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.rb_root))
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_cached(&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.rb_root) || head->extent_op) {
2452 spin_unlock(&head->lock);
2453 spin_unlock(&delayed_refs->lock);
2456 delayed_refs->num_heads--;
2457 rb_erase_cached(&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);
2505 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2506 struct btrfs_trans_handle *trans)
2508 struct btrfs_delayed_ref_root *delayed_refs =
2509 &trans->transaction->delayed_refs;
2510 struct btrfs_delayed_ref_head *head = NULL;
2513 spin_lock(&delayed_refs->lock);
2514 head = btrfs_select_ref_head(delayed_refs);
2516 spin_unlock(&delayed_refs->lock);
2521 * Grab the lock that says we are going to process all the refs for
2524 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2525 spin_unlock(&delayed_refs->lock);
2528 * We may have dropped the spin lock to get the head mutex lock, and
2529 * that might have given someone else time to free the head. If that's
2530 * true, it has been removed from our list and we can move on.
2533 head = ERR_PTR(-EAGAIN);
2538 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2539 struct btrfs_delayed_ref_head *locked_ref,
2540 unsigned long *run_refs)
2542 struct btrfs_fs_info *fs_info = trans->fs_info;
2543 struct btrfs_delayed_ref_root *delayed_refs;
2544 struct btrfs_delayed_extent_op *extent_op;
2545 struct btrfs_delayed_ref_node *ref;
2546 int must_insert_reserved = 0;
2549 delayed_refs = &trans->transaction->delayed_refs;
2551 lockdep_assert_held(&locked_ref->mutex);
2552 lockdep_assert_held(&locked_ref->lock);
2554 while ((ref = select_delayed_ref(locked_ref))) {
2556 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2557 spin_unlock(&locked_ref->lock);
2558 unselect_delayed_ref_head(delayed_refs, locked_ref);
2564 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2565 RB_CLEAR_NODE(&ref->ref_node);
2566 if (!list_empty(&ref->add_list))
2567 list_del(&ref->add_list);
2569 * When we play the delayed ref, also correct the ref_mod on
2572 switch (ref->action) {
2573 case BTRFS_ADD_DELAYED_REF:
2574 case BTRFS_ADD_DELAYED_EXTENT:
2575 locked_ref->ref_mod -= ref->ref_mod;
2577 case BTRFS_DROP_DELAYED_REF:
2578 locked_ref->ref_mod += ref->ref_mod;
2583 atomic_dec(&delayed_refs->num_entries);
2586 * Record the must_insert_reserved flag before we drop the
2589 must_insert_reserved = locked_ref->must_insert_reserved;
2590 locked_ref->must_insert_reserved = 0;
2592 extent_op = locked_ref->extent_op;
2593 locked_ref->extent_op = NULL;
2594 spin_unlock(&locked_ref->lock);
2596 ret = run_one_delayed_ref(trans, ref, extent_op,
2597 must_insert_reserved);
2599 btrfs_free_delayed_extent_op(extent_op);
2601 unselect_delayed_ref_head(delayed_refs, locked_ref);
2602 btrfs_put_delayed_ref(ref);
2603 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2608 btrfs_put_delayed_ref(ref);
2611 spin_lock(&locked_ref->lock);
2612 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2619 * Returns 0 on success or if called with an already aborted transaction.
2620 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2622 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2625 struct btrfs_fs_info *fs_info = trans->fs_info;
2626 struct btrfs_delayed_ref_root *delayed_refs;
2627 struct btrfs_delayed_ref_head *locked_ref = NULL;
2628 ktime_t start = ktime_get();
2630 unsigned long count = 0;
2631 unsigned long actual_count = 0;
2633 delayed_refs = &trans->transaction->delayed_refs;
2636 locked_ref = btrfs_obtain_ref_head(trans);
2637 if (IS_ERR_OR_NULL(locked_ref)) {
2638 if (PTR_ERR(locked_ref) == -EAGAIN) {
2647 * We need to try and merge add/drops of the same ref since we
2648 * can run into issues with relocate dropping the implicit ref
2649 * and then it being added back again before the drop can
2650 * finish. If we merged anything we need to re-loop so we can
2652 * Or we can get node references of the same type that weren't
2653 * merged when created due to bumps in the tree mod seq, and
2654 * we need to merge them to prevent adding an inline extent
2655 * backref before dropping it (triggering a BUG_ON at
2656 * insert_inline_extent_backref()).
2658 spin_lock(&locked_ref->lock);
2659 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2661 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2663 if (ret < 0 && ret != -EAGAIN) {
2665 * Error, btrfs_run_delayed_refs_for_head already
2666 * unlocked everything so just bail out
2671 * Success, perform the usual cleanup of a processed
2674 ret = cleanup_ref_head(trans, locked_ref);
2676 /* We dropped our lock, we need to loop. */
2685 * Either success case or btrfs_run_delayed_refs_for_head
2686 * returned -EAGAIN, meaning we need to select another head
2691 } while ((nr != -1 && count < nr) || locked_ref);
2694 * We don't want to include ref heads since we can have empty ref heads
2695 * and those will drastically skew our runtime down since we just do
2696 * accounting, no actual extent tree updates.
2698 if (actual_count > 0) {
2699 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2703 * We weigh the current average higher than our current runtime
2704 * to avoid large swings in the average.
2706 spin_lock(&delayed_refs->lock);
2707 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2708 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2709 spin_unlock(&delayed_refs->lock);
2714 #ifdef SCRAMBLE_DELAYED_REFS
2716 * Normally delayed refs get processed in ascending bytenr order. This
2717 * correlates in most cases to the order added. To expose dependencies on this
2718 * order, we start to process the tree in the middle instead of the beginning
2720 static u64 find_middle(struct rb_root *root)
2722 struct rb_node *n = root->rb_node;
2723 struct btrfs_delayed_ref_node *entry;
2726 u64 first = 0, last = 0;
2730 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2731 first = entry->bytenr;
2735 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2736 last = entry->bytenr;
2741 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2742 WARN_ON(!entry->in_tree);
2744 middle = entry->bytenr;
2757 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2761 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2762 sizeof(struct btrfs_extent_inline_ref));
2763 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2764 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2767 * We don't ever fill up leaves all the way so multiply by 2 just to be
2768 * closer to what we're really going to want to use.
2770 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2774 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2775 * would require to store the csums for that many bytes.
2777 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2780 u64 num_csums_per_leaf;
2783 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2784 num_csums_per_leaf = div64_u64(csum_size,
2785 (u64)btrfs_super_csum_size(fs_info->super_copy));
2786 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2787 num_csums += num_csums_per_leaf - 1;
2788 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2792 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans)
2794 struct btrfs_fs_info *fs_info = trans->fs_info;
2795 struct btrfs_block_rsv *global_rsv;
2796 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2797 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2798 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2799 u64 num_bytes, num_dirty_bgs_bytes;
2802 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2803 num_heads = heads_to_leaves(fs_info, num_heads);
2805 num_bytes += (num_heads - 1) * fs_info->nodesize;
2807 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2809 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2811 global_rsv = &fs_info->global_block_rsv;
2814 * If we can't allocate any more chunks lets make sure we have _lots_ of
2815 * wiggle room since running delayed refs can create more delayed refs.
2817 if (global_rsv->space_info->full) {
2818 num_dirty_bgs_bytes <<= 1;
2822 spin_lock(&global_rsv->lock);
2823 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2825 spin_unlock(&global_rsv->lock);
2829 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2832 atomic_read(&trans->transaction->delayed_refs.num_entries);
2837 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2838 val = num_entries * avg_runtime;
2839 if (val >= NSEC_PER_SEC)
2841 if (val >= NSEC_PER_SEC / 2)
2844 return btrfs_check_space_for_delayed_refs(trans);
2847 struct async_delayed_refs {
2848 struct btrfs_root *root;
2853 struct completion wait;
2854 struct btrfs_work work;
2857 static inline struct async_delayed_refs *
2858 to_async_delayed_refs(struct btrfs_work *work)
2860 return container_of(work, struct async_delayed_refs, work);
2863 static void delayed_ref_async_start(struct btrfs_work *work)
2865 struct async_delayed_refs *async = to_async_delayed_refs(work);
2866 struct btrfs_trans_handle *trans;
2867 struct btrfs_fs_info *fs_info = async->root->fs_info;
2870 /* if the commit is already started, we don't need to wait here */
2871 if (btrfs_transaction_blocked(fs_info))
2874 trans = btrfs_join_transaction(async->root);
2875 if (IS_ERR(trans)) {
2876 async->error = PTR_ERR(trans);
2881 * trans->sync means that when we call end_transaction, we won't
2882 * wait on delayed refs
2886 /* Don't bother flushing if we got into a different transaction */
2887 if (trans->transid > async->transid)
2890 ret = btrfs_run_delayed_refs(trans, async->count);
2894 ret = btrfs_end_transaction(trans);
2895 if (ret && !async->error)
2899 complete(&async->wait);
2904 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2905 unsigned long count, u64 transid, int wait)
2907 struct async_delayed_refs *async;
2910 async = kmalloc(sizeof(*async), GFP_NOFS);
2914 async->root = fs_info->tree_root;
2915 async->count = count;
2917 async->transid = transid;
2922 init_completion(&async->wait);
2924 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2925 delayed_ref_async_start, NULL, NULL);
2927 btrfs_queue_work(fs_info->extent_workers, &async->work);
2930 wait_for_completion(&async->wait);
2939 * this starts processing the delayed reference count updates and
2940 * extent insertions we have queued up so far. count can be
2941 * 0, which means to process everything in the tree at the start
2942 * of the run (but not newly added entries), or it can be some target
2943 * number you'd like to process.
2945 * Returns 0 on success or if called with an aborted transaction
2946 * Returns <0 on error and aborts the transaction
2948 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2949 unsigned long count)
2951 struct btrfs_fs_info *fs_info = trans->fs_info;
2952 struct rb_node *node;
2953 struct btrfs_delayed_ref_root *delayed_refs;
2954 struct btrfs_delayed_ref_head *head;
2956 int run_all = count == (unsigned long)-1;
2957 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2959 /* We'll clean this up in btrfs_cleanup_transaction */
2963 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2966 delayed_refs = &trans->transaction->delayed_refs;
2968 count = atomic_read(&delayed_refs->num_entries) * 2;
2971 #ifdef SCRAMBLE_DELAYED_REFS
2972 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2974 trans->can_flush_pending_bgs = false;
2975 ret = __btrfs_run_delayed_refs(trans, count);
2977 btrfs_abort_transaction(trans, ret);
2982 if (!list_empty(&trans->new_bgs))
2983 btrfs_create_pending_block_groups(trans);
2985 spin_lock(&delayed_refs->lock);
2986 node = rb_first_cached(&delayed_refs->href_root);
2988 spin_unlock(&delayed_refs->lock);
2991 head = rb_entry(node, struct btrfs_delayed_ref_head,
2993 refcount_inc(&head->refs);
2994 spin_unlock(&delayed_refs->lock);
2996 /* Mutex was contended, block until it's released and retry. */
2997 mutex_lock(&head->mutex);
2998 mutex_unlock(&head->mutex);
3000 btrfs_put_delayed_ref_head(head);
3005 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3009 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3010 struct btrfs_fs_info *fs_info,
3011 u64 bytenr, u64 num_bytes, u64 flags,
3012 int level, int is_data)
3014 struct btrfs_delayed_extent_op *extent_op;
3017 extent_op = btrfs_alloc_delayed_extent_op();
3021 extent_op->flags_to_set = flags;
3022 extent_op->update_flags = true;
3023 extent_op->update_key = false;
3024 extent_op->is_data = is_data ? true : false;
3025 extent_op->level = level;
3027 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3028 num_bytes, extent_op);
3030 btrfs_free_delayed_extent_op(extent_op);
3034 static noinline int check_delayed_ref(struct btrfs_root *root,
3035 struct btrfs_path *path,
3036 u64 objectid, u64 offset, u64 bytenr)
3038 struct btrfs_delayed_ref_head *head;
3039 struct btrfs_delayed_ref_node *ref;
3040 struct btrfs_delayed_data_ref *data_ref;
3041 struct btrfs_delayed_ref_root *delayed_refs;
3042 struct btrfs_transaction *cur_trans;
3043 struct rb_node *node;
3046 spin_lock(&root->fs_info->trans_lock);
3047 cur_trans = root->fs_info->running_transaction;
3049 refcount_inc(&cur_trans->use_count);
3050 spin_unlock(&root->fs_info->trans_lock);
3054 delayed_refs = &cur_trans->delayed_refs;
3055 spin_lock(&delayed_refs->lock);
3056 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3058 spin_unlock(&delayed_refs->lock);
3059 btrfs_put_transaction(cur_trans);
3063 if (!mutex_trylock(&head->mutex)) {
3064 refcount_inc(&head->refs);
3065 spin_unlock(&delayed_refs->lock);
3067 btrfs_release_path(path);
3070 * Mutex was contended, block until it's released and let
3073 mutex_lock(&head->mutex);
3074 mutex_unlock(&head->mutex);
3075 btrfs_put_delayed_ref_head(head);
3076 btrfs_put_transaction(cur_trans);
3079 spin_unlock(&delayed_refs->lock);
3081 spin_lock(&head->lock);
3083 * XXX: We should replace this with a proper search function in the
3086 for (node = rb_first_cached(&head->ref_tree); node;
3087 node = rb_next(node)) {
3088 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3089 /* If it's a shared ref we know a cross reference exists */
3090 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3095 data_ref = btrfs_delayed_node_to_data_ref(ref);
3098 * If our ref doesn't match the one we're currently looking at
3099 * then we have a cross reference.
3101 if (data_ref->root != root->root_key.objectid ||
3102 data_ref->objectid != objectid ||
3103 data_ref->offset != offset) {
3108 spin_unlock(&head->lock);
3109 mutex_unlock(&head->mutex);
3110 btrfs_put_transaction(cur_trans);
3114 static noinline int check_committed_ref(struct btrfs_root *root,
3115 struct btrfs_path *path,
3116 u64 objectid, u64 offset, u64 bytenr)
3118 struct btrfs_fs_info *fs_info = root->fs_info;
3119 struct btrfs_root *extent_root = fs_info->extent_root;
3120 struct extent_buffer *leaf;
3121 struct btrfs_extent_data_ref *ref;
3122 struct btrfs_extent_inline_ref *iref;
3123 struct btrfs_extent_item *ei;
3124 struct btrfs_key key;
3129 key.objectid = bytenr;
3130 key.offset = (u64)-1;
3131 key.type = BTRFS_EXTENT_ITEM_KEY;
3133 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3136 BUG_ON(ret == 0); /* Corruption */
3139 if (path->slots[0] == 0)
3143 leaf = path->nodes[0];
3144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3146 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3150 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3151 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3153 if (item_size != sizeof(*ei) +
3154 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3157 if (btrfs_extent_generation(leaf, ei) <=
3158 btrfs_root_last_snapshot(&root->root_item))
3161 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3163 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3164 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3167 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3168 if (btrfs_extent_refs(leaf, ei) !=
3169 btrfs_extent_data_ref_count(leaf, ref) ||
3170 btrfs_extent_data_ref_root(leaf, ref) !=
3171 root->root_key.objectid ||
3172 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3173 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3181 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3184 struct btrfs_path *path;
3187 path = btrfs_alloc_path();
3192 ret = check_committed_ref(root, path, objectid,
3194 if (ret && ret != -ENOENT)
3197 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3198 } while (ret == -EAGAIN);
3201 btrfs_free_path(path);
3202 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3207 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *root,
3209 struct extent_buffer *buf,
3210 int full_backref, int inc)
3212 struct btrfs_fs_info *fs_info = root->fs_info;
3218 struct btrfs_key key;
3219 struct btrfs_file_extent_item *fi;
3223 int (*process_func)(struct btrfs_trans_handle *,
3224 struct btrfs_root *,
3225 u64, u64, u64, u64, u64, u64);
3228 if (btrfs_is_testing(fs_info))
3231 ref_root = btrfs_header_owner(buf);
3232 nritems = btrfs_header_nritems(buf);
3233 level = btrfs_header_level(buf);
3235 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3239 process_func = btrfs_inc_extent_ref;
3241 process_func = btrfs_free_extent;
3244 parent = buf->start;
3248 for (i = 0; i < nritems; i++) {
3250 btrfs_item_key_to_cpu(buf, &key, i);
3251 if (key.type != BTRFS_EXTENT_DATA_KEY)
3253 fi = btrfs_item_ptr(buf, i,
3254 struct btrfs_file_extent_item);
3255 if (btrfs_file_extent_type(buf, fi) ==
3256 BTRFS_FILE_EXTENT_INLINE)
3258 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3262 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3263 key.offset -= btrfs_file_extent_offset(buf, fi);
3264 ret = process_func(trans, root, bytenr, num_bytes,
3265 parent, ref_root, key.objectid,
3270 bytenr = btrfs_node_blockptr(buf, i);
3271 num_bytes = fs_info->nodesize;
3272 ret = process_func(trans, root, bytenr, num_bytes,
3273 parent, ref_root, level - 1, 0);
3283 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3284 struct extent_buffer *buf, int full_backref)
3286 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3289 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3290 struct extent_buffer *buf, int full_backref)
3292 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3295 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3296 struct btrfs_fs_info *fs_info,
3297 struct btrfs_path *path,
3298 struct btrfs_block_group_cache *cache)
3301 struct btrfs_root *extent_root = fs_info->extent_root;
3303 struct extent_buffer *leaf;
3305 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3312 leaf = path->nodes[0];
3313 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3314 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3315 btrfs_mark_buffer_dirty(leaf);
3317 btrfs_release_path(path);
3322 static struct btrfs_block_group_cache *
3323 next_block_group(struct btrfs_fs_info *fs_info,
3324 struct btrfs_block_group_cache *cache)
3326 struct rb_node *node;
3328 spin_lock(&fs_info->block_group_cache_lock);
3330 /* If our block group was removed, we need a full search. */
3331 if (RB_EMPTY_NODE(&cache->cache_node)) {
3332 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3334 spin_unlock(&fs_info->block_group_cache_lock);
3335 btrfs_put_block_group(cache);
3336 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3338 node = rb_next(&cache->cache_node);
3339 btrfs_put_block_group(cache);
3341 cache = rb_entry(node, struct btrfs_block_group_cache,
3343 btrfs_get_block_group(cache);
3346 spin_unlock(&fs_info->block_group_cache_lock);
3350 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3351 struct btrfs_trans_handle *trans,
3352 struct btrfs_path *path)
3354 struct btrfs_fs_info *fs_info = block_group->fs_info;
3355 struct btrfs_root *root = fs_info->tree_root;
3356 struct inode *inode = NULL;
3357 struct extent_changeset *data_reserved = NULL;
3359 int dcs = BTRFS_DC_ERROR;
3365 * If this block group is smaller than 100 megs don't bother caching the
3368 if (block_group->key.offset < (100 * SZ_1M)) {
3369 spin_lock(&block_group->lock);
3370 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3371 spin_unlock(&block_group->lock);
3378 inode = lookup_free_space_inode(fs_info, block_group, path);
3379 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3380 ret = PTR_ERR(inode);
3381 btrfs_release_path(path);
3385 if (IS_ERR(inode)) {
3389 if (block_group->ro)
3392 ret = create_free_space_inode(fs_info, trans, block_group,
3400 * We want to set the generation to 0, that way if anything goes wrong
3401 * from here on out we know not to trust this cache when we load up next
3404 BTRFS_I(inode)->generation = 0;
3405 ret = btrfs_update_inode(trans, root, inode);
3408 * So theoretically we could recover from this, simply set the
3409 * super cache generation to 0 so we know to invalidate the
3410 * cache, but then we'd have to keep track of the block groups
3411 * that fail this way so we know we _have_ to reset this cache
3412 * before the next commit or risk reading stale cache. So to
3413 * limit our exposure to horrible edge cases lets just abort the
3414 * transaction, this only happens in really bad situations
3417 btrfs_abort_transaction(trans, ret);
3422 /* We've already setup this transaction, go ahead and exit */
3423 if (block_group->cache_generation == trans->transid &&
3424 i_size_read(inode)) {
3425 dcs = BTRFS_DC_SETUP;
3429 if (i_size_read(inode) > 0) {
3430 ret = btrfs_check_trunc_cache_free_space(fs_info,
3431 &fs_info->global_block_rsv);
3435 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3440 spin_lock(&block_group->lock);
3441 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3442 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3444 * don't bother trying to write stuff out _if_
3445 * a) we're not cached,
3446 * b) we're with nospace_cache mount option,
3447 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3449 dcs = BTRFS_DC_WRITTEN;
3450 spin_unlock(&block_group->lock);
3453 spin_unlock(&block_group->lock);
3456 * We hit an ENOSPC when setting up the cache in this transaction, just
3457 * skip doing the setup, we've already cleared the cache so we're safe.
3459 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3465 * Try to preallocate enough space based on how big the block group is.
3466 * Keep in mind this has to include any pinned space which could end up
3467 * taking up quite a bit since it's not folded into the other space
3470 num_pages = div_u64(block_group->key.offset, SZ_256M);
3475 num_pages *= PAGE_SIZE;
3477 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3481 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3482 num_pages, num_pages,
3485 * Our cache requires contiguous chunks so that we don't modify a bunch
3486 * of metadata or split extents when writing the cache out, which means
3487 * we can enospc if we are heavily fragmented in addition to just normal
3488 * out of space conditions. So if we hit this just skip setting up any
3489 * other block groups for this transaction, maybe we'll unpin enough
3490 * space the next time around.
3493 dcs = BTRFS_DC_SETUP;
3494 else if (ret == -ENOSPC)
3495 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3500 btrfs_release_path(path);
3502 spin_lock(&block_group->lock);
3503 if (!ret && dcs == BTRFS_DC_SETUP)
3504 block_group->cache_generation = trans->transid;
3505 block_group->disk_cache_state = dcs;
3506 spin_unlock(&block_group->lock);
3508 extent_changeset_free(data_reserved);
3512 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3513 struct btrfs_fs_info *fs_info)
3515 struct btrfs_block_group_cache *cache, *tmp;
3516 struct btrfs_transaction *cur_trans = trans->transaction;
3517 struct btrfs_path *path;
3519 if (list_empty(&cur_trans->dirty_bgs) ||
3520 !btrfs_test_opt(fs_info, SPACE_CACHE))
3523 path = btrfs_alloc_path();
3527 /* Could add new block groups, use _safe just in case */
3528 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3530 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3531 cache_save_setup(cache, trans, path);
3534 btrfs_free_path(path);
3539 * transaction commit does final block group cache writeback during a
3540 * critical section where nothing is allowed to change the FS. This is
3541 * required in order for the cache to actually match the block group,
3542 * but can introduce a lot of latency into the commit.
3544 * So, btrfs_start_dirty_block_groups is here to kick off block group
3545 * cache IO. There's a chance we'll have to redo some of it if the
3546 * block group changes again during the commit, but it greatly reduces
3547 * the commit latency by getting rid of the easy block groups while
3548 * we're still allowing others to join the commit.
3550 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3552 struct btrfs_fs_info *fs_info = trans->fs_info;
3553 struct btrfs_block_group_cache *cache;
3554 struct btrfs_transaction *cur_trans = trans->transaction;
3557 struct btrfs_path *path = NULL;
3559 struct list_head *io = &cur_trans->io_bgs;
3560 int num_started = 0;
3563 spin_lock(&cur_trans->dirty_bgs_lock);
3564 if (list_empty(&cur_trans->dirty_bgs)) {
3565 spin_unlock(&cur_trans->dirty_bgs_lock);
3568 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3569 spin_unlock(&cur_trans->dirty_bgs_lock);
3573 * make sure all the block groups on our dirty list actually
3576 btrfs_create_pending_block_groups(trans);
3579 path = btrfs_alloc_path();
3585 * cache_write_mutex is here only to save us from balance or automatic
3586 * removal of empty block groups deleting this block group while we are
3587 * writing out the cache
3589 mutex_lock(&trans->transaction->cache_write_mutex);
3590 while (!list_empty(&dirty)) {
3591 cache = list_first_entry(&dirty,
3592 struct btrfs_block_group_cache,
3595 * this can happen if something re-dirties a block
3596 * group that is already under IO. Just wait for it to
3597 * finish and then do it all again
3599 if (!list_empty(&cache->io_list)) {
3600 list_del_init(&cache->io_list);
3601 btrfs_wait_cache_io(trans, cache, path);
3602 btrfs_put_block_group(cache);
3607 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3608 * if it should update the cache_state. Don't delete
3609 * until after we wait.
3611 * Since we're not running in the commit critical section
3612 * we need the dirty_bgs_lock to protect from update_block_group
3614 spin_lock(&cur_trans->dirty_bgs_lock);
3615 list_del_init(&cache->dirty_list);
3616 spin_unlock(&cur_trans->dirty_bgs_lock);
3620 cache_save_setup(cache, trans, path);
3622 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3623 cache->io_ctl.inode = NULL;
3624 ret = btrfs_write_out_cache(fs_info, trans,
3626 if (ret == 0 && cache->io_ctl.inode) {
3631 * The cache_write_mutex is protecting the
3632 * io_list, also refer to the definition of
3633 * btrfs_transaction::io_bgs for more details
3635 list_add_tail(&cache->io_list, io);
3638 * if we failed to write the cache, the
3639 * generation will be bad and life goes on
3645 ret = write_one_cache_group(trans, fs_info,
3648 * Our block group might still be attached to the list
3649 * of new block groups in the transaction handle of some
3650 * other task (struct btrfs_trans_handle->new_bgs). This
3651 * means its block group item isn't yet in the extent
3652 * tree. If this happens ignore the error, as we will
3653 * try again later in the critical section of the
3654 * transaction commit.
