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;
1985 path = btrfs_alloc_path();
1989 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1993 path->leave_spinning = 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 bytenr, num_bytes, parent,
1997 root_objectid, owner, offset,
1998 refs_to_add, extent_op);
1999 if ((ret < 0 && ret != -EAGAIN) || !ret)
2003 * Ok we had -EAGAIN which means we didn't have space to insert and
2004 * inline extent ref, so just update the reference count and add a
2007 leaf = path->nodes[0];
2008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2010 refs = btrfs_extent_refs(leaf, item);
2011 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2013 __run_delayed_extent_op(extent_op, leaf, item);
2015 btrfs_mark_buffer_dirty(leaf);
2016 btrfs_release_path(path);
2019 path->leave_spinning = 1;
2020 /* now insert the actual backref */
2021 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2022 path, bytenr, parent, root_objectid,
2023 owner, offset, refs_to_add);
2025 btrfs_abort_transaction(trans, root, ret);
2027 btrfs_free_path(path);
2031 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2032 struct btrfs_root *root,
2033 struct btrfs_delayed_ref_node *node,
2034 struct btrfs_delayed_extent_op *extent_op,
2035 int insert_reserved)
2038 struct btrfs_delayed_data_ref *ref;
2039 struct btrfs_key ins;
2044 ins.objectid = node->bytenr;
2045 ins.offset = node->num_bytes;
2046 ins.type = BTRFS_EXTENT_ITEM_KEY;
2048 ref = btrfs_delayed_node_to_data_ref(node);
2049 trace_run_delayed_data_ref(node, ref, node->action);
2051 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2052 parent = ref->parent;
2053 ref_root = ref->root;
2055 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2057 flags |= extent_op->flags_to_set;
2058 ret = alloc_reserved_file_extent(trans, root,
2059 parent, ref_root, flags,
2060 ref->objectid, ref->offset,
2061 &ins, node->ref_mod);
2062 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2063 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2064 ref_root, ref->objectid,
2065 ref->offset, node->ref_mod,
2067 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2068 ret = __btrfs_free_extent(trans, root, node, parent,
2069 ref_root, ref->objectid,
2070 ref->offset, node->ref_mod,
2078 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2079 struct extent_buffer *leaf,
2080 struct btrfs_extent_item *ei)
2082 u64 flags = btrfs_extent_flags(leaf, ei);
2083 if (extent_op->update_flags) {
2084 flags |= extent_op->flags_to_set;
2085 btrfs_set_extent_flags(leaf, ei, flags);
2088 if (extent_op->update_key) {
2089 struct btrfs_tree_block_info *bi;
2090 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2091 bi = (struct btrfs_tree_block_info *)(ei + 1);
2092 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2096 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *root,
2098 struct btrfs_delayed_ref_node *node,
2099 struct btrfs_delayed_extent_op *extent_op)
2101 struct btrfs_key key;
2102 struct btrfs_path *path;
2103 struct btrfs_extent_item *ei;
2104 struct extent_buffer *leaf;
2108 int metadata = !extent_op->is_data;
2113 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2116 path = btrfs_alloc_path();
2120 key.objectid = node->bytenr;
2123 key.type = BTRFS_METADATA_ITEM_KEY;
2124 key.offset = extent_op->level;
2126 key.type = BTRFS_EXTENT_ITEM_KEY;
2127 key.offset = node->num_bytes;
2132 path->leave_spinning = 1;
2133 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2141 if (path->slots[0] > 0) {
2143 btrfs_item_key_to_cpu(path->nodes[0], &key,
2145 if (key.objectid == node->bytenr &&
2146 key.type == BTRFS_EXTENT_ITEM_KEY &&
2147 key.offset == node->num_bytes)
2151 btrfs_release_path(path);
2154 key.objectid = node->bytenr;
2155 key.offset = node->num_bytes;
2156 key.type = BTRFS_EXTENT_ITEM_KEY;
2165 leaf = path->nodes[0];
2166 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2167 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2168 if (item_size < sizeof(*ei)) {
2169 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2175 leaf = path->nodes[0];
2176 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2179 BUG_ON(item_size < sizeof(*ei));
2180 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2181 __run_delayed_extent_op(extent_op, leaf, ei);
2183 btrfs_mark_buffer_dirty(leaf);
2185 btrfs_free_path(path);
2189 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_delayed_ref_node *node,
2192 struct btrfs_delayed_extent_op *extent_op,
2193 int insert_reserved)
2196 struct btrfs_delayed_tree_ref *ref;
2197 struct btrfs_key ins;
2200 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2203 ref = btrfs_delayed_node_to_tree_ref(node);
2204 trace_run_delayed_tree_ref(node, ref, node->action);
2206 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2207 parent = ref->parent;
2208 ref_root = ref->root;
2210 ins.objectid = node->bytenr;
2211 if (skinny_metadata) {
2212 ins.offset = ref->level;
2213 ins.type = BTRFS_METADATA_ITEM_KEY;
2215 ins.offset = node->num_bytes;
2216 ins.type = BTRFS_EXTENT_ITEM_KEY;
2219 BUG_ON(node->ref_mod != 1);
2220 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2221 BUG_ON(!extent_op || !extent_op->update_flags);
2222 ret = alloc_reserved_tree_block(trans, root,
2224 extent_op->flags_to_set,
2228 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2229 ret = __btrfs_inc_extent_ref(trans, root, node,
2233 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2234 ret = __btrfs_free_extent(trans, root, node,
2236 ref->level, 0, 1, extent_op);
2243 /* helper function to actually process a single delayed ref entry */
2244 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *root,
2246 struct btrfs_delayed_ref_node *node,
2247 struct btrfs_delayed_extent_op *extent_op,
2248 int insert_reserved)
2252 if (trans->aborted) {
2253 if (insert_reserved)
2254 btrfs_pin_extent(root, node->bytenr,
2255 node->num_bytes, 1);
2259 if (btrfs_delayed_ref_is_head(node)) {
2260 struct btrfs_delayed_ref_head *head;
2262 * we've hit the end of the chain and we were supposed
2263 * to insert this extent into the tree. But, it got
2264 * deleted before we ever needed to insert it, so all
2265 * we have to do is clean up the accounting
2268 head = btrfs_delayed_node_to_head(node);
2269 trace_run_delayed_ref_head(node, head, node->action);
2271 if (insert_reserved) {
2272 btrfs_pin_extent(root, node->bytenr,
2273 node->num_bytes, 1);
2274 if (head->is_data) {
2275 ret = btrfs_del_csums(trans, root,
2283 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2284 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2285 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2287 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2288 node->type == BTRFS_SHARED_DATA_REF_KEY)
2289 ret = run_delayed_data_ref(trans, root, node, extent_op,
2296 static inline struct btrfs_delayed_ref_node *
2297 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2299 struct btrfs_delayed_ref_node *ref;
2301 if (list_empty(&head->ref_list))
2305 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2306 * This is to prevent a ref count from going down to zero, which deletes
2307 * the extent item from the extent tree, when there still are references
2308 * to add, which would fail because they would not find the extent item.
2310 list_for_each_entry(ref, &head->ref_list, list) {
2311 if (ref->action == BTRFS_ADD_DELAYED_REF)
2315 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2320 * Returns 0 on success or if called with an already aborted transaction.
2321 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2323 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2324 struct btrfs_root *root,
2327 struct btrfs_delayed_ref_root *delayed_refs;
2328 struct btrfs_delayed_ref_node *ref;
2329 struct btrfs_delayed_ref_head *locked_ref = NULL;
2330 struct btrfs_delayed_extent_op *extent_op;
2331 struct btrfs_fs_info *fs_info = root->fs_info;
2332 ktime_t start = ktime_get();
2334 unsigned long count = 0;
2335 unsigned long actual_count = 0;
2336 int must_insert_reserved = 0;
2338 delayed_refs = &trans->transaction->delayed_refs;
2344 spin_lock(&delayed_refs->lock);
2345 locked_ref = btrfs_select_ref_head(trans);
2347 spin_unlock(&delayed_refs->lock);
2351 /* grab the lock that says we are going to process
2352 * all the refs for this head */
2353 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2354 spin_unlock(&delayed_refs->lock);
2356 * we may have dropped the spin lock to get the head
2357 * mutex lock, and that might have given someone else
2358 * time to free the head. If that's true, it has been
2359 * removed from our list and we can move on.
2361 if (ret == -EAGAIN) {
2368 spin_lock(&locked_ref->lock);
2371 * locked_ref is the head node, so we have to go one
2372 * node back for any delayed ref updates
2374 ref = select_delayed_ref(locked_ref);
2376 if (ref && ref->seq &&
2377 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2378 spin_unlock(&locked_ref->lock);
2379 btrfs_delayed_ref_unlock(locked_ref);
2380 spin_lock(&delayed_refs->lock);
2381 locked_ref->processing = 0;
2382 delayed_refs->num_heads_ready++;
2383 spin_unlock(&delayed_refs->lock);
2391 * record the must insert reserved flag before we
2392 * drop the spin lock.
2394 must_insert_reserved = locked_ref->must_insert_reserved;
2395 locked_ref->must_insert_reserved = 0;
2397 extent_op = locked_ref->extent_op;
2398 locked_ref->extent_op = NULL;
2403 /* All delayed refs have been processed, Go ahead
2404 * and send the head node to run_one_delayed_ref,
2405 * so that any accounting fixes can happen
2407 ref = &locked_ref->node;
2409 if (extent_op && must_insert_reserved) {
2410 btrfs_free_delayed_extent_op(extent_op);
2415 spin_unlock(&locked_ref->lock);
2416 ret = run_delayed_extent_op(trans, root,
2418 btrfs_free_delayed_extent_op(extent_op);
2422 * Need to reset must_insert_reserved if
2423 * there was an error so the abort stuff
2424 * can cleanup the reserved space
2427 if (must_insert_reserved)
2428 locked_ref->must_insert_reserved = 1;
2429 locked_ref->processing = 0;
2430 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2431 btrfs_delayed_ref_unlock(locked_ref);
2438 * Need to drop our head ref lock and re-aqcuire the
2439 * delayed ref lock and then re-check to make sure
2442 spin_unlock(&locked_ref->lock);
2443 spin_lock(&delayed_refs->lock);
2444 spin_lock(&locked_ref->lock);
2445 if (!list_empty(&locked_ref->ref_list) ||
2446 locked_ref->extent_op) {
2447 spin_unlock(&locked_ref->lock);
2448 spin_unlock(&delayed_refs->lock);
2452 delayed_refs->num_heads--;
2453 rb_erase(&locked_ref->href_node,
2454 &delayed_refs->href_root);
2455 spin_unlock(&delayed_refs->lock);
2459 list_del(&ref->list);
2461 atomic_dec(&delayed_refs->num_entries);
2463 if (!btrfs_delayed_ref_is_head(ref)) {
2465 * when we play the delayed ref, also correct the
2468 switch (ref->action) {
2469 case BTRFS_ADD_DELAYED_REF:
2470 case BTRFS_ADD_DELAYED_EXTENT:
2471 locked_ref->node.ref_mod -= ref->ref_mod;
2473 case BTRFS_DROP_DELAYED_REF:
2474 locked_ref->node.ref_mod += ref->ref_mod;
2480 spin_unlock(&locked_ref->lock);
2482 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2483 must_insert_reserved);
2485 btrfs_free_delayed_extent_op(extent_op);
2487 locked_ref->processing = 0;
2488 btrfs_delayed_ref_unlock(locked_ref);
2489 btrfs_put_delayed_ref(ref);
2490 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2495 * If this node is a head, that means all the refs in this head
2496 * have been dealt with, and we will pick the next head to deal
2497 * with, so we must unlock the head and drop it from the cluster
2498 * list before we release it.
2500 if (btrfs_delayed_ref_is_head(ref)) {
2501 if (locked_ref->is_data &&
2502 locked_ref->total_ref_mod < 0) {
2503 spin_lock(&delayed_refs->lock);
2504 delayed_refs->pending_csums -= ref->num_bytes;
2505 spin_unlock(&delayed_refs->lock);
2507 btrfs_delayed_ref_unlock(locked_ref);
2510 btrfs_put_delayed_ref(ref);
2516 * We don't want to include ref heads since we can have empty ref heads
2517 * and those will drastically skew our runtime down since we just do
2518 * accounting, no actual extent tree updates.
2520 if (actual_count > 0) {
2521 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2525 * We weigh the current average higher than our current runtime
2526 * to avoid large swings in the average.
2528 spin_lock(&delayed_refs->lock);
2529 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2530 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2531 spin_unlock(&delayed_refs->lock);
2536 #ifdef SCRAMBLE_DELAYED_REFS
2538 * Normally delayed refs get processed in ascending bytenr order. This
2539 * correlates in most cases to the order added. To expose dependencies on this
2540 * order, we start to process the tree in the middle instead of the beginning
2542 static u64 find_middle(struct rb_root *root)
2544 struct rb_node *n = root->rb_node;
2545 struct btrfs_delayed_ref_node *entry;
2548 u64 first = 0, last = 0;
2552 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2553 first = entry->bytenr;
2557 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2558 last = entry->bytenr;
2563 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2564 WARN_ON(!entry->in_tree);
2566 middle = entry->bytenr;
2579 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2583 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2584 sizeof(struct btrfs_extent_inline_ref));
2585 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2586 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2589 * We don't ever fill up leaves all the way so multiply by 2 just to be
2590 * closer to what we're really going to want to ouse.
2592 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2596 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2597 * would require to store the csums for that many bytes.
