discarded_bytes;
2021 if (ret == -EOPNOTSUPP)
2026 /* Can return -ENOMEM */
2027 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2028 struct btrfs_root *root,
2029 u64 bytenr, u64 num_bytes, u64 parent,
2030 u64 root_objectid, u64 owner, u64 offset)
2032 struct btrfs_fs_info *fs_info = root->fs_info;
2033 int old_ref_mod, new_ref_mod;
2036 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2037 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2039 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2040 owner, offset, BTRFS_ADD_DELAYED_REF);
2042 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2043 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
2045 root_objectid, (int)owner,
2046 BTRFS_ADD_DELAYED_REF, NULL,
2047 &old_ref_mod, &new_ref_mod);
2049 ret = btrfs_add_delayed_data_ref(trans, bytenr,
2051 root_objectid, owner, offset,
2052 0, BTRFS_ADD_DELAYED_REF,
2053 &old_ref_mod, &new_ref_mod);
2056 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2057 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2059 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2066 * __btrfs_inc_extent_ref - insert backreference for a given extent
2068 * @trans: Handle of transaction
2070 * @node: The delayed ref node used to get the bytenr/length for
2071 * extent whose references are incremented.
2073 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2074 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2075 * bytenr of the parent block. Since new extents are always
2076 * created with indirect references, this will only be the case
2077 * when relocating a shared extent. In that case, root_objectid
2078 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2081 * @root_objectid: The id of the root where this modification has originated,
2082 * this can be either one of the well-known metadata trees or
2083 * the subvolume id which references this extent.
2085 * @owner: For data extents it is the inode number of the owning file.
2086 * For metadata extents this parameter holds the level in the
2087 * tree of the extent.
2089 * @offset: For metadata extents the offset is ignored and is currently
2090 * always passed as 0. For data extents it is the fileoffset
2091 * this extent belongs to.
2093 * @refs_to_add Number of references to add
2095 * @extent_op Pointer to a structure, holding information necessary when
2096 * updating a tree block's flags
2099 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_delayed_ref_node *node,
2101 u64 parent, u64 root_objectid,
2102 u64 owner, u64 offset, int refs_to_add,
2103 struct btrfs_delayed_extent_op *extent_op)
2105 struct btrfs_path *path;
2106 struct extent_buffer *leaf;
2107 struct btrfs_extent_item *item;
2108 struct btrfs_key key;
2109 u64 bytenr = node->bytenr;
2110 u64 num_bytes = node->num_bytes;
2114 path = btrfs_alloc_path();
2118 path->reada = READA_FORWARD;
2119 path->leave_spinning = 1;
2120 /* this will setup the path even if it fails to insert the back ref */
2121 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2122 parent, root_objectid, owner,
2123 offset, refs_to_add, extent_op);
2124 if ((ret < 0 && ret != -EAGAIN) || !ret)
2128 * Ok we had -EAGAIN which means we didn't have space to insert and
2129 * inline extent ref, so just update the reference count and add a
2132 leaf = path->nodes[0];
2133 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2134 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2135 refs = btrfs_extent_refs(leaf, item);
2136 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2138 __run_delayed_extent_op(extent_op, leaf, item);
2140 btrfs_mark_buffer_dirty(leaf);
2141 btrfs_release_path(path);
2143 path->reada = READA_FORWARD;
2144 path->leave_spinning = 1;
2145 /* now insert the actual backref */
2146 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2147 owner, offset, refs_to_add);
2149 btrfs_abort_transaction(trans, ret);
2151 btrfs_free_path(path);
2155 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2156 struct btrfs_delayed_ref_node *node,
2157 struct btrfs_delayed_extent_op *extent_op,
2158 int insert_reserved)
2161 struct btrfs_delayed_data_ref *ref;
2162 struct btrfs_key ins;
2167 ins.objectid = node->bytenr;
2168 ins.offset = node->num_bytes;
2169 ins.type = BTRFS_EXTENT_ITEM_KEY;
2171 ref = btrfs_delayed_node_to_data_ref(node);
2172 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2174 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2175 parent = ref->parent;
2176 ref_root = ref->root;
2178 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2180 flags |= extent_op->flags_to_set;
2181 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2182 flags, ref->objectid,
2185 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2186 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2187 ref->objectid, ref->offset,
2188 node->ref_mod, extent_op);
2189 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2190 ret = __btrfs_free_extent(trans, node, parent,
2191 ref_root, ref->objectid,
2192 ref->offset, node->ref_mod,
2200 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2201 struct extent_buffer *leaf,
2202 struct btrfs_extent_item *ei)
2204 u64 flags = btrfs_extent_flags(leaf, ei);
2205 if (extent_op->update_flags) {
2206 flags |= extent_op->flags_to_set;
2207 btrfs_set_extent_flags(leaf, ei, flags);
2210 if (extent_op->update_key) {
2211 struct btrfs_tree_block_info *bi;
2212 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2213 bi = (struct btrfs_tree_block_info *)(ei + 1);
2214 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2218 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2219 struct btrfs_delayed_ref_head *head,
2220 struct btrfs_delayed_extent_op *extent_op)
2222 struct btrfs_fs_info *fs_info = trans->fs_info;
2223 struct btrfs_key key;
2224 struct btrfs_path *path;
2225 struct btrfs_extent_item *ei;
2226 struct extent_buffer *leaf;
2230 int metadata = !extent_op->is_data;
2235 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2238 path = btrfs_alloc_path();
2242 key.objectid = head->bytenr;
2245 key.type = BTRFS_METADATA_ITEM_KEY;
2246 key.offset = extent_op->level;
2248 key.type = BTRFS_EXTENT_ITEM_KEY;
2249 key.offset = head->num_bytes;
2253 path->reada = READA_FORWARD;
2254 path->leave_spinning = 1;
2255 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2262 if (path->slots[0] > 0) {
2264 btrfs_item_key_to_cpu(path->nodes[0], &key,
2266 if (key.objectid == head->bytenr &&
2267 key.type == BTRFS_EXTENT_ITEM_KEY &&
2268 key.offset == head->num_bytes)
2272 btrfs_release_path(path);
2275 key.objectid = head->bytenr;
2276 key.offset = head->num_bytes;
2277 key.type = BTRFS_EXTENT_ITEM_KEY;
2286 leaf = path->nodes[0];
2287 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2289 if (unlikely(item_size < sizeof(*ei))) {
2291 btrfs_print_v0_err(fs_info);
2292 btrfs_abort_transaction(trans, err);
2296 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2297 __run_delayed_extent_op(extent_op, leaf, ei);
2299 btrfs_mark_buffer_dirty(leaf);
2301 btrfs_free_path(path);
2305 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2306 struct btrfs_delayed_ref_node *node,
2307 struct btrfs_delayed_extent_op *extent_op,
2308 int insert_reserved)
2311 struct btrfs_delayed_tree_ref *ref;
2315 ref = btrfs_delayed_node_to_tree_ref(node);
2316 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2318 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2319 parent = ref->parent;
2320 ref_root = ref->root;
2322 if (node->ref_mod != 1) {
2323 btrfs_err(trans->fs_info,
2324 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2325 node->bytenr, node->ref_mod, node->action, ref_root,
2329 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2330 BUG_ON(!extent_op || !extent_op->update_flags);
2331 ret = alloc_reserved_tree_block(trans, node, extent_op);
2332 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2333 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2334 ref->level, 0, 1, extent_op);
2335 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2336 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2337 ref->level, 0, 1, extent_op);
2344 /* helper function to actually process a single delayed ref entry */
2345 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2346 struct btrfs_delayed_ref_node *node,
2347 struct btrfs_delayed_extent_op *extent_op,
2348 int insert_reserved)
2352 if (trans->aborted) {
2353 if (insert_reserved)
2354 btrfs_pin_extent(trans->fs_info, node->bytenr,
2355 node->num_bytes, 1);
2359 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2360 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2361 ret = run_delayed_tree_ref(trans, node, extent_op,
2363 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2364 node->type == BTRFS_SHARED_DATA_REF_KEY)
2365 ret = run_delayed_data_ref(trans, node, extent_op,
2372 static inline struct btrfs_delayed_ref_node *
2373 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2375 struct btrfs_delayed_ref_node *ref;
2377 if (RB_EMPTY_ROOT(&head->ref_tree))
2381 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2382 * This is to prevent a ref count from going down to zero, which deletes
2383 * the extent item from the extent tree, when there still are references
2384 * to add, which would fail because they would not find the extent item.
2386 if (!list_empty(&head->ref_add_list))
2387 return list_first_entry(&head->ref_add_list,
2388 struct btrfs_delayed_ref_node, add_list);
2390 ref = rb_entry(rb_first(&head->ref_tree),
2391 struct btrfs_delayed_ref_node, ref_node);
2392 ASSERT(list_empty(&ref->add_list));
2396 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2397 struct btrfs_delayed_ref_head *head)
2399 spin_lock(&delayed_refs->lock);
2400 head->processing = 0;
2401 delayed_refs->num_heads_ready++;
2402 spin_unlock(&delayed_refs->lock);
2403 btrfs_delayed_ref_unlock(head);
2406 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2407 struct btrfs_delayed_ref_head *head)
2409 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2414 head->extent_op = NULL;
2415 if (head->must_insert_reserved) {
2416 btrfs_free_delayed_extent_op(extent_op);
2419 spin_unlock(&head->lock);
2420 ret = run_delayed_extent_op(trans, head, extent_op);
2421 btrfs_free_delayed_extent_op(extent_op);
2422 return ret ? ret : 1;
2425 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2426 struct btrfs_delayed_ref_head *head)
2429 struct btrfs_fs_info *fs_info = trans->fs_info;
2430 struct btrfs_delayed_ref_root *delayed_refs;
2433 delayed_refs = &trans->transaction->delayed_refs;
2435 ret = cleanup_extent_op(trans, head);
2437 unselect_delayed_ref_head(delayed_refs, head);
2438 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2445 * Need to drop our head ref lock and re-acquire the delayed ref lock
2446 * and then re-check to make sure nobody got added.
2448 spin_unlock(&head->lock);
2449 spin_lock(&delayed_refs->lock);
2450 spin_lock(&head->lock);
2451 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2452 spin_unlock(&head->lock);
2453 spin_unlock(&delayed_refs->lock);
2456 delayed_refs->num_heads--;
2457 rb_erase(&head->href_node, &delayed_refs->href_root);
2458 RB_CLEAR_NODE(&head->href_node);
2459 spin_unlock(&head->lock);
2460 spin_unlock(&delayed_refs->lock);
2461 atomic_dec(&delayed_refs->num_entries);
2463 trace_run_delayed_ref_head(fs_info, head, 0);
2465 if (head->total_ref_mod < 0) {
2466 struct btrfs_space_info *space_info;
2470 flags = BTRFS_BLOCK_GROUP_DATA;
2471 else if (head->is_system)
2472 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2474 flags = BTRFS_BLOCK_GROUP_METADATA;
2475 space_info = __find_space_info(fs_info, flags);
2477 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2479 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2481 if (head->is_data) {
2482 spin_lock(&delayed_refs->lock);
2483 delayed_refs->pending_csums -= head->num_bytes;
2484 spin_unlock(&delayed_refs->lock);
2488 if (head->must_insert_reserved) {
2489 btrfs_pin_extent(fs_info, head->bytenr,
2490 head->num_bytes, 1);
2491 if (head->is_data) {
2492 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2497 /* Also free its reserved qgroup space */
2498 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2499 head->qgroup_reserved);
2500 btrfs_delayed_ref_unlock(head);
2501 btrfs_put_delayed_ref_head(head);
2506 * Returns 0 on success or if called with an already aborted transaction.
2507 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2509 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2512 struct btrfs_fs_info *fs_info = trans->fs_info;
2513 struct btrfs_delayed_ref_root *delayed_refs;
2514 struct btrfs_delayed_ref_node *ref;
2515 struct btrfs_delayed_ref_head *locked_ref = NULL;
2516 struct btrfs_delayed_extent_op *extent_op;
2517 ktime_t start = ktime_get();
2519 unsigned long count = 0;
2520 unsigned long actual_count = 0;
2521 int must_insert_reserved = 0;
2523 delayed_refs = &trans->transaction->delayed_refs;
2529 spin_lock(&delayed_refs->lock);
2530 locked_ref = btrfs_select_ref_head(trans);
2532 spin_unlock(&delayed_refs->lock);
2536 /* grab the lock that says we are going to process
2537 * all the refs for this head */
2538 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2539 spin_unlock(&delayed_refs->lock);
2541 * we may have dropped the spin lock to get the head
2542 * mutex lock, and that might have given someone else
2543 * time to free the head. If that's true, it has been
2544 * removed from our list and we can move on.
2546 if (ret == -EAGAIN) {
2554 * We need to try and merge add/drops of the same ref since we
2555 * can run into issues with relocate dropping the implicit ref
2556 * and then it being added back again before the drop can
2557 * finish. If we merged anything we need to re-loop so we can
2559 * Or we can get node references of the same type that weren't
2560 * merged when created due to bumps in the tree mod seq, and
2561 * we need to merge them to prevent adding an inline extent
2562 * backref before dropping it (triggering a BUG_ON at
2563 * insert_inline_extent_backref()).
2565 spin_lock(&locked_ref->lock);
2566 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2568 ref = select_delayed_ref(locked_ref);
2570 if (ref && ref->seq &&
2571 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2572 spin_unlock(&locked_ref->lock);
2573 unselect_delayed_ref_head(delayed_refs, locked_ref);
2581 * We're done processing refs in this ref_head, clean everything
2582 * up and move on to the next ref_head.
2585 ret = cleanup_ref_head(trans, locked_ref);
2587 /* We dropped our lock, we need to loop. */
2600 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2601 RB_CLEAR_NODE(&ref->ref_node);
2602 if (!list_empty(&ref->add_list))
2603 list_del(&ref->add_list);
2605 * When we play the delayed ref, also correct the ref_mod on
2608 switch (ref->action) {
2609 case BTRFS_ADD_DELAYED_REF:
2610 case BTRFS_ADD_DELAYED_EXTENT:
2611 locked_ref->ref_mod -= ref->ref_mod;
2613 case BTRFS_DROP_DELAYED_REF:
2614 locked_ref->ref_mod += ref->ref_mod;
2619 atomic_dec(&delayed_refs->num_entries);
2622 * Record the must-insert_reserved flag before we drop the spin
2625 must_insert_reserved = locked_ref->must_insert_reserved;
2626 locked_ref->must_insert_reserved = 0;
2628 extent_op = locked_ref->extent_op;
2629 locked_ref->extent_op = NULL;
2630 spin_unlock(&locked_ref->lock);
2632 ret = run_one_delayed_ref(trans, ref, extent_op,
2633 must_insert_reserved);
2635 btrfs_free_delayed_extent_op(extent_op);
2637 unselect_delayed_ref_head(delayed_refs, locked_ref);
2638 btrfs_put_delayed_ref(ref);
2639 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2644 btrfs_put_delayed_ref(ref);
2650 * We don't want to include ref heads since we can have empty ref heads
2651 * and those will drastically skew our runtime down since we just do
2652 * accounting, no actual extent tree updates.
2654 if (actual_count > 0) {
2655 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2659 * We weigh the current average higher than our current runtime
2660 * to avoid large swings in the average.
2662 spin_lock(&delayed_refs->lock);
2663 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2664 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2665 spin_unlock(&delayed_refs->lock);
2670 #ifdef SCRAMBLE_DELAYED_REFS
2672 * Normally delayed refs get processed in ascending bytenr order. This
2673 * correlates in most cases to the order added. To expose dependencies on this
2674 * order, we start to process the tree in the middle instead of the beginning
2676 static u64 find_middle(struct rb_root *root)
2678 struct rb_node *n = root->rb_node;
2679 struct btrfs_delayed_ref_node *entry;
2682 u64 first = 0, last = 0;
2686 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2687 first = entry->bytenr;
2691 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2692 last = entry->bytenr;
2697 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2698 WARN_ON(!entry->in_tree);
2700 middle = entry->bytenr;
2713 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2717 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2718 sizeof(struct btrfs_extent_inline_ref));
2719 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2720 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2723 * We don't ever fill up leaves all the way so multiply by 2 just to be
2724 * closer to what we're really going to want to use.
2726 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2730 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2731 * would require to store the csums for that many bytes.
2733 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2736 u64 num_csums_per_leaf;
2739 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2740 num_csums_per_leaf = div64_u64(csum_size,
2741 (u64)btrfs_super_csum_size(fs_info->super_copy));
2742 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2743 num_csums += num_csums_per_leaf - 1;
2744 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2748 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2749 struct btrfs_fs_info *fs_info)
2751 struct btrfs_block_rsv *global_rsv;
2752 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2753 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2754 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2755 u64 num_bytes, num_dirty_bgs_bytes;
2758 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2759 num_heads = heads_to_leaves(fs_info, num_heads);
2761 num_bytes += (num_heads - 1) * fs_info->nodesize;
2763 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2765 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2767 global_rsv = &fs_info->global_block_rsv;
2770 * If we can't allocate any more chunks lets make sure we have _lots_ of
2771 * wiggle room since running delayed refs can create more delayed refs.
2773 if (global_rsv->space_info->full) {
2774 num_dirty_bgs_bytes <<= 1;
2778 spin_lock(&global_rsv->lock);
2779 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2781 spin_unlock(&global_rsv->lock);
2785 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2786 struct btrfs_fs_info *fs_info)
2789 atomic_read(&trans->transaction->delayed_refs.num_entries);
2794 avg_runtime = fs_info->avg_delayed_ref_runtime;
2795 val = num_entries * avg_runtime;
2796 if (val >= NSEC_PER_SEC)
2798 if (val >= NSEC_PER_SEC / 2)
2801 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2804 struct async_delayed_refs {
2805 struct btrfs_root *root;
2810 struct completion wait;
2811 struct btrfs_work work;
2814 static inline struct async_delayed_refs *
2815 to_async_delayed_refs(struct btrfs_work *work)
2817 return container_of(work, struct async_delayed_refs, work);
2820 static void delayed_ref_async_start(struct btrfs_work *work)
2822 struct async_delayed_refs *async = to_async_delayed_refs(work);
2823 struct btrfs_trans_handle *trans;
2824 struct btrfs_fs_info *fs_info = async->root->fs_info;
2827 /* if the commit is already started, we don't need to wait here */
2828 if (btrfs_transaction_blocked(fs_info))
2831 trans = btrfs_join_transaction(async->root);
2832 if (IS_ERR(trans)) {
2833 async->error = PTR_ERR(trans);
2838 * trans->sync means that when we call end_transaction, we won't
2839 * wait on delayed refs
2843 /* Don't bother flushing if we got into a different transaction */
2844 if (trans->transid > async->transid)
2847 ret = btrfs_run_delayed_refs(trans, async->count);
2851 ret = btrfs_end_transaction(trans);
2852 if (ret && !async->error)
2856 complete(&async->wait);
2861 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2862 unsigned long count, u64 transid, int wait)
2864 struct async_delayed_refs *async;
2867 async = kmalloc(sizeof(*async), GFP_NOFS);
2871 async->root = fs_info->tree_root;
2872 async->count = count;
2874 async->transid = transid;
2879 init_completion(&async->wait);
2881 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2882 delayed_ref_async_start, NULL, NULL);
2884 btrfs_queue_work(fs_info->extent_workers, &async->work);
2887 wait_for_completion(&async->wait);
2896 * this starts processing the delayed reference count updates and
2897 * extent insertions we have queued up so far. count can be
2898 * 0, which means to process everything in the tree at the start
2899 * of the run (but not newly added entries), or it can be some target
2900 * number you'd like to process.
2902 * Returns 0 on success or if called with an aborted transaction
2903 * Returns <0 on error and aborts the transaction
2905 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2906 unsigned long count)
2908 struct btrfs_fs_info *fs_info = trans->fs_info;
2909 struct rb_node *node;
2910 struct btrfs_delayed_ref_root *delayed_refs;
2911 struct btrfs_delayed_ref_head *head;
2913 int run_all = count == (unsigned long)-1;
2914 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2916 /* We'll clean this up in btrfs_cleanup_transaction */
2920 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2923 delayed_refs = &trans->transaction->delayed_refs;
2925 count = atomic_read(&delayed_refs->num_entries) * 2;
2928 #ifdef SCRAMBLE_DELAYED_REFS
2929 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2931 trans->can_flush_pending_bgs = false;
2932 ret = __btrfs_run_delayed_refs(trans, count);
2934 btrfs_abort_transaction(trans, ret);
2939 if (!list_empty(&trans->new_bgs))
2940 btrfs_create_pending_block_groups(trans);
2942 spin_lock(&delayed_refs->lock);
2943 node = rb_first(&delayed_refs->href_root);
2945 spin_unlock(&delayed_refs->lock);
2948 head = rb_entry(node, struct btrfs_delayed_ref_head,
2950 refcount_inc(&head->refs);
2951 spin_unlock(&delayed_refs->lock);
2953 /* Mutex was contended, block until it's released and retry. */
2954 mutex_lock(&head->mutex);
2955 mutex_unlock(&head->mutex);
2957 btrfs_put_delayed_ref_head(head);
2962 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2966 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2967 struct btrfs_fs_info *fs_info,
2968 u64 bytenr, u64 num_bytes, u64 flags,
2969 int level, int is_data)
2971 struct btrfs_delayed_extent_op *extent_op;
2974 extent_op = btrfs_alloc_delayed_extent_op();
2978 extent_op->flags_to_set = flags;
2979 extent_op->update_flags = true;
2980 extent_op->update_key = false;
2981 extent_op->is_data = is_data ? true : false;
2982 extent_op->level = level;
2984 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
2985 num_bytes, extent_op);
2987 btrfs_free_delayed_extent_op(extent_op);
2991 static noinline int check_delayed_ref(struct btrfs_root *root,
2992 struct btrfs_path *path,
2993 u64 objectid, u64 offset, u64 bytenr)
2995 struct btrfs_delayed_ref_head *head;
2996 struct btrfs_delayed_ref_node *ref;
2997 struct btrfs_delayed_data_ref *data_ref;
2998 struct btrfs_delayed_ref_root *delayed_refs;
2999 struct btrfs_transaction *cur_trans;
3000 struct rb_node *node;
3003 spin_lock(&root->fs_info->trans_lock);
3004 cur_trans = root->fs_info->running_transaction;
3006 refcount_inc(&cur_trans->use_count);
3007 spin_unlock(&root->fs_info->trans_lock);
3011 delayed_refs = &cur_trans->delayed_refs;
3012 spin_lock(&delayed_refs->lock);
3013 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3015 spin_unlock(&delayed_refs->lock);
3016 btrfs_put_transaction(cur_trans);
3020 if (!mutex_trylock(&head->mutex)) {
3021 refcount_inc(&head->refs);
3022 spin_unlock(&delayed_refs->lock);
3024 btrfs_release_path(path);
3027 * Mutex was contended, block until it's released and let
3030 mutex_lock(&head->mutex);
3031 mutex_unlock(&head->mutex);
3032 btrfs_put_delayed_ref_head(head);
3033 btrfs_put_transaction(cur_trans);
3036 spin_unlock(&delayed_refs->lock);
3038 spin_lock(&head->lock);
3040 * XXX: We should replace this with a proper search function in the
3043 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3044 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3045 /* If it's a shared ref we know a cross reference exists */
3046 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3051 data_ref = btrfs_delayed_node_to_data_ref(ref);
3054 * If our ref doesn't match the one we're currently looking at
3055 * then we have a cross reference.