3656 if (ret == -ENOENT) {
3658 spin_lock(&cur_trans->dirty_bgs_lock);
3659 if (list_empty(&cache->dirty_list)) {
3660 list_add_tail(&cache->dirty_list,
3661 &cur_trans->dirty_bgs);
3662 btrfs_get_block_group(cache);
3664 spin_unlock(&cur_trans->dirty_bgs_lock);
3666 btrfs_abort_transaction(trans, ret);
3670 /* if its not on the io list, we need to put the block group */
3672 btrfs_put_block_group(cache);
3678 * Avoid blocking other tasks for too long. It might even save
3679 * us from writing caches for block groups that are going to be
3682 mutex_unlock(&trans->transaction->cache_write_mutex);
3683 mutex_lock(&trans->transaction->cache_write_mutex);
3685 mutex_unlock(&trans->transaction->cache_write_mutex);
3688 * go through delayed refs for all the stuff we've just kicked off
3689 * and then loop back (just once)
3691 ret = btrfs_run_delayed_refs(trans, 0);
3692 if (!ret && loops == 0) {
3694 spin_lock(&cur_trans->dirty_bgs_lock);
3695 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3697 * dirty_bgs_lock protects us from concurrent block group
3698 * deletes too (not just cache_write_mutex).
3700 if (!list_empty(&dirty)) {
3701 spin_unlock(&cur_trans->dirty_bgs_lock);
3704 spin_unlock(&cur_trans->dirty_bgs_lock);
3705 } else if (ret < 0) {
3706 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3709 btrfs_free_path(path);
3713 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3714 struct btrfs_fs_info *fs_info)
3716 struct btrfs_block_group_cache *cache;
3717 struct btrfs_transaction *cur_trans = trans->transaction;
3720 struct btrfs_path *path;
3721 struct list_head *io = &cur_trans->io_bgs;
3722 int num_started = 0;
3724 path = btrfs_alloc_path();
3729 * Even though we are in the critical section of the transaction commit,
3730 * we can still have concurrent tasks adding elements to this
3731 * transaction's list of dirty block groups. These tasks correspond to
3732 * endio free space workers started when writeback finishes for a
3733 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3734 * allocate new block groups as a result of COWing nodes of the root
3735 * tree when updating the free space inode. The writeback for the space
3736 * caches is triggered by an earlier call to
3737 * btrfs_start_dirty_block_groups() and iterations of the following
3739 * Also we want to do the cache_save_setup first and then run the
3740 * delayed refs to make sure we have the best chance at doing this all
3743 spin_lock(&cur_trans->dirty_bgs_lock);
3744 while (!list_empty(&cur_trans->dirty_bgs)) {
3745 cache = list_first_entry(&cur_trans->dirty_bgs,
3746 struct btrfs_block_group_cache,
3750 * this can happen if cache_save_setup re-dirties a block
3751 * group that is already under IO. Just wait for it to
3752 * finish and then do it all again
3754 if (!list_empty(&cache->io_list)) {
3755 spin_unlock(&cur_trans->dirty_bgs_lock);
3756 list_del_init(&cache->io_list);
3757 btrfs_wait_cache_io(trans, cache, path);
3758 btrfs_put_block_group(cache);
3759 spin_lock(&cur_trans->dirty_bgs_lock);
3763 * don't remove from the dirty list until after we've waited
3766 list_del_init(&cache->dirty_list);
3767 spin_unlock(&cur_trans->dirty_bgs_lock);
3770 cache_save_setup(cache, trans, path);
3773 ret = btrfs_run_delayed_refs(trans,
3774 (unsigned long) -1);
3776 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3777 cache->io_ctl.inode = NULL;
3778 ret = btrfs_write_out_cache(fs_info, trans,
3780 if (ret == 0 && cache->io_ctl.inode) {
3783 list_add_tail(&cache->io_list, io);
3786 * if we failed to write the cache, the
3787 * generation will be bad and life goes on
3793 ret = write_one_cache_group(trans, fs_info,
3796 * One of the free space endio workers might have
3797 * created a new block group while updating a free space
3798 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3799 * and hasn't released its transaction handle yet, in
3800 * which case the new block group is still attached to
3801 * its transaction handle and its creation has not
3802 * finished yet (no block group item in the extent tree
3803 * yet, etc). If this is the case, wait for all free
3804 * space endio workers to finish and retry. This is a
3805 * a very rare case so no need for a more efficient and
3808 if (ret == -ENOENT) {
3809 wait_event(cur_trans->writer_wait,
3810 atomic_read(&cur_trans->num_writers) == 1);
3811 ret = write_one_cache_group(trans, fs_info,
3815 btrfs_abort_transaction(trans, ret);
3818 /* if its not on the io list, we need to put the block group */
3820 btrfs_put_block_group(cache);
3821 spin_lock(&cur_trans->dirty_bgs_lock);
3823 spin_unlock(&cur_trans->dirty_bgs_lock);
3826 * Refer to the definition of io_bgs member for details why it's safe
3827 * to use it without any locking
3829 while (!list_empty(io)) {
3830 cache = list_first_entry(io, struct btrfs_block_group_cache,
3832 list_del_init(&cache->io_list);
3833 btrfs_wait_cache_io(trans, cache, path);
3834 btrfs_put_block_group(cache);
3837 btrfs_free_path(path);
3841 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3843 struct btrfs_block_group_cache *block_group;
3846 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3847 if (!block_group || block_group->ro)
3850 btrfs_put_block_group(block_group);
3854 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3856 struct btrfs_block_group_cache *bg;
3859 bg = btrfs_lookup_block_group(fs_info, bytenr);
3863 spin_lock(&bg->lock);
3867 atomic_inc(&bg->nocow_writers);
3868 spin_unlock(&bg->lock);
3870 /* no put on block group, done by btrfs_dec_nocow_writers */
3872 btrfs_put_block_group(bg);
3878 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3880 struct btrfs_block_group_cache *bg;
3882 bg = btrfs_lookup_block_group(fs_info, bytenr);
3884 if (atomic_dec_and_test(&bg->nocow_writers))
3885 wake_up_var(&bg->nocow_writers);
3887 * Once for our lookup and once for the lookup done by a previous call
3888 * to btrfs_inc_nocow_writers()
3890 btrfs_put_block_group(bg);
3891 btrfs_put_block_group(bg);
3894 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3896 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3899 static const char *alloc_name(u64 flags)
3902 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3904 case BTRFS_BLOCK_GROUP_METADATA:
3906 case BTRFS_BLOCK_GROUP_DATA:
3908 case BTRFS_BLOCK_GROUP_SYSTEM:
3912 return "invalid-combination";
3916 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3919 struct btrfs_space_info *space_info;
3923 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3927 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3934 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3935 INIT_LIST_HEAD(&space_info->block_groups[i]);
3936 init_rwsem(&space_info->groups_sem);
3937 spin_lock_init(&space_info->lock);
3938 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3939 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3940 init_waitqueue_head(&space_info->wait);
3941 INIT_LIST_HEAD(&space_info->ro_bgs);
3942 INIT_LIST_HEAD(&space_info->tickets);
3943 INIT_LIST_HEAD(&space_info->priority_tickets);
3945 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3946 info->space_info_kobj, "%s",
3947 alloc_name(space_info->flags));
3949 percpu_counter_destroy(&space_info->total_bytes_pinned);
3954 list_add_rcu(&space_info->list, &info->space_info);
3955 if (flags & BTRFS_BLOCK_GROUP_DATA)
3956 info->data_sinfo = space_info;
3961 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
3962 u64 total_bytes, u64 bytes_used,
3964 struct btrfs_space_info **space_info)
3966 struct btrfs_space_info *found;
3969 factor = btrfs_bg_type_to_factor(flags);
3971 found = __find_space_info(info, flags);
3973 spin_lock(&found->lock);
3974 found->total_bytes += total_bytes;
3975 found->disk_total += total_bytes * factor;
3976 found->bytes_used += bytes_used;
3977 found->disk_used += bytes_used * factor;
3978 found->bytes_readonly += bytes_readonly;
3979 if (total_bytes > 0)
3981 space_info_add_new_bytes(info, found, total_bytes -
3982 bytes_used - bytes_readonly);
3983 spin_unlock(&found->lock);
3984 *space_info = found;
3987 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3989 u64 extra_flags = chunk_to_extended(flags) &
3990 BTRFS_EXTENDED_PROFILE_MASK;
3992 write_seqlock(&fs_info->profiles_lock);
3993 if (flags & BTRFS_BLOCK_GROUP_DATA)
3994 fs_info->avail_data_alloc_bits |= extra_flags;
3995 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3996 fs_info->avail_metadata_alloc_bits |= extra_flags;
3997 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3998 fs_info->avail_system_alloc_bits |= extra_flags;
3999 write_sequnlock(&fs_info->profiles_lock);
4003 * returns target flags in extended format or 0 if restripe for this
4004 * chunk_type is not in progress
4006 * should be called with balance_lock held
4008 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4010 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4016 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4017 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4018 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4019 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4020 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4021 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4022 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4023 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4024 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4031 * @flags: available profiles in extended format (see ctree.h)
4033 * Returns reduced profile in chunk format. If profile changing is in
4034 * progress (either running or paused) picks the target profile (if it's
4035 * already available), otherwise falls back to plain reducing.
4037 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4039 u64 num_devices = fs_info->fs_devices->rw_devices;
4045 * see if restripe for this chunk_type is in progress, if so
4046 * try to reduce to the target profile
4048 spin_lock(&fs_info->balance_lock);
4049 target = get_restripe_target(fs_info, flags);
4051 /* pick target profile only if it's already available */
4052 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4053 spin_unlock(&fs_info->balance_lock);
4054 return extended_to_chunk(target);
4057 spin_unlock(&fs_info->balance_lock);
4059 /* First, mask out the RAID levels which aren't possible */
4060 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4061 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4062 allowed |= btrfs_raid_array[raid_type].bg_flag;
4066 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4067 allowed = BTRFS_BLOCK_GROUP_RAID6;
4068 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4069 allowed = BTRFS_BLOCK_GROUP_RAID5;
4070 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4071 allowed = BTRFS_BLOCK_GROUP_RAID10;
4072 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4073 allowed = BTRFS_BLOCK_GROUP_RAID1;
4074 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4075 allowed = BTRFS_BLOCK_GROUP_RAID0;
4077 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4079 return extended_to_chunk(flags | allowed);
4082 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4089 seq = read_seqbegin(&fs_info->profiles_lock);
4091 if (flags & BTRFS_BLOCK_GROUP_DATA)
4092 flags |= fs_info->avail_data_alloc_bits;
4093 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4094 flags |= fs_info->avail_system_alloc_bits;
4095 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4096 flags |= fs_info->avail_metadata_alloc_bits;
4097 } while (read_seqretry(&fs_info->profiles_lock, seq));
4099 return btrfs_reduce_alloc_profile(fs_info, flags);
4102 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4104 struct btrfs_fs_info *fs_info = root->fs_info;
4109 flags = BTRFS_BLOCK_GROUP_DATA;
4110 else if (root == fs_info->chunk_root)
4111 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4113 flags = BTRFS_BLOCK_GROUP_METADATA;
4115 ret = get_alloc_profile(fs_info, flags);
4119 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4121 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4124 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4126 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4129 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4131 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4134 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4135 bool may_use_included)
4138 return s_info->bytes_used + s_info->bytes_reserved +
4139 s_info->bytes_pinned + s_info->bytes_readonly +
4140 (may_use_included ? s_info->bytes_may_use : 0);
4143 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4145 struct btrfs_root *root = inode->root;
4146 struct btrfs_fs_info *fs_info = root->fs_info;
4147 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4150 int need_commit = 2;
4151 int have_pinned_space;
4153 /* make sure bytes are sectorsize aligned */
4154 bytes = ALIGN(bytes, fs_info->sectorsize);
4156 if (btrfs_is_free_space_inode(inode)) {
4158 ASSERT(current->journal_info);
4162 /* make sure we have enough space to handle the data first */
4163 spin_lock(&data_sinfo->lock);
4164 used = btrfs_space_info_used(data_sinfo, true);
4166 if (used + bytes > data_sinfo->total_bytes) {
4167 struct btrfs_trans_handle *trans;
4170 * if we don't have enough free bytes in this space then we need
4171 * to alloc a new chunk.
4173 if (!data_sinfo->full) {
4176 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4177 spin_unlock(&data_sinfo->lock);
4179 alloc_target = btrfs_data_alloc_profile(fs_info);
4181 * It is ugly that we don't call nolock join
4182 * transaction for the free space inode case here.
4183 * But it is safe because we only do the data space
4184 * reservation for the free space cache in the
4185 * transaction context, the common join transaction
4186 * just increase the counter of the current transaction
4187 * handler, doesn't try to acquire the trans_lock of
4190 trans = btrfs_join_transaction(root);
4192 return PTR_ERR(trans);
4194 ret = do_chunk_alloc(trans, alloc_target,
4195 CHUNK_ALLOC_NO_FORCE);
4196 btrfs_end_transaction(trans);
4201 have_pinned_space = 1;
4210 * If we don't have enough pinned space to deal with this
4211 * allocation, and no removed chunk in current transaction,
4212 * don't bother committing the transaction.
4214 have_pinned_space = __percpu_counter_compare(
4215 &data_sinfo->total_bytes_pinned,
4216 used + bytes - data_sinfo->total_bytes,
4217 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4218 spin_unlock(&data_sinfo->lock);
4220 /* commit the current transaction and try again */
4225 if (need_commit > 0) {
4226 btrfs_start_delalloc_roots(fs_info, -1);
4227 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4231 trans = btrfs_join_transaction(root);
4233 return PTR_ERR(trans);
4234 if (have_pinned_space >= 0 ||
4235 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4236 &trans->transaction->flags) ||
4238 ret = btrfs_commit_transaction(trans);
4242 * The cleaner kthread might still be doing iput
4243 * operations. Wait for it to finish so that
4244 * more space is released.
4246 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4247 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4250 btrfs_end_transaction(trans);
4254 trace_btrfs_space_reservation(fs_info,
4255 "space_info:enospc",
4256 data_sinfo->flags, bytes, 1);
4259 data_sinfo->bytes_may_use += bytes;
4260 trace_btrfs_space_reservation(fs_info, "space_info",
4261 data_sinfo->flags, bytes, 1);
4262 spin_unlock(&data_sinfo->lock);
4267 int btrfs_check_data_free_space(struct inode *inode,
4268 struct extent_changeset **reserved, u64 start, u64 len)
4270 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4273 /* align the range */
4274 len = round_up(start + len, fs_info->sectorsize) -
4275 round_down(start, fs_info->sectorsize);
4276 start = round_down(start, fs_info->sectorsize);
4278 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4282 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4283 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4285 btrfs_free_reserved_data_space_noquota(inode, start, len);
4292 * Called if we need to clear a data reservation for this inode
4293 * Normally in a error case.
4295 * This one will *NOT* use accurate qgroup reserved space API, just for case
4296 * which we can't sleep and is sure it won't affect qgroup reserved space.
4297 * Like clear_bit_hook().
4299 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4302 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4303 struct btrfs_space_info *data_sinfo;
4305 /* Make sure the range is aligned to sectorsize */
4306 len = round_up(start + len, fs_info->sectorsize) -
4307 round_down(start, fs_info->sectorsize);
4308 start = round_down(start, fs_info->sectorsize);
4310 data_sinfo = fs_info->data_sinfo;
4311 spin_lock(&data_sinfo->lock);
4312 if (WARN_ON(data_sinfo->bytes_may_use < len))
4313 data_sinfo->bytes_may_use = 0;
4315 data_sinfo->bytes_may_use -= len;
4316 trace_btrfs_space_reservation(fs_info, "space_info",
4317 data_sinfo->flags, len, 0);
4318 spin_unlock(&data_sinfo->lock);
4322 * Called if we need to clear a data reservation for this inode
4323 * Normally in a error case.
4325 * This one will handle the per-inode data rsv map for accurate reserved
4328 void btrfs_free_reserved_data_space(struct inode *inode,
4329 struct extent_changeset *reserved, u64 start, u64 len)
4331 struct btrfs_root *root = BTRFS_I(inode)->root;
4333 /* Make sure the range is aligned to sectorsize */
4334 len = round_up(start + len, root->fs_info->sectorsize) -
4335 round_down(start, root->fs_info->sectorsize);
4336 start = round_down(start, root->fs_info->sectorsize);
4338 btrfs_free_reserved_data_space_noquota(inode, start, len);
4339 btrfs_qgroup_free_data(inode, reserved, start, len);
4342 static void force_metadata_allocation(struct btrfs_fs_info *info)
4344 struct list_head *head = &info->space_info;
4345 struct btrfs_space_info *found;
4348 list_for_each_entry_rcu(found, head, list) {
4349 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4350 found->force_alloc = CHUNK_ALLOC_FORCE;
4355 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4357 return (global->size << 1);
4360 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4361 struct btrfs_space_info *sinfo, int force)
4363 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4364 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4367 if (force == CHUNK_ALLOC_FORCE)
4371 * We need to take into account the global rsv because for all intents
4372 * and purposes it's used space. Don't worry about locking the
4373 * global_rsv, it doesn't change except when the transaction commits.
4375 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4376 bytes_used += calc_global_rsv_need_space(global_rsv);
4379 * in limited mode, we want to have some free space up to
4380 * about 1% of the FS size.
4382 if (force == CHUNK_ALLOC_LIMITED) {
4383 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4384 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4386 if (sinfo->total_bytes - bytes_used < thresh)
4390 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4395 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4399 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4400 BTRFS_BLOCK_GROUP_RAID0 |
4401 BTRFS_BLOCK_GROUP_RAID5 |
4402 BTRFS_BLOCK_GROUP_RAID6))
4403 num_dev = fs_info->fs_devices->rw_devices;
4404 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4407 num_dev = 1; /* DUP or single */
4413 * If @is_allocation is true, reserve space in the system space info necessary
4414 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4417 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4419 struct btrfs_fs_info *fs_info = trans->fs_info;
4420 struct btrfs_space_info *info;
4427 * Needed because we can end up allocating a system chunk and for an
4428 * atomic and race free space reservation in the chunk block reserve.
4430 lockdep_assert_held(&fs_info->chunk_mutex);
4432 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4433 spin_lock(&info->lock);
4434 left = info->total_bytes - btrfs_space_info_used(info, true);
4435 spin_unlock(&info->lock);
4437 num_devs = get_profile_num_devs(fs_info, type);
4439 /* num_devs device items to update and 1 chunk item to add or remove */
4440 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4441 btrfs_calc_trans_metadata_size(fs_info, 1);
4443 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4444 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4445 left, thresh, type);
4446 dump_space_info(fs_info, info, 0, 0);
4449 if (left < thresh) {
4450 u64 flags = btrfs_system_alloc_profile(fs_info);
4453 * Ignore failure to create system chunk. We might end up not
4454 * needing it, as we might not need to COW all nodes/leafs from
4455 * the paths we visit in the chunk tree (they were already COWed
4456 * or created in the current transaction for example).
4458 ret = btrfs_alloc_chunk(trans, flags);
4462 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4463 &fs_info->chunk_block_rsv,
4464 thresh, BTRFS_RESERVE_NO_FLUSH);
4466 trans->chunk_bytes_reserved += thresh;
4471 * If force is CHUNK_ALLOC_FORCE:
4472 * - return 1 if it successfully allocates a chunk,
4473 * - return errors including -ENOSPC otherwise.
4474 * If force is NOT CHUNK_ALLOC_FORCE:
4475 * - return 0 if it doesn't need to allocate a new chunk,
4476 * - return 1 if it successfully allocates a chunk,
4477 * - return errors including -ENOSPC otherwise.
4479 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4482 struct btrfs_fs_info *fs_info = trans->fs_info;
4483 struct btrfs_space_info *space_info;
4484 bool wait_for_alloc = false;
4485 bool should_alloc = false;
4488 /* Don't re-enter if we're already allocating a chunk */
4489 if (trans->allocating_chunk)
4492 space_info = __find_space_info(fs_info, flags);
4496 spin_lock(&space_info->lock);
4497 if (force < space_info->force_alloc)
4498 force = space_info->force_alloc;
4499 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4500 if (space_info->full) {
4501 /* No more free physical space */
4506 spin_unlock(&space_info->lock);
4508 } else if (!should_alloc) {
4509 spin_unlock(&space_info->lock);
4511 } else if (space_info->chunk_alloc) {
4513 * Someone is already allocating, so we need to block
4514 * until this someone is finished and then loop to
4515 * recheck if we should continue with our allocation
4518 wait_for_alloc = true;
4519 spin_unlock(&space_info->lock);
4520 mutex_lock(&fs_info->chunk_mutex);
4521 mutex_unlock(&fs_info->chunk_mutex);
4523 /* Proceed with allocation */
4524 space_info->chunk_alloc = 1;
4525 wait_for_alloc = false;
4526 spin_unlock(&space_info->lock);
4530 } while (wait_for_alloc);
4532 mutex_lock(&fs_info->chunk_mutex);
4533 trans->allocating_chunk = true;
4536 * If we have mixed data/metadata chunks we want to make sure we keep
4537 * allocating mixed chunks instead of individual chunks.
4539 if (btrfs_mixed_space_info(space_info))
4540 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4543 * if we're doing a data chunk, go ahead and make sure that
4544 * we keep a reasonable number of metadata chunks allocated in the
4547 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4548 fs_info->data_chunk_allocations++;
4549 if (!(fs_info->data_chunk_allocations %
4550 fs_info->metadata_ratio))
4551 force_metadata_allocation(fs_info);
4555 * Check if we have enough space in SYSTEM chunk because we may need
4556 * to update devices.
4558 check_system_chunk(trans, flags);
4560 ret = btrfs_alloc_chunk(trans, flags);
4561 trans->allocating_chunk = false;
4563 spin_lock(&space_info->lock);
4566 space_info->full = 1;
4573 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4575 space_info->chunk_alloc = 0;
4576 spin_unlock(&space_info->lock);
4577 mutex_unlock(&fs_info->chunk_mutex);
4579 * When we allocate a new chunk we reserve space in the chunk block
4580 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4581 * add new nodes/leafs to it if we end up needing to do it when
4582 * inserting the chunk item and updating device items as part of the
4583 * second phase of chunk allocation, performed by
4584 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4585 * large number of new block groups to create in our transaction
4586 * handle's new_bgs list to avoid exhausting the chunk block reserve
4587 * in extreme cases - like having a single transaction create many new
4588 * block groups when starting to write out the free space caches of all
4589 * the block groups that were made dirty during the lifetime of the
4592 if (trans->can_flush_pending_bgs &&
4593 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4594 btrfs_create_pending_block_groups(trans);
4595 btrfs_trans_release_chunk_metadata(trans);
4600 static int can_overcommit(struct btrfs_fs_info *fs_info,
4601 struct btrfs_space_info *space_info, u64 bytes,
4602 enum btrfs_reserve_flush_enum flush,
4605 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4612 /* Don't overcommit when in mixed mode. */
4613 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4617 profile = btrfs_system_alloc_profile(fs_info);
4619 profile = btrfs_metadata_alloc_profile(fs_info);
4621 used = btrfs_space_info_used(space_info, false);
4624 * We only want to allow over committing if we have lots of actual space
4625 * free, but if we don't have enough space to handle the global reserve
4626 * space then we could end up having a real enospc problem when trying
4627 * to allocate a chunk or some other such important allocation.
4629 spin_lock(&global_rsv->lock);
4630 space_size = calc_global_rsv_need_space(global_rsv);
4631 spin_unlock(&global_rsv->lock);
4632 if (used + space_size >= space_info->total_bytes)
4635 used += space_info->bytes_may_use;
4637 avail = atomic64_read(&fs_info->free_chunk_space);
4640 * If we have dup, raid1 or raid10 then only half of the free
4641 * space is actually useable. For raid56, the space info used
4642 * doesn't include the parity drive, so we don't have to
4645 factor = btrfs_bg_type_to_factor(profile);
4646 avail = div_u64(avail, factor);
4649 * If we aren't flushing all things, let us overcommit up to
4650 * 1/2th of the space. If we can flush, don't let us overcommit
4651 * too much, let it overcommit up to 1/8 of the space.
