2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
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 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
343 u64 extent_start, extent_end, size, total_added = 0;
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
359 ret = btrfs_add_free_space(block_group, start,
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
378 static noinline void caching_thread(struct btrfs_work *work)
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < last) {
465 key.type = BTRFS_EXTENT_ITEM_KEY;
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
472 if (key.objectid < block_group->key.objectid) {
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
490 last = key.objectid + key.offset;
492 if (total_found > (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl->wait);
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
515 free_excluded_extents(extent_root, block_group);
517 mutex_unlock(&caching_ctl->mutex);
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
525 wake_up(&caching_ctl->wait);
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
612 wake_up(&caching_ctl->wait);
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.objectid = bytenr;
1537 key.type = BTRFS_EXTENT_ITEM_KEY;
1538 key.offset = num_bytes;
1539 btrfs_release_path(path);
1544 if (ret && !insert) {
1547 } else if (WARN_ON(ret)) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1673 unsigned long item_offset;
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1685 btrfs_extend_item(root, path, size);
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 btrfs_mark_buffer_dirty(leaf);
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1737 btrfs_release_path(path);
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1758 struct btrfs_delayed_extent_op *extent_op,
1761 struct extent_buffer *leaf;
1762 struct btrfs_extent_item *ei;
1763 struct btrfs_extent_data_ref *dref = NULL;
1764 struct btrfs_shared_data_ref *sref = NULL;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 refs = btrfs_extent_refs(leaf, ei);
1775 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776 refs += refs_to_mod;
1777 btrfs_set_extent_refs(leaf, ei, refs);
1779 __run_delayed_extent_op(extent_op, leaf, ei);
1781 type = btrfs_extent_inline_ref_type(leaf, iref);
1783 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785 refs = btrfs_extent_data_ref_count(leaf, dref);
1786 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788 refs = btrfs_shared_data_ref_count(leaf, sref);
1791 BUG_ON(refs_to_mod != -1);
1794 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795 refs += refs_to_mod;
1798 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(root, path, iref,
1835 refs_to_add, extent_op, NULL);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data, int *last_ref)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(root, path, iref,
1875 -refs_to_drop, NULL, last_ref);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1881 ret = btrfs_del_item(trans, root, path);
1886 static int btrfs_issue_discard(struct block_device *bdev,
1889 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1892 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893 u64 num_bytes, u64 *actual_bytes)
1896 u64 discarded_bytes = 0;
1897 struct btrfs_bio *bbio = NULL;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902 bytenr, &num_bytes, &bbio, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe *stripe = bbio->stripes;
1909 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910 if (!stripe->dev->can_discard)
1913 ret = btrfs_issue_discard(stripe->dev->bdev,
1917 discarded_bytes += stripe->length;
1918 else if (ret != -EOPNOTSUPP)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1928 btrfs_put_bbio(bbio);
1932 *actual_bytes = discarded_bytes;
1935 if (ret == -EOPNOTSUPP)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 u64 bytenr, u64 num_bytes, u64 parent,
1944 u64 root_objectid, u64 owner, u64 offset,
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 struct btrfs_delayed_ref_node *node,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1972 struct btrfs_delayed_extent_op *extent_op)
1974 struct btrfs_fs_info *fs_info = root->fs_info;
1975 struct btrfs_path *path;
1976 struct extent_buffer *leaf;
1977 struct btrfs_extent_item *item;
1978 struct btrfs_key key;
1979 u64 bytenr = node->bytenr;
1980 u64 num_bytes = node->num_bytes;
1983 int no_quota = node->no_quota;
1984 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1986 path = btrfs_alloc_path();
1990 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1994 path->leave_spinning = 1;
1995 /* this will setup the path even if it fails to insert the back ref */
1996 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1997 bytenr, num_bytes, parent,
1998 root_objectid, owner, offset,
1999 refs_to_add, extent_op);
2000 if ((ret < 0 && ret != -EAGAIN) || !ret)
2004 * Ok we had -EAGAIN which means we didn't have space to insert and
2005 * inline extent ref, so just update the reference count and add a
2008 leaf = path->nodes[0];
2009 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2010 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2011 refs = btrfs_extent_refs(leaf, item);
2013 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2014 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2016 __run_delayed_extent_op(extent_op, leaf, item);
2018 btrfs_mark_buffer_dirty(leaf);
2019 btrfs_release_path(path);
2022 path->leave_spinning = 1;
2023 /* now insert the actual backref */
2024 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2025 path, bytenr, parent, root_objectid,
2026 owner, offset, refs_to_add);
2028 btrfs_abort_transaction(trans, root, ret);
2030 btrfs_free_path(path);
2034 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2035 struct btrfs_root *root,
2036 struct btrfs_delayed_ref_node *node,
2037 struct btrfs_delayed_extent_op *extent_op,
2038 int insert_reserved)
2041 struct btrfs_delayed_data_ref *ref;
2042 struct btrfs_key ins;
2047 ins.objectid = node->bytenr;
2048 ins.offset = node->num_bytes;
2049 ins.type = BTRFS_EXTENT_ITEM_KEY;
2051 ref = btrfs_delayed_node_to_data_ref(node);
2052 trace_run_delayed_data_ref(node, ref, node->action);
2054 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2055 parent = ref->parent;
2056 ref_root = ref->root;
2058 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2060 flags |= extent_op->flags_to_set;
2061 ret = alloc_reserved_file_extent(trans, root,
2062 parent, ref_root, flags,
2063 ref->objectid, ref->offset,
2064 &ins, node->ref_mod);
2065 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2066 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2067 ref_root, ref->objectid,
2068 ref->offset, node->ref_mod,
2070 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2071 ret = __btrfs_free_extent(trans, root, node, parent,
2072 ref_root, ref->objectid,
2073 ref->offset, node->ref_mod,
2081 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2082 struct extent_buffer *leaf,
2083 struct btrfs_extent_item *ei)
2085 u64 flags = btrfs_extent_flags(leaf, ei);
2086 if (extent_op->update_flags) {
2087 flags |= extent_op->flags_to_set;
2088 btrfs_set_extent_flags(leaf, ei, flags);
2091 if (extent_op->update_key) {
2092 struct btrfs_tree_block_info *bi;
2093 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2094 bi = (struct btrfs_tree_block_info *)(ei + 1);
2095 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2099 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 struct btrfs_delayed_extent_op *extent_op)
2104 struct btrfs_key key;
2105 struct btrfs_path *path;
2106 struct btrfs_extent_item *ei;
2107 struct extent_buffer *leaf;
2111 int metadata = !extent_op->is_data;
2116 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2119 path = btrfs_alloc_path();
2123 key.objectid = node->bytenr;
2126 key.type = BTRFS_METADATA_ITEM_KEY;
2127 key.offset = extent_op->level;
2129 key.type = BTRFS_EXTENT_ITEM_KEY;
2130 key.offset = node->num_bytes;
2135 path->leave_spinning = 1;
2136 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2144 if (path->slots[0] > 0) {
2146 btrfs_item_key_to_cpu(path->nodes[0], &key,
2148 if (key.objectid == node->bytenr &&
2149 key.type == BTRFS_EXTENT_ITEM_KEY &&
2150 key.offset == node->num_bytes)
2154 btrfs_release_path(path);
2157 key.objectid = node->bytenr;
2158 key.offset = node->num_bytes;
2159 key.type = BTRFS_EXTENT_ITEM_KEY;
2168 leaf = path->nodes[0];
2169 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2170 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2171 if (item_size < sizeof(*ei)) {
2172 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2178 leaf = path->nodes[0];
2179 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2182 BUG_ON(item_size < sizeof(*ei));
2183 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2184 __run_delayed_extent_op(extent_op, leaf, ei);
2186 btrfs_mark_buffer_dirty(leaf);
2188 btrfs_free_path(path);
2192 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2193 struct btrfs_root *root,
2194 struct btrfs_delayed_ref_node *node,
2195 struct btrfs_delayed_extent_op *extent_op,
2196 int insert_reserved)
2199 struct btrfs_delayed_tree_ref *ref;
2200 struct btrfs_key ins;
2203 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2206 ref = btrfs_delayed_node_to_tree_ref(node);
2207 trace_run_delayed_tree_ref(node, ref, node->action);
2209 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2210 parent = ref->parent;
2211 ref_root = ref->root;
2213 ins.objectid = node->bytenr;
2214 if (skinny_metadata) {
2215 ins.offset = ref->level;
2216 ins.type = BTRFS_METADATA_ITEM_KEY;
2218 ins.offset = node->num_bytes;
2219 ins.type = BTRFS_EXTENT_ITEM_KEY;
2222 BUG_ON(node->ref_mod != 1);
2223 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2224 BUG_ON(!extent_op || !extent_op->update_flags);
2225 ret = alloc_reserved_tree_block(trans, root,
2227 extent_op->flags_to_set,
2231 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2232 ret = __btrfs_inc_extent_ref(trans, root, node,
2236 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2237 ret = __btrfs_free_extent(trans, root, node,
2239 ref->level, 0, 1, extent_op);
2246 /* helper function to actually process a single delayed ref entry */
2247 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2248 struct btrfs_root *root,
2249 struct btrfs_delayed_ref_node *node,
2250 struct btrfs_delayed_extent_op *extent_op,
2251 int insert_reserved)
2255 if (trans->aborted) {
2256 if (insert_reserved)
2257 btrfs_pin_extent(root, node->bytenr,
2258 node->num_bytes, 1);
2262 if (btrfs_delayed_ref_is_head(node)) {
2263 struct btrfs_delayed_ref_head *head;
2265 * we've hit the end of the chain and we were supposed
2266 * to insert this extent into the tree. But, it got
2267 * deleted before we ever needed to insert it, so all
2268 * we have to do is clean up the accounting
2271 head = btrfs_delayed_node_to_head(node);
2272 trace_run_delayed_ref_head(node, head, node->action);
2274 if (insert_reserved) {
2275 btrfs_pin_extent(root, node->bytenr,
2276 node->num_bytes, 1);
2277 if (head->is_data) {
2278 ret = btrfs_del_csums(trans, root,
2286 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2287 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2288 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2290 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2291 node->type == BTRFS_SHARED_DATA_REF_KEY)
2292 ret = run_delayed_data_ref(trans, root, node, extent_op,
2299 static inline struct btrfs_delayed_ref_node *
2300 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2302 if (list_empty(&head->ref_list))
2305 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2310 * Returns 0 on success or if called with an already aborted transaction.
2311 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2313 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2314 struct btrfs_root *root,
2317 struct btrfs_delayed_ref_root *delayed_refs;
2318 struct btrfs_delayed_ref_node *ref;
2319 struct btrfs_delayed_ref_head *locked_ref = NULL;
2320 struct btrfs_delayed_extent_op *extent_op;
2321 struct btrfs_fs_info *fs_info = root->fs_info;
2322 ktime_t start = ktime_get();
2324 unsigned long count = 0;
2325 unsigned long actual_count = 0;
2326 int must_insert_reserved = 0;
2328 delayed_refs = &trans->transaction->delayed_refs;
2334 spin_lock(&delayed_refs->lock);
2335 locked_ref = btrfs_select_ref_head(trans);
2337 spin_unlock(&delayed_refs->lock);
2341 /* grab the lock that says we are going to process
2342 * all the refs for this head */
2343 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2344 spin_unlock(&delayed_refs->lock);
2346 * we may have dropped the spin lock to get the head
2347 * mutex lock, and that might have given someone else
2348 * time to free the head. If that's true, it has been
2349 * removed from our list and we can move on.
2351 if (ret == -EAGAIN) {
2358 spin_lock(&locked_ref->lock);
2361 * locked_ref is the head node, so we have to go one
2362 * node back for any delayed ref updates
2364 ref = select_delayed_ref(locked_ref);
2366 if (ref && ref->seq &&
2367 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2368 spin_unlock(&locked_ref->lock);
2369 btrfs_delayed_ref_unlock(locked_ref);
2370 spin_lock(&delayed_refs->lock);
2371 locked_ref->processing = 0;
2372 delayed_refs->num_heads_ready++;
2373 spin_unlock(&delayed_refs->lock);
2381 * record the must insert reserved flag before we
2382 * drop the spin lock.
2384 must_insert_reserved = locked_ref->must_insert_reserved;
2385 locked_ref->must_insert_reserved = 0;
2387 extent_op = locked_ref->extent_op;
2388 locked_ref->extent_op = NULL;
2393 /* All delayed refs have been processed, Go ahead
2394 * and send the head node to run_one_delayed_ref,
2395 * so that any accounting fixes can happen
2397 ref = &locked_ref->node;
2399 if (extent_op && must_insert_reserved) {
2400 btrfs_free_delayed_extent_op(extent_op);
2405 spin_unlock(&locked_ref->lock);
2406 ret = run_delayed_extent_op(trans, root,
2408 btrfs_free_delayed_extent_op(extent_op);
2412 * Need to reset must_insert_reserved if
2413 * there was an error so the abort stuff
2414 * can cleanup the reserved space
2417 if (must_insert_reserved)
2418 locked_ref->must_insert_reserved = 1;
2419 locked_ref->processing = 0;
2420 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2421 btrfs_delayed_ref_unlock(locked_ref);
2428 * Need to drop our head ref lock and re-aqcuire the
2429 * delayed ref lock and then re-check to make sure
2432 spin_unlock(&locked_ref->lock);
2433 spin_lock(&delayed_refs->lock);
2434 spin_lock(&locked_ref->lock);
2435 if (!list_empty(&locked_ref->ref_list) ||
2436 locked_ref->extent_op) {
2437 spin_unlock(&locked_ref->lock);
2438 spin_unlock(&delayed_refs->lock);
2442 delayed_refs->num_heads--;
2443 rb_erase(&locked_ref->href_node,
2444 &delayed_refs->href_root);
2445 spin_unlock(&delayed_refs->lock);
2449 list_del(&ref->list);
2451 atomic_dec(&delayed_refs->num_entries);
2453 if (!btrfs_delayed_ref_is_head(ref)) {
2455 * when we play the delayed ref, also correct the
2458 switch (ref->action) {
2459 case BTRFS_ADD_DELAYED_REF:
2460 case BTRFS_ADD_DELAYED_EXTENT:
2461 locked_ref->node.ref_mod -= ref->ref_mod;
2463 case BTRFS_DROP_DELAYED_REF:
2464 locked_ref->node.ref_mod += ref->ref_mod;
2470 spin_unlock(&locked_ref->lock);
2472 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2473 must_insert_reserved);
2475 btrfs_free_delayed_extent_op(extent_op);
2477 locked_ref->processing = 0;
2478 btrfs_delayed_ref_unlock(locked_ref);
2479 btrfs_put_delayed_ref(ref);
2480 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2485 * If this node is a head, that means all the refs in this head
2486 * have been dealt with, and we will pick the next head to deal
2487 * with, so we must unlock the head and drop it from the cluster
2488 * list before we release it.
2490 if (btrfs_delayed_ref_is_head(ref)) {
2491 if (locked_ref->is_data &&
2492 locked_ref->total_ref_mod < 0) {
2493 spin_lock(&delayed_refs->lock);
2494 delayed_refs->pending_csums -= ref->num_bytes;
2495 spin_unlock(&delayed_refs->lock);
2497 btrfs_delayed_ref_unlock(locked_ref);
2500 btrfs_put_delayed_ref(ref);
2506 * We don't want to include ref heads since we can have empty ref heads
2507 * and those will drastically skew our runtime down since we just do
2508 * accounting, no actual extent tree updates.
2510 if (actual_count > 0) {
2511 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2515 * We weigh the current average higher than our current runtime
2516 * to avoid large swings in the average.
2518 spin_lock(&delayed_refs->lock);
2519 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2520 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2521 spin_unlock(&delayed_refs->lock);
2526 #ifdef SCRAMBLE_DELAYED_REFS
2528 * Normally delayed refs get processed in ascending bytenr order. This
2529 * correlates in most cases to the order added. To expose dependencies on this
2530 * order, we start to process the tree in the middle instead of the beginning
2532 static u64 find_middle(struct rb_root *root)
2534 struct rb_node *n = root->rb_node;
2535 struct btrfs_delayed_ref_node *entry;
2538 u64 first = 0, last = 0;
2542 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2543 first = entry->bytenr;
2547 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2548 last = entry->bytenr;
2553 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2554 WARN_ON(!entry->in_tree);
2556 middle = entry->bytenr;
2569 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2573 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2574 sizeof(struct btrfs_extent_inline_ref));
2575 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2576 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2579 * We don't ever fill up leaves all the way so multiply by 2 just to be
2580 * closer to what we're really going to want to ouse.
2582 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2586 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2587 * would require to store the csums for that many bytes.
2589 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2592 u64 num_csums_per_leaf;
2595 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2596 num_csums_per_leaf = div64_u64(csum_size,
2597 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2598 num_csums = div64_u64(csum_bytes, root->sectorsize);
2599 num_csums += num_csums_per_leaf - 1;
2600 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2604 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root)
2607 struct btrfs_block_rsv *global_rsv;
2608 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2609 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2610 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2611 u64 num_bytes, num_dirty_bgs_bytes;
2614 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2615 num_heads = heads_to_leaves(root, num_heads);
2617 num_bytes += (num_heads - 1) * root->nodesize;
2619 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2620 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2622 global_rsv = &root->fs_info->global_block_rsv;
2625 * If we can't allocate any more chunks lets make sure we have _lots_ of
2626 * wiggle room since running delayed refs can create more delayed refs.
2628 if (global_rsv->space_info->full) {
2629 num_dirty_bgs_bytes <<= 1;
2633 spin_lock(&global_rsv->lock);
2634 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2636 spin_unlock(&global_rsv->lock);
2640 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root)
2643 struct btrfs_fs_info *fs_info = root->fs_info;
2645 atomic_read(&trans->transaction->delayed_refs.num_entries);
2650 avg_runtime = fs_info->avg_delayed_ref_runtime;
2651 val = num_entries * avg_runtime;
2652 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2654 if (val >= NSEC_PER_SEC / 2)
2657 return btrfs_check_space_for_delayed_refs(trans, root);
2660 struct async_delayed_refs {
2661 struct btrfs_root *root;
2665 struct completion wait;
2666 struct btrfs_work work;
2669 static void delayed_ref_async_start(struct btrfs_work *work)
2671 struct async_delayed_refs *async;
2672 struct btrfs_trans_handle *trans;
2675 async = container_of(work, struct async_delayed_refs, work);
2677 trans = btrfs_join_transaction(async->root);
2678 if (IS_ERR(trans)) {
2679 async->error = PTR_ERR(trans);
2684 * trans->sync means that when we call end_transaciton, we won't
2685 * wait on delayed refs
2688 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2692 ret = btrfs_end_transaction(trans, async->root);
2693 if (ret && !async->error)
2697 complete(&async->wait);
2702 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2703 unsigned long count, int wait)
2705 struct async_delayed_refs *async;
2708 async = kmalloc(sizeof(*async), GFP_NOFS);
2712 async->root = root->fs_info->tree_root;
2713 async->count = count;
2719 init_completion(&async->wait);
2721 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2722 delayed_ref_async_start, NULL, NULL);
2724 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2727 wait_for_completion(&async->wait);
2736 * this starts processing the delayed reference count updates and
2737 * extent insertions we have queued up so far. count can be
2738 * 0, which means to process everything in the tree at the start
2739 * of the run (but not newly added entries), or it can be some target
2740 * number you'd like to process.
2742 * Returns 0 on success or if called with an aborted transaction
2743 * Returns <0 on error and aborts the transaction
2745 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2746 struct btrfs_root *root, unsigned long count)
2748 struct rb_node *node;
2749 struct btrfs_delayed_ref_root *delayed_refs;
2750 struct btrfs_delayed_ref_head *head;
2752 int run_all = count == (unsigned long)-1;
2754 /* We'll clean this up in btrfs_cleanup_transaction */
2758 if (root == root->fs_info->extent_root)
2759 root = root->fs_info->tree_root;
2761 delayed_refs = &trans->transaction->delayed_refs;
2763 count = atomic_read(&delayed_refs->num_entries) * 2;
2766 #ifdef SCRAMBLE_DELAYED_REFS
2767 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2769 ret = __btrfs_run_delayed_refs(trans, root, count);
2771 btrfs_abort_transaction(trans, root, ret);
2776 if (!list_empty(&trans->new_bgs))
2777 btrfs_create_pending_block_groups(trans, root);
2779 spin_lock(&delayed_refs->lock);
2780 node = rb_first(&delayed_refs->href_root);
2782 spin_unlock(&delayed_refs->lock);
2785 count = (unsigned long)-1;
2788 head = rb_entry(node, struct btrfs_delayed_ref_head,
2790 if (btrfs_delayed_ref_is_head(&head->node)) {
2791 struct btrfs_delayed_ref_node *ref;
2794 atomic_inc(&ref->refs);
2796 spin_unlock(&delayed_refs->lock);
2798 * Mutex was contended, block until it's
2799 * released and try again
2801 mutex_lock(&head->mutex);
2802 mutex_unlock(&head->mutex);
2804 btrfs_put_delayed_ref(ref);
2810 node = rb_next(node);
2812 spin_unlock(&delayed_refs->lock);
2817 assert_qgroups_uptodate(trans);
2821 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2822 struct btrfs_root *root,
2823 u64 bytenr, u64 num_bytes, u64 flags,
2824 int level, int is_data)
2826 struct btrfs_delayed_extent_op *extent_op;
2829 extent_op = btrfs_alloc_delayed_extent_op();
2833 extent_op->flags_to_set = flags;
2834 extent_op->update_flags = 1;
2835 extent_op->update_key = 0;
2836 extent_op->is_data = is_data ? 1 : 0;
2837 extent_op->level = level;
2839 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2840 num_bytes, extent_op);
2842 btrfs_free_delayed_extent_op(extent_op);
2846 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2847 struct btrfs_root *root,
2848 struct btrfs_path *path,
2849 u64 objectid, u64 offset, u64 bytenr)
2851 struct btrfs_delayed_ref_head *head;
2852 struct btrfs_delayed_ref_node *ref;
2853 struct btrfs_delayed_data_ref *data_ref;
2854 struct btrfs_delayed_ref_root *delayed_refs;
2857 delayed_refs = &trans->transaction->delayed_refs;
2858 spin_lock(&delayed_refs->lock);
2859 head = btrfs_find_delayed_ref_head(trans, bytenr);
2861 spin_unlock(&delayed_refs->lock);
2865 if (!mutex_trylock(&head->mutex)) {
2866 atomic_inc(&head->node.refs);
2867 spin_unlock(&delayed_refs->lock);
2869 btrfs_release_path(path);
2872 * Mutex was contended, block until it's released and let
2875 mutex_lock(&head->mutex);
2876 mutex_unlock(&head->mutex);
2877 btrfs_put_delayed_ref(&head->node);
2880 spin_unlock(&delayed_refs->lock);
2882 spin_lock(&head->lock);
2883 list_for_each_entry(ref, &head->ref_list, list) {
2884 /* If it's a shared ref we know a cross reference exists */
2885 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2890 data_ref = btrfs_delayed_node_to_data_ref(ref);
2893 * If our ref doesn't match the one we're currently looking at
2894 * then we have a cross reference.