3057 if (data_ref->root != root->root_key.objectid ||
3058 data_ref->objectid != objectid ||
3059 data_ref->offset != offset) {
3064 spin_unlock(&head->lock);
3065 mutex_unlock(&head->mutex);
3066 btrfs_put_transaction(cur_trans);
3070 static noinline int check_committed_ref(struct btrfs_root *root,
3071 struct btrfs_path *path,
3072 u64 objectid, u64 offset, u64 bytenr)
3074 struct btrfs_fs_info *fs_info = root->fs_info;
3075 struct btrfs_root *extent_root = fs_info->extent_root;
3076 struct extent_buffer *leaf;
3077 struct btrfs_extent_data_ref *ref;
3078 struct btrfs_extent_inline_ref *iref;
3079 struct btrfs_extent_item *ei;
3080 struct btrfs_key key;
3085 key.objectid = bytenr;
3086 key.offset = (u64)-1;
3087 key.type = BTRFS_EXTENT_ITEM_KEY;
3089 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3092 BUG_ON(ret == 0); /* Corruption */
3095 if (path->slots[0] == 0)
3099 leaf = path->nodes[0];
3100 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3102 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3106 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3107 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3109 if (item_size != sizeof(*ei) +
3110 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3113 if (btrfs_extent_generation(leaf, ei) <=
3114 btrfs_root_last_snapshot(&root->root_item))
3117 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3119 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3120 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3123 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3124 if (btrfs_extent_refs(leaf, ei) !=
3125 btrfs_extent_data_ref_count(leaf, ref) ||
3126 btrfs_extent_data_ref_root(leaf, ref) !=
3127 root->root_key.objectid ||
3128 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3129 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3137 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3140 struct btrfs_path *path;
3144 path = btrfs_alloc_path();
3149 ret = check_committed_ref(root, path, objectid,
3151 if (ret && ret != -ENOENT)
3154 ret2 = check_delayed_ref(root, path, objectid,
3156 } while (ret2 == -EAGAIN);
3158 if (ret2 && ret2 != -ENOENT) {
3163 if (ret != -ENOENT || ret2 != -ENOENT)
3166 btrfs_free_path(path);
3167 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3172 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 struct extent_buffer *buf,
3175 int full_backref, int inc)
3177 struct btrfs_fs_info *fs_info = root->fs_info;
3183 struct btrfs_key key;
3184 struct btrfs_file_extent_item *fi;
3188 int (*process_func)(struct btrfs_trans_handle *,
3189 struct btrfs_root *,
3190 u64, u64, u64, u64, u64, u64);
3193 if (btrfs_is_testing(fs_info))
3196 ref_root = btrfs_header_owner(buf);
3197 nritems = btrfs_header_nritems(buf);
3198 level = btrfs_header_level(buf);
3200 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3204 process_func = btrfs_inc_extent_ref;
3206 process_func = btrfs_free_extent;
3209 parent = buf->start;
3213 for (i = 0; i < nritems; i++) {
3215 btrfs_item_key_to_cpu(buf, &key, i);
3216 if (key.type != BTRFS_EXTENT_DATA_KEY)
3218 fi = btrfs_item_ptr(buf, i,
3219 struct btrfs_file_extent_item);
3220 if (btrfs_file_extent_type(buf, fi) ==
3221 BTRFS_FILE_EXTENT_INLINE)
3223 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3227 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3228 key.offset -= btrfs_file_extent_offset(buf, fi);
3229 ret = process_func(trans, root, bytenr, num_bytes,
3230 parent, ref_root, key.objectid,
3235 bytenr = btrfs_node_blockptr(buf, i);
3236 num_bytes = fs_info->nodesize;
3237 ret = process_func(trans, root, bytenr, num_bytes,
3238 parent, ref_root, level - 1, 0);
3248 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3249 struct extent_buffer *buf, int full_backref)
3251 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3254 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3255 struct extent_buffer *buf, int full_backref)
3257 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3260 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3261 struct btrfs_fs_info *fs_info,
3262 struct btrfs_path *path,
3263 struct btrfs_block_group_cache *cache)
3266 struct btrfs_root *extent_root = fs_info->extent_root;
3268 struct extent_buffer *leaf;
3270 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3277 leaf = path->nodes[0];
3278 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3279 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3280 btrfs_mark_buffer_dirty(leaf);
3282 btrfs_release_path(path);
3287 static struct btrfs_block_group_cache *
3288 next_block_group(struct btrfs_fs_info *fs_info,
3289 struct btrfs_block_group_cache *cache)
3291 struct rb_node *node;
3293 spin_lock(&fs_info->block_group_cache_lock);
3295 /* If our block group was removed, we need a full search. */
3296 if (RB_EMPTY_NODE(&cache->cache_node)) {
3297 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3299 spin_unlock(&fs_info->block_group_cache_lock);
3300 btrfs_put_block_group(cache);
3301 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3303 node = rb_next(&cache->cache_node);
3304 btrfs_put_block_group(cache);
3306 cache = rb_entry(node, struct btrfs_block_group_cache,
3308 btrfs_get_block_group(cache);
3311 spin_unlock(&fs_info->block_group_cache_lock);
3315 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3316 struct btrfs_trans_handle *trans,
3317 struct btrfs_path *path)
3319 struct btrfs_fs_info *fs_info = block_group->fs_info;
3320 struct btrfs_root *root = fs_info->tree_root;
3321 struct inode *inode = NULL;
3322 struct extent_changeset *data_reserved = NULL;
3324 int dcs = BTRFS_DC_ERROR;
3330 * If this block group is smaller than 100 megs don't bother caching the
3333 if (block_group->key.offset < (100 * SZ_1M)) {
3334 spin_lock(&block_group->lock);
3335 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3336 spin_unlock(&block_group->lock);
3343 inode = lookup_free_space_inode(fs_info, block_group, path);
3344 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3345 ret = PTR_ERR(inode);
3346 btrfs_release_path(path);
3350 if (IS_ERR(inode)) {
3354 if (block_group->ro)
3357 ret = create_free_space_inode(fs_info, trans, block_group,
3365 * We want to set the generation to 0, that way if anything goes wrong
3366 * from here on out we know not to trust this cache when we load up next
3369 BTRFS_I(inode)->generation = 0;
3370 ret = btrfs_update_inode(trans, root, inode);
3373 * So theoretically we could recover from this, simply set the
3374 * super cache generation to 0 so we know to invalidate the
3375 * cache, but then we'd have to keep track of the block groups
3376 * that fail this way so we know we _have_ to reset this cache
3377 * before the next commit or risk reading stale cache. So to
3378 * limit our exposure to horrible edge cases lets just abort the
3379 * transaction, this only happens in really bad situations
3382 btrfs_abort_transaction(trans, ret);
3387 /* We've already setup this transaction, go ahead and exit */
3388 if (block_group->cache_generation == trans->transid &&
3389 i_size_read(inode)) {
3390 dcs = BTRFS_DC_SETUP;
3394 if (i_size_read(inode) > 0) {
3395 ret = btrfs_check_trunc_cache_free_space(fs_info,
3396 &fs_info->global_block_rsv);
3400 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3405 spin_lock(&block_group->lock);
3406 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3407 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3409 * don't bother trying to write stuff out _if_
3410 * a) we're not cached,
3411 * b) we're with nospace_cache mount option,
3412 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3414 dcs = BTRFS_DC_WRITTEN;
3415 spin_unlock(&block_group->lock);
3418 spin_unlock(&block_group->lock);
3421 * We hit an ENOSPC when setting up the cache in this transaction, just
3422 * skip doing the setup, we've already cleared the cache so we're safe.
3424 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3430 * Try to preallocate enough space based on how big the block group is.
3431 * Keep in mind this has to include any pinned space which could end up
3432 * taking up quite a bit since it's not folded into the other space
3435 num_pages = div_u64(block_group->key.offset, SZ_256M);
3440 num_pages *= PAGE_SIZE;
3442 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3446 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3447 num_pages, num_pages,
3450 * Our cache requires contiguous chunks so that we don't modify a bunch
3451 * of metadata or split extents when writing the cache out, which means
3452 * we can enospc if we are heavily fragmented in addition to just normal
3453 * out of space conditions. So if we hit this just skip setting up any
3454 * other block groups for this transaction, maybe we'll unpin enough
3455 * space the next time around.
3458 dcs = BTRFS_DC_SETUP;
3459 else if (ret == -ENOSPC)
3460 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3465 btrfs_release_path(path);
3467 spin_lock(&block_group->lock);
3468 if (!ret && dcs == BTRFS_DC_SETUP)
3469 block_group->cache_generation = trans->transid;
3470 block_group->disk_cache_state = dcs;
3471 spin_unlock(&block_group->lock);
3473 extent_changeset_free(data_reserved);
3477 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3478 struct btrfs_fs_info *fs_info)
3480 struct btrfs_block_group_cache *cache, *tmp;
3481 struct btrfs_transaction *cur_trans = trans->transaction;
3482 struct btrfs_path *path;
3484 if (list_empty(&cur_trans->dirty_bgs) ||
3485 !btrfs_test_opt(fs_info, SPACE_CACHE))
3488 path = btrfs_alloc_path();
3492 /* Could add new block groups, use _safe just in case */
3493 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3495 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3496 cache_save_setup(cache, trans, path);
3499 btrfs_free_path(path);
3504 * transaction commit does final block group cache writeback during a
3505 * critical section where nothing is allowed to change the FS. This is
3506 * required in order for the cache to actually match the block group,
3507 * but can introduce a lot of latency into the commit.
3509 * So, btrfs_start_dirty_block_groups is here to kick off block group
3510 * cache IO. There's a chance we'll have to redo some of it if the
3511 * block group changes again during the commit, but it greatly reduces
3512 * the commit latency by getting rid of the easy block groups while
3513 * we're still allowing others to join the commit.
3515 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3517 struct btrfs_fs_info *fs_info = trans->fs_info;
3518 struct btrfs_block_group_cache *cache;
3519 struct btrfs_transaction *cur_trans = trans->transaction;
3522 struct btrfs_path *path = NULL;
3524 struct list_head *io = &cur_trans->io_bgs;
3525 int num_started = 0;
3528 spin_lock(&cur_trans->dirty_bgs_lock);
3529 if (list_empty(&cur_trans->dirty_bgs)) {
3530 spin_unlock(&cur_trans->dirty_bgs_lock);
3533 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3534 spin_unlock(&cur_trans->dirty_bgs_lock);
3538 * make sure all the block groups on our dirty list actually
3541 btrfs_create_pending_block_groups(trans);
3544 path = btrfs_alloc_path();
3550 * cache_write_mutex is here only to save us from balance or automatic
3551 * removal of empty block groups deleting this block group while we are
3552 * writing out the cache
3554 mutex_lock(&trans->transaction->cache_write_mutex);
3555 while (!list_empty(&dirty)) {
3556 cache = list_first_entry(&dirty,
3557 struct btrfs_block_group_cache,
3560 * this can happen if something re-dirties a block
3561 * group that is already under IO. Just wait for it to
3562 * finish and then do it all again
3564 if (!list_empty(&cache->io_list)) {
3565 list_del_init(&cache->io_list);
3566 btrfs_wait_cache_io(trans, cache, path);
3567 btrfs_put_block_group(cache);
3572 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3573 * if it should update the cache_state. Don't delete
3574 * until after we wait.
3576 * Since we're not running in the commit critical section
3577 * we need the dirty_bgs_lock to protect from update_block_group
3579 spin_lock(&cur_trans->dirty_bgs_lock);
3580 list_del_init(&cache->dirty_list);
3581 spin_unlock(&cur_trans->dirty_bgs_lock);
3585 cache_save_setup(cache, trans, path);
3587 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3588 cache->io_ctl.inode = NULL;
3589 ret = btrfs_write_out_cache(fs_info, trans,
3591 if (ret == 0 && cache->io_ctl.inode) {
3596 * The cache_write_mutex is protecting the
3597 * io_list, also refer to the definition of
3598 * btrfs_transaction::io_bgs for more details
3600 list_add_tail(&cache->io_list, io);
3603 * if we failed to write the cache, the
3604 * generation will be bad and life goes on
3610 ret = write_one_cache_group(trans, fs_info,
3613 * Our block group might still be attached to the list
3614 * of new block groups in the transaction handle of some
3615 * other task (struct btrfs_trans_handle->new_bgs). This
3616 * means its block group item isn't yet in the extent
3617 * tree. If this happens ignore the error, as we will
3618 * try again later in the critical section of the
3619 * transaction commit.
3621 if (ret == -ENOENT) {
3623 spin_lock(&cur_trans->dirty_bgs_lock);
3624 if (list_empty(&cache->dirty_list)) {
3625 list_add_tail(&cache->dirty_list,
3626 &cur_trans->dirty_bgs);
3627 btrfs_get_block_group(cache);
3629 spin_unlock(&cur_trans->dirty_bgs_lock);
3631 btrfs_abort_transaction(trans, ret);
3635 /* if its not on the io list, we need to put the block group */
3637 btrfs_put_block_group(cache);
3643 * Avoid blocking other tasks for too long. It might even save
3644 * us from writing caches for block groups that are going to be
3647 mutex_unlock(&trans->transaction->cache_write_mutex);
3648 mutex_lock(&trans->transaction->cache_write_mutex);
3650 mutex_unlock(&trans->transaction->cache_write_mutex);
3653 * go through delayed refs for all the stuff we've just kicked off
3654 * and then loop back (just once)
3656 ret = btrfs_run_delayed_refs(trans, 0);
3657 if (!ret && loops == 0) {
3659 spin_lock(&cur_trans->dirty_bgs_lock);
3660 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3662 * dirty_bgs_lock protects us from concurrent block group
3663 * deletes too (not just cache_write_mutex).
3665 if (!list_empty(&dirty)) {
3666 spin_unlock(&cur_trans->dirty_bgs_lock);
3669 spin_unlock(&cur_trans->dirty_bgs_lock);
3670 } else if (ret < 0) {
3671 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3674 btrfs_free_path(path);
3678 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3679 struct btrfs_fs_info *fs_info)
3681 struct btrfs_block_group_cache *cache;
3682 struct btrfs_transaction *cur_trans = trans->transaction;
3685 struct btrfs_path *path;
3686 struct list_head *io = &cur_trans->io_bgs;
3687 int num_started = 0;
3689 path = btrfs_alloc_path();
3694 * Even though we are in the critical section of the transaction commit,
3695 * we can still have concurrent tasks adding elements to this
3696 * transaction's list of dirty block groups. These tasks correspond to
3697 * endio free space workers started when writeback finishes for a
3698 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3699 * allocate new block groups as a result of COWing nodes of the root
3700 * tree when updating the free space inode. The writeback for the space
3701 * caches is triggered by an earlier call to
3702 * btrfs_start_dirty_block_groups() and iterations of the following
3704 * Also we want to do the cache_save_setup first and then run the
3705 * delayed refs to make sure we have the best chance at doing this all
3708 spin_lock(&cur_trans->dirty_bgs_lock);
3709 while (!list_empty(&cur_trans->dirty_bgs)) {
3710 cache = list_first_entry(&cur_trans->dirty_bgs,
3711 struct btrfs_block_group_cache,
3715 * this can happen if cache_save_setup re-dirties a block
3716 * group that is already under IO. Just wait for it to
3717 * finish and then do it all again
3719 if (!list_empty(&cache->io_list)) {
3720 spin_unlock(&cur_trans->dirty_bgs_lock);
3721 list_del_init(&cache->io_list);
3722 btrfs_wait_cache_io(trans, cache, path);
3723 btrfs_put_block_group(cache);
3724 spin_lock(&cur_trans->dirty_bgs_lock);
3728 * don't remove from the dirty list until after we've waited
3731 list_del_init(&cache->dirty_list);
3732 spin_unlock(&cur_trans->dirty_bgs_lock);
3735 cache_save_setup(cache, trans, path);
3738 ret = btrfs_run_delayed_refs(trans,
3739 (unsigned long) -1);
3741 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3742 cache->io_ctl.inode = NULL;
3743 ret = btrfs_write_out_cache(fs_info, trans,
3745 if (ret == 0 && cache->io_ctl.inode) {
3748 list_add_tail(&cache->io_list, io);
3751 * if we failed to write the cache, the
3752 * generation will be bad and life goes on
3758 ret = write_one_cache_group(trans, fs_info,
3761 * One of the free space endio workers might have
3762 * created a new block group while updating a free space
3763 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3764 * and hasn't released its transaction handle yet, in
3765 * which case the new block group is still attached to
3766 * its transaction handle and its creation has not
3767 * finished yet (no block group item in the extent tree
3768 * yet, etc). If this is the case, wait for all free
3769 * space endio workers to finish and retry. This is a
3770 * a very rare case so no need for a more efficient and
3773 if (ret == -ENOENT) {
3774 wait_event(cur_trans->writer_wait,
3775 atomic_read(&cur_trans->num_writers) == 1);
3776 ret = write_one_cache_group(trans, fs_info,
3780 btrfs_abort_transaction(trans, ret);
3783 /* if its not on the io list, we need to put the block group */
3785 btrfs_put_block_group(cache);
3786 spin_lock(&cur_trans->dirty_bgs_lock);
3788 spin_unlock(&cur_trans->dirty_bgs_lock);
3791 * Refer to the definition of io_bgs member for details why it's safe
3792 * to use it without any locking
3794 while (!list_empty(io)) {
3795 cache = list_first_entry(io, struct btrfs_block_group_cache,
3797 list_del_init(&cache->io_list);
3798 btrfs_wait_cache_io(trans, cache, path);
3799 btrfs_put_block_group(cache);
3802 btrfs_free_path(path);
3806 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3808 struct btrfs_block_group_cache *block_group;
3811 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3812 if (!block_group || block_group->ro)
3815 btrfs_put_block_group(block_group);
3819 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3821 struct btrfs_block_group_cache *bg;
3824 bg = btrfs_lookup_block_group(fs_info, bytenr);
3828 spin_lock(&bg->lock);
3832 atomic_inc(&bg->nocow_writers);
3833 spin_unlock(&bg->lock);
3835 /* no put on block group, done by btrfs_dec_nocow_writers */
3837 btrfs_put_block_group(bg);
3843 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3845 struct btrfs_block_group_cache *bg;
3847 bg = btrfs_lookup_block_group(fs_info, bytenr);
3849 if (atomic_dec_and_test(&bg->nocow_writers))
3850 wake_up_var(&bg->nocow_writers);
3852 * Once for our lookup and once for the lookup done by a previous call
3853 * to btrfs_inc_nocow_writers()
3855 btrfs_put_block_group(bg);
3856 btrfs_put_block_group(bg);
3859 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3861 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3864 static const char *alloc_name(u64 flags)
3867 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3869 case BTRFS_BLOCK_GROUP_METADATA:
3871 case BTRFS_BLOCK_GROUP_DATA:
3873 case BTRFS_BLOCK_GROUP_SYSTEM:
3877 return "invalid-combination";
3881 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3884 struct btrfs_space_info *space_info;
3888 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3892 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3899 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3900 INIT_LIST_HEAD(&space_info->block_groups[i]);
3901 init_rwsem(&space_info->groups_sem);
3902 spin_lock_init(&space_info->lock);
3903 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3904 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3905 init_waitqueue_head(&space_info->wait);
3906 INIT_LIST_HEAD(&space_info->ro_bgs);
3907 INIT_LIST_HEAD(&space_info->tickets);
3908 INIT_LIST_HEAD(&space_info->priority_tickets);
3910 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3911 info->space_info_kobj, "%s",
3912 alloc_name(space_info->flags));
3914 percpu_counter_destroy(&space_info->total_bytes_pinned);
3919 list_add_rcu(&space_info->list, &info->space_info);
3920 if (flags & BTRFS_BLOCK_GROUP_DATA)
3921 info->data_sinfo = space_info;
3926 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3927 u64 total_bytes, u64 bytes_used,
3929 struct btrfs_space_info **space_info)
3931 struct btrfs_space_info *found;
3934 factor = btrfs_bg_type_to_factor(flags);
3936 found = __find_space_info(info, flags);
3938 spin_lock(&found->lock);
3939 found->total_bytes += total_bytes;
3940 found->disk_total += total_bytes * factor;
3941 found->bytes_used += bytes_used;
3942 found->disk_used += bytes_used * factor;
3943 found->bytes_readonly += bytes_readonly;
3944 if (total_bytes > 0)
3946 space_info_add_new_bytes(info, found, total_bytes -
3947 bytes_used - bytes_readonly);
3948 spin_unlock(&found->lock);
3949 *space_info = found;
3952 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3954 u64 extra_flags = chunk_to_extended(flags) &
3955 BTRFS_EXTENDED_PROFILE_MASK;
3957 write_seqlock(&fs_info->profiles_lock);
3958 if (flags & BTRFS_BLOCK_GROUP_DATA)
3959 fs_info->avail_data_alloc_bits |= extra_flags;
3960 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3961 fs_info->avail_metadata_alloc_bits |= extra_flags;
3962 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3963 fs_info->avail_system_alloc_bits |= extra_flags;
3964 write_sequnlock(&fs_info->profiles_lock);
3968 * returns target flags in extended format or 0 if restripe for this
3969 * chunk_type is not in progress
3971 * should be called with balance_lock held
3973 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3975 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3981 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3982 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3983 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3984 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3985 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3986 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3987 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3988 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3989 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3996 * @flags: available profiles in extended format (see ctree.h)
3998 * Returns reduced profile in chunk format. If profile changing is in
3999 * progress (either running or paused) picks the target profile (if it's
4000 * already available), otherwise falls back to plain reducing.
4002 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4004 u64 num_devices = fs_info->fs_devices->rw_devices;
4010 * see if restripe for this chunk_type is in progress, if so
4011 * try to reduce to the target profile
4013 spin_lock(&fs_info->balance_lock);
4014 target = get_restripe_target(fs_info, flags);
4016 /* pick target profile only if it's already available */
4017 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4018 spin_unlock(&fs_info->balance_lock);
4019 return extended_to_chunk(target);
4022 spin_unlock(&fs_info->balance_lock);
4024 /* First, mask out the RAID levels which aren't possible */
4025 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4026 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4027 allowed |= btrfs_raid_array[raid_type].bg_flag;
4031 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4032 allowed = BTRFS_BLOCK_GROUP_RAID6;
4033 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4034 allowed = BTRFS_BLOCK_GROUP_RAID5;
4035 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4036 allowed = BTRFS_BLOCK_GROUP_RAID10;
4037 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4038 allowed = BTRFS_BLOCK_GROUP_RAID1;
4039 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4040 allowed = BTRFS_BLOCK_GROUP_RAID0;
4042 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4044 return extended_to_chunk(flags | allowed);
4047 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4054 seq = read_seqbegin(&fs_info->profiles_lock);
4056 if (flags & BTRFS_BLOCK_GROUP_DATA)
4057 flags |= fs_info->avail_data_alloc_bits;
4058 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4059 flags |= fs_info->avail_system_alloc_bits;
4060 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4061 flags |= fs_info->avail_metadata_alloc_bits;
4062 } while (read_seqretry(&fs_info->profiles_lock, seq));
4064 return btrfs_reduce_alloc_profile(fs_info, flags);
4067 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4069 struct btrfs_fs_info *fs_info = root->fs_info;
4074 flags = BTRFS_BLOCK_GROUP_DATA;
4075 else if (root == fs_info->chunk_root)
4076 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4078 flags = BTRFS_BLOCK_GROUP_METADATA;
4080 ret = get_alloc_profile(fs_info, flags);
4084 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4086 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4089 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4091 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4094 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4096 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4099 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4100 bool may_use_included)
4103 return s_info->bytes_used + s_info->bytes_reserved +
4104 s_info->bytes_pinned + s_info->bytes_readonly +
4105 (may_use_included ? s_info->bytes_may_use : 0);
4108 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4110 struct btrfs_root *root = inode->root;
4111 struct btrfs_fs_info *fs_info = root->fs_info;
4112 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4115 int need_commit = 2;
4116 int have_pinned_space;
4118 /* make sure bytes are sectorsize aligned */
4119 bytes = ALIGN(bytes, fs_info->sectorsize);
4121 if (btrfs_is_free_space_inode(inode)) {
4123 ASSERT(current->journal_info);
4127 /* make sure we have enough space to handle the data first */
4128 spin_lock(&data_sinfo->lock);
4129 used = btrfs_space_info_used(data_sinfo, true);
4131 if (used + bytes > data_sinfo->total_bytes) {
4132 struct btrfs_trans_handle *trans;
4135 * if we don't have enough free bytes in this space then we need
4136 * to alloc a new chunk.
4138 if (!data_sinfo->full) {
4141 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4142 spin_unlock(&data_sinfo->lock);
4144 alloc_target = btrfs_data_alloc_profile(fs_info);
4146 * It is ugly that we don't call nolock join
4147 * transaction for the free space inode case here.
4148 * But it is safe because we only do the data space
4149 * reservation for the free space cache in the
4150 * transaction context, the common join transaction
4151 * just increase the counter of the current transaction
4152 * handler, doesn't try to acquire the trans_lock of
4155 trans = btrfs_join_transaction(root);
4157 return PTR_ERR(trans);
4159 ret = do_chunk_alloc(trans, alloc_target,
4160 CHUNK_ALLOC_NO_FORCE);
4161 btrfs_end_transaction(trans);
4166 have_pinned_space = 1;
4175 * If we don't have enough pinned space to deal with this
4176 * allocation, and no removed chunk in current transaction,
4177 * don't bother committing the transaction.
4179 have_pinned_space = __percpu_counter_compare(
4180 &data_sinfo->total_bytes_pinned,
4181 used + bytes - data_sinfo->total_bytes,
4182 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4183 spin_unlock(&data_sinfo->lock);
4185 /* commit the current transaction and try again */
4190 if (need_commit > 0) {
4191 btrfs_start_delalloc_roots(fs_info, -1);
4192 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4196 trans = btrfs_join_transaction(root);
4198 return PTR_ERR(trans);
4199 if (have_pinned_space >= 0 ||
4200 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4201 &trans->transaction->flags) ||
4203 ret = btrfs_commit_transaction(trans);
4207 * The cleaner kthread might still be doing iput
4208 * operations. Wait for it to finish so that
4209 * more space is released.
4211 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4212 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4215 btrfs_end_transaction(trans);
4219 trace_btrfs_space_reservation(fs_info,
4220 "space_info:enospc",
4221 data_sinfo->flags, bytes, 1);
4224 data_sinfo->bytes_may_use += bytes;
4225 trace_btrfs_space_reservation(fs_info, "space_info",
4226 data_sinfo->flags, bytes, 1);
4227 spin_unlock(&data_sinfo->lock);
4232 int btrfs_check_data_free_space(struct inode *inode,
4233 struct extent_changeset **reserved, u64 start, u64 len)
4235 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4238 /* align the range */
4239 len = round_up(start + len, fs_info->sectorsize) -
4240 round_down(start, fs_info->sectorsize);
4241 start = round_down(start, fs_info->sectorsize);
4243 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4247 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4248 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4250 btrfs_free_reserved_data_space_noquota(inode, start, len);
4257 * Called if we need to clear a data reservation for this inode
4258 * Normally in a error case.
4260 * This one will *NOT* use accurate qgroup reserved space API, just for case
4261 * which we can't sleep and is sure it won't affect qgroup reserved space.
4262 * Like clear_bit_hook().
4264 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4267 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4268 struct btrfs_space_info *data_sinfo;
4270 /* Make sure the range is aligned to sectorsize */
4271 len = round_up(start + len, fs_info->sectorsize) -
4272 round_down(start, fs_info->sectorsize);
4273 start = round_down(start, fs_info->sectorsize);
4275 data_sinfo = fs_info->data_sinfo;
4276 spin_lock(&data_sinfo->lock);
4277 if (WARN_ON(data_sinfo->bytes_may_use < len))
4278 data_sinfo->bytes_may_use = 0;
4280 data_sinfo->bytes_may_use -= len;
4281 trace_btrfs_space_reservation(fs_info, "space_info",
4282 data_sinfo->flags, len, 0);
4283 spin_unlock(&data_sinfo->lock);
4287 * Called if we need to clear a data reservation for this inode
4288 * Normally in a error case.
4290 * This one will handle the per-inode data rsv map for accurate reserved
4293 void btrfs_free_reserved_data_space(struct inode *inode,
4294 struct extent_changeset *reserved, u64 start, u64 len)
4296 struct btrfs_root *root = BTRFS_I(inode)->root;
4298 /* Make sure the range is aligned to sectorsize */
4299 len = round_up(start + len, root->fs_info->sectorsize) -
4300 round_down(start, root->fs_info->sectorsize);
4301 start = round_down(start, root->fs_info->sectorsize);
4303 btrfs_free_reserved_data_space_noquota(inode, start, len);
4304 btrfs_qgroup_free_data(inode, reserved, start, len);
4307 static void force_metadata_allocation(struct btrfs_fs_info *info)
4309 struct list_head *head = &info->space_info;
4310 struct btrfs_space_info *found;
4313 list_for_each_entry_rcu(found, head, list) {
4314 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4315 found->force_alloc = CHUNK_ALLOC_FORCE;
4320 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4322 return (global->size << 1);
4325 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4326 struct btrfs_space_info *sinfo, int force)
4328 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4329 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4332 if (force == CHUNK_ALLOC_FORCE)
4336 * We need to take into account the global rsv because for all intents
4337 * and purposes it's used space. Don't worry about locking the
4338 * global_rsv, it doesn't change except when the transaction commits.