2599 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2602 u64 num_csums_per_leaf;
2605 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2606 num_csums_per_leaf = div64_u64(csum_size,
2607 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2608 num_csums = div64_u64(csum_bytes, root->sectorsize);
2609 num_csums += num_csums_per_leaf - 1;
2610 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2614 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2615 struct btrfs_root *root)
2617 struct btrfs_block_rsv *global_rsv;
2618 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2619 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2620 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2621 u64 num_bytes, num_dirty_bgs_bytes;
2624 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2625 num_heads = heads_to_leaves(root, num_heads);
2627 num_bytes += (num_heads - 1) * root->nodesize;
2629 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2630 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2632 global_rsv = &root->fs_info->global_block_rsv;
2635 * If we can't allocate any more chunks lets make sure we have _lots_ of
2636 * wiggle room since running delayed refs can create more delayed refs.
2638 if (global_rsv->space_info->full) {
2639 num_dirty_bgs_bytes <<= 1;
2643 spin_lock(&global_rsv->lock);
2644 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2646 spin_unlock(&global_rsv->lock);
2650 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2651 struct btrfs_root *root)
2653 struct btrfs_fs_info *fs_info = root->fs_info;
2655 atomic_read(&trans->transaction->delayed_refs.num_entries);
2660 avg_runtime = fs_info->avg_delayed_ref_runtime;
2661 val = num_entries * avg_runtime;
2662 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2664 if (val >= NSEC_PER_SEC / 2)
2667 return btrfs_check_space_for_delayed_refs(trans, root);
2670 struct async_delayed_refs {
2671 struct btrfs_root *root;
2675 struct completion wait;
2676 struct btrfs_work work;
2679 static void delayed_ref_async_start(struct btrfs_work *work)
2681 struct async_delayed_refs *async;
2682 struct btrfs_trans_handle *trans;
2685 async = container_of(work, struct async_delayed_refs, work);
2687 trans = btrfs_join_transaction(async->root);
2688 if (IS_ERR(trans)) {
2689 async->error = PTR_ERR(trans);
2694 * trans->sync means that when we call end_transaciton, we won't
2695 * wait on delayed refs
2698 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2702 ret = btrfs_end_transaction(trans, async->root);
2703 if (ret && !async->error)
2707 complete(&async->wait);
2712 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2713 unsigned long count, int wait)
2715 struct async_delayed_refs *async;
2718 async = kmalloc(sizeof(*async), GFP_NOFS);
2722 async->root = root->fs_info->tree_root;
2723 async->count = count;
2729 init_completion(&async->wait);
2731 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2732 delayed_ref_async_start, NULL, NULL);
2734 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2737 wait_for_completion(&async->wait);
2746 * this starts processing the delayed reference count updates and
2747 * extent insertions we have queued up so far. count can be
2748 * 0, which means to process everything in the tree at the start
2749 * of the run (but not newly added entries), or it can be some target
2750 * number you'd like to process.
2752 * Returns 0 on success or if called with an aborted transaction
2753 * Returns <0 on error and aborts the transaction
2755 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2756 struct btrfs_root *root, unsigned long count)
2758 struct rb_node *node;
2759 struct btrfs_delayed_ref_root *delayed_refs;
2760 struct btrfs_delayed_ref_head *head;
2762 int run_all = count == (unsigned long)-1;
2764 /* We'll clean this up in btrfs_cleanup_transaction */
2768 if (root == root->fs_info->extent_root)
2769 root = root->fs_info->tree_root;
2771 delayed_refs = &trans->transaction->delayed_refs;
2773 count = atomic_read(&delayed_refs->num_entries) * 2;
2776 #ifdef SCRAMBLE_DELAYED_REFS
2777 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2779 ret = __btrfs_run_delayed_refs(trans, root, count);
2781 btrfs_abort_transaction(trans, root, ret);
2786 if (!list_empty(&trans->new_bgs))
2787 btrfs_create_pending_block_groups(trans, root);
2789 spin_lock(&delayed_refs->lock);
2790 node = rb_first(&delayed_refs->href_root);
2792 spin_unlock(&delayed_refs->lock);
2795 count = (unsigned long)-1;
2798 head = rb_entry(node, struct btrfs_delayed_ref_head,
2800 if (btrfs_delayed_ref_is_head(&head->node)) {
2801 struct btrfs_delayed_ref_node *ref;
2804 atomic_inc(&ref->refs);
2806 spin_unlock(&delayed_refs->lock);
2808 * Mutex was contended, block until it's
2809 * released and try again
2811 mutex_lock(&head->mutex);
2812 mutex_unlock(&head->mutex);
2814 btrfs_put_delayed_ref(ref);
2820 node = rb_next(node);
2822 spin_unlock(&delayed_refs->lock);
2827 assert_qgroups_uptodate(trans);
2831 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2832 struct btrfs_root *root,
2833 u64 bytenr, u64 num_bytes, u64 flags,
2834 int level, int is_data)
2836 struct btrfs_delayed_extent_op *extent_op;
2839 extent_op = btrfs_alloc_delayed_extent_op();
2843 extent_op->flags_to_set = flags;
2844 extent_op->update_flags = 1;
2845 extent_op->update_key = 0;
2846 extent_op->is_data = is_data ? 1 : 0;
2847 extent_op->level = level;
2849 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2850 num_bytes, extent_op);
2852 btrfs_free_delayed_extent_op(extent_op);
2856 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path,
2859 u64 objectid, u64 offset, u64 bytenr)
2861 struct btrfs_delayed_ref_head *head;
2862 struct btrfs_delayed_ref_node *ref;
2863 struct btrfs_delayed_data_ref *data_ref;
2864 struct btrfs_delayed_ref_root *delayed_refs;
2867 delayed_refs = &trans->transaction->delayed_refs;
2868 spin_lock(&delayed_refs->lock);
2869 head = btrfs_find_delayed_ref_head(trans, bytenr);
2871 spin_unlock(&delayed_refs->lock);
2875 if (!mutex_trylock(&head->mutex)) {
2876 atomic_inc(&head->node.refs);
2877 spin_unlock(&delayed_refs->lock);
2879 btrfs_release_path(path);
2882 * Mutex was contended, block until it's released and let
2885 mutex_lock(&head->mutex);
2886 mutex_unlock(&head->mutex);
2887 btrfs_put_delayed_ref(&head->node);
2890 spin_unlock(&delayed_refs->lock);
2892 spin_lock(&head->lock);
2893 list_for_each_entry(ref, &head->ref_list, list) {
2894 /* If it's a shared ref we know a cross reference exists */
2895 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2900 data_ref = btrfs_delayed_node_to_data_ref(ref);
2903 * If our ref doesn't match the one we're currently looking at
2904 * then we have a cross reference.
2906 if (data_ref->root != root->root_key.objectid ||
2907 data_ref->objectid != objectid ||
2908 data_ref->offset != offset) {
2913 spin_unlock(&head->lock);
2914 mutex_unlock(&head->mutex);
2918 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2919 struct btrfs_root *root,
2920 struct btrfs_path *path,
2921 u64 objectid, u64 offset, u64 bytenr)
2923 struct btrfs_root *extent_root = root->fs_info->extent_root;
2924 struct extent_buffer *leaf;
2925 struct btrfs_extent_data_ref *ref;
2926 struct btrfs_extent_inline_ref *iref;
2927 struct btrfs_extent_item *ei;
2928 struct btrfs_key key;
2932 key.objectid = bytenr;
2933 key.offset = (u64)-1;
2934 key.type = BTRFS_EXTENT_ITEM_KEY;
2936 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2939 BUG_ON(ret == 0); /* Corruption */
2942 if (path->slots[0] == 0)
2946 leaf = path->nodes[0];
2947 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2949 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2953 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2955 if (item_size < sizeof(*ei)) {
2956 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2960 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2962 if (item_size != sizeof(*ei) +
2963 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2966 if (btrfs_extent_generation(leaf, ei) <=
2967 btrfs_root_last_snapshot(&root->root_item))
2970 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2971 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2972 BTRFS_EXTENT_DATA_REF_KEY)
2975 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2976 if (btrfs_extent_refs(leaf, ei) !=
2977 btrfs_extent_data_ref_count(leaf, ref) ||
2978 btrfs_extent_data_ref_root(leaf, ref) !=
2979 root->root_key.objectid ||
2980 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2981 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2989 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2990 struct btrfs_root *root,
2991 u64 objectid, u64 offset, u64 bytenr)
2993 struct btrfs_path *path;
2997 path = btrfs_alloc_path();
3002 ret = check_committed_ref(trans, root, path, objectid,
3004 if (ret && ret != -ENOENT)
3007 ret2 = check_delayed_ref(trans, root, path, objectid,
3009 } while (ret2 == -EAGAIN);
3011 if (ret2 && ret2 != -ENOENT) {
3016 if (ret != -ENOENT || ret2 != -ENOENT)
3019 btrfs_free_path(path);
3020 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3025 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3026 struct btrfs_root *root,
3027 struct extent_buffer *buf,
3028 int full_backref, int inc)
3035 struct btrfs_key key;
3036 struct btrfs_file_extent_item *fi;
3040 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3041 u64, u64, u64, u64, u64, u64, int);
3044 if (btrfs_test_is_dummy_root(root))
3047 ref_root = btrfs_header_owner(buf);
3048 nritems = btrfs_header_nritems(buf);
3049 level = btrfs_header_level(buf);
3051 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3055 process_func = btrfs_inc_extent_ref;
3057 process_func = btrfs_free_extent;
3060 parent = buf->start;
3064 for (i = 0; i < nritems; i++) {
3066 btrfs_item_key_to_cpu(buf, &key, i);
3067 if (key.type != BTRFS_EXTENT_DATA_KEY)
3069 fi = btrfs_item_ptr(buf, i,
3070 struct btrfs_file_extent_item);
3071 if (btrfs_file_extent_type(buf, fi) ==
3072 BTRFS_FILE_EXTENT_INLINE)
3074 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3078 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3079 key.offset -= btrfs_file_extent_offset(buf, fi);
3080 ret = process_func(trans, root, bytenr, num_bytes,
3081 parent, ref_root, key.objectid,
3086 bytenr = btrfs_node_blockptr(buf, i);
3087 num_bytes = root->nodesize;
3088 ret = process_func(trans, root, bytenr, num_bytes,
3089 parent, ref_root, level - 1, 0,
3100 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3101 struct extent_buffer *buf, int full_backref)
3103 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3106 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3107 struct extent_buffer *buf, int full_backref)
3109 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3112 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3113 struct btrfs_root *root,
3114 struct btrfs_path *path,
3115 struct btrfs_block_group_cache *cache)
3118 struct btrfs_root *extent_root = root->fs_info->extent_root;
3120 struct extent_buffer *leaf;
3122 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3129 leaf = path->nodes[0];
3130 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3131 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3132 btrfs_mark_buffer_dirty(leaf);
3134 btrfs_release_path(path);
3139 static struct btrfs_block_group_cache *
3140 next_block_group(struct btrfs_root *root,
3141 struct btrfs_block_group_cache *cache)
3143 struct rb_node *node;
3145 spin_lock(&root->fs_info->block_group_cache_lock);
3147 /* If our block group was removed, we need a full search. */
3148 if (RB_EMPTY_NODE(&cache->cache_node)) {
3149 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3151 spin_unlock(&root->fs_info->block_group_cache_lock);
3152 btrfs_put_block_group(cache);
3153 cache = btrfs_lookup_first_block_group(root->fs_info,
3157 node = rb_next(&cache->cache_node);
3158 btrfs_put_block_group(cache);
3160 cache = rb_entry(node, struct btrfs_block_group_cache,
3162 btrfs_get_block_group(cache);
3165 spin_unlock(&root->fs_info->block_group_cache_lock);
3169 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3170 struct btrfs_trans_handle *trans,
3171 struct btrfs_path *path)
3173 struct btrfs_root *root = block_group->fs_info->tree_root;
3174 struct inode *inode = NULL;
3176 int dcs = BTRFS_DC_ERROR;
3182 * If this block group is smaller than 100 megs don't bother caching the
3185 if (block_group->key.offset < (100 * 1024 * 1024)) {
3186 spin_lock(&block_group->lock);
3187 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3188 spin_unlock(&block_group->lock);
3195 inode = lookup_free_space_inode(root, block_group, path);
3196 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3197 ret = PTR_ERR(inode);
3198 btrfs_release_path(path);
3202 if (IS_ERR(inode)) {
3206 if (block_group->ro)
3209 ret = create_free_space_inode(root, trans, block_group, path);
3215 /* We've already setup this transaction, go ahead and exit */
3216 if (block_group->cache_generation == trans->transid &&
3217 i_size_read(inode)) {
3218 dcs = BTRFS_DC_SETUP;
3223 * We want to set the generation to 0, that way if anything goes wrong
3224 * from here on out we know not to trust this cache when we load up next
3227 BTRFS_I(inode)->generation = 0;
3228 ret = btrfs_update_inode(trans, root, inode);
3231 * So theoretically we could recover from this, simply set the
3232 * super cache generation to 0 so we know to invalidate the
3233 * cache, but then we'd have to keep track of the block groups
3234 * that fail this way so we know we _have_ to reset this cache
3235 * before the next commit or risk reading stale cache. So to
3236 * limit our exposure to horrible edge cases lets just abort the
3237 * transaction, this only happens in really bad situations
3240 btrfs_abort_transaction(trans, root, ret);
3245 if (i_size_read(inode) > 0) {
3246 ret = btrfs_check_trunc_cache_free_space(root,
3247 &root->fs_info->global_block_rsv);
3251 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3256 spin_lock(&block_group->lock);
3257 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3258 !btrfs_test_opt(root, SPACE_CACHE)) {
3260 * don't bother trying to write stuff out _if_
3261 * a) we're not cached,
3262 * b) we're with nospace_cache mount option.
3264 dcs = BTRFS_DC_WRITTEN;
3265 spin_unlock(&block_group->lock);
3268 spin_unlock(&block_group->lock);
3271 * Try to preallocate enough space based on how big the block group is.