2896 if (data_ref->root != root->root_key.objectid ||
2897 data_ref->objectid != objectid ||
2898 data_ref->offset != offset) {
2903 spin_unlock(&head->lock);
2904 mutex_unlock(&head->mutex);
2908 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2909 struct btrfs_root *root,
2910 struct btrfs_path *path,
2911 u64 objectid, u64 offset, u64 bytenr)
2913 struct btrfs_root *extent_root = root->fs_info->extent_root;
2914 struct extent_buffer *leaf;
2915 struct btrfs_extent_data_ref *ref;
2916 struct btrfs_extent_inline_ref *iref;
2917 struct btrfs_extent_item *ei;
2918 struct btrfs_key key;
2922 key.objectid = bytenr;
2923 key.offset = (u64)-1;
2924 key.type = BTRFS_EXTENT_ITEM_KEY;
2926 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2929 BUG_ON(ret == 0); /* Corruption */
2932 if (path->slots[0] == 0)
2936 leaf = path->nodes[0];
2937 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2939 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2943 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2944 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2945 if (item_size < sizeof(*ei)) {
2946 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2950 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2952 if (item_size != sizeof(*ei) +
2953 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2956 if (btrfs_extent_generation(leaf, ei) <=
2957 btrfs_root_last_snapshot(&root->root_item))
2960 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2961 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2962 BTRFS_EXTENT_DATA_REF_KEY)
2965 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2966 if (btrfs_extent_refs(leaf, ei) !=
2967 btrfs_extent_data_ref_count(leaf, ref) ||
2968 btrfs_extent_data_ref_root(leaf, ref) !=
2969 root->root_key.objectid ||
2970 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2971 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2979 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2980 struct btrfs_root *root,
2981 u64 objectid, u64 offset, u64 bytenr)
2983 struct btrfs_path *path;
2987 path = btrfs_alloc_path();
2992 ret = check_committed_ref(trans, root, path, objectid,
2994 if (ret && ret != -ENOENT)
2997 ret2 = check_delayed_ref(trans, root, path, objectid,
2999 } while (ret2 == -EAGAIN);
3001 if (ret2 && ret2 != -ENOENT) {
3006 if (ret != -ENOENT || ret2 != -ENOENT)
3009 btrfs_free_path(path);
3010 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3015 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3016 struct btrfs_root *root,
3017 struct extent_buffer *buf,
3018 int full_backref, int inc)
3025 struct btrfs_key key;
3026 struct btrfs_file_extent_item *fi;
3030 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3031 u64, u64, u64, u64, u64, u64, int);
3034 if (btrfs_test_is_dummy_root(root))
3037 ref_root = btrfs_header_owner(buf);
3038 nritems = btrfs_header_nritems(buf);
3039 level = btrfs_header_level(buf);
3041 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3045 process_func = btrfs_inc_extent_ref;
3047 process_func = btrfs_free_extent;
3050 parent = buf->start;
3054 for (i = 0; i < nritems; i++) {
3056 btrfs_item_key_to_cpu(buf, &key, i);
3057 if (key.type != BTRFS_EXTENT_DATA_KEY)
3059 fi = btrfs_item_ptr(buf, i,
3060 struct btrfs_file_extent_item);
3061 if (btrfs_file_extent_type(buf, fi) ==
3062 BTRFS_FILE_EXTENT_INLINE)
3064 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3068 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3069 key.offset -= btrfs_file_extent_offset(buf, fi);
3070 ret = process_func(trans, root, bytenr, num_bytes,
3071 parent, ref_root, key.objectid,
3076 bytenr = btrfs_node_blockptr(buf, i);
3077 num_bytes = root->nodesize;
3078 ret = process_func(trans, root, bytenr, num_bytes,
3079 parent, ref_root, level - 1, 0,
3090 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3091 struct extent_buffer *buf, int full_backref)
3093 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3096 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3097 struct extent_buffer *buf, int full_backref)
3099 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3102 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3103 struct btrfs_root *root,
3104 struct btrfs_path *path,
3105 struct btrfs_block_group_cache *cache)
3108 struct btrfs_root *extent_root = root->fs_info->extent_root;
3110 struct extent_buffer *leaf;
3112 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3119 leaf = path->nodes[0];
3120 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3121 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3122 btrfs_mark_buffer_dirty(leaf);
3124 btrfs_release_path(path);
3129 static struct btrfs_block_group_cache *
3130 next_block_group(struct btrfs_root *root,
3131 struct btrfs_block_group_cache *cache)
3133 struct rb_node *node;
3135 spin_lock(&root->fs_info->block_group_cache_lock);
3137 /* If our block group was removed, we need a full search. */
3138 if (RB_EMPTY_NODE(&cache->cache_node)) {
3139 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3141 spin_unlock(&root->fs_info->block_group_cache_lock);
3142 btrfs_put_block_group(cache);
3143 cache = btrfs_lookup_first_block_group(root->fs_info,
3147 node = rb_next(&cache->cache_node);
3148 btrfs_put_block_group(cache);
3150 cache = rb_entry(node, struct btrfs_block_group_cache,
3152 btrfs_get_block_group(cache);
3155 spin_unlock(&root->fs_info->block_group_cache_lock);
3159 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3160 struct btrfs_trans_handle *trans,
3161 struct btrfs_path *path)
3163 struct btrfs_root *root = block_group->fs_info->tree_root;
3164 struct inode *inode = NULL;
3166 int dcs = BTRFS_DC_ERROR;
3172 * If this block group is smaller than 100 megs don't bother caching the
3175 if (block_group->key.offset < (100 * 1024 * 1024)) {
3176 spin_lock(&block_group->lock);
3177 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3178 spin_unlock(&block_group->lock);
3185 inode = lookup_free_space_inode(root, block_group, path);
3186 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3187 ret = PTR_ERR(inode);
3188 btrfs_release_path(path);
3192 if (IS_ERR(inode)) {
3196 if (block_group->ro)
3199 ret = create_free_space_inode(root, trans, block_group, path);
3205 /* We've already setup this transaction, go ahead and exit */
3206 if (block_group->cache_generation == trans->transid &&
3207 i_size_read(inode)) {
3208 dcs = BTRFS_DC_SETUP;
3213 * We want to set the generation to 0, that way if anything goes wrong
3214 * from here on out we know not to trust this cache when we load up next
3217 BTRFS_I(inode)->generation = 0;
3218 ret = btrfs_update_inode(trans, root, inode);
3221 * So theoretically we could recover from this, simply set the
3222 * super cache generation to 0 so we know to invalidate the
3223 * cache, but then we'd have to keep track of the block groups
3224 * that fail this way so we know we _have_ to reset this cache
3225 * before the next commit or risk reading stale cache. So to
3226 * limit our exposure to horrible edge cases lets just abort the
3227 * transaction, this only happens in really bad situations
3230 btrfs_abort_transaction(trans, root, ret);
3235 if (i_size_read(inode) > 0) {
3236 ret = btrfs_check_trunc_cache_free_space(root,
3237 &root->fs_info->global_block_rsv);
3241 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3246 spin_lock(&block_group->lock);
3247 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3248 !btrfs_test_opt(root, SPACE_CACHE)) {
3250 * don't bother trying to write stuff out _if_
3251 * a) we're not cached,
3252 * b) we're with nospace_cache mount option.
3254 dcs = BTRFS_DC_WRITTEN;
3255 spin_unlock(&block_group->lock);
3258 spin_unlock(&block_group->lock);
3261 * Try to preallocate enough space based on how big the block group is.
3262 * Keep in mind this has to include any pinned space which could end up
3263 * taking up quite a bit since it's not folded into the other space
3266 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3271 num_pages *= PAGE_CACHE_SIZE;
3273 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3277 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3278 num_pages, num_pages,
3281 dcs = BTRFS_DC_SETUP;
3282 btrfs_free_reserved_data_space(inode, num_pages);
3287 btrfs_release_path(path);
3289 spin_lock(&block_group->lock);
3290 if (!ret && dcs == BTRFS_DC_SETUP)
3291 block_group->cache_generation = trans->transid;
3292 block_group->disk_cache_state = dcs;
3293 spin_unlock(&block_group->lock);
3298 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3299 struct btrfs_root *root)
3301 struct btrfs_block_group_cache *cache, *tmp;
3302 struct btrfs_transaction *cur_trans = trans->transaction;
3303 struct btrfs_path *path;
3305 if (list_empty(&cur_trans->dirty_bgs) ||
3306 !btrfs_test_opt(root, SPACE_CACHE))
3309 path = btrfs_alloc_path();
3313 /* Could add new block groups, use _safe just in case */
3314 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3316 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3317 cache_save_setup(cache, trans, path);
3320 btrfs_free_path(path);
3325 * transaction commit does final block group cache writeback during a
3326 * critical section where nothing is allowed to change the FS. This is
3327 * required in order for the cache to actually match the block group,
3328 * but can introduce a lot of latency into the commit.
3330 * So, btrfs_start_dirty_block_groups is here to kick off block group
3331 * cache IO. There's a chance we'll have to redo some of it if the
3332 * block group changes again during the commit, but it greatly reduces
3333 * the commit latency by getting rid of the easy block groups while
3334 * we're still allowing others to join the commit.
3336 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3337 struct btrfs_root *root)
3339 struct btrfs_block_group_cache *cache;
3340 struct btrfs_transaction *cur_trans = trans->transaction;
3343 struct btrfs_path *path = NULL;
3345 struct list_head *io = &cur_trans->io_bgs;
3346 int num_started = 0;
3349 spin_lock(&cur_trans->dirty_bgs_lock);
3350 if (list_empty(&cur_trans->dirty_bgs)) {
3351 spin_unlock(&cur_trans->dirty_bgs_lock);
3354 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3355 spin_unlock(&cur_trans->dirty_bgs_lock);
3359 * make sure all the block groups on our dirty list actually
3362 btrfs_create_pending_block_groups(trans, root);
3365 path = btrfs_alloc_path();
3371 * cache_write_mutex is here only to save us from balance or automatic
3372 * removal of empty block groups deleting this block group while we are
3373 * writing out the cache
3375 mutex_lock(&trans->transaction->cache_write_mutex);
3376 while (!list_empty(&dirty)) {
3377 cache = list_first_entry(&dirty,
3378 struct btrfs_block_group_cache,
3381 * this can happen if something re-dirties a block
3382 * group that is already under IO. Just wait for it to
3383 * finish and then do it all again
3385 if (!list_empty(&cache->io_list)) {
3386 list_del_init(&cache->io_list);
3387 btrfs_wait_cache_io(root, trans, cache,
3388 &cache->io_ctl, path,
3389 cache->key.objectid);
3390 btrfs_put_block_group(cache);
3395 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3396 * if it should update the cache_state. Don't delete
3397 * until after we wait.
3399 * Since we're not running in the commit critical section
3400 * we need the dirty_bgs_lock to protect from update_block_group
3402 spin_lock(&cur_trans->dirty_bgs_lock);
3403 list_del_init(&cache->dirty_list);
3404 spin_unlock(&cur_trans->dirty_bgs_lock);
3408 cache_save_setup(cache, trans, path);
3410 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3411 cache->io_ctl.inode = NULL;
3412 ret = btrfs_write_out_cache(root, trans, cache, path);
3413 if (ret == 0 && cache->io_ctl.inode) {
3418 * the cache_write_mutex is protecting
3421 list_add_tail(&cache->io_list, io);
3424 * if we failed to write the cache, the
3425 * generation will be bad and life goes on
3431 ret = write_one_cache_group(trans, root, path, cache);
3433 * Our block group might still be attached to the list
3434 * of new block groups in the transaction handle of some
3435 * other task (struct btrfs_trans_handle->new_bgs). This
3436 * means its block group item isn't yet in the extent
3437 * tree. If this happens ignore the error, as we will
3438 * try again later in the critical section of the
3439 * transaction commit.
3441 if (ret == -ENOENT) {
3443 spin_lock(&cur_trans->dirty_bgs_lock);
3444 if (list_empty(&cache->dirty_list)) {
3445 list_add_tail(&cache->dirty_list,
3446 &cur_trans->dirty_bgs);
3447 btrfs_get_block_group(cache);
3449 spin_unlock(&cur_trans->dirty_bgs_lock);
3451 btrfs_abort_transaction(trans, root, ret);
3455 /* if its not on the io list, we need to put the block group */
3457 btrfs_put_block_group(cache);
3463 * Avoid blocking other tasks for too long. It might even save
3464 * us from writing caches for block groups that are going to be
3467 mutex_unlock(&trans->transaction->cache_write_mutex);
3468 mutex_lock(&trans->transaction->cache_write_mutex);
3470 mutex_unlock(&trans->transaction->cache_write_mutex);
3473 * go through delayed refs for all the stuff we've just kicked off
3474 * and then loop back (just once)
3476 ret = btrfs_run_delayed_refs(trans, root, 0);
3477 if (!ret && loops == 0) {
3479 spin_lock(&cur_trans->dirty_bgs_lock);
3480 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3482 * dirty_bgs_lock protects us from concurrent block group
3483 * deletes too (not just cache_write_mutex).
3485 if (!list_empty(&dirty)) {
3486 spin_unlock(&cur_trans->dirty_bgs_lock);
3489 spin_unlock(&cur_trans->dirty_bgs_lock);
3492 btrfs_free_path(path);
3496 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3497 struct btrfs_root *root)
3499 struct btrfs_block_group_cache *cache;
3500 struct btrfs_transaction *cur_trans = trans->transaction;
3503 struct btrfs_path *path;
3504 struct list_head *io = &cur_trans->io_bgs;
3505 int num_started = 0;
3507 path = btrfs_alloc_path();
3512 * We don't need the lock here since we are protected by the transaction
3513 * commit. We want to do the cache_save_setup first and then run the
3514 * delayed refs to make sure we have the best chance at doing this all
3517 while (!list_empty(&cur_trans->dirty_bgs)) {
3518 cache = list_first_entry(&cur_trans->dirty_bgs,
3519 struct btrfs_block_group_cache,
3523 * this can happen if cache_save_setup re-dirties a block
3524 * group that is already under IO. Just wait for it to
3525 * finish and then do it all again
3527 if (!list_empty(&cache->io_list)) {
3528 list_del_init(&cache->io_list);
3529 btrfs_wait_cache_io(root, trans, cache,
3530 &cache->io_ctl, path,
3531 cache->key.objectid);
3532 btrfs_put_block_group(cache);
3536 * don't remove from the dirty list until after we've waited
3539 list_del_init(&cache->dirty_list);
3542 cache_save_setup(cache, trans, path);
3545 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3547 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3548 cache->io_ctl.inode = NULL;
3549 ret = btrfs_write_out_cache(root, trans, cache, path);
3550 if (ret == 0 && cache->io_ctl.inode) {
3553 list_add_tail(&cache->io_list, io);
3556 * if we failed to write the cache, the
3557 * generation will be bad and life goes on
3563 ret = write_one_cache_group(trans, root, path, cache);
3565 btrfs_abort_transaction(trans, root, ret);
3568 /* if its not on the io list, we need to put the block group */
3570 btrfs_put_block_group(cache);
3573 while (!list_empty(io)) {
3574 cache = list_first_entry(io, struct btrfs_block_group_cache,
3576 list_del_init(&cache->io_list);
3577 btrfs_wait_cache_io(root, trans, cache,
3578 &cache->io_ctl, path, cache->key.objectid);
3579 btrfs_put_block_group(cache);
3582 btrfs_free_path(path);
3586 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3588 struct btrfs_block_group_cache *block_group;
3591 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3592 if (!block_group || block_group->ro)
3595 btrfs_put_block_group(block_group);
3599 static const char *alloc_name(u64 flags)
3602 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3604 case BTRFS_BLOCK_GROUP_METADATA:
3606 case BTRFS_BLOCK_GROUP_DATA:
3608 case BTRFS_BLOCK_GROUP_SYSTEM:
3612 return "invalid-combination";
3616 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3617 u64 total_bytes, u64 bytes_used,
3618 struct btrfs_space_info **space_info)
3620 struct btrfs_space_info *found;
3625 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3626 BTRFS_BLOCK_GROUP_RAID10))
3631 found = __find_space_info(info, flags);
3633 spin_lock(&found->lock);
3634 found->total_bytes += total_bytes;
3635 found->disk_total += total_bytes * factor;
3636 found->bytes_used += bytes_used;
3637 found->disk_used += bytes_used * factor;
3638 if (total_bytes > 0)
3640 spin_unlock(&found->lock);
3641 *space_info = found;
3644 found = kzalloc(sizeof(*found), GFP_NOFS);
3648 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3654 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3655 INIT_LIST_HEAD(&found->block_groups[i]);
3656 init_rwsem(&found->groups_sem);
3657 spin_lock_init(&found->lock);
3658 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3659 found->total_bytes = total_bytes;
3660 found->disk_total = total_bytes * factor;
3661 found->bytes_used = bytes_used;
3662 found->disk_used = bytes_used * factor;
3663 found->bytes_pinned = 0;
3664 found->bytes_reserved = 0;
3665 found->bytes_readonly = 0;
3666 found->bytes_may_use = 0;
3667 if (total_bytes > 0)
3671 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3672 found->chunk_alloc = 0;
3674 init_waitqueue_head(&found->wait);
3675 INIT_LIST_HEAD(&found->ro_bgs);
3677 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3678 info->space_info_kobj, "%s",
3679 alloc_name(found->flags));
3685 *space_info = found;
3686 list_add_rcu(&found->list, &info->space_info);
3687 if (flags & BTRFS_BLOCK_GROUP_DATA)
3688 info->data_sinfo = found;
3693 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3695 u64 extra_flags = chunk_to_extended(flags) &
3696 BTRFS_EXTENDED_PROFILE_MASK;
3698 write_seqlock(&fs_info->profiles_lock);
3699 if (flags & BTRFS_BLOCK_GROUP_DATA)
3700 fs_info->avail_data_alloc_bits |= extra_flags;
3701 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3702 fs_info->avail_metadata_alloc_bits |= extra_flags;
3703 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3704 fs_info->avail_system_alloc_bits |= extra_flags;
3705 write_sequnlock(&fs_info->profiles_lock);
3709 * returns target flags in extended format or 0 if restripe for this
3710 * chunk_type is not in progress
3712 * should be called with either volume_mutex or balance_lock held
3714 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3716 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3722 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3723 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3724 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3725 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3726 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3727 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3728 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3729 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3730 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3737 * @flags: available profiles in extended format (see ctree.h)
3739 * Returns reduced profile in chunk format. If profile changing is in
3740 * progress (either running or paused) picks the target profile (if it's
3741 * already available), otherwise falls back to plain reducing.
3743 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3745 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3750 * see if restripe for this chunk_type is in progress, if so
3751 * try to reduce to the target profile
3753 spin_lock(&root->fs_info->balance_lock);
3754 target = get_restripe_target(root->fs_info, flags);
3756 /* pick target profile only if it's already available */
3757 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3758 spin_unlock(&root->fs_info->balance_lock);
3759 return extended_to_chunk(target);
3762 spin_unlock(&root->fs_info->balance_lock);
3764 /* First, mask out the RAID levels which aren't possible */
3765 if (num_devices == 1)
3766 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3767 BTRFS_BLOCK_GROUP_RAID5);
3768 if (num_devices < 3)
3769 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3770 if (num_devices < 4)
3771 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3773 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3774 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3775 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3778 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3779 tmp = BTRFS_BLOCK_GROUP_RAID6;
3780 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3781 tmp = BTRFS_BLOCK_GROUP_RAID5;
3782 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3783 tmp = BTRFS_BLOCK_GROUP_RAID10;
3784 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3785 tmp = BTRFS_BLOCK_GROUP_RAID1;
3786 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3787 tmp = BTRFS_BLOCK_GROUP_RAID0;
3789 return extended_to_chunk(flags | tmp);
3792 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3799 seq = read_seqbegin(&root->fs_info->profiles_lock);
3801 if (flags & BTRFS_BLOCK_GROUP_DATA)
3802 flags |= root->fs_info->avail_data_alloc_bits;
3803 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3804 flags |= root->fs_info->avail_system_alloc_bits;
3805 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3806 flags |= root->fs_info->avail_metadata_alloc_bits;
3807 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3809 return btrfs_reduce_alloc_profile(root, flags);
3812 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3818 flags = BTRFS_BLOCK_GROUP_DATA;
3819 else if (root == root->fs_info->chunk_root)
3820 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3822 flags = BTRFS_BLOCK_GROUP_METADATA;
3824 ret = get_alloc_profile(root, flags);
3829 * This will check the space that the inode allocates from to make sure we have
3830 * enough space for bytes.
3832 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3834 struct btrfs_space_info *data_sinfo;
3835 struct btrfs_root *root = BTRFS_I(inode)->root;
3836 struct btrfs_fs_info *fs_info = root->fs_info;
3839 int need_commit = 2;
3840 int have_pinned_space;
3842 /* make sure bytes are sectorsize aligned */
3843 bytes = ALIGN(bytes, root->sectorsize);
3845 if (btrfs_is_free_space_inode(inode)) {
3847 ASSERT(current->journal_info);
3850 data_sinfo = fs_info->data_sinfo;
3855 /* make sure we have enough space to handle the data first */
3856 spin_lock(&data_sinfo->lock);
3857 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3858 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3859 data_sinfo->bytes_may_use;
3861 if (used + bytes > data_sinfo->total_bytes) {
3862 struct btrfs_trans_handle *trans;
3865 * if we don't have enough free bytes in this space then we need
3866 * to alloc a new chunk.
3868 if (!data_sinfo->full) {
3871 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3872 spin_unlock(&data_sinfo->lock);
3874 alloc_target = btrfs_get_alloc_profile(root, 1);
3876 * It is ugly that we don't call nolock join
3877 * transaction for the free space inode case here.
3878 * But it is safe because we only do the data space
3879 * reservation for the free space cache in the
3880 * transaction context, the common join transaction
3881 * just increase the counter of the current transaction
3882 * handler, doesn't try to acquire the trans_lock of
3885 trans = btrfs_join_transaction(root);
3887 return PTR_ERR(trans);
3889 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3891 CHUNK_ALLOC_NO_FORCE);
3892 btrfs_end_transaction(trans, root);
3897 have_pinned_space = 1;
3903 data_sinfo = fs_info->data_sinfo;
3909 * If we don't have enough pinned space to deal with this
3910 * allocation, and no removed chunk in current transaction,
3911 * don't bother committing the transaction.
3913 have_pinned_space = percpu_counter_compare(
3914 &data_sinfo->total_bytes_pinned,
3915 used + bytes - data_sinfo->total_bytes);
3916 spin_unlock(&data_sinfo->lock);
3918 /* commit the current transaction and try again */
3921 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3924 trans = btrfs_join_transaction(root);
3926 return PTR_ERR(trans);
3927 if (have_pinned_space >= 0 ||
3928 trans->transaction->have_free_bgs ||
3930 ret = btrfs_commit_transaction(trans, root);
3934 * make sure that all running delayed iput are
3937 down_write(&root->fs_info->delayed_iput_sem);
3938 up_write(&root->fs_info->delayed_iput_sem);
3941 btrfs_end_transaction(trans, root);
3945 trace_btrfs_space_reservation(root->fs_info,
3946 "space_info:enospc",
3947 data_sinfo->flags, bytes, 1);
3950 ret = btrfs_qgroup_reserve(root, write_bytes);
3953 data_sinfo->bytes_may_use += bytes;
3954 trace_btrfs_space_reservation(root->fs_info, "space_info",
3955 data_sinfo->flags, bytes, 1);
3957 spin_unlock(&data_sinfo->lock);
3963 * Called if we need to clear a data reservation for this inode.
3965 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3967 struct btrfs_root *root = BTRFS_I(inode)->root;
3968 struct btrfs_space_info *data_sinfo;
3970 /* make sure bytes are sectorsize aligned */
3971 bytes = ALIGN(bytes, root->sectorsize);
3973 data_sinfo = root->fs_info->data_sinfo;
3974 spin_lock(&data_sinfo->lock);
3975 WARN_ON(data_sinfo->bytes_may_use < bytes);
3976 data_sinfo->bytes_may_use -= bytes;
3977 trace_btrfs_space_reservation(root->fs_info, "space_info",
3978 data_sinfo->flags, bytes, 0);
3979 spin_unlock(&data_sinfo->lock);
3982 static void force_metadata_allocation(struct btrfs_fs_info *info)
3984 struct list_head *head = &info->space_info;
3985 struct btrfs_space_info *found;
3988 list_for_each_entry_rcu(found, head, list) {
3989 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3990 found->force_alloc = CHUNK_ALLOC_FORCE;
3995 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3997 return (global->size << 1);
4000 static int should_alloc_chunk(struct btrfs_root *root,
4001 struct btrfs_space_info *sinfo, int force)
4003 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4004 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4005 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4008 if (force == CHUNK_ALLOC_FORCE)
4012 * We need to take into account the global rsv because for all intents
4013 * and purposes it's used space. Don't worry about locking the
4014 * global_rsv, it doesn't change except when the transaction commits.