4653 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4658 if (used + bytes < space_info->total_bytes + avail)
4663 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4664 unsigned long nr_pages, int nr_items)
4666 struct super_block *sb = fs_info->sb;
4668 if (down_read_trylock(&sb->s_umount)) {
4669 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4670 up_read(&sb->s_umount);
4673 * We needn't worry the filesystem going from r/w to r/o though
4674 * we don't acquire ->s_umount mutex, because the filesystem
4675 * should guarantee the delalloc inodes list be empty after
4676 * the filesystem is readonly(all dirty pages are written to
4679 btrfs_start_delalloc_roots(fs_info, nr_items);
4680 if (!current->journal_info)
4681 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4685 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4691 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4692 nr = div64_u64(to_reclaim, bytes);
4698 #define EXTENT_SIZE_PER_ITEM SZ_256K
4701 * shrink metadata reservation for delalloc
4703 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4704 u64 orig, bool wait_ordered)
4706 struct btrfs_space_info *space_info;
4707 struct btrfs_trans_handle *trans;
4712 unsigned long nr_pages;
4715 /* Calc the number of the pages we need flush for space reservation */
4716 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4717 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4719 trans = (struct btrfs_trans_handle *)current->journal_info;
4720 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4722 delalloc_bytes = percpu_counter_sum_positive(
4723 &fs_info->delalloc_bytes);
4724 if (delalloc_bytes == 0) {
4728 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4733 while (delalloc_bytes && loops < 3) {
4734 max_reclaim = min(delalloc_bytes, to_reclaim);
4735 nr_pages = max_reclaim >> PAGE_SHIFT;
4736 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4738 * We need to wait for the async pages to actually start before
4741 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4745 if (max_reclaim <= nr_pages)
4748 max_reclaim -= nr_pages;
4750 wait_event(fs_info->async_submit_wait,
4751 atomic_read(&fs_info->async_delalloc_pages) <=
4754 spin_lock(&space_info->lock);
4755 if (list_empty(&space_info->tickets) &&
4756 list_empty(&space_info->priority_tickets)) {
4757 spin_unlock(&space_info->lock);
4760 spin_unlock(&space_info->lock);
4763 if (wait_ordered && !trans) {
4764 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4766 time_left = schedule_timeout_killable(1);
4770 delalloc_bytes = percpu_counter_sum_positive(
4771 &fs_info->delalloc_bytes);
4775 struct reserve_ticket {
4778 struct list_head list;
4779 wait_queue_head_t wait;
4783 * maybe_commit_transaction - possibly commit the transaction if its ok to
4784 * @root - the root we're allocating for
4785 * @bytes - the number of bytes we want to reserve
4786 * @force - force the commit
4788 * This will check to make sure that committing the transaction will actually
4789 * get us somewhere and then commit the transaction if it does. Otherwise it
4790 * will return -ENOSPC.
4792 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4793 struct btrfs_space_info *space_info)
4795 struct reserve_ticket *ticket = NULL;
4796 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4797 struct btrfs_trans_handle *trans;
4800 trans = (struct btrfs_trans_handle *)current->journal_info;
4804 spin_lock(&space_info->lock);
4805 if (!list_empty(&space_info->priority_tickets))
4806 ticket = list_first_entry(&space_info->priority_tickets,
4807 struct reserve_ticket, list);
4808 else if (!list_empty(&space_info->tickets))
4809 ticket = list_first_entry(&space_info->tickets,
4810 struct reserve_ticket, list);
4811 bytes = (ticket) ? ticket->bytes : 0;
4812 spin_unlock(&space_info->lock);
4817 /* See if there is enough pinned space to make this reservation */
4818 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4820 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4824 * See if there is some space in the delayed insertion reservation for
4827 if (space_info != delayed_rsv->space_info)
4830 spin_lock(&delayed_rsv->lock);
4831 if (delayed_rsv->size > bytes)
4834 bytes -= delayed_rsv->size;
4835 spin_unlock(&delayed_rsv->lock);
4837 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4839 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4844 trans = btrfs_join_transaction(fs_info->extent_root);
4848 return btrfs_commit_transaction(trans);
4852 * Try to flush some data based on policy set by @state. This is only advisory
4853 * and may fail for various reasons. The caller is supposed to examine the
4854 * state of @space_info to detect the outcome.
4856 static void flush_space(struct btrfs_fs_info *fs_info,
4857 struct btrfs_space_info *space_info, u64 num_bytes,
4860 struct btrfs_root *root = fs_info->extent_root;
4861 struct btrfs_trans_handle *trans;
4866 case FLUSH_DELAYED_ITEMS_NR:
4867 case FLUSH_DELAYED_ITEMS:
4868 if (state == FLUSH_DELAYED_ITEMS_NR)
4869 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4873 trans = btrfs_join_transaction(root);
4874 if (IS_ERR(trans)) {
4875 ret = PTR_ERR(trans);
4878 ret = btrfs_run_delayed_items_nr(trans, nr);
4879 btrfs_end_transaction(trans);
4881 case FLUSH_DELALLOC:
4882 case FLUSH_DELALLOC_WAIT:
4883 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4884 state == FLUSH_DELALLOC_WAIT);
4887 trans = btrfs_join_transaction(root);
4888 if (IS_ERR(trans)) {
4889 ret = PTR_ERR(trans);
4892 ret = do_chunk_alloc(trans,
4893 btrfs_metadata_alloc_profile(fs_info),
4894 CHUNK_ALLOC_NO_FORCE);
4895 btrfs_end_transaction(trans);
4896 if (ret > 0 || ret == -ENOSPC)
4900 ret = may_commit_transaction(fs_info, space_info);
4907 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4913 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4914 struct btrfs_space_info *space_info,
4917 struct reserve_ticket *ticket;
4922 list_for_each_entry(ticket, &space_info->tickets, list)
4923 to_reclaim += ticket->bytes;
4924 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4925 to_reclaim += ticket->bytes;
4929 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4930 if (can_overcommit(fs_info, space_info, to_reclaim,
4931 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4934 used = btrfs_space_info_used(space_info, true);
4936 if (can_overcommit(fs_info, space_info, SZ_1M,
4937 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4938 expected = div_factor_fine(space_info->total_bytes, 95);
4940 expected = div_factor_fine(space_info->total_bytes, 90);
4942 if (used > expected)
4943 to_reclaim = used - expected;
4946 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4947 space_info->bytes_reserved);
4951 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
4952 struct btrfs_space_info *space_info,
4953 u64 used, bool system_chunk)
4955 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4957 /* If we're just plain full then async reclaim just slows us down. */
4958 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4961 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
4965 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4966 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4969 static void wake_all_tickets(struct list_head *head)
4971 struct reserve_ticket *ticket;
4973 while (!list_empty(head)) {
4974 ticket = list_first_entry(head, struct reserve_ticket, list);
4975 list_del_init(&ticket->list);
4976 ticket->error = -ENOSPC;
4977 wake_up(&ticket->wait);
4982 * This is for normal flushers, we can wait all goddamned day if we want to. We
4983 * will loop and continuously try to flush as long as we are making progress.
4984 * We count progress as clearing off tickets each time we have to loop.
4986 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4988 struct btrfs_fs_info *fs_info;
4989 struct btrfs_space_info *space_info;
4992 int commit_cycles = 0;
4993 u64 last_tickets_id;
4995 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4996 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4998 spin_lock(&space_info->lock);
4999 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5002 space_info->flush = 0;
5003 spin_unlock(&space_info->lock);
5006 last_tickets_id = space_info->tickets_id;
5007 spin_unlock(&space_info->lock);
5009 flush_state = FLUSH_DELAYED_ITEMS_NR;
5011 flush_space(fs_info, space_info, to_reclaim, flush_state);
5012 spin_lock(&space_info->lock);
5013 if (list_empty(&space_info->tickets)) {
5014 space_info->flush = 0;
5015 spin_unlock(&space_info->lock);
5018 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5021 if (last_tickets_id == space_info->tickets_id) {
5024 last_tickets_id = space_info->tickets_id;
5025 flush_state = FLUSH_DELAYED_ITEMS_NR;
5030 if (flush_state > COMMIT_TRANS) {
5032 if (commit_cycles > 2) {
5033 wake_all_tickets(&space_info->tickets);
5034 space_info->flush = 0;
5036 flush_state = FLUSH_DELAYED_ITEMS_NR;
5039 spin_unlock(&space_info->lock);
5040 } while (flush_state <= COMMIT_TRANS);
5043 void btrfs_init_async_reclaim_work(struct work_struct *work)
5045 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5048 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5049 struct btrfs_space_info *space_info,
5050 struct reserve_ticket *ticket)
5053 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5055 spin_lock(&space_info->lock);
5056 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5059 spin_unlock(&space_info->lock);
5062 spin_unlock(&space_info->lock);
5065 flush_space(fs_info, space_info, to_reclaim, flush_state);
5067 spin_lock(&space_info->lock);
5068 if (ticket->bytes == 0) {
5069 spin_unlock(&space_info->lock);
5072 spin_unlock(&space_info->lock);
5075 * Priority flushers can't wait on delalloc without
5078 if (flush_state == FLUSH_DELALLOC ||
5079 flush_state == FLUSH_DELALLOC_WAIT)
5080 flush_state = ALLOC_CHUNK;
5081 } while (flush_state < COMMIT_TRANS);
5084 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5085 struct btrfs_space_info *space_info,
5086 struct reserve_ticket *ticket, u64 orig_bytes)
5092 spin_lock(&space_info->lock);
5093 while (ticket->bytes > 0 && ticket->error == 0) {
5094 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5099 spin_unlock(&space_info->lock);
5103 finish_wait(&ticket->wait, &wait);
5104 spin_lock(&space_info->lock);
5107 ret = ticket->error;
5108 if (!list_empty(&ticket->list))
5109 list_del_init(&ticket->list);
5110 if (ticket->bytes && ticket->bytes < orig_bytes) {
5111 u64 num_bytes = orig_bytes - ticket->bytes;
5112 space_info->bytes_may_use -= num_bytes;
5113 trace_btrfs_space_reservation(fs_info, "space_info",
5114 space_info->flags, num_bytes, 0);
5116 spin_unlock(&space_info->lock);
5122 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5123 * @root - the root we're allocating for
5124 * @space_info - the space info we want to allocate from
5125 * @orig_bytes - the number of bytes we want
5126 * @flush - whether or not we can flush to make our reservation
5128 * This will reserve orig_bytes number of bytes from the space info associated
5129 * with the block_rsv. If there is not enough space it will make an attempt to
5130 * flush out space to make room. It will do this by flushing delalloc if
5131 * possible or committing the transaction. If flush is 0 then no attempts to
5132 * regain reservations will be made and this will fail if there is not enough
5135 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5136 struct btrfs_space_info *space_info,
5138 enum btrfs_reserve_flush_enum flush,
5141 struct reserve_ticket ticket;
5146 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5148 spin_lock(&space_info->lock);
5150 used = btrfs_space_info_used(space_info, true);
5153 * If we have enough space then hooray, make our reservation and carry
5154 * on. If not see if we can overcommit, and if we can, hooray carry on.
5155 * If not things get more complicated.
5157 if (used + orig_bytes <= space_info->total_bytes) {
5158 space_info->bytes_may_use += orig_bytes;
5159 trace_btrfs_space_reservation(fs_info, "space_info",
5160 space_info->flags, orig_bytes, 1);
5162 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5164 space_info->bytes_may_use += orig_bytes;
5165 trace_btrfs_space_reservation(fs_info, "space_info",
5166 space_info->flags, orig_bytes, 1);
5171 * If we couldn't make a reservation then setup our reservation ticket
5172 * and kick the async worker if it's not already running.
5174 * If we are a priority flusher then we just need to add our ticket to
5175 * the list and we will do our own flushing further down.
5177 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5178 ticket.bytes = orig_bytes;
5180 init_waitqueue_head(&ticket.wait);
5181 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5182 list_add_tail(&ticket.list, &space_info->tickets);
5183 if (!space_info->flush) {
5184 space_info->flush = 1;
5185 trace_btrfs_trigger_flush(fs_info,
5189 queue_work(system_unbound_wq,
5190 &fs_info->async_reclaim_work);
5193 list_add_tail(&ticket.list,
5194 &space_info->priority_tickets);
5196 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5199 * We will do the space reservation dance during log replay,
5200 * which means we won't have fs_info->fs_root set, so don't do
5201 * the async reclaim as we will panic.
5203 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5204 need_do_async_reclaim(fs_info, space_info,
5205 used, system_chunk) &&
5206 !work_busy(&fs_info->async_reclaim_work)) {
5207 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5208 orig_bytes, flush, "preempt");
5209 queue_work(system_unbound_wq,
5210 &fs_info->async_reclaim_work);
5213 spin_unlock(&space_info->lock);
5214 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5217 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5218 return wait_reserve_ticket(fs_info, space_info, &ticket,
5222 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5223 spin_lock(&space_info->lock);
5225 if (ticket.bytes < orig_bytes) {
5226 u64 num_bytes = orig_bytes - ticket.bytes;
5227 space_info->bytes_may_use -= num_bytes;
5228 trace_btrfs_space_reservation(fs_info, "space_info",
5233 list_del_init(&ticket.list);
5236 spin_unlock(&space_info->lock);
5237 ASSERT(list_empty(&ticket.list));
5242 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5243 * @root - the root we're allocating for
5244 * @block_rsv - the block_rsv we're allocating for
5245 * @orig_bytes - the number of bytes we want
5246 * @flush - whether or not we can flush to make our reservation
5248 * This will reserve orgi_bytes number of bytes from the space info associated
5249 * with the block_rsv. If there is not enough space it will make an attempt to
5250 * flush out space to make room. It will do this by flushing delalloc if
5251 * possible or committing the transaction. If flush is 0 then no attempts to
5252 * regain reservations will be made and this will fail if there is not enough
5255 static int reserve_metadata_bytes(struct btrfs_root *root,
5256 struct btrfs_block_rsv *block_rsv,
5258 enum btrfs_reserve_flush_enum flush)
5260 struct btrfs_fs_info *fs_info = root->fs_info;
5261 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5263 bool system_chunk = (root == fs_info->chunk_root);
5265 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5266 orig_bytes, flush, system_chunk);
5267 if (ret == -ENOSPC &&
5268 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5269 if (block_rsv != global_rsv &&
5270 !block_rsv_use_bytes(global_rsv, orig_bytes))
5273 if (ret == -ENOSPC) {
5274 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5275 block_rsv->space_info->flags,
5278 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5279 dump_space_info(fs_info, block_rsv->space_info,
5285 static struct btrfs_block_rsv *get_block_rsv(
5286 const struct btrfs_trans_handle *trans,
5287 const struct btrfs_root *root)
5289 struct btrfs_fs_info *fs_info = root->fs_info;
5290 struct btrfs_block_rsv *block_rsv = NULL;
5292 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5293 (root == fs_info->csum_root && trans->adding_csums) ||
5294 (root == fs_info->uuid_root))
5295 block_rsv = trans->block_rsv;
5298 block_rsv = root->block_rsv;
5301 block_rsv = &fs_info->empty_block_rsv;
5306 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5310 spin_lock(&block_rsv->lock);
5311 if (block_rsv->reserved >= num_bytes) {
5312 block_rsv->reserved -= num_bytes;
5313 if (block_rsv->reserved < block_rsv->size)
5314 block_rsv->full = 0;
5317 spin_unlock(&block_rsv->lock);
5321 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5322 u64 num_bytes, bool update_size)
5324 spin_lock(&block_rsv->lock);
5325 block_rsv->reserved += num_bytes;
5327 block_rsv->size += num_bytes;
5328 else if (block_rsv->reserved >= block_rsv->size)
5329 block_rsv->full = 1;
5330 spin_unlock(&block_rsv->lock);
5333 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5334 struct btrfs_block_rsv *dest, u64 num_bytes,
5337 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5340 if (global_rsv->space_info != dest->space_info)
5343 spin_lock(&global_rsv->lock);
5344 min_bytes = div_factor(global_rsv->size, min_factor);
5345 if (global_rsv->reserved < min_bytes + num_bytes) {
5346 spin_unlock(&global_rsv->lock);
5349 global_rsv->reserved -= num_bytes;
5350 if (global_rsv->reserved < global_rsv->size)
5351 global_rsv->full = 0;
5352 spin_unlock(&global_rsv->lock);
5354 block_rsv_add_bytes(dest, num_bytes, true);
5359 * This is for space we already have accounted in space_info->bytes_may_use, so
5360 * basically when we're returning space from block_rsv's.
5362 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5363 struct btrfs_space_info *space_info,
5366 struct reserve_ticket *ticket;
5367 struct list_head *head;
5369 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5370 bool check_overcommit = false;
5372 spin_lock(&space_info->lock);
5373 head = &space_info->priority_tickets;
5376 * If we are over our limit then we need to check and see if we can
5377 * overcommit, and if we can't then we just need to free up our space
5378 * and not satisfy any requests.
5380 used = btrfs_space_info_used(space_info, true);
5381 if (used - num_bytes >= space_info->total_bytes)
5382 check_overcommit = true;
5384 while (!list_empty(head) && num_bytes) {
5385 ticket = list_first_entry(head, struct reserve_ticket,
5388 * We use 0 bytes because this space is already reserved, so
5389 * adding the ticket space would be a double count.
5391 if (check_overcommit &&
5392 !can_overcommit(fs_info, space_info, 0, flush, false))
5394 if (num_bytes >= ticket->bytes) {
5395 list_del_init(&ticket->list);
5396 num_bytes -= ticket->bytes;
5398 space_info->tickets_id++;
5399 wake_up(&ticket->wait);
5401 ticket->bytes -= num_bytes;
5406 if (num_bytes && head == &space_info->priority_tickets) {
5407 head = &space_info->tickets;
5408 flush = BTRFS_RESERVE_FLUSH_ALL;
5411 space_info->bytes_may_use -= num_bytes;
5412 trace_btrfs_space_reservation(fs_info, "space_info",
5413 space_info->flags, num_bytes, 0);
5414 spin_unlock(&space_info->lock);
5418 * This is for newly allocated space that isn't accounted in
5419 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5420 * we use this helper.
5422 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5423 struct btrfs_space_info *space_info,
5426 struct reserve_ticket *ticket;
5427 struct list_head *head = &space_info->priority_tickets;
5430 while (!list_empty(head) && num_bytes) {
5431 ticket = list_first_entry(head, struct reserve_ticket,
5433 if (num_bytes >= ticket->bytes) {
5434 trace_btrfs_space_reservation(fs_info, "space_info",
5437 list_del_init(&ticket->list);
5438 num_bytes -= ticket->bytes;
5439 space_info->bytes_may_use += ticket->bytes;
5441 space_info->tickets_id++;
5442 wake_up(&ticket->wait);
5444 trace_btrfs_space_reservation(fs_info, "space_info",
5447 space_info->bytes_may_use += num_bytes;
5448 ticket->bytes -= num_bytes;
5453 if (num_bytes && head == &space_info->priority_tickets) {
5454 head = &space_info->tickets;
5459 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5460 struct btrfs_block_rsv *block_rsv,
5461 struct btrfs_block_rsv *dest, u64 num_bytes,
5462 u64 *qgroup_to_release_ret)
5464 struct btrfs_space_info *space_info = block_rsv->space_info;
5465 u64 qgroup_to_release = 0;
5468 spin_lock(&block_rsv->lock);
5469 if (num_bytes == (u64)-1) {
5470 num_bytes = block_rsv->size;
5471 qgroup_to_release = block_rsv->qgroup_rsv_size;
5473 block_rsv->size -= num_bytes;
5474 if (block_rsv->reserved >= block_rsv->size) {
5475 num_bytes = block_rsv->reserved - block_rsv->size;
5476 block_rsv->reserved = block_rsv->size;
5477 block_rsv->full = 1;
5481 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5482 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5483 block_rsv->qgroup_rsv_size;
5484 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5486 qgroup_to_release = 0;
5488 spin_unlock(&block_rsv->lock);
5491 if (num_bytes > 0) {
5493 spin_lock(&dest->lock);
5497 bytes_to_add = dest->size - dest->reserved;
5498 bytes_to_add = min(num_bytes, bytes_to_add);
5499 dest->reserved += bytes_to_add;
5500 if (dest->reserved >= dest->size)
5502 num_bytes -= bytes_to_add;
5504 spin_unlock(&dest->lock);
5507 space_info_add_old_bytes(fs_info, space_info,
5510 if (qgroup_to_release_ret)
5511 *qgroup_to_release_ret = qgroup_to_release;
5515 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5516 struct btrfs_block_rsv *dst, u64 num_bytes,
5521 ret = block_rsv_use_bytes(src, num_bytes);
5525 block_rsv_add_bytes(dst, num_bytes, update_size);
5529 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5531 memset(rsv, 0, sizeof(*rsv));
5532 spin_lock_init(&rsv->lock);
5536 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5537 struct btrfs_block_rsv *rsv,
5538 unsigned short type)
5540 btrfs_init_block_rsv(rsv, type);
5541 rsv->space_info = __find_space_info(fs_info,
5542 BTRFS_BLOCK_GROUP_METADATA);
5545 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5546 unsigned short type)
5548 struct btrfs_block_rsv *block_rsv;
5550 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5554 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5558 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5559 struct btrfs_block_rsv *rsv)
5563 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5567 int btrfs_block_rsv_add(struct btrfs_root *root,
5568 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5569 enum btrfs_reserve_flush_enum flush)
5576 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5578 block_rsv_add_bytes(block_rsv, num_bytes, true);
5583 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5591 spin_lock(&block_rsv->lock);
5592 num_bytes = div_factor(block_rsv->size, min_factor);
5593 if (block_rsv->reserved >= num_bytes)
5595 spin_unlock(&block_rsv->lock);
5600 int btrfs_block_rsv_refill(struct btrfs_root *root,
5601 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5602 enum btrfs_reserve_flush_enum flush)
5610 spin_lock(&block_rsv->lock);
5611 num_bytes = min_reserved;
5612 if (block_rsv->reserved >= num_bytes)
5615 num_bytes -= block_rsv->reserved;
5616 spin_unlock(&block_rsv->lock);
5621 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5623 block_rsv_add_bytes(block_rsv, num_bytes, false);
5631 * btrfs_inode_rsv_refill - refill the inode block rsv.
5632 * @inode - the inode we are refilling.
5633 * @flush - the flusing restriction.
5635 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5636 * block_rsv->size as the minimum size. We'll either refill the missing amount
5637 * or return if we already have enough space. This will also handle the resreve
5638 * tracepoint for the reserved amount.
5640 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5641 enum btrfs_reserve_flush_enum flush)
5643 struct btrfs_root *root = inode->root;
5644 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5646 u64 qgroup_num_bytes = 0;
5649 spin_lock(&block_rsv->lock);
5650 if (block_rsv->reserved < block_rsv->size)
5651 num_bytes = block_rsv->size - block_rsv->reserved;
5652 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5653 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5654 block_rsv->qgroup_rsv_reserved;
5655 spin_unlock(&block_rsv->lock);
5660 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5663 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5665 block_rsv_add_bytes(block_rsv, num_bytes, false);
5666 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5667 btrfs_ino(inode), num_bytes, 1);
5669 /* Don't forget to increase qgroup_rsv_reserved */
5670 spin_lock(&block_rsv->lock);
5671 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5672 spin_unlock(&block_rsv->lock);
5674 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5679 * btrfs_inode_rsv_release - release any excessive reservation.
5680 * @inode - the inode we need to release from.
5681 * @qgroup_free - free or convert qgroup meta.
5682 * Unlike normal operation, qgroup meta reservation needs to know if we are
5683 * freeing qgroup reservation or just converting it into per-trans. Normally
5684 * @qgroup_free is true for error handling, and false for normal release.
5686 * This is the same as btrfs_block_rsv_release, except that it handles the
5687 * tracepoint for the reservation.
5689 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5691 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5692 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5693 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5695 u64 qgroup_to_release = 0;
5698 * Since we statically set the block_rsv->size we just want to say we
5699 * are releasing 0 bytes, and then we'll just get the reservation over
5702 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5703 &qgroup_to_release);
5705 trace_btrfs_space_reservation(fs_info, "delalloc",
5706 btrfs_ino(inode), released, 0);
5708 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5710 btrfs_qgroup_convert_reserved_meta(inode->root,
5714 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5715 struct btrfs_block_rsv *block_rsv,
5718 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5720 if (global_rsv == block_rsv ||
5721 block_rsv->space_info != global_rsv->space_info)
5723 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5726 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5728 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5729 struct btrfs_space_info *sinfo = block_rsv->space_info;
5733 * The global block rsv is based on the size of the extent tree, the
5734 * checksum tree and the root tree. If the fs is empty we want to set
5735 * it to a minimal amount for safety.