4340 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4341 bytes_used += calc_global_rsv_need_space(global_rsv);
4344 * in limited mode, we want to have some free space up to
4345 * about 1% of the FS size.
4347 if (force == CHUNK_ALLOC_LIMITED) {
4348 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4349 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4351 if (sinfo->total_bytes - bytes_used < thresh)
4355 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4360 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4364 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4365 BTRFS_BLOCK_GROUP_RAID0 |
4366 BTRFS_BLOCK_GROUP_RAID5 |
4367 BTRFS_BLOCK_GROUP_RAID6))
4368 num_dev = fs_info->fs_devices->rw_devices;
4369 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4372 num_dev = 1; /* DUP or single */
4378 * If @is_allocation is true, reserve space in the system space info necessary
4379 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4382 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4384 struct btrfs_fs_info *fs_info = trans->fs_info;
4385 struct btrfs_space_info *info;
4392 * Needed because we can end up allocating a system chunk and for an
4393 * atomic and race free space reservation in the chunk block reserve.
4395 lockdep_assert_held(&fs_info->chunk_mutex);
4397 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4398 spin_lock(&info->lock);
4399 left = info->total_bytes - btrfs_space_info_used(info, true);
4400 spin_unlock(&info->lock);
4402 num_devs = get_profile_num_devs(fs_info, type);
4404 /* num_devs device items to update and 1 chunk item to add or remove */
4405 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4406 btrfs_calc_trans_metadata_size(fs_info, 1);
4408 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4409 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4410 left, thresh, type);
4411 dump_space_info(fs_info, info, 0, 0);
4414 if (left < thresh) {
4415 u64 flags = btrfs_system_alloc_profile(fs_info);
4418 * Ignore failure to create system chunk. We might end up not
4419 * needing it, as we might not need to COW all nodes/leafs from
4420 * the paths we visit in the chunk tree (they were already COWed
4421 * or created in the current transaction for example).
4423 ret = btrfs_alloc_chunk(trans, flags);
4427 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4428 &fs_info->chunk_block_rsv,
4429 thresh, BTRFS_RESERVE_NO_FLUSH);
4431 trans->chunk_bytes_reserved += thresh;
4436 * If force is CHUNK_ALLOC_FORCE:
4437 * - return 1 if it successfully allocates a chunk,
4438 * - return errors including -ENOSPC otherwise.
4439 * If force is NOT CHUNK_ALLOC_FORCE:
4440 * - return 0 if it doesn't need to allocate a new chunk,
4441 * - return 1 if it successfully allocates a chunk,
4442 * - return errors including -ENOSPC otherwise.
4444 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4447 struct btrfs_fs_info *fs_info = trans->fs_info;
4448 struct btrfs_space_info *space_info;
4449 bool wait_for_alloc = false;
4450 bool should_alloc = false;
4453 /* Don't re-enter if we're already allocating a chunk */
4454 if (trans->allocating_chunk)
4457 space_info = __find_space_info(fs_info, flags);
4461 spin_lock(&space_info->lock);
4462 if (force < space_info->force_alloc)
4463 force = space_info->force_alloc;
4464 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4465 if (space_info->full) {
4466 /* No more free physical space */
4471 spin_unlock(&space_info->lock);
4473 } else if (!should_alloc) {
4474 spin_unlock(&space_info->lock);
4476 } else if (space_info->chunk_alloc) {
4478 * Someone is already allocating, so we need to block
4479 * until this someone is finished and then loop to
4480 * recheck if we should continue with our allocation
4483 wait_for_alloc = true;
4484 spin_unlock(&space_info->lock);
4485 mutex_lock(&fs_info->chunk_mutex);
4486 mutex_unlock(&fs_info->chunk_mutex);
4488 /* Proceed with allocation */
4489 space_info->chunk_alloc = 1;
4490 wait_for_alloc = false;
4491 spin_unlock(&space_info->lock);
4495 } while (wait_for_alloc);
4497 mutex_lock(&fs_info->chunk_mutex);
4498 trans->allocating_chunk = true;
4501 * If we have mixed data/metadata chunks we want to make sure we keep
4502 * allocating mixed chunks instead of individual chunks.
4504 if (btrfs_mixed_space_info(space_info))
4505 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4508 * if we're doing a data chunk, go ahead and make sure that
4509 * we keep a reasonable number of metadata chunks allocated in the
4512 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4513 fs_info->data_chunk_allocations++;
4514 if (!(fs_info->data_chunk_allocations %
4515 fs_info->metadata_ratio))
4516 force_metadata_allocation(fs_info);
4520 * Check if we have enough space in SYSTEM chunk because we may need
4521 * to update devices.
4523 check_system_chunk(trans, flags);
4525 ret = btrfs_alloc_chunk(trans, flags);
4526 trans->allocating_chunk = false;
4528 spin_lock(&space_info->lock);
4531 space_info->full = 1;
4538 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4540 space_info->chunk_alloc = 0;
4541 spin_unlock(&space_info->lock);
4542 mutex_unlock(&fs_info->chunk_mutex);
4544 * When we allocate a new chunk we reserve space in the chunk block
4545 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4546 * add new nodes/leafs to it if we end up needing to do it when
4547 * inserting the chunk item and updating device items as part of the
4548 * second phase of chunk allocation, performed by
4549 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4550 * large number of new block groups to create in our transaction
4551 * handle's new_bgs list to avoid exhausting the chunk block reserve
4552 * in extreme cases - like having a single transaction create many new
4553 * block groups when starting to write out the free space caches of all
4554 * the block groups that were made dirty during the lifetime of the
4557 if (trans->can_flush_pending_bgs &&
4558 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4559 btrfs_create_pending_block_groups(trans);
4560 btrfs_trans_release_chunk_metadata(trans);
4565 static int can_overcommit(struct btrfs_fs_info *fs_info,
4566 struct btrfs_space_info *space_info, u64 bytes,
4567 enum btrfs_reserve_flush_enum flush,
4570 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4577 /* Don't overcommit when in mixed mode. */
4578 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4582 profile = btrfs_system_alloc_profile(fs_info);
4584 profile = btrfs_metadata_alloc_profile(fs_info);
4586 used = btrfs_space_info_used(space_info, false);
4589 * We only want to allow over committing if we have lots of actual space
4590 * free, but if we don't have enough space to handle the global reserve
4591 * space then we could end up having a real enospc problem when trying
4592 * to allocate a chunk or some other such important allocation.
4594 spin_lock(&global_rsv->lock);
4595 space_size = calc_global_rsv_need_space(global_rsv);
4596 spin_unlock(&global_rsv->lock);
4597 if (used + space_size >= space_info->total_bytes)
4600 used += space_info->bytes_may_use;
4602 avail = atomic64_read(&fs_info->free_chunk_space);
4605 * If we have dup, raid1 or raid10 then only half of the free
4606 * space is actually useable. For raid56, the space info used
4607 * doesn't include the parity drive, so we don't have to
4610 factor = btrfs_bg_type_to_factor(profile);
4611 avail = div_u64(avail, factor);
4614 * If we aren't flushing all things, let us overcommit up to
4615 * 1/2th of the space. If we can flush, don't let us overcommit
4616 * too much, let it overcommit up to 1/8 of the space.
4618 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4623 if (used + bytes < space_info->total_bytes + avail)
4628 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4629 unsigned long nr_pages, int nr_items)
4631 struct super_block *sb = fs_info->sb;
4633 if (down_read_trylock(&sb->s_umount)) {
4634 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4635 up_read(&sb->s_umount);
4638 * We needn't worry the filesystem going from r/w to r/o though
4639 * we don't acquire ->s_umount mutex, because the filesystem
4640 * should guarantee the delalloc inodes list be empty after
4641 * the filesystem is readonly(all dirty pages are written to
4644 btrfs_start_delalloc_roots(fs_info, nr_items);
4645 if (!current->journal_info)
4646 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4650 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4656 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4657 nr = div64_u64(to_reclaim, bytes);
4663 #define EXTENT_SIZE_PER_ITEM SZ_256K
4666 * shrink metadata reservation for delalloc
4668 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4669 u64 orig, bool wait_ordered)
4671 struct btrfs_space_info *space_info;
4672 struct btrfs_trans_handle *trans;
4677 unsigned long nr_pages;
4680 /* Calc the number of the pages we need flush for space reservation */
4681 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4682 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4684 trans = (struct btrfs_trans_handle *)current->journal_info;
4685 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4687 delalloc_bytes = percpu_counter_sum_positive(
4688 &fs_info->delalloc_bytes);
4689 if (delalloc_bytes == 0) {
4693 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4698 while (delalloc_bytes && loops < 3) {
4699 max_reclaim = min(delalloc_bytes, to_reclaim);
4700 nr_pages = max_reclaim >> PAGE_SHIFT;
4701 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4703 * We need to wait for the async pages to actually start before
4706 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4710 if (max_reclaim <= nr_pages)
4713 max_reclaim -= nr_pages;
4715 wait_event(fs_info->async_submit_wait,
4716 atomic_read(&fs_info->async_delalloc_pages) <=
4719 spin_lock(&space_info->lock);
4720 if (list_empty(&space_info->tickets) &&
4721 list_empty(&space_info->priority_tickets)) {
4722 spin_unlock(&space_info->lock);
4725 spin_unlock(&space_info->lock);
4728 if (wait_ordered && !trans) {
4729 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4731 time_left = schedule_timeout_killable(1);
4735 delalloc_bytes = percpu_counter_sum_positive(
4736 &fs_info->delalloc_bytes);
4740 struct reserve_ticket {
4743 struct list_head list;
4744 wait_queue_head_t wait;
4748 * maybe_commit_transaction - possibly commit the transaction if its ok to
4749 * @root - the root we're allocating for
4750 * @bytes - the number of bytes we want to reserve
4751 * @force - force the commit
4753 * This will check to make sure that committing the transaction will actually
4754 * get us somewhere and then commit the transaction if it does. Otherwise it
4755 * will return -ENOSPC.
4757 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4758 struct btrfs_space_info *space_info)
4760 struct reserve_ticket *ticket = NULL;
4761 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4762 struct btrfs_trans_handle *trans;
4765 trans = (struct btrfs_trans_handle *)current->journal_info;
4769 spin_lock(&space_info->lock);
4770 if (!list_empty(&space_info->priority_tickets))
4771 ticket = list_first_entry(&space_info->priority_tickets,
4772 struct reserve_ticket, list);
4773 else if (!list_empty(&space_info->tickets))
4774 ticket = list_first_entry(&space_info->tickets,
4775 struct reserve_ticket, list);
4776 bytes = (ticket) ? ticket->bytes : 0;
4777 spin_unlock(&space_info->lock);
4782 /* See if there is enough pinned space to make this reservation */
4783 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4785 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4789 * See if there is some space in the delayed insertion reservation for
4792 if (space_info != delayed_rsv->space_info)
4795 spin_lock(&delayed_rsv->lock);
4796 if (delayed_rsv->size > bytes)
4799 bytes -= delayed_rsv->size;
4800 spin_unlock(&delayed_rsv->lock);
4802 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4804 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4809 trans = btrfs_join_transaction(fs_info->extent_root);
4813 return btrfs_commit_transaction(trans);
4817 * Try to flush some data based on policy set by @state. This is only advisory
4818 * and may fail for various reasons. The caller is supposed to examine the
4819 * state of @space_info to detect the outcome.
4821 static void flush_space(struct btrfs_fs_info *fs_info,
4822 struct btrfs_space_info *space_info, u64 num_bytes,
4825 struct btrfs_root *root = fs_info->extent_root;
4826 struct btrfs_trans_handle *trans;
4831 case FLUSH_DELAYED_ITEMS_NR:
4832 case FLUSH_DELAYED_ITEMS:
4833 if (state == FLUSH_DELAYED_ITEMS_NR)
4834 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4838 trans = btrfs_join_transaction(root);
4839 if (IS_ERR(trans)) {
4840 ret = PTR_ERR(trans);
4843 ret = btrfs_run_delayed_items_nr(trans, nr);
4844 btrfs_end_transaction(trans);
4846 case FLUSH_DELALLOC:
4847 case FLUSH_DELALLOC_WAIT:
4848 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4849 state == FLUSH_DELALLOC_WAIT);
4852 trans = btrfs_join_transaction(root);
4853 if (IS_ERR(trans)) {
4854 ret = PTR_ERR(trans);
4857 ret = do_chunk_alloc(trans,
4858 btrfs_metadata_alloc_profile(fs_info),
4859 CHUNK_ALLOC_NO_FORCE);
4860 btrfs_end_transaction(trans);
4861 if (ret > 0 || ret == -ENOSPC)
4865 ret = may_commit_transaction(fs_info, space_info);
4872 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4878 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4879 struct btrfs_space_info *space_info,
4882 struct reserve_ticket *ticket;
4887 list_for_each_entry(ticket, &space_info->tickets, list)
4888 to_reclaim += ticket->bytes;
4889 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4890 to_reclaim += ticket->bytes;
4894 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4895 if (can_overcommit(fs_info, space_info, to_reclaim,
4896 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4899 used = btrfs_space_info_used(space_info, true);
4901 if (can_overcommit(fs_info, space_info, SZ_1M,
4902 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4903 expected = div_factor_fine(space_info->total_bytes, 95);
4905 expected = div_factor_fine(space_info->total_bytes, 90);
4907 if (used > expected)
4908 to_reclaim = used - expected;
4911 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4912 space_info->bytes_reserved);
4916 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
4917 struct btrfs_space_info *space_info,
4918 u64 used, bool system_chunk)
4920 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4922 /* If we're just plain full then async reclaim just slows us down. */
4923 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4926 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4930 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4931 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4934 static void wake_all_tickets(struct list_head *head)
4936 struct reserve_ticket *ticket;
4938 while (!list_empty(head)) {
4939 ticket = list_first_entry(head, struct reserve_ticket, list);
4940 list_del_init(&ticket->list);
4941 ticket->error = -ENOSPC;
4942 wake_up(&ticket->wait);
4947 * This is for normal flushers, we can wait all goddamned day if we want to. We
4948 * will loop and continuously try to flush as long as we are making progress.
4949 * We count progress as clearing off tickets each time we have to loop.
4951 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4953 struct btrfs_fs_info *fs_info;
4954 struct btrfs_space_info *space_info;
4957 int commit_cycles = 0;
4958 u64 last_tickets_id;
4960 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4961 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4963 spin_lock(&space_info->lock);
4964 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4967 space_info->flush = 0;
4968 spin_unlock(&space_info->lock);
4971 last_tickets_id = space_info->tickets_id;
4972 spin_unlock(&space_info->lock);
4974 flush_state = FLUSH_DELAYED_ITEMS_NR;
4976 flush_space(fs_info, space_info, to_reclaim, flush_state);
4977 spin_lock(&space_info->lock);
4978 if (list_empty(&space_info->tickets)) {
4979 space_info->flush = 0;
4980 spin_unlock(&space_info->lock);
4983 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
4986 if (last_tickets_id == space_info->tickets_id) {
4989 last_tickets_id = space_info->tickets_id;
4990 flush_state = FLUSH_DELAYED_ITEMS_NR;
4995 if (flush_state > COMMIT_TRANS) {
4997 if (commit_cycles > 2) {
4998 wake_all_tickets(&space_info->tickets);
4999 space_info->flush = 0;
5001 flush_state = FLUSH_DELAYED_ITEMS_NR;
5004 spin_unlock(&space_info->lock);
5005 } while (flush_state <= COMMIT_TRANS);
5008 void btrfs_init_async_reclaim_work(struct work_struct *work)
5010 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5013 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5014 struct btrfs_space_info *space_info,
5015 struct reserve_ticket *ticket)
5018 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5020 spin_lock(&space_info->lock);
5021 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5024 spin_unlock(&space_info->lock);
5027 spin_unlock(&space_info->lock);
5030 flush_space(fs_info, space_info, to_reclaim, flush_state);
5032 spin_lock(&space_info->lock);
5033 if (ticket->bytes == 0) {
5034 spin_unlock(&space_info->lock);
5037 spin_unlock(&space_info->lock);
5040 * Priority flushers can't wait on delalloc without
5043 if (flush_state == FLUSH_DELALLOC ||
5044 flush_state == FLUSH_DELALLOC_WAIT)
5045 flush_state = ALLOC_CHUNK;
5046 } while (flush_state < COMMIT_TRANS);
5049 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5050 struct btrfs_space_info *space_info,
5051 struct reserve_ticket *ticket, u64 orig_bytes)
5057 spin_lock(&space_info->lock);
5058 while (ticket->bytes > 0 && ticket->error == 0) {
5059 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5064 spin_unlock(&space_info->lock);
5068 finish_wait(&ticket->wait, &wait);
5069 spin_lock(&space_info->lock);
5072 ret = ticket->error;
5073 if (!list_empty(&ticket->list))
5074 list_del_init(&ticket->list);
5075 if (ticket->bytes && ticket->bytes < orig_bytes) {
5076 u64 num_bytes = orig_bytes - ticket->bytes;
5077 space_info->bytes_may_use -= num_bytes;
5078 trace_btrfs_space_reservation(fs_info, "space_info",
5079 space_info->flags, num_bytes, 0);
5081 spin_unlock(&space_info->lock);
5087 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5088 * @root - the root we're allocating for
5089 * @space_info - the space info we want to allocate from
5090 * @orig_bytes - the number of bytes we want
5091 * @flush - whether or not we can flush to make our reservation
5093 * This will reserve orig_bytes number of bytes from the space info associated
5094 * with the block_rsv. If there is not enough space it will make an attempt to
5095 * flush out space to make room. It will do this by flushing delalloc if
5096 * possible or committing the transaction. If flush is 0 then no attempts to
5097 * regain reservations will be made and this will fail if there is not enough
5100 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5101 struct btrfs_space_info *space_info,
5103 enum btrfs_reserve_flush_enum flush,
5106 struct reserve_ticket ticket;
5111 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5113 spin_lock(&space_info->lock);
5115 used = btrfs_space_info_used(space_info, true);
5118 * If we have enough space then hooray, make our reservation and carry
5119 * on. If not see if we can overcommit, and if we can, hooray carry on.
5120 * If not things get more complicated.
5122 if (used + orig_bytes <= space_info->total_bytes) {
5123 space_info->bytes_may_use += orig_bytes;
5124 trace_btrfs_space_reservation(fs_info, "space_info",
5125 space_info->flags, orig_bytes, 1);
5127 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5129 space_info->bytes_may_use += orig_bytes;
5130 trace_btrfs_space_reservation(fs_info, "space_info",
5131 space_info->flags, orig_bytes, 1);
5136 * If we couldn't make a reservation then setup our reservation ticket
5137 * and kick the async worker if it's not already running.
5139 * If we are a priority flusher then we just need to add our ticket to
5140 * the list and we will do our own flushing further down.
5142 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5143 ticket.bytes = orig_bytes;
5145 init_waitqueue_head(&ticket.wait);
5146 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5147 list_add_tail(&ticket.list, &space_info->tickets);
5148 if (!space_info->flush) {
5149 space_info->flush = 1;
5150 trace_btrfs_trigger_flush(fs_info,
5154 queue_work(system_unbound_wq,
5155 &fs_info->async_reclaim_work);
5158 list_add_tail(&ticket.list,
5159 &space_info->priority_tickets);
5161 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5164 * We will do the space reservation dance during log replay,
5165 * which means we won't have fs_info->fs_root set, so don't do
5166 * the async reclaim as we will panic.
5168 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5169 need_do_async_reclaim(fs_info, space_info,
5170 used, system_chunk) &&
5171 !work_busy(&fs_info->async_reclaim_work)) {
5172 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5173 orig_bytes, flush, "preempt");
5174 queue_work(system_unbound_wq,
5175 &fs_info->async_reclaim_work);
5178 spin_unlock(&space_info->lock);
5179 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5182 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5183 return wait_reserve_ticket(fs_info, space_info, &ticket,
5187 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5188 spin_lock(&space_info->lock);
5190 if (ticket.bytes < orig_bytes) {
5191 u64 num_bytes = orig_bytes - ticket.bytes;
5192 space_info->bytes_may_use -= num_bytes;
5193 trace_btrfs_space_reservation(fs_info, "space_info",
5198 list_del_init(&ticket.list);
5201 spin_unlock(&space_info->lock);
5202 ASSERT(list_empty(&ticket.list));
5207 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5208 * @root - the root we're allocating for
5209 * @block_rsv - the block_rsv we're allocating for
5210 * @orig_bytes - the number of bytes we want
5211 * @flush - whether or not we can flush to make our reservation
5213 * This will reserve orgi_bytes number of bytes from the space info associated
5214 * with the block_rsv. If there is not enough space it will make an attempt to
5215 * flush out space to make room. It will do this by flushing delalloc if
5216 * possible or committing the transaction. If flush is 0 then no attempts to
5217 * regain reservations will be made and this will fail if there is not enough
5220 static int reserve_metadata_bytes(struct btrfs_root *root,
5221 struct btrfs_block_rsv *block_rsv,
5223 enum btrfs_reserve_flush_enum flush)
5225 struct btrfs_fs_info *fs_info = root->fs_info;
5226 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5228 bool system_chunk = (root == fs_info->chunk_root);
5230 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5231 orig_bytes, flush, system_chunk);
5232 if (ret == -ENOSPC &&
5233 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5234 if (block_rsv != global_rsv &&
5235 !block_rsv_use_bytes(global_rsv, orig_bytes))
5238 if (ret == -ENOSPC) {
5239 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5240 block_rsv->space_info->flags,
5243 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5244 dump_space_info(fs_info, block_rsv->space_info,
5250 static struct btrfs_block_rsv *get_block_rsv(
5251 const struct btrfs_trans_handle *trans,
5252 const struct btrfs_root *root)
5254 struct btrfs_fs_info *fs_info = root->fs_info;
5255 struct btrfs_block_rsv *block_rsv = NULL;
5257 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5258 (root == fs_info->csum_root && trans->adding_csums) ||
5259 (root == fs_info->uuid_root))
5260 block_rsv = trans->block_rsv;
5263 block_rsv = root->block_rsv;
5266 block_rsv = &fs_info->empty_block_rsv;
5271 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5275 spin_lock(&block_rsv->lock);
5276 if (block_rsv->reserved >= num_bytes) {
5277 block_rsv->reserved -= num_bytes;
5278 if (block_rsv->reserved < block_rsv->size)
5279 block_rsv->full = 0;
5282 spin_unlock(&block_rsv->lock);
5286 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5287 u64 num_bytes, int update_size)
5289 spin_lock(&block_rsv->lock);
5290 block_rsv->reserved += num_bytes;
5292 block_rsv->size += num_bytes;
5293 else if (block_rsv->reserved >= block_rsv->size)
5294 block_rsv->full = 1;
5295 spin_unlock(&block_rsv->lock);
5298 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5299 struct btrfs_block_rsv *dest, u64 num_bytes,
5302 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5305 if (global_rsv->space_info != dest->space_info)
5308 spin_lock(&global_rsv->lock);
5309 min_bytes = div_factor(global_rsv->size, min_factor);
5310 if (global_rsv->reserved < min_bytes + num_bytes) {
5311 spin_unlock(&global_rsv->lock);
5314 global_rsv->reserved -= num_bytes;
5315 if (global_rsv->reserved < global_rsv->size)
5316 global_rsv->full = 0;
5317 spin_unlock(&global_rsv->lock);
5319 block_rsv_add_bytes(dest, num_bytes, 1);
5324 * This is for space we already have accounted in space_info->bytes_may_use, so
5325 * basically when we're returning space from block_rsv's.
5327 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5328 struct btrfs_space_info *space_info,
5331 struct reserve_ticket *ticket;
5332 struct list_head *head;
5334 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5335 bool check_overcommit = false;
5337 spin_lock(&space_info->lock);
5338 head = &space_info->priority_tickets;
5341 * If we are over our limit then we need to check and see if we can
5342 * overcommit, and if we can't then we just need to free up our space
5343 * and not satisfy any requests.
5345 used = btrfs_space_info_used(space_info, true);
5346 if (used - num_bytes >= space_info->total_bytes)
5347 check_overcommit = true;
5349 while (!list_empty(head) && num_bytes) {
5350 ticket = list_first_entry(head, struct reserve_ticket,
5353 * We use 0 bytes because this space is already reserved, so
5354 * adding the ticket space would be a double count.
5356 if (check_overcommit &&
5357 !can_overcommit(fs_info, space_info, 0, flush, false))
5359 if (num_bytes >= ticket->bytes) {
5360 list_del_init(&ticket->list);
5361 num_bytes -= ticket->bytes;
5363 space_info->tickets_id++;
5364 wake_up(&ticket->wait);
5366 ticket->bytes -= num_bytes;
5371 if (num_bytes && head == &space_info->priority_tickets) {
5372 head = &space_info->tickets;
5373 flush = BTRFS_RESERVE_FLUSH_ALL;
5376 space_info->bytes_may_use -= num_bytes;
5377 trace_btrfs_space_reservation(fs_info, "space_info",
5378 space_info->flags, num_bytes, 0);
5379 spin_unlock(&space_info->lock);
5383 * This is for newly allocated space that isn't accounted in
5384 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5385 * we use this helper.