4016 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4017 num_allocated += calc_global_rsv_need_space(global_rsv);
4020 * in limited mode, we want to have some free space up to
4021 * about 1% of the FS size.
4023 if (force == CHUNK_ALLOC_LIMITED) {
4024 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4025 thresh = max_t(u64, 64 * 1024 * 1024,
4026 div_factor_fine(thresh, 1));
4028 if (num_bytes - num_allocated < thresh)
4032 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4037 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4041 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4042 BTRFS_BLOCK_GROUP_RAID0 |
4043 BTRFS_BLOCK_GROUP_RAID5 |
4044 BTRFS_BLOCK_GROUP_RAID6))
4045 num_dev = root->fs_info->fs_devices->rw_devices;
4046 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4049 num_dev = 1; /* DUP or single */
4055 * If @is_allocation is true, reserve space in the system space info necessary
4056 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4059 void check_system_chunk(struct btrfs_trans_handle *trans,
4060 struct btrfs_root *root,
4063 struct btrfs_space_info *info;
4070 * Needed because we can end up allocating a system chunk and for an
4071 * atomic and race free space reservation in the chunk block reserve.
4073 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4075 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4076 spin_lock(&info->lock);
4077 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4078 info->bytes_reserved - info->bytes_readonly -
4079 info->bytes_may_use;
4080 spin_unlock(&info->lock);
4082 num_devs = get_profile_num_devs(root, type);
4084 /* num_devs device items to update and 1 chunk item to add or remove */
4085 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4086 btrfs_calc_trans_metadata_size(root, 1);
4088 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4089 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4090 left, thresh, type);
4091 dump_space_info(info, 0, 0);
4094 if (left < thresh) {
4097 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4099 * Ignore failure to create system chunk. We might end up not
4100 * needing it, as we might not need to COW all nodes/leafs from
4101 * the paths we visit in the chunk tree (they were already COWed
4102 * or created in the current transaction for example).
4104 ret = btrfs_alloc_chunk(trans, root, flags);
4108 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4109 &root->fs_info->chunk_block_rsv,
4110 thresh, BTRFS_RESERVE_NO_FLUSH);
4112 trans->chunk_bytes_reserved += thresh;
4116 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4117 struct btrfs_root *extent_root, u64 flags, int force)
4119 struct btrfs_space_info *space_info;
4120 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4121 int wait_for_alloc = 0;
4124 /* Don't re-enter if we're already allocating a chunk */
4125 if (trans->allocating_chunk)
4128 space_info = __find_space_info(extent_root->fs_info, flags);
4130 ret = update_space_info(extent_root->fs_info, flags,
4132 BUG_ON(ret); /* -ENOMEM */
4134 BUG_ON(!space_info); /* Logic error */
4137 spin_lock(&space_info->lock);
4138 if (force < space_info->force_alloc)
4139 force = space_info->force_alloc;
4140 if (space_info->full) {
4141 if (should_alloc_chunk(extent_root, space_info, force))
4145 spin_unlock(&space_info->lock);
4149 if (!should_alloc_chunk(extent_root, space_info, force)) {
4150 spin_unlock(&space_info->lock);
4152 } else if (space_info->chunk_alloc) {
4155 space_info->chunk_alloc = 1;
4158 spin_unlock(&space_info->lock);
4160 mutex_lock(&fs_info->chunk_mutex);
4163 * The chunk_mutex is held throughout the entirety of a chunk
4164 * allocation, so once we've acquired the chunk_mutex we know that the
4165 * other guy is done and we need to recheck and see if we should
4168 if (wait_for_alloc) {
4169 mutex_unlock(&fs_info->chunk_mutex);
4174 trans->allocating_chunk = true;
4177 * If we have mixed data/metadata chunks we want to make sure we keep
4178 * allocating mixed chunks instead of individual chunks.
4180 if (btrfs_mixed_space_info(space_info))
4181 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4184 * if we're doing a data chunk, go ahead and make sure that
4185 * we keep a reasonable number of metadata chunks allocated in the
4188 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4189 fs_info->data_chunk_allocations++;
4190 if (!(fs_info->data_chunk_allocations %
4191 fs_info->metadata_ratio))
4192 force_metadata_allocation(fs_info);
4196 * Check if we have enough space in SYSTEM chunk because we may need
4197 * to update devices.
4199 check_system_chunk(trans, extent_root, flags);
4201 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4202 trans->allocating_chunk = false;
4204 spin_lock(&space_info->lock);
4205 if (ret < 0 && ret != -ENOSPC)
4208 space_info->full = 1;
4212 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4214 space_info->chunk_alloc = 0;
4215 spin_unlock(&space_info->lock);
4216 mutex_unlock(&fs_info->chunk_mutex);
4220 static int can_overcommit(struct btrfs_root *root,
4221 struct btrfs_space_info *space_info, u64 bytes,
4222 enum btrfs_reserve_flush_enum flush)
4224 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4225 u64 profile = btrfs_get_alloc_profile(root, 0);
4230 used = space_info->bytes_used + space_info->bytes_reserved +
4231 space_info->bytes_pinned + space_info->bytes_readonly;
4234 * We only want to allow over committing if we have lots of actual space
4235 * free, but if we don't have enough space to handle the global reserve
4236 * space then we could end up having a real enospc problem when trying
4237 * to allocate a chunk or some other such important allocation.
4239 spin_lock(&global_rsv->lock);
4240 space_size = calc_global_rsv_need_space(global_rsv);
4241 spin_unlock(&global_rsv->lock);
4242 if (used + space_size >= space_info->total_bytes)
4245 used += space_info->bytes_may_use;
4247 spin_lock(&root->fs_info->free_chunk_lock);
4248 avail = root->fs_info->free_chunk_space;
4249 spin_unlock(&root->fs_info->free_chunk_lock);
4252 * If we have dup, raid1 or raid10 then only half of the free
4253 * space is actually useable. For raid56, the space info used
4254 * doesn't include the parity drive, so we don't have to
4257 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4258 BTRFS_BLOCK_GROUP_RAID1 |
4259 BTRFS_BLOCK_GROUP_RAID10))
4263 * If we aren't flushing all things, let us overcommit up to
4264 * 1/2th of the space. If we can flush, don't let us overcommit
4265 * too much, let it overcommit up to 1/8 of the space.
4267 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4272 if (used + bytes < space_info->total_bytes + avail)
4277 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4278 unsigned long nr_pages, int nr_items)
4280 struct super_block *sb = root->fs_info->sb;
4282 if (down_read_trylock(&sb->s_umount)) {
4283 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4284 up_read(&sb->s_umount);
4287 * We needn't worry the filesystem going from r/w to r/o though
4288 * we don't acquire ->s_umount mutex, because the filesystem
4289 * should guarantee the delalloc inodes list be empty after
4290 * the filesystem is readonly(all dirty pages are written to
4293 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4294 if (!current->journal_info)
4295 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4299 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4304 bytes = btrfs_calc_trans_metadata_size(root, 1);
4305 nr = (int)div64_u64(to_reclaim, bytes);
4311 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4314 * shrink metadata reservation for delalloc
4316 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4319 struct btrfs_block_rsv *block_rsv;
4320 struct btrfs_space_info *space_info;
4321 struct btrfs_trans_handle *trans;
4325 unsigned long nr_pages;
4328 enum btrfs_reserve_flush_enum flush;
4330 /* Calc the number of the pages we need flush for space reservation */
4331 items = calc_reclaim_items_nr(root, to_reclaim);
4332 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4334 trans = (struct btrfs_trans_handle *)current->journal_info;
4335 block_rsv = &root->fs_info->delalloc_block_rsv;
4336 space_info = block_rsv->space_info;
4338 delalloc_bytes = percpu_counter_sum_positive(
4339 &root->fs_info->delalloc_bytes);
4340 if (delalloc_bytes == 0) {
4344 btrfs_wait_ordered_roots(root->fs_info, items);
4349 while (delalloc_bytes && loops < 3) {
4350 max_reclaim = min(delalloc_bytes, to_reclaim);
4351 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4352 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4354 * We need to wait for the async pages to actually start before
4357 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4361 if (max_reclaim <= nr_pages)
4364 max_reclaim -= nr_pages;
4366 wait_event(root->fs_info->async_submit_wait,
4367 atomic_read(&root->fs_info->async_delalloc_pages) <=
4371 flush = BTRFS_RESERVE_FLUSH_ALL;
4373 flush = BTRFS_RESERVE_NO_FLUSH;
4374 spin_lock(&space_info->lock);
4375 if (can_overcommit(root, space_info, orig, flush)) {
4376 spin_unlock(&space_info->lock);
4379 spin_unlock(&space_info->lock);
4382 if (wait_ordered && !trans) {
4383 btrfs_wait_ordered_roots(root->fs_info, items);
4385 time_left = schedule_timeout_killable(1);
4389 delalloc_bytes = percpu_counter_sum_positive(
4390 &root->fs_info->delalloc_bytes);
4395 * maybe_commit_transaction - possibly commit the transaction if its ok to
4396 * @root - the root we're allocating for
4397 * @bytes - the number of bytes we want to reserve
4398 * @force - force the commit
4400 * This will check to make sure that committing the transaction will actually
4401 * get us somewhere and then commit the transaction if it does. Otherwise it
4402 * will return -ENOSPC.
4404 static int may_commit_transaction(struct btrfs_root *root,
4405 struct btrfs_space_info *space_info,
4406 u64 bytes, int force)
4408 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4409 struct btrfs_trans_handle *trans;
4411 trans = (struct btrfs_trans_handle *)current->journal_info;
4418 /* See if there is enough pinned space to make this reservation */
4419 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4424 * See if there is some space in the delayed insertion reservation for
4427 if (space_info != delayed_rsv->space_info)
4430 spin_lock(&delayed_rsv->lock);
4431 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4432 bytes - delayed_rsv->size) >= 0) {
4433 spin_unlock(&delayed_rsv->lock);
4436 spin_unlock(&delayed_rsv->lock);
4439 trans = btrfs_join_transaction(root);
4443 return btrfs_commit_transaction(trans, root);
4447 FLUSH_DELAYED_ITEMS_NR = 1,
4448 FLUSH_DELAYED_ITEMS = 2,
4450 FLUSH_DELALLOC_WAIT = 4,
4455 static int flush_space(struct btrfs_root *root,
4456 struct btrfs_space_info *space_info, u64 num_bytes,
4457 u64 orig_bytes, int state)
4459 struct btrfs_trans_handle *trans;
4464 case FLUSH_DELAYED_ITEMS_NR:
4465 case FLUSH_DELAYED_ITEMS:
4466 if (state == FLUSH_DELAYED_ITEMS_NR)
4467 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4471 trans = btrfs_join_transaction(root);
4472 if (IS_ERR(trans)) {
4473 ret = PTR_ERR(trans);
4476 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4477 btrfs_end_transaction(trans, root);
4479 case FLUSH_DELALLOC:
4480 case FLUSH_DELALLOC_WAIT:
4481 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4482 state == FLUSH_DELALLOC_WAIT);
4485 trans = btrfs_join_transaction(root);
4486 if (IS_ERR(trans)) {
4487 ret = PTR_ERR(trans);
4490 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4491 btrfs_get_alloc_profile(root, 0),
4492 CHUNK_ALLOC_NO_FORCE);
4493 btrfs_end_transaction(trans, root);
4498 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4509 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4510 struct btrfs_space_info *space_info)
4516 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4518 spin_lock(&space_info->lock);
4519 if (can_overcommit(root, space_info, to_reclaim,
4520 BTRFS_RESERVE_FLUSH_ALL)) {
4525 used = space_info->bytes_used + space_info->bytes_reserved +
4526 space_info->bytes_pinned + space_info->bytes_readonly +
4527 space_info->bytes_may_use;
4528 if (can_overcommit(root, space_info, 1024 * 1024,
4529 BTRFS_RESERVE_FLUSH_ALL))
4530 expected = div_factor_fine(space_info->total_bytes, 95);
4532 expected = div_factor_fine(space_info->total_bytes, 90);
4534 if (used > expected)
4535 to_reclaim = used - expected;
4538 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4539 space_info->bytes_reserved);
4541 spin_unlock(&space_info->lock);
4546 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4547 struct btrfs_fs_info *fs_info, u64 used)
4549 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4551 /* If we're just plain full then async reclaim just slows us down. */
4552 if (space_info->bytes_used >= thresh)
4555 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4556 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4559 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4560 struct btrfs_fs_info *fs_info,
4565 spin_lock(&space_info->lock);
4567 * We run out of space and have not got any free space via flush_space,
4568 * so don't bother doing async reclaim.
4570 if (flush_state > COMMIT_TRANS && space_info->full) {
4571 spin_unlock(&space_info->lock);
4575 used = space_info->bytes_used + space_info->bytes_reserved +
4576 space_info->bytes_pinned + space_info->bytes_readonly +
4577 space_info->bytes_may_use;
4578 if (need_do_async_reclaim(space_info, fs_info, used)) {
4579 spin_unlock(&space_info->lock);
4582 spin_unlock(&space_info->lock);
4587 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4589 struct btrfs_fs_info *fs_info;
4590 struct btrfs_space_info *space_info;
4594 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4595 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4597 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4602 flush_state = FLUSH_DELAYED_ITEMS_NR;
4604 flush_space(fs_info->fs_root, space_info, to_reclaim,
4605 to_reclaim, flush_state);
4607 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4610 } while (flush_state < COMMIT_TRANS);
4613 void btrfs_init_async_reclaim_work(struct work_struct *work)
4615 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4619 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4620 * @root - the root we're allocating for
4621 * @block_rsv - the block_rsv we're allocating for
4622 * @orig_bytes - the number of bytes we want
4623 * @flush - whether or not we can flush to make our reservation
4625 * This will reserve orgi_bytes number of bytes from the space info associated
4626 * with the block_rsv. If there is not enough space it will make an attempt to
4627 * flush out space to make room. It will do this by flushing delalloc if
4628 * possible or committing the transaction. If flush is 0 then no attempts to
4629 * regain reservations will be made and this will fail if there is not enough
4632 static int reserve_metadata_bytes(struct btrfs_root *root,
4633 struct btrfs_block_rsv *block_rsv,
4635 enum btrfs_reserve_flush_enum flush)
4637 struct btrfs_space_info *space_info = block_rsv->space_info;
4639 u64 num_bytes = orig_bytes;
4640 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4642 bool flushing = false;
4646 spin_lock(&space_info->lock);
4648 * We only want to wait if somebody other than us is flushing and we
4649 * are actually allowed to flush all things.
4651 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4652 space_info->flush) {
4653 spin_unlock(&space_info->lock);
4655 * If we have a trans handle we can't wait because the flusher
4656 * may have to commit the transaction, which would mean we would
4657 * deadlock since we are waiting for the flusher to finish, but
4658 * hold the current transaction open.
4660 if (current->journal_info)
4662 ret = wait_event_killable(space_info->wait, !space_info->flush);
4663 /* Must have been killed, return */
4667 spin_lock(&space_info->lock);
4671 used = space_info->bytes_used + space_info->bytes_reserved +
4672 space_info->bytes_pinned + space_info->bytes_readonly +
4673 space_info->bytes_may_use;
4676 * The idea here is that we've not already over-reserved the block group
4677 * then we can go ahead and save our reservation first and then start
4678 * flushing if we need to. Otherwise if we've already overcommitted
4679 * lets start flushing stuff first and then come back and try to make
4682 if (used <= space_info->total_bytes) {
4683 if (used + orig_bytes <= space_info->total_bytes) {
4684 space_info->bytes_may_use += orig_bytes;
4685 trace_btrfs_space_reservation(root->fs_info,
4686 "space_info", space_info->flags, orig_bytes, 1);
4690 * Ok set num_bytes to orig_bytes since we aren't
4691 * overocmmitted, this way we only try and reclaim what
4694 num_bytes = orig_bytes;
4698 * Ok we're over committed, set num_bytes to the overcommitted
4699 * amount plus the amount of bytes that we need for this
4702 num_bytes = used - space_info->total_bytes +
4706 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4707 space_info->bytes_may_use += orig_bytes;
4708 trace_btrfs_space_reservation(root->fs_info, "space_info",
4709 space_info->flags, orig_bytes,
4715 * Couldn't make our reservation, save our place so while we're trying
4716 * to reclaim space we can actually use it instead of somebody else
4717 * stealing it from us.
4719 * We make the other tasks wait for the flush only when we can flush
4722 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4724 space_info->flush = 1;
4725 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4728 * We will do the space reservation dance during log replay,
4729 * which means we won't have fs_info->fs_root set, so don't do
4730 * the async reclaim as we will panic.
4732 if (!root->fs_info->log_root_recovering &&
4733 need_do_async_reclaim(space_info, root->fs_info, used) &&
4734 !work_busy(&root->fs_info->async_reclaim_work))
4735 queue_work(system_unbound_wq,
4736 &root->fs_info->async_reclaim_work);
4738 spin_unlock(&space_info->lock);
4740 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4743 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4748 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4749 * would happen. So skip delalloc flush.
4751 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4752 (flush_state == FLUSH_DELALLOC ||
4753 flush_state == FLUSH_DELALLOC_WAIT))
4754 flush_state = ALLOC_CHUNK;
4758 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4759 flush_state < COMMIT_TRANS)
4761 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4762 flush_state <= COMMIT_TRANS)
4766 if (ret == -ENOSPC &&
4767 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4768 struct btrfs_block_rsv *global_rsv =
4769 &root->fs_info->global_block_rsv;
4771 if (block_rsv != global_rsv &&
4772 !block_rsv_use_bytes(global_rsv, orig_bytes))
4776 trace_btrfs_space_reservation(root->fs_info,
4777 "space_info:enospc",
4778 space_info->flags, orig_bytes, 1);
4780 spin_lock(&space_info->lock);
4781 space_info->flush = 0;
4782 wake_up_all(&space_info->wait);
4783 spin_unlock(&space_info->lock);
4788 static struct btrfs_block_rsv *get_block_rsv(
4789 const struct btrfs_trans_handle *trans,
4790 const struct btrfs_root *root)
4792 struct btrfs_block_rsv *block_rsv = NULL;
4794 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4795 block_rsv = trans->block_rsv;
4797 if (root == root->fs_info->csum_root && trans->adding_csums)
4798 block_rsv = trans->block_rsv;
4800 if (root == root->fs_info->uuid_root)
4801 block_rsv = trans->block_rsv;
4804 block_rsv = root->block_rsv;
4807 block_rsv = &root->fs_info->empty_block_rsv;
4812 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4816 spin_lock(&block_rsv->lock);
4817 if (block_rsv->reserved >= num_bytes) {
4818 block_rsv->reserved -= num_bytes;
4819 if (block_rsv->reserved < block_rsv->size)
4820 block_rsv->full = 0;
4823 spin_unlock(&block_rsv->lock);
4827 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4828 u64 num_bytes, int update_size)
4830 spin_lock(&block_rsv->lock);
4831 block_rsv->reserved += num_bytes;
4833 block_rsv->size += num_bytes;
4834 else if (block_rsv->reserved >= block_rsv->size)
4835 block_rsv->full = 1;
4836 spin_unlock(&block_rsv->lock);
4839 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4840 struct btrfs_block_rsv *dest, u64 num_bytes,
4843 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4846 if (global_rsv->space_info != dest->space_info)
4849 spin_lock(&global_rsv->lock);
4850 min_bytes = div_factor(global_rsv->size, min_factor);
4851 if (global_rsv->reserved < min_bytes + num_bytes) {
4852 spin_unlock(&global_rsv->lock);
4855 global_rsv->reserved -= num_bytes;
4856 if (global_rsv->reserved < global_rsv->size)
4857 global_rsv->full = 0;
4858 spin_unlock(&global_rsv->lock);
4860 block_rsv_add_bytes(dest, num_bytes, 1);
4864 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4865 struct btrfs_block_rsv *block_rsv,
4866 struct btrfs_block_rsv *dest, u64 num_bytes)
4868 struct btrfs_space_info *space_info = block_rsv->space_info;
4870 spin_lock(&block_rsv->lock);
4871 if (num_bytes == (u64)-1)
4872 num_bytes = block_rsv->size;
4873 block_rsv->size -= num_bytes;
4874 if (block_rsv->reserved >= block_rsv->size) {
4875 num_bytes = block_rsv->reserved - block_rsv->size;
4876 block_rsv->reserved = block_rsv->size;
4877 block_rsv->full = 1;
4881 spin_unlock(&block_rsv->lock);
4883 if (num_bytes > 0) {
4885 spin_lock(&dest->lock);
4889 bytes_to_add = dest->size - dest->reserved;
4890 bytes_to_add = min(num_bytes, bytes_to_add);
4891 dest->reserved += bytes_to_add;
4892 if (dest->reserved >= dest->size)
4894 num_bytes -= bytes_to_add;
4896 spin_unlock(&dest->lock);
4899 spin_lock(&space_info->lock);
4900 space_info->bytes_may_use -= num_bytes;
4901 trace_btrfs_space_reservation(fs_info, "space_info",
4902 space_info->flags, num_bytes, 0);
4903 spin_unlock(&space_info->lock);
4908 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4909 struct btrfs_block_rsv *dst, u64 num_bytes)
4913 ret = block_rsv_use_bytes(src, num_bytes);
4917 block_rsv_add_bytes(dst, num_bytes, 1);
4921 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4923 memset(rsv, 0, sizeof(*rsv));
4924 spin_lock_init(&rsv->lock);
4928 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4929 unsigned short type)
4931 struct btrfs_block_rsv *block_rsv;
4932 struct btrfs_fs_info *fs_info = root->fs_info;
4934 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4938 btrfs_init_block_rsv(block_rsv, type);
4939 block_rsv->space_info = __find_space_info(fs_info,
4940 BTRFS_BLOCK_GROUP_METADATA);
4944 void btrfs_free_block_rsv(struct btrfs_root *root,
4945 struct btrfs_block_rsv *rsv)
4949 btrfs_block_rsv_release(root, rsv, (u64)-1);
4953 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4958 int btrfs_block_rsv_add(struct btrfs_root *root,
4959 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4960 enum btrfs_reserve_flush_enum flush)
4967 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4969 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4976 int btrfs_block_rsv_check(struct btrfs_root *root,
4977 struct btrfs_block_rsv *block_rsv, int min_factor)
4985 spin_lock(&block_rsv->lock);
4986 num_bytes = div_factor(block_rsv->size, min_factor);
4987 if (block_rsv->reserved >= num_bytes)
4989 spin_unlock(&block_rsv->lock);
4994 int btrfs_block_rsv_refill(struct btrfs_root *root,
4995 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4996 enum btrfs_reserve_flush_enum flush)
5004 spin_lock(&block_rsv->lock);
5005 num_bytes = min_reserved;
5006 if (block_rsv->reserved >= num_bytes)
5009 num_bytes -= block_rsv->reserved;
5010 spin_unlock(&block_rsv->lock);
5015 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5017 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5024 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5025 struct btrfs_block_rsv *dst_rsv,
5028 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5031 void btrfs_block_rsv_release(struct btrfs_root *root,
5032 struct btrfs_block_rsv *block_rsv,
5035 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5036 if (global_rsv == block_rsv ||
5037 block_rsv->space_info != global_rsv->space_info)
5039 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5044 * helper to calculate size of global block reservation.