5737 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5738 btrfs_root_used(&fs_info->csum_root->root_item) +
5739 btrfs_root_used(&fs_info->tree_root->root_item);
5740 num_bytes = max_t(u64, num_bytes, SZ_16M);
5742 spin_lock(&sinfo->lock);
5743 spin_lock(&block_rsv->lock);
5745 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5747 if (block_rsv->reserved < block_rsv->size) {
5748 num_bytes = btrfs_space_info_used(sinfo, true);
5749 if (sinfo->total_bytes > num_bytes) {
5750 num_bytes = sinfo->total_bytes - num_bytes;
5751 num_bytes = min(num_bytes,
5752 block_rsv->size - block_rsv->reserved);
5753 block_rsv->reserved += num_bytes;
5754 sinfo->bytes_may_use += num_bytes;
5755 trace_btrfs_space_reservation(fs_info, "space_info",
5756 sinfo->flags, num_bytes,
5759 } else if (block_rsv->reserved > block_rsv->size) {
5760 num_bytes = block_rsv->reserved - block_rsv->size;
5761 sinfo->bytes_may_use -= num_bytes;
5762 trace_btrfs_space_reservation(fs_info, "space_info",
5763 sinfo->flags, num_bytes, 0);
5764 block_rsv->reserved = block_rsv->size;
5767 if (block_rsv->reserved == block_rsv->size)
5768 block_rsv->full = 1;
5770 block_rsv->full = 0;
5772 spin_unlock(&block_rsv->lock);
5773 spin_unlock(&sinfo->lock);
5776 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5778 struct btrfs_space_info *space_info;
5780 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5781 fs_info->chunk_block_rsv.space_info = space_info;
5783 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5784 fs_info->global_block_rsv.space_info = space_info;
5785 fs_info->trans_block_rsv.space_info = space_info;
5786 fs_info->empty_block_rsv.space_info = space_info;
5787 fs_info->delayed_block_rsv.space_info = space_info;
5789 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5790 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5791 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5792 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5793 if (fs_info->quota_root)
5794 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5795 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5797 update_global_block_rsv(fs_info);
5800 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5802 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5804 WARN_ON(fs_info->trans_block_rsv.size > 0);
5805 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5806 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5807 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5808 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5809 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5814 * To be called after all the new block groups attached to the transaction
5815 * handle have been created (btrfs_create_pending_block_groups()).
5817 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5819 struct btrfs_fs_info *fs_info = trans->fs_info;
5821 if (!trans->chunk_bytes_reserved)
5824 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5826 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5827 trans->chunk_bytes_reserved, NULL);
5828 trans->chunk_bytes_reserved = 0;
5832 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5833 * root: the root of the parent directory
5834 * rsv: block reservation
5835 * items: the number of items that we need do reservation
5836 * use_global_rsv: allow fallback to the global block reservation
5838 * This function is used to reserve the space for snapshot/subvolume
5839 * creation and deletion. Those operations are different with the
5840 * common file/directory operations, they change two fs/file trees
5841 * and root tree, the number of items that the qgroup reserves is
5842 * different with the free space reservation. So we can not use
5843 * the space reservation mechanism in start_transaction().
5845 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5846 struct btrfs_block_rsv *rsv, int items,
5847 bool use_global_rsv)
5849 u64 qgroup_num_bytes = 0;
5852 struct btrfs_fs_info *fs_info = root->fs_info;
5853 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5855 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5856 /* One for parent inode, two for dir entries */
5857 qgroup_num_bytes = 3 * fs_info->nodesize;
5858 ret = btrfs_qgroup_reserve_meta_prealloc(root,
5859 qgroup_num_bytes, true);
5864 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5865 rsv->space_info = __find_space_info(fs_info,
5866 BTRFS_BLOCK_GROUP_METADATA);
5867 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5868 BTRFS_RESERVE_FLUSH_ALL);
5870 if (ret == -ENOSPC && use_global_rsv)
5871 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
5873 if (ret && qgroup_num_bytes)
5874 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5879 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5880 struct btrfs_block_rsv *rsv)
5882 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5885 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
5886 struct btrfs_inode *inode)
5888 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5889 u64 reserve_size = 0;
5890 u64 qgroup_rsv_size = 0;
5892 unsigned outstanding_extents;
5894 lockdep_assert_held(&inode->lock);
5895 outstanding_extents = inode->outstanding_extents;
5896 if (outstanding_extents)
5897 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
5898 outstanding_extents + 1);
5899 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
5901 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
5904 * For qgroup rsv, the calculation is very simple:
5905 * account one nodesize for each outstanding extent
5907 * This is overestimating in most cases.
5909 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
5911 spin_lock(&block_rsv->lock);
5912 block_rsv->size = reserve_size;
5913 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
5914 spin_unlock(&block_rsv->lock);
5917 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5919 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5920 unsigned nr_extents;
5921 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5923 bool delalloc_lock = true;
5925 /* If we are a free space inode we need to not flush since we will be in
5926 * the middle of a transaction commit. We also don't need the delalloc
5927 * mutex since we won't race with anybody. We need this mostly to make
5928 * lockdep shut its filthy mouth.
5930 * If we have a transaction open (can happen if we call truncate_block
5931 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5933 if (btrfs_is_free_space_inode(inode)) {
5934 flush = BTRFS_RESERVE_NO_FLUSH;
5935 delalloc_lock = false;
5937 if (current->journal_info)
5938 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5940 if (btrfs_transaction_in_commit(fs_info))
5941 schedule_timeout(1);
5945 mutex_lock(&inode->delalloc_mutex);
5947 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5949 /* Add our new extents and calculate the new rsv size. */
5950 spin_lock(&inode->lock);
5951 nr_extents = count_max_extents(num_bytes);
5952 btrfs_mod_outstanding_extents(inode, nr_extents);
5953 inode->csum_bytes += num_bytes;
5954 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5955 spin_unlock(&inode->lock);
5957 ret = btrfs_inode_rsv_refill(inode, flush);
5962 mutex_unlock(&inode->delalloc_mutex);
5966 spin_lock(&inode->lock);
5967 nr_extents = count_max_extents(num_bytes);
5968 btrfs_mod_outstanding_extents(inode, -nr_extents);
5969 inode->csum_bytes -= num_bytes;
5970 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5971 spin_unlock(&inode->lock);
5973 btrfs_inode_rsv_release(inode, true);
5975 mutex_unlock(&inode->delalloc_mutex);
5980 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5981 * @inode: the inode to release the reservation for.
5982 * @num_bytes: the number of bytes we are releasing.
5983 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
5985 * This will release the metadata reservation for an inode. This can be called
5986 * once we complete IO for a given set of bytes to release their metadata
5987 * reservations, or on error for the same reason.
5989 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
5992 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5994 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5995 spin_lock(&inode->lock);
5996 inode->csum_bytes -= num_bytes;
5997 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
5998 spin_unlock(&inode->lock);
6000 if (btrfs_is_testing(fs_info))
6003 btrfs_inode_rsv_release(inode, qgroup_free);
6007 * btrfs_delalloc_release_extents - release our outstanding_extents
6008 * @inode: the inode to balance the reservation for.
6009 * @num_bytes: the number of bytes we originally reserved with
6010 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6012 * When we reserve space we increase outstanding_extents for the extents we may
6013 * add. Once we've set the range as delalloc or created our ordered extents we
6014 * have outstanding_extents to track the real usage, so we use this to free our
6015 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6016 * with btrfs_delalloc_reserve_metadata.
6018 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6021 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6022 unsigned num_extents;
6024 spin_lock(&inode->lock);
6025 num_extents = count_max_extents(num_bytes);
6026 btrfs_mod_outstanding_extents(inode, -num_extents);
6027 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6028 spin_unlock(&inode->lock);
6030 if (btrfs_is_testing(fs_info))
6033 btrfs_inode_rsv_release(inode, qgroup_free);
6037 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6039 * @inode: inode we're writing to
6040 * @start: start range we are writing to
6041 * @len: how long the range we are writing to
6042 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6043 * current reservation.
6045 * This will do the following things
6047 * o reserve space in data space info for num bytes
6048 * and reserve precious corresponding qgroup space
6049 * (Done in check_data_free_space)
6051 * o reserve space for metadata space, based on the number of outstanding
6052 * extents and how much csums will be needed
6053 * also reserve metadata space in a per root over-reserve method.
6054 * o add to the inodes->delalloc_bytes
6055 * o add it to the fs_info's delalloc inodes list.
6056 * (Above 3 all done in delalloc_reserve_metadata)
6058 * Return 0 for success
6059 * Return <0 for error(-ENOSPC or -EQUOT)
6061 int btrfs_delalloc_reserve_space(struct inode *inode,
6062 struct extent_changeset **reserved, u64 start, u64 len)
6066 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6069 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6071 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6076 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6077 * @inode: inode we're releasing space for
6078 * @start: start position of the space already reserved
6079 * @len: the len of the space already reserved
6080 * @release_bytes: the len of the space we consumed or didn't use
6082 * This function will release the metadata space that was not used and will
6083 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6084 * list if there are no delalloc bytes left.
6085 * Also it will handle the qgroup reserved space.
6087 void btrfs_delalloc_release_space(struct inode *inode,
6088 struct extent_changeset *reserved,
6089 u64 start, u64 len, bool qgroup_free)
6091 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6092 btrfs_free_reserved_data_space(inode, reserved, start, len);
6095 static int update_block_group(struct btrfs_trans_handle *trans,
6096 struct btrfs_fs_info *info, u64 bytenr,
6097 u64 num_bytes, int alloc)
6099 struct btrfs_block_group_cache *cache = NULL;
6100 u64 total = num_bytes;
6105 /* block accounting for super block */
6106 spin_lock(&info->delalloc_root_lock);
6107 old_val = btrfs_super_bytes_used(info->super_copy);
6109 old_val += num_bytes;
6111 old_val -= num_bytes;
6112 btrfs_set_super_bytes_used(info->super_copy, old_val);
6113 spin_unlock(&info->delalloc_root_lock);
6116 cache = btrfs_lookup_block_group(info, bytenr);
6119 factor = btrfs_bg_type_to_factor(cache->flags);
6122 * If this block group has free space cache written out, we
6123 * need to make sure to load it if we are removing space. This
6124 * is because we need the unpinning stage to actually add the
6125 * space back to the block group, otherwise we will leak space.
6127 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6128 cache_block_group(cache, 1);
6130 byte_in_group = bytenr - cache->key.objectid;
6131 WARN_ON(byte_in_group > cache->key.offset);
6133 spin_lock(&cache->space_info->lock);
6134 spin_lock(&cache->lock);
6136 if (btrfs_test_opt(info, SPACE_CACHE) &&
6137 cache->disk_cache_state < BTRFS_DC_CLEAR)
6138 cache->disk_cache_state = BTRFS_DC_CLEAR;
6140 old_val = btrfs_block_group_used(&cache->item);
6141 num_bytes = min(total, cache->key.offset - byte_in_group);
6143 old_val += num_bytes;
6144 btrfs_set_block_group_used(&cache->item, old_val);
6145 cache->reserved -= num_bytes;
6146 cache->space_info->bytes_reserved -= num_bytes;
6147 cache->space_info->bytes_used += num_bytes;
6148 cache->space_info->disk_used += num_bytes * factor;
6149 spin_unlock(&cache->lock);
6150 spin_unlock(&cache->space_info->lock);
6152 old_val -= num_bytes;
6153 btrfs_set_block_group_used(&cache->item, old_val);
6154 cache->pinned += num_bytes;
6155 cache->space_info->bytes_pinned += num_bytes;
6156 cache->space_info->bytes_used -= num_bytes;
6157 cache->space_info->disk_used -= num_bytes * factor;
6158 spin_unlock(&cache->lock);
6159 spin_unlock(&cache->space_info->lock);
6161 trace_btrfs_space_reservation(info, "pinned",
6162 cache->space_info->flags,
6164 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6166 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6167 set_extent_dirty(info->pinned_extents,
6168 bytenr, bytenr + num_bytes - 1,
6169 GFP_NOFS | __GFP_NOFAIL);
6172 spin_lock(&trans->transaction->dirty_bgs_lock);
6173 if (list_empty(&cache->dirty_list)) {
6174 list_add_tail(&cache->dirty_list,
6175 &trans->transaction->dirty_bgs);
6176 trans->transaction->num_dirty_bgs++;
6177 btrfs_get_block_group(cache);
6179 spin_unlock(&trans->transaction->dirty_bgs_lock);
6182 * No longer have used bytes in this block group, queue it for
6183 * deletion. We do this after adding the block group to the
6184 * dirty list to avoid races between cleaner kthread and space
6187 if (!alloc && old_val == 0)
6188 btrfs_mark_bg_unused(cache);
6190 btrfs_put_block_group(cache);
6192 bytenr += num_bytes;
6197 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6199 struct btrfs_block_group_cache *cache;
6202 spin_lock(&fs_info->block_group_cache_lock);
6203 bytenr = fs_info->first_logical_byte;
6204 spin_unlock(&fs_info->block_group_cache_lock);
6206 if (bytenr < (u64)-1)
6209 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6213 bytenr = cache->key.objectid;
6214 btrfs_put_block_group(cache);
6219 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6220 struct btrfs_block_group_cache *cache,
6221 u64 bytenr, u64 num_bytes, int reserved)
6223 spin_lock(&cache->space_info->lock);
6224 spin_lock(&cache->lock);
6225 cache->pinned += num_bytes;
6226 cache->space_info->bytes_pinned += num_bytes;
6228 cache->reserved -= num_bytes;
6229 cache->space_info->bytes_reserved -= num_bytes;
6231 spin_unlock(&cache->lock);
6232 spin_unlock(&cache->space_info->lock);
6234 trace_btrfs_space_reservation(fs_info, "pinned",
6235 cache->space_info->flags, num_bytes, 1);
6236 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6237 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6238 set_extent_dirty(fs_info->pinned_extents, bytenr,
6239 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6244 * this function must be called within transaction
6246 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6247 u64 bytenr, u64 num_bytes, int reserved)
6249 struct btrfs_block_group_cache *cache;
6251 cache = btrfs_lookup_block_group(fs_info, bytenr);
6252 BUG_ON(!cache); /* Logic error */
6254 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6256 btrfs_put_block_group(cache);
6261 * this function must be called within transaction
6263 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6264 u64 bytenr, u64 num_bytes)
6266 struct btrfs_block_group_cache *cache;
6269 cache = btrfs_lookup_block_group(fs_info, bytenr);
6274 * pull in the free space cache (if any) so that our pin
6275 * removes the free space from the cache. We have load_only set
6276 * to one because the slow code to read in the free extents does check
6277 * the pinned extents.
6279 cache_block_group(cache, 1);
6281 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6283 /* remove us from the free space cache (if we're there at all) */
6284 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6285 btrfs_put_block_group(cache);
6289 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6290 u64 start, u64 num_bytes)
6293 struct btrfs_block_group_cache *block_group;
6294 struct btrfs_caching_control *caching_ctl;
6296 block_group = btrfs_lookup_block_group(fs_info, start);
6300 cache_block_group(block_group, 0);
6301 caching_ctl = get_caching_control(block_group);
6305 BUG_ON(!block_group_cache_done(block_group));
6306 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6308 mutex_lock(&caching_ctl->mutex);
6310 if (start >= caching_ctl->progress) {
6311 ret = add_excluded_extent(fs_info, start, num_bytes);
6312 } else if (start + num_bytes <= caching_ctl->progress) {
6313 ret = btrfs_remove_free_space(block_group,
6316 num_bytes = caching_ctl->progress - start;
6317 ret = btrfs_remove_free_space(block_group,
6322 num_bytes = (start + num_bytes) -
6323 caching_ctl->progress;
6324 start = caching_ctl->progress;
6325 ret = add_excluded_extent(fs_info, start, num_bytes);
6328 mutex_unlock(&caching_ctl->mutex);
6329 put_caching_control(caching_ctl);
6331 btrfs_put_block_group(block_group);
6335 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6336 struct extent_buffer *eb)
6338 struct btrfs_file_extent_item *item;
6339 struct btrfs_key key;
6344 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6347 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6348 btrfs_item_key_to_cpu(eb, &key, i);
6349 if (key.type != BTRFS_EXTENT_DATA_KEY)
6351 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6352 found_type = btrfs_file_extent_type(eb, item);
6353 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6355 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6357 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6358 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6359 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6368 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6370 atomic_inc(&bg->reservations);
6373 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6376 struct btrfs_block_group_cache *bg;
6378 bg = btrfs_lookup_block_group(fs_info, start);
6380 if (atomic_dec_and_test(&bg->reservations))
6381 wake_up_var(&bg->reservations);
6382 btrfs_put_block_group(bg);
6385 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6387 struct btrfs_space_info *space_info = bg->space_info;
6391 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6395 * Our block group is read only but before we set it to read only,
6396 * some task might have had allocated an extent from it already, but it
6397 * has not yet created a respective ordered extent (and added it to a
6398 * root's list of ordered extents).
6399 * Therefore wait for any task currently allocating extents, since the
6400 * block group's reservations counter is incremented while a read lock
6401 * on the groups' semaphore is held and decremented after releasing
6402 * the read access on that semaphore and creating the ordered extent.
6404 down_write(&space_info->groups_sem);
6405 up_write(&space_info->groups_sem);
6407 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6411 * btrfs_add_reserved_bytes - update the block_group and space info counters
6412 * @cache: The cache we are manipulating
6413 * @ram_bytes: The number of bytes of file content, and will be same to
6414 * @num_bytes except for the compress path.
6415 * @num_bytes: The number of bytes in question
6416 * @delalloc: The blocks are allocated for the delalloc write
6418 * This is called by the allocator when it reserves space. If this is a
6419 * reservation and the block group has become read only we cannot make the
6420 * reservation and return -EAGAIN, otherwise this function always succeeds.
6422 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6423 u64 ram_bytes, u64 num_bytes, int delalloc)
6425 struct btrfs_space_info *space_info = cache->space_info;
6428 spin_lock(&space_info->lock);
6429 spin_lock(&cache->lock);
6433 cache->reserved += num_bytes;
6434 space_info->bytes_reserved += num_bytes;
6435 space_info->bytes_may_use -= ram_bytes;
6437 cache->delalloc_bytes += num_bytes;
6439 spin_unlock(&cache->lock);
6440 spin_unlock(&space_info->lock);
6445 * btrfs_free_reserved_bytes - update the block_group and space info counters
6446 * @cache: The cache we are manipulating
6447 * @num_bytes: The number of bytes in question
6448 * @delalloc: The blocks are allocated for the delalloc write
6450 * This is called by somebody who is freeing space that was never actually used
6451 * on disk. For example if you reserve some space for a new leaf in transaction
6452 * A and before transaction A commits you free that leaf, you call this with
6453 * reserve set to 0 in order to clear the reservation.
6456 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6457 u64 num_bytes, int delalloc)
6459 struct btrfs_space_info *space_info = cache->space_info;
6461 spin_lock(&space_info->lock);
6462 spin_lock(&cache->lock);
6464 space_info->bytes_readonly += num_bytes;
6465 cache->reserved -= num_bytes;
6466 space_info->bytes_reserved -= num_bytes;
6469 cache->delalloc_bytes -= num_bytes;
6470 spin_unlock(&cache->lock);
6471 spin_unlock(&space_info->lock);
6473 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6475 struct btrfs_caching_control *next;
6476 struct btrfs_caching_control *caching_ctl;
6477 struct btrfs_block_group_cache *cache;
6479 down_write(&fs_info->commit_root_sem);
6481 list_for_each_entry_safe(caching_ctl, next,
6482 &fs_info->caching_block_groups, list) {
6483 cache = caching_ctl->block_group;
6484 if (block_group_cache_done(cache)) {
6485 cache->last_byte_to_unpin = (u64)-1;
6486 list_del_init(&caching_ctl->list);
6487 put_caching_control(caching_ctl);
6489 cache->last_byte_to_unpin = caching_ctl->progress;
6493 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6494 fs_info->pinned_extents = &fs_info->freed_extents[1];
6496 fs_info->pinned_extents = &fs_info->freed_extents[0];
6498 up_write(&fs_info->commit_root_sem);
6500 update_global_block_rsv(fs_info);
6504 * Returns the free cluster for the given space info and sets empty_cluster to
6505 * what it should be based on the mount options.
6507 static struct btrfs_free_cluster *
6508 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6509 struct btrfs_space_info *space_info, u64 *empty_cluster)
6511 struct btrfs_free_cluster *ret = NULL;
6514 if (btrfs_mixed_space_info(space_info))
6517 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6518 ret = &fs_info->meta_alloc_cluster;
6519 if (btrfs_test_opt(fs_info, SSD))
6520 *empty_cluster = SZ_2M;
6522 *empty_cluster = SZ_64K;
6523 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6524 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6525 *empty_cluster = SZ_2M;
6526 ret = &fs_info->data_alloc_cluster;
6532 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6534 const bool return_free_space)
6536 struct btrfs_block_group_cache *cache = NULL;
6537 struct btrfs_space_info *space_info;
6538 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6539 struct btrfs_free_cluster *cluster = NULL;
6541 u64 total_unpinned = 0;
6542 u64 empty_cluster = 0;
6545 while (start <= end) {
6548 start >= cache->key.objectid + cache->key.offset) {
6550 btrfs_put_block_group(cache);
6552 cache = btrfs_lookup_block_group(fs_info, start);
6553 BUG_ON(!cache); /* Logic error */
6555 cluster = fetch_cluster_info(fs_info,
6558 empty_cluster <<= 1;
6561 len = cache->key.objectid + cache->key.offset - start;
6562 len = min(len, end + 1 - start);
6564 if (start < cache->last_byte_to_unpin) {
6565 len = min(len, cache->last_byte_to_unpin - start);
6566 if (return_free_space)
6567 btrfs_add_free_space(cache, start, len);
6571 total_unpinned += len;
6572 space_info = cache->space_info;
6575 * If this space cluster has been marked as fragmented and we've
6576 * unpinned enough in this block group to potentially allow a
6577 * cluster to be created inside of it go ahead and clear the
6580 if (cluster && cluster->fragmented &&
6581 total_unpinned > empty_cluster) {
6582 spin_lock(&cluster->lock);
6583 cluster->fragmented = 0;
6584 spin_unlock(&cluster->lock);
6587 spin_lock(&space_info->lock);
6588 spin_lock(&cache->lock);
6589 cache->pinned -= len;
6590 space_info->bytes_pinned -= len;
6592 trace_btrfs_space_reservation(fs_info, "pinned",
6593 space_info->flags, len, 0);
6594 space_info->max_extent_size = 0;
6595 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6596 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6598 space_info->bytes_readonly += len;
6601 spin_unlock(&cache->lock);
6602 if (!readonly && return_free_space &&
6603 global_rsv->space_info == space_info) {
6606 spin_lock(&global_rsv->lock);
6607 if (!global_rsv->full) {
6608 to_add = min(len, global_rsv->size -
6609 global_rsv->reserved);
6610 global_rsv->reserved += to_add;
6611 space_info->bytes_may_use += to_add;
6612 if (global_rsv->reserved >= global_rsv->size)
6613 global_rsv->full = 1;
6614 trace_btrfs_space_reservation(fs_info,
6620 spin_unlock(&global_rsv->lock);
6621 /* Add to any tickets we may have */
6623 space_info_add_new_bytes(fs_info, space_info,
6626 spin_unlock(&space_info->lock);
6630 btrfs_put_block_group(cache);
6634 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6636 struct btrfs_fs_info *fs_info = trans->fs_info;
6637 struct btrfs_block_group_cache *block_group, *tmp;
6638 struct list_head *deleted_bgs;
6639 struct extent_io_tree *unpin;
6644 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6645 unpin = &fs_info->freed_extents[1];
6647 unpin = &fs_info->freed_extents[0];
6649 while (!trans->aborted) {
6650 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6651 ret = find_first_extent_bit(unpin, 0, &start, &end,
6652 EXTENT_DIRTY, NULL);
6654 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6658 if (btrfs_test_opt(fs_info, DISCARD))
6659 ret = btrfs_discard_extent(fs_info, start,
6660 end + 1 - start, NULL);
6662 clear_extent_dirty(unpin, start, end);
6663 unpin_extent_range(fs_info, start, end, true);
6664 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6669 * Transaction is finished. We don't need the lock anymore. We
6670 * do need to clean up the block groups in case of a transaction
6673 deleted_bgs = &trans->transaction->deleted_bgs;
6674 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6678 if (!trans->aborted)
6679 ret = btrfs_discard_extent(fs_info,
6680 block_group->key.objectid,
6681 block_group->key.offset,
6684 list_del_init(&block_group->bg_list);
6685 btrfs_put_block_group_trimming(block_group);
6686 btrfs_put_block_group(block_group);
6689 const char *errstr = btrfs_decode_error(ret);
6691 "discard failed while removing blockgroup: errno=%d %s",
6699 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6700 struct btrfs_delayed_ref_node *node, u64 parent,
6701 u64 root_objectid, u64 owner_objectid,
6702 u64 owner_offset, int refs_to_drop,
6703 struct btrfs_delayed_extent_op *extent_op)
6705 struct btrfs_fs_info *info = trans->fs_info;
6706 struct btrfs_key key;
6707 struct btrfs_path *path;
6708 struct btrfs_root *extent_root = info->extent_root;
6709 struct extent_buffer *leaf;
6710 struct btrfs_extent_item *ei;
6711 struct btrfs_extent_inline_ref *iref;
6714 int extent_slot = 0;
6715 int found_extent = 0;
6719 u64 bytenr = node->bytenr;
6720 u64 num_bytes = node->num_bytes;
6722 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6724 path = btrfs_alloc_path();
6728 path->reada = READA_FORWARD;
6729 path->leave_spinning = 1;
6731 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6732 BUG_ON(!is_data && refs_to_drop != 1);
6735 skinny_metadata = false;
6737 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6738 parent, root_objectid, owner_objectid,
6741 extent_slot = path->slots[0];
6742 while (extent_slot >= 0) {
6743 btrfs_item_key_to_cpu(path->nodes[0], &key,
6745 if (key.objectid != bytenr)
6747 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6748 key.offset == num_bytes) {
6752 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6753 key.offset == owner_objectid) {
6757 if (path->slots[0] - extent_slot > 5)
6762 if (!found_extent) {
6764 ret = remove_extent_backref(trans, path, NULL,
6766 is_data, &last_ref);
6768 btrfs_abort_transaction(trans, ret);
6771 btrfs_release_path(path);
6772 path->leave_spinning = 1;
6774 key.objectid = bytenr;
6775 key.type = BTRFS_EXTENT_ITEM_KEY;
6776 key.offset = num_bytes;
6778 if (!is_data && skinny_metadata) {
6779 key.type = BTRFS_METADATA_ITEM_KEY;
6780 key.offset = owner_objectid;
6783 ret = btrfs_search_slot(trans, extent_root,
6785 if (ret > 0 && skinny_metadata && path->slots[0]) {
6787 * Couldn't find our skinny metadata item,
6788 * see if we have ye olde extent item.