3272 * Keep in mind this has to include any pinned space which could end up
3273 * taking up quite a bit since it's not folded into the other space
3276 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3281 num_pages *= PAGE_CACHE_SIZE;
3283 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3287 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3288 num_pages, num_pages,
3291 dcs = BTRFS_DC_SETUP;
3292 btrfs_free_reserved_data_space(inode, num_pages);
3297 btrfs_release_path(path);
3299 spin_lock(&block_group->lock);
3300 if (!ret && dcs == BTRFS_DC_SETUP)
3301 block_group->cache_generation = trans->transid;
3302 block_group->disk_cache_state = dcs;
3303 spin_unlock(&block_group->lock);
3308 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3309 struct btrfs_root *root)
3311 struct btrfs_block_group_cache *cache, *tmp;
3312 struct btrfs_transaction *cur_trans = trans->transaction;
3313 struct btrfs_path *path;
3315 if (list_empty(&cur_trans->dirty_bgs) ||
3316 !btrfs_test_opt(root, SPACE_CACHE))
3319 path = btrfs_alloc_path();
3323 /* Could add new block groups, use _safe just in case */
3324 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3326 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3327 cache_save_setup(cache, trans, path);
3330 btrfs_free_path(path);
3335 * transaction commit does final block group cache writeback during a
3336 * critical section where nothing is allowed to change the FS. This is
3337 * required in order for the cache to actually match the block group,
3338 * but can introduce a lot of latency into the commit.
3340 * So, btrfs_start_dirty_block_groups is here to kick off block group
3341 * cache IO. There's a chance we'll have to redo some of it if the
3342 * block group changes again during the commit, but it greatly reduces
3343 * the commit latency by getting rid of the easy block groups while
3344 * we're still allowing others to join the commit.
3346 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3347 struct btrfs_root *root)
3349 struct btrfs_block_group_cache *cache;
3350 struct btrfs_transaction *cur_trans = trans->transaction;
3353 struct btrfs_path *path = NULL;
3355 struct list_head *io = &cur_trans->io_bgs;
3356 int num_started = 0;
3359 spin_lock(&cur_trans->dirty_bgs_lock);
3360 if (list_empty(&cur_trans->dirty_bgs)) {
3361 spin_unlock(&cur_trans->dirty_bgs_lock);
3364 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3365 spin_unlock(&cur_trans->dirty_bgs_lock);
3369 * make sure all the block groups on our dirty list actually
3372 btrfs_create_pending_block_groups(trans, root);
3375 path = btrfs_alloc_path();
3381 * cache_write_mutex is here only to save us from balance or automatic
3382 * removal of empty block groups deleting this block group while we are
3383 * writing out the cache
3385 mutex_lock(&trans->transaction->cache_write_mutex);
3386 while (!list_empty(&dirty)) {
3387 cache = list_first_entry(&dirty,
3388 struct btrfs_block_group_cache,
3391 * this can happen if something re-dirties a block
3392 * group that is already under IO. Just wait for it to
3393 * finish and then do it all again
3395 if (!list_empty(&cache->io_list)) {
3396 list_del_init(&cache->io_list);
3397 btrfs_wait_cache_io(root, trans, cache,
3398 &cache->io_ctl, path,
3399 cache->key.objectid);
3400 btrfs_put_block_group(cache);
3405 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3406 * if it should update the cache_state. Don't delete
3407 * until after we wait.
3409 * Since we're not running in the commit critical section
3410 * we need the dirty_bgs_lock to protect from update_block_group
3412 spin_lock(&cur_trans->dirty_bgs_lock);
3413 list_del_init(&cache->dirty_list);
3414 spin_unlock(&cur_trans->dirty_bgs_lock);
3418 cache_save_setup(cache, trans, path);
3420 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3421 cache->io_ctl.inode = NULL;
3422 ret = btrfs_write_out_cache(root, trans, cache, path);
3423 if (ret == 0 && cache->io_ctl.inode) {
3428 * the cache_write_mutex is protecting
3431 list_add_tail(&cache->io_list, io);
3434 * if we failed to write the cache, the
3435 * generation will be bad and life goes on
3441 ret = write_one_cache_group(trans, root, path, cache);
3443 * Our block group might still be attached to the list
3444 * of new block groups in the transaction handle of some
3445 * other task (struct btrfs_trans_handle->new_bgs). This
3446 * means its block group item isn't yet in the extent
3447 * tree. If this happens ignore the error, as we will
3448 * try again later in the critical section of the
3449 * transaction commit.
3451 if (ret == -ENOENT) {
3453 spin_lock(&cur_trans->dirty_bgs_lock);
3454 if (list_empty(&cache->dirty_list)) {
3455 list_add_tail(&cache->dirty_list,
3456 &cur_trans->dirty_bgs);
3457 btrfs_get_block_group(cache);
3459 spin_unlock(&cur_trans->dirty_bgs_lock);
3461 btrfs_abort_transaction(trans, root, ret);
3465 /* if its not on the io list, we need to put the block group */
3467 btrfs_put_block_group(cache);
3473 * Avoid blocking other tasks for too long. It might even save
3474 * us from writing caches for block groups that are going to be
3477 mutex_unlock(&trans->transaction->cache_write_mutex);
3478 mutex_lock(&trans->transaction->cache_write_mutex);
3480 mutex_unlock(&trans->transaction->cache_write_mutex);
3483 * go through delayed refs for all the stuff we've just kicked off
3484 * and then loop back (just once)
3486 ret = btrfs_run_delayed_refs(trans, root, 0);
3487 if (!ret && loops == 0) {
3489 spin_lock(&cur_trans->dirty_bgs_lock);
3490 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3492 * dirty_bgs_lock protects us from concurrent block group
3493 * deletes too (not just cache_write_mutex).
3495 if (!list_empty(&dirty)) {
3496 spin_unlock(&cur_trans->dirty_bgs_lock);
3499 spin_unlock(&cur_trans->dirty_bgs_lock);
3502 btrfs_free_path(path);
3506 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3507 struct btrfs_root *root)
3509 struct btrfs_block_group_cache *cache;
3510 struct btrfs_transaction *cur_trans = trans->transaction;
3513 struct btrfs_path *path;
3514 struct list_head *io = &cur_trans->io_bgs;
3515 int num_started = 0;
3517 path = btrfs_alloc_path();
3522 * We don't need the lock here since we are protected by the transaction
3523 * commit. We want to do the cache_save_setup first and then run the
3524 * delayed refs to make sure we have the best chance at doing this all
3527 while (!list_empty(&cur_trans->dirty_bgs)) {
3528 cache = list_first_entry(&cur_trans->dirty_bgs,
3529 struct btrfs_block_group_cache,
3533 * this can happen if cache_save_setup re-dirties a block
3534 * group that is already under IO. Just wait for it to
3535 * finish and then do it all again
3537 if (!list_empty(&cache->io_list)) {
3538 list_del_init(&cache->io_list);
3539 btrfs_wait_cache_io(root, trans, cache,
3540 &cache->io_ctl, path,
3541 cache->key.objectid);
3542 btrfs_put_block_group(cache);
3546 * don't remove from the dirty list until after we've waited
3549 list_del_init(&cache->dirty_list);
3552 cache_save_setup(cache, trans, path);
3555 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3557 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3558 cache->io_ctl.inode = NULL;
3559 ret = btrfs_write_out_cache(root, trans, cache, path);
3560 if (ret == 0 && cache->io_ctl.inode) {
3563 list_add_tail(&cache->io_list, io);
3566 * if we failed to write the cache, the
3567 * generation will be bad and life goes on
3573 ret = write_one_cache_group(trans, root, path, cache);
3575 btrfs_abort_transaction(trans, root, ret);
3578 /* if its not on the io list, we need to put the block group */
3580 btrfs_put_block_group(cache);
3583 while (!list_empty(io)) {
3584 cache = list_first_entry(io, struct btrfs_block_group_cache,
3586 list_del_init(&cache->io_list);
3587 btrfs_wait_cache_io(root, trans, cache,
3588 &cache->io_ctl, path, cache->key.objectid);
3589 btrfs_put_block_group(cache);
3592 btrfs_free_path(path);
3596 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3598 struct btrfs_block_group_cache *block_group;
3601 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3602 if (!block_group || block_group->ro)
3605 btrfs_put_block_group(block_group);
3609 static const char *alloc_name(u64 flags)
3612 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3614 case BTRFS_BLOCK_GROUP_METADATA:
3616 case BTRFS_BLOCK_GROUP_DATA:
3618 case BTRFS_BLOCK_GROUP_SYSTEM:
3622 return "invalid-combination";
3626 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3627 u64 total_bytes, u64 bytes_used,
3628 struct btrfs_space_info **space_info)
3630 struct btrfs_space_info *found;
3635 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3636 BTRFS_BLOCK_GROUP_RAID10))
3641 found = __find_space_info(info, flags);
3643 spin_lock(&found->lock);
3644 found->total_bytes += total_bytes;
3645 found->disk_total += total_bytes * factor;
3646 found->bytes_used += bytes_used;
3647 found->disk_used += bytes_used * factor;
3648 if (total_bytes > 0)
3650 spin_unlock(&found->lock);
3651 *space_info = found;
3654 found = kzalloc(sizeof(*found), GFP_NOFS);
3658 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3664 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3665 INIT_LIST_HEAD(&found->block_groups[i]);
3666 init_rwsem(&found->groups_sem);
3667 spin_lock_init(&found->lock);
3668 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3669 found->total_bytes = total_bytes;
3670 found->disk_total = total_bytes * factor;
3671 found->bytes_used = bytes_used;
3672 found->disk_used = bytes_used * factor;
3673 found->bytes_pinned = 0;
3674 found->bytes_reserved = 0;
3675 found->bytes_readonly = 0;
3676 found->bytes_may_use = 0;
3677 if (total_bytes > 0)
3681 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3682 found->chunk_alloc = 0;
3684 init_waitqueue_head(&found->wait);
3685 INIT_LIST_HEAD(&found->ro_bgs);
3687 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3688 info->space_info_kobj, "%s",
3689 alloc_name(found->flags));
3695 *space_info = found;
3696 list_add_rcu(&found->list, &info->space_info);
3697 if (flags & BTRFS_BLOCK_GROUP_DATA)
3698 info->data_sinfo = found;
3703 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3705 u64 extra_flags = chunk_to_extended(flags) &
3706 BTRFS_EXTENDED_PROFILE_MASK;
3708 write_seqlock(&fs_info->profiles_lock);
3709 if (flags & BTRFS_BLOCK_GROUP_DATA)
3710 fs_info->avail_data_alloc_bits |= extra_flags;
3711 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3712 fs_info->avail_metadata_alloc_bits |= extra_flags;
3713 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3714 fs_info->avail_system_alloc_bits |= extra_flags;
3715 write_sequnlock(&fs_info->profiles_lock);
3719 * returns target flags in extended format or 0 if restripe for this
3720 * chunk_type is not in progress
3722 * should be called with either volume_mutex or balance_lock held
3724 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3726 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3732 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3733 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3734 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3735 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3736 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3737 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3738 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3739 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3740 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3747 * @flags: available profiles in extended format (see ctree.h)
3749 * Returns reduced profile in chunk format. If profile changing is in
3750 * progress (either running or paused) picks the target profile (if it's
3751 * already available), otherwise falls back to plain reducing.
3753 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3755 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3760 * see if restripe for this chunk_type is in progress, if so
3761 * try to reduce to the target profile
3763 spin_lock(&root->fs_info->balance_lock);
3764 target = get_restripe_target(root->fs_info, flags);
3766 /* pick target profile only if it's already available */
3767 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3768 spin_unlock(&root->fs_info->balance_lock);
3769 return extended_to_chunk(target);
3772 spin_unlock(&root->fs_info->balance_lock);
3774 /* First, mask out the RAID levels which aren't possible */
3775 if (num_devices == 1)
3776 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3777 BTRFS_BLOCK_GROUP_RAID5);
3778 if (num_devices < 3)
3779 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3780 if (num_devices < 4)
3781 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3783 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3784 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3785 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3788 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3789 tmp = BTRFS_BLOCK_GROUP_RAID6;
3790 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3791 tmp = BTRFS_BLOCK_GROUP_RAID5;
3792 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3793 tmp = BTRFS_BLOCK_GROUP_RAID10;
3794 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3795 tmp = BTRFS_BLOCK_GROUP_RAID1;
3796 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3797 tmp = BTRFS_BLOCK_GROUP_RAID0;
3799 return extended_to_chunk(flags | tmp);
3802 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3809 seq = read_seqbegin(&root->fs_info->profiles_lock);
3811 if (flags & BTRFS_BLOCK_GROUP_DATA)
3812 flags |= root->fs_info->avail_data_alloc_bits;
3813 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3814 flags |= root->fs_info->avail_system_alloc_bits;
3815 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3816 flags |= root->fs_info->avail_metadata_alloc_bits;
3817 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3819 return btrfs_reduce_alloc_profile(root, flags);
3822 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3828 flags = BTRFS_BLOCK_GROUP_DATA;
3829 else if (root == root->fs_info->chunk_root)
3830 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3832 flags = BTRFS_BLOCK_GROUP_METADATA;
3834 ret = get_alloc_profile(root, flags);
3839 * This will check the space that the inode allocates from to make sure we have
3840 * enough space for bytes.
3842 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3844 struct btrfs_space_info *data_sinfo;
3845 struct btrfs_root *root = BTRFS_I(inode)->root;
3846 struct btrfs_fs_info *fs_info = root->fs_info;
3849 int need_commit = 2;
3850 int have_pinned_space;
3852 /* make sure bytes are sectorsize aligned */
3853 bytes = ALIGN(bytes, root->sectorsize);
3855 if (btrfs_is_free_space_inode(inode)) {
3857 ASSERT(current->journal_info);
3860 data_sinfo = fs_info->data_sinfo;
3865 /* make sure we have enough space to handle the data first */
3866 spin_lock(&data_sinfo->lock);
3867 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3868 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3869 data_sinfo->bytes_may_use;
3871 if (used + bytes > data_sinfo->total_bytes) {
3872 struct btrfs_trans_handle *trans;
3875 * if we don't have enough free bytes in this space then we need
3876 * to alloc a new chunk.