5387 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5388 struct btrfs_space_info *space_info,
5391 struct reserve_ticket *ticket;
5392 struct list_head *head = &space_info->priority_tickets;
5395 while (!list_empty(head) && num_bytes) {
5396 ticket = list_first_entry(head, struct reserve_ticket,
5398 if (num_bytes >= ticket->bytes) {
5399 trace_btrfs_space_reservation(fs_info, "space_info",
5402 list_del_init(&ticket->list);
5403 num_bytes -= ticket->bytes;
5404 space_info->bytes_may_use += ticket->bytes;
5406 space_info->tickets_id++;
5407 wake_up(&ticket->wait);
5409 trace_btrfs_space_reservation(fs_info, "space_info",
5412 space_info->bytes_may_use += num_bytes;
5413 ticket->bytes -= num_bytes;
5418 if (num_bytes && head == &space_info->priority_tickets) {
5419 head = &space_info->tickets;
5424 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5425 struct btrfs_block_rsv *block_rsv,
5426 struct btrfs_block_rsv *dest, u64 num_bytes,
5427 u64 *qgroup_to_release_ret)
5429 struct btrfs_space_info *space_info = block_rsv->space_info;
5430 u64 qgroup_to_release = 0;
5433 spin_lock(&block_rsv->lock);
5434 if (num_bytes == (u64)-1) {
5435 num_bytes = block_rsv->size;
5436 qgroup_to_release = block_rsv->qgroup_rsv_size;
5438 block_rsv->size -= num_bytes;
5439 if (block_rsv->reserved >= block_rsv->size) {
5440 num_bytes = block_rsv->reserved - block_rsv->size;
5441 block_rsv->reserved = block_rsv->size;
5442 block_rsv->full = 1;
5446 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5447 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5448 block_rsv->qgroup_rsv_size;
5449 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5451 qgroup_to_release = 0;
5453 spin_unlock(&block_rsv->lock);
5456 if (num_bytes > 0) {
5458 spin_lock(&dest->lock);
5462 bytes_to_add = dest->size - dest->reserved;
5463 bytes_to_add = min(num_bytes, bytes_to_add);
5464 dest->reserved += bytes_to_add;
5465 if (dest->reserved >= dest->size)
5467 num_bytes -= bytes_to_add;
5469 spin_unlock(&dest->lock);
5472 space_info_add_old_bytes(fs_info, space_info,
5475 if (qgroup_to_release_ret)
5476 *qgroup_to_release_ret = qgroup_to_release;
5480 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5481 struct btrfs_block_rsv *dst, u64 num_bytes,
5486 ret = block_rsv_use_bytes(src, num_bytes);
5490 block_rsv_add_bytes(dst, num_bytes, update_size);
5494 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5496 memset(rsv, 0, sizeof(*rsv));
5497 spin_lock_init(&rsv->lock);
5501 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5502 struct btrfs_block_rsv *rsv,
5503 unsigned short type)
5505 btrfs_init_block_rsv(rsv, type);
5506 rsv->space_info = __find_space_info(fs_info,
5507 BTRFS_BLOCK_GROUP_METADATA);
5510 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5511 unsigned short type)
5513 struct btrfs_block_rsv *block_rsv;
5515 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5519 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5523 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5524 struct btrfs_block_rsv *rsv)
5528 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5532 int btrfs_block_rsv_add(struct btrfs_root *root,
5533 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5534 enum btrfs_reserve_flush_enum flush)
5541 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5543 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5550 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5558 spin_lock(&block_rsv->lock);
5559 num_bytes = div_factor(block_rsv->size, min_factor);
5560 if (block_rsv->reserved >= num_bytes)
5562 spin_unlock(&block_rsv->lock);
5567 int btrfs_block_rsv_refill(struct btrfs_root *root,
5568 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5569 enum btrfs_reserve_flush_enum flush)
5577 spin_lock(&block_rsv->lock);
5578 num_bytes = min_reserved;
5579 if (block_rsv->reserved >= num_bytes)
5582 num_bytes -= block_rsv->reserved;
5583 spin_unlock(&block_rsv->lock);
5588 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5590 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5598 * btrfs_inode_rsv_refill - refill the inode block rsv.
5599 * @inode - the inode we are refilling.
5600 * @flush - the flusing restriction.
5602 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5603 * block_rsv->size as the minimum size. We'll either refill the missing amount
5604 * or return if we already have enough space. This will also handle the resreve
5605 * tracepoint for the reserved amount.
5607 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5608 enum btrfs_reserve_flush_enum flush)
5610 struct btrfs_root *root = inode->root;
5611 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5613 u64 qgroup_num_bytes = 0;
5616 spin_lock(&block_rsv->lock);
5617 if (block_rsv->reserved < block_rsv->size)
5618 num_bytes = block_rsv->size - block_rsv->reserved;
5619 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5620 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5621 block_rsv->qgroup_rsv_reserved;
5622 spin_unlock(&block_rsv->lock);
5627 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5630 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5632 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5633 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5634 btrfs_ino(inode), num_bytes, 1);
5636 /* Don't forget to increase qgroup_rsv_reserved */
5637 spin_lock(&block_rsv->lock);
5638 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5639 spin_unlock(&block_rsv->lock);
5641 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5646 * btrfs_inode_rsv_release - release any excessive reservation.
5647 * @inode - the inode we need to release from.
5648 * @qgroup_free - free or convert qgroup meta.
5649 * Unlike normal operation, qgroup meta reservation needs to know if we are
5650 * freeing qgroup reservation or just converting it into per-trans. Normally
5651 * @qgroup_free is true for error handling, and false for normal release.
5653 * This is the same as btrfs_block_rsv_release, except that it handles the
5654 * tracepoint for the reservation.
5656 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5658 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5659 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5660 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5662 u64 qgroup_to_release = 0;
5665 * Since we statically set the block_rsv->size we just want to say we
5666 * are releasing 0 bytes, and then we'll just get the reservation over
5669 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5670 &qgroup_to_release);
5672 trace_btrfs_space_reservation(fs_info, "delalloc",
5673 btrfs_ino(inode), released, 0);
5675 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5677 btrfs_qgroup_convert_reserved_meta(inode->root,
5681 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5682 struct btrfs_block_rsv *block_rsv,
5685 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5687 if (global_rsv == block_rsv ||
5688 block_rsv->space_info != global_rsv->space_info)
5690 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5693 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5695 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5696 struct btrfs_space_info *sinfo = block_rsv->space_info;
5700 * The global block rsv is based on the size of the extent tree, the
5701 * checksum tree and the root tree. If the fs is empty we want to set
5702 * it to a minimal amount for safety.
5704 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5705 btrfs_root_used(&fs_info->csum_root->root_item) +
5706 btrfs_root_used(&fs_info->tree_root->root_item);
5707 num_bytes = max_t(u64, num_bytes, SZ_16M);
5709 spin_lock(&sinfo->lock);
5710 spin_lock(&block_rsv->lock);
5712 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5714 if (block_rsv->reserved < block_rsv->size) {
5715 num_bytes = btrfs_space_info_used(sinfo, true);
5716 if (sinfo->total_bytes > num_bytes) {
5717 num_bytes = sinfo->total_bytes - num_bytes;
5718 num_bytes = min(num_bytes,
5719 block_rsv->size - block_rsv->reserved);
5720 block_rsv->reserved += num_bytes;
5721 sinfo->bytes_may_use += num_bytes;
5722 trace_btrfs_space_reservation(fs_info, "space_info",
5723 sinfo->flags, num_bytes,
5726 } else if (block_rsv->reserved > block_rsv->size) {
5727 num_bytes = block_rsv->reserved - block_rsv->size;
5728 sinfo->bytes_may_use -= num_bytes;
5729 trace_btrfs_space_reservation(fs_info, "space_info",
5730 sinfo->flags, num_bytes, 0);
5731 block_rsv->reserved = block_rsv->size;
5734 if (block_rsv->reserved == block_rsv->size)
5735 block_rsv->full = 1;
5737 block_rsv->full = 0;
5739 spin_unlock(&block_rsv->lock);
5740 spin_unlock(&sinfo->lock);
5743 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5745 struct btrfs_space_info *space_info;
5747 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5748 fs_info->chunk_block_rsv.space_info = space_info;
5750 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5751 fs_info->global_block_rsv.space_info = space_info;
5752 fs_info->trans_block_rsv.space_info = space_info;
5753 fs_info->empty_block_rsv.space_info = space_info;
5754 fs_info->delayed_block_rsv.space_info = space_info;
5756 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5757 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5758 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5759 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5760 if (fs_info->quota_root)
5761 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5762 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5764 update_global_block_rsv(fs_info);
5767 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5769 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5771 WARN_ON(fs_info->trans_block_rsv.size > 0);
5772 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5773 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5774 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5775 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5776 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5781 * To be called after all the new block groups attached to the transaction
5782 * handle have been created (btrfs_create_pending_block_groups()).
5784 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5786 struct btrfs_fs_info *fs_info = trans->fs_info;
5788 if (!trans->chunk_bytes_reserved)
5791 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5793 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5794 trans->chunk_bytes_reserved, NULL);
5795 trans->chunk_bytes_reserved = 0;
5799 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5800 * root: the root of the parent directory
5801 * rsv: block reservation
5802 * items: the number of items that we need do reservation
5803 * qgroup_reserved: used to return the reserved size in qgroup
5805 * This function is used to reserve the space for snapshot/subvolume
5806 * creation and deletion. Those operations are different with the
5807 * common file/directory operations, they change two fs/file trees
5808 * and root tree, the number of items that the qgroup reserves is
5809 * different with the free space reservation. So we can not use
5810 * the space reservation mechanism in start_transaction().
5812 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5813 struct btrfs_block_rsv *rsv,
5815 bool use_global_rsv)
5819 struct btrfs_fs_info *fs_info = root->fs_info;
5820 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5822 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5823 /* One for parent inode, two for dir entries */
5824 num_bytes = 3 * fs_info->nodesize;
5825 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5832 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5833 rsv->space_info = __find_space_info(fs_info,
5834 BTRFS_BLOCK_GROUP_METADATA);
5835 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5836 BTRFS_RESERVE_FLUSH_ALL);
5838 if (ret == -ENOSPC && use_global_rsv)
5839 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5841 if (ret && num_bytes)
5842 btrfs_qgroup_free_meta_prealloc(root, num_bytes);
5847 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5848 struct btrfs_block_rsv *rsv)
5850 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5853 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5854 struct btrfs_inode *inode)
5856 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5857 u64 reserve_size = 0;
5858 u64 qgroup_rsv_size = 0;
5860 unsigned outstanding_extents;
5862 lockdep_assert_held(&inode->lock);
5863 outstanding_extents = inode->outstanding_extents;
5864 if (outstanding_extents)
5865 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
5866 outstanding_extents + 1);
5867 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
5869 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
5872 * For qgroup rsv, the calculation is very simple:
5873 * account one nodesize for each outstanding extent
5875 * This is overestimating in most cases.
5877 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
5879 spin_lock(&block_rsv->lock);
5880 block_rsv->size = reserve_size;
5881 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
5882 spin_unlock(&block_rsv->lock);
5885 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5887 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5888 unsigned nr_extents;
5889 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5891 bool delalloc_lock = true;
5893 /* If we are a free space inode we need to not flush since we will be in
5894 * the middle of a transaction commit. We also don't need the delalloc
5895 * mutex since we won't race with anybody. We need this mostly to make
5896 * lockdep shut its filthy mouth.
5898 * If we have a transaction open (can happen if we call truncate_block
5899 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5901 if (btrfs_is_free_space_inode(inode)) {
5902 flush = BTRFS_RESERVE_NO_FLUSH;
5903 delalloc_lock = false;
5905 if (current->journal_info)
5906 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5908 if (btrfs_transaction_in_commit(fs_info))
5909 schedule_timeout(1);
5913 mutex_lock(&inode->delalloc_mutex);
5915 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5917 /* Add our new extents and calculate the new rsv size. */
5918 spin_lock(&inode->lock);
5919 nr_extents = count_max_extents(num_bytes);
5920 btrfs_mod_outstanding_extents(inode, nr_extents);
5921 inode->csum_bytes += num_bytes;
5922 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5923 spin_unlock(&inode->lock);
5925 ret = btrfs_inode_rsv_refill(inode, flush);
5930 mutex_unlock(&inode->delalloc_mutex);
5934 spin_lock(&inode->lock);
5935 nr_extents = count_max_extents(num_bytes);
5936 btrfs_mod_outstanding_extents(inode, -nr_extents);
5937 inode->csum_bytes -= num_bytes;
5938 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5939 spin_unlock(&inode->lock);
5941 btrfs_inode_rsv_release(inode, true);
5943 mutex_unlock(&inode->delalloc_mutex);
5948 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5949 * @inode: the inode to release the reservation for.
5950 * @num_bytes: the number of bytes we are releasing.
5951 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
5953 * This will release the metadata reservation for an inode. This can be called
5954 * once we complete IO for a given set of bytes to release their metadata
5955 * reservations, or on error for the same reason.
5957 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
5960 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5962 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5963 spin_lock(&inode->lock);
5964 inode->csum_bytes -= num_bytes;
5965 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5966 spin_unlock(&inode->lock);
5968 if (btrfs_is_testing(fs_info))
5971 btrfs_inode_rsv_release(inode, qgroup_free);
5975 * btrfs_delalloc_release_extents - release our outstanding_extents
5976 * @inode: the inode to balance the reservation for.
5977 * @num_bytes: the number of bytes we originally reserved with
5978 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
5980 * When we reserve space we increase outstanding_extents for the extents we may
5981 * add. Once we've set the range as delalloc or created our ordered extents we
5982 * have outstanding_extents to track the real usage, so we use this to free our
5983 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
5984 * with btrfs_delalloc_reserve_metadata.
5986 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
5989 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5990 unsigned num_extents;
5992 spin_lock(&inode->lock);
5993 num_extents = count_max_extents(num_bytes);
5994 btrfs_mod_outstanding_extents(inode, -num_extents);
5995 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5996 spin_unlock(&inode->lock);
5998 if (btrfs_is_testing(fs_info))
6001 btrfs_inode_rsv_release(inode, qgroup_free);
6005 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6007 * @inode: inode we're writing to
6008 * @start: start range we are writing to
6009 * @len: how long the range we are writing to
6010 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6011 * current reservation.
6013 * This will do the following things
6015 * o reserve space in data space info for num bytes
6016 * and reserve precious corresponding qgroup space
6017 * (Done in check_data_free_space)
6019 * o reserve space for metadata space, based on the number of outstanding
6020 * extents and how much csums will be needed
6021 * also reserve metadata space in a per root over-reserve method.
6022 * o add to the inodes->delalloc_bytes
6023 * o add it to the fs_info's delalloc inodes list.
6024 * (Above 3 all done in delalloc_reserve_metadata)
6026 * Return 0 for success
6027 * Return <0 for error(-ENOSPC or -EQUOT)
6029 int btrfs_delalloc_reserve_space(struct inode *inode,
6030 struct extent_changeset **reserved, u64 start, u64 len)
6034 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6037 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6039 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6044 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6045 * @inode: inode we're releasing space for
6046 * @start: start position of the space already reserved
6047 * @len: the len of the space already reserved
6048 * @release_bytes: the len of the space we consumed or didn't use
6050 * This function will release the metadata space that was not used and will
6051 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6052 * list if there are no delalloc bytes left.
6053 * Also it will handle the qgroup reserved space.
6055 void btrfs_delalloc_release_space(struct inode *inode,
6056 struct extent_changeset *reserved,
6057 u64 start, u64 len, bool qgroup_free)
6059 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6060 btrfs_free_reserved_data_space(inode, reserved, start, len);
6063 static int update_block_group(struct btrfs_trans_handle *trans,
6064 struct btrfs_fs_info *info, u64 bytenr,
6065 u64 num_bytes, int alloc)
6067 struct btrfs_block_group_cache *cache = NULL;
6068 u64 total = num_bytes;
6073 /* block accounting for super block */
6074 spin_lock(&info->delalloc_root_lock);
6075 old_val = btrfs_super_bytes_used(info->super_copy);
6077 old_val += num_bytes;
6079 old_val -= num_bytes;
6080 btrfs_set_super_bytes_used(info->super_copy, old_val);
6081 spin_unlock(&info->delalloc_root_lock);
6084 cache = btrfs_lookup_block_group(info, bytenr);
6087 factor = btrfs_bg_type_to_factor(cache->flags);
6090 * If this block group has free space cache written out, we
6091 * need to make sure to load it if we are removing space. This
6092 * is because we need the unpinning stage to actually add the
6093 * space back to the block group, otherwise we will leak space.
6095 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6096 cache_block_group(cache, 1);
6098 byte_in_group = bytenr - cache->key.objectid;
6099 WARN_ON(byte_in_group > cache->key.offset);
6101 spin_lock(&cache->space_info->lock);
6102 spin_lock(&cache->lock);
6104 if (btrfs_test_opt(info, SPACE_CACHE) &&
6105 cache->disk_cache_state < BTRFS_DC_CLEAR)
6106 cache->disk_cache_state = BTRFS_DC_CLEAR;
6108 old_val = btrfs_block_group_used(&cache->item);
6109 num_bytes = min(total, cache->key.offset - byte_in_group);
6111 old_val += num_bytes;
6112 btrfs_set_block_group_used(&cache->item, old_val);
6113 cache->reserved -= num_bytes;
6114 cache->space_info->bytes_reserved -= num_bytes;
6115 cache->space_info->bytes_used += num_bytes;
6116 cache->space_info->disk_used += num_bytes * factor;
6117 spin_unlock(&cache->lock);
6118 spin_unlock(&cache->space_info->lock);
6120 old_val -= num_bytes;
6121 btrfs_set_block_group_used(&cache->item, old_val);
6122 cache->pinned += num_bytes;
6123 cache->space_info->bytes_pinned += num_bytes;
6124 cache->space_info->bytes_used -= num_bytes;
6125 cache->space_info->disk_used -= num_bytes * factor;
6126 spin_unlock(&cache->lock);
6127 spin_unlock(&cache->space_info->lock);
6129 trace_btrfs_space_reservation(info, "pinned",
6130 cache->space_info->flags,
6132 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6134 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6135 set_extent_dirty(info->pinned_extents,
6136 bytenr, bytenr + num_bytes - 1,
6137 GFP_NOFS | __GFP_NOFAIL);
6140 spin_lock(&trans->transaction->dirty_bgs_lock);
6141 if (list_empty(&cache->dirty_list)) {
6142 list_add_tail(&cache->dirty_list,
6143 &trans->transaction->dirty_bgs);
6144 trans->transaction->num_dirty_bgs++;
6145 btrfs_get_block_group(cache);
6147 spin_unlock(&trans->transaction->dirty_bgs_lock);
6150 * No longer have used bytes in this block group, queue it for
6151 * deletion. We do this after adding the block group to the
6152 * dirty list to avoid races between cleaner kthread and space
6155 if (!alloc && old_val == 0)
6156 btrfs_mark_bg_unused(cache);
6158 btrfs_put_block_group(cache);
6160 bytenr += num_bytes;
6165 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6167 struct btrfs_block_group_cache *cache;
6170 spin_lock(&fs_info->block_group_cache_lock);
6171 bytenr = fs_info->first_logical_byte;
6172 spin_unlock(&fs_info->block_group_cache_lock);
6174 if (bytenr < (u64)-1)
6177 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6181 bytenr = cache->key.objectid;
6182 btrfs_put_block_group(cache);
6187 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6188 struct btrfs_block_group_cache *cache,
6189 u64 bytenr, u64 num_bytes, int reserved)
6191 spin_lock(&cache->space_info->lock);
6192 spin_lock(&cache->lock);
6193 cache->pinned += num_bytes;
6194 cache->space_info->bytes_pinned += num_bytes;
6196 cache->reserved -= num_bytes;
6197 cache->space_info->bytes_reserved -= num_bytes;
6199 spin_unlock(&cache->lock);
6200 spin_unlock(&cache->space_info->lock);
6202 trace_btrfs_space_reservation(fs_info, "pinned",
6203 cache->space_info->flags, num_bytes, 1);
6204 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6205 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6206 set_extent_dirty(fs_info->pinned_extents, bytenr,
6207 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6212 * this function must be called within transaction
6214 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6215 u64 bytenr, u64 num_bytes, int reserved)
6217 struct btrfs_block_group_cache *cache;
6219 cache = btrfs_lookup_block_group(fs_info, bytenr);
6220 BUG_ON(!cache); /* Logic error */
6222 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6224 btrfs_put_block_group(cache);
6229 * this function must be called within transaction
6231 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6232 u64 bytenr, u64 num_bytes)
6234 struct btrfs_block_group_cache *cache;
6237 cache = btrfs_lookup_block_group(fs_info, bytenr);
6242 * pull in the free space cache (if any) so that our pin
6243 * removes the free space from the cache. We have load_only set
6244 * to one because the slow code to read in the free extents does check
6245 * the pinned extents.
6247 cache_block_group(cache, 1);
6249 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6251 /* remove us from the free space cache (if we're there at all) */
6252 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6253 btrfs_put_block_group(cache);
6257 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6258 u64 start, u64 num_bytes)
6261 struct btrfs_block_group_cache *block_group;
6262 struct btrfs_caching_control *caching_ctl;
6264 block_group = btrfs_lookup_block_group(fs_info, start);
6268 cache_block_group(block_group, 0);
6269 caching_ctl = get_caching_control(block_group);
6273 BUG_ON(!block_group_cache_done(block_group));
6274 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6276 mutex_lock(&caching_ctl->mutex);
6278 if (start >= caching_ctl->progress) {
6279 ret = add_excluded_extent(fs_info, start, num_bytes);
6280 } else if (start + num_bytes <= caching_ctl->progress) {
6281 ret = btrfs_remove_free_space(block_group,
6284 num_bytes = caching_ctl->progress - start;
6285 ret = btrfs_remove_free_space(block_group,
6290 num_bytes = (start + num_bytes) -
6291 caching_ctl->progress;
6292 start = caching_ctl->progress;
6293 ret = add_excluded_extent(fs_info, start, num_bytes);
6296 mutex_unlock(&caching_ctl->mutex);
6297 put_caching_control(caching_ctl);
6299 btrfs_put_block_group(block_group);
6303 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6304 struct extent_buffer *eb)
6306 struct btrfs_file_extent_item *item;
6307 struct btrfs_key key;
6312 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6315 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6316 btrfs_item_key_to_cpu(eb, &key, i);
6317 if (key.type != BTRFS_EXTENT_DATA_KEY)
6319 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6320 found_type = btrfs_file_extent_type(eb, item);
6321 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6323 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6325 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6326 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6327 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6336 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6338 atomic_inc(&bg->reservations);
6341 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6344 struct btrfs_block_group_cache *bg;
6346 bg = btrfs_lookup_block_group(fs_info, start);
6348 if (atomic_dec_and_test(&bg->reservations))
6349 wake_up_var(&bg->reservations);
6350 btrfs_put_block_group(bg);
6353 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6355 struct btrfs_space_info *space_info = bg->space_info;
6359 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6363 * Our block group is read only but before we set it to read only,
6364 * some task might have had allocated an extent from it already, but it
6365 * has not yet created a respective ordered extent (and added it to a
6366 * root's list of ordered extents).
6367 * Therefore wait for any task currently allocating extents, since the
6368 * block group's reservations counter is incremented while a read lock
6369 * on the groups' semaphore is held and decremented after releasing
6370 * the read access on that semaphore and creating the ordered extent.
6372 down_write(&space_info->groups_sem);
6373 up_write(&space_info->groups_sem);
6375 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6379 * btrfs_add_reserved_bytes - update the block_group and space info counters
6380 * @cache: The cache we are manipulating
6381 * @ram_bytes: The number of bytes of file content, and will be same to
6382 * @num_bytes except for the compress path.
6383 * @num_bytes: The number of bytes in question
6384 * @delalloc: The blocks are allocated for the delalloc write
6386 * This is called by the allocator when it reserves space. If this is a
6387 * reservation and the block group has become read only we cannot make the
6388 * reservation and return -EAGAIN, otherwise this function always succeeds.
6390 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6391 u64 ram_bytes, u64 num_bytes, int delalloc)
6393 struct btrfs_space_info *space_info = cache->space_info;
6396 spin_lock(&space_info->lock);
6397 spin_lock(&cache->lock);
6401 cache->reserved += num_bytes;
6402 space_info->bytes_reserved += num_bytes;
6404 trace_btrfs_space_reservation(cache->fs_info,
6405 "space_info", space_info->flags,
6407 space_info->bytes_may_use -= ram_bytes;
6409 cache->delalloc_bytes += num_bytes;
6411 spin_unlock(&cache->lock);
6412 spin_unlock(&space_info->lock);
6417 * btrfs_free_reserved_bytes - update the block_group and space info counters
6418 * @cache: The cache we are manipulating
6419 * @num_bytes: The number of bytes in question
6420 * @delalloc: The blocks are allocated for the delalloc write
6422 * This is called by somebody who is freeing space that was never actually used
6423 * on disk. For example if you reserve some space for a new leaf in transaction
6424 * A and before transaction A commits you free that leaf, you call this with
6425 * reserve set to 0 in order to clear the reservation.
6428 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6429 u64 num_bytes, int delalloc)
6431 struct btrfs_space_info *space_info = cache->space_info;
6434 spin_lock(&space_info->lock);
6435 spin_lock(&cache->lock);
6437 space_info->bytes_readonly += num_bytes;
6438 cache->reserved -= num_bytes;
6439 space_info->bytes_reserved -= num_bytes;
6442 cache->delalloc_bytes -= num_bytes;
6443 spin_unlock(&cache->lock);
6444 spin_unlock(&space_info->lock);
6447 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6449 struct btrfs_caching_control *next;
6450 struct btrfs_caching_control *caching_ctl;
6451 struct btrfs_block_group_cache *cache;
6453 down_write(&fs_info->commit_root_sem);
6455 list_for_each_entry_safe(caching_ctl, next,
6456 &fs_info->caching_block_groups, list) {
6457 cache = caching_ctl->block_group;
6458 if (block_group_cache_done(cache)) {
6459 cache->last_byte_to_unpin = (u64)-1;
6460 list_del_init(&caching_ctl->list);
6461 put_caching_control(caching_ctl);
6463 cache->last_byte_to_unpin = caching_ctl->progress;
6467 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6468 fs_info->pinned_extents = &fs_info->freed_extents[1];
6470 fs_info->pinned_extents = &fs_info->freed_extents[0];
6472 up_write(&fs_info->commit_root_sem);
6474 update_global_block_rsv(fs_info);
6478 * Returns the free cluster for the given space info and sets empty_cluster to
6479 * what it should be based on the mount options.