6791 btrfs_item_key_to_cpu(path->nodes[0], &key,
6793 if (key.objectid == bytenr &&
6794 key.type == BTRFS_EXTENT_ITEM_KEY &&
6795 key.offset == num_bytes)
6799 if (ret > 0 && skinny_metadata) {
6800 skinny_metadata = false;
6801 key.objectid = bytenr;
6802 key.type = BTRFS_EXTENT_ITEM_KEY;
6803 key.offset = num_bytes;
6804 btrfs_release_path(path);
6805 ret = btrfs_search_slot(trans, extent_root,
6811 "umm, got %d back from search, was looking for %llu",
6814 btrfs_print_leaf(path->nodes[0]);
6817 btrfs_abort_transaction(trans, ret);
6820 extent_slot = path->slots[0];
6822 } else if (WARN_ON(ret == -ENOENT)) {
6823 btrfs_print_leaf(path->nodes[0]);
6825 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6826 bytenr, parent, root_objectid, owner_objectid,
6828 btrfs_abort_transaction(trans, ret);
6831 btrfs_abort_transaction(trans, ret);
6835 leaf = path->nodes[0];
6836 item_size = btrfs_item_size_nr(leaf, extent_slot);
6837 if (unlikely(item_size < sizeof(*ei))) {
6839 btrfs_print_v0_err(info);
6840 btrfs_abort_transaction(trans, ret);
6843 ei = btrfs_item_ptr(leaf, extent_slot,
6844 struct btrfs_extent_item);
6845 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6846 key.type == BTRFS_EXTENT_ITEM_KEY) {
6847 struct btrfs_tree_block_info *bi;
6848 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6849 bi = (struct btrfs_tree_block_info *)(ei + 1);
6850 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6853 refs = btrfs_extent_refs(leaf, ei);
6854 if (refs < refs_to_drop) {
6856 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
6857 refs_to_drop, refs, bytenr);
6859 btrfs_abort_transaction(trans, ret);
6862 refs -= refs_to_drop;
6866 __run_delayed_extent_op(extent_op, leaf, ei);
6868 * In the case of inline back ref, reference count will
6869 * be updated by remove_extent_backref
6872 BUG_ON(!found_extent);
6874 btrfs_set_extent_refs(leaf, ei, refs);
6875 btrfs_mark_buffer_dirty(leaf);
6878 ret = remove_extent_backref(trans, path, iref,
6879 refs_to_drop, is_data,
6882 btrfs_abort_transaction(trans, ret);
6888 BUG_ON(is_data && refs_to_drop !=
6889 extent_data_ref_count(path, iref));
6891 BUG_ON(path->slots[0] != extent_slot);
6893 BUG_ON(path->slots[0] != extent_slot + 1);
6894 path->slots[0] = extent_slot;
6900 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6903 btrfs_abort_transaction(trans, ret);
6906 btrfs_release_path(path);
6909 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
6911 btrfs_abort_transaction(trans, ret);
6916 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
6918 btrfs_abort_transaction(trans, ret);
6922 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
6924 btrfs_abort_transaction(trans, ret);
6928 btrfs_release_path(path);
6931 btrfs_free_path(path);
6936 * when we free an block, it is possible (and likely) that we free the last
6937 * delayed ref for that extent as well. This searches the delayed ref tree for
6938 * a given extent, and if there are no other delayed refs to be processed, it
6939 * removes it from the tree.
6941 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6944 struct btrfs_delayed_ref_head *head;
6945 struct btrfs_delayed_ref_root *delayed_refs;
6948 delayed_refs = &trans->transaction->delayed_refs;
6949 spin_lock(&delayed_refs->lock);
6950 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
6952 goto out_delayed_unlock;
6954 spin_lock(&head->lock);
6955 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
6958 if (head->extent_op) {
6959 if (!head->must_insert_reserved)
6961 btrfs_free_delayed_extent_op(head->extent_op);
6962 head->extent_op = NULL;
6966 * waiting for the lock here would deadlock. If someone else has it
6967 * locked they are already in the process of dropping it anyway
6969 if (!mutex_trylock(&head->mutex))
6973 * at this point we have a head with no other entries. Go
6974 * ahead and process it.
6976 rb_erase_cached(&head->href_node, &delayed_refs->href_root);
6977 RB_CLEAR_NODE(&head->href_node);
6978 atomic_dec(&delayed_refs->num_entries);
6981 * we don't take a ref on the node because we're removing it from the
6982 * tree, so we just steal the ref the tree was holding.
6984 delayed_refs->num_heads--;
6985 if (head->processing == 0)
6986 delayed_refs->num_heads_ready--;
6987 head->processing = 0;
6988 spin_unlock(&head->lock);
6989 spin_unlock(&delayed_refs->lock);
6991 BUG_ON(head->extent_op);
6992 if (head->must_insert_reserved)
6995 mutex_unlock(&head->mutex);
6996 btrfs_put_delayed_ref_head(head);
6999 spin_unlock(&head->lock);
7002 spin_unlock(&delayed_refs->lock);
7006 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7007 struct btrfs_root *root,
7008 struct extent_buffer *buf,
7009 u64 parent, int last_ref)
7011 struct btrfs_fs_info *fs_info = root->fs_info;
7015 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7016 int old_ref_mod, new_ref_mod;
7018 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7019 root->root_key.objectid,
7020 btrfs_header_level(buf), 0,
7021 BTRFS_DROP_DELAYED_REF);
7022 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7024 root->root_key.objectid,
7025 btrfs_header_level(buf),
7026 BTRFS_DROP_DELAYED_REF, NULL,
7027 &old_ref_mod, &new_ref_mod);
7028 BUG_ON(ret); /* -ENOMEM */
7029 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7032 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7033 struct btrfs_block_group_cache *cache;
7035 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7036 ret = check_ref_cleanup(trans, buf->start);
7042 cache = btrfs_lookup_block_group(fs_info, buf->start);
7044 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7045 pin_down_extent(fs_info, cache, buf->start,
7047 btrfs_put_block_group(cache);
7051 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7053 btrfs_add_free_space(cache, buf->start, buf->len);
7054 btrfs_free_reserved_bytes(cache, buf->len, 0);
7055 btrfs_put_block_group(cache);
7056 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7060 add_pinned_bytes(fs_info, buf->len, true,
7061 root->root_key.objectid);
7065 * Deleting the buffer, clear the corrupt flag since it doesn't
7068 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7072 /* Can return -ENOMEM */
7073 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7074 struct btrfs_root *root,
7075 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7076 u64 owner, u64 offset)
7078 struct btrfs_fs_info *fs_info = root->fs_info;
7079 int old_ref_mod, new_ref_mod;
7082 if (btrfs_is_testing(fs_info))
7085 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7086 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7087 root_objectid, owner, offset,
7088 BTRFS_DROP_DELAYED_REF);
7091 * tree log blocks never actually go into the extent allocation
7092 * tree, just update pinning info and exit early.
7094 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7095 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7096 /* unlocks the pinned mutex */
7097 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7098 old_ref_mod = new_ref_mod = 0;
7100 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7101 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7103 root_objectid, (int)owner,
7104 BTRFS_DROP_DELAYED_REF, NULL,
7105 &old_ref_mod, &new_ref_mod);
7107 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7109 root_objectid, owner, offset,
7110 0, BTRFS_DROP_DELAYED_REF,
7111 &old_ref_mod, &new_ref_mod);
7114 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7115 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7117 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7124 * when we wait for progress in the block group caching, its because
7125 * our allocation attempt failed at least once. So, we must sleep
7126 * and let some progress happen before we try again.
7128 * This function will sleep at least once waiting for new free space to
7129 * show up, and then it will check the block group free space numbers
7130 * for our min num_bytes. Another option is to have it go ahead
7131 * and look in the rbtree for a free extent of a given size, but this
7134 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7135 * any of the information in this block group.
7137 static noinline void
7138 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7141 struct btrfs_caching_control *caching_ctl;
7143 caching_ctl = get_caching_control(cache);
7147 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7148 (cache->free_space_ctl->free_space >= num_bytes));
7150 put_caching_control(caching_ctl);
7154 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7156 struct btrfs_caching_control *caching_ctl;
7159 caching_ctl = get_caching_control(cache);
7161 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7163 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7164 if (cache->cached == BTRFS_CACHE_ERROR)
7166 put_caching_control(caching_ctl);
7170 enum btrfs_loop_type {
7171 LOOP_CACHING_NOWAIT = 0,
7172 LOOP_CACHING_WAIT = 1,
7173 LOOP_ALLOC_CHUNK = 2,
7174 LOOP_NO_EMPTY_SIZE = 3,
7178 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7182 down_read(&cache->data_rwsem);
7186 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7189 btrfs_get_block_group(cache);
7191 down_read(&cache->data_rwsem);
7194 static struct btrfs_block_group_cache *
7195 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7196 struct btrfs_free_cluster *cluster,
7199 struct btrfs_block_group_cache *used_bg = NULL;
7201 spin_lock(&cluster->refill_lock);
7203 used_bg = cluster->block_group;
7207 if (used_bg == block_group)
7210 btrfs_get_block_group(used_bg);
7215 if (down_read_trylock(&used_bg->data_rwsem))
7218 spin_unlock(&cluster->refill_lock);
7220 /* We should only have one-level nested. */
7221 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7223 spin_lock(&cluster->refill_lock);
7224 if (used_bg == cluster->block_group)
7227 up_read(&used_bg->data_rwsem);
7228 btrfs_put_block_group(used_bg);
7233 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7237 up_read(&cache->data_rwsem);
7238 btrfs_put_block_group(cache);
7242 * walks the btree of allocated extents and find a hole of a given size.
7243 * The key ins is changed to record the hole:
7244 * ins->objectid == start position
7245 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7246 * ins->offset == the size of the hole.
7247 * Any available blocks before search_start are skipped.
7249 * If there is no suitable free space, we will record the max size of
7250 * the free space extent currently.
7252 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7253 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7254 u64 hint_byte, struct btrfs_key *ins,
7255 u64 flags, int delalloc)
7258 struct btrfs_root *root = fs_info->extent_root;
7259 struct btrfs_free_cluster *last_ptr = NULL;
7260 struct btrfs_block_group_cache *block_group = NULL;
7261 u64 search_start = 0;
7262 u64 max_extent_size = 0;
7263 u64 empty_cluster = 0;
7264 struct btrfs_space_info *space_info;
7266 int index = btrfs_bg_flags_to_raid_index(flags);
7267 bool failed_cluster_refill = false;
7268 bool failed_alloc = false;
7269 bool use_cluster = true;
7270 bool have_caching_bg = false;
7271 bool orig_have_caching_bg = false;
7272 bool full_search = false;
7274 WARN_ON(num_bytes < fs_info->sectorsize);
7275 ins->type = BTRFS_EXTENT_ITEM_KEY;
7279 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7281 space_info = __find_space_info(fs_info, flags);
7283 btrfs_err(fs_info, "No space info for %llu", flags);
7288 * If our free space is heavily fragmented we may not be able to make
7289 * big contiguous allocations, so instead of doing the expensive search
7290 * for free space, simply return ENOSPC with our max_extent_size so we
7291 * can go ahead and search for a more manageable chunk.
7293 * If our max_extent_size is large enough for our allocation simply
7294 * disable clustering since we will likely not be able to find enough
7295 * space to create a cluster and induce latency trying.
7297 if (unlikely(space_info->max_extent_size)) {
7298 spin_lock(&space_info->lock);
7299 if (space_info->max_extent_size &&
7300 num_bytes > space_info->max_extent_size) {
7301 ins->offset = space_info->max_extent_size;
7302 spin_unlock(&space_info->lock);
7304 } else if (space_info->max_extent_size) {
7305 use_cluster = false;
7307 spin_unlock(&space_info->lock);
7310 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7312 spin_lock(&last_ptr->lock);
7313 if (last_ptr->block_group)
7314 hint_byte = last_ptr->window_start;
7315 if (last_ptr->fragmented) {
7317 * We still set window_start so we can keep track of the
7318 * last place we found an allocation to try and save
7321 hint_byte = last_ptr->window_start;
7322 use_cluster = false;
7324 spin_unlock(&last_ptr->lock);
7327 search_start = max(search_start, first_logical_byte(fs_info, 0));
7328 search_start = max(search_start, hint_byte);
7329 if (search_start == hint_byte) {
7330 block_group = btrfs_lookup_block_group(fs_info, search_start);
7332 * we don't want to use the block group if it doesn't match our
7333 * allocation bits, or if its not cached.
7335 * However if we are re-searching with an ideal block group
7336 * picked out then we don't care that the block group is cached.
7338 if (block_group && block_group_bits(block_group, flags) &&
7339 block_group->cached != BTRFS_CACHE_NO) {
7340 down_read(&space_info->groups_sem);
7341 if (list_empty(&block_group->list) ||
7344 * someone is removing this block group,
7345 * we can't jump into the have_block_group
7346 * target because our list pointers are not
7349 btrfs_put_block_group(block_group);
7350 up_read(&space_info->groups_sem);
7352 index = btrfs_bg_flags_to_raid_index(
7353 block_group->flags);
7354 btrfs_lock_block_group(block_group, delalloc);
7355 goto have_block_group;
7357 } else if (block_group) {
7358 btrfs_put_block_group(block_group);
7362 have_caching_bg = false;
7363 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7365 down_read(&space_info->groups_sem);
7366 list_for_each_entry(block_group, &space_info->block_groups[index],
7371 /* If the block group is read-only, we can skip it entirely. */
7372 if (unlikely(block_group->ro))
7375 btrfs_grab_block_group(block_group, delalloc);
7376 search_start = block_group->key.objectid;
7379 * this can happen if we end up cycling through all the
7380 * raid types, but we want to make sure we only allocate
7381 * for the proper type.
7383 if (!block_group_bits(block_group, flags)) {
7384 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7385 BTRFS_BLOCK_GROUP_RAID1 |
7386 BTRFS_BLOCK_GROUP_RAID5 |
7387 BTRFS_BLOCK_GROUP_RAID6 |
7388 BTRFS_BLOCK_GROUP_RAID10;
7391 * if they asked for extra copies and this block group
7392 * doesn't provide them, bail. This does allow us to
7393 * fill raid0 from raid1.
7395 if ((flags & extra) && !(block_group->flags & extra))
7400 cached = block_group_cache_done(block_group);
7401 if (unlikely(!cached)) {
7402 have_caching_bg = true;
7403 ret = cache_block_group(block_group, 0);
7408 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7412 * Ok we want to try and use the cluster allocator, so
7415 if (last_ptr && use_cluster) {
7416 struct btrfs_block_group_cache *used_block_group;
7417 unsigned long aligned_cluster;
7419 * the refill lock keeps out other
7420 * people trying to start a new cluster
7422 used_block_group = btrfs_lock_cluster(block_group,
7425 if (!used_block_group)
7426 goto refill_cluster;
7428 if (used_block_group != block_group &&
7429 (used_block_group->ro ||
7430 !block_group_bits(used_block_group, flags)))
7431 goto release_cluster;
7433 offset = btrfs_alloc_from_cluster(used_block_group,
7436 used_block_group->key.objectid,
7439 /* we have a block, we're done */
7440 spin_unlock(&last_ptr->refill_lock);
7441 trace_btrfs_reserve_extent_cluster(
7443 search_start, num_bytes);
7444 if (used_block_group != block_group) {
7445 btrfs_release_block_group(block_group,
7447 block_group = used_block_group;
7452 WARN_ON(last_ptr->block_group != used_block_group);
7454 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7455 * set up a new clusters, so lets just skip it
7456 * and let the allocator find whatever block
7457 * it can find. If we reach this point, we
7458 * will have tried the cluster allocator
7459 * plenty of times and not have found
7460 * anything, so we are likely way too
7461 * fragmented for the clustering stuff to find
7464 * However, if the cluster is taken from the
7465 * current block group, release the cluster
7466 * first, so that we stand a better chance of
7467 * succeeding in the unclustered
7469 if (loop >= LOOP_NO_EMPTY_SIZE &&
7470 used_block_group != block_group) {
7471 spin_unlock(&last_ptr->refill_lock);
7472 btrfs_release_block_group(used_block_group,
7474 goto unclustered_alloc;
7478 * this cluster didn't work out, free it and
7481 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7483 if (used_block_group != block_group)
7484 btrfs_release_block_group(used_block_group,
7487 if (loop >= LOOP_NO_EMPTY_SIZE) {
7488 spin_unlock(&last_ptr->refill_lock);
7489 goto unclustered_alloc;
7492 aligned_cluster = max_t(unsigned long,
7493 empty_cluster + empty_size,
7494 block_group->full_stripe_len);
7496 /* allocate a cluster in this block group */
7497 ret = btrfs_find_space_cluster(fs_info, block_group,
7498 last_ptr, search_start,
7503 * now pull our allocation out of this
7506 offset = btrfs_alloc_from_cluster(block_group,
7512 /* we found one, proceed */
7513 spin_unlock(&last_ptr->refill_lock);
7514 trace_btrfs_reserve_extent_cluster(
7515 block_group, search_start,
7519 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7520 && !failed_cluster_refill) {
7521 spin_unlock(&last_ptr->refill_lock);
7523 failed_cluster_refill = true;
7524 wait_block_group_cache_progress(block_group,
7525 num_bytes + empty_cluster + empty_size);
7526 goto have_block_group;
7530 * at this point we either didn't find a cluster
7531 * or we weren't able to allocate a block from our
7532 * cluster. Free the cluster we've been trying
7533 * to use, and go to the next block group
7535 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7536 spin_unlock(&last_ptr->refill_lock);
7542 * We are doing an unclustered alloc, set the fragmented flag so
7543 * we don't bother trying to setup a cluster again until we get
7546 if (unlikely(last_ptr)) {
7547 spin_lock(&last_ptr->lock);
7548 last_ptr->fragmented = 1;
7549 spin_unlock(&last_ptr->lock);
7552 struct btrfs_free_space_ctl *ctl =
7553 block_group->free_space_ctl;
7555 spin_lock(&ctl->tree_lock);
7556 if (ctl->free_space <
7557 num_bytes + empty_cluster + empty_size) {
7558 if (ctl->free_space > max_extent_size)
7559 max_extent_size = ctl->free_space;
7560 spin_unlock(&ctl->tree_lock);
7563 spin_unlock(&ctl->tree_lock);
7566 offset = btrfs_find_space_for_alloc(block_group, search_start,
7567 num_bytes, empty_size,
7570 * If we didn't find a chunk, and we haven't failed on this
7571 * block group before, and this block group is in the middle of
7572 * caching and we are ok with waiting, then go ahead and wait
7573 * for progress to be made, and set failed_alloc to true.
7575 * If failed_alloc is true then we've already waited on this
7576 * block group once and should move on to the next block group.
7578 if (!offset && !failed_alloc && !cached &&
7579 loop > LOOP_CACHING_NOWAIT) {
7580 wait_block_group_cache_progress(block_group,
7581 num_bytes + empty_size);
7582 failed_alloc = true;
7583 goto have_block_group;
7584 } else if (!offset) {
7588 search_start = round_up(offset, fs_info->stripesize);
7590 /* move on to the next group */
7591 if (search_start + num_bytes >
7592 block_group->key.objectid + block_group->key.offset) {
7593 btrfs_add_free_space(block_group, offset, num_bytes);
7597 if (offset < search_start)
7598 btrfs_add_free_space(block_group, offset,
7599 search_start - offset);
7601 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7602 num_bytes, delalloc);
7603 if (ret == -EAGAIN) {
7604 btrfs_add_free_space(block_group, offset, num_bytes);
7607 btrfs_inc_block_group_reservations(block_group);
7609 /* we are all good, lets return */
7610 ins->objectid = search_start;
7611 ins->offset = num_bytes;
7613 trace_btrfs_reserve_extent(block_group, search_start, num_bytes);
7614 btrfs_release_block_group(block_group, delalloc);
7617 failed_cluster_refill = false;
7618 failed_alloc = false;
7619 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7621 btrfs_release_block_group(block_group, delalloc);
7624 up_read(&space_info->groups_sem);
7626 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7627 && !orig_have_caching_bg)
7628 orig_have_caching_bg = true;
7630 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7633 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7637 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7638 * caching kthreads as we move along
7639 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7640 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7641 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7644 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7646 if (loop == LOOP_CACHING_NOWAIT) {
7648 * We want to skip the LOOP_CACHING_WAIT step if we
7649 * don't have any uncached bgs and we've already done a
7650 * full search through.
7652 if (orig_have_caching_bg || !full_search)
7653 loop = LOOP_CACHING_WAIT;
7655 loop = LOOP_ALLOC_CHUNK;
7660 if (loop == LOOP_ALLOC_CHUNK) {
7661 struct btrfs_trans_handle *trans;
7664 trans = current->journal_info;
7668 trans = btrfs_join_transaction(root);
7670 if (IS_ERR(trans)) {
7671 ret = PTR_ERR(trans);
7675 ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE);
7678 * If we can't allocate a new chunk we've already looped
7679 * through at least once, move on to the NO_EMPTY_SIZE
7683 loop = LOOP_NO_EMPTY_SIZE;
7686 * Do not bail out on ENOSPC since we
7687 * can do more things.
7689 if (ret < 0 && ret != -ENOSPC)
7690 btrfs_abort_transaction(trans, ret);
7694 btrfs_end_transaction(trans);
7699 if (loop == LOOP_NO_EMPTY_SIZE) {
7701 * Don't loop again if we already have no empty_size and
7704 if (empty_size == 0 &&
7705 empty_cluster == 0) {
7714 } else if (!ins->objectid) {
7716 } else if (ins->objectid) {
7717 if (!use_cluster && last_ptr) {
7718 spin_lock(&last_ptr->lock);
7719 last_ptr->window_start = ins->objectid;
7720 spin_unlock(&last_ptr->lock);
7725 if (ret == -ENOSPC) {
7726 spin_lock(&space_info->lock);
7727 space_info->max_extent_size = max_extent_size;
7728 spin_unlock(&space_info->lock);
7729 ins->offset = max_extent_size;
7734 static void dump_space_info(struct btrfs_fs_info *fs_info,
7735 struct btrfs_space_info *info, u64 bytes,
7736 int dump_block_groups)
7738 struct btrfs_block_group_cache *cache;
7741 spin_lock(&info->lock);
7742 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7744 info->total_bytes - btrfs_space_info_used(info, true),
7745 info->full ? "" : "not ");
7747 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7748 info->total_bytes, info->bytes_used, info->bytes_pinned,
7749 info->bytes_reserved, info->bytes_may_use,
7750 info->bytes_readonly);
7751 spin_unlock(&info->lock);
7753 if (!dump_block_groups)
7756 down_read(&info->groups_sem);
7758 list_for_each_entry(cache, &info->block_groups[index], list) {
7759 spin_lock(&cache->lock);
7761 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7762 cache->key.objectid, cache->key.offset,
7763 btrfs_block_group_used(&cache->item), cache->pinned,
7764 cache->reserved, cache->ro ? "[readonly]" : "");
7765 btrfs_dump_free_space(cache, bytes);
7766 spin_unlock(&cache->lock);
7768 if (++index < BTRFS_NR_RAID_TYPES)
7770 up_read(&info->groups_sem);
7774 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7775 * hole that is at least as big as @num_bytes.
7777 * @root - The root that will contain this extent
7779 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7780 * is used for accounting purposes. This value differs
7781 * from @num_bytes only in the case of compressed extents.
7783 * @num_bytes - Number of bytes to allocate on-disk.
7785 * @min_alloc_size - Indicates the minimum amount of space that the
7786 * allocator should try to satisfy. In some cases
7787 * @num_bytes may be larger than what is required and if
7788 * the filesystem is fragmented then allocation fails.
7789 * However, the presence of @min_alloc_size gives a
7790 * chance to try and satisfy the smaller allocation.
7792 * @empty_size - A hint that you plan on doing more COW. This is the
7793 * size in bytes the allocator should try to find free
7794 * next to the block it returns. This is just a hint and
7795 * may be ignored by the allocator.
7797 * @hint_byte - Hint to the allocator to start searching above the byte
7798 * address passed. It might be ignored.
7800 * @ins - This key is modified to record the found hole. It will
7801 * have the following values:
7802 * ins->objectid == start position
7803 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7804 * ins->offset == the size of the hole.