3878 if (!data_sinfo->full) {
3881 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3882 spin_unlock(&data_sinfo->lock);
3884 alloc_target = btrfs_get_alloc_profile(root, 1);
3886 * It is ugly that we don't call nolock join
3887 * transaction for the free space inode case here.
3888 * But it is safe because we only do the data space
3889 * reservation for the free space cache in the
3890 * transaction context, the common join transaction
3891 * just increase the counter of the current transaction
3892 * handler, doesn't try to acquire the trans_lock of
3895 trans = btrfs_join_transaction(root);
3897 return PTR_ERR(trans);
3899 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3901 CHUNK_ALLOC_NO_FORCE);
3902 btrfs_end_transaction(trans, root);
3907 have_pinned_space = 1;
3913 data_sinfo = fs_info->data_sinfo;
3919 * If we don't have enough pinned space to deal with this
3920 * allocation, and no removed chunk in current transaction,
3921 * don't bother committing the transaction.
3923 have_pinned_space = percpu_counter_compare(
3924 &data_sinfo->total_bytes_pinned,
3925 used + bytes - data_sinfo->total_bytes);
3926 spin_unlock(&data_sinfo->lock);
3928 /* commit the current transaction and try again */
3931 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3934 if (need_commit > 0)
3935 btrfs_wait_ordered_roots(fs_info, -1);
3937 trans = btrfs_join_transaction(root);
3939 return PTR_ERR(trans);
3940 if (have_pinned_space >= 0 ||
3941 trans->transaction->have_free_bgs ||
3943 ret = btrfs_commit_transaction(trans, root);
3947 * make sure that all running delayed iput are
3950 down_write(&root->fs_info->delayed_iput_sem);
3951 up_write(&root->fs_info->delayed_iput_sem);
3954 btrfs_end_transaction(trans, root);
3958 trace_btrfs_space_reservation(root->fs_info,
3959 "space_info:enospc",
3960 data_sinfo->flags, bytes, 1);
3963 ret = btrfs_qgroup_reserve(root, write_bytes);
3966 data_sinfo->bytes_may_use += bytes;
3967 trace_btrfs_space_reservation(root->fs_info, "space_info",
3968 data_sinfo->flags, bytes, 1);
3970 spin_unlock(&data_sinfo->lock);
3976 * Called if we need to clear a data reservation for this inode.
3978 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3980 struct btrfs_root *root = BTRFS_I(inode)->root;
3981 struct btrfs_space_info *data_sinfo;
3983 /* make sure bytes are sectorsize aligned */
3984 bytes = ALIGN(bytes, root->sectorsize);
3986 data_sinfo = root->fs_info->data_sinfo;
3987 spin_lock(&data_sinfo->lock);
3988 WARN_ON(data_sinfo->bytes_may_use < bytes);
3989 data_sinfo->bytes_may_use -= bytes;
3990 trace_btrfs_space_reservation(root->fs_info, "space_info",
3991 data_sinfo->flags, bytes, 0);
3992 spin_unlock(&data_sinfo->lock);
3995 static void force_metadata_allocation(struct btrfs_fs_info *info)
3997 struct list_head *head = &info->space_info;
3998 struct btrfs_space_info *found;
4001 list_for_each_entry_rcu(found, head, list) {
4002 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4003 found->force_alloc = CHUNK_ALLOC_FORCE;
4008 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4010 return (global->size << 1);
4013 static int should_alloc_chunk(struct btrfs_root *root,
4014 struct btrfs_space_info *sinfo, int force)
4016 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4017 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4018 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4021 if (force == CHUNK_ALLOC_FORCE)
4025 * We need to take into account the global rsv because for all intents
4026 * and purposes it's used space. Don't worry about locking the
4027 * global_rsv, it doesn't change except when the transaction commits.
4029 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4030 num_allocated += calc_global_rsv_need_space(global_rsv);
4033 * in limited mode, we want to have some free space up to
4034 * about 1% of the FS size.
4036 if (force == CHUNK_ALLOC_LIMITED) {
4037 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4038 thresh = max_t(u64, 64 * 1024 * 1024,
4039 div_factor_fine(thresh, 1));
4041 if (num_bytes - num_allocated < thresh)
4045 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4050 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4054 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4055 BTRFS_BLOCK_GROUP_RAID0 |
4056 BTRFS_BLOCK_GROUP_RAID5 |
4057 BTRFS_BLOCK_GROUP_RAID6))
4058 num_dev = root->fs_info->fs_devices->rw_devices;
4059 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4062 num_dev = 1; /* DUP or single */
4068 * If @is_allocation is true, reserve space in the system space info necessary
4069 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4072 void check_system_chunk(struct btrfs_trans_handle *trans,
4073 struct btrfs_root *root,
4076 struct btrfs_space_info *info;
4083 * Needed because we can end up allocating a system chunk and for an
4084 * atomic and race free space reservation in the chunk block reserve.
4086 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4088 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4089 spin_lock(&info->lock);
4090 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4091 info->bytes_reserved - info->bytes_readonly -
4092 info->bytes_may_use;
4093 spin_unlock(&info->lock);
4095 num_devs = get_profile_num_devs(root, type);
4097 /* num_devs device items to update and 1 chunk item to add or remove */
4098 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4099 btrfs_calc_trans_metadata_size(root, 1);
4101 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4102 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4103 left, thresh, type);
4104 dump_space_info(info, 0, 0);
4107 if (left < thresh) {
4110 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4112 * Ignore failure to create system chunk. We might end up not
4113 * needing it, as we might not need to COW all nodes/leafs from
4114 * the paths we visit in the chunk tree (they were already COWed
4115 * or created in the current transaction for example).
4117 ret = btrfs_alloc_chunk(trans, root, flags);
4121 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4122 &root->fs_info->chunk_block_rsv,
4123 thresh, BTRFS_RESERVE_NO_FLUSH);
4125 trans->chunk_bytes_reserved += thresh;
4129 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4130 struct btrfs_root *extent_root, u64 flags, int force)
4132 struct btrfs_space_info *space_info;
4133 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4134 int wait_for_alloc = 0;
4137 /* Don't re-enter if we're already allocating a chunk */
4138 if (trans->allocating_chunk)
4141 space_info = __find_space_info(extent_root->fs_info, flags);
4143 ret = update_space_info(extent_root->fs_info, flags,
4145 BUG_ON(ret); /* -ENOMEM */
4147 BUG_ON(!space_info); /* Logic error */
4150 spin_lock(&space_info->lock);
4151 if (force < space_info->force_alloc)
4152 force = space_info->force_alloc;
4153 if (space_info->full) {
4154 if (should_alloc_chunk(extent_root, space_info, force))
4158 spin_unlock(&space_info->lock);
4162 if (!should_alloc_chunk(extent_root, space_info, force)) {
4163 spin_unlock(&space_info->lock);
4165 } else if (space_info->chunk_alloc) {
4168 space_info->chunk_alloc = 1;
4171 spin_unlock(&space_info->lock);
4173 mutex_lock(&fs_info->chunk_mutex);
4176 * The chunk_mutex is held throughout the entirety of a chunk
4177 * allocation, so once we've acquired the chunk_mutex we know that the
4178 * other guy is done and we need to recheck and see if we should
4181 if (wait_for_alloc) {
4182 mutex_unlock(&fs_info->chunk_mutex);
4187 trans->allocating_chunk = true;
4190 * If we have mixed data/metadata chunks we want to make sure we keep
4191 * allocating mixed chunks instead of individual chunks.
4193 if (btrfs_mixed_space_info(space_info))
4194 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4197 * if we're doing a data chunk, go ahead and make sure that
4198 * we keep a reasonable number of metadata chunks allocated in the
4201 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4202 fs_info->data_chunk_allocations++;
4203 if (!(fs_info->data_chunk_allocations %
4204 fs_info->metadata_ratio))
4205 force_metadata_allocation(fs_info);
4209 * Check if we have enough space in SYSTEM chunk because we may need
4210 * to update devices.
4212 check_system_chunk(trans, extent_root, flags);
4214 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4215 trans->allocating_chunk = false;
4217 spin_lock(&space_info->lock);
4218 if (ret < 0 && ret != -ENOSPC)
4221 space_info->full = 1;
4225 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4227 space_info->chunk_alloc = 0;
4228 spin_unlock(&space_info->lock);
4229 mutex_unlock(&fs_info->chunk_mutex);
4233 static int can_overcommit(struct btrfs_root *root,
4234 struct btrfs_space_info *space_info, u64 bytes,
4235 enum btrfs_reserve_flush_enum flush)
4237 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4238 u64 profile = btrfs_get_alloc_profile(root, 0);
4243 used = space_info->bytes_used + space_info->bytes_reserved +
4244 space_info->bytes_pinned + space_info->bytes_readonly;
4247 * We only want to allow over committing if we have lots of actual space
4248 * free, but if we don't have enough space to handle the global reserve
4249 * space then we could end up having a real enospc problem when trying
4250 * to allocate a chunk or some other such important allocation.
4252 spin_lock(&global_rsv->lock);
4253 space_size = calc_global_rsv_need_space(global_rsv);
4254 spin_unlock(&global_rsv->lock);
4255 if (used + space_size >= space_info->total_bytes)
4258 used += space_info->bytes_may_use;
4260 spin_lock(&root->fs_info->free_chunk_lock);
4261 avail = root->fs_info->free_chunk_space;
4262 spin_unlock(&root->fs_info->free_chunk_lock);
4265 * If we have dup, raid1 or raid10 then only half of the free
4266 * space is actually useable. For raid56, the space info used
4267 * doesn't include the parity drive, so we don't have to
4270 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4271 BTRFS_BLOCK_GROUP_RAID1 |
4272 BTRFS_BLOCK_GROUP_RAID10))
4276 * If we aren't flushing all things, let us overcommit up to
4277 * 1/2th of the space. If we can flush, don't let us overcommit
4278 * too much, let it overcommit up to 1/8 of the space.
4280 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4285 if (used + bytes < space_info->total_bytes + avail)
4290 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4291 unsigned long nr_pages, int nr_items)
4293 struct super_block *sb = root->fs_info->sb;
4295 if (down_read_trylock(&sb->s_umount)) {
4296 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4297 up_read(&sb->s_umount);
4300 * We needn't worry the filesystem going from r/w to r/o though
4301 * we don't acquire ->s_umount mutex, because the filesystem
4302 * should guarantee the delalloc inodes list be empty after
4303 * the filesystem is readonly(all dirty pages are written to
4306 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4307 if (!current->journal_info)
4308 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4312 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4317 bytes = btrfs_calc_trans_metadata_size(root, 1);
4318 nr = (int)div64_u64(to_reclaim, bytes);
4324 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4327 * shrink metadata reservation for delalloc
4329 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4332 struct btrfs_block_rsv *block_rsv;
4333 struct btrfs_space_info *space_info;
4334 struct btrfs_trans_handle *trans;
4338 unsigned long nr_pages;
4341 enum btrfs_reserve_flush_enum flush;
4343 /* Calc the number of the pages we need flush for space reservation */
4344 items = calc_reclaim_items_nr(root, to_reclaim);
4345 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4347 trans = (struct btrfs_trans_handle *)current->journal_info;
4348 block_rsv = &root->fs_info->delalloc_block_rsv;
4349 space_info = block_rsv->space_info;
4351 delalloc_bytes = percpu_counter_sum_positive(
4352 &root->fs_info->delalloc_bytes);
4353 if (delalloc_bytes == 0) {
4357 btrfs_wait_ordered_roots(root->fs_info, items);
4362 while (delalloc_bytes && loops < 3) {
4363 max_reclaim = min(delalloc_bytes, to_reclaim);
4364 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4365 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4367 * We need to wait for the async pages to actually start before
4370 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4374 if (max_reclaim <= nr_pages)
4377 max_reclaim -= nr_pages;
4379 wait_event(root->fs_info->async_submit_wait,
4380 atomic_read(&root->fs_info->async_delalloc_pages) <=
4384 flush = BTRFS_RESERVE_FLUSH_ALL;
4386 flush = BTRFS_RESERVE_NO_FLUSH;
4387 spin_lock(&space_info->lock);
4388 if (can_overcommit(root, space_info, orig, flush)) {
4389 spin_unlock(&space_info->lock);
4392 spin_unlock(&space_info->lock);
4395 if (wait_ordered && !trans) {
4396 btrfs_wait_ordered_roots(root->fs_info, items);
4398 time_left = schedule_timeout_killable(1);
4402 delalloc_bytes = percpu_counter_sum_positive(
4403 &root->fs_info->delalloc_bytes);
4408 * maybe_commit_transaction - possibly commit the transaction if its ok to
4409 * @root - the root we're allocating for
4410 * @bytes - the number of bytes we want to reserve
4411 * @force - force the commit
4413 * This will check to make sure that committing the transaction will actually
4414 * get us somewhere and then commit the transaction if it does. Otherwise it
4415 * will return -ENOSPC.