5045 * the desired value is sum of space used by extent tree,
5046 * checksum tree and root tree
5048 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5050 struct btrfs_space_info *sinfo;
5054 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5056 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5057 spin_lock(&sinfo->lock);
5058 data_used = sinfo->bytes_used;
5059 spin_unlock(&sinfo->lock);
5061 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5062 spin_lock(&sinfo->lock);
5063 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5065 meta_used = sinfo->bytes_used;
5066 spin_unlock(&sinfo->lock);
5068 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5070 num_bytes += div_u64(data_used + meta_used, 50);
5072 if (num_bytes * 3 > meta_used)
5073 num_bytes = div_u64(meta_used, 3);
5075 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5078 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5080 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5081 struct btrfs_space_info *sinfo = block_rsv->space_info;
5084 num_bytes = calc_global_metadata_size(fs_info);
5086 spin_lock(&sinfo->lock);
5087 spin_lock(&block_rsv->lock);
5089 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5091 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5092 sinfo->bytes_reserved + sinfo->bytes_readonly +
5093 sinfo->bytes_may_use;
5095 if (sinfo->total_bytes > num_bytes) {
5096 num_bytes = sinfo->total_bytes - num_bytes;
5097 block_rsv->reserved += num_bytes;
5098 sinfo->bytes_may_use += num_bytes;
5099 trace_btrfs_space_reservation(fs_info, "space_info",
5100 sinfo->flags, num_bytes, 1);
5103 if (block_rsv->reserved >= block_rsv->size) {
5104 num_bytes = block_rsv->reserved - block_rsv->size;
5105 sinfo->bytes_may_use -= num_bytes;
5106 trace_btrfs_space_reservation(fs_info, "space_info",
5107 sinfo->flags, num_bytes, 0);
5108 block_rsv->reserved = block_rsv->size;
5109 block_rsv->full = 1;
5112 spin_unlock(&block_rsv->lock);
5113 spin_unlock(&sinfo->lock);
5116 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5118 struct btrfs_space_info *space_info;
5120 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5121 fs_info->chunk_block_rsv.space_info = space_info;
5123 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5124 fs_info->global_block_rsv.space_info = space_info;
5125 fs_info->delalloc_block_rsv.space_info = space_info;
5126 fs_info->trans_block_rsv.space_info = space_info;
5127 fs_info->empty_block_rsv.space_info = space_info;
5128 fs_info->delayed_block_rsv.space_info = space_info;
5130 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5131 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5132 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5133 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5134 if (fs_info->quota_root)
5135 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5136 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5138 update_global_block_rsv(fs_info);
5141 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5143 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5145 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5146 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5147 WARN_ON(fs_info->trans_block_rsv.size > 0);
5148 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5149 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5150 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5151 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5152 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5155 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5156 struct btrfs_root *root)
5158 if (!trans->block_rsv)
5161 if (!trans->bytes_reserved)
5164 trace_btrfs_space_reservation(root->fs_info, "transaction",
5165 trans->transid, trans->bytes_reserved, 0);
5166 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5167 trans->bytes_reserved = 0;
5171 * To be called after all the new block groups attached to the transaction
5172 * handle have been created (btrfs_create_pending_block_groups()).
5174 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5176 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5178 if (!trans->chunk_bytes_reserved)
5181 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5183 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5184 trans->chunk_bytes_reserved);
5185 trans->chunk_bytes_reserved = 0;
5188 /* Can only return 0 or -ENOSPC */
5189 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5190 struct inode *inode)
5192 struct btrfs_root *root = BTRFS_I(inode)->root;
5193 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5194 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5197 * We need to hold space in order to delete our orphan item once we've
5198 * added it, so this takes the reservation so we can release it later
5199 * when we are truly done with the orphan item.
5201 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5202 trace_btrfs_space_reservation(root->fs_info, "orphan",
5203 btrfs_ino(inode), num_bytes, 1);
5204 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5207 void btrfs_orphan_release_metadata(struct inode *inode)
5209 struct btrfs_root *root = BTRFS_I(inode)->root;
5210 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5211 trace_btrfs_space_reservation(root->fs_info, "orphan",
5212 btrfs_ino(inode), num_bytes, 0);
5213 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5217 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5218 * root: the root of the parent directory
5219 * rsv: block reservation
5220 * items: the number of items that we need do reservation
5221 * qgroup_reserved: used to return the reserved size in qgroup
5223 * This function is used to reserve the space for snapshot/subvolume
5224 * creation and deletion. Those operations are different with the
5225 * common file/directory operations, they change two fs/file trees
5226 * and root tree, the number of items that the qgroup reserves is
5227 * different with the free space reservation. So we can not use
5228 * the space reseravtion mechanism in start_transaction().
5230 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5231 struct btrfs_block_rsv *rsv,
5233 u64 *qgroup_reserved,
5234 bool use_global_rsv)
5238 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5240 if (root->fs_info->quota_enabled) {
5241 /* One for parent inode, two for dir entries */
5242 num_bytes = 3 * root->nodesize;
5243 ret = btrfs_qgroup_reserve(root, num_bytes);
5250 *qgroup_reserved = num_bytes;
5252 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5253 rsv->space_info = __find_space_info(root->fs_info,
5254 BTRFS_BLOCK_GROUP_METADATA);
5255 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5256 BTRFS_RESERVE_FLUSH_ALL);
5258 if (ret == -ENOSPC && use_global_rsv)
5259 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5262 if (*qgroup_reserved)
5263 btrfs_qgroup_free(root, *qgroup_reserved);
5269 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5270 struct btrfs_block_rsv *rsv,
5271 u64 qgroup_reserved)
5273 btrfs_block_rsv_release(root, rsv, (u64)-1);
5277 * drop_outstanding_extent - drop an outstanding extent
5278 * @inode: the inode we're dropping the extent for
5279 * @num_bytes: the number of bytes we're relaseing.
5281 * This is called when we are freeing up an outstanding extent, either called
5282 * after an error or after an extent is written. This will return the number of
5283 * reserved extents that need to be freed. This must be called with
5284 * BTRFS_I(inode)->lock held.
5286 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5288 unsigned drop_inode_space = 0;
5289 unsigned dropped_extents = 0;
5290 unsigned num_extents = 0;
5292 num_extents = (unsigned)div64_u64(num_bytes +
5293 BTRFS_MAX_EXTENT_SIZE - 1,
5294 BTRFS_MAX_EXTENT_SIZE);
5295 ASSERT(num_extents);
5296 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5297 BTRFS_I(inode)->outstanding_extents -= num_extents;
5299 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5300 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5301 &BTRFS_I(inode)->runtime_flags))
5302 drop_inode_space = 1;
5305 * If we have more or the same amount of outsanding extents than we have
5306 * reserved then we need to leave the reserved extents count alone.
5308 if (BTRFS_I(inode)->outstanding_extents >=
5309 BTRFS_I(inode)->reserved_extents)
5310 return drop_inode_space;
5312 dropped_extents = BTRFS_I(inode)->reserved_extents -
5313 BTRFS_I(inode)->outstanding_extents;
5314 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5315 return dropped_extents + drop_inode_space;
5319 * calc_csum_metadata_size - return the amount of metada space that must be
5320 * reserved/free'd for the given bytes.
5321 * @inode: the inode we're manipulating
5322 * @num_bytes: the number of bytes in question
5323 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5325 * This adjusts the number of csum_bytes in the inode and then returns the
5326 * correct amount of metadata that must either be reserved or freed. We
5327 * calculate how many checksums we can fit into one leaf and then divide the
5328 * number of bytes that will need to be checksumed by this value to figure out
5329 * how many checksums will be required. If we are adding bytes then the number
5330 * may go up and we will return the number of additional bytes that must be
5331 * reserved. If it is going down we will return the number of bytes that must
5334 * This must be called with BTRFS_I(inode)->lock held.
5336 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5339 struct btrfs_root *root = BTRFS_I(inode)->root;
5340 u64 old_csums, num_csums;
5342 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5343 BTRFS_I(inode)->csum_bytes == 0)
5346 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5348 BTRFS_I(inode)->csum_bytes += num_bytes;
5350 BTRFS_I(inode)->csum_bytes -= num_bytes;
5351 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5353 /* No change, no need to reserve more */
5354 if (old_csums == num_csums)
5358 return btrfs_calc_trans_metadata_size(root,
5359 num_csums - old_csums);
5361 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5364 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5366 struct btrfs_root *root = BTRFS_I(inode)->root;
5367 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5370 unsigned nr_extents = 0;
5371 int extra_reserve = 0;
5372 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5374 bool delalloc_lock = true;
5378 /* If we are a free space inode we need to not flush since we will be in
5379 * the middle of a transaction commit. We also don't need the delalloc
5380 * mutex since we won't race with anybody. We need this mostly to make
5381 * lockdep shut its filthy mouth.
5383 if (btrfs_is_free_space_inode(inode)) {
5384 flush = BTRFS_RESERVE_NO_FLUSH;
5385 delalloc_lock = false;
5388 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5389 btrfs_transaction_in_commit(root->fs_info))
5390 schedule_timeout(1);
5393 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5395 num_bytes = ALIGN(num_bytes, root->sectorsize);
5397 spin_lock(&BTRFS_I(inode)->lock);
5398 nr_extents = (unsigned)div64_u64(num_bytes +
5399 BTRFS_MAX_EXTENT_SIZE - 1,
5400 BTRFS_MAX_EXTENT_SIZE);
5401 BTRFS_I(inode)->outstanding_extents += nr_extents;
5404 if (BTRFS_I(inode)->outstanding_extents >
5405 BTRFS_I(inode)->reserved_extents)
5406 nr_extents = BTRFS_I(inode)->outstanding_extents -
5407 BTRFS_I(inode)->reserved_extents;
5410 * Add an item to reserve for updating the inode when we complete the
5413 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5414 &BTRFS_I(inode)->runtime_flags)) {
5419 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5420 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5421 csum_bytes = BTRFS_I(inode)->csum_bytes;
5422 spin_unlock(&BTRFS_I(inode)->lock);
5424 if (root->fs_info->quota_enabled) {
5425 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5430 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5431 if (unlikely(ret)) {
5432 if (root->fs_info->quota_enabled)
5433 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5437 spin_lock(&BTRFS_I(inode)->lock);
5438 if (extra_reserve) {
5439 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5440 &BTRFS_I(inode)->runtime_flags);
5443 BTRFS_I(inode)->reserved_extents += nr_extents;
5444 spin_unlock(&BTRFS_I(inode)->lock);
5447 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5450 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5451 btrfs_ino(inode), to_reserve, 1);
5452 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5457 spin_lock(&BTRFS_I(inode)->lock);
5458 dropped = drop_outstanding_extent(inode, num_bytes);
5460 * If the inodes csum_bytes is the same as the original
5461 * csum_bytes then we know we haven't raced with any free()ers
5462 * so we can just reduce our inodes csum bytes and carry on.
5464 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5465 calc_csum_metadata_size(inode, num_bytes, 0);
5467 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5471 * This is tricky, but first we need to figure out how much we
5472 * free'd from any free-ers that occured during this
5473 * reservation, so we reset ->csum_bytes to the csum_bytes
5474 * before we dropped our lock, and then call the free for the
5475 * number of bytes that were freed while we were trying our
5478 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5479 BTRFS_I(inode)->csum_bytes = csum_bytes;
5480 to_free = calc_csum_metadata_size(inode, bytes, 0);
5484 * Now we need to see how much we would have freed had we not
5485 * been making this reservation and our ->csum_bytes were not
5486 * artificially inflated.
5488 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5489 bytes = csum_bytes - orig_csum_bytes;
5490 bytes = calc_csum_metadata_size(inode, bytes, 0);
5493 * Now reset ->csum_bytes to what it should be. If bytes is
5494 * more than to_free then we would have free'd more space had we
5495 * not had an artificially high ->csum_bytes, so we need to free
5496 * the remainder. If bytes is the same or less then we don't
5497 * need to do anything, the other free-ers did the correct
5500 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5501 if (bytes > to_free)
5502 to_free = bytes - to_free;
5506 spin_unlock(&BTRFS_I(inode)->lock);
5508 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5511 btrfs_block_rsv_release(root, block_rsv, to_free);
5512 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5513 btrfs_ino(inode), to_free, 0);
5516 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5521 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5522 * @inode: the inode to release the reservation for
5523 * @num_bytes: the number of bytes we're releasing
5525 * This will release the metadata reservation for an inode. This can be called
5526 * once we complete IO for a given set of bytes to release their metadata
5529 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5531 struct btrfs_root *root = BTRFS_I(inode)->root;
5535 num_bytes = ALIGN(num_bytes, root->sectorsize);
5536 spin_lock(&BTRFS_I(inode)->lock);
5537 dropped = drop_outstanding_extent(inode, num_bytes);
5540 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5541 spin_unlock(&BTRFS_I(inode)->lock);
5543 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5545 if (btrfs_test_is_dummy_root(root))
5548 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5549 btrfs_ino(inode), to_free, 0);
5551 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5556 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5557 * @inode: inode we're writing to
5558 * @num_bytes: the number of bytes we want to allocate
5560 * This will do the following things
5562 * o reserve space in the data space info for num_bytes
5563 * o reserve space in the metadata space info based on number of outstanding
5564 * extents and how much csums will be needed
5565 * o add to the inodes ->delalloc_bytes
5566 * o add it to the fs_info's delalloc inodes list.
5568 * This will return 0 for success and -ENOSPC if there is no space left.
5570 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5574 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5578 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5580 btrfs_free_reserved_data_space(inode, num_bytes);
5588 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5589 * @inode: inode we're releasing space for
5590 * @num_bytes: the number of bytes we want to free up
5592 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5593 * called in the case that we don't need the metadata AND data reservations
5594 * anymore. So if there is an error or we insert an inline extent.
5596 * This function will release the metadata space that was not used and will
5597 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5598 * list if there are no delalloc bytes left.
5600 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5602 btrfs_delalloc_release_metadata(inode, num_bytes);
5603 btrfs_free_reserved_data_space(inode, num_bytes);
5606 static int update_block_group(struct btrfs_trans_handle *trans,
5607 struct btrfs_root *root, u64 bytenr,
5608 u64 num_bytes, int alloc)
5610 struct btrfs_block_group_cache *cache = NULL;
5611 struct btrfs_fs_info *info = root->fs_info;
5612 u64 total = num_bytes;
5617 /* block accounting for super block */
5618 spin_lock(&info->delalloc_root_lock);
5619 old_val = btrfs_super_bytes_used(info->super_copy);
5621 old_val += num_bytes;
5623 old_val -= num_bytes;
5624 btrfs_set_super_bytes_used(info->super_copy, old_val);
5625 spin_unlock(&info->delalloc_root_lock);
5628 cache = btrfs_lookup_block_group(info, bytenr);
5631 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5632 BTRFS_BLOCK_GROUP_RAID1 |
5633 BTRFS_BLOCK_GROUP_RAID10))
5638 * If this block group has free space cache written out, we
5639 * need to make sure to load it if we are removing space. This
5640 * is because we need the unpinning stage to actually add the
5641 * space back to the block group, otherwise we will leak space.
5643 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5644 cache_block_group(cache, 1);
5646 byte_in_group = bytenr - cache->key.objectid;
5647 WARN_ON(byte_in_group > cache->key.offset);
5649 spin_lock(&cache->space_info->lock);
5650 spin_lock(&cache->lock);
5652 if (btrfs_test_opt(root, SPACE_CACHE) &&
5653 cache->disk_cache_state < BTRFS_DC_CLEAR)
5654 cache->disk_cache_state = BTRFS_DC_CLEAR;
5656 old_val = btrfs_block_group_used(&cache->item);
5657 num_bytes = min(total, cache->key.offset - byte_in_group);
5659 old_val += num_bytes;
5660 btrfs_set_block_group_used(&cache->item, old_val);
5661 cache->reserved -= num_bytes;
5662 cache->space_info->bytes_reserved -= num_bytes;
5663 cache->space_info->bytes_used += num_bytes;
5664 cache->space_info->disk_used += num_bytes * factor;
5665 spin_unlock(&cache->lock);
5666 spin_unlock(&cache->space_info->lock);
5668 old_val -= num_bytes;
5669 btrfs_set_block_group_used(&cache->item, old_val);
5670 cache->pinned += num_bytes;
5671 cache->space_info->bytes_pinned += num_bytes;
5672 cache->space_info->bytes_used -= num_bytes;
5673 cache->space_info->disk_used -= num_bytes * factor;
5674 spin_unlock(&cache->lock);
5675 spin_unlock(&cache->space_info->lock);
5677 set_extent_dirty(info->pinned_extents,
5678 bytenr, bytenr + num_bytes - 1,
5679 GFP_NOFS | __GFP_NOFAIL);
5681 * No longer have used bytes in this block group, queue
5685 spin_lock(&info->unused_bgs_lock);
5686 if (list_empty(&cache->bg_list)) {
5687 btrfs_get_block_group(cache);
5688 list_add_tail(&cache->bg_list,
5691 spin_unlock(&info->unused_bgs_lock);
5695 spin_lock(&trans->transaction->dirty_bgs_lock);
5696 if (list_empty(&cache->dirty_list)) {
5697 list_add_tail(&cache->dirty_list,
5698 &trans->transaction->dirty_bgs);
5699 trans->transaction->num_dirty_bgs++;
5700 btrfs_get_block_group(cache);
5702 spin_unlock(&trans->transaction->dirty_bgs_lock);
5704 btrfs_put_block_group(cache);
5706 bytenr += num_bytes;
5711 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5713 struct btrfs_block_group_cache *cache;
5716 spin_lock(&root->fs_info->block_group_cache_lock);
5717 bytenr = root->fs_info->first_logical_byte;
5718 spin_unlock(&root->fs_info->block_group_cache_lock);
5720 if (bytenr < (u64)-1)
5723 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5727 bytenr = cache->key.objectid;
5728 btrfs_put_block_group(cache);
5733 static int pin_down_extent(struct btrfs_root *root,
5734 struct btrfs_block_group_cache *cache,
5735 u64 bytenr, u64 num_bytes, int reserved)
5737 spin_lock(&cache->space_info->lock);
5738 spin_lock(&cache->lock);
5739 cache->pinned += num_bytes;
5740 cache->space_info->bytes_pinned += num_bytes;
5742 cache->reserved -= num_bytes;
5743 cache->space_info->bytes_reserved -= num_bytes;
5745 spin_unlock(&cache->lock);
5746 spin_unlock(&cache->space_info->lock);
5748 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5749 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5751 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5756 * this function must be called within transaction
5758 int btrfs_pin_extent(struct btrfs_root *root,
5759 u64 bytenr, u64 num_bytes, int reserved)
5761 struct btrfs_block_group_cache *cache;
5763 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5764 BUG_ON(!cache); /* Logic error */
5766 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5768 btrfs_put_block_group(cache);
5773 * this function must be called within transaction
5775 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5776 u64 bytenr, u64 num_bytes)
5778 struct btrfs_block_group_cache *cache;
5781 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5786 * pull in the free space cache (if any) so that our pin
5787 * removes the free space from the cache. We have load_only set
5788 * to one because the slow code to read in the free extents does check
5789 * the pinned extents.
5791 cache_block_group(cache, 1);
5793 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5795 /* remove us from the free space cache (if we're there at all) */
5796 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5797 btrfs_put_block_group(cache);
5801 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5804 struct btrfs_block_group_cache *block_group;
5805 struct btrfs_caching_control *caching_ctl;
5807 block_group = btrfs_lookup_block_group(root->fs_info, start);
5811 cache_block_group(block_group, 0);
5812 caching_ctl = get_caching_control(block_group);
5816 BUG_ON(!block_group_cache_done(block_group));
5817 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5819 mutex_lock(&caching_ctl->mutex);
5821 if (start >= caching_ctl->progress) {
5822 ret = add_excluded_extent(root, start, num_bytes);
5823 } else if (start + num_bytes <= caching_ctl->progress) {
5824 ret = btrfs_remove_free_space(block_group,
5827 num_bytes = caching_ctl->progress - start;
5828 ret = btrfs_remove_free_space(block_group,
5833 num_bytes = (start + num_bytes) -
5834 caching_ctl->progress;
5835 start = caching_ctl->progress;
5836 ret = add_excluded_extent(root, start, num_bytes);
5839 mutex_unlock(&caching_ctl->mutex);
5840 put_caching_control(caching_ctl);
5842 btrfs_put_block_group(block_group);
5846 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5847 struct extent_buffer *eb)
5849 struct btrfs_file_extent_item *item;
5850 struct btrfs_key key;
5854 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5857 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5858 btrfs_item_key_to_cpu(eb, &key, i);
5859 if (key.type != BTRFS_EXTENT_DATA_KEY)
5861 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5862 found_type = btrfs_file_extent_type(eb, item);
5863 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5865 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5867 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5868 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5869 __exclude_logged_extent(log, key.objectid, key.offset);
5876 * btrfs_update_reserved_bytes - update the block_group and space info counters
5877 * @cache: The cache we are manipulating
5878 * @num_bytes: The number of bytes in question
5879 * @reserve: One of the reservation enums
5880 * @delalloc: The blocks are allocated for the delalloc write
5882 * This is called by the allocator when it reserves space, or by somebody who is
5883 * freeing space that was never actually used on disk. For example if you
5884 * reserve some space for a new leaf in transaction A and before transaction A
5885 * commits you free that leaf, you call this with reserve set to 0 in order to
5886 * clear the reservation.
5888 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5889 * ENOSPC accounting. For data we handle the reservation through clearing the
5890 * delalloc bits in the io_tree. We have to do this since we could end up
5891 * allocating less disk space for the amount of data we have reserved in the
5892 * case of compression.