6481 static struct btrfs_free_cluster *
6482 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6483 struct btrfs_space_info *space_info, u64 *empty_cluster)
6485 struct btrfs_free_cluster *ret = NULL;
6488 if (btrfs_mixed_space_info(space_info))
6491 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6492 ret = &fs_info->meta_alloc_cluster;
6493 if (btrfs_test_opt(fs_info, SSD))
6494 *empty_cluster = SZ_2M;
6496 *empty_cluster = SZ_64K;
6497 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6498 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6499 *empty_cluster = SZ_2M;
6500 ret = &fs_info->data_alloc_cluster;
6506 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6508 const bool return_free_space)
6510 struct btrfs_block_group_cache *cache = NULL;
6511 struct btrfs_space_info *space_info;
6512 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6513 struct btrfs_free_cluster *cluster = NULL;
6515 u64 total_unpinned = 0;
6516 u64 empty_cluster = 0;
6519 while (start <= end) {
6522 start >= cache->key.objectid + cache->key.offset) {
6524 btrfs_put_block_group(cache);
6526 cache = btrfs_lookup_block_group(fs_info, start);
6527 BUG_ON(!cache); /* Logic error */
6529 cluster = fetch_cluster_info(fs_info,
6532 empty_cluster <<= 1;
6535 len = cache->key.objectid + cache->key.offset - start;
6536 len = min(len, end + 1 - start);
6538 if (start < cache->last_byte_to_unpin) {
6539 len = min(len, cache->last_byte_to_unpin - start);
6540 if (return_free_space)
6541 btrfs_add_free_space(cache, start, len);
6545 total_unpinned += len;
6546 space_info = cache->space_info;
6549 * If this space cluster has been marked as fragmented and we've
6550 * unpinned enough in this block group to potentially allow a
6551 * cluster to be created inside of it go ahead and clear the
6554 if (cluster && cluster->fragmented &&
6555 total_unpinned > empty_cluster) {
6556 spin_lock(&cluster->lock);
6557 cluster->fragmented = 0;
6558 spin_unlock(&cluster->lock);
6561 spin_lock(&space_info->lock);
6562 spin_lock(&cache->lock);
6563 cache->pinned -= len;
6564 space_info->bytes_pinned -= len;
6566 trace_btrfs_space_reservation(fs_info, "pinned",
6567 space_info->flags, len, 0);
6568 space_info->max_extent_size = 0;
6569 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6570 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6572 space_info->bytes_readonly += len;
6575 spin_unlock(&cache->lock);
6576 if (!readonly && return_free_space &&
6577 global_rsv->space_info == space_info) {
6580 spin_lock(&global_rsv->lock);
6581 if (!global_rsv->full) {
6582 to_add = min(len, global_rsv->size -
6583 global_rsv->reserved);
6584 global_rsv->reserved += to_add;
6585 space_info->bytes_may_use += to_add;
6586 if (global_rsv->reserved >= global_rsv->size)
6587 global_rsv->full = 1;
6588 trace_btrfs_space_reservation(fs_info,
6594 spin_unlock(&global_rsv->lock);
6595 /* Add to any tickets we may have */
6597 space_info_add_new_bytes(fs_info, space_info,
6600 spin_unlock(&space_info->lock);
6604 btrfs_put_block_group(cache);
6608 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6610 struct btrfs_fs_info *fs_info = trans->fs_info;
6611 struct btrfs_block_group_cache *block_group, *tmp;
6612 struct list_head *deleted_bgs;
6613 struct extent_io_tree *unpin;
6618 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6619 unpin = &fs_info->freed_extents[1];
6621 unpin = &fs_info->freed_extents[0];
6623 while (!trans->aborted) {
6624 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6625 ret = find_first_extent_bit(unpin, 0, &start, &end,
6626 EXTENT_DIRTY, NULL);
6628 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6632 if (btrfs_test_opt(fs_info, DISCARD))
6633 ret = btrfs_discard_extent(fs_info, start,
6634 end + 1 - start, NULL);
6636 clear_extent_dirty(unpin, start, end);
6637 unpin_extent_range(fs_info, start, end, true);
6638 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6643 * Transaction is finished. We don't need the lock anymore. We
6644 * do need to clean up the block groups in case of a transaction
6647 deleted_bgs = &trans->transaction->deleted_bgs;
6648 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6652 if (!trans->aborted)
6653 ret = btrfs_discard_extent(fs_info,
6654 block_group->key.objectid,
6655 block_group->key.offset,
6658 list_del_init(&block_group->bg_list);
6659 btrfs_put_block_group_trimming(block_group);
6660 btrfs_put_block_group(block_group);
6663 const char *errstr = btrfs_decode_error(ret);
6665 "discard failed while removing blockgroup: errno=%d %s",
6673 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6674 struct btrfs_delayed_ref_node *node, u64 parent,
6675 u64 root_objectid, u64 owner_objectid,
6676 u64 owner_offset, int refs_to_drop,
6677 struct btrfs_delayed_extent_op *extent_op)
6679 struct btrfs_fs_info *info = trans->fs_info;
6680 struct btrfs_key key;
6681 struct btrfs_path *path;
6682 struct btrfs_root *extent_root = info->extent_root;
6683 struct extent_buffer *leaf;
6684 struct btrfs_extent_item *ei;
6685 struct btrfs_extent_inline_ref *iref;
6688 int extent_slot = 0;
6689 int found_extent = 0;
6693 u64 bytenr = node->bytenr;
6694 u64 num_bytes = node->num_bytes;
6696 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6698 path = btrfs_alloc_path();
6702 path->reada = READA_FORWARD;
6703 path->leave_spinning = 1;
6705 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6706 BUG_ON(!is_data && refs_to_drop != 1);
6709 skinny_metadata = false;
6711 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6712 parent, root_objectid, owner_objectid,
6715 extent_slot = path->slots[0];
6716 while (extent_slot >= 0) {
6717 btrfs_item_key_to_cpu(path->nodes[0], &key,
6719 if (key.objectid != bytenr)
6721 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6722 key.offset == num_bytes) {
6726 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6727 key.offset == owner_objectid) {
6731 if (path->slots[0] - extent_slot > 5)
6736 if (!found_extent) {
6738 ret = remove_extent_backref(trans, path, NULL,
6740 is_data, &last_ref);
6742 btrfs_abort_transaction(trans, ret);
6745 btrfs_release_path(path);
6746 path->leave_spinning = 1;
6748 key.objectid = bytenr;
6749 key.type = BTRFS_EXTENT_ITEM_KEY;
6750 key.offset = num_bytes;
6752 if (!is_data && skinny_metadata) {
6753 key.type = BTRFS_METADATA_ITEM_KEY;
6754 key.offset = owner_objectid;
6757 ret = btrfs_search_slot(trans, extent_root,
6759 if (ret > 0 && skinny_metadata && path->slots[0]) {
6761 * Couldn't find our skinny metadata item,
6762 * see if we have ye olde extent item.
6765 btrfs_item_key_to_cpu(path->nodes[0], &key,
6767 if (key.objectid == bytenr &&
6768 key.type == BTRFS_EXTENT_ITEM_KEY &&
6769 key.offset == num_bytes)
6773 if (ret > 0 && skinny_metadata) {
6774 skinny_metadata = false;
6775 key.objectid = bytenr;
6776 key.type = BTRFS_EXTENT_ITEM_KEY;
6777 key.offset = num_bytes;
6778 btrfs_release_path(path);
6779 ret = btrfs_search_slot(trans, extent_root,
6785 "umm, got %d back from search, was looking for %llu",
6788 btrfs_print_leaf(path->nodes[0]);
6791 btrfs_abort_transaction(trans, ret);
6794 extent_slot = path->slots[0];
6796 } else if (WARN_ON(ret == -ENOENT)) {
6797 btrfs_print_leaf(path->nodes[0]);
6799 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6800 bytenr, parent, root_objectid, owner_objectid,
6802 btrfs_abort_transaction(trans, ret);
6805 btrfs_abort_transaction(trans, ret);
6809 leaf = path->nodes[0];
6810 item_size = btrfs_item_size_nr(leaf, extent_slot);
6811 if (unlikely(item_size < sizeof(*ei))) {
6813 btrfs_print_v0_err(info);
6814 btrfs_abort_transaction(trans, ret);
6817 ei = btrfs_item_ptr(leaf, extent_slot,
6818 struct btrfs_extent_item);
6819 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6820 key.type == BTRFS_EXTENT_ITEM_KEY) {
6821 struct btrfs_tree_block_info *bi;
6822 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6823 bi = (struct btrfs_tree_block_info *)(ei + 1);
6824 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6827 refs = btrfs_extent_refs(leaf, ei);
6828 if (refs < refs_to_drop) {
6830 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
6831 refs_to_drop, refs, bytenr);
6833 btrfs_abort_transaction(trans, ret);
6836 refs -= refs_to_drop;
6840 __run_delayed_extent_op(extent_op, leaf, ei);
6842 * In the case of inline back ref, reference count will
6843 * be updated by remove_extent_backref
6846 BUG_ON(!found_extent);
6848 btrfs_set_extent_refs(leaf, ei, refs);
6849 btrfs_mark_buffer_dirty(leaf);
6852 ret = remove_extent_backref(trans, path, iref,
6853 refs_to_drop, is_data,
6856 btrfs_abort_transaction(trans, ret);
6862 BUG_ON(is_data && refs_to_drop !=
6863 extent_data_ref_count(path, iref));
6865 BUG_ON(path->slots[0] != extent_slot);
6867 BUG_ON(path->slots[0] != extent_slot + 1);
6868 path->slots[0] = extent_slot;
6874 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6877 btrfs_abort_transaction(trans, ret);
6880 btrfs_release_path(path);
6883 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
6885 btrfs_abort_transaction(trans, ret);
6890 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
6892 btrfs_abort_transaction(trans, ret);
6896 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
6898 btrfs_abort_transaction(trans, ret);
6902 btrfs_release_path(path);
6905 btrfs_free_path(path);
6910 * when we free an block, it is possible (and likely) that we free the last
6911 * delayed ref for that extent as well. This searches the delayed ref tree for
6912 * a given extent, and if there are no other delayed refs to be processed, it
6913 * removes it from the tree.
6915 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6918 struct btrfs_delayed_ref_head *head;
6919 struct btrfs_delayed_ref_root *delayed_refs;
6922 delayed_refs = &trans->transaction->delayed_refs;
6923 spin_lock(&delayed_refs->lock);
6924 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
6926 goto out_delayed_unlock;
6928 spin_lock(&head->lock);
6929 if (!RB_EMPTY_ROOT(&head->ref_tree))
6932 if (head->extent_op) {
6933 if (!head->must_insert_reserved)
6935 btrfs_free_delayed_extent_op(head->extent_op);
6936 head->extent_op = NULL;
6940 * waiting for the lock here would deadlock. If someone else has it
6941 * locked they are already in the process of dropping it anyway
6943 if (!mutex_trylock(&head->mutex))
6947 * at this point we have a head with no other entries. Go
6948 * ahead and process it.
6950 rb_erase(&head->href_node, &delayed_refs->href_root);
6951 RB_CLEAR_NODE(&head->href_node);
6952 atomic_dec(&delayed_refs->num_entries);
6955 * we don't take a ref on the node because we're removing it from the
6956 * tree, so we just steal the ref the tree was holding.
6958 delayed_refs->num_heads--;
6959 if (head->processing == 0)
6960 delayed_refs->num_heads_ready--;
6961 head->processing = 0;
6962 spin_unlock(&head->lock);
6963 spin_unlock(&delayed_refs->lock);
6965 BUG_ON(head->extent_op);
6966 if (head->must_insert_reserved)
6969 mutex_unlock(&head->mutex);
6970 btrfs_put_delayed_ref_head(head);
6973 spin_unlock(&head->lock);
6976 spin_unlock(&delayed_refs->lock);
6980 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6981 struct btrfs_root *root,
6982 struct extent_buffer *buf,
6983 u64 parent, int last_ref)
6985 struct btrfs_fs_info *fs_info = root->fs_info;
6989 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6990 int old_ref_mod, new_ref_mod;
6992 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
6993 root->root_key.objectid,
6994 btrfs_header_level(buf), 0,
6995 BTRFS_DROP_DELAYED_REF);
6996 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
6998 root->root_key.objectid,
6999 btrfs_header_level(buf),
7000 BTRFS_DROP_DELAYED_REF, NULL,
7001 &old_ref_mod, &new_ref_mod);
7002 BUG_ON(ret); /* -ENOMEM */
7003 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7006 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7007 struct btrfs_block_group_cache *cache;
7009 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7010 ret = check_ref_cleanup(trans, buf->start);
7016 cache = btrfs_lookup_block_group(fs_info, buf->start);
7018 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7019 pin_down_extent(fs_info, cache, buf->start,
7021 btrfs_put_block_group(cache);
7025 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7027 btrfs_add_free_space(cache, buf->start, buf->len);
7028 btrfs_free_reserved_bytes(cache, buf->len, 0);
7029 btrfs_put_block_group(cache);
7030 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7034 add_pinned_bytes(fs_info, buf->len, true,
7035 root->root_key.objectid);
7039 * Deleting the buffer, clear the corrupt flag since it doesn't
7042 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7046 /* Can return -ENOMEM */
7047 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7048 struct btrfs_root *root,
7049 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7050 u64 owner, u64 offset)
7052 struct btrfs_fs_info *fs_info = root->fs_info;
7053 int old_ref_mod, new_ref_mod;
7056 if (btrfs_is_testing(fs_info))
7059 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7060 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7061 root_objectid, owner, offset,
7062 BTRFS_DROP_DELAYED_REF);
7065 * tree log blocks never actually go into the extent allocation
7066 * tree, just update pinning info and exit early.
7068 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7069 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7070 /* unlocks the pinned mutex */
7071 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7072 old_ref_mod = new_ref_mod = 0;
7074 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7075 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7077 root_objectid, (int)owner,
7078 BTRFS_DROP_DELAYED_REF, NULL,
7079 &old_ref_mod, &new_ref_mod);
7081 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7083 root_objectid, owner, offset,
7084 0, BTRFS_DROP_DELAYED_REF,
7085 &old_ref_mod, &new_ref_mod);
7088 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7089 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7091 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7098 * when we wait for progress in the block group caching, its because
7099 * our allocation attempt failed at least once. So, we must sleep
7100 * and let some progress happen before we try again.
7102 * This function will sleep at least once waiting for new free space to
7103 * show up, and then it will check the block group free space numbers
7104 * for our min num_bytes. Another option is to have it go ahead
7105 * and look in the rbtree for a free extent of a given size, but this
7108 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7109 * any of the information in this block group.
7111 static noinline void
7112 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7115 struct btrfs_caching_control *caching_ctl;
7117 caching_ctl = get_caching_control(cache);
7121 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7122 (cache->free_space_ctl->free_space >= num_bytes));
7124 put_caching_control(caching_ctl);
7128 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7130 struct btrfs_caching_control *caching_ctl;
7133 caching_ctl = get_caching_control(cache);
7135 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7137 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7138 if (cache->cached == BTRFS_CACHE_ERROR)
7140 put_caching_control(caching_ctl);
7144 enum btrfs_loop_type {
7145 LOOP_CACHING_NOWAIT = 0,
7146 LOOP_CACHING_WAIT = 1,
7147 LOOP_ALLOC_CHUNK = 2,
7148 LOOP_NO_EMPTY_SIZE = 3,
7152 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7156 down_read(&cache->data_rwsem);
7160 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7163 btrfs_get_block_group(cache);
7165 down_read(&cache->data_rwsem);
7168 static struct btrfs_block_group_cache *
7169 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7170 struct btrfs_free_cluster *cluster,
7173 struct btrfs_block_group_cache *used_bg = NULL;
7175 spin_lock(&cluster->refill_lock);
7177 used_bg = cluster->block_group;
7181 if (used_bg == block_group)
7184 btrfs_get_block_group(used_bg);
7189 if (down_read_trylock(&used_bg->data_rwsem))
7192 spin_unlock(&cluster->refill_lock);
7194 /* We should only have one-level nested. */
7195 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7197 spin_lock(&cluster->refill_lock);
7198 if (used_bg == cluster->block_group)
7201 up_read(&used_bg->data_rwsem);
7202 btrfs_put_block_group(used_bg);
7207 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7211 up_read(&cache->data_rwsem);
7212 btrfs_put_block_group(cache);
7216 * walks the btree of allocated extents and find a hole of a given size.
7217 * The key ins is changed to record the hole:
7218 * ins->objectid == start position
7219 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7220 * ins->offset == the size of the hole.
7221 * Any available blocks before search_start are skipped.
7223 * If there is no suitable free space, we will record the max size of
7224 * the free space extent currently.
7226 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7227 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7228 u64 hint_byte, struct btrfs_key *ins,
7229 u64 flags, int delalloc)
7232 struct btrfs_root *root = fs_info->extent_root;
7233 struct btrfs_free_cluster *last_ptr = NULL;
7234 struct btrfs_block_group_cache *block_group = NULL;
7235 u64 search_start = 0;
7236 u64 max_extent_size = 0;
7237 u64 empty_cluster = 0;
7238 struct btrfs_space_info *space_info;
7240 int index = btrfs_bg_flags_to_raid_index(flags);
7241 bool failed_cluster_refill = false;
7242 bool failed_alloc = false;
7243 bool use_cluster = true;
7244 bool have_caching_bg = false;
7245 bool orig_have_caching_bg = false;
7246 bool full_search = false;
7248 WARN_ON(num_bytes < fs_info->sectorsize);
7249 ins->type = BTRFS_EXTENT_ITEM_KEY;
7253 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7255 space_info = __find_space_info(fs_info, flags);
7257 btrfs_err(fs_info, "No space info for %llu", flags);
7262 * If our free space is heavily fragmented we may not be able to make
7263 * big contiguous allocations, so instead of doing the expensive search
7264 * for free space, simply return ENOSPC with our max_extent_size so we
7265 * can go ahead and search for a more manageable chunk.
7267 * If our max_extent_size is large enough for our allocation simply
7268 * disable clustering since we will likely not be able to find enough
7269 * space to create a cluster and induce latency trying.
7271 if (unlikely(space_info->max_extent_size)) {
7272 spin_lock(&space_info->lock);
7273 if (space_info->max_extent_size &&
7274 num_bytes > space_info->max_extent_size) {
7275 ins->offset = space_info->max_extent_size;
7276 spin_unlock(&space_info->lock);
7278 } else if (space_info->max_extent_size) {
7279 use_cluster = false;
7281 spin_unlock(&space_info->lock);
7284 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7286 spin_lock(&last_ptr->lock);
7287 if (last_ptr->block_group)
7288 hint_byte = last_ptr->window_start;
7289 if (last_ptr->fragmented) {
7291 * We still set window_start so we can keep track of the
7292 * last place we found an allocation to try and save
7295 hint_byte = last_ptr->window_start;
7296 use_cluster = false;
7298 spin_unlock(&last_ptr->lock);
7301 search_start = max(search_start, first_logical_byte(fs_info, 0));
7302 search_start = max(search_start, hint_byte);
7303 if (search_start == hint_byte) {
7304 block_group = btrfs_lookup_block_group(fs_info, search_start);
7306 * we don't want to use the block group if it doesn't match our
7307 * allocation bits, or if its not cached.
7309 * However if we are re-searching with an ideal block group
7310 * picked out then we don't care that the block group is cached.
7312 if (block_group && block_group_bits(block_group, flags) &&
7313 block_group->cached != BTRFS_CACHE_NO) {
7314 down_read(&space_info->groups_sem);
7315 if (list_empty(&block_group->list) ||
7318 * someone is removing this block group,
7319 * we can't jump into the have_block_group
7320 * target because our list pointers are not
7323 btrfs_put_block_group(block_group);
7324 up_read(&space_info->groups_sem);
7326 index = btrfs_bg_flags_to_raid_index(
7327 block_group->flags);
7328 btrfs_lock_block_group(block_group, delalloc);
7329 goto have_block_group;
7331 } else if (block_group) {
7332 btrfs_put_block_group(block_group);
7336 have_caching_bg = false;
7337 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7339 down_read(&space_info->groups_sem);
7340 list_for_each_entry(block_group, &space_info->block_groups[index],
7345 /* If the block group is read-only, we can skip it entirely. */
7346 if (unlikely(block_group->ro))
7349 btrfs_grab_block_group(block_group, delalloc);
7350 search_start = block_group->key.objectid;
7353 * this can happen if we end up cycling through all the
7354 * raid types, but we want to make sure we only allocate
7355 * for the proper type.
7357 if (!block_group_bits(block_group, flags)) {
7358 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7359 BTRFS_BLOCK_GROUP_RAID1 |
7360 BTRFS_BLOCK_GROUP_RAID5 |
7361 BTRFS_BLOCK_GROUP_RAID6 |
7362 BTRFS_BLOCK_GROUP_RAID10;
7365 * if they asked for extra copies and this block group
7366 * doesn't provide them, bail. This does allow us to
7367 * fill raid0 from raid1.
7369 if ((flags & extra) && !(block_group->flags & extra))
7374 cached = block_group_cache_done(block_group);
7375 if (unlikely(!cached)) {
7376 have_caching_bg = true;
7377 ret = cache_block_group(block_group, 0);
7382 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7386 * Ok we want to try and use the cluster allocator, so
7389 if (last_ptr && use_cluster) {
7390 struct btrfs_block_group_cache *used_block_group;
7391 unsigned long aligned_cluster;
7393 * the refill lock keeps out other
7394 * people trying to start a new cluster
7396 used_block_group = btrfs_lock_cluster(block_group,
7399 if (!used_block_group)
7400 goto refill_cluster;
7402 if (used_block_group != block_group &&
7403 (used_block_group->ro ||
7404 !block_group_bits(used_block_group, flags)))
7405 goto release_cluster;
7407 offset = btrfs_alloc_from_cluster(used_block_group,
7410 used_block_group->key.objectid,
7413 /* we have a block, we're done */
7414 spin_unlock(&last_ptr->refill_lock);
7415 trace_btrfs_reserve_extent_cluster(
7417 search_start, num_bytes);
7418 if (used_block_group != block_group) {
7419 btrfs_release_block_group(block_group,
7421 block_group = used_block_group;
7426 WARN_ON(last_ptr->block_group != used_block_group);
7428 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7429 * set up a new clusters, so lets just skip it
7430 * and let the allocator find whatever block
7431 * it can find. If we reach this point, we
7432 * will have tried the cluster allocator
7433 * plenty of times and not have found
7434 * anything, so we are likely way too
7435 * fragmented for the clustering stuff to find
7438 * However, if the cluster is taken from the
7439 * current block group, release the cluster
7440 * first, so that we stand a better chance of
7441 * succeeding in the unclustered
7443 if (loop >= LOOP_NO_EMPTY_SIZE &&
7444 used_block_group != block_group) {
7445 spin_unlock(&last_ptr->refill_lock);
7446 btrfs_release_block_group(used_block_group,
7448 goto unclustered_alloc;
7452 * this cluster didn't work out, free it and
7455 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7457 if (used_block_group != block_group)
7458 btrfs_release_block_group(used_block_group,
7461 if (loop >= LOOP_NO_EMPTY_SIZE) {
7462 spin_unlock(&last_ptr->refill_lock);
7463 goto unclustered_alloc;
7466 aligned_cluster = max_t(unsigned long,
7467 empty_cluster + empty_size,
7468 block_group->full_stripe_len);
7470 /* allocate a cluster in this block group */
7471 ret = btrfs_find_space_cluster(fs_info, block_group,
7472 last_ptr, search_start,
7477 * now pull our allocation out of this
7480 offset = btrfs_alloc_from_cluster(block_group,
7486 /* we found one, proceed */
7487 spin_unlock(&last_ptr->refill_lock);
7488 trace_btrfs_reserve_extent_cluster(
7489 block_group, search_start,
7493 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7494 && !failed_cluster_refill) {
7495 spin_unlock(&last_ptr->refill_lock);
7497 failed_cluster_refill = true;
7498 wait_block_group_cache_progress(block_group,
7499 num_bytes + empty_cluster + empty_size);
7500 goto have_block_group;
7504 * at this point we either didn't find a cluster
7505 * or we weren't able to allocate a block from our
7506 * cluster. Free the cluster we've been trying
7507 * to use, and go to the next block group
7509 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7510 spin_unlock(&last_ptr->refill_lock);
7516 * We are doing an unclustered alloc, set the fragmented flag so
7517 * we don't bother trying to setup a cluster again until we get
7520 if (unlikely(last_ptr)) {
7521 spin_lock(&last_ptr->lock);
7522 last_ptr->fragmented = 1;
7523 spin_unlock(&last_ptr->lock);
7526 struct btrfs_free_space_ctl *ctl =
7527 block_group->free_space_ctl;
7529 spin_lock(&ctl->tree_lock);
7530 if (ctl->free_space <
7531 num_bytes + empty_cluster + empty_size) {
7532 if (ctl->free_space > max_extent_size)
7533 max_extent_size = ctl->free_space;
7534 spin_unlock(&ctl->tree_lock);
7537 spin_unlock(&ctl->tree_lock);
7540 offset = btrfs_find_space_for_alloc(block_group, search_start,
7541 num_bytes, empty_size,
7544 * If we didn't find a chunk, and we haven't failed on this
7545 * block group before, and this block group is in the middle of
7546 * caching and we are ok with waiting, then go ahead and wait
7547 * for progress to be made, and set failed_alloc to true.
7549 * If failed_alloc is true then we've already waited on this
7550 * block group once and should move on to the next block group.
7552 if (!offset && !failed_alloc && !cached &&
7553 loop > LOOP_CACHING_NOWAIT) {
7554 wait_block_group_cache_progress(block_group,
7555 num_bytes + empty_size);
7556 failed_alloc = true;
7557 goto have_block_group;
7558 } else if (!offset) {
7562 search_start = round_up(offset, fs_info->stripesize);
7564 /* move on to the next group */
7565 if (search_start + num_bytes >
7566 block_group->key.objectid + block_group->key.offset) {
7567 btrfs_add_free_space(block_group, offset, num_bytes);
7571 if (offset < search_start)
7572 btrfs_add_free_space(block_group, offset,
7573 search_start - offset);
7575 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7576 num_bytes, delalloc);
7577 if (ret == -EAGAIN) {
7578 btrfs_add_free_space(block_group, offset, num_bytes);
7581 btrfs_inc_block_group_reservations(block_group);
7583 /* we are all good, lets return */
7584 ins->objectid = search_start;
7585 ins->offset = num_bytes;
7587 trace_btrfs_reserve_extent(block_group, search_start, num_bytes);
7588 btrfs_release_block_group(block_group, delalloc);
7591 failed_cluster_refill = false;
7592 failed_alloc = false;
7593 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7595 btrfs_release_block_group(block_group, delalloc);
7598 up_read(&space_info->groups_sem);
7600 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7601 && !orig_have_caching_bg)
7602 orig_have_caching_bg = true;
7604 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7607 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7611 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7612 * caching kthreads as we move along
7613 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7614 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7615 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7618 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7620 if (loop == LOOP_CACHING_NOWAIT) {
7622 * We want to skip the LOOP_CACHING_WAIT step if we
7623 * don't have any uncached bgs and we've already done a
7624 * full search through.