4417 static int may_commit_transaction(struct btrfs_root *root,
4418 struct btrfs_space_info *space_info,
4419 u64 bytes, int force)
4421 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4422 struct btrfs_trans_handle *trans;
4424 trans = (struct btrfs_trans_handle *)current->journal_info;
4431 /* See if there is enough pinned space to make this reservation */
4432 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4437 * See if there is some space in the delayed insertion reservation for
4440 if (space_info != delayed_rsv->space_info)
4443 spin_lock(&delayed_rsv->lock);
4444 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4445 bytes - delayed_rsv->size) >= 0) {
4446 spin_unlock(&delayed_rsv->lock);
4449 spin_unlock(&delayed_rsv->lock);
4452 trans = btrfs_join_transaction(root);
4456 return btrfs_commit_transaction(trans, root);
4460 FLUSH_DELAYED_ITEMS_NR = 1,
4461 FLUSH_DELAYED_ITEMS = 2,
4463 FLUSH_DELALLOC_WAIT = 4,
4468 static int flush_space(struct btrfs_root *root,
4469 struct btrfs_space_info *space_info, u64 num_bytes,
4470 u64 orig_bytes, int state)
4472 struct btrfs_trans_handle *trans;
4477 case FLUSH_DELAYED_ITEMS_NR:
4478 case FLUSH_DELAYED_ITEMS:
4479 if (state == FLUSH_DELAYED_ITEMS_NR)
4480 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4484 trans = btrfs_join_transaction(root);
4485 if (IS_ERR(trans)) {
4486 ret = PTR_ERR(trans);
4489 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4490 btrfs_end_transaction(trans, root);
4492 case FLUSH_DELALLOC:
4493 case FLUSH_DELALLOC_WAIT:
4494 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4495 state == FLUSH_DELALLOC_WAIT);
4498 trans = btrfs_join_transaction(root);
4499 if (IS_ERR(trans)) {
4500 ret = PTR_ERR(trans);
4503 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4504 btrfs_get_alloc_profile(root, 0),
4505 CHUNK_ALLOC_NO_FORCE);
4506 btrfs_end_transaction(trans, root);
4511 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4522 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4523 struct btrfs_space_info *space_info)
4529 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4531 spin_lock(&space_info->lock);
4532 if (can_overcommit(root, space_info, to_reclaim,
4533 BTRFS_RESERVE_FLUSH_ALL)) {
4538 used = space_info->bytes_used + space_info->bytes_reserved +
4539 space_info->bytes_pinned + space_info->bytes_readonly +
4540 space_info->bytes_may_use;
4541 if (can_overcommit(root, space_info, 1024 * 1024,
4542 BTRFS_RESERVE_FLUSH_ALL))
4543 expected = div_factor_fine(space_info->total_bytes, 95);
4545 expected = div_factor_fine(space_info->total_bytes, 90);
4547 if (used > expected)
4548 to_reclaim = used - expected;
4551 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4552 space_info->bytes_reserved);
4554 spin_unlock(&space_info->lock);
4559 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4560 struct btrfs_fs_info *fs_info, u64 used)
4562 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4564 /* If we're just plain full then async reclaim just slows us down. */
4565 if (space_info->bytes_used >= thresh)
4568 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4569 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4572 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4573 struct btrfs_fs_info *fs_info,
4578 spin_lock(&space_info->lock);
4580 * We run out of space and have not got any free space via flush_space,
4581 * so don't bother doing async reclaim.
4583 if (flush_state > COMMIT_TRANS && space_info->full) {
4584 spin_unlock(&space_info->lock);
4588 used = space_info->bytes_used + space_info->bytes_reserved +
4589 space_info->bytes_pinned + space_info->bytes_readonly +
4590 space_info->bytes_may_use;
4591 if (need_do_async_reclaim(space_info, fs_info, used)) {
4592 spin_unlock(&space_info->lock);
4595 spin_unlock(&space_info->lock);
4600 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4602 struct btrfs_fs_info *fs_info;
4603 struct btrfs_space_info *space_info;
4607 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4608 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4610 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4615 flush_state = FLUSH_DELAYED_ITEMS_NR;
4617 flush_space(fs_info->fs_root, space_info, to_reclaim,
4618 to_reclaim, flush_state);
4620 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4623 } while (flush_state < COMMIT_TRANS);
4626 void btrfs_init_async_reclaim_work(struct work_struct *work)
4628 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4632 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4633 * @root - the root we're allocating for
4634 * @block_rsv - the block_rsv we're allocating for
4635 * @orig_bytes - the number of bytes we want
4636 * @flush - whether or not we can flush to make our reservation
4638 * This will reserve orgi_bytes number of bytes from the space info associated
4639 * with the block_rsv. If there is not enough space it will make an attempt to
4640 * flush out space to make room. It will do this by flushing delalloc if
4641 * possible or committing the transaction. If flush is 0 then no attempts to
4642 * regain reservations will be made and this will fail if there is not enough
4645 static int reserve_metadata_bytes(struct btrfs_root *root,
4646 struct btrfs_block_rsv *block_rsv,
4648 enum btrfs_reserve_flush_enum flush)
4650 struct btrfs_space_info *space_info = block_rsv->space_info;
4652 u64 num_bytes = orig_bytes;
4653 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4655 bool flushing = false;
4659 spin_lock(&space_info->lock);
4661 * We only want to wait if somebody other than us is flushing and we
4662 * are actually allowed to flush all things.
4664 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4665 space_info->flush) {
4666 spin_unlock(&space_info->lock);
4668 * If we have a trans handle we can't wait because the flusher
4669 * may have to commit the transaction, which would mean we would
4670 * deadlock since we are waiting for the flusher to finish, but
4671 * hold the current transaction open.
4673 if (current->journal_info)
4675 ret = wait_event_killable(space_info->wait, !space_info->flush);
4676 /* Must have been killed, return */
4680 spin_lock(&space_info->lock);
4684 used = space_info->bytes_used + space_info->bytes_reserved +
4685 space_info->bytes_pinned + space_info->bytes_readonly +
4686 space_info->bytes_may_use;
4689 * The idea here is that we've not already over-reserved the block group
4690 * then we can go ahead and save our reservation first and then start
4691 * flushing if we need to. Otherwise if we've already overcommitted
4692 * lets start flushing stuff first and then come back and try to make
4695 if (used <= space_info->total_bytes) {
4696 if (used + orig_bytes <= space_info->total_bytes) {
4697 space_info->bytes_may_use += orig_bytes;
4698 trace_btrfs_space_reservation(root->fs_info,
4699 "space_info", space_info->flags, orig_bytes, 1);
4703 * Ok set num_bytes to orig_bytes since we aren't
4704 * overocmmitted, this way we only try and reclaim what
4707 num_bytes = orig_bytes;
4711 * Ok we're over committed, set num_bytes to the overcommitted
4712 * amount plus the amount of bytes that we need for this
4715 num_bytes = used - space_info->total_bytes +
4719 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4720 space_info->bytes_may_use += orig_bytes;
4721 trace_btrfs_space_reservation(root->fs_info, "space_info",
4722 space_info->flags, orig_bytes,
4728 * Couldn't make our reservation, save our place so while we're trying
4729 * to reclaim space we can actually use it instead of somebody else
4730 * stealing it from us.
4732 * We make the other tasks wait for the flush only when we can flush
4735 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4737 space_info->flush = 1;
4738 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4741 * We will do the space reservation dance during log replay,
4742 * which means we won't have fs_info->fs_root set, so don't do
4743 * the async reclaim as we will panic.
4745 if (!root->fs_info->log_root_recovering &&
4746 need_do_async_reclaim(space_info, root->fs_info, used) &&
4747 !work_busy(&root->fs_info->async_reclaim_work))
4748 queue_work(system_unbound_wq,
4749 &root->fs_info->async_reclaim_work);
4751 spin_unlock(&space_info->lock);
4753 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4756 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4761 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4762 * would happen. So skip delalloc flush.
4764 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4765 (flush_state == FLUSH_DELALLOC ||
4766 flush_state == FLUSH_DELALLOC_WAIT))
4767 flush_state = ALLOC_CHUNK;
4771 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4772 flush_state < COMMIT_TRANS)
4774 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4775 flush_state <= COMMIT_TRANS)
4779 if (ret == -ENOSPC &&
4780 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4781 struct btrfs_block_rsv *global_rsv =
4782 &root->fs_info->global_block_rsv;
4784 if (block_rsv != global_rsv &&
4785 !block_rsv_use_bytes(global_rsv, orig_bytes))
4789 trace_btrfs_space_reservation(root->fs_info,
4790 "space_info:enospc",
4791 space_info->flags, orig_bytes, 1);
4793 spin_lock(&space_info->lock);
4794 space_info->flush = 0;
4795 wake_up_all(&space_info->wait);
4796 spin_unlock(&space_info->lock);
4801 static struct btrfs_block_rsv *get_block_rsv(
4802 const struct btrfs_trans_handle *trans,
4803 const struct btrfs_root *root)
4805 struct btrfs_block_rsv *block_rsv = NULL;
4807 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4808 block_rsv = trans->block_rsv;
4810 if (root == root->fs_info->csum_root && trans->adding_csums)
4811 block_rsv = trans->block_rsv;
4813 if (root == root->fs_info->uuid_root)
4814 block_rsv = trans->block_rsv;
4817 block_rsv = root->block_rsv;
4820 block_rsv = &root->fs_info->empty_block_rsv;
4825 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4829 spin_lock(&block_rsv->lock);
4830 if (block_rsv->reserved >= num_bytes) {
4831 block_rsv->reserved -= num_bytes;
4832 if (block_rsv->reserved < block_rsv->size)
4833 block_rsv->full = 0;
4836 spin_unlock(&block_rsv->lock);
4840 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4841 u64 num_bytes, int update_size)
4843 spin_lock(&block_rsv->lock);
4844 block_rsv->reserved += num_bytes;
4846 block_rsv->size += num_bytes;
4847 else if (block_rsv->reserved >= block_rsv->size)
4848 block_rsv->full = 1;
4849 spin_unlock(&block_rsv->lock);
4852 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4853 struct btrfs_block_rsv *dest, u64 num_bytes,
4856 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4859 if (global_rsv->space_info != dest->space_info)
4862 spin_lock(&global_rsv->lock);
4863 min_bytes = div_factor(global_rsv->size, min_factor);
4864 if (global_rsv->reserved < min_bytes + num_bytes) {
4865 spin_unlock(&global_rsv->lock);
4868 global_rsv->reserved -= num_bytes;
4869 if (global_rsv->reserved < global_rsv->size)
4870 global_rsv->full = 0;
4871 spin_unlock(&global_rsv->lock);
4873 block_rsv_add_bytes(dest, num_bytes, 1);
4877 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4878 struct btrfs_block_rsv *block_rsv,
4879 struct btrfs_block_rsv *dest, u64 num_bytes)
4881 struct btrfs_space_info *space_info = block_rsv->space_info;
4883 spin_lock(&block_rsv->lock);
4884 if (num_bytes == (u64)-1)
4885 num_bytes = block_rsv->size;
4886 block_rsv->size -= num_bytes;
4887 if (block_rsv->reserved >= block_rsv->size) {
4888 num_bytes = block_rsv->reserved - block_rsv->size;
4889 block_rsv->reserved = block_rsv->size;
4890 block_rsv->full = 1;
4894 spin_unlock(&block_rsv->lock);
4896 if (num_bytes > 0) {
4898 spin_lock(&dest->lock);
4902 bytes_to_add = dest->size - dest->reserved;
4903 bytes_to_add = min(num_bytes, bytes_to_add);
4904 dest->reserved += bytes_to_add;
4905 if (dest->reserved >= dest->size)
4907 num_bytes -= bytes_to_add;
4909 spin_unlock(&dest->lock);
4912 spin_lock(&space_info->lock);
4913 space_info->bytes_may_use -= num_bytes;
4914 trace_btrfs_space_reservation(fs_info, "space_info",
4915 space_info->flags, num_bytes, 0);
4916 spin_unlock(&space_info->lock);
4921 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4922 struct btrfs_block_rsv *dst, u64 num_bytes)
4926 ret = block_rsv_use_bytes(src, num_bytes);
4930 block_rsv_add_bytes(dst, num_bytes, 1);
4934 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4936 memset(rsv, 0, sizeof(*rsv));
4937 spin_lock_init(&rsv->lock);
4941 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4942 unsigned short type)
4944 struct btrfs_block_rsv *block_rsv;
4945 struct btrfs_fs_info *fs_info = root->fs_info;
4947 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4951 btrfs_init_block_rsv(block_rsv, type);
4952 block_rsv->space_info = __find_space_info(fs_info,
4953 BTRFS_BLOCK_GROUP_METADATA);
4957 void btrfs_free_block_rsv(struct btrfs_root *root,
4958 struct btrfs_block_rsv *rsv)
4962 btrfs_block_rsv_release(root, rsv, (u64)-1);
4966 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4971 int btrfs_block_rsv_add(struct btrfs_root *root,
4972 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4973 enum btrfs_reserve_flush_enum flush)
4980 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4982 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4989 int btrfs_block_rsv_check(struct btrfs_root *root,
4990 struct btrfs_block_rsv *block_rsv, int min_factor)
4998 spin_lock(&block_rsv->lock);
4999 num_bytes = div_factor(block_rsv->size, min_factor);
5000 if (block_rsv->reserved >= num_bytes)
5002 spin_unlock(&block_rsv->lock);
5007 int btrfs_block_rsv_refill(struct btrfs_root *root,
5008 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5009 enum btrfs_reserve_flush_enum flush)
5017 spin_lock(&block_rsv->lock);
5018 num_bytes = min_reserved;
5019 if (block_rsv->reserved >= num_bytes)
5022 num_bytes -= block_rsv->reserved;
5023 spin_unlock(&block_rsv->lock);
5028 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5030 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5037 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5038 struct btrfs_block_rsv *dst_rsv,
5041 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5044 void btrfs_block_rsv_release(struct btrfs_root *root,
5045 struct btrfs_block_rsv *block_rsv,
5048 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5049 if (global_rsv == block_rsv ||
5050 block_rsv->space_info != global_rsv->space_info)
5052 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5057 * helper to calculate size of global block reservation.