5894 * If this is a reservation and the block group has become read only we cannot
5895 * make the reservation and return -EAGAIN, otherwise this function always
5898 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5899 u64 num_bytes, int reserve, int delalloc)
5901 struct btrfs_space_info *space_info = cache->space_info;
5904 spin_lock(&space_info->lock);
5905 spin_lock(&cache->lock);
5906 if (reserve != RESERVE_FREE) {
5910 cache->reserved += num_bytes;
5911 space_info->bytes_reserved += num_bytes;
5912 if (reserve == RESERVE_ALLOC) {
5913 trace_btrfs_space_reservation(cache->fs_info,
5914 "space_info", space_info->flags,
5916 space_info->bytes_may_use -= num_bytes;
5920 cache->delalloc_bytes += num_bytes;
5924 space_info->bytes_readonly += num_bytes;
5925 cache->reserved -= num_bytes;
5926 space_info->bytes_reserved -= num_bytes;
5929 cache->delalloc_bytes -= num_bytes;
5931 spin_unlock(&cache->lock);
5932 spin_unlock(&space_info->lock);
5936 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5937 struct btrfs_root *root)
5939 struct btrfs_fs_info *fs_info = root->fs_info;
5940 struct btrfs_caching_control *next;
5941 struct btrfs_caching_control *caching_ctl;
5942 struct btrfs_block_group_cache *cache;
5944 down_write(&fs_info->commit_root_sem);
5946 list_for_each_entry_safe(caching_ctl, next,
5947 &fs_info->caching_block_groups, list) {
5948 cache = caching_ctl->block_group;
5949 if (block_group_cache_done(cache)) {
5950 cache->last_byte_to_unpin = (u64)-1;
5951 list_del_init(&caching_ctl->list);
5952 put_caching_control(caching_ctl);
5954 cache->last_byte_to_unpin = caching_ctl->progress;
5958 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5959 fs_info->pinned_extents = &fs_info->freed_extents[1];
5961 fs_info->pinned_extents = &fs_info->freed_extents[0];
5963 up_write(&fs_info->commit_root_sem);
5965 update_global_block_rsv(fs_info);
5968 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5969 const bool return_free_space)
5971 struct btrfs_fs_info *fs_info = root->fs_info;
5972 struct btrfs_block_group_cache *cache = NULL;
5973 struct btrfs_space_info *space_info;
5974 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5978 while (start <= end) {
5981 start >= cache->key.objectid + cache->key.offset) {
5983 btrfs_put_block_group(cache);
5984 cache = btrfs_lookup_block_group(fs_info, start);
5985 BUG_ON(!cache); /* Logic error */
5988 len = cache->key.objectid + cache->key.offset - start;
5989 len = min(len, end + 1 - start);
5991 if (start < cache->last_byte_to_unpin) {
5992 len = min(len, cache->last_byte_to_unpin - start);
5993 if (return_free_space)
5994 btrfs_add_free_space(cache, start, len);
5998 space_info = cache->space_info;
6000 spin_lock(&space_info->lock);
6001 spin_lock(&cache->lock);
6002 cache->pinned -= len;
6003 space_info->bytes_pinned -= len;
6004 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6006 space_info->bytes_readonly += len;
6009 spin_unlock(&cache->lock);
6010 if (!readonly && global_rsv->space_info == space_info) {
6011 spin_lock(&global_rsv->lock);
6012 if (!global_rsv->full) {
6013 len = min(len, global_rsv->size -
6014 global_rsv->reserved);
6015 global_rsv->reserved += len;
6016 space_info->bytes_may_use += len;
6017 if (global_rsv->reserved >= global_rsv->size)
6018 global_rsv->full = 1;
6020 spin_unlock(&global_rsv->lock);
6022 spin_unlock(&space_info->lock);
6026 btrfs_put_block_group(cache);
6030 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6031 struct btrfs_root *root)
6033 struct btrfs_fs_info *fs_info = root->fs_info;
6034 struct extent_io_tree *unpin;
6042 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6043 unpin = &fs_info->freed_extents[1];
6045 unpin = &fs_info->freed_extents[0];
6048 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6049 ret = find_first_extent_bit(unpin, 0, &start, &end,
6050 EXTENT_DIRTY, NULL);
6052 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6056 if (btrfs_test_opt(root, DISCARD))
6057 ret = btrfs_discard_extent(root, start,
6058 end + 1 - start, NULL);
6060 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6061 unpin_extent_range(root, start, end, true);
6062 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6069 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6070 u64 owner, u64 root_objectid)
6072 struct btrfs_space_info *space_info;
6075 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6076 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6077 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6079 flags = BTRFS_BLOCK_GROUP_METADATA;
6081 flags = BTRFS_BLOCK_GROUP_DATA;
6084 space_info = __find_space_info(fs_info, flags);
6085 BUG_ON(!space_info); /* Logic bug */
6086 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6090 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6091 struct btrfs_root *root,
6092 struct btrfs_delayed_ref_node *node, u64 parent,
6093 u64 root_objectid, u64 owner_objectid,
6094 u64 owner_offset, int refs_to_drop,
6095 struct btrfs_delayed_extent_op *extent_op)
6097 struct btrfs_key key;
6098 struct btrfs_path *path;
6099 struct btrfs_fs_info *info = root->fs_info;
6100 struct btrfs_root *extent_root = info->extent_root;
6101 struct extent_buffer *leaf;
6102 struct btrfs_extent_item *ei;
6103 struct btrfs_extent_inline_ref *iref;
6106 int extent_slot = 0;
6107 int found_extent = 0;
6109 int no_quota = node->no_quota;
6112 u64 bytenr = node->bytenr;
6113 u64 num_bytes = node->num_bytes;
6115 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6116 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6119 if (!info->quota_enabled || !is_fstree(root_objectid))
6122 path = btrfs_alloc_path();
6127 path->leave_spinning = 1;
6129 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6130 BUG_ON(!is_data && refs_to_drop != 1);
6133 skinny_metadata = 0;
6135 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6136 bytenr, num_bytes, parent,
6137 root_objectid, owner_objectid,
6140 extent_slot = path->slots[0];
6141 while (extent_slot >= 0) {
6142 btrfs_item_key_to_cpu(path->nodes[0], &key,
6144 if (key.objectid != bytenr)
6146 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6147 key.offset == num_bytes) {
6151 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6152 key.offset == owner_objectid) {
6156 if (path->slots[0] - extent_slot > 5)
6160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6161 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6162 if (found_extent && item_size < sizeof(*ei))
6165 if (!found_extent) {
6167 ret = remove_extent_backref(trans, extent_root, path,
6169 is_data, &last_ref);
6171 btrfs_abort_transaction(trans, extent_root, ret);
6174 btrfs_release_path(path);
6175 path->leave_spinning = 1;
6177 key.objectid = bytenr;
6178 key.type = BTRFS_EXTENT_ITEM_KEY;
6179 key.offset = num_bytes;
6181 if (!is_data && skinny_metadata) {
6182 key.type = BTRFS_METADATA_ITEM_KEY;
6183 key.offset = owner_objectid;
6186 ret = btrfs_search_slot(trans, extent_root,
6188 if (ret > 0 && skinny_metadata && path->slots[0]) {
6190 * Couldn't find our skinny metadata item,
6191 * see if we have ye olde extent item.
6194 btrfs_item_key_to_cpu(path->nodes[0], &key,
6196 if (key.objectid == bytenr &&
6197 key.type == BTRFS_EXTENT_ITEM_KEY &&
6198 key.offset == num_bytes)
6202 if (ret > 0 && skinny_metadata) {
6203 skinny_metadata = false;
6204 key.objectid = bytenr;
6205 key.type = BTRFS_EXTENT_ITEM_KEY;
6206 key.offset = num_bytes;
6207 btrfs_release_path(path);
6208 ret = btrfs_search_slot(trans, extent_root,
6213 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6216 btrfs_print_leaf(extent_root,
6220 btrfs_abort_transaction(trans, extent_root, ret);
6223 extent_slot = path->slots[0];
6225 } else if (WARN_ON(ret == -ENOENT)) {
6226 btrfs_print_leaf(extent_root, path->nodes[0]);
6228 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6229 bytenr, parent, root_objectid, owner_objectid,
6231 btrfs_abort_transaction(trans, extent_root, ret);
6234 btrfs_abort_transaction(trans, extent_root, ret);
6238 leaf = path->nodes[0];
6239 item_size = btrfs_item_size_nr(leaf, extent_slot);
6240 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6241 if (item_size < sizeof(*ei)) {
6242 BUG_ON(found_extent || extent_slot != path->slots[0]);
6243 ret = convert_extent_item_v0(trans, extent_root, path,
6246 btrfs_abort_transaction(trans, extent_root, ret);
6250 btrfs_release_path(path);
6251 path->leave_spinning = 1;
6253 key.objectid = bytenr;
6254 key.type = BTRFS_EXTENT_ITEM_KEY;
6255 key.offset = num_bytes;
6257 ret = btrfs_search_slot(trans, extent_root, &key, path,
6260 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6262 btrfs_print_leaf(extent_root, path->nodes[0]);
6265 btrfs_abort_transaction(trans, extent_root, ret);
6269 extent_slot = path->slots[0];
6270 leaf = path->nodes[0];
6271 item_size = btrfs_item_size_nr(leaf, extent_slot);
6274 BUG_ON(item_size < sizeof(*ei));
6275 ei = btrfs_item_ptr(leaf, extent_slot,
6276 struct btrfs_extent_item);
6277 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6278 key.type == BTRFS_EXTENT_ITEM_KEY) {
6279 struct btrfs_tree_block_info *bi;
6280 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6281 bi = (struct btrfs_tree_block_info *)(ei + 1);
6282 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6285 refs = btrfs_extent_refs(leaf, ei);
6286 if (refs < refs_to_drop) {
6287 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6288 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6290 btrfs_abort_transaction(trans, extent_root, ret);
6293 refs -= refs_to_drop;
6296 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6298 __run_delayed_extent_op(extent_op, leaf, ei);
6300 * In the case of inline back ref, reference count will
6301 * be updated by remove_extent_backref
6304 BUG_ON(!found_extent);
6306 btrfs_set_extent_refs(leaf, ei, refs);
6307 btrfs_mark_buffer_dirty(leaf);
6310 ret = remove_extent_backref(trans, extent_root, path,
6312 is_data, &last_ref);
6314 btrfs_abort_transaction(trans, extent_root, ret);
6318 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6322 BUG_ON(is_data && refs_to_drop !=
6323 extent_data_ref_count(root, path, iref));
6325 BUG_ON(path->slots[0] != extent_slot);
6327 BUG_ON(path->slots[0] != extent_slot + 1);
6328 path->slots[0] = extent_slot;
6334 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6337 btrfs_abort_transaction(trans, extent_root, ret);
6340 btrfs_release_path(path);
6343 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6345 btrfs_abort_transaction(trans, extent_root, ret);
6350 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6352 btrfs_abort_transaction(trans, extent_root, ret);
6356 btrfs_release_path(path);
6359 btrfs_free_path(path);
6364 * when we free an block, it is possible (and likely) that we free the last
6365 * delayed ref for that extent as well. This searches the delayed ref tree for
6366 * a given extent, and if there are no other delayed refs to be processed, it
6367 * removes it from the tree.
6369 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6370 struct btrfs_root *root, u64 bytenr)
6372 struct btrfs_delayed_ref_head *head;
6373 struct btrfs_delayed_ref_root *delayed_refs;
6376 delayed_refs = &trans->transaction->delayed_refs;
6377 spin_lock(&delayed_refs->lock);
6378 head = btrfs_find_delayed_ref_head(trans, bytenr);
6380 goto out_delayed_unlock;
6382 spin_lock(&head->lock);
6383 if (!list_empty(&head->ref_list))
6386 if (head->extent_op) {
6387 if (!head->must_insert_reserved)
6389 btrfs_free_delayed_extent_op(head->extent_op);
6390 head->extent_op = NULL;
6394 * waiting for the lock here would deadlock. If someone else has it
6395 * locked they are already in the process of dropping it anyway
6397 if (!mutex_trylock(&head->mutex))
6401 * at this point we have a head with no other entries. Go
6402 * ahead and process it.
6404 head->node.in_tree = 0;
6405 rb_erase(&head->href_node, &delayed_refs->href_root);
6407 atomic_dec(&delayed_refs->num_entries);
6410 * we don't take a ref on the node because we're removing it from the
6411 * tree, so we just steal the ref the tree was holding.
6413 delayed_refs->num_heads--;
6414 if (head->processing == 0)
6415 delayed_refs->num_heads_ready--;
6416 head->processing = 0;
6417 spin_unlock(&head->lock);
6418 spin_unlock(&delayed_refs->lock);
6420 BUG_ON(head->extent_op);
6421 if (head->must_insert_reserved)
6424 mutex_unlock(&head->mutex);
6425 btrfs_put_delayed_ref(&head->node);
6428 spin_unlock(&head->lock);
6431 spin_unlock(&delayed_refs->lock);
6435 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6436 struct btrfs_root *root,
6437 struct extent_buffer *buf,
6438 u64 parent, int last_ref)
6443 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6444 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6445 buf->start, buf->len,
6446 parent, root->root_key.objectid,
6447 btrfs_header_level(buf),
6448 BTRFS_DROP_DELAYED_REF, NULL, 0);
6449 BUG_ON(ret); /* -ENOMEM */
6455 if (btrfs_header_generation(buf) == trans->transid) {
6456 struct btrfs_block_group_cache *cache;
6458 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6459 ret = check_ref_cleanup(trans, root, buf->start);
6464 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6466 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6467 pin_down_extent(root, cache, buf->start, buf->len, 1);
6468 btrfs_put_block_group(cache);
6472 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6474 btrfs_add_free_space(cache, buf->start, buf->len);
6475 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6476 btrfs_put_block_group(cache);
6477 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6482 add_pinned_bytes(root->fs_info, buf->len,
6483 btrfs_header_level(buf),
6484 root->root_key.objectid);
6487 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6490 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6493 /* Can return -ENOMEM */
6494 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6495 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6496 u64 owner, u64 offset, int no_quota)
6499 struct btrfs_fs_info *fs_info = root->fs_info;
6501 if (btrfs_test_is_dummy_root(root))
6504 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6507 * tree log blocks never actually go into the extent allocation
6508 * tree, just update pinning info and exit early.
6510 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6511 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6512 /* unlocks the pinned mutex */
6513 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6515 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6516 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6518 parent, root_objectid, (int)owner,
6519 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6521 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6523 parent, root_objectid, owner,
6524 offset, BTRFS_DROP_DELAYED_REF,
6531 * when we wait for progress in the block group caching, its because
6532 * our allocation attempt failed at least once. So, we must sleep
6533 * and let some progress happen before we try again.
6535 * This function will sleep at least once waiting for new free space to
6536 * show up, and then it will check the block group free space numbers
6537 * for our min num_bytes. Another option is to have it go ahead
6538 * and look in the rbtree for a free extent of a given size, but this
6541 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6542 * any of the information in this block group.
6544 static noinline void
6545 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6548 struct btrfs_caching_control *caching_ctl;
6550 caching_ctl = get_caching_control(cache);
6554 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6555 (cache->free_space_ctl->free_space >= num_bytes));
6557 put_caching_control(caching_ctl);
6561 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6563 struct btrfs_caching_control *caching_ctl;
6566 caching_ctl = get_caching_control(cache);
6568 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6570 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6571 if (cache->cached == BTRFS_CACHE_ERROR)
6573 put_caching_control(caching_ctl);
6577 int __get_raid_index(u64 flags)
6579 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6580 return BTRFS_RAID_RAID10;
6581 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6582 return BTRFS_RAID_RAID1;
6583 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6584 return BTRFS_RAID_DUP;
6585 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6586 return BTRFS_RAID_RAID0;
6587 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6588 return BTRFS_RAID_RAID5;
6589 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6590 return BTRFS_RAID_RAID6;
6592 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6595 int get_block_group_index(struct btrfs_block_group_cache *cache)
6597 return __get_raid_index(cache->flags);
6600 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6601 [BTRFS_RAID_RAID10] = "raid10",
6602 [BTRFS_RAID_RAID1] = "raid1",
6603 [BTRFS_RAID_DUP] = "dup",
6604 [BTRFS_RAID_RAID0] = "raid0",
6605 [BTRFS_RAID_SINGLE] = "single",
6606 [BTRFS_RAID_RAID5] = "raid5",
6607 [BTRFS_RAID_RAID6] = "raid6",
6610 static const char *get_raid_name(enum btrfs_raid_types type)
6612 if (type >= BTRFS_NR_RAID_TYPES)
6615 return btrfs_raid_type_names[type];
6618 enum btrfs_loop_type {
6619 LOOP_CACHING_NOWAIT = 0,
6620 LOOP_CACHING_WAIT = 1,
6621 LOOP_ALLOC_CHUNK = 2,
6622 LOOP_NO_EMPTY_SIZE = 3,
6626 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6630 down_read(&cache->data_rwsem);
6634 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6637 btrfs_get_block_group(cache);
6639 down_read(&cache->data_rwsem);
6642 static struct btrfs_block_group_cache *
6643 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6644 struct btrfs_free_cluster *cluster,
6647 struct btrfs_block_group_cache *used_bg;
6648 bool locked = false;
6650 spin_lock(&cluster->refill_lock);
6652 if (used_bg == cluster->block_group)
6655 up_read(&used_bg->data_rwsem);
6656 btrfs_put_block_group(used_bg);
6659 used_bg = cluster->block_group;
6663 if (used_bg == block_group)
6666 btrfs_get_block_group(used_bg);
6671 if (down_read_trylock(&used_bg->data_rwsem))
6674 spin_unlock(&cluster->refill_lock);
6675 down_read(&used_bg->data_rwsem);
6681 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6685 up_read(&cache->data_rwsem);
6686 btrfs_put_block_group(cache);
6690 * walks the btree of allocated extents and find a hole of a given size.
6691 * The key ins is changed to record the hole:
6692 * ins->objectid == start position
6693 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6694 * ins->offset == the size of the hole.
6695 * Any available blocks before search_start are skipped.
6697 * If there is no suitable free space, we will record the max size of
6698 * the free space extent currently.
6700 static noinline int find_free_extent(struct btrfs_root *orig_root,
6701 u64 num_bytes, u64 empty_size,
6702 u64 hint_byte, struct btrfs_key *ins,
6703 u64 flags, int delalloc)
6706 struct btrfs_root *root = orig_root->fs_info->extent_root;
6707 struct btrfs_free_cluster *last_ptr = NULL;
6708 struct btrfs_block_group_cache *block_group = NULL;
6709 u64 search_start = 0;
6710 u64 max_extent_size = 0;
6711 int empty_cluster = 2 * 1024 * 1024;
6712 struct btrfs_space_info *space_info;
6714 int index = __get_raid_index(flags);
6715 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6716 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6717 bool failed_cluster_refill = false;
6718 bool failed_alloc = false;
6719 bool use_cluster = true;
6720 bool have_caching_bg = false;
6722 WARN_ON(num_bytes < root->sectorsize);
6723 ins->type = BTRFS_EXTENT_ITEM_KEY;
6727 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6729 space_info = __find_space_info(root->fs_info, flags);
6731 btrfs_err(root->fs_info, "No space info for %llu", flags);
6736 * If the space info is for both data and metadata it means we have a
6737 * small filesystem and we can't use the clustering stuff.
6739 if (btrfs_mixed_space_info(space_info))
6740 use_cluster = false;
6742 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6743 last_ptr = &root->fs_info->meta_alloc_cluster;
6744 if (!btrfs_test_opt(root, SSD))
6745 empty_cluster = 64 * 1024;
6748 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6749 btrfs_test_opt(root, SSD)) {
6750 last_ptr = &root->fs_info->data_alloc_cluster;
6754 spin_lock(&last_ptr->lock);
6755 if (last_ptr->block_group)
6756 hint_byte = last_ptr->window_start;
6757 spin_unlock(&last_ptr->lock);
6760 search_start = max(search_start, first_logical_byte(root, 0));
6761 search_start = max(search_start, hint_byte);
6766 if (search_start == hint_byte) {
6767 block_group = btrfs_lookup_block_group(root->fs_info,
6770 * we don't want to use the block group if it doesn't match our
6771 * allocation bits, or if its not cached.
6773 * However if we are re-searching with an ideal block group
6774 * picked out then we don't care that the block group is cached.
6776 if (block_group && block_group_bits(block_group, flags) &&
6777 block_group->cached != BTRFS_CACHE_NO) {
6778 down_read(&space_info->groups_sem);
6779 if (list_empty(&block_group->list) ||
6782 * someone is removing this block group,
6783 * we can't jump into the have_block_group
6784 * target because our list pointers are not
6787 btrfs_put_block_group(block_group);
6788 up_read(&space_info->groups_sem);
6790 index = get_block_group_index(block_group);
6791 btrfs_lock_block_group(block_group, delalloc);
6792 goto have_block_group;
6794 } else if (block_group) {
6795 btrfs_put_block_group(block_group);
6799 have_caching_bg = false;
6800 down_read(&space_info->groups_sem);
6801 list_for_each_entry(block_group, &space_info->block_groups[index],
6806 btrfs_grab_block_group(block_group, delalloc);
6807 search_start = block_group->key.objectid;
6810 * this can happen if we end up cycling through all the
6811 * raid types, but we want to make sure we only allocate
6812 * for the proper type.
6814 if (!block_group_bits(block_group, flags)) {
6815 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6816 BTRFS_BLOCK_GROUP_RAID1 |
6817 BTRFS_BLOCK_GROUP_RAID5 |
6818 BTRFS_BLOCK_GROUP_RAID6 |
6819 BTRFS_BLOCK_GROUP_RAID10;
6822 * if they asked for extra copies and this block group
6823 * doesn't provide them, bail. This does allow us to
6824 * fill raid0 from raid1.
6826 if ((flags & extra) && !(block_group->flags & extra))
6831 cached = block_group_cache_done(block_group);
6832 if (unlikely(!cached)) {
6833 ret = cache_block_group(block_group, 0);
6838 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6840 if (unlikely(block_group->ro))
6844 * Ok we want to try and use the cluster allocator, so
6848 struct btrfs_block_group_cache *used_block_group;
6849 unsigned long aligned_cluster;
6851 * the refill lock keeps out other
6852 * people trying to start a new cluster
6854 used_block_group = btrfs_lock_cluster(block_group,
6857 if (!used_block_group)
6858 goto refill_cluster;
6860 if (used_block_group != block_group &&
6861 (used_block_group->ro ||
6862 !block_group_bits(used_block_group, flags)))
6863 goto release_cluster;
6865 offset = btrfs_alloc_from_cluster(used_block_group,
6868 used_block_group->key.objectid,
6871 /* we have a block, we're done */
6872 spin_unlock(&last_ptr->refill_lock);
6873 trace_btrfs_reserve_extent_cluster(root,
6875 search_start, num_bytes);
6876 if (used_block_group != block_group) {
6877 btrfs_release_block_group(block_group,
6879 block_group = used_block_group;
6884 WARN_ON(last_ptr->block_group != used_block_group);
6886 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6887 * set up a new clusters, so lets just skip it
6888 * and let the allocator find whatever block
6889 * it can find. If we reach this point, we
6890 * will have tried the cluster allocator
6891 * plenty of times and not have found
6892 * anything, so we are likely way too
6893 * fragmented for the clustering stuff to find
6896 * However, if the cluster is taken from the
6897 * current block group, release the cluster
6898 * first, so that we stand a better chance of
6899 * succeeding in the unclustered
6901 if (loop >= LOOP_NO_EMPTY_SIZE &&
6902 used_block_group != block_group) {
6903 spin_unlock(&last_ptr->refill_lock);
6904 btrfs_release_block_group(used_block_group,
6906 goto unclustered_alloc;
6910 * this cluster didn't work out, free it and
6913 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6915 if (used_block_group != block_group)
6916 btrfs_release_block_group(used_block_group,
6919 if (loop >= LOOP_NO_EMPTY_SIZE) {
6920 spin_unlock(&last_ptr->refill_lock);
6921 goto unclustered_alloc;
6924 aligned_cluster = max_t(unsigned long,
6925 empty_cluster + empty_size,
6926 block_group->full_stripe_len);
6928 /* allocate a cluster in this block group */
6929 ret = btrfs_find_space_cluster(root, block_group,
6930 last_ptr, search_start,
6935 * now pull our allocation out of this
6938 offset = btrfs_alloc_from_cluster(block_group,
6944 /* we found one, proceed */
6945 spin_unlock(&last_ptr->refill_lock);
6946 trace_btrfs_reserve_extent_cluster(root,
6947 block_group, search_start,
6951 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6952 && !failed_cluster_refill) {
6953 spin_unlock(&last_ptr->refill_lock);
6955 failed_cluster_refill = true;
6956 wait_block_group_cache_progress(block_group,
6957 num_bytes + empty_cluster + empty_size);
6958 goto have_block_group;
6962 * at this point we either didn't find a cluster
6963 * or we weren't able to allocate a block from our
6964 * cluster. Free the cluster we've been trying
6965 * to use, and go to the next block group
6967 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6968 spin_unlock(&last_ptr->refill_lock);
6973 spin_lock(&block_group->free_space_ctl->tree_lock);
6975 block_group->free_space_ctl->free_space <
6976 num_bytes + empty_cluster + empty_size) {
6977 if (block_group->free_space_ctl->free_space >
6980 block_group->free_space_ctl->free_space;
6981 spin_unlock(&block_group->free_space_ctl->tree_lock);
6984 spin_unlock(&block_group->free_space_ctl->tree_lock);
6986 offset = btrfs_find_space_for_alloc(block_group, search_start,
6987 num_bytes, empty_size,
6990 * If we didn't find a chunk, and we haven't failed on this
6991 * block group before, and this block group is in the middle of
6992 * caching and we are ok with waiting, then go ahead and wait
6993 * for progress to be made, and set failed_alloc to true.
6995 * If failed_alloc is true then we've already waited on this
6996 * block group once and should move on to the next block group.