7806 * @is_data - Boolean flag indicating whether an extent is
7807 * allocated for data (true) or metadata (false)
7809 * @delalloc - Boolean flag indicating whether this allocation is for
7810 * delalloc or not. If 'true' data_rwsem of block groups
7811 * is going to be acquired.
7814 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7815 * case -ENOSPC is returned then @ins->offset will contain the size of the
7816 * largest available hole the allocator managed to find.
7818 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7819 u64 num_bytes, u64 min_alloc_size,
7820 u64 empty_size, u64 hint_byte,
7821 struct btrfs_key *ins, int is_data, int delalloc)
7823 struct btrfs_fs_info *fs_info = root->fs_info;
7824 bool final_tried = num_bytes == min_alloc_size;
7828 flags = get_alloc_profile_by_root(root, is_data);
7830 WARN_ON(num_bytes < fs_info->sectorsize);
7831 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7832 hint_byte, ins, flags, delalloc);
7833 if (!ret && !is_data) {
7834 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7835 } else if (ret == -ENOSPC) {
7836 if (!final_tried && ins->offset) {
7837 num_bytes = min(num_bytes >> 1, ins->offset);
7838 num_bytes = round_down(num_bytes,
7839 fs_info->sectorsize);
7840 num_bytes = max(num_bytes, min_alloc_size);
7841 ram_bytes = num_bytes;
7842 if (num_bytes == min_alloc_size)
7845 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7846 struct btrfs_space_info *sinfo;
7848 sinfo = __find_space_info(fs_info, flags);
7850 "allocation failed flags %llu, wanted %llu",
7853 dump_space_info(fs_info, sinfo, num_bytes, 1);
7860 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
7862 int pin, int delalloc)
7864 struct btrfs_block_group_cache *cache;
7867 cache = btrfs_lookup_block_group(fs_info, start);
7869 btrfs_err(fs_info, "Unable to find block group for %llu",
7875 pin_down_extent(fs_info, cache, start, len, 1);
7877 if (btrfs_test_opt(fs_info, DISCARD))
7878 ret = btrfs_discard_extent(fs_info, start, len, NULL);
7879 btrfs_add_free_space(cache, start, len);
7880 btrfs_free_reserved_bytes(cache, len, delalloc);
7881 trace_btrfs_reserved_extent_free(fs_info, start, len);
7884 btrfs_put_block_group(cache);
7888 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
7889 u64 start, u64 len, int delalloc)
7891 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
7894 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
7897 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
7900 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7901 u64 parent, u64 root_objectid,
7902 u64 flags, u64 owner, u64 offset,
7903 struct btrfs_key *ins, int ref_mod)
7905 struct btrfs_fs_info *fs_info = trans->fs_info;
7907 struct btrfs_extent_item *extent_item;
7908 struct btrfs_extent_inline_ref *iref;
7909 struct btrfs_path *path;
7910 struct extent_buffer *leaf;
7915 type = BTRFS_SHARED_DATA_REF_KEY;
7917 type = BTRFS_EXTENT_DATA_REF_KEY;
7919 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7921 path = btrfs_alloc_path();
7925 path->leave_spinning = 1;
7926 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7929 btrfs_free_path(path);
7933 leaf = path->nodes[0];
7934 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7935 struct btrfs_extent_item);
7936 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7937 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7938 btrfs_set_extent_flags(leaf, extent_item,
7939 flags | BTRFS_EXTENT_FLAG_DATA);
7941 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7942 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7944 struct btrfs_shared_data_ref *ref;
7945 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7946 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7947 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7949 struct btrfs_extent_data_ref *ref;
7950 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7951 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7952 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7953 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7954 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7957 btrfs_mark_buffer_dirty(path->nodes[0]);
7958 btrfs_free_path(path);
7960 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
7964 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
7965 if (ret) { /* -ENOENT, logic error */
7966 btrfs_err(fs_info, "update block group failed for %llu %llu",
7967 ins->objectid, ins->offset);
7970 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
7974 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7975 struct btrfs_delayed_ref_node *node,
7976 struct btrfs_delayed_extent_op *extent_op)
7978 struct btrfs_fs_info *fs_info = trans->fs_info;
7980 struct btrfs_extent_item *extent_item;
7981 struct btrfs_key extent_key;
7982 struct btrfs_tree_block_info *block_info;
7983 struct btrfs_extent_inline_ref *iref;
7984 struct btrfs_path *path;
7985 struct extent_buffer *leaf;
7986 struct btrfs_delayed_tree_ref *ref;
7987 u32 size = sizeof(*extent_item) + sizeof(*iref);
7989 u64 flags = extent_op->flags_to_set;
7990 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
7992 ref = btrfs_delayed_node_to_tree_ref(node);
7994 extent_key.objectid = node->bytenr;
7995 if (skinny_metadata) {
7996 extent_key.offset = ref->level;
7997 extent_key.type = BTRFS_METADATA_ITEM_KEY;
7998 num_bytes = fs_info->nodesize;
8000 extent_key.offset = node->num_bytes;
8001 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8002 size += sizeof(*block_info);
8003 num_bytes = node->num_bytes;
8006 path = btrfs_alloc_path();
8008 btrfs_free_and_pin_reserved_extent(fs_info,
8009 extent_key.objectid,
8014 path->leave_spinning = 1;
8015 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8018 btrfs_free_path(path);
8019 btrfs_free_and_pin_reserved_extent(fs_info,
8020 extent_key.objectid,
8025 leaf = path->nodes[0];
8026 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8027 struct btrfs_extent_item);
8028 btrfs_set_extent_refs(leaf, extent_item, 1);
8029 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8030 btrfs_set_extent_flags(leaf, extent_item,
8031 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8033 if (skinny_metadata) {
8034 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8036 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8037 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8038 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8039 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8042 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8043 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8044 btrfs_set_extent_inline_ref_type(leaf, iref,
8045 BTRFS_SHARED_BLOCK_REF_KEY);
8046 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8048 btrfs_set_extent_inline_ref_type(leaf, iref,
8049 BTRFS_TREE_BLOCK_REF_KEY);
8050 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8053 btrfs_mark_buffer_dirty(leaf);
8054 btrfs_free_path(path);
8056 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8061 ret = update_block_group(trans, fs_info, extent_key.objectid,
8062 fs_info->nodesize, 1);
8063 if (ret) { /* -ENOENT, logic error */
8064 btrfs_err(fs_info, "update block group failed for %llu %llu",
8065 extent_key.objectid, extent_key.offset);
8069 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8074 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8075 struct btrfs_root *root, u64 owner,
8076 u64 offset, u64 ram_bytes,
8077 struct btrfs_key *ins)
8081 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8083 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8084 root->root_key.objectid, owner, offset,
8085 BTRFS_ADD_DELAYED_EXTENT);
8087 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8089 root->root_key.objectid, owner,
8091 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8096 * this is used by the tree logging recovery code. It records that
8097 * an extent has been allocated and makes sure to clear the free
8098 * space cache bits as well
8100 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8101 u64 root_objectid, u64 owner, u64 offset,
8102 struct btrfs_key *ins)
8104 struct btrfs_fs_info *fs_info = trans->fs_info;
8106 struct btrfs_block_group_cache *block_group;
8107 struct btrfs_space_info *space_info;
8110 * Mixed block groups will exclude before processing the log so we only
8111 * need to do the exclude dance if this fs isn't mixed.
8113 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8114 ret = __exclude_logged_extent(fs_info, ins->objectid,
8120 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8124 space_info = block_group->space_info;
8125 spin_lock(&space_info->lock);
8126 spin_lock(&block_group->lock);
8127 space_info->bytes_reserved += ins->offset;
8128 block_group->reserved += ins->offset;
8129 spin_unlock(&block_group->lock);
8130 spin_unlock(&space_info->lock);
8132 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8134 btrfs_put_block_group(block_group);
8138 static struct extent_buffer *
8139 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8140 u64 bytenr, int level, u64 owner)
8142 struct btrfs_fs_info *fs_info = root->fs_info;
8143 struct extent_buffer *buf;
8145 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8150 * Extra safety check in case the extent tree is corrupted and extent
8151 * allocator chooses to use a tree block which is already used and
8154 if (buf->lock_owner == current->pid) {
8155 btrfs_err_rl(fs_info,
8156 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8157 buf->start, btrfs_header_owner(buf), current->pid);
8158 free_extent_buffer(buf);
8159 return ERR_PTR(-EUCLEAN);
8162 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8163 btrfs_tree_lock(buf);
8164 clean_tree_block(fs_info, buf);
8165 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8167 btrfs_set_lock_blocking(buf);
8168 set_extent_buffer_uptodate(buf);
8170 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8171 btrfs_set_header_level(buf, level);
8172 btrfs_set_header_bytenr(buf, buf->start);
8173 btrfs_set_header_generation(buf, trans->transid);
8174 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8175 btrfs_set_header_owner(buf, owner);
8176 write_extent_buffer_fsid(buf, fs_info->fsid);
8177 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8178 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8179 buf->log_index = root->log_transid % 2;
8181 * we allow two log transactions at a time, use different
8182 * EXENT bit to differentiate dirty pages.
8184 if (buf->log_index == 0)
8185 set_extent_dirty(&root->dirty_log_pages, buf->start,
8186 buf->start + buf->len - 1, GFP_NOFS);
8188 set_extent_new(&root->dirty_log_pages, buf->start,
8189 buf->start + buf->len - 1);
8191 buf->log_index = -1;
8192 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8193 buf->start + buf->len - 1, GFP_NOFS);
8195 trans->dirty = true;
8196 /* this returns a buffer locked for blocking */
8200 static struct btrfs_block_rsv *
8201 use_block_rsv(struct btrfs_trans_handle *trans,
8202 struct btrfs_root *root, u32 blocksize)
8204 struct btrfs_fs_info *fs_info = root->fs_info;
8205 struct btrfs_block_rsv *block_rsv;
8206 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8208 bool global_updated = false;
8210 block_rsv = get_block_rsv(trans, root);
8212 if (unlikely(block_rsv->size == 0))
8215 ret = block_rsv_use_bytes(block_rsv, blocksize);
8219 if (block_rsv->failfast)
8220 return ERR_PTR(ret);
8222 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8223 global_updated = true;
8224 update_global_block_rsv(fs_info);
8228 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8229 static DEFINE_RATELIMIT_STATE(_rs,
8230 DEFAULT_RATELIMIT_INTERVAL * 10,
8231 /*DEFAULT_RATELIMIT_BURST*/ 1);
8232 if (__ratelimit(&_rs))
8234 "BTRFS: block rsv returned %d\n", ret);
8237 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8238 BTRFS_RESERVE_NO_FLUSH);
8242 * If we couldn't reserve metadata bytes try and use some from
8243 * the global reserve if its space type is the same as the global
8246 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8247 block_rsv->space_info == global_rsv->space_info) {
8248 ret = block_rsv_use_bytes(global_rsv, blocksize);
8252 return ERR_PTR(ret);
8255 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8256 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8258 block_rsv_add_bytes(block_rsv, blocksize, false);
8259 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8263 * finds a free extent and does all the dirty work required for allocation
8264 * returns the tree buffer or an ERR_PTR on error.
8266 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8267 struct btrfs_root *root,
8268 u64 parent, u64 root_objectid,
8269 const struct btrfs_disk_key *key,
8270 int level, u64 hint,
8273 struct btrfs_fs_info *fs_info = root->fs_info;
8274 struct btrfs_key ins;
8275 struct btrfs_block_rsv *block_rsv;
8276 struct extent_buffer *buf;
8277 struct btrfs_delayed_extent_op *extent_op;
8280 u32 blocksize = fs_info->nodesize;
8281 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8283 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8284 if (btrfs_is_testing(fs_info)) {
8285 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8286 level, root_objectid);
8288 root->alloc_bytenr += blocksize;
8293 block_rsv = use_block_rsv(trans, root, blocksize);
8294 if (IS_ERR(block_rsv))
8295 return ERR_CAST(block_rsv);
8297 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8298 empty_size, hint, &ins, 0, 0);
8302 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8306 goto out_free_reserved;
8309 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8311 parent = ins.objectid;
8312 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8316 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8317 extent_op = btrfs_alloc_delayed_extent_op();
8323 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8325 memset(&extent_op->key, 0, sizeof(extent_op->key));
8326 extent_op->flags_to_set = flags;
8327 extent_op->update_key = skinny_metadata ? false : true;
8328 extent_op->update_flags = true;
8329 extent_op->is_data = false;
8330 extent_op->level = level;
8332 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8333 root_objectid, level, 0,
8334 BTRFS_ADD_DELAYED_EXTENT);
8335 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8337 root_objectid, level,
8338 BTRFS_ADD_DELAYED_EXTENT,
8339 extent_op, NULL, NULL);
8341 goto out_free_delayed;
8346 btrfs_free_delayed_extent_op(extent_op);
8348 free_extent_buffer(buf);
8350 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8352 unuse_block_rsv(fs_info, block_rsv, blocksize);
8353 return ERR_PTR(ret);
8356 struct walk_control {
8357 u64 refs[BTRFS_MAX_LEVEL];
8358 u64 flags[BTRFS_MAX_LEVEL];
8359 struct btrfs_key update_progress;
8369 #define DROP_REFERENCE 1
8370 #define UPDATE_BACKREF 2
8372 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8373 struct btrfs_root *root,
8374 struct walk_control *wc,
8375 struct btrfs_path *path)
8377 struct btrfs_fs_info *fs_info = root->fs_info;
8383 struct btrfs_key key;
8384 struct extent_buffer *eb;
8389 if (path->slots[wc->level] < wc->reada_slot) {
8390 wc->reada_count = wc->reada_count * 2 / 3;
8391 wc->reada_count = max(wc->reada_count, 2);
8393 wc->reada_count = wc->reada_count * 3 / 2;
8394 wc->reada_count = min_t(int, wc->reada_count,
8395 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8398 eb = path->nodes[wc->level];
8399 nritems = btrfs_header_nritems(eb);
8401 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8402 if (nread >= wc->reada_count)
8406 bytenr = btrfs_node_blockptr(eb, slot);
8407 generation = btrfs_node_ptr_generation(eb, slot);
8409 if (slot == path->slots[wc->level])
8412 if (wc->stage == UPDATE_BACKREF &&
8413 generation <= root->root_key.offset)
8416 /* We don't lock the tree block, it's OK to be racy here */
8417 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8418 wc->level - 1, 1, &refs,
8420 /* We don't care about errors in readahead. */
8425 if (wc->stage == DROP_REFERENCE) {
8429 if (wc->level == 1 &&
8430 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8432 if (!wc->update_ref ||
8433 generation <= root->root_key.offset)
8435 btrfs_node_key_to_cpu(eb, &key, slot);
8436 ret = btrfs_comp_cpu_keys(&key,
8437 &wc->update_progress);
8441 if (wc->level == 1 &&
8442 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8446 readahead_tree_block(fs_info, bytenr);
8449 wc->reada_slot = slot;
8453 * helper to process tree block while walking down the tree.
8455 * when wc->stage == UPDATE_BACKREF, this function updates
8456 * back refs for pointers in the block.
8458 * NOTE: return value 1 means we should stop walking down.
8460 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8461 struct btrfs_root *root,
8462 struct btrfs_path *path,
8463 struct walk_control *wc, int lookup_info)
8465 struct btrfs_fs_info *fs_info = root->fs_info;
8466 int level = wc->level;
8467 struct extent_buffer *eb = path->nodes[level];
8468 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8471 if (wc->stage == UPDATE_BACKREF &&
8472 btrfs_header_owner(eb) != root->root_key.objectid)
8476 * when reference count of tree block is 1, it won't increase
8477 * again. once full backref flag is set, we never clear it.
8480 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8481 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8482 BUG_ON(!path->locks[level]);
8483 ret = btrfs_lookup_extent_info(trans, fs_info,
8484 eb->start, level, 1,
8487 BUG_ON(ret == -ENOMEM);
8490 BUG_ON(wc->refs[level] == 0);
8493 if (wc->stage == DROP_REFERENCE) {
8494 if (wc->refs[level] > 1)
8497 if (path->locks[level] && !wc->keep_locks) {
8498 btrfs_tree_unlock_rw(eb, path->locks[level]);
8499 path->locks[level] = 0;
8504 /* wc->stage == UPDATE_BACKREF */
8505 if (!(wc->flags[level] & flag)) {
8506 BUG_ON(!path->locks[level]);
8507 ret = btrfs_inc_ref(trans, root, eb, 1);
8508 BUG_ON(ret); /* -ENOMEM */
8509 ret = btrfs_dec_ref(trans, root, eb, 0);
8510 BUG_ON(ret); /* -ENOMEM */
8511 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8513 btrfs_header_level(eb), 0);
8514 BUG_ON(ret); /* -ENOMEM */
8515 wc->flags[level] |= flag;
8519 * the block is shared by multiple trees, so it's not good to
8520 * keep the tree lock
8522 if (path->locks[level] && level > 0) {
8523 btrfs_tree_unlock_rw(eb, path->locks[level]);
8524 path->locks[level] = 0;
8530 * helper to process tree block pointer.
8532 * when wc->stage == DROP_REFERENCE, this function checks
8533 * reference count of the block pointed to. if the block
8534 * is shared and we need update back refs for the subtree
8535 * rooted at the block, this function changes wc->stage to
8536 * UPDATE_BACKREF. if the block is shared and there is no
8537 * need to update back, this function drops the reference
8540 * NOTE: return value 1 means we should stop walking down.
8542 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8543 struct btrfs_root *root,
8544 struct btrfs_path *path,
8545 struct walk_control *wc, int *lookup_info)
8547 struct btrfs_fs_info *fs_info = root->fs_info;
8552 struct btrfs_key key;
8553 struct btrfs_key first_key;
8554 struct extent_buffer *next;
8555 int level = wc->level;
8558 bool need_account = false;
8560 generation = btrfs_node_ptr_generation(path->nodes[level],
8561 path->slots[level]);
8563 * if the lower level block was created before the snapshot
8564 * was created, we know there is no need to update back refs
8567 if (wc->stage == UPDATE_BACKREF &&
8568 generation <= root->root_key.offset) {
8573 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8574 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8575 path->slots[level]);
8576 blocksize = fs_info->nodesize;
8578 next = find_extent_buffer(fs_info, bytenr);
8580 next = btrfs_find_create_tree_block(fs_info, bytenr);
8582 return PTR_ERR(next);
8584 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8588 btrfs_tree_lock(next);
8589 btrfs_set_lock_blocking(next);
8591 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8592 &wc->refs[level - 1],
8593 &wc->flags[level - 1]);
8597 if (unlikely(wc->refs[level - 1] == 0)) {
8598 btrfs_err(fs_info, "Missing references.");
8604 if (wc->stage == DROP_REFERENCE) {
8605 if (wc->refs[level - 1] > 1) {
8606 need_account = true;
8608 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8611 if (!wc->update_ref ||
8612 generation <= root->root_key.offset)
8615 btrfs_node_key_to_cpu(path->nodes[level], &key,
8616 path->slots[level]);
8617 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8621 wc->stage = UPDATE_BACKREF;
8622 wc->shared_level = level - 1;
8626 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8630 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8631 btrfs_tree_unlock(next);
8632 free_extent_buffer(next);
8638 if (reada && level == 1)
8639 reada_walk_down(trans, root, wc, path);
8640 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8643 return PTR_ERR(next);
8644 } else if (!extent_buffer_uptodate(next)) {
8645 free_extent_buffer(next);
8648 btrfs_tree_lock(next);
8649 btrfs_set_lock_blocking(next);
8653 ASSERT(level == btrfs_header_level(next));
8654 if (level != btrfs_header_level(next)) {
8655 btrfs_err(root->fs_info, "mismatched level");
8659 path->nodes[level] = next;
8660 path->slots[level] = 0;
8661 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8667 wc->refs[level - 1] = 0;
8668 wc->flags[level - 1] = 0;
8669 if (wc->stage == DROP_REFERENCE) {
8670 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8671 parent = path->nodes[level]->start;
8673 ASSERT(root->root_key.objectid ==
8674 btrfs_header_owner(path->nodes[level]));
8675 if (root->root_key.objectid !=
8676 btrfs_header_owner(path->nodes[level])) {
8677 btrfs_err(root->fs_info,
8678 "mismatched block owner");
8686 * Reloc tree doesn't contribute to qgroup numbers, and we have
8687 * already accounted them at merge time (replace_path),
8688 * thus we could skip expensive subtree trace here.
8690 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
8692 ret = btrfs_qgroup_trace_subtree(trans, next,
8693 generation, level - 1);
8695 btrfs_err_rl(fs_info,
8696 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8700 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8701 parent, root->root_key.objectid,
8711 btrfs_tree_unlock(next);
8712 free_extent_buffer(next);
8718 * helper to process tree block while walking up the tree.
8720 * when wc->stage == DROP_REFERENCE, this function drops
8721 * reference count on the block.
8723 * when wc->stage == UPDATE_BACKREF, this function changes
8724 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8725 * to UPDATE_BACKREF previously while processing the block.
8727 * NOTE: return value 1 means we should stop walking up.
8729 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8730 struct btrfs_root *root,
8731 struct btrfs_path *path,
8732 struct walk_control *wc)
8734 struct btrfs_fs_info *fs_info = root->fs_info;
8736 int level = wc->level;
8737 struct extent_buffer *eb = path->nodes[level];
8740 if (wc->stage == UPDATE_BACKREF) {
8741 BUG_ON(wc->shared_level < level);
8742 if (level < wc->shared_level)
8745 ret = find_next_key(path, level + 1, &wc->update_progress);
8749 wc->stage = DROP_REFERENCE;
8750 wc->shared_level = -1;
8751 path->slots[level] = 0;
8754 * check reference count again if the block isn't locked.
8755 * we should start walking down the tree again if reference
8758 if (!path->locks[level]) {
8760 btrfs_tree_lock(eb);
8761 btrfs_set_lock_blocking(eb);
8762 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8764 ret = btrfs_lookup_extent_info(trans, fs_info,
8765 eb->start, level, 1,
8769 btrfs_tree_unlock_rw(eb, path->locks[level]);
8770 path->locks[level] = 0;
8773 BUG_ON(wc->refs[level] == 0);
8774 if (wc->refs[level] == 1) {
8775 btrfs_tree_unlock_rw(eb, path->locks[level]);
8776 path->locks[level] = 0;
8782 /* wc->stage == DROP_REFERENCE */
8783 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8785 if (wc->refs[level] == 1) {
8787 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8788 ret = btrfs_dec_ref(trans, root, eb, 1);
8790 ret = btrfs_dec_ref(trans, root, eb, 0);
8791 BUG_ON(ret); /* -ENOMEM */
8792 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
8794 btrfs_err_rl(fs_info,
8795 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8799 /* make block locked assertion in clean_tree_block happy */
8800 if (!path->locks[level] &&
8801 btrfs_header_generation(eb) == trans->transid) {
8802 btrfs_tree_lock(eb);
8803 btrfs_set_lock_blocking(eb);
8804 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8806 clean_tree_block(fs_info, eb);
8809 if (eb == root->node) {
8810 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8812 else if (root->root_key.objectid != btrfs_header_owner(eb))
8813 goto owner_mismatch;
8815 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8816 parent = path->nodes[level + 1]->start;
8817 else if (root->root_key.objectid !=
8818 btrfs_header_owner(path->nodes[level + 1]))
8819 goto owner_mismatch;
8822 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8824 wc->refs[level] = 0;
8825 wc->flags[level] = 0;
8829 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
8830 btrfs_header_owner(eb), root->root_key.objectid);
8834 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8835 struct btrfs_root *root,
8836 struct btrfs_path *path,
8837 struct walk_control *wc)
8839 int level = wc->level;
8840 int lookup_info = 1;
8843 while (level >= 0) {
8844 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8851 if (path->slots[level] >=
8852 btrfs_header_nritems(path->nodes[level]))
8855 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8857 path->slots[level]++;
8866 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8867 struct btrfs_root *root,
8868 struct btrfs_path *path,
8869 struct walk_control *wc, int max_level)
8871 int level = wc->level;
8874 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8875 while (level < max_level && path->nodes[level]) {
8877 if (path->slots[level] + 1 <
8878 btrfs_header_nritems(path->nodes[level])) {
8879 path->slots[level]++;
8882 ret = walk_up_proc(trans, root, path, wc);
8888 if (path->locks[level]) {
8889 btrfs_tree_unlock_rw(path->nodes[level],
8890 path->locks[level]);
8891 path->locks[level] = 0;
8893 free_extent_buffer(path->nodes[level]);
8894 path->nodes[level] = NULL;
8902 * drop a subvolume tree.
8904 * this function traverses the tree freeing any blocks that only
8905 * referenced by the tree.
8907 * when a shared tree block is found. this function decreases its
8908 * reference count by one. if update_ref is true, this function
8909 * also make sure backrefs for the shared block and all lower level
8910 * blocks are properly updated.