5058 * the desired value is sum of space used by extent tree,
5059 * checksum tree and root tree
5061 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5063 struct btrfs_space_info *sinfo;
5067 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5069 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5070 spin_lock(&sinfo->lock);
5071 data_used = sinfo->bytes_used;
5072 spin_unlock(&sinfo->lock);
5074 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5075 spin_lock(&sinfo->lock);
5076 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5078 meta_used = sinfo->bytes_used;
5079 spin_unlock(&sinfo->lock);
5081 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5083 num_bytes += div_u64(data_used + meta_used, 50);
5085 if (num_bytes * 3 > meta_used)
5086 num_bytes = div_u64(meta_used, 3);
5088 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5091 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5093 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5094 struct btrfs_space_info *sinfo = block_rsv->space_info;
5097 num_bytes = calc_global_metadata_size(fs_info);
5099 spin_lock(&sinfo->lock);
5100 spin_lock(&block_rsv->lock);
5102 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5104 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5105 sinfo->bytes_reserved + sinfo->bytes_readonly +
5106 sinfo->bytes_may_use;
5108 if (sinfo->total_bytes > num_bytes) {
5109 num_bytes = sinfo->total_bytes - num_bytes;
5110 block_rsv->reserved += num_bytes;
5111 sinfo->bytes_may_use += num_bytes;
5112 trace_btrfs_space_reservation(fs_info, "space_info",
5113 sinfo->flags, num_bytes, 1);
5116 if (block_rsv->reserved >= block_rsv->size) {
5117 num_bytes = block_rsv->reserved - block_rsv->size;
5118 sinfo->bytes_may_use -= num_bytes;
5119 trace_btrfs_space_reservation(fs_info, "space_info",
5120 sinfo->flags, num_bytes, 0);
5121 block_rsv->reserved = block_rsv->size;
5122 block_rsv->full = 1;
5125 spin_unlock(&block_rsv->lock);
5126 spin_unlock(&sinfo->lock);
5129 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5131 struct btrfs_space_info *space_info;
5133 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5134 fs_info->chunk_block_rsv.space_info = space_info;
5136 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5137 fs_info->global_block_rsv.space_info = space_info;
5138 fs_info->delalloc_block_rsv.space_info = space_info;
5139 fs_info->trans_block_rsv.space_info = space_info;
5140 fs_info->empty_block_rsv.space_info = space_info;
5141 fs_info->delayed_block_rsv.space_info = space_info;
5143 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5144 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5145 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5146 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5147 if (fs_info->quota_root)
5148 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5149 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5151 update_global_block_rsv(fs_info);
5154 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5156 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5158 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5159 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5160 WARN_ON(fs_info->trans_block_rsv.size > 0);
5161 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5162 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5163 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5164 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5165 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5168 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5169 struct btrfs_root *root)
5171 if (!trans->block_rsv)
5174 if (!trans->bytes_reserved)
5177 trace_btrfs_space_reservation(root->fs_info, "transaction",
5178 trans->transid, trans->bytes_reserved, 0);
5179 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5180 trans->bytes_reserved = 0;
5184 * To be called after all the new block groups attached to the transaction
5185 * handle have been created (btrfs_create_pending_block_groups()).
5187 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5189 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5191 if (!trans->chunk_bytes_reserved)
5194 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5196 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5197 trans->chunk_bytes_reserved);
5198 trans->chunk_bytes_reserved = 0;
5201 /* Can only return 0 or -ENOSPC */
5202 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5203 struct inode *inode)
5205 struct btrfs_root *root = BTRFS_I(inode)->root;
5206 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5207 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5210 * We need to hold space in order to delete our orphan item once we've
5211 * added it, so this takes the reservation so we can release it later
5212 * when we are truly done with the orphan item.
5214 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5215 trace_btrfs_space_reservation(root->fs_info, "orphan",
5216 btrfs_ino(inode), num_bytes, 1);
5217 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5220 void btrfs_orphan_release_metadata(struct inode *inode)
5222 struct btrfs_root *root = BTRFS_I(inode)->root;
5223 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5224 trace_btrfs_space_reservation(root->fs_info, "orphan",
5225 btrfs_ino(inode), num_bytes, 0);
5226 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5230 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5231 * root: the root of the parent directory
5232 * rsv: block reservation
5233 * items: the number of items that we need do reservation
5234 * qgroup_reserved: used to return the reserved size in qgroup
5236 * This function is used to reserve the space for snapshot/subvolume
5237 * creation and deletion. Those operations are different with the
5238 * common file/directory operations, they change two fs/file trees
5239 * and root tree, the number of items that the qgroup reserves is
5240 * different with the free space reservation. So we can not use
5241 * the space reseravtion mechanism in start_transaction().
5243 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5244 struct btrfs_block_rsv *rsv,
5246 u64 *qgroup_reserved,
5247 bool use_global_rsv)
5251 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5253 if (root->fs_info->quota_enabled) {
5254 /* One for parent inode, two for dir entries */
5255 num_bytes = 3 * root->nodesize;
5256 ret = btrfs_qgroup_reserve(root, num_bytes);
5263 *qgroup_reserved = num_bytes;
5265 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5266 rsv->space_info = __find_space_info(root->fs_info,
5267 BTRFS_BLOCK_GROUP_METADATA);
5268 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5269 BTRFS_RESERVE_FLUSH_ALL);
5271 if (ret == -ENOSPC && use_global_rsv)
5272 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5275 if (*qgroup_reserved)
5276 btrfs_qgroup_free(root, *qgroup_reserved);
5282 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5283 struct btrfs_block_rsv *rsv,
5284 u64 qgroup_reserved)
5286 btrfs_block_rsv_release(root, rsv, (u64)-1);
5290 * drop_outstanding_extent - drop an outstanding extent
5291 * @inode: the inode we're dropping the extent for
5292 * @num_bytes: the number of bytes we're relaseing.
5294 * This is called when we are freeing up an outstanding extent, either called
5295 * after an error or after an extent is written. This will return the number of
5296 * reserved extents that need to be freed. This must be called with
5297 * BTRFS_I(inode)->lock held.
5299 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5301 unsigned drop_inode_space = 0;
5302 unsigned dropped_extents = 0;
5303 unsigned num_extents = 0;
5305 num_extents = (unsigned)div64_u64(num_bytes +
5306 BTRFS_MAX_EXTENT_SIZE - 1,
5307 BTRFS_MAX_EXTENT_SIZE);
5308 ASSERT(num_extents);
5309 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5310 BTRFS_I(inode)->outstanding_extents -= num_extents;
5312 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5313 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5314 &BTRFS_I(inode)->runtime_flags))
5315 drop_inode_space = 1;
5318 * If we have more or the same amount of outsanding extents than we have
5319 * reserved then we need to leave the reserved extents count alone.
5321 if (BTRFS_I(inode)->outstanding_extents >=
5322 BTRFS_I(inode)->reserved_extents)
5323 return drop_inode_space;
5325 dropped_extents = BTRFS_I(inode)->reserved_extents -
5326 BTRFS_I(inode)->outstanding_extents;
5327 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5328 return dropped_extents + drop_inode_space;
5332 * calc_csum_metadata_size - return the amount of metada space that must be
5333 * reserved/free'd for the given bytes.
5334 * @inode: the inode we're manipulating
5335 * @num_bytes: the number of bytes in question
5336 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5338 * This adjusts the number of csum_bytes in the inode and then returns the
5339 * correct amount of metadata that must either be reserved or freed. We
5340 * calculate how many checksums we can fit into one leaf and then divide the
5341 * number of bytes that will need to be checksumed by this value to figure out
5342 * how many checksums will be required. If we are adding bytes then the number
5343 * may go up and we will return the number of additional bytes that must be
5344 * reserved. If it is going down we will return the number of bytes that must
5347 * This must be called with BTRFS_I(inode)->lock held.
5349 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5352 struct btrfs_root *root = BTRFS_I(inode)->root;
5353 u64 old_csums, num_csums;
5355 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5356 BTRFS_I(inode)->csum_bytes == 0)
5359 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5361 BTRFS_I(inode)->csum_bytes += num_bytes;
5363 BTRFS_I(inode)->csum_bytes -= num_bytes;
5364 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5366 /* No change, no need to reserve more */
5367 if (old_csums == num_csums)
5371 return btrfs_calc_trans_metadata_size(root,
5372 num_csums - old_csums);
5374 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5377 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5379 struct btrfs_root *root = BTRFS_I(inode)->root;
5380 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5383 unsigned nr_extents = 0;
5384 int extra_reserve = 0;
5385 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5387 bool delalloc_lock = true;
5391 /* If we are a free space inode we need to not flush since we will be in
5392 * the middle of a transaction commit. We also don't need the delalloc
5393 * mutex since we won't race with anybody. We need this mostly to make
5394 * lockdep shut its filthy mouth.
5396 if (btrfs_is_free_space_inode(inode)) {
5397 flush = BTRFS_RESERVE_NO_FLUSH;
5398 delalloc_lock = false;
5401 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5402 btrfs_transaction_in_commit(root->fs_info))
5403 schedule_timeout(1);
5406 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5408 num_bytes = ALIGN(num_bytes, root->sectorsize);
5410 spin_lock(&BTRFS_I(inode)->lock);
5411 nr_extents = (unsigned)div64_u64(num_bytes +
5412 BTRFS_MAX_EXTENT_SIZE - 1,
5413 BTRFS_MAX_EXTENT_SIZE);
5414 BTRFS_I(inode)->outstanding_extents += nr_extents;
5417 if (BTRFS_I(inode)->outstanding_extents >
5418 BTRFS_I(inode)->reserved_extents)
5419 nr_extents = BTRFS_I(inode)->outstanding_extents -
5420 BTRFS_I(inode)->reserved_extents;
5423 * Add an item to reserve for updating the inode when we complete the
5426 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5427 &BTRFS_I(inode)->runtime_flags)) {
5432 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5433 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5434 csum_bytes = BTRFS_I(inode)->csum_bytes;
5435 spin_unlock(&BTRFS_I(inode)->lock);
5437 if (root->fs_info->quota_enabled) {
5438 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5443 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5444 if (unlikely(ret)) {
5445 if (root->fs_info->quota_enabled)
5446 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5450 spin_lock(&BTRFS_I(inode)->lock);
5451 if (extra_reserve) {
5452 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5453 &BTRFS_I(inode)->runtime_flags);
5456 BTRFS_I(inode)->reserved_extents += nr_extents;
5457 spin_unlock(&BTRFS_I(inode)->lock);
5460 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5463 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5464 btrfs_ino(inode), to_reserve, 1);
5465 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5470 spin_lock(&BTRFS_I(inode)->lock);
5471 dropped = drop_outstanding_extent(inode, num_bytes);
5473 * If the inodes csum_bytes is the same as the original
5474 * csum_bytes then we know we haven't raced with any free()ers
5475 * so we can just reduce our inodes csum bytes and carry on.
5477 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5478 calc_csum_metadata_size(inode, num_bytes, 0);
5480 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5484 * This is tricky, but first we need to figure out how much we
5485 * free'd from any free-ers that occured during this
5486 * reservation, so we reset ->csum_bytes to the csum_bytes
5487 * before we dropped our lock, and then call the free for the
5488 * number of bytes that were freed while we were trying our
5491 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5492 BTRFS_I(inode)->csum_bytes = csum_bytes;
5493 to_free = calc_csum_metadata_size(inode, bytes, 0);
5497 * Now we need to see how much we would have freed had we not
5498 * been making this reservation and our ->csum_bytes were not
5499 * artificially inflated.
5501 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5502 bytes = csum_bytes - orig_csum_bytes;
5503 bytes = calc_csum_metadata_size(inode, bytes, 0);
5506 * Now reset ->csum_bytes to what it should be. If bytes is
5507 * more than to_free then we would have free'd more space had we
5508 * not had an artificially high ->csum_bytes, so we need to free
5509 * the remainder. If bytes is the same or less then we don't
5510 * need to do anything, the other free-ers did the correct
5513 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5514 if (bytes > to_free)
5515 to_free = bytes - to_free;
5519 spin_unlock(&BTRFS_I(inode)->lock);
5521 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5524 btrfs_block_rsv_release(root, block_rsv, to_free);
5525 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5526 btrfs_ino(inode), to_free, 0);
5529 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5534 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5535 * @inode: the inode to release the reservation for
5536 * @num_bytes: the number of bytes we're releasing
5538 * This will release the metadata reservation for an inode. This can be called
5539 * once we complete IO for a given set of bytes to release their metadata
5542 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5544 struct btrfs_root *root = BTRFS_I(inode)->root;
5548 num_bytes = ALIGN(num_bytes, root->sectorsize);
5549 spin_lock(&BTRFS_I(inode)->lock);
5550 dropped = drop_outstanding_extent(inode, num_bytes);
5553 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5554 spin_unlock(&BTRFS_I(inode)->lock);
5556 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5558 if (btrfs_test_is_dummy_root(root))
5561 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5562 btrfs_ino(inode), to_free, 0);
5564 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5569 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5570 * @inode: inode we're writing to
5571 * @num_bytes: the number of bytes we want to allocate
5573 * This will do the following things
5575 * o reserve space in the data space info for num_bytes
5576 * o reserve space in the metadata space info based on number of outstanding
5577 * extents and how much csums will be needed
5578 * o add to the inodes ->delalloc_bytes
5579 * o add it to the fs_info's delalloc inodes list.
5581 * This will return 0 for success and -ENOSPC if there is no space left.
5583 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5587 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5591 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5593 btrfs_free_reserved_data_space(inode, num_bytes);
5601 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5602 * @inode: inode we're releasing space for
5603 * @num_bytes: the number of bytes we want to free up
5605 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5606 * called in the case that we don't need the metadata AND data reservations
5607 * anymore. So if there is an error or we insert an inline extent.
5609 * This function will release the metadata space that was not used and will
5610 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5611 * list if there are no delalloc bytes left.
5613 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5615 btrfs_delalloc_release_metadata(inode, num_bytes);
5616 btrfs_free_reserved_data_space(inode, num_bytes);
5619 static int update_block_group(struct btrfs_trans_handle *trans,
5620 struct btrfs_root *root, u64 bytenr,
5621 u64 num_bytes, int alloc)
5623 struct btrfs_block_group_cache *cache = NULL;
5624 struct btrfs_fs_info *info = root->fs_info;
5625 u64 total = num_bytes;
5630 /* block accounting for super block */
5631 spin_lock(&info->delalloc_root_lock);
5632 old_val = btrfs_super_bytes_used(info->super_copy);
5634 old_val += num_bytes;
5636 old_val -= num_bytes;
5637 btrfs_set_super_bytes_used(info->super_copy, old_val);
5638 spin_unlock(&info->delalloc_root_lock);
5641 cache = btrfs_lookup_block_group(info, bytenr);
5644 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5645 BTRFS_BLOCK_GROUP_RAID1 |
5646 BTRFS_BLOCK_GROUP_RAID10))
5651 * If this block group has free space cache written out, we
5652 * need to make sure to load it if we are removing space. This
5653 * is because we need the unpinning stage to actually add the
5654 * space back to the block group, otherwise we will leak space.