6998 if (!offset && !failed_alloc && !cached &&
6999 loop > LOOP_CACHING_NOWAIT) {
7000 wait_block_group_cache_progress(block_group,
7001 num_bytes + empty_size);
7002 failed_alloc = true;
7003 goto have_block_group;
7004 } else if (!offset) {
7006 have_caching_bg = true;
7010 search_start = ALIGN(offset, root->stripesize);
7012 /* move on to the next group */
7013 if (search_start + num_bytes >
7014 block_group->key.objectid + block_group->key.offset) {
7015 btrfs_add_free_space(block_group, offset, num_bytes);
7019 if (offset < search_start)
7020 btrfs_add_free_space(block_group, offset,
7021 search_start - offset);
7022 BUG_ON(offset > search_start);
7024 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7025 alloc_type, delalloc);
7026 if (ret == -EAGAIN) {
7027 btrfs_add_free_space(block_group, offset, num_bytes);
7031 /* we are all good, lets return */
7032 ins->objectid = search_start;
7033 ins->offset = num_bytes;
7035 trace_btrfs_reserve_extent(orig_root, block_group,
7036 search_start, num_bytes);
7037 btrfs_release_block_group(block_group, delalloc);
7040 failed_cluster_refill = false;
7041 failed_alloc = false;
7042 BUG_ON(index != get_block_group_index(block_group));
7043 btrfs_release_block_group(block_group, delalloc);
7045 up_read(&space_info->groups_sem);
7047 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7050 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7054 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7055 * caching kthreads as we move along
7056 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7057 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7058 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7061 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7064 if (loop == LOOP_ALLOC_CHUNK) {
7065 struct btrfs_trans_handle *trans;
7068 trans = current->journal_info;
7072 trans = btrfs_join_transaction(root);
7074 if (IS_ERR(trans)) {
7075 ret = PTR_ERR(trans);
7079 ret = do_chunk_alloc(trans, root, flags,
7082 * Do not bail out on ENOSPC since we
7083 * can do more things.
7085 if (ret < 0 && ret != -ENOSPC)
7086 btrfs_abort_transaction(trans,
7091 btrfs_end_transaction(trans, root);
7096 if (loop == LOOP_NO_EMPTY_SIZE) {
7102 } else if (!ins->objectid) {
7104 } else if (ins->objectid) {
7109 ins->offset = max_extent_size;
7113 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7114 int dump_block_groups)
7116 struct btrfs_block_group_cache *cache;
7119 spin_lock(&info->lock);
7120 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7122 info->total_bytes - info->bytes_used - info->bytes_pinned -
7123 info->bytes_reserved - info->bytes_readonly,
7124 (info->full) ? "" : "not ");
7125 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7126 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7127 info->total_bytes, info->bytes_used, info->bytes_pinned,
7128 info->bytes_reserved, info->bytes_may_use,
7129 info->bytes_readonly);
7130 spin_unlock(&info->lock);
7132 if (!dump_block_groups)
7135 down_read(&info->groups_sem);
7137 list_for_each_entry(cache, &info->block_groups[index], list) {
7138 spin_lock(&cache->lock);
7139 printk(KERN_INFO "BTRFS: "
7140 "block group %llu has %llu bytes, "
7141 "%llu used %llu pinned %llu reserved %s\n",
7142 cache->key.objectid, cache->key.offset,
7143 btrfs_block_group_used(&cache->item), cache->pinned,
7144 cache->reserved, cache->ro ? "[readonly]" : "");
7145 btrfs_dump_free_space(cache, bytes);
7146 spin_unlock(&cache->lock);
7148 if (++index < BTRFS_NR_RAID_TYPES)
7150 up_read(&info->groups_sem);
7153 int btrfs_reserve_extent(struct btrfs_root *root,
7154 u64 num_bytes, u64 min_alloc_size,
7155 u64 empty_size, u64 hint_byte,
7156 struct btrfs_key *ins, int is_data, int delalloc)
7158 bool final_tried = false;
7162 flags = btrfs_get_alloc_profile(root, is_data);
7164 WARN_ON(num_bytes < root->sectorsize);
7165 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7168 if (ret == -ENOSPC) {
7169 if (!final_tried && ins->offset) {
7170 num_bytes = min(num_bytes >> 1, ins->offset);
7171 num_bytes = round_down(num_bytes, root->sectorsize);
7172 num_bytes = max(num_bytes, min_alloc_size);
7173 if (num_bytes == min_alloc_size)
7176 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7177 struct btrfs_space_info *sinfo;
7179 sinfo = __find_space_info(root->fs_info, flags);
7180 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7183 dump_space_info(sinfo, num_bytes, 1);
7190 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7192 int pin, int delalloc)
7194 struct btrfs_block_group_cache *cache;
7197 cache = btrfs_lookup_block_group(root->fs_info, start);
7199 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7205 pin_down_extent(root, cache, start, len, 1);
7207 if (btrfs_test_opt(root, DISCARD))
7208 ret = btrfs_discard_extent(root, start, len, NULL);
7209 btrfs_add_free_space(cache, start, len);
7210 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7213 btrfs_put_block_group(cache);
7215 trace_btrfs_reserved_extent_free(root, start, len);
7220 int btrfs_free_reserved_extent(struct btrfs_root *root,
7221 u64 start, u64 len, int delalloc)
7223 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7226 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7229 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7232 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7233 struct btrfs_root *root,
7234 u64 parent, u64 root_objectid,
7235 u64 flags, u64 owner, u64 offset,
7236 struct btrfs_key *ins, int ref_mod)
7239 struct btrfs_fs_info *fs_info = root->fs_info;
7240 struct btrfs_extent_item *extent_item;
7241 struct btrfs_extent_inline_ref *iref;
7242 struct btrfs_path *path;
7243 struct extent_buffer *leaf;
7248 type = BTRFS_SHARED_DATA_REF_KEY;
7250 type = BTRFS_EXTENT_DATA_REF_KEY;
7252 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7254 path = btrfs_alloc_path();
7258 path->leave_spinning = 1;
7259 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7262 btrfs_free_path(path);
7266 leaf = path->nodes[0];
7267 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7268 struct btrfs_extent_item);
7269 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7270 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7271 btrfs_set_extent_flags(leaf, extent_item,
7272 flags | BTRFS_EXTENT_FLAG_DATA);
7274 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7275 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7277 struct btrfs_shared_data_ref *ref;
7278 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7279 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7280 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7282 struct btrfs_extent_data_ref *ref;
7283 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7284 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7285 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7286 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7287 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7290 btrfs_mark_buffer_dirty(path->nodes[0]);
7291 btrfs_free_path(path);
7293 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7294 if (ret) { /* -ENOENT, logic error */
7295 btrfs_err(fs_info, "update block group failed for %llu %llu",
7296 ins->objectid, ins->offset);
7299 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7303 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7304 struct btrfs_root *root,
7305 u64 parent, u64 root_objectid,
7306 u64 flags, struct btrfs_disk_key *key,
7307 int level, struct btrfs_key *ins,
7311 struct btrfs_fs_info *fs_info = root->fs_info;
7312 struct btrfs_extent_item *extent_item;
7313 struct btrfs_tree_block_info *block_info;
7314 struct btrfs_extent_inline_ref *iref;
7315 struct btrfs_path *path;
7316 struct extent_buffer *leaf;
7317 u32 size = sizeof(*extent_item) + sizeof(*iref);
7318 u64 num_bytes = ins->offset;
7319 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7322 if (!skinny_metadata)
7323 size += sizeof(*block_info);
7325 path = btrfs_alloc_path();
7327 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7332 path->leave_spinning = 1;
7333 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7336 btrfs_free_path(path);
7337 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7342 leaf = path->nodes[0];
7343 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7344 struct btrfs_extent_item);
7345 btrfs_set_extent_refs(leaf, extent_item, 1);
7346 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7347 btrfs_set_extent_flags(leaf, extent_item,
7348 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7350 if (skinny_metadata) {
7351 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7352 num_bytes = root->nodesize;
7354 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7355 btrfs_set_tree_block_key(leaf, block_info, key);
7356 btrfs_set_tree_block_level(leaf, block_info, level);
7357 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7361 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7362 btrfs_set_extent_inline_ref_type(leaf, iref,
7363 BTRFS_SHARED_BLOCK_REF_KEY);
7364 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7366 btrfs_set_extent_inline_ref_type(leaf, iref,
7367 BTRFS_TREE_BLOCK_REF_KEY);
7368 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7371 btrfs_mark_buffer_dirty(leaf);
7372 btrfs_free_path(path);
7374 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7376 if (ret) { /* -ENOENT, logic error */
7377 btrfs_err(fs_info, "update block group failed for %llu %llu",
7378 ins->objectid, ins->offset);
7382 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7386 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7387 struct btrfs_root *root,
7388 u64 root_objectid, u64 owner,
7389 u64 offset, struct btrfs_key *ins)
7393 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7395 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7397 root_objectid, owner, offset,
7398 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7403 * this is used by the tree logging recovery code. It records that
7404 * an extent has been allocated and makes sure to clear the free
7405 * space cache bits as well
7407 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7408 struct btrfs_root *root,
7409 u64 root_objectid, u64 owner, u64 offset,
7410 struct btrfs_key *ins)
7413 struct btrfs_block_group_cache *block_group;
7416 * Mixed block groups will exclude before processing the log so we only
7417 * need to do the exlude dance if this fs isn't mixed.
7419 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7420 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7425 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7429 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7430 RESERVE_ALLOC_NO_ACCOUNT, 0);
7431 BUG_ON(ret); /* logic error */
7432 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7433 0, owner, offset, ins, 1);
7434 btrfs_put_block_group(block_group);
7438 static struct extent_buffer *
7439 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7440 u64 bytenr, int level)
7442 struct extent_buffer *buf;
7444 buf = btrfs_find_create_tree_block(root, bytenr);
7446 return ERR_PTR(-ENOMEM);
7447 btrfs_set_header_generation(buf, trans->transid);
7448 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7449 btrfs_tree_lock(buf);
7450 clean_tree_block(trans, root->fs_info, buf);
7451 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7453 btrfs_set_lock_blocking(buf);
7454 btrfs_set_buffer_uptodate(buf);
7456 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7457 buf->log_index = root->log_transid % 2;
7459 * we allow two log transactions at a time, use different
7460 * EXENT bit to differentiate dirty pages.
7462 if (buf->log_index == 0)
7463 set_extent_dirty(&root->dirty_log_pages, buf->start,
7464 buf->start + buf->len - 1, GFP_NOFS);
7466 set_extent_new(&root->dirty_log_pages, buf->start,
7467 buf->start + buf->len - 1, GFP_NOFS);
7469 buf->log_index = -1;
7470 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7471 buf->start + buf->len - 1, GFP_NOFS);
7473 trans->blocks_used++;
7474 /* this returns a buffer locked for blocking */
7478 static struct btrfs_block_rsv *
7479 use_block_rsv(struct btrfs_trans_handle *trans,
7480 struct btrfs_root *root, u32 blocksize)
7482 struct btrfs_block_rsv *block_rsv;
7483 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7485 bool global_updated = false;
7487 block_rsv = get_block_rsv(trans, root);
7489 if (unlikely(block_rsv->size == 0))
7492 ret = block_rsv_use_bytes(block_rsv, blocksize);
7496 if (block_rsv->failfast)
7497 return ERR_PTR(ret);
7499 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7500 global_updated = true;
7501 update_global_block_rsv(root->fs_info);
7505 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7506 static DEFINE_RATELIMIT_STATE(_rs,
7507 DEFAULT_RATELIMIT_INTERVAL * 10,
7508 /*DEFAULT_RATELIMIT_BURST*/ 1);
7509 if (__ratelimit(&_rs))
7511 "BTRFS: block rsv returned %d\n", ret);
7514 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7515 BTRFS_RESERVE_NO_FLUSH);
7519 * If we couldn't reserve metadata bytes try and use some from
7520 * the global reserve if its space type is the same as the global
7523 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7524 block_rsv->space_info == global_rsv->space_info) {
7525 ret = block_rsv_use_bytes(global_rsv, blocksize);
7529 return ERR_PTR(ret);
7532 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7533 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7535 block_rsv_add_bytes(block_rsv, blocksize, 0);
7536 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7540 * finds a free extent and does all the dirty work required for allocation
7541 * returns the key for the extent through ins, and a tree buffer for
7542 * the first block of the extent through buf.
7544 * returns the tree buffer or an ERR_PTR on error.
7546 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7547 struct btrfs_root *root,
7548 u64 parent, u64 root_objectid,
7549 struct btrfs_disk_key *key, int level,
7550 u64 hint, u64 empty_size)
7552 struct btrfs_key ins;
7553 struct btrfs_block_rsv *block_rsv;
7554 struct extent_buffer *buf;
7555 struct btrfs_delayed_extent_op *extent_op;
7558 u32 blocksize = root->nodesize;
7559 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7562 if (btrfs_test_is_dummy_root(root)) {
7563 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7566 root->alloc_bytenr += blocksize;
7570 block_rsv = use_block_rsv(trans, root, blocksize);
7571 if (IS_ERR(block_rsv))
7572 return ERR_CAST(block_rsv);
7574 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7575 empty_size, hint, &ins, 0, 0);
7579 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7582 goto out_free_reserved;
7585 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7587 parent = ins.objectid;
7588 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7592 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7593 extent_op = btrfs_alloc_delayed_extent_op();
7599 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7601 memset(&extent_op->key, 0, sizeof(extent_op->key));
7602 extent_op->flags_to_set = flags;
7603 if (skinny_metadata)
7604 extent_op->update_key = 0;
7606 extent_op->update_key = 1;
7607 extent_op->update_flags = 1;
7608 extent_op->is_data = 0;
7609 extent_op->level = level;
7611 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7612 ins.objectid, ins.offset,
7613 parent, root_objectid, level,
7614 BTRFS_ADD_DELAYED_EXTENT,
7617 goto out_free_delayed;
7622 btrfs_free_delayed_extent_op(extent_op);
7624 free_extent_buffer(buf);
7626 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7628 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7629 return ERR_PTR(ret);
7632 struct walk_control {
7633 u64 refs[BTRFS_MAX_LEVEL];
7634 u64 flags[BTRFS_MAX_LEVEL];
7635 struct btrfs_key update_progress;
7646 #define DROP_REFERENCE 1
7647 #define UPDATE_BACKREF 2
7649 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7650 struct btrfs_root *root,
7651 struct walk_control *wc,
7652 struct btrfs_path *path)
7660 struct btrfs_key key;
7661 struct extent_buffer *eb;
7666 if (path->slots[wc->level] < wc->reada_slot) {
7667 wc->reada_count = wc->reada_count * 2 / 3;
7668 wc->reada_count = max(wc->reada_count, 2);
7670 wc->reada_count = wc->reada_count * 3 / 2;
7671 wc->reada_count = min_t(int, wc->reada_count,
7672 BTRFS_NODEPTRS_PER_BLOCK(root));
7675 eb = path->nodes[wc->level];
7676 nritems = btrfs_header_nritems(eb);
7677 blocksize = root->nodesize;
7679 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7680 if (nread >= wc->reada_count)
7684 bytenr = btrfs_node_blockptr(eb, slot);
7685 generation = btrfs_node_ptr_generation(eb, slot);
7687 if (slot == path->slots[wc->level])
7690 if (wc->stage == UPDATE_BACKREF &&
7691 generation <= root->root_key.offset)
7694 /* We don't lock the tree block, it's OK to be racy here */
7695 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7696 wc->level - 1, 1, &refs,
7698 /* We don't care about errors in readahead. */
7703 if (wc->stage == DROP_REFERENCE) {
7707 if (wc->level == 1 &&
7708 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7710 if (!wc->update_ref ||
7711 generation <= root->root_key.offset)
7713 btrfs_node_key_to_cpu(eb, &key, slot);
7714 ret = btrfs_comp_cpu_keys(&key,
7715 &wc->update_progress);
7719 if (wc->level == 1 &&
7720 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7724 readahead_tree_block(root, bytenr);
7727 wc->reada_slot = slot;
7731 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7732 * for later qgroup accounting.
7734 * Current, this function does nothing.
7736 static int account_leaf_items(struct btrfs_trans_handle *trans,
7737 struct btrfs_root *root,
7738 struct extent_buffer *eb)
7740 int nr = btrfs_header_nritems(eb);
7742 struct btrfs_key key;
7743 struct btrfs_file_extent_item *fi;
7744 u64 bytenr, num_bytes;
7746 for (i = 0; i < nr; i++) {
7747 btrfs_item_key_to_cpu(eb, &key, i);
7749 if (key.type != BTRFS_EXTENT_DATA_KEY)
7752 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7753 /* filter out non qgroup-accountable extents */
7754 extent_type = btrfs_file_extent_type(eb, fi);
7756 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7759 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7763 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7769 * Walk up the tree from the bottom, freeing leaves and any interior
7770 * nodes which have had all slots visited. If a node (leaf or
7771 * interior) is freed, the node above it will have it's slot
7772 * incremented. The root node will never be freed.
7774 * At the end of this function, we should have a path which has all
7775 * slots incremented to the next position for a search. If we need to
7776 * read a new node it will be NULL and the node above it will have the
7777 * correct slot selected for a later read.
7779 * If we increment the root nodes slot counter past the number of
7780 * elements, 1 is returned to signal completion of the search.
7782 static int adjust_slots_upwards(struct btrfs_root *root,
7783 struct btrfs_path *path, int root_level)
7787 struct extent_buffer *eb;
7789 if (root_level == 0)
7792 while (level <= root_level) {
7793 eb = path->nodes[level];
7794 nr = btrfs_header_nritems(eb);
7795 path->slots[level]++;
7796 slot = path->slots[level];
7797 if (slot >= nr || level == 0) {
7799 * Don't free the root - we will detect this
7800 * condition after our loop and return a
7801 * positive value for caller to stop walking the tree.
7803 if (level != root_level) {
7804 btrfs_tree_unlock_rw(eb, path->locks[level]);
7805 path->locks[level] = 0;
7807 free_extent_buffer(eb);
7808 path->nodes[level] = NULL;
7809 path->slots[level] = 0;
7813 * We have a valid slot to walk back down
7814 * from. Stop here so caller can process these
7823 eb = path->nodes[root_level];
7824 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7831 * root_eb is the subtree root and is locked before this function is called.
7832 * TODO: Modify this function to mark all (including complete shared node)
7833 * to dirty_extent_root to allow it get accounted in qgroup.
7835 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7836 struct btrfs_root *root,
7837 struct extent_buffer *root_eb,
7843 struct extent_buffer *eb = root_eb;
7844 struct btrfs_path *path = NULL;
7846 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7847 BUG_ON(root_eb == NULL);
7849 if (!root->fs_info->quota_enabled)
7852 if (!extent_buffer_uptodate(root_eb)) {
7853 ret = btrfs_read_buffer(root_eb, root_gen);
7858 if (root_level == 0) {
7859 ret = account_leaf_items(trans, root, root_eb);
7863 path = btrfs_alloc_path();
7868 * Walk down the tree. Missing extent blocks are filled in as
7869 * we go. Metadata is accounted every time we read a new
7872 * When we reach a leaf, we account for file extent items in it,
7873 * walk back up the tree (adjusting slot pointers as we go)
7874 * and restart the search process.
7876 extent_buffer_get(root_eb); /* For path */
7877 path->nodes[root_level] = root_eb;
7878 path->slots[root_level] = 0;
7879 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7882 while (level >= 0) {
7883 if (path->nodes[level] == NULL) {
7888 /* We need to get child blockptr/gen from
7889 * parent before we can read it. */
7890 eb = path->nodes[level + 1];
7891 parent_slot = path->slots[level + 1];
7892 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7893 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7895 eb = read_tree_block(root, child_bytenr, child_gen);
7899 } else if (!extent_buffer_uptodate(eb)) {
7900 free_extent_buffer(eb);
7905 path->nodes[level] = eb;
7906 path->slots[level] = 0;
7908 btrfs_tree_read_lock(eb);
7909 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7910 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7914 ret = account_leaf_items(trans, root, path->nodes[level]);
7918 /* Nonzero return here means we completed our search */
7919 ret = adjust_slots_upwards(root, path, root_level);
7923 /* Restart search with new slots */
7932 btrfs_free_path(path);
7938 * helper to process tree block while walking down the tree.
7940 * when wc->stage == UPDATE_BACKREF, this function updates
7941 * back refs for pointers in the block.
7943 * NOTE: return value 1 means we should stop walking down.
7945 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7946 struct btrfs_root *root,
7947 struct btrfs_path *path,
7948 struct walk_control *wc, int lookup_info)
7950 int level = wc->level;
7951 struct extent_buffer *eb = path->nodes[level];
7952 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7955 if (wc->stage == UPDATE_BACKREF &&
7956 btrfs_header_owner(eb) != root->root_key.objectid)
7960 * when reference count of tree block is 1, it won't increase
7961 * again. once full backref flag is set, we never clear it.
7964 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7965 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7966 BUG_ON(!path->locks[level]);
7967 ret = btrfs_lookup_extent_info(trans, root,
7968 eb->start, level, 1,
7971 BUG_ON(ret == -ENOMEM);
7974 BUG_ON(wc->refs[level] == 0);
7977 if (wc->stage == DROP_REFERENCE) {
7978 if (wc->refs[level] > 1)
7981 if (path->locks[level] && !wc->keep_locks) {
7982 btrfs_tree_unlock_rw(eb, path->locks[level]);
7983 path->locks[level] = 0;
7988 /* wc->stage == UPDATE_BACKREF */
7989 if (!(wc->flags[level] & flag)) {
7990 BUG_ON(!path->locks[level]);
7991 ret = btrfs_inc_ref(trans, root, eb, 1);
7992 BUG_ON(ret); /* -ENOMEM */
7993 ret = btrfs_dec_ref(trans, root, eb, 0);
7994 BUG_ON(ret); /* -ENOMEM */
7995 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7997 btrfs_header_level(eb), 0);
7998 BUG_ON(ret); /* -ENOMEM */
7999 wc->flags[level] |= flag;
8003 * the block is shared by multiple trees, so it's not good to
8004 * keep the tree lock
8006 if (path->locks[level] && level > 0) {
8007 btrfs_tree_unlock_rw(eb, path->locks[level]);
8008 path->locks[level] = 0;
8014 * helper to process tree block pointer.
8016 * when wc->stage == DROP_REFERENCE, this function checks
8017 * reference count of the block pointed to. if the block
8018 * is shared and we need update back refs for the subtree
8019 * rooted at the block, this function changes wc->stage to
8020 * UPDATE_BACKREF. if the block is shared and there is no
8021 * need to update back, this function drops the reference
8024 * NOTE: return value 1 means we should stop walking down.
8026 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8027 struct btrfs_root *root,
8028 struct btrfs_path *path,
8029 struct walk_control *wc, int *lookup_info)
8035 struct btrfs_key key;
8036 struct extent_buffer *next;
8037 int level = wc->level;
8040 bool need_account = false;
8042 generation = btrfs_node_ptr_generation(path->nodes[level],
8043 path->slots[level]);
8045 * if the lower level block was created before the snapshot
8046 * was created, we know there is no need to update back refs
8049 if (wc->stage == UPDATE_BACKREF &&
8050 generation <= root->root_key.offset) {
8055 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8056 blocksize = root->nodesize;
8058 next = btrfs_find_tree_block(root->fs_info, bytenr);
8060 next = btrfs_find_create_tree_block(root, bytenr);
8063 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8067 btrfs_tree_lock(next);
8068 btrfs_set_lock_blocking(next);
8070 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8071 &wc->refs[level - 1],
8072 &wc->flags[level - 1]);
8074 btrfs_tree_unlock(next);
8078 if (unlikely(wc->refs[level - 1] == 0)) {
8079 btrfs_err(root->fs_info, "Missing references.");
8084 if (wc->stage == DROP_REFERENCE) {
8085 if (wc->refs[level - 1] > 1) {
8086 need_account = true;
8088 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8091 if (!wc->update_ref ||
8092 generation <= root->root_key.offset)
8095 btrfs_node_key_to_cpu(path->nodes[level], &key,
8096 path->slots[level]);
8097 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8101 wc->stage = UPDATE_BACKREF;
8102 wc->shared_level = level - 1;
8106 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8110 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8111 btrfs_tree_unlock(next);
8112 free_extent_buffer(next);
8118 if (reada && level == 1)
8119 reada_walk_down(trans, root, wc, path);
8120 next = read_tree_block(root, bytenr, generation);
8122 return PTR_ERR(next);
8123 } else if (!extent_buffer_uptodate(next)) {
8124 free_extent_buffer(next);
8127 btrfs_tree_lock(next);
8128 btrfs_set_lock_blocking(next);
8132 BUG_ON(level != btrfs_header_level(next));
8133 path->nodes[level] = next;
8134 path->slots[level] = 0;
8135 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8141 wc->refs[level - 1] = 0;
8142 wc->flags[level - 1] = 0;
8143 if (wc->stage == DROP_REFERENCE) {
8144 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8145 parent = path->nodes[level]->start;
8147 BUG_ON(root->root_key.objectid !=
8148 btrfs_header_owner(path->nodes[level]));
8153 ret = account_shared_subtree(trans, root, next,
8154 generation, level - 1);
8156 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8157 "%d accounting shared subtree. Quota "
8158 "is out of sync, rescan required.\n",
8159 root->fs_info->sb->s_id, ret);
8162 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8163 root->root_key.objectid, level - 1, 0, 0);
8164 BUG_ON(ret); /* -ENOMEM */
8166 btrfs_tree_unlock(next);
8167 free_extent_buffer(next);
8173 * helper to process tree block while walking up the tree.