8912 * If called with for_reloc == 0, may exit early with -EAGAIN
8914 int btrfs_drop_snapshot(struct btrfs_root *root,
8915 struct btrfs_block_rsv *block_rsv, int update_ref,
8918 struct btrfs_fs_info *fs_info = root->fs_info;
8919 struct btrfs_path *path;
8920 struct btrfs_trans_handle *trans;
8921 struct btrfs_root *tree_root = fs_info->tree_root;
8922 struct btrfs_root_item *root_item = &root->root_item;
8923 struct walk_control *wc;
8924 struct btrfs_key key;
8928 bool root_dropped = false;
8930 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
8932 path = btrfs_alloc_path();
8938 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8940 btrfs_free_path(path);
8945 trans = btrfs_start_transaction(tree_root, 0);
8946 if (IS_ERR(trans)) {
8947 err = PTR_ERR(trans);
8952 trans->block_rsv = block_rsv;
8954 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8955 level = btrfs_header_level(root->node);
8956 path->nodes[level] = btrfs_lock_root_node(root);
8957 btrfs_set_lock_blocking(path->nodes[level]);
8958 path->slots[level] = 0;
8959 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8960 memset(&wc->update_progress, 0,
8961 sizeof(wc->update_progress));
8963 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8964 memcpy(&wc->update_progress, &key,
8965 sizeof(wc->update_progress));
8967 level = root_item->drop_level;
8969 path->lowest_level = level;
8970 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8971 path->lowest_level = 0;
8979 * unlock our path, this is safe because only this
8980 * function is allowed to delete this snapshot
8982 btrfs_unlock_up_safe(path, 0);
8984 level = btrfs_header_level(root->node);
8986 btrfs_tree_lock(path->nodes[level]);
8987 btrfs_set_lock_blocking(path->nodes[level]);
8988 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8990 ret = btrfs_lookup_extent_info(trans, fs_info,
8991 path->nodes[level]->start,
8992 level, 1, &wc->refs[level],
8998 BUG_ON(wc->refs[level] == 0);
9000 if (level == root_item->drop_level)
9003 btrfs_tree_unlock(path->nodes[level]);
9004 path->locks[level] = 0;
9005 WARN_ON(wc->refs[level] != 1);
9011 wc->shared_level = -1;
9012 wc->stage = DROP_REFERENCE;
9013 wc->update_ref = update_ref;
9015 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9019 ret = walk_down_tree(trans, root, path, wc);
9025 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9032 BUG_ON(wc->stage != DROP_REFERENCE);
9036 if (wc->stage == DROP_REFERENCE) {
9038 btrfs_node_key(path->nodes[level],
9039 &root_item->drop_progress,
9040 path->slots[level]);
9041 root_item->drop_level = level;
9044 BUG_ON(wc->level == 0);
9045 if (btrfs_should_end_transaction(trans) ||
9046 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9047 ret = btrfs_update_root(trans, tree_root,
9051 btrfs_abort_transaction(trans, ret);
9056 btrfs_end_transaction_throttle(trans);
9057 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9058 btrfs_debug(fs_info,
9059 "drop snapshot early exit");
9064 trans = btrfs_start_transaction(tree_root, 0);
9065 if (IS_ERR(trans)) {
9066 err = PTR_ERR(trans);
9070 trans->block_rsv = block_rsv;
9073 btrfs_release_path(path);
9077 ret = btrfs_del_root(trans, &root->root_key);
9079 btrfs_abort_transaction(trans, ret);
9084 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9085 ret = btrfs_find_root(tree_root, &root->root_key, path,
9088 btrfs_abort_transaction(trans, ret);
9091 } else if (ret > 0) {
9092 /* if we fail to delete the orphan item this time
9093 * around, it'll get picked up the next time.
9095 * The most common failure here is just -ENOENT.
9097 btrfs_del_orphan_item(trans, tree_root,
9098 root->root_key.objectid);
9102 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9103 btrfs_add_dropped_root(trans, root);
9105 free_extent_buffer(root->node);
9106 free_extent_buffer(root->commit_root);
9107 btrfs_put_fs_root(root);
9109 root_dropped = true;
9111 btrfs_end_transaction_throttle(trans);
9114 btrfs_free_path(path);
9117 * So if we need to stop dropping the snapshot for whatever reason we
9118 * need to make sure to add it back to the dead root list so that we
9119 * keep trying to do the work later. This also cleans up roots if we
9120 * don't have it in the radix (like when we recover after a power fail
9121 * or unmount) so we don't leak memory.
9123 if (!for_reloc && !root_dropped)
9124 btrfs_add_dead_root(root);
9125 if (err && err != -EAGAIN)
9126 btrfs_handle_fs_error(fs_info, err, NULL);
9131 * drop subtree rooted at tree block 'node'.
9133 * NOTE: this function will unlock and release tree block 'node'
9134 * only used by relocation code
9136 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9137 struct btrfs_root *root,
9138 struct extent_buffer *node,
9139 struct extent_buffer *parent)
9141 struct btrfs_fs_info *fs_info = root->fs_info;
9142 struct btrfs_path *path;
9143 struct walk_control *wc;
9149 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9151 path = btrfs_alloc_path();
9155 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9157 btrfs_free_path(path);
9161 btrfs_assert_tree_locked(parent);
9162 parent_level = btrfs_header_level(parent);
9163 extent_buffer_get(parent);
9164 path->nodes[parent_level] = parent;
9165 path->slots[parent_level] = btrfs_header_nritems(parent);
9167 btrfs_assert_tree_locked(node);
9168 level = btrfs_header_level(node);
9169 path->nodes[level] = node;
9170 path->slots[level] = 0;
9171 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9173 wc->refs[parent_level] = 1;
9174 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9176 wc->shared_level = -1;
9177 wc->stage = DROP_REFERENCE;
9180 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9183 wret = walk_down_tree(trans, root, path, wc);
9189 wret = walk_up_tree(trans, root, path, wc, parent_level);
9197 btrfs_free_path(path);
9201 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9207 * if restripe for this chunk_type is on pick target profile and
9208 * return, otherwise do the usual balance
9210 stripped = get_restripe_target(fs_info, flags);
9212 return extended_to_chunk(stripped);
9214 num_devices = fs_info->fs_devices->rw_devices;
9216 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9217 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9218 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9220 if (num_devices == 1) {
9221 stripped |= BTRFS_BLOCK_GROUP_DUP;
9222 stripped = flags & ~stripped;
9224 /* turn raid0 into single device chunks */
9225 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9228 /* turn mirroring into duplication */
9229 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9230 BTRFS_BLOCK_GROUP_RAID10))
9231 return stripped | BTRFS_BLOCK_GROUP_DUP;
9233 /* they already had raid on here, just return */
9234 if (flags & stripped)
9237 stripped |= BTRFS_BLOCK_GROUP_DUP;
9238 stripped = flags & ~stripped;
9240 /* switch duplicated blocks with raid1 */
9241 if (flags & BTRFS_BLOCK_GROUP_DUP)
9242 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9244 /* this is drive concat, leave it alone */
9250 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9252 struct btrfs_space_info *sinfo = cache->space_info;
9254 u64 min_allocable_bytes;
9258 * We need some metadata space and system metadata space for
9259 * allocating chunks in some corner cases until we force to set
9260 * it to be readonly.
9263 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9265 min_allocable_bytes = SZ_1M;
9267 min_allocable_bytes = 0;
9269 spin_lock(&sinfo->lock);
9270 spin_lock(&cache->lock);
9278 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9279 cache->bytes_super - btrfs_block_group_used(&cache->item);
9281 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9282 min_allocable_bytes <= sinfo->total_bytes) {
9283 sinfo->bytes_readonly += num_bytes;
9285 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9289 spin_unlock(&cache->lock);
9290 spin_unlock(&sinfo->lock);
9294 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9297 struct btrfs_fs_info *fs_info = cache->fs_info;
9298 struct btrfs_trans_handle *trans;
9303 trans = btrfs_join_transaction(fs_info->extent_root);
9305 return PTR_ERR(trans);
9308 * we're not allowed to set block groups readonly after the dirty
9309 * block groups cache has started writing. If it already started,
9310 * back off and let this transaction commit
9312 mutex_lock(&fs_info->ro_block_group_mutex);
9313 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9314 u64 transid = trans->transid;
9316 mutex_unlock(&fs_info->ro_block_group_mutex);
9317 btrfs_end_transaction(trans);
9319 ret = btrfs_wait_for_commit(fs_info, transid);
9326 * if we are changing raid levels, try to allocate a corresponding
9327 * block group with the new raid level.
9329 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9330 if (alloc_flags != cache->flags) {
9331 ret = do_chunk_alloc(trans, alloc_flags,
9334 * ENOSPC is allowed here, we may have enough space
9335 * already allocated at the new raid level to
9344 ret = inc_block_group_ro(cache, 0);
9347 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9348 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9351 ret = inc_block_group_ro(cache, 0);
9353 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9354 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9355 mutex_lock(&fs_info->chunk_mutex);
9356 check_system_chunk(trans, alloc_flags);
9357 mutex_unlock(&fs_info->chunk_mutex);
9359 mutex_unlock(&fs_info->ro_block_group_mutex);
9361 btrfs_end_transaction(trans);
9365 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9367 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9369 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9373 * helper to account the unused space of all the readonly block group in the
9374 * space_info. takes mirrors into account.
9376 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9378 struct btrfs_block_group_cache *block_group;
9382 /* It's df, we don't care if it's racy */
9383 if (list_empty(&sinfo->ro_bgs))
9386 spin_lock(&sinfo->lock);
9387 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9388 spin_lock(&block_group->lock);
9390 if (!block_group->ro) {
9391 spin_unlock(&block_group->lock);
9395 factor = btrfs_bg_type_to_factor(block_group->flags);
9396 free_bytes += (block_group->key.offset -
9397 btrfs_block_group_used(&block_group->item)) *
9400 spin_unlock(&block_group->lock);
9402 spin_unlock(&sinfo->lock);
9407 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9409 struct btrfs_space_info *sinfo = cache->space_info;
9414 spin_lock(&sinfo->lock);
9415 spin_lock(&cache->lock);
9417 num_bytes = cache->key.offset - cache->reserved -
9418 cache->pinned - cache->bytes_super -
9419 btrfs_block_group_used(&cache->item);
9420 sinfo->bytes_readonly -= num_bytes;
9421 list_del_init(&cache->ro_list);
9423 spin_unlock(&cache->lock);
9424 spin_unlock(&sinfo->lock);
9428 * checks to see if its even possible to relocate this block group.
9430 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9431 * ok to go ahead and try.
9433 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9435 struct btrfs_root *root = fs_info->extent_root;
9436 struct btrfs_block_group_cache *block_group;
9437 struct btrfs_space_info *space_info;
9438 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9439 struct btrfs_device *device;
9440 struct btrfs_trans_handle *trans;
9450 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9452 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9454 /* odd, couldn't find the block group, leave it alone */
9458 "can't find block group for bytenr %llu",
9463 min_free = btrfs_block_group_used(&block_group->item);
9465 /* no bytes used, we're good */
9469 space_info = block_group->space_info;
9470 spin_lock(&space_info->lock);
9472 full = space_info->full;
9475 * if this is the last block group we have in this space, we can't
9476 * relocate it unless we're able to allocate a new chunk below.
9478 * Otherwise, we need to make sure we have room in the space to handle
9479 * all of the extents from this block group. If we can, we're good
9481 if ((space_info->total_bytes != block_group->key.offset) &&
9482 (btrfs_space_info_used(space_info, false) + min_free <
9483 space_info->total_bytes)) {
9484 spin_unlock(&space_info->lock);
9487 spin_unlock(&space_info->lock);
9490 * ok we don't have enough space, but maybe we have free space on our
9491 * devices to allocate new chunks for relocation, so loop through our
9492 * alloc devices and guess if we have enough space. if this block
9493 * group is going to be restriped, run checks against the target
9494 * profile instead of the current one.
9506 target = get_restripe_target(fs_info, block_group->flags);
9508 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9511 * this is just a balance, so if we were marked as full
9512 * we know there is no space for a new chunk
9517 "no space to alloc new chunk for block group %llu",
9518 block_group->key.objectid);
9522 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9525 if (index == BTRFS_RAID_RAID10) {
9529 } else if (index == BTRFS_RAID_RAID1) {
9531 } else if (index == BTRFS_RAID_DUP) {
9534 } else if (index == BTRFS_RAID_RAID0) {
9535 dev_min = fs_devices->rw_devices;
9536 min_free = div64_u64(min_free, dev_min);
9539 /* We need to do this so that we can look at pending chunks */
9540 trans = btrfs_join_transaction(root);
9541 if (IS_ERR(trans)) {
9542 ret = PTR_ERR(trans);
9546 mutex_lock(&fs_info->chunk_mutex);
9547 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9551 * check to make sure we can actually find a chunk with enough
9552 * space to fit our block group in.
9554 if (device->total_bytes > device->bytes_used + min_free &&
9555 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9556 ret = find_free_dev_extent(trans, device, min_free,
9561 if (dev_nr >= dev_min)
9567 if (debug && ret == -1)
9569 "no space to allocate a new chunk for block group %llu",
9570 block_group->key.objectid);
9571 mutex_unlock(&fs_info->chunk_mutex);
9572 btrfs_end_transaction(trans);
9574 btrfs_put_block_group(block_group);
9578 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9579 struct btrfs_path *path,
9580 struct btrfs_key *key)
9582 struct btrfs_root *root = fs_info->extent_root;
9584 struct btrfs_key found_key;
9585 struct extent_buffer *leaf;
9586 struct btrfs_block_group_item bg;
9590 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9595 slot = path->slots[0];
9596 leaf = path->nodes[0];
9597 if (slot >= btrfs_header_nritems(leaf)) {
9598 ret = btrfs_next_leaf(root, path);
9605 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9607 if (found_key.objectid >= key->objectid &&
9608 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9609 struct extent_map_tree *em_tree;
9610 struct extent_map *em;
9612 em_tree = &root->fs_info->mapping_tree.map_tree;
9613 read_lock(&em_tree->lock);
9614 em = lookup_extent_mapping(em_tree, found_key.objectid,
9616 read_unlock(&em_tree->lock);
9619 "logical %llu len %llu found bg but no related chunk",
9620 found_key.objectid, found_key.offset);
9622 } else if (em->start != found_key.objectid ||
9623 em->len != found_key.offset) {
9625 "block group %llu len %llu mismatch with chunk %llu len %llu",
9626 found_key.objectid, found_key.offset,
9627 em->start, em->len);
9630 read_extent_buffer(leaf, &bg,
9631 btrfs_item_ptr_offset(leaf, slot),
9633 flags = btrfs_block_group_flags(&bg) &
9634 BTRFS_BLOCK_GROUP_TYPE_MASK;
9636 if (flags != (em->map_lookup->type &
9637 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9639 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9641 found_key.offset, flags,
9642 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9643 em->map_lookup->type));
9649 free_extent_map(em);
9658 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9660 struct btrfs_block_group_cache *block_group;
9664 struct inode *inode;
9666 block_group = btrfs_lookup_first_block_group(info, last);
9667 while (block_group) {
9668 wait_block_group_cache_done(block_group);
9669 spin_lock(&block_group->lock);
9670 if (block_group->iref)
9672 spin_unlock(&block_group->lock);
9673 block_group = next_block_group(info, block_group);
9682 inode = block_group->inode;
9683 block_group->iref = 0;
9684 block_group->inode = NULL;
9685 spin_unlock(&block_group->lock);
9686 ASSERT(block_group->io_ctl.inode == NULL);
9688 last = block_group->key.objectid + block_group->key.offset;
9689 btrfs_put_block_group(block_group);
9694 * Must be called only after stopping all workers, since we could have block
9695 * group caching kthreads running, and therefore they could race with us if we
9696 * freed the block groups before stopping them.
9698 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9700 struct btrfs_block_group_cache *block_group;
9701 struct btrfs_space_info *space_info;
9702 struct btrfs_caching_control *caching_ctl;
9705 down_write(&info->commit_root_sem);
9706 while (!list_empty(&info->caching_block_groups)) {
9707 caching_ctl = list_entry(info->caching_block_groups.next,
9708 struct btrfs_caching_control, list);
9709 list_del(&caching_ctl->list);
9710 put_caching_control(caching_ctl);
9712 up_write(&info->commit_root_sem);
9714 spin_lock(&info->unused_bgs_lock);
9715 while (!list_empty(&info->unused_bgs)) {
9716 block_group = list_first_entry(&info->unused_bgs,
9717 struct btrfs_block_group_cache,
9719 list_del_init(&block_group->bg_list);
9720 btrfs_put_block_group(block_group);
9722 spin_unlock(&info->unused_bgs_lock);
9724 spin_lock(&info->block_group_cache_lock);
9725 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9726 block_group = rb_entry(n, struct btrfs_block_group_cache,
9728 rb_erase(&block_group->cache_node,
9729 &info->block_group_cache_tree);
9730 RB_CLEAR_NODE(&block_group->cache_node);
9731 spin_unlock(&info->block_group_cache_lock);
9733 down_write(&block_group->space_info->groups_sem);
9734 list_del(&block_group->list);
9735 up_write(&block_group->space_info->groups_sem);
9738 * We haven't cached this block group, which means we could
9739 * possibly have excluded extents on this block group.
9741 if (block_group->cached == BTRFS_CACHE_NO ||
9742 block_group->cached == BTRFS_CACHE_ERROR)
9743 free_excluded_extents(block_group);
9745 btrfs_remove_free_space_cache(block_group);
9746 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9747 ASSERT(list_empty(&block_group->dirty_list));
9748 ASSERT(list_empty(&block_group->io_list));
9749 ASSERT(list_empty(&block_group->bg_list));
9750 ASSERT(atomic_read(&block_group->count) == 1);
9751 btrfs_put_block_group(block_group);
9753 spin_lock(&info->block_group_cache_lock);
9755 spin_unlock(&info->block_group_cache_lock);
9757 /* now that all the block groups are freed, go through and
9758 * free all the space_info structs. This is only called during
9759 * the final stages of unmount, and so we know nobody is
9760 * using them. We call synchronize_rcu() once before we start,
9761 * just to be on the safe side.
9765 release_global_block_rsv(info);
9767 while (!list_empty(&info->space_info)) {
9770 space_info = list_entry(info->space_info.next,
9771 struct btrfs_space_info,
9775 * Do not hide this behind enospc_debug, this is actually
9776 * important and indicates a real bug if this happens.
9778 if (WARN_ON(space_info->bytes_pinned > 0 ||
9779 space_info->bytes_reserved > 0 ||
9780 space_info->bytes_may_use > 0))
9781 dump_space_info(info, space_info, 0, 0);
9782 list_del(&space_info->list);
9783 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9784 struct kobject *kobj;
9785 kobj = space_info->block_group_kobjs[i];
9786 space_info->block_group_kobjs[i] = NULL;
9792 kobject_del(&space_info->kobj);
9793 kobject_put(&space_info->kobj);
9798 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9799 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9801 struct btrfs_space_info *space_info;
9802 struct raid_kobject *rkobj;
9807 spin_lock(&fs_info->pending_raid_kobjs_lock);
9808 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9809 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9811 list_for_each_entry(rkobj, &list, list) {
9812 space_info = __find_space_info(fs_info, rkobj->flags);
9813 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9815 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9816 "%s", get_raid_name(index));
9818 kobject_put(&rkobj->kobj);
9824 "failed to add kobject for block cache, ignoring");
9827 static void link_block_group(struct btrfs_block_group_cache *cache)
9829 struct btrfs_space_info *space_info = cache->space_info;
9830 struct btrfs_fs_info *fs_info = cache->fs_info;
9831 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9834 down_write(&space_info->groups_sem);
9835 if (list_empty(&space_info->block_groups[index]))
9837 list_add_tail(&cache->list, &space_info->block_groups[index]);
9838 up_write(&space_info->groups_sem);
9841 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9843 btrfs_warn(cache->fs_info,
9844 "couldn't alloc memory for raid level kobject");
9847 rkobj->flags = cache->flags;
9848 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9850 spin_lock(&fs_info->pending_raid_kobjs_lock);
9851 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9852 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9853 space_info->block_group_kobjs[index] = &rkobj->kobj;
9857 static struct btrfs_block_group_cache *
9858 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9859 u64 start, u64 size)
9861 struct btrfs_block_group_cache *cache;
9863 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9867 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9869 if (!cache->free_space_ctl) {
9874 cache->key.objectid = start;
9875 cache->key.offset = size;
9876 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9878 cache->fs_info = fs_info;
9879 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9880 set_free_space_tree_thresholds(cache);
9882 atomic_set(&cache->count, 1);
9883 spin_lock_init(&cache->lock);
9884 init_rwsem(&cache->data_rwsem);
9885 INIT_LIST_HEAD(&cache->list);
9886 INIT_LIST_HEAD(&cache->cluster_list);
9887 INIT_LIST_HEAD(&cache->bg_list);
9888 INIT_LIST_HEAD(&cache->ro_list);
9889 INIT_LIST_HEAD(&cache->dirty_list);
9890 INIT_LIST_HEAD(&cache->io_list);
9891 btrfs_init_free_space_ctl(cache);
9892 atomic_set(&cache->trimming, 0);
9893 mutex_init(&cache->free_space_lock);
9894 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9901 * Iterate all chunks and verify that each of them has the corresponding block
9904 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
9906 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
9907 struct extent_map *em;
9908 struct btrfs_block_group_cache *bg;
9913 read_lock(&map_tree->map_tree.lock);
9915 * lookup_extent_mapping will return the first extent map
9916 * intersecting the range, so setting @len to 1 is enough to
9917 * get the first chunk.
9919 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
9920 read_unlock(&map_tree->map_tree.lock);
9924 bg = btrfs_lookup_block_group(fs_info, em->start);
9927 "chunk start=%llu len=%llu doesn't have corresponding block group",
9928 em->start, em->len);
9930 free_extent_map(em);
9933 if (bg->key.objectid != em->start ||
9934 bg->key.offset != em->len ||
9935 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
9936 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9938 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
9940 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
9941 bg->key.objectid, bg->key.offset,
9942 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
9944 free_extent_map(em);
9945 btrfs_put_block_group(bg);
9948 start = em->start + em->len;
9949 free_extent_map(em);
9950 btrfs_put_block_group(bg);
9955 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9957 struct btrfs_path *path;
9959 struct btrfs_block_group_cache *cache;
9960 struct btrfs_space_info *space_info;
9961 struct btrfs_key key;
9962 struct btrfs_key found_key;
9963 struct extent_buffer *leaf;
9969 feature = btrfs_super_incompat_flags(info->super_copy);
9970 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9974 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9975 path = btrfs_alloc_path();
9978 path->reada = READA_FORWARD;
9980 cache_gen = btrfs_super_cache_generation(info->super_copy);
9981 if (btrfs_test_opt(info, SPACE_CACHE) &&
9982 btrfs_super_generation(info->super_copy) != cache_gen)
9984 if (btrfs_test_opt(info, CLEAR_CACHE))
9988 ret = find_first_block_group(info, path, &key);
9994 leaf = path->nodes[0];
9995 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9997 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10006 * When we mount with old space cache, we need to
10007 * set BTRFS_DC_CLEAR and set dirty flag.
10009 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10010 * truncate the old free space cache inode and
10012 * b) Setting 'dirty flag' makes sure that we flush
10013 * the new space cache info onto disk.
10015 if (btrfs_test_opt(info, SPACE_CACHE))
10016 cache->disk_cache_state = BTRFS_DC_CLEAR;
10019 read_extent_buffer(leaf, &cache->item,
10020 btrfs_item_ptr_offset(leaf, path->slots[0]),
10021 sizeof(cache->item));
10022 cache->flags = btrfs_block_group_flags(&cache->item);
10024 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10025 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10027 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10028 cache->key.objectid);
10033 key.objectid = found_key.objectid + found_key.offset;
10034 btrfs_release_path(path);
10037 * We need to exclude the super stripes now so that the space
10038 * info has super bytes accounted for, otherwise we'll think
10039 * we have more space than we actually do.
10041 ret = exclude_super_stripes(cache);
10044 * We may have excluded something, so call this just in
10047 free_excluded_extents(cache);
10048 btrfs_put_block_group(cache);
10053 * check for two cases, either we are full, and therefore
10054 * don't need to bother with the caching work since we won't
10055 * find any space, or we are empty, and we can just add all
10056 * the space in and be done with it. This saves us _alot_ of
10057 * time, particularly in the full case.
10059 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10060 cache->last_byte_to_unpin = (u64)-1;
10061 cache->cached = BTRFS_CACHE_FINISHED;
10062 free_excluded_extents(cache);
10063 } else if (btrfs_block_group_used(&cache->item) == 0) {
10064 cache->last_byte_to_unpin = (u64)-1;
10065 cache->cached = BTRFS_CACHE_FINISHED;
10066 add_new_free_space(cache, found_key.objectid,
10067 found_key.objectid +
10069 free_excluded_extents(cache);
10072 ret = btrfs_add_block_group_cache(info, cache);
10074 btrfs_remove_free_space_cache(cache);
10075 btrfs_put_block_group(cache);
10079 trace_btrfs_add_block_group(info, cache, 0);
10080 update_space_info(info, cache->flags, found_key.offset,
10081 btrfs_block_group_used(&cache->item),
10082 cache->bytes_super, &space_info);
10084 cache->space_info = space_info;
10086 link_block_group(cache);
10088 set_avail_alloc_bits(info, cache->flags);
10089 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10090 inc_block_group_ro(cache, 1);
10091 } else if (btrfs_block_group_used(&cache->item) == 0) {
10092 ASSERT(list_empty(&cache->bg_list));
10093 btrfs_mark_bg_unused(cache);
10097 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10098 if (!(get_alloc_profile(info, space_info->flags) &
10099 (BTRFS_BLOCK_GROUP_RAID10 |
10100 BTRFS_BLOCK_GROUP_RAID1 |
10101 BTRFS_BLOCK_GROUP_RAID5 |
10102 BTRFS_BLOCK_GROUP_RAID6 |
10103 BTRFS_BLOCK_GROUP_DUP)))
10106 * avoid allocating from un-mirrored block group if there are
10107 * mirrored block groups.