5656 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5657 cache_block_group(cache, 1);
5659 byte_in_group = bytenr - cache->key.objectid;
5660 WARN_ON(byte_in_group > cache->key.offset);
5662 spin_lock(&cache->space_info->lock);
5663 spin_lock(&cache->lock);
5665 if (btrfs_test_opt(root, SPACE_CACHE) &&
5666 cache->disk_cache_state < BTRFS_DC_CLEAR)
5667 cache->disk_cache_state = BTRFS_DC_CLEAR;
5669 old_val = btrfs_block_group_used(&cache->item);
5670 num_bytes = min(total, cache->key.offset - byte_in_group);
5672 old_val += num_bytes;
5673 btrfs_set_block_group_used(&cache->item, old_val);
5674 cache->reserved -= num_bytes;
5675 cache->space_info->bytes_reserved -= num_bytes;
5676 cache->space_info->bytes_used += num_bytes;
5677 cache->space_info->disk_used += num_bytes * factor;
5678 spin_unlock(&cache->lock);
5679 spin_unlock(&cache->space_info->lock);
5681 old_val -= num_bytes;
5682 btrfs_set_block_group_used(&cache->item, old_val);
5683 cache->pinned += num_bytes;
5684 cache->space_info->bytes_pinned += num_bytes;
5685 cache->space_info->bytes_used -= num_bytes;
5686 cache->space_info->disk_used -= num_bytes * factor;
5687 spin_unlock(&cache->lock);
5688 spin_unlock(&cache->space_info->lock);
5690 set_extent_dirty(info->pinned_extents,
5691 bytenr, bytenr + num_bytes - 1,
5692 GFP_NOFS | __GFP_NOFAIL);
5694 * No longer have used bytes in this block group, queue
5698 spin_lock(&info->unused_bgs_lock);
5699 if (list_empty(&cache->bg_list)) {
5700 btrfs_get_block_group(cache);
5701 list_add_tail(&cache->bg_list,
5704 spin_unlock(&info->unused_bgs_lock);
5708 spin_lock(&trans->transaction->dirty_bgs_lock);
5709 if (list_empty(&cache->dirty_list)) {
5710 list_add_tail(&cache->dirty_list,
5711 &trans->transaction->dirty_bgs);
5712 trans->transaction->num_dirty_bgs++;
5713 btrfs_get_block_group(cache);
5715 spin_unlock(&trans->transaction->dirty_bgs_lock);
5717 btrfs_put_block_group(cache);
5719 bytenr += num_bytes;
5724 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5726 struct btrfs_block_group_cache *cache;
5729 spin_lock(&root->fs_info->block_group_cache_lock);
5730 bytenr = root->fs_info->first_logical_byte;
5731 spin_unlock(&root->fs_info->block_group_cache_lock);
5733 if (bytenr < (u64)-1)
5736 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5740 bytenr = cache->key.objectid;
5741 btrfs_put_block_group(cache);
5746 static int pin_down_extent(struct btrfs_root *root,
5747 struct btrfs_block_group_cache *cache,
5748 u64 bytenr, u64 num_bytes, int reserved)
5750 spin_lock(&cache->space_info->lock);
5751 spin_lock(&cache->lock);
5752 cache->pinned += num_bytes;
5753 cache->space_info->bytes_pinned += num_bytes;
5755 cache->reserved -= num_bytes;
5756 cache->space_info->bytes_reserved -= num_bytes;
5758 spin_unlock(&cache->lock);
5759 spin_unlock(&cache->space_info->lock);
5761 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5762 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5764 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5769 * this function must be called within transaction
5771 int btrfs_pin_extent(struct btrfs_root *root,
5772 u64 bytenr, u64 num_bytes, int reserved)
5774 struct btrfs_block_group_cache *cache;
5776 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5777 BUG_ON(!cache); /* Logic error */
5779 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5781 btrfs_put_block_group(cache);
5786 * this function must be called within transaction
5788 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5789 u64 bytenr, u64 num_bytes)
5791 struct btrfs_block_group_cache *cache;
5794 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5799 * pull in the free space cache (if any) so that our pin
5800 * removes the free space from the cache. We have load_only set
5801 * to one because the slow code to read in the free extents does check
5802 * the pinned extents.
5804 cache_block_group(cache, 1);
5806 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5808 /* remove us from the free space cache (if we're there at all) */
5809 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5810 btrfs_put_block_group(cache);
5814 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5817 struct btrfs_block_group_cache *block_group;
5818 struct btrfs_caching_control *caching_ctl;
5820 block_group = btrfs_lookup_block_group(root->fs_info, start);
5824 cache_block_group(block_group, 0);
5825 caching_ctl = get_caching_control(block_group);
5829 BUG_ON(!block_group_cache_done(block_group));
5830 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5832 mutex_lock(&caching_ctl->mutex);
5834 if (start >= caching_ctl->progress) {
5835 ret = add_excluded_extent(root, start, num_bytes);
5836 } else if (start + num_bytes <= caching_ctl->progress) {
5837 ret = btrfs_remove_free_space(block_group,
5840 num_bytes = caching_ctl->progress - start;
5841 ret = btrfs_remove_free_space(block_group,
5846 num_bytes = (start + num_bytes) -
5847 caching_ctl->progress;
5848 start = caching_ctl->progress;
5849 ret = add_excluded_extent(root, start, num_bytes);
5852 mutex_unlock(&caching_ctl->mutex);
5853 put_caching_control(caching_ctl);
5855 btrfs_put_block_group(block_group);
5859 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5860 struct extent_buffer *eb)
5862 struct btrfs_file_extent_item *item;
5863 struct btrfs_key key;
5867 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5870 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5871 btrfs_item_key_to_cpu(eb, &key, i);
5872 if (key.type != BTRFS_EXTENT_DATA_KEY)
5874 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5875 found_type = btrfs_file_extent_type(eb, item);
5876 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5878 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5880 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5881 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5882 __exclude_logged_extent(log, key.objectid, key.offset);
5889 * btrfs_update_reserved_bytes - update the block_group and space info counters
5890 * @cache: The cache we are manipulating
5891 * @num_bytes: The number of bytes in question
5892 * @reserve: One of the reservation enums
5893 * @delalloc: The blocks are allocated for the delalloc write
5895 * This is called by the allocator when it reserves space, or by somebody who is
5896 * freeing space that was never actually used on disk. For example if you
5897 * reserve some space for a new leaf in transaction A and before transaction A
5898 * commits you free that leaf, you call this with reserve set to 0 in order to
5899 * clear the reservation.
5901 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5902 * ENOSPC accounting. For data we handle the reservation through clearing the
5903 * delalloc bits in the io_tree. We have to do this since we could end up
5904 * allocating less disk space for the amount of data we have reserved in the
5905 * case of compression.
5907 * If this is a reservation and the block group has become read only we cannot
5908 * make the reservation and return -EAGAIN, otherwise this function always
5911 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5912 u64 num_bytes, int reserve, int delalloc)
5914 struct btrfs_space_info *space_info = cache->space_info;
5917 spin_lock(&space_info->lock);
5918 spin_lock(&cache->lock);
5919 if (reserve != RESERVE_FREE) {
5923 cache->reserved += num_bytes;
5924 space_info->bytes_reserved += num_bytes;
5925 if (reserve == RESERVE_ALLOC) {
5926 trace_btrfs_space_reservation(cache->fs_info,
5927 "space_info", space_info->flags,
5929 space_info->bytes_may_use -= num_bytes;
5933 cache->delalloc_bytes += num_bytes;
5937 space_info->bytes_readonly += num_bytes;
5938 cache->reserved -= num_bytes;
5939 space_info->bytes_reserved -= num_bytes;
5942 cache->delalloc_bytes -= num_bytes;
5944 spin_unlock(&cache->lock);
5945 spin_unlock(&space_info->lock);
5949 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5950 struct btrfs_root *root)
5952 struct btrfs_fs_info *fs_info = root->fs_info;
5953 struct btrfs_caching_control *next;
5954 struct btrfs_caching_control *caching_ctl;
5955 struct btrfs_block_group_cache *cache;
5957 down_write(&fs_info->commit_root_sem);
5959 list_for_each_entry_safe(caching_ctl, next,
5960 &fs_info->caching_block_groups, list) {
5961 cache = caching_ctl->block_group;
5962 if (block_group_cache_done(cache)) {
5963 cache->last_byte_to_unpin = (u64)-1;
5964 list_del_init(&caching_ctl->list);
5965 put_caching_control(caching_ctl);
5967 cache->last_byte_to_unpin = caching_ctl->progress;
5971 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5972 fs_info->pinned_extents = &fs_info->freed_extents[1];
5974 fs_info->pinned_extents = &fs_info->freed_extents[0];
5976 up_write(&fs_info->commit_root_sem);
5978 update_global_block_rsv(fs_info);
5981 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5982 const bool return_free_space)
5984 struct btrfs_fs_info *fs_info = root->fs_info;
5985 struct btrfs_block_group_cache *cache = NULL;
5986 struct btrfs_space_info *space_info;
5987 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5991 while (start <= end) {
5994 start >= cache->key.objectid + cache->key.offset) {
5996 btrfs_put_block_group(cache);
5997 cache = btrfs_lookup_block_group(fs_info, start);
5998 BUG_ON(!cache); /* Logic error */
6001 len = cache->key.objectid + cache->key.offset - start;
6002 len = min(len, end + 1 - start);
6004 if (start < cache->last_byte_to_unpin) {
6005 len = min(len, cache->last_byte_to_unpin - start);
6006 if (return_free_space)
6007 btrfs_add_free_space(cache, start, len);
6011 space_info = cache->space_info;
6013 spin_lock(&space_info->lock);
6014 spin_lock(&cache->lock);
6015 cache->pinned -= len;
6016 space_info->bytes_pinned -= len;
6017 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6019 space_info->bytes_readonly += len;
6022 spin_unlock(&cache->lock);
6023 if (!readonly && global_rsv->space_info == space_info) {
6024 spin_lock(&global_rsv->lock);
6025 if (!global_rsv->full) {
6026 len = min(len, global_rsv->size -
6027 global_rsv->reserved);
6028 global_rsv->reserved += len;
6029 space_info->bytes_may_use += len;
6030 if (global_rsv->reserved >= global_rsv->size)
6031 global_rsv->full = 1;
6033 spin_unlock(&global_rsv->lock);
6035 spin_unlock(&space_info->lock);
6039 btrfs_put_block_group(cache);
6043 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6044 struct btrfs_root *root)
6046 struct btrfs_fs_info *fs_info = root->fs_info;
6047 struct extent_io_tree *unpin;
6055 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6056 unpin = &fs_info->freed_extents[1];
6058 unpin = &fs_info->freed_extents[0];
6061 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6062 ret = find_first_extent_bit(unpin, 0, &start, &end,
6063 EXTENT_DIRTY, NULL);
6065 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6069 if (btrfs_test_opt(root, DISCARD))
6070 ret = btrfs_discard_extent(root, start,
6071 end + 1 - start, NULL);
6073 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6074 unpin_extent_range(root, start, end, true);
6075 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6082 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6083 u64 owner, u64 root_objectid)
6085 struct btrfs_space_info *space_info;
6088 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6089 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6090 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6092 flags = BTRFS_BLOCK_GROUP_METADATA;
6094 flags = BTRFS_BLOCK_GROUP_DATA;
6097 space_info = __find_space_info(fs_info, flags);
6098 BUG_ON(!space_info); /* Logic bug */
6099 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6103 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6104 struct btrfs_root *root,
6105 struct btrfs_delayed_ref_node *node, u64 parent,
6106 u64 root_objectid, u64 owner_objectid,
6107 u64 owner_offset, int refs_to_drop,
6108 struct btrfs_delayed_extent_op *extent_op)
6110 struct btrfs_key key;
6111 struct btrfs_path *path;
6112 struct btrfs_fs_info *info = root->fs_info;
6113 struct btrfs_root *extent_root = info->extent_root;
6114 struct extent_buffer *leaf;
6115 struct btrfs_extent_item *ei;
6116 struct btrfs_extent_inline_ref *iref;
6119 int extent_slot = 0;
6120 int found_extent = 0;
6122 int no_quota = node->no_quota;
6125 u64 bytenr = node->bytenr;
6126 u64 num_bytes = node->num_bytes;
6128 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6131 if (!info->quota_enabled || !is_fstree(root_objectid))
6134 path = btrfs_alloc_path();
6139 path->leave_spinning = 1;
6141 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6142 BUG_ON(!is_data && refs_to_drop != 1);
6145 skinny_metadata = 0;
6147 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6148 bytenr, num_bytes, parent,
6149 root_objectid, owner_objectid,
6152 extent_slot = path->slots[0];
6153 while (extent_slot >= 0) {
6154 btrfs_item_key_to_cpu(path->nodes[0], &key,
6156 if (key.objectid != bytenr)
6158 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6159 key.offset == num_bytes) {
6163 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6164 key.offset == owner_objectid) {
6168 if (path->slots[0] - extent_slot > 5)
6172 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6173 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6174 if (found_extent && item_size < sizeof(*ei))
6177 if (!found_extent) {
6179 ret = remove_extent_backref(trans, extent_root, path,
6181 is_data, &last_ref);
6183 btrfs_abort_transaction(trans, extent_root, ret);
6186 btrfs_release_path(path);
6187 path->leave_spinning = 1;
6189 key.objectid = bytenr;
6190 key.type = BTRFS_EXTENT_ITEM_KEY;
6191 key.offset = num_bytes;
6193 if (!is_data && skinny_metadata) {
6194 key.type = BTRFS_METADATA_ITEM_KEY;
6195 key.offset = owner_objectid;
6198 ret = btrfs_search_slot(trans, extent_root,
6200 if (ret > 0 && skinny_metadata && path->slots[0]) {
6202 * Couldn't find our skinny metadata item,
6203 * see if we have ye olde extent item.