8175 * when wc->stage == DROP_REFERENCE, this function drops
8176 * reference count on the block.
8178 * when wc->stage == UPDATE_BACKREF, this function changes
8179 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8180 * to UPDATE_BACKREF previously while processing the block.
8182 * NOTE: return value 1 means we should stop walking up.
8184 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8185 struct btrfs_root *root,
8186 struct btrfs_path *path,
8187 struct walk_control *wc)
8190 int level = wc->level;
8191 struct extent_buffer *eb = path->nodes[level];
8194 if (wc->stage == UPDATE_BACKREF) {
8195 BUG_ON(wc->shared_level < level);
8196 if (level < wc->shared_level)
8199 ret = find_next_key(path, level + 1, &wc->update_progress);
8203 wc->stage = DROP_REFERENCE;
8204 wc->shared_level = -1;
8205 path->slots[level] = 0;
8208 * check reference count again if the block isn't locked.
8209 * we should start walking down the tree again if reference
8212 if (!path->locks[level]) {
8214 btrfs_tree_lock(eb);
8215 btrfs_set_lock_blocking(eb);
8216 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8218 ret = btrfs_lookup_extent_info(trans, root,
8219 eb->start, level, 1,
8223 btrfs_tree_unlock_rw(eb, path->locks[level]);
8224 path->locks[level] = 0;
8227 BUG_ON(wc->refs[level] == 0);
8228 if (wc->refs[level] == 1) {
8229 btrfs_tree_unlock_rw(eb, path->locks[level]);
8230 path->locks[level] = 0;
8236 /* wc->stage == DROP_REFERENCE */
8237 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8239 if (wc->refs[level] == 1) {
8241 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8242 ret = btrfs_dec_ref(trans, root, eb, 1);
8244 ret = btrfs_dec_ref(trans, root, eb, 0);
8245 BUG_ON(ret); /* -ENOMEM */
8246 ret = account_leaf_items(trans, root, eb);
8248 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8249 "%d accounting leaf items. Quota "
8250 "is out of sync, rescan required.\n",
8251 root->fs_info->sb->s_id, ret);
8254 /* make block locked assertion in clean_tree_block happy */
8255 if (!path->locks[level] &&
8256 btrfs_header_generation(eb) == trans->transid) {
8257 btrfs_tree_lock(eb);
8258 btrfs_set_lock_blocking(eb);
8259 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8261 clean_tree_block(trans, root->fs_info, eb);
8264 if (eb == root->node) {
8265 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8268 BUG_ON(root->root_key.objectid !=
8269 btrfs_header_owner(eb));
8271 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8272 parent = path->nodes[level + 1]->start;
8274 BUG_ON(root->root_key.objectid !=
8275 btrfs_header_owner(path->nodes[level + 1]));
8278 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8280 wc->refs[level] = 0;
8281 wc->flags[level] = 0;
8285 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8286 struct btrfs_root *root,
8287 struct btrfs_path *path,
8288 struct walk_control *wc)
8290 int level = wc->level;
8291 int lookup_info = 1;
8294 while (level >= 0) {
8295 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8302 if (path->slots[level] >=
8303 btrfs_header_nritems(path->nodes[level]))
8306 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8308 path->slots[level]++;
8317 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8318 struct btrfs_root *root,
8319 struct btrfs_path *path,
8320 struct walk_control *wc, int max_level)
8322 int level = wc->level;
8325 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8326 while (level < max_level && path->nodes[level]) {
8328 if (path->slots[level] + 1 <
8329 btrfs_header_nritems(path->nodes[level])) {
8330 path->slots[level]++;
8333 ret = walk_up_proc(trans, root, path, wc);
8337 if (path->locks[level]) {
8338 btrfs_tree_unlock_rw(path->nodes[level],
8339 path->locks[level]);
8340 path->locks[level] = 0;
8342 free_extent_buffer(path->nodes[level]);
8343 path->nodes[level] = NULL;
8351 * drop a subvolume tree.
8353 * this function traverses the tree freeing any blocks that only
8354 * referenced by the tree.
8356 * when a shared tree block is found. this function decreases its
8357 * reference count by one. if update_ref is true, this function
8358 * also make sure backrefs for the shared block and all lower level
8359 * blocks are properly updated.
8361 * If called with for_reloc == 0, may exit early with -EAGAIN
8363 int btrfs_drop_snapshot(struct btrfs_root *root,
8364 struct btrfs_block_rsv *block_rsv, int update_ref,
8367 struct btrfs_path *path;
8368 struct btrfs_trans_handle *trans;
8369 struct btrfs_root *tree_root = root->fs_info->tree_root;
8370 struct btrfs_root_item *root_item = &root->root_item;
8371 struct walk_control *wc;
8372 struct btrfs_key key;
8376 bool root_dropped = false;
8378 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8380 path = btrfs_alloc_path();
8386 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8388 btrfs_free_path(path);
8393 trans = btrfs_start_transaction(tree_root, 0);
8394 if (IS_ERR(trans)) {
8395 err = PTR_ERR(trans);
8400 trans->block_rsv = block_rsv;
8402 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8403 level = btrfs_header_level(root->node);
8404 path->nodes[level] = btrfs_lock_root_node(root);
8405 btrfs_set_lock_blocking(path->nodes[level]);
8406 path->slots[level] = 0;
8407 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8408 memset(&wc->update_progress, 0,
8409 sizeof(wc->update_progress));
8411 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8412 memcpy(&wc->update_progress, &key,
8413 sizeof(wc->update_progress));
8415 level = root_item->drop_level;
8417 path->lowest_level = level;
8418 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8419 path->lowest_level = 0;
8427 * unlock our path, this is safe because only this
8428 * function is allowed to delete this snapshot
8430 btrfs_unlock_up_safe(path, 0);
8432 level = btrfs_header_level(root->node);
8434 btrfs_tree_lock(path->nodes[level]);
8435 btrfs_set_lock_blocking(path->nodes[level]);
8436 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8438 ret = btrfs_lookup_extent_info(trans, root,
8439 path->nodes[level]->start,
8440 level, 1, &wc->refs[level],
8446 BUG_ON(wc->refs[level] == 0);
8448 if (level == root_item->drop_level)
8451 btrfs_tree_unlock(path->nodes[level]);
8452 path->locks[level] = 0;
8453 WARN_ON(wc->refs[level] != 1);
8459 wc->shared_level = -1;
8460 wc->stage = DROP_REFERENCE;
8461 wc->update_ref = update_ref;
8463 wc->for_reloc = for_reloc;
8464 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8468 ret = walk_down_tree(trans, root, path, wc);
8474 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8481 BUG_ON(wc->stage != DROP_REFERENCE);
8485 if (wc->stage == DROP_REFERENCE) {
8487 btrfs_node_key(path->nodes[level],
8488 &root_item->drop_progress,
8489 path->slots[level]);
8490 root_item->drop_level = level;
8493 BUG_ON(wc->level == 0);
8494 if (btrfs_should_end_transaction(trans, tree_root) ||
8495 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8496 ret = btrfs_update_root(trans, tree_root,
8500 btrfs_abort_transaction(trans, tree_root, ret);
8505 btrfs_end_transaction_throttle(trans, tree_root);
8506 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8507 pr_debug("BTRFS: drop snapshot early exit\n");
8512 trans = btrfs_start_transaction(tree_root, 0);
8513 if (IS_ERR(trans)) {
8514 err = PTR_ERR(trans);
8518 trans->block_rsv = block_rsv;
8521 btrfs_release_path(path);
8525 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8527 btrfs_abort_transaction(trans, tree_root, ret);
8531 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8532 ret = btrfs_find_root(tree_root, &root->root_key, path,
8535 btrfs_abort_transaction(trans, tree_root, ret);
8538 } else if (ret > 0) {
8539 /* if we fail to delete the orphan item this time
8540 * around, it'll get picked up the next time.
8542 * The most common failure here is just -ENOENT.
8544 btrfs_del_orphan_item(trans, tree_root,
8545 root->root_key.objectid);
8549 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8550 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8552 free_extent_buffer(root->node);
8553 free_extent_buffer(root->commit_root);
8554 btrfs_put_fs_root(root);
8556 root_dropped = true;
8558 btrfs_end_transaction_throttle(trans, tree_root);
8561 btrfs_free_path(path);
8564 * So if we need to stop dropping the snapshot for whatever reason we
8565 * need to make sure to add it back to the dead root list so that we
8566 * keep trying to do the work later. This also cleans up roots if we
8567 * don't have it in the radix (like when we recover after a power fail
8568 * or unmount) so we don't leak memory.
8570 if (!for_reloc && root_dropped == false)
8571 btrfs_add_dead_root(root);
8572 if (err && err != -EAGAIN)
8573 btrfs_std_error(root->fs_info, err);
8578 * drop subtree rooted at tree block 'node'.
8580 * NOTE: this function will unlock and release tree block 'node'
8581 * only used by relocation code
8583 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8584 struct btrfs_root *root,
8585 struct extent_buffer *node,
8586 struct extent_buffer *parent)
8588 struct btrfs_path *path;
8589 struct walk_control *wc;
8595 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8597 path = btrfs_alloc_path();
8601 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8603 btrfs_free_path(path);
8607 btrfs_assert_tree_locked(parent);
8608 parent_level = btrfs_header_level(parent);
8609 extent_buffer_get(parent);
8610 path->nodes[parent_level] = parent;
8611 path->slots[parent_level] = btrfs_header_nritems(parent);
8613 btrfs_assert_tree_locked(node);
8614 level = btrfs_header_level(node);
8615 path->nodes[level] = node;
8616 path->slots[level] = 0;
8617 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8619 wc->refs[parent_level] = 1;
8620 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8622 wc->shared_level = -1;
8623 wc->stage = DROP_REFERENCE;
8627 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8630 wret = walk_down_tree(trans, root, path, wc);
8636 wret = walk_up_tree(trans, root, path, wc, parent_level);
8644 btrfs_free_path(path);
8648 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8654 * if restripe for this chunk_type is on pick target profile and
8655 * return, otherwise do the usual balance
8657 stripped = get_restripe_target(root->fs_info, flags);
8659 return extended_to_chunk(stripped);
8661 num_devices = root->fs_info->fs_devices->rw_devices;
8663 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8664 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8665 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8667 if (num_devices == 1) {
8668 stripped |= BTRFS_BLOCK_GROUP_DUP;
8669 stripped = flags & ~stripped;
8671 /* turn raid0 into single device chunks */
8672 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8675 /* turn mirroring into duplication */
8676 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8677 BTRFS_BLOCK_GROUP_RAID10))
8678 return stripped | BTRFS_BLOCK_GROUP_DUP;
8680 /* they already had raid on here, just return */
8681 if (flags & stripped)
8684 stripped |= BTRFS_BLOCK_GROUP_DUP;
8685 stripped = flags & ~stripped;
8687 /* switch duplicated blocks with raid1 */
8688 if (flags & BTRFS_BLOCK_GROUP_DUP)
8689 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8691 /* this is drive concat, leave it alone */
8697 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8699 struct btrfs_space_info *sinfo = cache->space_info;
8701 u64 min_allocable_bytes;
8706 * We need some metadata space and system metadata space for
8707 * allocating chunks in some corner cases until we force to set
8708 * it to be readonly.
8711 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8713 min_allocable_bytes = 1 * 1024 * 1024;
8715 min_allocable_bytes = 0;
8717 spin_lock(&sinfo->lock);
8718 spin_lock(&cache->lock);
8725 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8726 cache->bytes_super - btrfs_block_group_used(&cache->item);
8728 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8729 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8730 min_allocable_bytes <= sinfo->total_bytes) {
8731 sinfo->bytes_readonly += num_bytes;
8733 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8737 spin_unlock(&cache->lock);
8738 spin_unlock(&sinfo->lock);
8742 int btrfs_set_block_group_ro(struct btrfs_root *root,
8743 struct btrfs_block_group_cache *cache)
8746 struct btrfs_trans_handle *trans;
8753 trans = btrfs_join_transaction(root);
8755 return PTR_ERR(trans);
8758 * we're not allowed to set block groups readonly after the dirty
8759 * block groups cache has started writing. If it already started,
8760 * back off and let this transaction commit
8762 mutex_lock(&root->fs_info->ro_block_group_mutex);
8763 if (trans->transaction->dirty_bg_run) {
8764 u64 transid = trans->transid;
8766 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8767 btrfs_end_transaction(trans, root);
8769 ret = btrfs_wait_for_commit(root, transid);
8776 * if we are changing raid levels, try to allocate a corresponding
8777 * block group with the new raid level.
8779 alloc_flags = update_block_group_flags(root, cache->flags);
8780 if (alloc_flags != cache->flags) {
8781 ret = do_chunk_alloc(trans, root, alloc_flags,
8784 * ENOSPC is allowed here, we may have enough space
8785 * already allocated at the new raid level to
8794 ret = set_block_group_ro(cache, 0);
8797 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8798 ret = do_chunk_alloc(trans, root, alloc_flags,
8802 ret = set_block_group_ro(cache, 0);
8804 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8805 alloc_flags = update_block_group_flags(root, cache->flags);
8806 lock_chunks(root->fs_info->chunk_root);
8807 check_system_chunk(trans, root, alloc_flags);
8808 unlock_chunks(root->fs_info->chunk_root);
8810 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8812 btrfs_end_transaction(trans, root);
8816 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8817 struct btrfs_root *root, u64 type)
8819 u64 alloc_flags = get_alloc_profile(root, type);
8820 return do_chunk_alloc(trans, root, alloc_flags,
8825 * helper to account the unused space of all the readonly block group in the
8826 * space_info. takes mirrors into account.
8828 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8830 struct btrfs_block_group_cache *block_group;
8834 /* It's df, we don't care if it's racey */
8835 if (list_empty(&sinfo->ro_bgs))
8838 spin_lock(&sinfo->lock);
8839 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8840 spin_lock(&block_group->lock);
8842 if (!block_group->ro) {
8843 spin_unlock(&block_group->lock);
8847 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8848 BTRFS_BLOCK_GROUP_RAID10 |
8849 BTRFS_BLOCK_GROUP_DUP))
8854 free_bytes += (block_group->key.offset -
8855 btrfs_block_group_used(&block_group->item)) *
8858 spin_unlock(&block_group->lock);
8860 spin_unlock(&sinfo->lock);
8865 void btrfs_set_block_group_rw(struct btrfs_root *root,
8866 struct btrfs_block_group_cache *cache)
8868 struct btrfs_space_info *sinfo = cache->space_info;
8873 spin_lock(&sinfo->lock);
8874 spin_lock(&cache->lock);
8875 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8876 cache->bytes_super - btrfs_block_group_used(&cache->item);
8877 sinfo->bytes_readonly -= num_bytes;
8879 list_del_init(&cache->ro_list);
8880 spin_unlock(&cache->lock);
8881 spin_unlock(&sinfo->lock);
8885 * checks to see if its even possible to relocate this block group.
8887 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8888 * ok to go ahead and try.
8890 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8892 struct btrfs_block_group_cache *block_group;
8893 struct btrfs_space_info *space_info;
8894 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8895 struct btrfs_device *device;
8896 struct btrfs_trans_handle *trans;
8905 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8907 /* odd, couldn't find the block group, leave it alone */
8911 min_free = btrfs_block_group_used(&block_group->item);
8913 /* no bytes used, we're good */
8917 space_info = block_group->space_info;
8918 spin_lock(&space_info->lock);
8920 full = space_info->full;
8923 * if this is the last block group we have in this space, we can't
8924 * relocate it unless we're able to allocate a new chunk below.
8926 * Otherwise, we need to make sure we have room in the space to handle
8927 * all of the extents from this block group. If we can, we're good
8929 if ((space_info->total_bytes != block_group->key.offset) &&
8930 (space_info->bytes_used + space_info->bytes_reserved +
8931 space_info->bytes_pinned + space_info->bytes_readonly +
8932 min_free < space_info->total_bytes)) {
8933 spin_unlock(&space_info->lock);
8936 spin_unlock(&space_info->lock);
8939 * ok we don't have enough space, but maybe we have free space on our
8940 * devices to allocate new chunks for relocation, so loop through our
8941 * alloc devices and guess if we have enough space. if this block
8942 * group is going to be restriped, run checks against the target
8943 * profile instead of the current one.
8955 target = get_restripe_target(root->fs_info, block_group->flags);
8957 index = __get_raid_index(extended_to_chunk(target));
8960 * this is just a balance, so if we were marked as full
8961 * we know there is no space for a new chunk
8966 index = get_block_group_index(block_group);
8969 if (index == BTRFS_RAID_RAID10) {
8973 } else if (index == BTRFS_RAID_RAID1) {
8975 } else if (index == BTRFS_RAID_DUP) {
8978 } else if (index == BTRFS_RAID_RAID0) {
8979 dev_min = fs_devices->rw_devices;
8980 min_free = div64_u64(min_free, dev_min);
8983 /* We need to do this so that we can look at pending chunks */
8984 trans = btrfs_join_transaction(root);
8985 if (IS_ERR(trans)) {
8986 ret = PTR_ERR(trans);
8990 mutex_lock(&root->fs_info->chunk_mutex);
8991 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8995 * check to make sure we can actually find a chunk with enough
8996 * space to fit our block group in.
8998 if (device->total_bytes > device->bytes_used + min_free &&
8999 !device->is_tgtdev_for_dev_replace) {
9000 ret = find_free_dev_extent(trans, device, min_free,
9005 if (dev_nr >= dev_min)
9011 mutex_unlock(&root->fs_info->chunk_mutex);
9012 btrfs_end_transaction(trans, root);
9014 btrfs_put_block_group(block_group);
9018 static int find_first_block_group(struct btrfs_root *root,
9019 struct btrfs_path *path, struct btrfs_key *key)
9022 struct btrfs_key found_key;
9023 struct extent_buffer *leaf;
9026 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9031 slot = path->slots[0];
9032 leaf = path->nodes[0];
9033 if (slot >= btrfs_header_nritems(leaf)) {
9034 ret = btrfs_next_leaf(root, path);
9041 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9043 if (found_key.objectid >= key->objectid &&
9044 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9054 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9056 struct btrfs_block_group_cache *block_group;
9060 struct inode *inode;
9062 block_group = btrfs_lookup_first_block_group(info, last);
9063 while (block_group) {
9064 spin_lock(&block_group->lock);
9065 if (block_group->iref)
9067 spin_unlock(&block_group->lock);
9068 block_group = next_block_group(info->tree_root,
9078 inode = block_group->inode;
9079 block_group->iref = 0;
9080 block_group->inode = NULL;
9081 spin_unlock(&block_group->lock);
9083 last = block_group->key.objectid + block_group->key.offset;
9084 btrfs_put_block_group(block_group);
9088 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9090 struct btrfs_block_group_cache *block_group;
9091 struct btrfs_space_info *space_info;
9092 struct btrfs_caching_control *caching_ctl;
9095 down_write(&info->commit_root_sem);
9096 while (!list_empty(&info->caching_block_groups)) {
9097 caching_ctl = list_entry(info->caching_block_groups.next,
9098 struct btrfs_caching_control, list);
9099 list_del(&caching_ctl->list);
9100 put_caching_control(caching_ctl);
9102 up_write(&info->commit_root_sem);
9104 spin_lock(&info->unused_bgs_lock);
9105 while (!list_empty(&info->unused_bgs)) {
9106 block_group = list_first_entry(&info->unused_bgs,
9107 struct btrfs_block_group_cache,
9109 list_del_init(&block_group->bg_list);
9110 btrfs_put_block_group(block_group);
9112 spin_unlock(&info->unused_bgs_lock);
9114 spin_lock(&info->block_group_cache_lock);
9115 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9116 block_group = rb_entry(n, struct btrfs_block_group_cache,
9118 rb_erase(&block_group->cache_node,
9119 &info->block_group_cache_tree);
9120 RB_CLEAR_NODE(&block_group->cache_node);
9121 spin_unlock(&info->block_group_cache_lock);
9123 down_write(&block_group->space_info->groups_sem);
9124 list_del(&block_group->list);
9125 up_write(&block_group->space_info->groups_sem);
9127 if (block_group->cached == BTRFS_CACHE_STARTED)
9128 wait_block_group_cache_done(block_group);
9131 * We haven't cached this block group, which means we could
9132 * possibly have excluded extents on this block group.
9134 if (block_group->cached == BTRFS_CACHE_NO ||
9135 block_group->cached == BTRFS_CACHE_ERROR)
9136 free_excluded_extents(info->extent_root, block_group);
9138 btrfs_remove_free_space_cache(block_group);
9139 btrfs_put_block_group(block_group);
9141 spin_lock(&info->block_group_cache_lock);
9143 spin_unlock(&info->block_group_cache_lock);
9145 /* now that all the block groups are freed, go through and
9146 * free all the space_info structs. This is only called during
9147 * the final stages of unmount, and so we know nobody is
9148 * using them. We call synchronize_rcu() once before we start,
9149 * just to be on the safe side.
9153 release_global_block_rsv(info);
9155 while (!list_empty(&info->space_info)) {
9158 space_info = list_entry(info->space_info.next,
9159 struct btrfs_space_info,
9161 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9162 if (WARN_ON(space_info->bytes_pinned > 0 ||
9163 space_info->bytes_reserved > 0 ||
9164 space_info->bytes_may_use > 0)) {
9165 dump_space_info(space_info, 0, 0);
9168 list_del(&space_info->list);
9169 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9170 struct kobject *kobj;
9171 kobj = space_info->block_group_kobjs[i];
9172 space_info->block_group_kobjs[i] = NULL;
9178 kobject_del(&space_info->kobj);
9179 kobject_put(&space_info->kobj);
9184 static void __link_block_group(struct btrfs_space_info *space_info,
9185 struct btrfs_block_group_cache *cache)
9187 int index = get_block_group_index(cache);
9190 down_write(&space_info->groups_sem);
9191 if (list_empty(&space_info->block_groups[index]))
9193 list_add_tail(&cache->list, &space_info->block_groups[index]);
9194 up_write(&space_info->groups_sem);
9197 struct raid_kobject *rkobj;
9200 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9203 rkobj->raid_type = index;
9204 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9205 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9206 "%s", get_raid_name(index));
9208 kobject_put(&rkobj->kobj);
9211 space_info->block_group_kobjs[index] = &rkobj->kobj;
9216 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9219 static struct btrfs_block_group_cache *
9220 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9222 struct btrfs_block_group_cache *cache;
9224 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9228 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9230 if (!cache->free_space_ctl) {
9235 cache->key.objectid = start;
9236 cache->key.offset = size;
9237 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9239 cache->sectorsize = root->sectorsize;
9240 cache->fs_info = root->fs_info;
9241 cache->full_stripe_len = btrfs_full_stripe_len(root,
9242 &root->fs_info->mapping_tree,
9244 atomic_set(&cache->count, 1);
9245 spin_lock_init(&cache->lock);
9246 init_rwsem(&cache->data_rwsem);
9247 INIT_LIST_HEAD(&cache->list);
9248 INIT_LIST_HEAD(&cache->cluster_list);
9249 INIT_LIST_HEAD(&cache->bg_list);
9250 INIT_LIST_HEAD(&cache->ro_list);
9251 INIT_LIST_HEAD(&cache->dirty_list);
9252 INIT_LIST_HEAD(&cache->io_list);
9253 btrfs_init_free_space_ctl(cache);
9254 atomic_set(&cache->trimming, 0);
9259 int btrfs_read_block_groups(struct btrfs_root *root)
9261 struct btrfs_path *path;
9263 struct btrfs_block_group_cache *cache;
9264 struct btrfs_fs_info *info = root->fs_info;
9265 struct btrfs_space_info *space_info;
9266 struct btrfs_key key;
9267 struct btrfs_key found_key;
9268 struct extent_buffer *leaf;
9272 root = info->extent_root;
9275 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9276 path = btrfs_alloc_path();
9281 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9282 if (btrfs_test_opt(root, SPACE_CACHE) &&
9283 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9285 if (btrfs_test_opt(root, CLEAR_CACHE))
9289 ret = find_first_block_group(root, path, &key);
9295 leaf = path->nodes[0];
9296 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9298 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9307 * When we mount with old space cache, we need to
9308 * set BTRFS_DC_CLEAR and set dirty flag.