10109 list_for_each_entry(cache,
10110 &space_info->block_groups[BTRFS_RAID_RAID0],
10112 inc_block_group_ro(cache, 1);
10113 list_for_each_entry(cache,
10114 &space_info->block_groups[BTRFS_RAID_SINGLE],
10116 inc_block_group_ro(cache, 1);
10119 btrfs_add_raid_kobjects(info);
10120 init_global_block_rsv(info);
10121 ret = check_chunk_block_group_mappings(info);
10123 btrfs_free_path(path);
10127 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10129 struct btrfs_fs_info *fs_info = trans->fs_info;
10130 struct btrfs_block_group_cache *block_group;
10131 struct btrfs_root *extent_root = fs_info->extent_root;
10132 struct btrfs_block_group_item item;
10133 struct btrfs_key key;
10135 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10137 trans->can_flush_pending_bgs = false;
10138 while (!list_empty(&trans->new_bgs)) {
10139 block_group = list_first_entry(&trans->new_bgs,
10140 struct btrfs_block_group_cache,
10145 spin_lock(&block_group->lock);
10146 memcpy(&item, &block_group->item, sizeof(item));
10147 memcpy(&key, &block_group->key, sizeof(key));
10148 spin_unlock(&block_group->lock);
10150 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10153 btrfs_abort_transaction(trans, ret);
10154 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10156 btrfs_abort_transaction(trans, ret);
10157 add_block_group_free_space(trans, block_group);
10158 /* already aborted the transaction if it failed. */
10160 list_del_init(&block_group->bg_list);
10162 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10165 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10166 u64 type, u64 chunk_offset, u64 size)
10168 struct btrfs_fs_info *fs_info = trans->fs_info;
10169 struct btrfs_block_group_cache *cache;
10172 btrfs_set_log_full_commit(fs_info, trans);
10174 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10178 btrfs_set_block_group_used(&cache->item, bytes_used);
10179 btrfs_set_block_group_chunk_objectid(&cache->item,
10180 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10181 btrfs_set_block_group_flags(&cache->item, type);
10183 cache->flags = type;
10184 cache->last_byte_to_unpin = (u64)-1;
10185 cache->cached = BTRFS_CACHE_FINISHED;
10186 cache->needs_free_space = 1;
10187 ret = exclude_super_stripes(cache);
10190 * We may have excluded something, so call this just in
10193 free_excluded_extents(cache);
10194 btrfs_put_block_group(cache);
10198 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10200 free_excluded_extents(cache);
10202 #ifdef CONFIG_BTRFS_DEBUG
10203 if (btrfs_should_fragment_free_space(cache)) {
10204 u64 new_bytes_used = size - bytes_used;
10206 bytes_used += new_bytes_used >> 1;
10207 fragment_free_space(cache);
10211 * Ensure the corresponding space_info object is created and
10212 * assigned to our block group. We want our bg to be added to the rbtree
10213 * with its ->space_info set.
10215 cache->space_info = __find_space_info(fs_info, cache->flags);
10216 ASSERT(cache->space_info);
10218 ret = btrfs_add_block_group_cache(fs_info, cache);
10220 btrfs_remove_free_space_cache(cache);
10221 btrfs_put_block_group(cache);
10226 * Now that our block group has its ->space_info set and is inserted in
10227 * the rbtree, update the space info's counters.
10229 trace_btrfs_add_block_group(fs_info, cache, 1);
10230 update_space_info(fs_info, cache->flags, size, bytes_used,
10231 cache->bytes_super, &cache->space_info);
10232 update_global_block_rsv(fs_info);
10234 link_block_group(cache);
10236 list_add_tail(&cache->bg_list, &trans->new_bgs);
10238 set_avail_alloc_bits(fs_info, type);
10242 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10244 u64 extra_flags = chunk_to_extended(flags) &
10245 BTRFS_EXTENDED_PROFILE_MASK;
10247 write_seqlock(&fs_info->profiles_lock);
10248 if (flags & BTRFS_BLOCK_GROUP_DATA)
10249 fs_info->avail_data_alloc_bits &= ~extra_flags;
10250 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10251 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10252 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10253 fs_info->avail_system_alloc_bits &= ~extra_flags;
10254 write_sequnlock(&fs_info->profiles_lock);
10257 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10258 u64 group_start, struct extent_map *em)
10260 struct btrfs_fs_info *fs_info = trans->fs_info;
10261 struct btrfs_root *root = fs_info->extent_root;
10262 struct btrfs_path *path;
10263 struct btrfs_block_group_cache *block_group;
10264 struct btrfs_free_cluster *cluster;
10265 struct btrfs_root *tree_root = fs_info->tree_root;
10266 struct btrfs_key key;
10267 struct inode *inode;
10268 struct kobject *kobj = NULL;
10272 struct btrfs_caching_control *caching_ctl = NULL;
10275 block_group = btrfs_lookup_block_group(fs_info, group_start);
10276 BUG_ON(!block_group);
10277 BUG_ON(!block_group->ro);
10279 trace_btrfs_remove_block_group(block_group);
10281 * Free the reserved super bytes from this block group before
10284 free_excluded_extents(block_group);
10285 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10286 block_group->key.offset);
10288 memcpy(&key, &block_group->key, sizeof(key));
10289 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10290 factor = btrfs_bg_type_to_factor(block_group->flags);
10292 /* make sure this block group isn't part of an allocation cluster */
10293 cluster = &fs_info->data_alloc_cluster;
10294 spin_lock(&cluster->refill_lock);
10295 btrfs_return_cluster_to_free_space(block_group, cluster);
10296 spin_unlock(&cluster->refill_lock);
10299 * make sure this block group isn't part of a metadata
10300 * allocation cluster
10302 cluster = &fs_info->meta_alloc_cluster;
10303 spin_lock(&cluster->refill_lock);
10304 btrfs_return_cluster_to_free_space(block_group, cluster);
10305 spin_unlock(&cluster->refill_lock);
10307 path = btrfs_alloc_path();
10314 * get the inode first so any iput calls done for the io_list
10315 * aren't the final iput (no unlinks allowed now)
10317 inode = lookup_free_space_inode(fs_info, block_group, path);
10319 mutex_lock(&trans->transaction->cache_write_mutex);
10321 * make sure our free spache cache IO is done before remove the
10324 spin_lock(&trans->transaction->dirty_bgs_lock);
10325 if (!list_empty(&block_group->io_list)) {
10326 list_del_init(&block_group->io_list);
10328 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10330 spin_unlock(&trans->transaction->dirty_bgs_lock);
10331 btrfs_wait_cache_io(trans, block_group, path);
10332 btrfs_put_block_group(block_group);
10333 spin_lock(&trans->transaction->dirty_bgs_lock);
10336 if (!list_empty(&block_group->dirty_list)) {
10337 list_del_init(&block_group->dirty_list);
10338 btrfs_put_block_group(block_group);
10340 spin_unlock(&trans->transaction->dirty_bgs_lock);
10341 mutex_unlock(&trans->transaction->cache_write_mutex);
10343 if (!IS_ERR(inode)) {
10344 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10346 btrfs_add_delayed_iput(inode);
10349 clear_nlink(inode);
10350 /* One for the block groups ref */
10351 spin_lock(&block_group->lock);
10352 if (block_group->iref) {
10353 block_group->iref = 0;
10354 block_group->inode = NULL;
10355 spin_unlock(&block_group->lock);
10358 spin_unlock(&block_group->lock);
10360 /* One for our lookup ref */
10361 btrfs_add_delayed_iput(inode);
10364 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10365 key.offset = block_group->key.objectid;
10368 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10372 btrfs_release_path(path);
10374 ret = btrfs_del_item(trans, tree_root, path);
10377 btrfs_release_path(path);
10380 spin_lock(&fs_info->block_group_cache_lock);
10381 rb_erase(&block_group->cache_node,
10382 &fs_info->block_group_cache_tree);
10383 RB_CLEAR_NODE(&block_group->cache_node);
10385 if (fs_info->first_logical_byte == block_group->key.objectid)
10386 fs_info->first_logical_byte = (u64)-1;
10387 spin_unlock(&fs_info->block_group_cache_lock);
10389 down_write(&block_group->space_info->groups_sem);
10391 * we must use list_del_init so people can check to see if they
10392 * are still on the list after taking the semaphore
10394 list_del_init(&block_group->list);
10395 if (list_empty(&block_group->space_info->block_groups[index])) {
10396 kobj = block_group->space_info->block_group_kobjs[index];
10397 block_group->space_info->block_group_kobjs[index] = NULL;
10398 clear_avail_alloc_bits(fs_info, block_group->flags);
10400 up_write(&block_group->space_info->groups_sem);
10406 if (block_group->has_caching_ctl)
10407 caching_ctl = get_caching_control(block_group);
10408 if (block_group->cached == BTRFS_CACHE_STARTED)
10409 wait_block_group_cache_done(block_group);
10410 if (block_group->has_caching_ctl) {
10411 down_write(&fs_info->commit_root_sem);
10412 if (!caching_ctl) {
10413 struct btrfs_caching_control *ctl;
10415 list_for_each_entry(ctl,
10416 &fs_info->caching_block_groups, list)
10417 if (ctl->block_group == block_group) {
10419 refcount_inc(&caching_ctl->count);
10424 list_del_init(&caching_ctl->list);
10425 up_write(&fs_info->commit_root_sem);
10427 /* Once for the caching bgs list and once for us. */
10428 put_caching_control(caching_ctl);
10429 put_caching_control(caching_ctl);
10433 spin_lock(&trans->transaction->dirty_bgs_lock);
10434 if (!list_empty(&block_group->dirty_list)) {
10437 if (!list_empty(&block_group->io_list)) {
10440 spin_unlock(&trans->transaction->dirty_bgs_lock);
10441 btrfs_remove_free_space_cache(block_group);
10443 spin_lock(&block_group->space_info->lock);
10444 list_del_init(&block_group->ro_list);
10446 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10447 WARN_ON(block_group->space_info->total_bytes
10448 < block_group->key.offset);
10449 WARN_ON(block_group->space_info->bytes_readonly
10450 < block_group->key.offset);
10451 WARN_ON(block_group->space_info->disk_total
10452 < block_group->key.offset * factor);
10454 block_group->space_info->total_bytes -= block_group->key.offset;
10455 block_group->space_info->bytes_readonly -= block_group->key.offset;
10456 block_group->space_info->disk_total -= block_group->key.offset * factor;
10458 spin_unlock(&block_group->space_info->lock);
10460 memcpy(&key, &block_group->key, sizeof(key));
10462 mutex_lock(&fs_info->chunk_mutex);
10463 if (!list_empty(&em->list)) {
10464 /* We're in the transaction->pending_chunks list. */
10465 free_extent_map(em);
10467 spin_lock(&block_group->lock);
10468 block_group->removed = 1;
10470 * At this point trimming can't start on this block group, because we
10471 * removed the block group from the tree fs_info->block_group_cache_tree
10472 * so no one can't find it anymore and even if someone already got this
10473 * block group before we removed it from the rbtree, they have already
10474 * incremented block_group->trimming - if they didn't, they won't find
10475 * any free space entries because we already removed them all when we
10476 * called btrfs_remove_free_space_cache().
10478 * And we must not remove the extent map from the fs_info->mapping_tree
10479 * to prevent the same logical address range and physical device space
10480 * ranges from being reused for a new block group. This is because our
10481 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10482 * completely transactionless, so while it is trimming a range the
10483 * currently running transaction might finish and a new one start,
10484 * allowing for new block groups to be created that can reuse the same
10485 * physical device locations unless we take this special care.
10487 * There may also be an implicit trim operation if the file system
10488 * is mounted with -odiscard. The same protections must remain
10489 * in place until the extents have been discarded completely when
10490 * the transaction commit has completed.
10492 remove_em = (atomic_read(&block_group->trimming) == 0);
10494 * Make sure a trimmer task always sees the em in the pinned_chunks list
10495 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10496 * before checking block_group->removed).
10500 * Our em might be in trans->transaction->pending_chunks which
10501 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10502 * and so is the fs_info->pinned_chunks list.
10504 * So at this point we must be holding the chunk_mutex to avoid
10505 * any races with chunk allocation (more specifically at
10506 * volumes.c:contains_pending_extent()), to ensure it always
10507 * sees the em, either in the pending_chunks list or in the
10508 * pinned_chunks list.
10510 list_move_tail(&em->list, &fs_info->pinned_chunks);
10512 spin_unlock(&block_group->lock);
10515 struct extent_map_tree *em_tree;
10517 em_tree = &fs_info->mapping_tree.map_tree;
10518 write_lock(&em_tree->lock);
10520 * The em might be in the pending_chunks list, so make sure the
10521 * chunk mutex is locked, since remove_extent_mapping() will
10522 * delete us from that list.
10524 remove_extent_mapping(em_tree, em);
10525 write_unlock(&em_tree->lock);
10526 /* once for the tree */
10527 free_extent_map(em);
10530 mutex_unlock(&fs_info->chunk_mutex);
10532 ret = remove_block_group_free_space(trans, block_group);
10536 btrfs_put_block_group(block_group);
10537 btrfs_put_block_group(block_group);
10539 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10545 ret = btrfs_del_item(trans, root, path);
10547 btrfs_free_path(path);
10551 struct btrfs_trans_handle *
10552 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10553 const u64 chunk_offset)
10555 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10556 struct extent_map *em;
10557 struct map_lookup *map;
10558 unsigned int num_items;
10560 read_lock(&em_tree->lock);
10561 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10562 read_unlock(&em_tree->lock);
10563 ASSERT(em && em->start == chunk_offset);
10566 * We need to reserve 3 + N units from the metadata space info in order
10567 * to remove a block group (done at btrfs_remove_chunk() and at
10568 * btrfs_remove_block_group()), which are used for:
10570 * 1 unit for adding the free space inode's orphan (located in the tree
10572 * 1 unit for deleting the block group item (located in the extent
10574 * 1 unit for deleting the free space item (located in tree of tree
10576 * N units for deleting N device extent items corresponding to each
10577 * stripe (located in the device tree).
10579 * In order to remove a block group we also need to reserve units in the
10580 * system space info in order to update the chunk tree (update one or
10581 * more device items and remove one chunk item), but this is done at
10582 * btrfs_remove_chunk() through a call to check_system_chunk().
10584 map = em->map_lookup;
10585 num_items = 3 + map->num_stripes;
10586 free_extent_map(em);
10588 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10593 * Process the unused_bgs list and remove any that don't have any allocated
10594 * space inside of them.
10596 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10598 struct btrfs_block_group_cache *block_group;
10599 struct btrfs_space_info *space_info;
10600 struct btrfs_trans_handle *trans;
10603 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10606 spin_lock(&fs_info->unused_bgs_lock);
10607 while (!list_empty(&fs_info->unused_bgs)) {
10611 block_group = list_first_entry(&fs_info->unused_bgs,
10612 struct btrfs_block_group_cache,
10614 list_del_init(&block_group->bg_list);
10616 space_info = block_group->space_info;
10618 if (ret || btrfs_mixed_space_info(space_info)) {
10619 btrfs_put_block_group(block_group);
10622 spin_unlock(&fs_info->unused_bgs_lock);
10624 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10626 /* Don't want to race with allocators so take the groups_sem */
10627 down_write(&space_info->groups_sem);
10628 spin_lock(&block_group->lock);
10629 if (block_group->reserved || block_group->pinned ||
10630 btrfs_block_group_used(&block_group->item) ||
10632 list_is_singular(&block_group->list)) {
10634 * We want to bail if we made new allocations or have
10635 * outstanding allocations in this block group. We do
10636 * the ro check in case balance is currently acting on
10637 * this block group.
10639 trace_btrfs_skip_unused_block_group(block_group);
10640 spin_unlock(&block_group->lock);
10641 up_write(&space_info->groups_sem);
10644 spin_unlock(&block_group->lock);
10646 /* We don't want to force the issue, only flip if it's ok. */
10647 ret = inc_block_group_ro(block_group, 0);
10648 up_write(&space_info->groups_sem);
10655 * Want to do this before we do anything else so we can recover
10656 * properly if we fail to join the transaction.
10658 trans = btrfs_start_trans_remove_block_group(fs_info,
10659 block_group->key.objectid);
10660 if (IS_ERR(trans)) {
10661 btrfs_dec_block_group_ro(block_group);
10662 ret = PTR_ERR(trans);
10667 * We could have pending pinned extents for this block group,
10668 * just delete them, we don't care about them anymore.
10670 start = block_group->key.objectid;
10671 end = start + block_group->key.offset - 1;
10673 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10674 * btrfs_finish_extent_commit(). If we are at transaction N,
10675 * another task might be running finish_extent_commit() for the
10676 * previous transaction N - 1, and have seen a range belonging
10677 * to the block group in freed_extents[] before we were able to
10678 * clear the whole block group range from freed_extents[]. This
10679 * means that task can lookup for the block group after we
10680 * unpinned it from freed_extents[] and removed it, leading to
10681 * a BUG_ON() at btrfs_unpin_extent_range().
10683 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10684 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10687 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10688 btrfs_dec_block_group_ro(block_group);
10691 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10694 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10695 btrfs_dec_block_group_ro(block_group);
10698 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10700 /* Reset pinned so btrfs_put_block_group doesn't complain */
10701 spin_lock(&space_info->lock);
10702 spin_lock(&block_group->lock);
10704 space_info->bytes_pinned -= block_group->pinned;
10705 space_info->bytes_readonly += block_group->pinned;
10706 percpu_counter_add_batch(&space_info->total_bytes_pinned,
10707 -block_group->pinned,
10708 BTRFS_TOTAL_BYTES_PINNED_BATCH);
10709 block_group->pinned = 0;
10711 spin_unlock(&block_group->lock);
10712 spin_unlock(&space_info->lock);
10714 /* DISCARD can flip during remount */
10715 trimming = btrfs_test_opt(fs_info, DISCARD);
10717 /* Implicit trim during transaction commit. */
10719 btrfs_get_block_group_trimming(block_group);
10722 * Btrfs_remove_chunk will abort the transaction if things go
10725 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
10729 btrfs_put_block_group_trimming(block_group);
10734 * If we're not mounted with -odiscard, we can just forget
10735 * about this block group. Otherwise we'll need to wait
10736 * until transaction commit to do the actual discard.
10739 spin_lock(&fs_info->unused_bgs_lock);
10741 * A concurrent scrub might have added us to the list
10742 * fs_info->unused_bgs, so use a list_move operation
10743 * to add the block group to the deleted_bgs list.
10745 list_move(&block_group->bg_list,
10746 &trans->transaction->deleted_bgs);
10747 spin_unlock(&fs_info->unused_bgs_lock);
10748 btrfs_get_block_group(block_group);
10751 btrfs_end_transaction(trans);
10753 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10754 btrfs_put_block_group(block_group);
10755 spin_lock(&fs_info->unused_bgs_lock);
10757 spin_unlock(&fs_info->unused_bgs_lock);
10760 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10762 struct btrfs_super_block *disk_super;
10768 disk_super = fs_info->super_copy;
10769 if (!btrfs_super_root(disk_super))
10772 features = btrfs_super_incompat_flags(disk_super);
10773 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10776 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10777 ret = create_space_info(fs_info, flags);
10782 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10783 ret = create_space_info(fs_info, flags);
10785 flags = BTRFS_BLOCK_GROUP_METADATA;
10786 ret = create_space_info(fs_info, flags);
10790 flags = BTRFS_BLOCK_GROUP_DATA;
10791 ret = create_space_info(fs_info, flags);
10797 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10798 u64 start, u64 end)
10800 return unpin_extent_range(fs_info, start, end, false);
10804 * It used to be that old block groups would be left around forever.
10805 * Iterating over them would be enough to trim unused space. Since we
10806 * now automatically remove them, we also need to iterate over unallocated
10809 * We don't want a transaction for this since the discard may take a
10810 * substantial amount of time. We don't require that a transaction be
10811 * running, but we do need to take a running transaction into account
10812 * to ensure that we're not discarding chunks that were released or
10813 * allocated in the current transaction.
10815 * Holding the chunks lock will prevent other threads from allocating
10816 * or releasing chunks, but it won't prevent a running transaction
10817 * from committing and releasing the memory that the pending chunks
10818 * list head uses. For that, we need to take a reference to the
10819 * transaction and hold the commit root sem. We only need to hold
10820 * it while performing the free space search since we have already
10821 * held back allocations.
10823 static int btrfs_trim_free_extents(struct btrfs_device *device,
10824 u64 minlen, u64 *trimmed)
10826 u64 start = 0, len = 0;
10831 /* Discard not supported = nothing to do. */
10832 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
10835 /* Not writeable = nothing to do. */
10836 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10839 /* No free space = nothing to do. */
10840 if (device->total_bytes <= device->bytes_used)
10846 struct btrfs_fs_info *fs_info = device->fs_info;
10847 struct btrfs_transaction *trans;
10850 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10854 ret = down_read_killable(&fs_info->commit_root_sem);
10856 mutex_unlock(&fs_info->chunk_mutex);
10860 spin_lock(&fs_info->trans_lock);
10861 trans = fs_info->running_transaction;
10863 refcount_inc(&trans->use_count);
10864 spin_unlock(&fs_info->trans_lock);
10867 up_read(&fs_info->commit_root_sem);
10869 ret = find_free_dev_extent_start(trans, device, minlen, start,
10872 up_read(&fs_info->commit_root_sem);
10873 btrfs_put_transaction(trans);
10877 mutex_unlock(&fs_info->chunk_mutex);
10878 if (ret == -ENOSPC)
10883 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10884 mutex_unlock(&fs_info->chunk_mutex);
10892 if (fatal_signal_pending(current)) {
10893 ret = -ERESTARTSYS;
10904 * Trim the whole filesystem by:
10905 * 1) trimming the free space in each block group
10906 * 2) trimming the unallocated space on each device
10908 * This will also continue trimming even if a block group or device encounters
10909 * an error. The return value will be the last error, or 0 if nothing bad
10912 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10914 struct btrfs_block_group_cache *cache = NULL;
10915 struct btrfs_device *device;
10916 struct list_head *devices;
10922 u64 dev_failed = 0;
10927 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10928 for (; cache; cache = next_block_group(fs_info, cache)) {
10929 if (cache->key.objectid >= (range->start + range->len)) {
10930 btrfs_put_block_group(cache);
10934 start = max(range->start, cache->key.objectid);
10935 end = min(range->start + range->len,
10936 cache->key.objectid + cache->key.offset);
10938 if (end - start >= range->minlen) {
10939 if (!block_group_cache_done(cache)) {
10940 ret = cache_block_group(cache, 0);
10946 ret = wait_block_group_cache_done(cache);
10953 ret = btrfs_trim_block_group(cache,
10959 trimmed += group_trimmed;
10969 btrfs_warn(fs_info,
10970 "failed to trim %llu block group(s), last error %d",
10971 bg_failed, bg_ret);
10972 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10973 devices = &fs_info->fs_devices->devices;
10974 list_for_each_entry(device, devices, dev_list) {
10975 ret = btrfs_trim_free_extents(device, range->minlen,
10983 trimmed += group_trimmed;
10985 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10988 btrfs_warn(fs_info,
10989 "failed to trim %llu device(s), last error %d",
10990 dev_failed, dev_ret);
10991 range->len = trimmed;
10998 * btrfs_{start,end}_write_no_snapshotting() are similar to
10999 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11000 * data into the page cache through nocow before the subvolume is snapshoted,
11001 * but flush the data into disk after the snapshot creation, or to prevent
11002 * operations while snapshotting is ongoing and that cause the snapshot to be
11003 * inconsistent (writes followed by expanding truncates for example).
11005 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11007 percpu_counter_dec(&root->subv_writers->counter);
11008 cond_wake_up(&root->subv_writers->wait);
11011 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11013 if (atomic_read(&root->will_be_snapshotted))
11016 percpu_counter_inc(&root->subv_writers->counter);
11018 * Make sure counter is updated before we check for snapshot creation.
11021 if (atomic_read(&root->will_be_snapshotted)) {
11022 btrfs_end_write_no_snapshotting(root);
11028 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11033 ret = btrfs_start_write_no_snapshotting(root);
11036 wait_var_event(&root->will_be_snapshotted,
11037 !atomic_read(&root->will_be_snapshotted));
11041 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11043 struct btrfs_fs_info *fs_info = bg->fs_info;
11045 spin_lock(&fs_info->unused_bgs_lock);
11046 if (list_empty(&bg->bg_list)) {
11047 btrfs_get_block_group(bg);
11048 trace_btrfs_add_unused_block_group(bg);
11049 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11051 spin_unlock(&fs_info->unused_bgs_lock);