7626 if (orig_have_caching_bg || !full_search)
7627 loop = LOOP_CACHING_WAIT;
7629 loop = LOOP_ALLOC_CHUNK;
7634 if (loop == LOOP_ALLOC_CHUNK) {
7635 struct btrfs_trans_handle *trans;
7638 trans = current->journal_info;
7642 trans = btrfs_join_transaction(root);
7644 if (IS_ERR(trans)) {
7645 ret = PTR_ERR(trans);
7649 ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE);
7652 * If we can't allocate a new chunk we've already looped
7653 * through at least once, move on to the NO_EMPTY_SIZE
7657 loop = LOOP_NO_EMPTY_SIZE;
7660 * Do not bail out on ENOSPC since we
7661 * can do more things.
7663 if (ret < 0 && ret != -ENOSPC)
7664 btrfs_abort_transaction(trans, ret);
7668 btrfs_end_transaction(trans);
7673 if (loop == LOOP_NO_EMPTY_SIZE) {
7675 * Don't loop again if we already have no empty_size and
7678 if (empty_size == 0 &&
7679 empty_cluster == 0) {
7688 } else if (!ins->objectid) {
7690 } else if (ins->objectid) {
7691 if (!use_cluster && last_ptr) {
7692 spin_lock(&last_ptr->lock);
7693 last_ptr->window_start = ins->objectid;
7694 spin_unlock(&last_ptr->lock);
7699 if (ret == -ENOSPC) {
7700 spin_lock(&space_info->lock);
7701 space_info->max_extent_size = max_extent_size;
7702 spin_unlock(&space_info->lock);
7703 ins->offset = max_extent_size;
7708 static void dump_space_info(struct btrfs_fs_info *fs_info,
7709 struct btrfs_space_info *info, u64 bytes,
7710 int dump_block_groups)
7712 struct btrfs_block_group_cache *cache;
7715 spin_lock(&info->lock);
7716 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7718 info->total_bytes - btrfs_space_info_used(info, true),
7719 info->full ? "" : "not ");
7721 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7722 info->total_bytes, info->bytes_used, info->bytes_pinned,
7723 info->bytes_reserved, info->bytes_may_use,
7724 info->bytes_readonly);
7725 spin_unlock(&info->lock);
7727 if (!dump_block_groups)
7730 down_read(&info->groups_sem);
7732 list_for_each_entry(cache, &info->block_groups[index], list) {
7733 spin_lock(&cache->lock);
7735 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7736 cache->key.objectid, cache->key.offset,
7737 btrfs_block_group_used(&cache->item), cache->pinned,
7738 cache->reserved, cache->ro ? "[readonly]" : "");
7739 btrfs_dump_free_space(cache, bytes);
7740 spin_unlock(&cache->lock);
7742 if (++index < BTRFS_NR_RAID_TYPES)
7744 up_read(&info->groups_sem);
7748 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7749 * hole that is at least as big as @num_bytes.
7751 * @root - The root that will contain this extent
7753 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7754 * is used for accounting purposes. This value differs
7755 * from @num_bytes only in the case of compressed extents.
7757 * @num_bytes - Number of bytes to allocate on-disk.
7759 * @min_alloc_size - Indicates the minimum amount of space that the
7760 * allocator should try to satisfy. In some cases
7761 * @num_bytes may be larger than what is required and if
7762 * the filesystem is fragmented then allocation fails.
7763 * However, the presence of @min_alloc_size gives a
7764 * chance to try and satisfy the smaller allocation.
7766 * @empty_size - A hint that you plan on doing more COW. This is the
7767 * size in bytes the allocator should try to find free
7768 * next to the block it returns. This is just a hint and
7769 * may be ignored by the allocator.
7771 * @hint_byte - Hint to the allocator to start searching above the byte
7772 * address passed. It might be ignored.
7774 * @ins - This key is modified to record the found hole. It will
7775 * have the following values:
7776 * ins->objectid == start position
7777 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7778 * ins->offset == the size of the hole.
7780 * @is_data - Boolean flag indicating whether an extent is
7781 * allocated for data (true) or metadata (false)
7783 * @delalloc - Boolean flag indicating whether this allocation is for
7784 * delalloc or not. If 'true' data_rwsem of block groups
7785 * is going to be acquired.
7788 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7789 * case -ENOSPC is returned then @ins->offset will contain the size of the
7790 * largest available hole the allocator managed to find.
7792 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7793 u64 num_bytes, u64 min_alloc_size,
7794 u64 empty_size, u64 hint_byte,
7795 struct btrfs_key *ins, int is_data, int delalloc)
7797 struct btrfs_fs_info *fs_info = root->fs_info;
7798 bool final_tried = num_bytes == min_alloc_size;
7802 flags = get_alloc_profile_by_root(root, is_data);
7804 WARN_ON(num_bytes < fs_info->sectorsize);
7805 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7806 hint_byte, ins, flags, delalloc);
7807 if (!ret && !is_data) {
7808 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7809 } else if (ret == -ENOSPC) {
7810 if (!final_tried && ins->offset) {
7811 num_bytes = min(num_bytes >> 1, ins->offset);
7812 num_bytes = round_down(num_bytes,
7813 fs_info->sectorsize);
7814 num_bytes = max(num_bytes, min_alloc_size);
7815 ram_bytes = num_bytes;
7816 if (num_bytes == min_alloc_size)
7819 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7820 struct btrfs_space_info *sinfo;
7822 sinfo = __find_space_info(fs_info, flags);
7824 "allocation failed flags %llu, wanted %llu",
7827 dump_space_info(fs_info, sinfo, num_bytes, 1);
7834 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
7836 int pin, int delalloc)
7838 struct btrfs_block_group_cache *cache;
7841 cache = btrfs_lookup_block_group(fs_info, start);
7843 btrfs_err(fs_info, "Unable to find block group for %llu",
7849 pin_down_extent(fs_info, cache, start, len, 1);
7851 if (btrfs_test_opt(fs_info, DISCARD))
7852 ret = btrfs_discard_extent(fs_info, start, len, NULL);
7853 btrfs_add_free_space(cache, start, len);
7854 btrfs_free_reserved_bytes(cache, len, delalloc);
7855 trace_btrfs_reserved_extent_free(fs_info, start, len);
7858 btrfs_put_block_group(cache);
7862 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
7863 u64 start, u64 len, int delalloc)
7865 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
7868 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
7871 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
7874 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7875 u64 parent, u64 root_objectid,
7876 u64 flags, u64 owner, u64 offset,
7877 struct btrfs_key *ins, int ref_mod)
7879 struct btrfs_fs_info *fs_info = trans->fs_info;
7881 struct btrfs_extent_item *extent_item;
7882 struct btrfs_extent_inline_ref *iref;
7883 struct btrfs_path *path;
7884 struct extent_buffer *leaf;
7889 type = BTRFS_SHARED_DATA_REF_KEY;
7891 type = BTRFS_EXTENT_DATA_REF_KEY;
7893 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7895 path = btrfs_alloc_path();
7899 path->leave_spinning = 1;
7900 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7903 btrfs_free_path(path);
7907 leaf = path->nodes[0];
7908 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7909 struct btrfs_extent_item);
7910 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7911 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7912 btrfs_set_extent_flags(leaf, extent_item,
7913 flags | BTRFS_EXTENT_FLAG_DATA);
7915 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7916 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7918 struct btrfs_shared_data_ref *ref;
7919 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7920 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7921 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7923 struct btrfs_extent_data_ref *ref;
7924 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7925 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7926 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7927 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7928 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7931 btrfs_mark_buffer_dirty(path->nodes[0]);
7932 btrfs_free_path(path);
7934 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
7938 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
7939 if (ret) { /* -ENOENT, logic error */
7940 btrfs_err(fs_info, "update block group failed for %llu %llu",
7941 ins->objectid, ins->offset);
7944 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
7948 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7949 struct btrfs_delayed_ref_node *node,
7950 struct btrfs_delayed_extent_op *extent_op)
7952 struct btrfs_fs_info *fs_info = trans->fs_info;
7954 struct btrfs_extent_item *extent_item;
7955 struct btrfs_key extent_key;
7956 struct btrfs_tree_block_info *block_info;
7957 struct btrfs_extent_inline_ref *iref;
7958 struct btrfs_path *path;
7959 struct extent_buffer *leaf;
7960 struct btrfs_delayed_tree_ref *ref;
7961 u32 size = sizeof(*extent_item) + sizeof(*iref);
7963 u64 flags = extent_op->flags_to_set;
7964 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
7966 ref = btrfs_delayed_node_to_tree_ref(node);
7968 extent_key.objectid = node->bytenr;
7969 if (skinny_metadata) {
7970 extent_key.offset = ref->level;
7971 extent_key.type = BTRFS_METADATA_ITEM_KEY;
7972 num_bytes = fs_info->nodesize;
7974 extent_key.offset = node->num_bytes;
7975 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7976 size += sizeof(*block_info);
7977 num_bytes = node->num_bytes;
7980 path = btrfs_alloc_path();
7982 btrfs_free_and_pin_reserved_extent(fs_info,
7983 extent_key.objectid,
7988 path->leave_spinning = 1;
7989 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7992 btrfs_free_path(path);
7993 btrfs_free_and_pin_reserved_extent(fs_info,
7994 extent_key.objectid,
7999 leaf = path->nodes[0];
8000 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8001 struct btrfs_extent_item);
8002 btrfs_set_extent_refs(leaf, extent_item, 1);
8003 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8004 btrfs_set_extent_flags(leaf, extent_item,
8005 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8007 if (skinny_metadata) {
8008 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8010 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8011 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8012 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8013 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8016 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8017 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8018 btrfs_set_extent_inline_ref_type(leaf, iref,
8019 BTRFS_SHARED_BLOCK_REF_KEY);
8020 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8022 btrfs_set_extent_inline_ref_type(leaf, iref,
8023 BTRFS_TREE_BLOCK_REF_KEY);
8024 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8027 btrfs_mark_buffer_dirty(leaf);
8028 btrfs_free_path(path);
8030 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8035 ret = update_block_group(trans, fs_info, extent_key.objectid,
8036 fs_info->nodesize, 1);
8037 if (ret) { /* -ENOENT, logic error */
8038 btrfs_err(fs_info, "update block group failed for %llu %llu",
8039 extent_key.objectid, extent_key.offset);
8043 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8048 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8049 struct btrfs_root *root, u64 owner,
8050 u64 offset, u64 ram_bytes,
8051 struct btrfs_key *ins)
8055 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8057 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8058 root->root_key.objectid, owner, offset,
8059 BTRFS_ADD_DELAYED_EXTENT);
8061 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8063 root->root_key.objectid, owner,
8065 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8070 * this is used by the tree logging recovery code. It records that
8071 * an extent has been allocated and makes sure to clear the free
8072 * space cache bits as well
8074 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8075 u64 root_objectid, u64 owner, u64 offset,
8076 struct btrfs_key *ins)
8078 struct btrfs_fs_info *fs_info = trans->fs_info;
8080 struct btrfs_block_group_cache *block_group;
8081 struct btrfs_space_info *space_info;
8084 * Mixed block groups will exclude before processing the log so we only
8085 * need to do the exclude dance if this fs isn't mixed.
8087 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8088 ret = __exclude_logged_extent(fs_info, ins->objectid,
8094 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8098 space_info = block_group->space_info;
8099 spin_lock(&space_info->lock);
8100 spin_lock(&block_group->lock);
8101 space_info->bytes_reserved += ins->offset;
8102 block_group->reserved += ins->offset;
8103 spin_unlock(&block_group->lock);
8104 spin_unlock(&space_info->lock);
8106 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8108 btrfs_put_block_group(block_group);
8112 static struct extent_buffer *
8113 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8114 u64 bytenr, int level, u64 owner)
8116 struct btrfs_fs_info *fs_info = root->fs_info;
8117 struct extent_buffer *buf;
8119 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8123 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8124 btrfs_tree_lock(buf);
8125 clean_tree_block(fs_info, buf);
8126 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8128 btrfs_set_lock_blocking(buf);
8129 set_extent_buffer_uptodate(buf);
8131 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8132 btrfs_set_header_level(buf, level);
8133 btrfs_set_header_bytenr(buf, buf->start);
8134 btrfs_set_header_generation(buf, trans->transid);
8135 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8136 btrfs_set_header_owner(buf, owner);
8137 write_extent_buffer_fsid(buf, fs_info->fsid);
8138 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8139 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8140 buf->log_index = root->log_transid % 2;
8142 * we allow two log transactions at a time, use different
8143 * EXENT bit to differentiate dirty pages.
8145 if (buf->log_index == 0)
8146 set_extent_dirty(&root->dirty_log_pages, buf->start,
8147 buf->start + buf->len - 1, GFP_NOFS);
8149 set_extent_new(&root->dirty_log_pages, buf->start,
8150 buf->start + buf->len - 1);
8152 buf->log_index = -1;
8153 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8154 buf->start + buf->len - 1, GFP_NOFS);
8156 trans->dirty = true;
8157 /* this returns a buffer locked for blocking */
8161 static struct btrfs_block_rsv *
8162 use_block_rsv(struct btrfs_trans_handle *trans,
8163 struct btrfs_root *root, u32 blocksize)
8165 struct btrfs_fs_info *fs_info = root->fs_info;
8166 struct btrfs_block_rsv *block_rsv;
8167 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8169 bool global_updated = false;
8171 block_rsv = get_block_rsv(trans, root);
8173 if (unlikely(block_rsv->size == 0))
8176 ret = block_rsv_use_bytes(block_rsv, blocksize);
8180 if (block_rsv->failfast)
8181 return ERR_PTR(ret);
8183 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8184 global_updated = true;
8185 update_global_block_rsv(fs_info);
8189 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8190 static DEFINE_RATELIMIT_STATE(_rs,
8191 DEFAULT_RATELIMIT_INTERVAL * 10,
8192 /*DEFAULT_RATELIMIT_BURST*/ 1);
8193 if (__ratelimit(&_rs))
8195 "BTRFS: block rsv returned %d\n", ret);
8198 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8199 BTRFS_RESERVE_NO_FLUSH);
8203 * If we couldn't reserve metadata bytes try and use some from
8204 * the global reserve if its space type is the same as the global
8207 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8208 block_rsv->space_info == global_rsv->space_info) {
8209 ret = block_rsv_use_bytes(global_rsv, blocksize);
8213 return ERR_PTR(ret);
8216 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8217 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8219 block_rsv_add_bytes(block_rsv, blocksize, 0);
8220 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8224 * finds a free extent and does all the dirty work required for allocation
8225 * returns the tree buffer or an ERR_PTR on error.
8227 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8228 struct btrfs_root *root,
8229 u64 parent, u64 root_objectid,
8230 const struct btrfs_disk_key *key,
8231 int level, u64 hint,
8234 struct btrfs_fs_info *fs_info = root->fs_info;
8235 struct btrfs_key ins;
8236 struct btrfs_block_rsv *block_rsv;
8237 struct extent_buffer *buf;
8238 struct btrfs_delayed_extent_op *extent_op;
8241 u32 blocksize = fs_info->nodesize;
8242 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8245 if (btrfs_is_testing(fs_info)) {
8246 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8247 level, root_objectid);
8249 root->alloc_bytenr += blocksize;
8254 block_rsv = use_block_rsv(trans, root, blocksize);
8255 if (IS_ERR(block_rsv))
8256 return ERR_CAST(block_rsv);
8258 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8259 empty_size, hint, &ins, 0, 0);
8263 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8267 goto out_free_reserved;
8270 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8272 parent = ins.objectid;
8273 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8277 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8278 extent_op = btrfs_alloc_delayed_extent_op();
8284 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8286 memset(&extent_op->key, 0, sizeof(extent_op->key));
8287 extent_op->flags_to_set = flags;
8288 extent_op->update_key = skinny_metadata ? false : true;
8289 extent_op->update_flags = true;
8290 extent_op->is_data = false;
8291 extent_op->level = level;
8293 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8294 root_objectid, level, 0,
8295 BTRFS_ADD_DELAYED_EXTENT);
8296 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8298 root_objectid, level,
8299 BTRFS_ADD_DELAYED_EXTENT,
8300 extent_op, NULL, NULL);
8302 goto out_free_delayed;
8307 btrfs_free_delayed_extent_op(extent_op);
8309 free_extent_buffer(buf);
8311 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8313 unuse_block_rsv(fs_info, block_rsv, blocksize);
8314 return ERR_PTR(ret);
8317 struct walk_control {
8318 u64 refs[BTRFS_MAX_LEVEL];
8319 u64 flags[BTRFS_MAX_LEVEL];
8320 struct btrfs_key update_progress;
8330 #define DROP_REFERENCE 1
8331 #define UPDATE_BACKREF 2
8333 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8334 struct btrfs_root *root,
8335 struct walk_control *wc,
8336 struct btrfs_path *path)
8338 struct btrfs_fs_info *fs_info = root->fs_info;
8344 struct btrfs_key key;
8345 struct extent_buffer *eb;
8350 if (path->slots[wc->level] < wc->reada_slot) {
8351 wc->reada_count = wc->reada_count * 2 / 3;
8352 wc->reada_count = max(wc->reada_count, 2);
8354 wc->reada_count = wc->reada_count * 3 / 2;
8355 wc->reada_count = min_t(int, wc->reada_count,
8356 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8359 eb = path->nodes[wc->level];
8360 nritems = btrfs_header_nritems(eb);
8362 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8363 if (nread >= wc->reada_count)
8367 bytenr = btrfs_node_blockptr(eb, slot);
8368 generation = btrfs_node_ptr_generation(eb, slot);
8370 if (slot == path->slots[wc->level])
8373 if (wc->stage == UPDATE_BACKREF &&
8374 generation <= root->root_key.offset)
8377 /* We don't lock the tree block, it's OK to be racy here */
8378 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8379 wc->level - 1, 1, &refs,
8381 /* We don't care about errors in readahead. */
8386 if (wc->stage == DROP_REFERENCE) {
8390 if (wc->level == 1 &&
8391 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8393 if (!wc->update_ref ||
8394 generation <= root->root_key.offset)
8396 btrfs_node_key_to_cpu(eb, &key, slot);
8397 ret = btrfs_comp_cpu_keys(&key,
8398 &wc->update_progress);
8402 if (wc->level == 1 &&
8403 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8407 readahead_tree_block(fs_info, bytenr);
8410 wc->reada_slot = slot;
8414 * helper to process tree block while walking down the tree.
8416 * when wc->stage == UPDATE_BACKREF, this function updates
8417 * back refs for pointers in the block.
8419 * NOTE: return value 1 means we should stop walking down.
8421 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8422 struct btrfs_root *root,
8423 struct btrfs_path *path,
8424 struct walk_control *wc, int lookup_info)
8426 struct btrfs_fs_info *fs_info = root->fs_info;
8427 int level = wc->level;
8428 struct extent_buffer *eb = path->nodes[level];
8429 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8432 if (wc->stage == UPDATE_BACKREF &&
8433 btrfs_header_owner(eb) != root->root_key.objectid)
8437 * when reference count of tree block is 1, it won't increase
8438 * again. once full backref flag is set, we never clear it.
8441 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8442 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8443 BUG_ON(!path->locks[level]);
8444 ret = btrfs_lookup_extent_info(trans, fs_info,
8445 eb->start, level, 1,
8448 BUG_ON(ret == -ENOMEM);
8451 BUG_ON(wc->refs[level] == 0);
8454 if (wc->stage == DROP_REFERENCE) {
8455 if (wc->refs[level] > 1)
8458 if (path->locks[level] && !wc->keep_locks) {
8459 btrfs_tree_unlock_rw(eb, path->locks[level]);
8460 path->locks[level] = 0;
8465 /* wc->stage == UPDATE_BACKREF */
8466 if (!(wc->flags[level] & flag)) {
8467 BUG_ON(!path->locks[level]);
8468 ret = btrfs_inc_ref(trans, root, eb, 1);
8469 BUG_ON(ret); /* -ENOMEM */
8470 ret = btrfs_dec_ref(trans, root, eb, 0);
8471 BUG_ON(ret); /* -ENOMEM */
8472 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8474 btrfs_header_level(eb), 0);
8475 BUG_ON(ret); /* -ENOMEM */
8476 wc->flags[level] |= flag;
8480 * the block is shared by multiple trees, so it's not good to
8481 * keep the tree lock
8483 if (path->locks[level] && level > 0) {
8484 btrfs_tree_unlock_rw(eb, path->locks[level]);
8485 path->locks[level] = 0;
8491 * helper to process tree block pointer.
8493 * when wc->stage == DROP_REFERENCE, this function checks
8494 * reference count of the block pointed to. if the block
8495 * is shared and we need update back refs for the subtree
8496 * rooted at the block, this function changes wc->stage to
8497 * UPDATE_BACKREF. if the block is shared and there is no
8498 * need to update back, this function drops the reference
8501 * NOTE: return value 1 means we should stop walking down.
8503 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8504 struct btrfs_root *root,
8505 struct btrfs_path *path,
8506 struct walk_control *wc, int *lookup_info)
8508 struct btrfs_fs_info *fs_info = root->fs_info;
8513 struct btrfs_key key;
8514 struct btrfs_key first_key;
8515 struct extent_buffer *next;
8516 int level = wc->level;
8519 bool need_account = false;
8521 generation = btrfs_node_ptr_generation(path->nodes[level],
8522 path->slots[level]);
8524 * if the lower level block was created before the snapshot
8525 * was created, we know there is no need to update back refs
8528 if (wc->stage == UPDATE_BACKREF &&
8529 generation <= root->root_key.offset) {
8534 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8535 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8536 path->slots[level]);
8537 blocksize = fs_info->nodesize;
8539 next = find_extent_buffer(fs_info, bytenr);
8541 next = btrfs_find_create_tree_block(fs_info, bytenr);
8543 return PTR_ERR(next);
8545 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8549 btrfs_tree_lock(next);
8550 btrfs_set_lock_blocking(next);
8552 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8553 &wc->refs[level - 1],
8554 &wc->flags[level - 1]);
8558 if (unlikely(wc->refs[level - 1] == 0)) {
8559 btrfs_err(fs_info, "Missing references.");
8565 if (wc->stage == DROP_REFERENCE) {
8566 if (wc->refs[level - 1] > 1) {
8567 need_account = true;
8569 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8572 if (!wc->update_ref ||
8573 generation <= root->root_key.offset)
8576 btrfs_node_key_to_cpu(path->nodes[level], &key,
8577 path->slots[level]);
8578 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8582 wc->stage = UPDATE_BACKREF;
8583 wc->shared_level = level - 1;
8587 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8591 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8592 btrfs_tree_unlock(next);
8593 free_extent_buffer(next);
8599 if (reada && level == 1)
8600 reada_walk_down(trans, root, wc, path);
8601 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8604 return PTR_ERR(next);
8605 } else if (!extent_buffer_uptodate(next)) {
8606 free_extent_buffer(next);
8609 btrfs_tree_lock(next);
8610 btrfs_set_lock_blocking(next);
8614 ASSERT(level == btrfs_header_level(next));
8615 if (level != btrfs_header_level(next)) {
8616 btrfs_err(root->fs_info, "mismatched level");
8620 path->nodes[level] = next;
8621 path->slots[level] = 0;
8622 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8628 wc->refs[level - 1] = 0;
8629 wc->flags[level - 1] = 0;
8630 if (wc->stage == DROP_REFERENCE) {
8631 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8632 parent = path->nodes[level]->start;
8634 ASSERT(root->root_key.objectid ==
8635 btrfs_header_owner(path->nodes[level]));
8636 if (root->root_key.objectid !=
8637 btrfs_header_owner(path->nodes[level])) {
8638 btrfs_err(root->fs_info,
8639 "mismatched block owner");
8647 ret = btrfs_qgroup_trace_subtree(trans, next,
8648 generation, level - 1);
8650 btrfs_err_rl(fs_info,
8651 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8655 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8656 parent, root->root_key.objectid,
8666 btrfs_tree_unlock(next);
8667 free_extent_buffer(next);
8673 * helper to process tree block while walking up the tree.
8675 * when wc->stage == DROP_REFERENCE, this function drops
8676 * reference count on the block.
8678 * when wc->stage == UPDATE_BACKREF, this function changes
8679 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8680 * to UPDATE_BACKREF previously while processing the block.
8682 * NOTE: return value 1 means we should stop walking up.
8684 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8685 struct btrfs_root *root,
8686 struct btrfs_path *path,
8687 struct walk_control *wc)
8689 struct btrfs_fs_info *fs_info = root->fs_info;
8691 int level = wc->level;
8692 struct extent_buffer *eb = path->nodes[level];
8695 if (wc->stage == UPDATE_BACKREF) {
8696 BUG_ON(wc->shared_level < level);
8697 if (level < wc->shared_level)
8700 ret = find_next_key(path, level + 1, &wc->update_progress);
8704 wc->stage = DROP_REFERENCE;
8705 wc->shared_level = -1;
8706 path->slots[level] = 0;
8709 * check reference count again if the block isn't locked.