9310 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9311 * truncate the old free space cache inode and
9313 * b) Setting 'dirty flag' makes sure that we flush
9314 * the new space cache info onto disk.
9316 if (btrfs_test_opt(root, SPACE_CACHE))
9317 cache->disk_cache_state = BTRFS_DC_CLEAR;
9320 read_extent_buffer(leaf, &cache->item,
9321 btrfs_item_ptr_offset(leaf, path->slots[0]),
9322 sizeof(cache->item));
9323 cache->flags = btrfs_block_group_flags(&cache->item);
9325 key.objectid = found_key.objectid + found_key.offset;
9326 btrfs_release_path(path);
9329 * We need to exclude the super stripes now so that the space
9330 * info has super bytes accounted for, otherwise we'll think
9331 * we have more space than we actually do.
9333 ret = exclude_super_stripes(root, cache);
9336 * We may have excluded something, so call this just in
9339 free_excluded_extents(root, cache);
9340 btrfs_put_block_group(cache);
9345 * check for two cases, either we are full, and therefore
9346 * don't need to bother with the caching work since we won't
9347 * find any space, or we are empty, and we can just add all
9348 * the space in and be done with it. This saves us _alot_ of
9349 * time, particularly in the full case.
9351 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9352 cache->last_byte_to_unpin = (u64)-1;
9353 cache->cached = BTRFS_CACHE_FINISHED;
9354 free_excluded_extents(root, cache);
9355 } else if (btrfs_block_group_used(&cache->item) == 0) {
9356 cache->last_byte_to_unpin = (u64)-1;
9357 cache->cached = BTRFS_CACHE_FINISHED;
9358 add_new_free_space(cache, root->fs_info,
9360 found_key.objectid +
9362 free_excluded_extents(root, cache);
9365 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9367 btrfs_remove_free_space_cache(cache);
9368 btrfs_put_block_group(cache);
9372 ret = update_space_info(info, cache->flags, found_key.offset,
9373 btrfs_block_group_used(&cache->item),
9376 btrfs_remove_free_space_cache(cache);
9377 spin_lock(&info->block_group_cache_lock);
9378 rb_erase(&cache->cache_node,
9379 &info->block_group_cache_tree);
9380 RB_CLEAR_NODE(&cache->cache_node);
9381 spin_unlock(&info->block_group_cache_lock);
9382 btrfs_put_block_group(cache);
9386 cache->space_info = space_info;
9387 spin_lock(&cache->space_info->lock);
9388 cache->space_info->bytes_readonly += cache->bytes_super;
9389 spin_unlock(&cache->space_info->lock);
9391 __link_block_group(space_info, cache);
9393 set_avail_alloc_bits(root->fs_info, cache->flags);
9394 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9395 set_block_group_ro(cache, 1);
9396 } else if (btrfs_block_group_used(&cache->item) == 0) {
9397 spin_lock(&info->unused_bgs_lock);
9398 /* Should always be true but just in case. */
9399 if (list_empty(&cache->bg_list)) {
9400 btrfs_get_block_group(cache);
9401 list_add_tail(&cache->bg_list,
9404 spin_unlock(&info->unused_bgs_lock);
9408 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9409 if (!(get_alloc_profile(root, space_info->flags) &
9410 (BTRFS_BLOCK_GROUP_RAID10 |
9411 BTRFS_BLOCK_GROUP_RAID1 |
9412 BTRFS_BLOCK_GROUP_RAID5 |
9413 BTRFS_BLOCK_GROUP_RAID6 |
9414 BTRFS_BLOCK_GROUP_DUP)))
9417 * avoid allocating from un-mirrored block group if there are
9418 * mirrored block groups.
9420 list_for_each_entry(cache,
9421 &space_info->block_groups[BTRFS_RAID_RAID0],
9423 set_block_group_ro(cache, 1);
9424 list_for_each_entry(cache,
9425 &space_info->block_groups[BTRFS_RAID_SINGLE],
9427 set_block_group_ro(cache, 1);
9430 init_global_block_rsv(info);
9433 btrfs_free_path(path);
9437 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9438 struct btrfs_root *root)
9440 struct btrfs_block_group_cache *block_group, *tmp;
9441 struct btrfs_root *extent_root = root->fs_info->extent_root;
9442 struct btrfs_block_group_item item;
9443 struct btrfs_key key;
9446 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9450 spin_lock(&block_group->lock);
9451 memcpy(&item, &block_group->item, sizeof(item));
9452 memcpy(&key, &block_group->key, sizeof(key));
9453 spin_unlock(&block_group->lock);
9455 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9458 btrfs_abort_transaction(trans, extent_root, ret);
9459 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9460 key.objectid, key.offset);
9462 btrfs_abort_transaction(trans, extent_root, ret);
9464 list_del_init(&block_group->bg_list);
9468 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9469 struct btrfs_root *root, u64 bytes_used,
9470 u64 type, u64 chunk_objectid, u64 chunk_offset,
9474 struct btrfs_root *extent_root;
9475 struct btrfs_block_group_cache *cache;
9477 extent_root = root->fs_info->extent_root;
9479 btrfs_set_log_full_commit(root->fs_info, trans);
9481 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9485 btrfs_set_block_group_used(&cache->item, bytes_used);
9486 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9487 btrfs_set_block_group_flags(&cache->item, type);
9489 cache->flags = type;
9490 cache->last_byte_to_unpin = (u64)-1;
9491 cache->cached = BTRFS_CACHE_FINISHED;
9492 ret = exclude_super_stripes(root, cache);
9495 * We may have excluded something, so call this just in
9498 free_excluded_extents(root, cache);
9499 btrfs_put_block_group(cache);
9503 add_new_free_space(cache, root->fs_info, chunk_offset,
9504 chunk_offset + size);
9506 free_excluded_extents(root, cache);
9509 * Call to ensure the corresponding space_info object is created and
9510 * assigned to our block group, but don't update its counters just yet.
9511 * We want our bg to be added to the rbtree with its ->space_info set.
9513 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9514 &cache->space_info);
9516 btrfs_remove_free_space_cache(cache);
9517 btrfs_put_block_group(cache);
9521 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9523 btrfs_remove_free_space_cache(cache);
9524 btrfs_put_block_group(cache);
9529 * Now that our block group has its ->space_info set and is inserted in
9530 * the rbtree, update the space info's counters.
9532 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9533 &cache->space_info);
9535 btrfs_remove_free_space_cache(cache);
9536 spin_lock(&root->fs_info->block_group_cache_lock);
9537 rb_erase(&cache->cache_node,
9538 &root->fs_info->block_group_cache_tree);
9539 RB_CLEAR_NODE(&cache->cache_node);
9540 spin_unlock(&root->fs_info->block_group_cache_lock);
9541 btrfs_put_block_group(cache);
9544 update_global_block_rsv(root->fs_info);
9546 spin_lock(&cache->space_info->lock);
9547 cache->space_info->bytes_readonly += cache->bytes_super;
9548 spin_unlock(&cache->space_info->lock);
9550 __link_block_group(cache->space_info, cache);
9552 list_add_tail(&cache->bg_list, &trans->new_bgs);
9554 set_avail_alloc_bits(extent_root->fs_info, type);
9559 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9561 u64 extra_flags = chunk_to_extended(flags) &
9562 BTRFS_EXTENDED_PROFILE_MASK;
9564 write_seqlock(&fs_info->profiles_lock);
9565 if (flags & BTRFS_BLOCK_GROUP_DATA)
9566 fs_info->avail_data_alloc_bits &= ~extra_flags;
9567 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9568 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9569 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9570 fs_info->avail_system_alloc_bits &= ~extra_flags;
9571 write_sequnlock(&fs_info->profiles_lock);
9574 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9575 struct btrfs_root *root, u64 group_start,
9576 struct extent_map *em)
9578 struct btrfs_path *path;
9579 struct btrfs_block_group_cache *block_group;
9580 struct btrfs_free_cluster *cluster;
9581 struct btrfs_root *tree_root = root->fs_info->tree_root;
9582 struct btrfs_key key;
9583 struct inode *inode;
9584 struct kobject *kobj = NULL;
9588 struct btrfs_caching_control *caching_ctl = NULL;
9591 root = root->fs_info->extent_root;
9593 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9594 BUG_ON(!block_group);
9595 BUG_ON(!block_group->ro);
9598 * Free the reserved super bytes from this block group before
9601 free_excluded_extents(root, block_group);
9603 memcpy(&key, &block_group->key, sizeof(key));
9604 index = get_block_group_index(block_group);
9605 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9606 BTRFS_BLOCK_GROUP_RAID1 |
9607 BTRFS_BLOCK_GROUP_RAID10))
9612 /* make sure this block group isn't part of an allocation cluster */
9613 cluster = &root->fs_info->data_alloc_cluster;
9614 spin_lock(&cluster->refill_lock);
9615 btrfs_return_cluster_to_free_space(block_group, cluster);
9616 spin_unlock(&cluster->refill_lock);
9619 * make sure this block group isn't part of a metadata
9620 * allocation cluster
9622 cluster = &root->fs_info->meta_alloc_cluster;
9623 spin_lock(&cluster->refill_lock);
9624 btrfs_return_cluster_to_free_space(block_group, cluster);
9625 spin_unlock(&cluster->refill_lock);
9627 path = btrfs_alloc_path();
9634 * get the inode first so any iput calls done for the io_list
9635 * aren't the final iput (no unlinks allowed now)
9637 inode = lookup_free_space_inode(tree_root, block_group, path);
9639 mutex_lock(&trans->transaction->cache_write_mutex);
9641 * make sure our free spache cache IO is done before remove the
9644 spin_lock(&trans->transaction->dirty_bgs_lock);
9645 if (!list_empty(&block_group->io_list)) {
9646 list_del_init(&block_group->io_list);
9648 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9650 spin_unlock(&trans->transaction->dirty_bgs_lock);
9651 btrfs_wait_cache_io(root, trans, block_group,
9652 &block_group->io_ctl, path,
9653 block_group->key.objectid);
9654 btrfs_put_block_group(block_group);
9655 spin_lock(&trans->transaction->dirty_bgs_lock);
9658 if (!list_empty(&block_group->dirty_list)) {
9659 list_del_init(&block_group->dirty_list);
9660 btrfs_put_block_group(block_group);
9662 spin_unlock(&trans->transaction->dirty_bgs_lock);
9663 mutex_unlock(&trans->transaction->cache_write_mutex);
9665 if (!IS_ERR(inode)) {
9666 ret = btrfs_orphan_add(trans, inode);
9668 btrfs_add_delayed_iput(inode);
9672 /* One for the block groups ref */
9673 spin_lock(&block_group->lock);
9674 if (block_group->iref) {
9675 block_group->iref = 0;
9676 block_group->inode = NULL;
9677 spin_unlock(&block_group->lock);
9680 spin_unlock(&block_group->lock);
9682 /* One for our lookup ref */
9683 btrfs_add_delayed_iput(inode);
9686 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9687 key.offset = block_group->key.objectid;
9690 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9694 btrfs_release_path(path);
9696 ret = btrfs_del_item(trans, tree_root, path);
9699 btrfs_release_path(path);
9702 spin_lock(&root->fs_info->block_group_cache_lock);
9703 rb_erase(&block_group->cache_node,
9704 &root->fs_info->block_group_cache_tree);
9705 RB_CLEAR_NODE(&block_group->cache_node);
9707 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9708 root->fs_info->first_logical_byte = (u64)-1;
9709 spin_unlock(&root->fs_info->block_group_cache_lock);
9711 down_write(&block_group->space_info->groups_sem);
9713 * we must use list_del_init so people can check to see if they
9714 * are still on the list after taking the semaphore
9716 list_del_init(&block_group->list);
9717 if (list_empty(&block_group->space_info->block_groups[index])) {
9718 kobj = block_group->space_info->block_group_kobjs[index];
9719 block_group->space_info->block_group_kobjs[index] = NULL;
9720 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9722 up_write(&block_group->space_info->groups_sem);
9728 if (block_group->has_caching_ctl)
9729 caching_ctl = get_caching_control(block_group);
9730 if (block_group->cached == BTRFS_CACHE_STARTED)
9731 wait_block_group_cache_done(block_group);
9732 if (block_group->has_caching_ctl) {
9733 down_write(&root->fs_info->commit_root_sem);
9735 struct btrfs_caching_control *ctl;
9737 list_for_each_entry(ctl,
9738 &root->fs_info->caching_block_groups, list)
9739 if (ctl->block_group == block_group) {
9741 atomic_inc(&caching_ctl->count);
9746 list_del_init(&caching_ctl->list);
9747 up_write(&root->fs_info->commit_root_sem);
9749 /* Once for the caching bgs list and once for us. */
9750 put_caching_control(caching_ctl);
9751 put_caching_control(caching_ctl);
9755 spin_lock(&trans->transaction->dirty_bgs_lock);
9756 if (!list_empty(&block_group->dirty_list)) {
9759 if (!list_empty(&block_group->io_list)) {
9762 spin_unlock(&trans->transaction->dirty_bgs_lock);
9763 btrfs_remove_free_space_cache(block_group);
9765 spin_lock(&block_group->space_info->lock);
9766 list_del_init(&block_group->ro_list);
9768 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9769 WARN_ON(block_group->space_info->total_bytes
9770 < block_group->key.offset);
9771 WARN_ON(block_group->space_info->bytes_readonly
9772 < block_group->key.offset);
9773 WARN_ON(block_group->space_info->disk_total
9774 < block_group->key.offset * factor);
9776 block_group->space_info->total_bytes -= block_group->key.offset;
9777 block_group->space_info->bytes_readonly -= block_group->key.offset;
9778 block_group->space_info->disk_total -= block_group->key.offset * factor;
9780 spin_unlock(&block_group->space_info->lock);
9782 memcpy(&key, &block_group->key, sizeof(key));
9785 if (!list_empty(&em->list)) {
9786 /* We're in the transaction->pending_chunks list. */
9787 free_extent_map(em);
9789 spin_lock(&block_group->lock);
9790 block_group->removed = 1;
9792 * At this point trimming can't start on this block group, because we
9793 * removed the block group from the tree fs_info->block_group_cache_tree
9794 * so no one can't find it anymore and even if someone already got this
9795 * block group before we removed it from the rbtree, they have already
9796 * incremented block_group->trimming - if they didn't, they won't find
9797 * any free space entries because we already removed them all when we
9798 * called btrfs_remove_free_space_cache().
9800 * And we must not remove the extent map from the fs_info->mapping_tree
9801 * to prevent the same logical address range and physical device space
9802 * ranges from being reused for a new block group. This is because our
9803 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9804 * completely transactionless, so while it is trimming a range the
9805 * currently running transaction might finish and a new one start,
9806 * allowing for new block groups to be created that can reuse the same
9807 * physical device locations unless we take this special care.
9809 remove_em = (atomic_read(&block_group->trimming) == 0);
9811 * Make sure a trimmer task always sees the em in the pinned_chunks list
9812 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9813 * before checking block_group->removed).
9817 * Our em might be in trans->transaction->pending_chunks which
9818 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9819 * and so is the fs_info->pinned_chunks list.
9821 * So at this point we must be holding the chunk_mutex to avoid
9822 * any races with chunk allocation (more specifically at
9823 * volumes.c:contains_pending_extent()), to ensure it always
9824 * sees the em, either in the pending_chunks list or in the
9825 * pinned_chunks list.
9827 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9829 spin_unlock(&block_group->lock);
9832 struct extent_map_tree *em_tree;
9834 em_tree = &root->fs_info->mapping_tree.map_tree;
9835 write_lock(&em_tree->lock);
9837 * The em might be in the pending_chunks list, so make sure the
9838 * chunk mutex is locked, since remove_extent_mapping() will
9839 * delete us from that list.
9841 remove_extent_mapping(em_tree, em);
9842 write_unlock(&em_tree->lock);
9843 /* once for the tree */
9844 free_extent_map(em);
9847 unlock_chunks(root);
9849 btrfs_put_block_group(block_group);
9850 btrfs_put_block_group(block_group);
9852 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9858 ret = btrfs_del_item(trans, root, path);
9860 btrfs_free_path(path);
9865 * Process the unused_bgs list and remove any that don't have any allocated
9866 * space inside of them.
9868 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9870 struct btrfs_block_group_cache *block_group;
9871 struct btrfs_space_info *space_info;
9872 struct btrfs_root *root = fs_info->extent_root;
9873 struct btrfs_trans_handle *trans;
9879 spin_lock(&fs_info->unused_bgs_lock);
9880 while (!list_empty(&fs_info->unused_bgs)) {
9883 block_group = list_first_entry(&fs_info->unused_bgs,
9884 struct btrfs_block_group_cache,
9886 space_info = block_group->space_info;
9887 list_del_init(&block_group->bg_list);
9888 if (ret || btrfs_mixed_space_info(space_info)) {
9889 btrfs_put_block_group(block_group);
9892 spin_unlock(&fs_info->unused_bgs_lock);
9894 /* Don't want to race with allocators so take the groups_sem */
9895 down_write(&space_info->groups_sem);
9896 spin_lock(&block_group->lock);
9897 if (block_group->reserved ||
9898 btrfs_block_group_used(&block_group->item) ||
9901 * We want to bail if we made new allocations or have
9902 * outstanding allocations in this block group. We do
9903 * the ro check in case balance is currently acting on
9906 spin_unlock(&block_group->lock);
9907 up_write(&space_info->groups_sem);
9910 spin_unlock(&block_group->lock);
9912 /* We don't want to force the issue, only flip if it's ok. */
9913 ret = set_block_group_ro(block_group, 0);
9914 up_write(&space_info->groups_sem);
9921 * Want to do this before we do anything else so we can recover
9922 * properly if we fail to join the transaction.
9924 /* 1 for btrfs_orphan_reserve_metadata() */
9925 trans = btrfs_start_transaction(root, 1);
9926 if (IS_ERR(trans)) {
9927 btrfs_set_block_group_rw(root, block_group);
9928 ret = PTR_ERR(trans);
9933 * We could have pending pinned extents for this block group,
9934 * just delete them, we don't care about them anymore.
9936 start = block_group->key.objectid;
9937 end = start + block_group->key.offset - 1;
9939 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9940 * btrfs_finish_extent_commit(). If we are at transaction N,
9941 * another task might be running finish_extent_commit() for the
9942 * previous transaction N - 1, and have seen a range belonging
9943 * to the block group in freed_extents[] before we were able to
9944 * clear the whole block group range from freed_extents[]. This
9945 * means that task can lookup for the block group after we
9946 * unpinned it from freed_extents[] and removed it, leading to
9947 * a BUG_ON() at btrfs_unpin_extent_range().
9949 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9950 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9951 EXTENT_DIRTY, GFP_NOFS);
9953 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9954 btrfs_set_block_group_rw(root, block_group);
9957 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9958 EXTENT_DIRTY, GFP_NOFS);
9960 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9961 btrfs_set_block_group_rw(root, block_group);
9964 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9966 /* Reset pinned so btrfs_put_block_group doesn't complain */
9967 spin_lock(&space_info->lock);
9968 spin_lock(&block_group->lock);
9970 space_info->bytes_pinned -= block_group->pinned;
9971 space_info->bytes_readonly += block_group->pinned;
9972 percpu_counter_add(&space_info->total_bytes_pinned,
9973 -block_group->pinned);
9974 block_group->pinned = 0;
9976 spin_unlock(&block_group->lock);
9977 spin_unlock(&space_info->lock);
9980 * Btrfs_remove_chunk will abort the transaction if things go
9983 ret = btrfs_remove_chunk(trans, root,
9984 block_group->key.objectid);
9986 btrfs_end_transaction(trans, root);
9988 btrfs_put_block_group(block_group);
9989 spin_lock(&fs_info->unused_bgs_lock);
9991 spin_unlock(&fs_info->unused_bgs_lock);
9994 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9996 struct btrfs_space_info *space_info;
9997 struct btrfs_super_block *disk_super;
10003 disk_super = fs_info->super_copy;
10004 if (!btrfs_super_root(disk_super))
10007 features = btrfs_super_incompat_flags(disk_super);
10008 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10011 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10012 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10017 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10018 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10020 flags = BTRFS_BLOCK_GROUP_METADATA;
10021 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10025 flags = BTRFS_BLOCK_GROUP_DATA;
10026 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10032 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10034 return unpin_extent_range(root, start, end, false);
10037 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10039 struct btrfs_fs_info *fs_info = root->fs_info;
10040 struct btrfs_block_group_cache *cache = NULL;
10045 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10049 * try to trim all FS space, our block group may start from non-zero.
10051 if (range->len == total_bytes)
10052 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10054 cache = btrfs_lookup_block_group(fs_info, range->start);
10057 if (cache->key.objectid >= (range->start + range->len)) {
10058 btrfs_put_block_group(cache);
10062 start = max(range->start, cache->key.objectid);
10063 end = min(range->start + range->len,
10064 cache->key.objectid + cache->key.offset);
10066 if (end - start >= range->minlen) {
10067 if (!block_group_cache_done(cache)) {
10068 ret = cache_block_group(cache, 0);
10070 btrfs_put_block_group(cache);
10073 ret = wait_block_group_cache_done(cache);
10075 btrfs_put_block_group(cache);
10079 ret = btrfs_trim_block_group(cache,
10085 trimmed += group_trimmed;
10087 btrfs_put_block_group(cache);
10092 cache = next_block_group(fs_info->tree_root, cache);
10095 range->len = trimmed;
10100 * btrfs_{start,end}_write_no_snapshoting() are similar to
10101 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10102 * data into the page cache through nocow before the subvolume is snapshoted,
10103 * but flush the data into disk after the snapshot creation, or to prevent
10104 * operations while snapshoting is ongoing and that cause the snapshot to be
10105 * inconsistent (writes followed by expanding truncates for example).
10107 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10109 percpu_counter_dec(&root->subv_writers->counter);
10111 * Make sure counter is updated before we wake up
10115 if (waitqueue_active(&root->subv_writers->wait))
10116 wake_up(&root->subv_writers->wait);
10119 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10121 if (atomic_read(&root->will_be_snapshoted))
10124 percpu_counter_inc(&root->subv_writers->counter);
10126 * Make sure counter is updated before we check for snapshot creation.
10129 if (atomic_read(&root->will_be_snapshoted)) {
10130 btrfs_end_write_no_snapshoting(root);