8710 * we should start walking down the tree again if reference
8713 if (!path->locks[level]) {
8715 btrfs_tree_lock(eb);
8716 btrfs_set_lock_blocking(eb);
8717 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8719 ret = btrfs_lookup_extent_info(trans, fs_info,
8720 eb->start, level, 1,
8724 btrfs_tree_unlock_rw(eb, path->locks[level]);
8725 path->locks[level] = 0;
8728 BUG_ON(wc->refs[level] == 0);
8729 if (wc->refs[level] == 1) {
8730 btrfs_tree_unlock_rw(eb, path->locks[level]);
8731 path->locks[level] = 0;
8737 /* wc->stage == DROP_REFERENCE */
8738 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8740 if (wc->refs[level] == 1) {
8742 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8743 ret = btrfs_dec_ref(trans, root, eb, 1);
8745 ret = btrfs_dec_ref(trans, root, eb, 0);
8746 BUG_ON(ret); /* -ENOMEM */
8747 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
8749 btrfs_err_rl(fs_info,
8750 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8754 /* make block locked assertion in clean_tree_block happy */
8755 if (!path->locks[level] &&
8756 btrfs_header_generation(eb) == trans->transid) {
8757 btrfs_tree_lock(eb);
8758 btrfs_set_lock_blocking(eb);
8759 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8761 clean_tree_block(fs_info, eb);
8764 if (eb == root->node) {
8765 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8768 BUG_ON(root->root_key.objectid !=
8769 btrfs_header_owner(eb));
8771 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8772 parent = path->nodes[level + 1]->start;
8774 BUG_ON(root->root_key.objectid !=
8775 btrfs_header_owner(path->nodes[level + 1]));
8778 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8780 wc->refs[level] = 0;
8781 wc->flags[level] = 0;
8785 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8786 struct btrfs_root *root,
8787 struct btrfs_path *path,
8788 struct walk_control *wc)
8790 int level = wc->level;
8791 int lookup_info = 1;
8794 while (level >= 0) {
8795 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8802 if (path->slots[level] >=
8803 btrfs_header_nritems(path->nodes[level]))
8806 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8808 path->slots[level]++;
8817 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8818 struct btrfs_root *root,
8819 struct btrfs_path *path,
8820 struct walk_control *wc, int max_level)
8822 int level = wc->level;
8825 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8826 while (level < max_level && path->nodes[level]) {
8828 if (path->slots[level] + 1 <
8829 btrfs_header_nritems(path->nodes[level])) {
8830 path->slots[level]++;
8833 ret = walk_up_proc(trans, root, path, wc);
8837 if (path->locks[level]) {
8838 btrfs_tree_unlock_rw(path->nodes[level],
8839 path->locks[level]);
8840 path->locks[level] = 0;
8842 free_extent_buffer(path->nodes[level]);
8843 path->nodes[level] = NULL;
8851 * drop a subvolume tree.
8853 * this function traverses the tree freeing any blocks that only
8854 * referenced by the tree.
8856 * when a shared tree block is found. this function decreases its
8857 * reference count by one. if update_ref is true, this function
8858 * also make sure backrefs for the shared block and all lower level
8859 * blocks are properly updated.
8861 * If called with for_reloc == 0, may exit early with -EAGAIN
8863 int btrfs_drop_snapshot(struct btrfs_root *root,
8864 struct btrfs_block_rsv *block_rsv, int update_ref,
8867 struct btrfs_fs_info *fs_info = root->fs_info;
8868 struct btrfs_path *path;
8869 struct btrfs_trans_handle *trans;
8870 struct btrfs_root *tree_root = fs_info->tree_root;
8871 struct btrfs_root_item *root_item = &root->root_item;
8872 struct walk_control *wc;
8873 struct btrfs_key key;
8877 bool root_dropped = false;
8879 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
8881 path = btrfs_alloc_path();
8887 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8889 btrfs_free_path(path);
8894 trans = btrfs_start_transaction(tree_root, 0);
8895 if (IS_ERR(trans)) {
8896 err = PTR_ERR(trans);
8901 trans->block_rsv = block_rsv;
8903 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8904 level = btrfs_header_level(root->node);
8905 path->nodes[level] = btrfs_lock_root_node(root);
8906 btrfs_set_lock_blocking(path->nodes[level]);
8907 path->slots[level] = 0;
8908 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8909 memset(&wc->update_progress, 0,
8910 sizeof(wc->update_progress));
8912 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8913 memcpy(&wc->update_progress, &key,
8914 sizeof(wc->update_progress));
8916 level = root_item->drop_level;
8918 path->lowest_level = level;
8919 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8920 path->lowest_level = 0;
8928 * unlock our path, this is safe because only this
8929 * function is allowed to delete this snapshot
8931 btrfs_unlock_up_safe(path, 0);
8933 level = btrfs_header_level(root->node);
8935 btrfs_tree_lock(path->nodes[level]);
8936 btrfs_set_lock_blocking(path->nodes[level]);
8937 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8939 ret = btrfs_lookup_extent_info(trans, fs_info,
8940 path->nodes[level]->start,
8941 level, 1, &wc->refs[level],
8947 BUG_ON(wc->refs[level] == 0);
8949 if (level == root_item->drop_level)
8952 btrfs_tree_unlock(path->nodes[level]);
8953 path->locks[level] = 0;
8954 WARN_ON(wc->refs[level] != 1);
8960 wc->shared_level = -1;
8961 wc->stage = DROP_REFERENCE;
8962 wc->update_ref = update_ref;
8964 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
8968 ret = walk_down_tree(trans, root, path, wc);
8974 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8981 BUG_ON(wc->stage != DROP_REFERENCE);
8985 if (wc->stage == DROP_REFERENCE) {
8987 btrfs_node_key(path->nodes[level],
8988 &root_item->drop_progress,
8989 path->slots[level]);
8990 root_item->drop_level = level;
8993 BUG_ON(wc->level == 0);
8994 if (btrfs_should_end_transaction(trans) ||
8995 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
8996 ret = btrfs_update_root(trans, tree_root,
9000 btrfs_abort_transaction(trans, ret);
9005 btrfs_end_transaction_throttle(trans);
9006 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9007 btrfs_debug(fs_info,
9008 "drop snapshot early exit");
9013 trans = btrfs_start_transaction(tree_root, 0);
9014 if (IS_ERR(trans)) {
9015 err = PTR_ERR(trans);
9019 trans->block_rsv = block_rsv;
9022 btrfs_release_path(path);
9026 ret = btrfs_del_root(trans, &root->root_key);
9028 btrfs_abort_transaction(trans, ret);
9033 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9034 ret = btrfs_find_root(tree_root, &root->root_key, path,
9037 btrfs_abort_transaction(trans, ret);
9040 } else if (ret > 0) {
9041 /* if we fail to delete the orphan item this time
9042 * around, it'll get picked up the next time.
9044 * The most common failure here is just -ENOENT.
9046 btrfs_del_orphan_item(trans, tree_root,
9047 root->root_key.objectid);
9051 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9052 btrfs_add_dropped_root(trans, root);
9054 free_extent_buffer(root->node);
9055 free_extent_buffer(root->commit_root);
9056 btrfs_put_fs_root(root);
9058 root_dropped = true;
9060 btrfs_end_transaction_throttle(trans);
9063 btrfs_free_path(path);
9066 * So if we need to stop dropping the snapshot for whatever reason we
9067 * need to make sure to add it back to the dead root list so that we
9068 * keep trying to do the work later. This also cleans up roots if we
9069 * don't have it in the radix (like when we recover after a power fail
9070 * or unmount) so we don't leak memory.
9072 if (!for_reloc && !root_dropped)
9073 btrfs_add_dead_root(root);
9074 if (err && err != -EAGAIN)
9075 btrfs_handle_fs_error(fs_info, err, NULL);
9080 * drop subtree rooted at tree block 'node'.
9082 * NOTE: this function will unlock and release tree block 'node'
9083 * only used by relocation code
9085 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9086 struct btrfs_root *root,
9087 struct extent_buffer *node,
9088 struct extent_buffer *parent)
9090 struct btrfs_fs_info *fs_info = root->fs_info;
9091 struct btrfs_path *path;
9092 struct walk_control *wc;
9098 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9100 path = btrfs_alloc_path();
9104 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9106 btrfs_free_path(path);
9110 btrfs_assert_tree_locked(parent);
9111 parent_level = btrfs_header_level(parent);
9112 extent_buffer_get(parent);
9113 path->nodes[parent_level] = parent;
9114 path->slots[parent_level] = btrfs_header_nritems(parent);
9116 btrfs_assert_tree_locked(node);
9117 level = btrfs_header_level(node);
9118 path->nodes[level] = node;
9119 path->slots[level] = 0;
9120 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9122 wc->refs[parent_level] = 1;
9123 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9125 wc->shared_level = -1;
9126 wc->stage = DROP_REFERENCE;
9129 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9132 wret = walk_down_tree(trans, root, path, wc);
9138 wret = walk_up_tree(trans, root, path, wc, parent_level);
9146 btrfs_free_path(path);
9150 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9156 * if restripe for this chunk_type is on pick target profile and
9157 * return, otherwise do the usual balance
9159 stripped = get_restripe_target(fs_info, flags);
9161 return extended_to_chunk(stripped);
9163 num_devices = fs_info->fs_devices->rw_devices;
9165 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9166 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9167 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9169 if (num_devices == 1) {
9170 stripped |= BTRFS_BLOCK_GROUP_DUP;
9171 stripped = flags & ~stripped;
9173 /* turn raid0 into single device chunks */
9174 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9177 /* turn mirroring into duplication */
9178 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9179 BTRFS_BLOCK_GROUP_RAID10))
9180 return stripped | BTRFS_BLOCK_GROUP_DUP;
9182 /* they already had raid on here, just return */
9183 if (flags & stripped)
9186 stripped |= BTRFS_BLOCK_GROUP_DUP;
9187 stripped = flags & ~stripped;
9189 /* switch duplicated blocks with raid1 */
9190 if (flags & BTRFS_BLOCK_GROUP_DUP)
9191 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9193 /* this is drive concat, leave it alone */
9199 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9201 struct btrfs_space_info *sinfo = cache->space_info;
9203 u64 min_allocable_bytes;
9207 * We need some metadata space and system metadata space for
9208 * allocating chunks in some corner cases until we force to set
9209 * it to be readonly.
9212 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9214 min_allocable_bytes = SZ_1M;
9216 min_allocable_bytes = 0;
9218 spin_lock(&sinfo->lock);
9219 spin_lock(&cache->lock);
9227 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9228 cache->bytes_super - btrfs_block_group_used(&cache->item);
9230 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9231 min_allocable_bytes <= sinfo->total_bytes) {
9232 sinfo->bytes_readonly += num_bytes;
9234 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9238 spin_unlock(&cache->lock);
9239 spin_unlock(&sinfo->lock);
9243 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9246 struct btrfs_fs_info *fs_info = cache->fs_info;
9247 struct btrfs_trans_handle *trans;
9252 trans = btrfs_join_transaction(fs_info->extent_root);
9254 return PTR_ERR(trans);
9257 * we're not allowed to set block groups readonly after the dirty
9258 * block groups cache has started writing. If it already started,
9259 * back off and let this transaction commit
9261 mutex_lock(&fs_info->ro_block_group_mutex);
9262 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9263 u64 transid = trans->transid;
9265 mutex_unlock(&fs_info->ro_block_group_mutex);
9266 btrfs_end_transaction(trans);
9268 ret = btrfs_wait_for_commit(fs_info, transid);
9275 * if we are changing raid levels, try to allocate a corresponding
9276 * block group with the new raid level.
9278 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9279 if (alloc_flags != cache->flags) {
9280 ret = do_chunk_alloc(trans, alloc_flags,
9283 * ENOSPC is allowed here, we may have enough space
9284 * already allocated at the new raid level to
9293 ret = inc_block_group_ro(cache, 0);
9296 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9297 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9300 ret = inc_block_group_ro(cache, 0);
9302 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9303 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9304 mutex_lock(&fs_info->chunk_mutex);
9305 check_system_chunk(trans, alloc_flags);
9306 mutex_unlock(&fs_info->chunk_mutex);
9308 mutex_unlock(&fs_info->ro_block_group_mutex);
9310 btrfs_end_transaction(trans);
9314 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9316 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9318 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9322 * helper to account the unused space of all the readonly block group in the
9323 * space_info. takes mirrors into account.
9325 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9327 struct btrfs_block_group_cache *block_group;
9331 /* It's df, we don't care if it's racy */
9332 if (list_empty(&sinfo->ro_bgs))
9335 spin_lock(&sinfo->lock);
9336 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9337 spin_lock(&block_group->lock);
9339 if (!block_group->ro) {
9340 spin_unlock(&block_group->lock);
9344 factor = btrfs_bg_type_to_factor(block_group->flags);
9345 free_bytes += (block_group->key.offset -
9346 btrfs_block_group_used(&block_group->item)) *
9349 spin_unlock(&block_group->lock);
9351 spin_unlock(&sinfo->lock);
9356 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9358 struct btrfs_space_info *sinfo = cache->space_info;
9363 spin_lock(&sinfo->lock);
9364 spin_lock(&cache->lock);
9366 num_bytes = cache->key.offset - cache->reserved -
9367 cache->pinned - cache->bytes_super -
9368 btrfs_block_group_used(&cache->item);
9369 sinfo->bytes_readonly -= num_bytes;
9370 list_del_init(&cache->ro_list);
9372 spin_unlock(&cache->lock);
9373 spin_unlock(&sinfo->lock);
9377 * checks to see if its even possible to relocate this block group.
9379 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9380 * ok to go ahead and try.
9382 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9384 struct btrfs_root *root = fs_info->extent_root;
9385 struct btrfs_block_group_cache *block_group;
9386 struct btrfs_space_info *space_info;
9387 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9388 struct btrfs_device *device;
9389 struct btrfs_trans_handle *trans;
9399 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9401 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9403 /* odd, couldn't find the block group, leave it alone */
9407 "can't find block group for bytenr %llu",
9412 min_free = btrfs_block_group_used(&block_group->item);
9414 /* no bytes used, we're good */
9418 space_info = block_group->space_info;
9419 spin_lock(&space_info->lock);
9421 full = space_info->full;
9424 * if this is the last block group we have in this space, we can't
9425 * relocate it unless we're able to allocate a new chunk below.
9427 * Otherwise, we need to make sure we have room in the space to handle
9428 * all of the extents from this block group. If we can, we're good
9430 if ((space_info->total_bytes != block_group->key.offset) &&
9431 (btrfs_space_info_used(space_info, false) + min_free <
9432 space_info->total_bytes)) {
9433 spin_unlock(&space_info->lock);
9436 spin_unlock(&space_info->lock);
9439 * ok we don't have enough space, but maybe we have free space on our
9440 * devices to allocate new chunks for relocation, so loop through our
9441 * alloc devices and guess if we have enough space. if this block
9442 * group is going to be restriped, run checks against the target
9443 * profile instead of the current one.
9455 target = get_restripe_target(fs_info, block_group->flags);
9457 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9460 * this is just a balance, so if we were marked as full
9461 * we know there is no space for a new chunk
9466 "no space to alloc new chunk for block group %llu",
9467 block_group->key.objectid);
9471 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9474 if (index == BTRFS_RAID_RAID10) {
9478 } else if (index == BTRFS_RAID_RAID1) {
9480 } else if (index == BTRFS_RAID_DUP) {
9483 } else if (index == BTRFS_RAID_RAID0) {
9484 dev_min = fs_devices->rw_devices;
9485 min_free = div64_u64(min_free, dev_min);
9488 /* We need to do this so that we can look at pending chunks */
9489 trans = btrfs_join_transaction(root);
9490 if (IS_ERR(trans)) {
9491 ret = PTR_ERR(trans);
9495 mutex_lock(&fs_info->chunk_mutex);
9496 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9500 * check to make sure we can actually find a chunk with enough
9501 * space to fit our block group in.
9503 if (device->total_bytes > device->bytes_used + min_free &&
9504 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9505 ret = find_free_dev_extent(trans, device, min_free,
9510 if (dev_nr >= dev_min)
9516 if (debug && ret == -1)
9518 "no space to allocate a new chunk for block group %llu",
9519 block_group->key.objectid);
9520 mutex_unlock(&fs_info->chunk_mutex);
9521 btrfs_end_transaction(trans);
9523 btrfs_put_block_group(block_group);
9527 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9528 struct btrfs_path *path,
9529 struct btrfs_key *key)
9531 struct btrfs_root *root = fs_info->extent_root;
9533 struct btrfs_key found_key;
9534 struct extent_buffer *leaf;
9535 struct btrfs_block_group_item bg;
9539 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9544 slot = path->slots[0];
9545 leaf = path->nodes[0];
9546 if (slot >= btrfs_header_nritems(leaf)) {
9547 ret = btrfs_next_leaf(root, path);
9554 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9556 if (found_key.objectid >= key->objectid &&
9557 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9558 struct extent_map_tree *em_tree;
9559 struct extent_map *em;
9561 em_tree = &root->fs_info->mapping_tree.map_tree;
9562 read_lock(&em_tree->lock);
9563 em = lookup_extent_mapping(em_tree, found_key.objectid,
9565 read_unlock(&em_tree->lock);
9568 "logical %llu len %llu found bg but no related chunk",
9569 found_key.objectid, found_key.offset);
9571 } else if (em->start != found_key.objectid ||
9572 em->len != found_key.offset) {
9574 "block group %llu len %llu mismatch with chunk %llu len %llu",
9575 found_key.objectid, found_key.offset,
9576 em->start, em->len);
9579 read_extent_buffer(leaf, &bg,
9580 btrfs_item_ptr_offset(leaf, slot),
9582 flags = btrfs_block_group_flags(&bg) &
9583 BTRFS_BLOCK_GROUP_TYPE_MASK;
9585 if (flags != (em->map_lookup->type &
9586 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9588 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9590 found_key.offset, flags,
9591 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9592 em->map_lookup->type));
9598 free_extent_map(em);
9607 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9609 struct btrfs_block_group_cache *block_group;
9613 struct inode *inode;
9615 block_group = btrfs_lookup_first_block_group(info, last);
9616 while (block_group) {
9617 spin_lock(&block_group->lock);
9618 if (block_group->iref)
9620 spin_unlock(&block_group->lock);
9621 block_group = next_block_group(info, block_group);
9630 inode = block_group->inode;
9631 block_group->iref = 0;
9632 block_group->inode = NULL;
9633 spin_unlock(&block_group->lock);
9634 ASSERT(block_group->io_ctl.inode == NULL);
9636 last = block_group->key.objectid + block_group->key.offset;
9637 btrfs_put_block_group(block_group);
9642 * Must be called only after stopping all workers, since we could have block
9643 * group caching kthreads running, and therefore they could race with us if we
9644 * freed the block groups before stopping them.
9646 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9648 struct btrfs_block_group_cache *block_group;
9649 struct btrfs_space_info *space_info;
9650 struct btrfs_caching_control *caching_ctl;
9653 down_write(&info->commit_root_sem);
9654 while (!list_empty(&info->caching_block_groups)) {
9655 caching_ctl = list_entry(info->caching_block_groups.next,
9656 struct btrfs_caching_control, list);
9657 list_del(&caching_ctl->list);
9658 put_caching_control(caching_ctl);
9660 up_write(&info->commit_root_sem);
9662 spin_lock(&info->unused_bgs_lock);
9663 while (!list_empty(&info->unused_bgs)) {
9664 block_group = list_first_entry(&info->unused_bgs,
9665 struct btrfs_block_group_cache,
9667 list_del_init(&block_group->bg_list);
9668 btrfs_put_block_group(block_group);
9670 spin_unlock(&info->unused_bgs_lock);
9672 spin_lock(&info->block_group_cache_lock);
9673 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9674 block_group = rb_entry(n, struct btrfs_block_group_cache,
9676 rb_erase(&block_group->cache_node,
9677 &info->block_group_cache_tree);
9678 RB_CLEAR_NODE(&block_group->cache_node);
9679 spin_unlock(&info->block_group_cache_lock);
9681 down_write(&block_group->space_info->groups_sem);
9682 list_del(&block_group->list);
9683 up_write(&block_group->space_info->groups_sem);
9686 * We haven't cached this block group, which means we could
9687 * possibly have excluded extents on this block group.
9689 if (block_group->cached == BTRFS_CACHE_NO ||
9690 block_group->cached == BTRFS_CACHE_ERROR)
9691 free_excluded_extents(block_group);
9693 btrfs_remove_free_space_cache(block_group);
9694 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9695 ASSERT(list_empty(&block_group->dirty_list));
9696 ASSERT(list_empty(&block_group->io_list));
9697 ASSERT(list_empty(&block_group->bg_list));
9698 ASSERT(atomic_read(&block_group->count) == 1);
9699 btrfs_put_block_group(block_group);
9701 spin_lock(&info->block_group_cache_lock);
9703 spin_unlock(&info->block_group_cache_lock);
9705 /* now that all the block groups are freed, go through and
9706 * free all the space_info structs. This is only called during
9707 * the final stages of unmount, and so we know nobody is
9708 * using them. We call synchronize_rcu() once before we start,
9709 * just to be on the safe side.
9713 release_global_block_rsv(info);
9715 while (!list_empty(&info->space_info)) {
9718 space_info = list_entry(info->space_info.next,
9719 struct btrfs_space_info,
9723 * Do not hide this behind enospc_debug, this is actually
9724 * important and indicates a real bug if this happens.
9726 if (WARN_ON(space_info->bytes_pinned > 0 ||
9727 space_info->bytes_reserved > 0 ||
9728 space_info->bytes_may_use > 0))
9729 dump_space_info(info, space_info, 0, 0);
9730 list_del(&space_info->list);
9731 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9732 struct kobject *kobj;
9733 kobj = space_info->block_group_kobjs[i];
9734 space_info->block_group_kobjs[i] = NULL;
9740 kobject_del(&space_info->kobj);
9741 kobject_put(&space_info->kobj);
9746 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9747 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9749 struct btrfs_space_info *space_info;
9750 struct raid_kobject *rkobj;
9755 spin_lock(&fs_info->pending_raid_kobjs_lock);
9756 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9757 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9759 list_for_each_entry(rkobj, &list, list) {
9760 space_info = __find_space_info(fs_info, rkobj->flags);
9761 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9763 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9764 "%s", get_raid_name(index));
9766 kobject_put(&rkobj->kobj);
9772 "failed to add kobject for block cache, ignoring");
9775 static void link_block_group(struct btrfs_block_group_cache *cache)
9777 struct btrfs_space_info *space_info = cache->space_info;
9778 struct btrfs_fs_info *fs_info = cache->fs_info;
9779 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9782 down_write(&space_info->groups_sem);
9783 if (list_empty(&space_info->block_groups[index]))
9785 list_add_tail(&cache->list, &space_info->block_groups[index]);
9786 up_write(&space_info->groups_sem);
9789 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9791 btrfs_warn(cache->fs_info,
9792 "couldn't alloc memory for raid level kobject");
9795 rkobj->flags = cache->flags;
9796 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9798 spin_lock(&fs_info->pending_raid_kobjs_lock);
9799 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9800 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9801 space_info->block_group_kobjs[index] = &rkobj->kobj;
9805 static struct btrfs_block_group_cache *
9806 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9807 u64 start, u64 size)
9809 struct btrfs_block_group_cache *cache;
9811 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9815 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9817 if (!cache->free_space_ctl) {
9822 cache->key.objectid = start;
9823 cache->key.offset = size;
9824 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9826 cache->fs_info = fs_info;
9827 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9828 set_free_space_tree_thresholds(cache);
9830 atomic_set(&cache->count, 1);
9831 spin_lock_init(&cache->lock);
9832 init_rwsem(&cache->data_rwsem);
9833 INIT_LIST_HEAD(&cache->list);
9834 INIT_LIST_HEAD(&cache->cluster_list);
9835 INIT_LIST_HEAD(&cache->bg_list);
9836 INIT_LIST_HEAD(&cache->ro_list);
9837 INIT_LIST_HEAD(&cache->dirty_list);
9838 INIT_LIST_HEAD(&cache->io_list);
9839 btrfs_init_free_space_ctl(cache);
9840 atomic_set(&cache->trimming, 0);
9841 mutex_init(&cache->free_space_lock);
9842 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9849 * Iterate all chunks and verify that each of them has the corresponding block
9852 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
9854 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
9855 struct extent_map *em;
9856 struct btrfs_block_group_cache *bg;
9861 read_lock(&map_tree->map_tree.lock);
9863 * lookup_extent_mapping will return the first extent map
9864 * intersecting the range, so setting @len to 1 is enough to
9865 * get the first chunk.
9867 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
9868 read_unlock(&map_tree->map_tree.lock);
9872 bg = btrfs_lookup_block_group(fs_info, em->start);
9875 "chunk start=%llu len=%llu doesn't have corresponding block group",
9876 em->start, em->len);
9878 free_extent_map(em);
9881 if (bg->key.objectid != em->start ||
9882 bg->key.offset != em->len ||
9883 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
9884 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9886 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
9888 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
9889 bg->key.objectid, bg->key.offset,
9890 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
9892 free_extent_map(em);
9893 btrfs_put_block_group(bg);
9896 start = em->start + em->len;
9897 free_extent_map(em);
9898 btrfs_put_block_group(bg);
9903 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9905 struct btrfs_path *path;
9907 struct btrfs_block_group_cache *cache;
9908 struct btrfs_space_info *space_info;
9909 struct btrfs_key key;
9910 struct btrfs_key found_key;
9911 struct extent_buffer *leaf;
9917 feature = btrfs_super_incompat_flags(info->super_copy);
9918 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9922 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9923 path = btrfs_alloc_path();
9926 path->reada = READA_FORWARD;
9928 cache_gen = btrfs_super_cache_generation(info->super_copy);
9929 if (btrfs_test_opt(info, SPACE_CACHE) &&
9930 btrfs_super_generation(info->super_copy) != cache_gen)
9932 if (btrfs_test_opt(info, CLEAR_CACHE))
9936 ret = find_first_block_group(info, path, &key);
9942 leaf = path->nodes[0];
9943 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9945 cache = btrfs_create_block_group_cache(info, found_key.objectid,
9954 * When we mount with old space cache, we need to
9955 * set BTRFS_DC_CLEAR and set dirty flag.
9957 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9958 * truncate the old free space cache inode and
9960 * b) Setting 'dirty flag' makes sure that we flush
9961 * the new space cache info onto disk.
9963 if (btrfs_test_opt(info, SPACE_CACHE))
9964 cache->disk_cache_state = BTRFS_DC_CLEAR;
9967 read_extent_buffer(leaf, &cache->item,
9968 btrfs_item_ptr_offset(leaf, path->slots[0]),
9969 sizeof(cache->item));
9970 cache->flags = btrfs_block_group_flags(&cache->item);
9972 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
9973 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
9975 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
9976 cache->key.objectid);
9981 key.objectid = found_key.objectid + found_key.offset;
9982 btrfs_release_path(path);
9985 * We need to exclude the super stripes now so that the space
9986 * info has super bytes accounted for, otherwise we'll think
9987 * we have more space than we actually do.
9989 ret = exclude_super_stripes(cache);
9992 * We may have excluded something, so call this just in
9995 free_excluded_extents(cache);
9996 btrfs_put_block_group(cache);
10001 * check for two cases, either we are full, and therefore
10002 * don't need to bother with the caching work since we won't
10003 * find any space, or we are empty, and we can just add all
10004 * the space in and be done with it. This saves us _alot_ of
10005 * time, particularly in the full case.
10007 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10008 cache->last_byte_to_unpin = (u64)-1;
10009 cache->cached = BTRFS_CACHE_FINISHED;
10010 free_excluded_extents(cache);
10011 } else if (btrfs_block_group_used(&cache->item) == 0) {
10012 cache->last_byte_to_unpin = (u64)-1;
10013 cache->cached = BTRFS_CACHE_FINISHED;
10014 add_new_free_space(cache, found_key.objectid,
10015 found_key.objectid +
10017 free_excluded_extents(cache);
10020 ret = btrfs_add_block_group_cache(info, cache);
10022 btrfs_remove_free_space_cache(cache);
10023 btrfs_put_block_group(cache);
10027 trace_btrfs_add_block_group(info, cache, 0);
10028 update_space_info(info, cache->flags, found_key.offset,
10029 btrfs_block_group_used(&cache->item),
10030 cache->bytes_super, &space_info);
10032 cache->space_info = space_info;
10034 link_block_group(cache);
10036 set_avail_alloc_bits(info, cache->flags);
10037 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10038 inc_block_group_ro(cache, 1);
10039 } else if (btrfs_block_group_used(&cache->item) == 0) {
10040 ASSERT(list_empty(&cache->bg_list));
10041 btrfs_mark_bg_unused(cache);
10045 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10046 if (!(get_alloc_profile(info, space_info->flags) &
10047 (BTRFS_BLOCK_GROUP_RAID10 |
10048 BTRFS_BLOCK_GROUP_RAID1 |
10049 BTRFS_BLOCK_GROUP_RAID5 |
10050 BTRFS_BLOCK_GROUP_RAID6 |
10051 BTRFS_BLOCK_GROUP_DUP)))
10054 * avoid allocating from un-mirrored block group if there are
10055 * mirrored block groups.
10057 list_for_each_entry(cache,
10058 &space_info->block_groups[BTRFS_RAID_RAID0],
10060 inc_block_group_ro(cache, 1);
10061 list_for_each_entry(cache,
10062 &space_info->block_groups[BTRFS_RAID_SINGLE],
10064 inc_block_group_ro(cache, 1);
10067 btrfs_add_raid_kobjects(info);
10068 init_global_block_rsv(info);
10069 ret = check_chunk_block_group_mappings(info);
10071 btrfs_free_path(path);
10075 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10077 struct btrfs_fs_info *fs_info = trans->fs_info;
10078 struct btrfs_block_group_cache *block_group, *tmp;
10079 struct btrfs_root *extent_root = fs_info->extent_root;
10080 struct btrfs_block_group_item item;
10081 struct btrfs_key key;
10083 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10085 trans->can_flush_pending_bgs = false;
10086 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10090 spin_lock(&block_group->lock);
10091 memcpy(&item, &block_group->item, sizeof(item));
10092 memcpy(&key, &block_group->key, sizeof(key));
10093 spin_unlock(&block_group->lock);
10095 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10098 btrfs_abort_transaction(trans, ret);
10099 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10101 btrfs_abort_transaction(trans, ret);
10102 add_block_group_free_space(trans, block_group);
10103 /* already aborted the transaction if it failed. */
10105 list_del_init(&block_group->bg_list);
10107 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10110 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10111 u64 type, u64 chunk_offset, u64 size)
10113 struct btrfs_fs_info *fs_info = trans->fs_info;
10114 struct btrfs_block_group_cache *cache;
10117 btrfs_set_log_full_commit(fs_info, trans);
10119 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10123 btrfs_set_block_group_used(&cache->item, bytes_used);
10124 btrfs_set_block_group_chunk_objectid(&cache->item,
10125 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10126 btrfs_set_block_group_flags(&cache->item, type);
10128 cache->flags = type;
10129 cache->last_byte_to_unpin = (u64)-1;
10130 cache->cached = BTRFS_CACHE_FINISHED;
10131 cache->needs_free_space = 1;
10132 ret = exclude_super_stripes(cache);
10135 * We may have excluded something, so call this just in
10138 free_excluded_extents(cache);
10139 btrfs_put_block_group(cache);
10143 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10145 free_excluded_extents(cache);
10147 #ifdef CONFIG_BTRFS_DEBUG
10148 if (btrfs_should_fragment_free_space(cache)) {
10149 u64 new_bytes_used = size - bytes_used;
10151 bytes_used += new_bytes_used >> 1;
10152 fragment_free_space(cache);
10156 * Ensure the corresponding space_info object is created and
10157 * assigned to our block group. We want our bg to be added to the rbtree
10158 * with its ->space_info set.
10160 cache->space_info = __find_space_info(fs_info, cache->flags);
10161 ASSERT(cache->space_info);
10163 ret = btrfs_add_block_group_cache(fs_info, cache);
10165 btrfs_remove_free_space_cache(cache);
10166 btrfs_put_block_group(cache);
10171 * Now that our block group has its ->space_info set and is inserted in
10172 * the rbtree, update the space info's counters.
10174 trace_btrfs_add_block_group(fs_info, cache, 1);
10175 update_space_info(fs_info, cache->flags, size, bytes_used,
10176 cache->bytes_super, &cache->space_info);
10177 update_global_block_rsv(fs_info);
10179 link_block_group(cache);
10181 list_add_tail(&cache->bg_list, &trans->new_bgs);
10183 set_avail_alloc_bits(fs_info, type);
10187 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10189 u64 extra_flags = chunk_to_extended(flags) &
10190 BTRFS_EXTENDED_PROFILE_MASK;
10192 write_seqlock(&fs_info->profiles_lock);
10193 if (flags & BTRFS_BLOCK_GROUP_DATA)
10194 fs_info->avail_data_alloc_bits &= ~extra_flags;
10195 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10196 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10197 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10198 fs_info->avail_system_alloc_bits &= ~extra_flags;
10199 write_sequnlock(&fs_info->profiles_lock);
10202 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10203 u64 group_start, struct extent_map *em)
10205 struct btrfs_fs_info *fs_info = trans->fs_info;
10206 struct btrfs_root *root = fs_info->extent_root;
10207 struct btrfs_path *path;
10208 struct btrfs_block_group_cache *block_group;
10209 struct btrfs_free_cluster *cluster;
10210 struct btrfs_root *tree_root = fs_info->tree_root;
10211 struct btrfs_key key;
10212 struct inode *inode;
10213 struct kobject *kobj = NULL;
10217 struct btrfs_caching_control *caching_ctl = NULL;
10220 block_group = btrfs_lookup_block_group(fs_info, group_start);
10221 BUG_ON(!block_group);
10222 BUG_ON(!block_group->ro);
10224 trace_btrfs_remove_block_group(block_group);
10226 * Free the reserved super bytes from this block group before
10229 free_excluded_extents(block_group);
10230 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10231 block_group->key.offset);
10233 memcpy(&key, &block_group->key, sizeof(key));
10234 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10235 factor = btrfs_bg_type_to_factor(block_group->flags);
10237 /* make sure this block group isn't part of an allocation cluster */
10238 cluster = &fs_info->data_alloc_cluster;
10239 spin_lock(&cluster->refill_lock);
10240 btrfs_return_cluster_to_free_space(block_group, cluster);
10241 spin_unlock(&cluster->refill_lock);
10244 * make sure this block group isn't part of a metadata
10245 * allocation cluster
10247 cluster = &fs_info->meta_alloc_cluster;
10248 spin_lock(&cluster->refill_lock);
10249 btrfs_return_cluster_to_free_space(block_group, cluster);
10250 spin_unlock(&cluster->refill_lock);
10252 path = btrfs_alloc_path();
10259 * get the inode first so any iput calls done for the io_list
10260 * aren't the final iput (no unlinks allowed now)
10262 inode = lookup_free_space_inode(fs_info, block_group, path);
10264 mutex_lock(&trans->transaction->cache_write_mutex);
10266 * make sure our free spache cache IO is done before remove the
10269 spin_lock(&trans->transaction->dirty_bgs_lock);
10270 if (!list_empty(&block_group->io_list)) {
10271 list_del_init(&block_group->io_list);
10273 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10275 spin_unlock(&trans->transaction->dirty_bgs_lock);
10276 btrfs_wait_cache_io(trans, block_group, path);
10277 btrfs_put_block_group(block_group);
10278 spin_lock(&trans->transaction->dirty_bgs_lock);
10281 if (!list_empty(&block_group->dirty_list)) {
10282 list_del_init(&block_group->dirty_list);
10283 btrfs_put_block_group(block_group);
10285 spin_unlock(&trans->transaction->dirty_bgs_lock);
10286 mutex_unlock(&trans->transaction->cache_write_mutex);
10288 if (!IS_ERR(inode)) {
10289 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10291 btrfs_add_delayed_iput(inode);
10294 clear_nlink(inode);
10295 /* One for the block groups ref */
10296 spin_lock(&block_group->lock);
10297 if (block_group->iref) {
10298 block_group->iref = 0;
10299 block_group->inode = NULL;
10300 spin_unlock(&block_group->lock);
10303 spin_unlock(&block_group->lock);
10305 /* One for our lookup ref */
10306 btrfs_add_delayed_iput(inode);
10309 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10310 key.offset = block_group->key.objectid;
10313 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10317 btrfs_release_path(path);
10319 ret = btrfs_del_item(trans, tree_root, path);
10322 btrfs_release_path(path);
10325 spin_lock(&fs_info->block_group_cache_lock);
10326 rb_erase(&block_group->cache_node,
10327 &fs_info->block_group_cache_tree);
10328 RB_CLEAR_NODE(&block_group->cache_node);
10330 if (fs_info->first_logical_byte == block_group->key.objectid)
10331 fs_info->first_logical_byte = (u64)-1;
10332 spin_unlock(&fs_info->block_group_cache_lock);
10334 down_write(&block_group->space_info->groups_sem);
10336 * we must use list_del_init so people can check to see if they
10337 * are still on the list after taking the semaphore
10339 list_del_init(&block_group->list);
10340 if (list_empty(&block_group->space_info->block_groups[index])) {
10341 kobj = block_group->space_info->block_group_kobjs[index];
10342 block_group->space_info->block_group_kobjs[index] = NULL;
10343 clear_avail_alloc_bits(fs_info, block_group->flags);
10345 up_write(&block_group->space_info->groups_sem);
10351 if (block_group->has_caching_ctl)
10352 caching_ctl = get_caching_control(block_group);
10353 if (block_group->cached == BTRFS_CACHE_STARTED)
10354 wait_block_group_cache_done(block_group);
10355 if (block_group->has_caching_ctl) {
10356 down_write(&fs_info->commit_root_sem);
10357 if (!caching_ctl) {
10358 struct btrfs_caching_control *ctl;
10360 list_for_each_entry(ctl,
10361 &fs_info->caching_block_groups, list)
10362 if (ctl->block_group == block_group) {
10364 refcount_inc(&caching_ctl->count);
10369 list_del_init(&caching_ctl->list);
10370 up_write(&fs_info->commit_root_sem);
10372 /* Once for the caching bgs list and once for us. */
10373 put_caching_control(caching_ctl);
10374 put_caching_control(caching_ctl);
10378 spin_lock(&trans->transaction->dirty_bgs_lock);
10379 if (!list_empty(&block_group->dirty_list)) {
10382 if (!list_empty(&block_group->io_list)) {
10385 spin_unlock(&trans->transaction->dirty_bgs_lock);
10386 btrfs_remove_free_space_cache(block_group);
10388 spin_lock(&block_group->space_info->lock);
10389 list_del_init(&block_group->ro_list);
10391 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10392 WARN_ON(block_group->space_info->total_bytes
10393 < block_group->key.offset);
10394 WARN_ON(block_group->space_info->bytes_readonly
10395 < block_group->key.offset);
10396 WARN_ON(block_group->space_info->disk_total
10397 < block_group->key.offset * factor);
10399 block_group->space_info->total_bytes -= block_group->key.offset;
10400 block_group->space_info->bytes_readonly -= block_group->key.offset;
10401 block_group->space_info->disk_total -= block_group->key.offset * factor;
10403 spin_unlock(&block_group->space_info->lock);
10405 memcpy(&key, &block_group->key, sizeof(key));
10407 mutex_lock(&fs_info->chunk_mutex);
10408 if (!list_empty(&em->list)) {
10409 /* We're in the transaction->pending_chunks list. */
10410 free_extent_map(em);
10412 spin_lock(&block_group->lock);
10413 block_group->removed = 1;
10415 * At this point trimming can't start on this block group, because we
10416 * removed the block group from the tree fs_info->block_group_cache_tree
10417 * so no one can't find it anymore and even if someone already got this
10418 * block group before we removed it from the rbtree, they have already
10419 * incremented block_group->trimming - if they didn't, they won't find
10420 * any free space entries because we already removed them all when we
10421 * called btrfs_remove_free_space_cache().
10423 * And we must not remove the extent map from the fs_info->mapping_tree
10424 * to prevent the same logical address range and physical device space
10425 * ranges from being reused for a new block group. This is because our
10426 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10427 * completely transactionless, so while it is trimming a range the
10428 * currently running transaction might finish and a new one start,
10429 * allowing for new block groups to be created that can reuse the same
10430 * physical device locations unless we take this special care.
10432 * There may also be an implicit trim operation if the file system
10433 * is mounted with -odiscard. The same protections must remain
10434 * in place until the extents have been discarded completely when
10435 * the transaction commit has completed.
10437 remove_em = (atomic_read(&block_group->trimming) == 0);
10439 * Make sure a trimmer task always sees the em in the pinned_chunks list
10440 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10441 * before checking block_group->removed).
10445 * Our em might be in trans->transaction->pending_chunks which
10446 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10447 * and so is the fs_info->pinned_chunks list.
10449 * So at this point we must be holding the chunk_mutex to avoid
10450 * any races with chunk allocation (more specifically at
10451 * volumes.c:contains_pending_extent()), to ensure it always
10452 * sees the em, either in the pending_chunks list or in the
10453 * pinned_chunks list.
10455 list_move_tail(&em->list, &fs_info->pinned_chunks);
10457 spin_unlock(&block_group->lock);
10460 struct extent_map_tree *em_tree;
10462 em_tree = &fs_info->mapping_tree.map_tree;
10463 write_lock(&em_tree->lock);
10465 * The em might be in the pending_chunks list, so make sure the
10466 * chunk mutex is locked, since remove_extent_mapping() will
10467 * delete us from that list.
10469 remove_extent_mapping(em_tree, em);
10470 write_unlock(&em_tree->lock);
10471 /* once for the tree */
10472 free_extent_map(em);
10475 mutex_unlock(&fs_info->chunk_mutex);
10477 ret = remove_block_group_free_space(trans, block_group);
10481 btrfs_put_block_group(block_group);
10482 btrfs_put_block_group(block_group);
10484 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10490 ret = btrfs_del_item(trans, root, path);
10492 btrfs_free_path(path);
10496 struct btrfs_trans_handle *
10497 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10498 const u64 chunk_offset)
10500 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10501 struct extent_map *em;
10502 struct map_lookup *map;
10503 unsigned int num_items;
10505 read_lock(&em_tree->lock);
10506 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10507 read_unlock(&em_tree->lock);
10508 ASSERT(em && em->start == chunk_offset);
10511 * We need to reserve 3 + N units from the metadata space info in order
10512 * to remove a block group (done at btrfs_remove_chunk() and at
10513 * btrfs_remove_block_group()), which are used for:
10515 * 1 unit for adding the free space inode's orphan (located in the tree
10517 * 1 unit for deleting the block group item (located in the extent
10519 * 1 unit for deleting the free space item (located in tree of tree
10521 * N units for deleting N device extent items corresponding to each
10522 * stripe (located in the device tree).
10524 * In order to remove a block group we also need to reserve units in the
10525 * system space info in order to update the chunk tree (update one or
10526 * more device items and remove one chunk item), but this is done at
10527 * btrfs_remove_chunk() through a call to check_system_chunk().
10529 map = em->map_lookup;
10530 num_items = 3 + map->num_stripes;
10531 free_extent_map(em);
10533 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10538 * Process the unused_bgs list and remove any that don't have any allocated
10539 * space inside of them.
10541 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10543 struct btrfs_block_group_cache *block_group;
10544 struct btrfs_space_info *space_info;
10545 struct btrfs_trans_handle *trans;
10548 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10551 spin_lock(&fs_info->unused_bgs_lock);
10552 while (!list_empty(&fs_info->unused_bgs)) {
10556 block_group = list_first_entry(&fs_info->unused_bgs,
10557 struct btrfs_block_group_cache,
10559 list_del_init(&block_group->bg_list);
10561 space_info = block_group->space_info;
10563 if (ret || btrfs_mixed_space_info(space_info)) {
10564 btrfs_put_block_group(block_group);
10567 spin_unlock(&fs_info->unused_bgs_lock);
10569 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10571 /* Don't want to race with allocators so take the groups_sem */
10572 down_write(&space_info->groups_sem);
10573 spin_lock(&block_group->lock);
10574 if (block_group->reserved || block_group->pinned ||
10575 btrfs_block_group_used(&block_group->item) ||
10577 list_is_singular(&block_group->list)) {
10579 * We want to bail if we made new allocations or have
10580 * outstanding allocations in this block group. We do
10581 * the ro check in case balance is currently acting on
10582 * this block group.
10584 trace_btrfs_skip_unused_block_group(block_group);
10585 spin_unlock(&block_group->lock);
10586 up_write(&space_info->groups_sem);
10589 spin_unlock(&block_group->lock);
10591 /* We don't want to force the issue, only flip if it's ok. */
10592 ret = inc_block_group_ro(block_group, 0);
10593 up_write(&space_info->groups_sem);
10600 * Want to do this before we do anything else so we can recover
10601 * properly if we fail to join the transaction.
10603 trans = btrfs_start_trans_remove_block_group(fs_info,
10604 block_group->key.objectid);
10605 if (IS_ERR(trans)) {
10606 btrfs_dec_block_group_ro(block_group);
10607 ret = PTR_ERR(trans);
10612 * We could have pending pinned extents for this block group,
10613 * just delete them, we don't care about them anymore.
10615 start = block_group->key.objectid;
10616 end = start + block_group->key.offset - 1;
10618 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10619 * btrfs_finish_extent_commit(). If we are at transaction N,
10620 * another task might be running finish_extent_commit() for the
10621 * previous transaction N - 1, and have seen a range belonging
10622 * to the block group in freed_extents[] before we were able to
10623 * clear the whole block group range from freed_extents[]. This
10624 * means that task can lookup for the block group after we
10625 * unpinned it from freed_extents[] and removed it, leading to
10626 * a BUG_ON() at btrfs_unpin_extent_range().
10628 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10629 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10632 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10633 btrfs_dec_block_group_ro(block_group);
10636 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10639 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10640 btrfs_dec_block_group_ro(block_group);
10643 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10645 /* Reset pinned so btrfs_put_block_group doesn't complain */
10646 spin_lock(&space_info->lock);
10647 spin_lock(&block_group->lock);
10649 space_info->bytes_pinned -= block_group->pinned;
10650 space_info->bytes_readonly += block_group->pinned;
10651 percpu_counter_add_batch(&space_info->total_bytes_pinned,
10652 -block_group->pinned,
10653 BTRFS_TOTAL_BYTES_PINNED_BATCH);
10654 block_group->pinned = 0;
10656 spin_unlock(&block_group->lock);
10657 spin_unlock(&space_info->lock);
10659 /* DISCARD can flip during remount */
10660 trimming = btrfs_test_opt(fs_info, DISCARD);
10662 /* Implicit trim during transaction commit. */
10664 btrfs_get_block_group_trimming(block_group);
10667 * Btrfs_remove_chunk will abort the transaction if things go
10670 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
10674 btrfs_put_block_group_trimming(block_group);
10679 * If we're not mounted with -odiscard, we can just forget
10680 * about this block group. Otherwise we'll need to wait
10681 * until transaction commit to do the actual discard.
10684 spin_lock(&fs_info->unused_bgs_lock);
10686 * A concurrent scrub might have added us to the list
10687 * fs_info->unused_bgs, so use a list_move operation
10688 * to add the block group to the deleted_bgs list.
10690 list_move(&block_group->bg_list,
10691 &trans->transaction->deleted_bgs);
10692 spin_unlock(&fs_info->unused_bgs_lock);
10693 btrfs_get_block_group(block_group);
10696 btrfs_end_transaction(trans);
10698 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10699 btrfs_put_block_group(block_group);
10700 spin_lock(&fs_info->unused_bgs_lock);
10702 spin_unlock(&fs_info->unused_bgs_lock);
10705 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10707 struct btrfs_super_block *disk_super;
10713 disk_super = fs_info->super_copy;
10714 if (!btrfs_super_root(disk_super))
10717 features = btrfs_super_incompat_flags(disk_super);
10718 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10721 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10722 ret = create_space_info(fs_info, flags);
10727 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10728 ret = create_space_info(fs_info, flags);
10730 flags = BTRFS_BLOCK_GROUP_METADATA;
10731 ret = create_space_info(fs_info, flags);
10735 flags = BTRFS_BLOCK_GROUP_DATA;
10736 ret = create_space_info(fs_info, flags);
10742 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10743 u64 start, u64 end)
10745 return unpin_extent_range(fs_info, start, end, false);
10749 * It used to be that old block groups would be left around forever.
10750 * Iterating over them would be enough to trim unused space. Since we
10751 * now automatically remove them, we also need to iterate over unallocated
10754 * We don't want a transaction for this since the discard may take a
10755 * substantial amount of time. We don't require that a transaction be
10756 * running, but we do need to take a running transaction into account
10757 * to ensure that we're not discarding chunks that were released in
10758 * the current transaction.
10760 * Holding the chunks lock will prevent other threads from allocating
10761 * or releasing chunks, but it won't prevent a running transaction
10762 * from committing and releasing the memory that the pending chunks
10763 * list head uses. For that, we need to take a reference to the
10766 static int btrfs_trim_free_extents(struct btrfs_device *device,
10767 u64 minlen, u64 *trimmed)
10769 u64 start = 0, len = 0;
10774 /* Not writeable = nothing to do. */
10775 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10778 /* No free space = nothing to do. */
10779 if (device->total_bytes <= device->bytes_used)
10785 struct btrfs_fs_info *fs_info = device->fs_info;
10786 struct btrfs_transaction *trans;
10789 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10793 down_read(&fs_info->commit_root_sem);
10795 spin_lock(&fs_info->trans_lock);
10796 trans = fs_info->running_transaction;
10798 refcount_inc(&trans->use_count);
10799 spin_unlock(&fs_info->trans_lock);
10801 ret = find_free_dev_extent_start(trans, device, minlen, start,
10804 btrfs_put_transaction(trans);
10807 up_read(&fs_info->commit_root_sem);
10808 mutex_unlock(&fs_info->chunk_mutex);
10809 if (ret == -ENOSPC)
10814 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10815 up_read(&fs_info->commit_root_sem);
10816 mutex_unlock(&fs_info->chunk_mutex);
10824 if (fatal_signal_pending(current)) {
10825 ret = -ERESTARTSYS;
10835 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10837 struct btrfs_block_group_cache *cache = NULL;
10838 struct btrfs_device *device;
10839 struct list_head *devices;
10844 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10848 * try to trim all FS space, our block group may start from non-zero.
10850 if (range->len == total_bytes)
10851 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10853 cache = btrfs_lookup_block_group(fs_info, range->start);
10856 if (cache->key.objectid >= (range->start + range->len)) {
10857 btrfs_put_block_group(cache);
10861 start = max(range->start, cache->key.objectid);
10862 end = min(range->start + range->len,
10863 cache->key.objectid + cache->key.offset);
10865 if (end - start >= range->minlen) {
10866 if (!block_group_cache_done(cache)) {
10867 ret = cache_block_group(cache, 0);
10869 btrfs_put_block_group(cache);
10872 ret = wait_block_group_cache_done(cache);
10874 btrfs_put_block_group(cache);
10878 ret = btrfs_trim_block_group(cache,
10884 trimmed += group_trimmed;
10886 btrfs_put_block_group(cache);
10891 cache = next_block_group(fs_info, cache);
10894 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10895 devices = &fs_info->fs_devices->alloc_list;
10896 list_for_each_entry(device, devices, dev_alloc_list) {
10897 ret = btrfs_trim_free_extents(device, range->minlen,
10902 trimmed += group_trimmed;
10904 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10906 range->len = trimmed;
10911 * btrfs_{start,end}_write_no_snapshotting() are similar to
10912 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10913 * data into the page cache through nocow before the subvolume is snapshoted,
10914 * but flush the data into disk after the snapshot creation, or to prevent
10915 * operations while snapshotting is ongoing and that cause the snapshot to be
10916 * inconsistent (writes followed by expanding truncates for example).
10918 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
10920 percpu_counter_dec(&root->subv_writers->counter);
10921 cond_wake_up(&root->subv_writers->wait);
10924 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
10926 if (atomic_read(&root->will_be_snapshotted))
10929 percpu_counter_inc(&root->subv_writers->counter);
10931 * Make sure counter is updated before we check for snapshot creation.
10934 if (atomic_read(&root->will_be_snapshotted)) {
10935 btrfs_end_write_no_snapshotting(root);
10941 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
10946 ret = btrfs_start_write_no_snapshotting(root);
10949 wait_var_event(&root->will_be_snapshotted,
10950 !atomic_read(&root->will_be_snapshotted));
10954 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
10956 struct btrfs_fs_info *fs_info = bg->fs_info;
10958 spin_lock(&fs_info->unused_bgs_lock);
10959 if (list_empty(&bg->bg_list)) {
10960 btrfs_get_block_group(bg);
10961 trace_btrfs_add_unused_block_group(bg);
10962 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
10964 spin_unlock(&fs_info->unused_bgs_lock);