6206 btrfs_item_key_to_cpu(path->nodes[0], &key,
6208 if (key.objectid == bytenr &&
6209 key.type == BTRFS_EXTENT_ITEM_KEY &&
6210 key.offset == num_bytes)
6214 if (ret > 0 && skinny_metadata) {
6215 skinny_metadata = false;
6216 key.objectid = bytenr;
6217 key.type = BTRFS_EXTENT_ITEM_KEY;
6218 key.offset = num_bytes;
6219 btrfs_release_path(path);
6220 ret = btrfs_search_slot(trans, extent_root,
6225 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6228 btrfs_print_leaf(extent_root,
6232 btrfs_abort_transaction(trans, extent_root, ret);
6235 extent_slot = path->slots[0];
6237 } else if (WARN_ON(ret == -ENOENT)) {
6238 btrfs_print_leaf(extent_root, path->nodes[0]);
6240 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6241 bytenr, parent, root_objectid, owner_objectid,
6243 btrfs_abort_transaction(trans, extent_root, ret);
6246 btrfs_abort_transaction(trans, extent_root, ret);
6250 leaf = path->nodes[0];
6251 item_size = btrfs_item_size_nr(leaf, extent_slot);
6252 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6253 if (item_size < sizeof(*ei)) {
6254 BUG_ON(found_extent || extent_slot != path->slots[0]);
6255 ret = convert_extent_item_v0(trans, extent_root, path,
6258 btrfs_abort_transaction(trans, extent_root, ret);
6262 btrfs_release_path(path);
6263 path->leave_spinning = 1;
6265 key.objectid = bytenr;
6266 key.type = BTRFS_EXTENT_ITEM_KEY;
6267 key.offset = num_bytes;
6269 ret = btrfs_search_slot(trans, extent_root, &key, path,
6272 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6274 btrfs_print_leaf(extent_root, path->nodes[0]);
6277 btrfs_abort_transaction(trans, extent_root, ret);
6281 extent_slot = path->slots[0];
6282 leaf = path->nodes[0];
6283 item_size = btrfs_item_size_nr(leaf, extent_slot);
6286 BUG_ON(item_size < sizeof(*ei));
6287 ei = btrfs_item_ptr(leaf, extent_slot,
6288 struct btrfs_extent_item);
6289 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6290 key.type == BTRFS_EXTENT_ITEM_KEY) {
6291 struct btrfs_tree_block_info *bi;
6292 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6293 bi = (struct btrfs_tree_block_info *)(ei + 1);
6294 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6297 refs = btrfs_extent_refs(leaf, ei);
6298 if (refs < refs_to_drop) {
6299 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6300 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6302 btrfs_abort_transaction(trans, extent_root, ret);
6305 refs -= refs_to_drop;
6309 __run_delayed_extent_op(extent_op, leaf, ei);
6311 * In the case of inline back ref, reference count will
6312 * be updated by remove_extent_backref
6315 BUG_ON(!found_extent);
6317 btrfs_set_extent_refs(leaf, ei, refs);
6318 btrfs_mark_buffer_dirty(leaf);
6321 ret = remove_extent_backref(trans, extent_root, path,
6323 is_data, &last_ref);
6325 btrfs_abort_transaction(trans, extent_root, ret);
6329 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6333 BUG_ON(is_data && refs_to_drop !=
6334 extent_data_ref_count(root, path, iref));
6336 BUG_ON(path->slots[0] != extent_slot);
6338 BUG_ON(path->slots[0] != extent_slot + 1);
6339 path->slots[0] = extent_slot;
6345 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6348 btrfs_abort_transaction(trans, extent_root, ret);
6351 btrfs_release_path(path);
6354 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6356 btrfs_abort_transaction(trans, extent_root, ret);
6361 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6363 btrfs_abort_transaction(trans, extent_root, ret);
6367 btrfs_release_path(path);
6370 btrfs_free_path(path);
6375 * when we free an block, it is possible (and likely) that we free the last
6376 * delayed ref for that extent as well. This searches the delayed ref tree for
6377 * a given extent, and if there are no other delayed refs to be processed, it
6378 * removes it from the tree.
6380 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6381 struct btrfs_root *root, u64 bytenr)
6383 struct btrfs_delayed_ref_head *head;
6384 struct btrfs_delayed_ref_root *delayed_refs;
6387 delayed_refs = &trans->transaction->delayed_refs;
6388 spin_lock(&delayed_refs->lock);
6389 head = btrfs_find_delayed_ref_head(trans, bytenr);
6391 goto out_delayed_unlock;
6393 spin_lock(&head->lock);
6394 if (!list_empty(&head->ref_list))
6397 if (head->extent_op) {
6398 if (!head->must_insert_reserved)
6400 btrfs_free_delayed_extent_op(head->extent_op);
6401 head->extent_op = NULL;
6405 * waiting for the lock here would deadlock. If someone else has it
6406 * locked they are already in the process of dropping it anyway
6408 if (!mutex_trylock(&head->mutex))
6412 * at this point we have a head with no other entries. Go
6413 * ahead and process it.
6415 head->node.in_tree = 0;
6416 rb_erase(&head->href_node, &delayed_refs->href_root);
6418 atomic_dec(&delayed_refs->num_entries);
6421 * we don't take a ref on the node because we're removing it from the
6422 * tree, so we just steal the ref the tree was holding.
6424 delayed_refs->num_heads--;
6425 if (head->processing == 0)
6426 delayed_refs->num_heads_ready--;
6427 head->processing = 0;
6428 spin_unlock(&head->lock);
6429 spin_unlock(&delayed_refs->lock);
6431 BUG_ON(head->extent_op);
6432 if (head->must_insert_reserved)
6435 mutex_unlock(&head->mutex);
6436 btrfs_put_delayed_ref(&head->node);
6439 spin_unlock(&head->lock);
6442 spin_unlock(&delayed_refs->lock);
6446 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6447 struct btrfs_root *root,
6448 struct extent_buffer *buf,
6449 u64 parent, int last_ref)
6454 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6455 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6456 buf->start, buf->len,
6457 parent, root->root_key.objectid,
6458 btrfs_header_level(buf),
6459 BTRFS_DROP_DELAYED_REF, NULL, 0);
6460 BUG_ON(ret); /* -ENOMEM */
6466 if (btrfs_header_generation(buf) == trans->transid) {
6467 struct btrfs_block_group_cache *cache;
6469 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6470 ret = check_ref_cleanup(trans, root, buf->start);
6475 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6477 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6478 pin_down_extent(root, cache, buf->start, buf->len, 1);
6479 btrfs_put_block_group(cache);
6483 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6485 btrfs_add_free_space(cache, buf->start, buf->len);
6486 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6487 btrfs_put_block_group(cache);
6488 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6493 add_pinned_bytes(root->fs_info, buf->len,
6494 btrfs_header_level(buf),
6495 root->root_key.objectid);
6498 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6501 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6504 /* Can return -ENOMEM */
6505 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6506 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6507 u64 owner, u64 offset, int no_quota)
6510 struct btrfs_fs_info *fs_info = root->fs_info;
6512 if (btrfs_test_is_dummy_root(root))
6515 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6518 * tree log blocks never actually go into the extent allocation
6519 * tree, just update pinning info and exit early.
6521 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6522 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6523 /* unlocks the pinned mutex */
6524 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6526 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6527 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6529 parent, root_objectid, (int)owner,
6530 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6532 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6534 parent, root_objectid, owner,
6535 offset, BTRFS_DROP_DELAYED_REF,
6542 * when we wait for progress in the block group caching, its because
6543 * our allocation attempt failed at least once. So, we must sleep
6544 * and let some progress happen before we try again.
6546 * This function will sleep at least once waiting for new free space to
6547 * show up, and then it will check the block group free space numbers
6548 * for our min num_bytes. Another option is to have it go ahead
6549 * and look in the rbtree for a free extent of a given size, but this
6552 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6553 * any of the information in this block group.
6555 static noinline void
6556 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6559 struct btrfs_caching_control *caching_ctl;
6561 caching_ctl = get_caching_control(cache);
6565 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6566 (cache->free_space_ctl->free_space >= num_bytes));
6568 put_caching_control(caching_ctl);
6572 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6574 struct btrfs_caching_control *caching_ctl;
6577 caching_ctl = get_caching_control(cache);
6579 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6581 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6582 if (cache->cached == BTRFS_CACHE_ERROR)
6584 put_caching_control(caching_ctl);
6588 int __get_raid_index(u64 flags)
6590 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6591 return BTRFS_RAID_RAID10;
6592 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6593 return BTRFS_RAID_RAID1;
6594 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6595 return BTRFS_RAID_DUP;
6596 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6597 return BTRFS_RAID_RAID0;
6598 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6599 return BTRFS_RAID_RAID5;
6600 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6601 return BTRFS_RAID_RAID6;
6603 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6606 int get_block_group_index(struct btrfs_block_group_cache *cache)
6608 return __get_raid_index(cache->flags);
6611 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6612 [BTRFS_RAID_RAID10] = "raid10",
6613 [BTRFS_RAID_RAID1] = "raid1",
6614 [BTRFS_RAID_DUP] = "dup",
6615 [BTRFS_RAID_RAID0] = "raid0",
6616 [BTRFS_RAID_SINGLE] = "single",
6617 [BTRFS_RAID_RAID5] = "raid5",
6618 [BTRFS_RAID_RAID6] = "raid6",
6621 static const char *get_raid_name(enum btrfs_raid_types type)
6623 if (type >= BTRFS_NR_RAID_TYPES)
6626 return btrfs_raid_type_names[type];
6629 enum btrfs_loop_type {
6630 LOOP_CACHING_NOWAIT = 0,
6631 LOOP_CACHING_WAIT = 1,
6632 LOOP_ALLOC_CHUNK = 2,
6633 LOOP_NO_EMPTY_SIZE = 3,
6637 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6641 down_read(&cache->data_rwsem);
6645 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6648 btrfs_get_block_group(cache);
6650 down_read(&cache->data_rwsem);
6653 static struct btrfs_block_group_cache *
6654 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6655 struct btrfs_free_cluster *cluster,
6658 struct btrfs_block_group_cache *used_bg;
6659 bool locked = false;
6661 spin_lock(&cluster->refill_lock);
6663 if (used_bg == cluster->block_group)
6666 up_read(&used_bg->data_rwsem);
6667 btrfs_put_block_group(used_bg);
6670 used_bg = cluster->block_group;
6674 if (used_bg == block_group)
6677 btrfs_get_block_group(used_bg);
6682 if (down_read_trylock(&used_bg->data_rwsem))
6685 spin_unlock(&cluster->refill_lock);
6686 down_read(&used_bg->data_rwsem);
6692 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6696 up_read(&cache->data_rwsem);
6697 btrfs_put_block_group(cache);
6701 * walks the btree of allocated extents and find a hole of a given size.
6702 * The key ins is changed to record the hole:
6703 * ins->objectid == start position
6704 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6705 * ins->offset == the size of the hole.
6706 * Any available blocks before search_start are skipped.
6708 * If there is no suitable free space, we will record the max size of
6709 * the free space extent currently.
6711 static noinline int find_free_extent(struct btrfs_root *orig_root,
6712 u64 num_bytes, u64 empty_size,
6713 u64 hint_byte, struct btrfs_key *ins,
6714 u64 flags, int delalloc)
6717 struct btrfs_root *root = orig_root->fs_info->extent_root;
6718 struct btrfs_free_cluster *last_ptr = NULL;
6719 struct btrfs_block_group_cache *block_group = NULL;
6720 u64 search_start = 0;
6721 u64 max_extent_size = 0;
6722 int empty_cluster = 2 * 1024 * 1024;
6723 struct btrfs_space_info *space_info;
6725 int index = __get_raid_index(flags);
6726 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6727 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6728 bool failed_cluster_refill = false;
6729 bool failed_alloc = false;
6730 bool use_cluster = true;
6731 bool have_caching_bg = false;
6733 WARN_ON(num_bytes < root->sectorsize);
6734 ins->type = BTRFS_EXTENT_ITEM_KEY;
6738 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6740 space_info = __find_space_info(root->fs_info, flags);
6742 btrfs_err(root->fs_info, "No space info for %llu", flags);
6747 * If the space info is for both data and metadata it means we have a
6748 * small filesystem and we can't use the clustering stuff.
6750 if (btrfs_mixed_space_info(space_info))
6751 use_cluster = false;
6753 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6754 last_ptr = &root->fs_info->meta_alloc_cluster;
6755 if (!btrfs_test_opt(root, SSD))
6756 empty_cluster = 64 * 1024;
6759 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6760 btrfs_test_opt(root, SSD)) {
6761 last_ptr = &root->fs_info->data_alloc_cluster;
6765 spin_lock(&last_ptr->lock);
6766 if (last_ptr->block_group)
6767 hint_byte = last_ptr->window_start;
6768 spin_unlock(&last_ptr->lock);
6771 search_start = max(search_start, first_logical_byte(root, 0));
6772 search_start = max(search_start, hint_byte);
6777 if (search_start == hint_byte) {
6778 block_group = btrfs_lookup_block_group(root->fs_info,
6781 * we don't want to use the block group if it doesn't match our
6782 * allocation bits, or if its not cached.
6784 * However if we are re-searching with an ideal block group
6785 * picked out then we don't care that the block group is cached.
6787 if (block_group && block_group_bits(block_group, flags) &&
6788 block_group->cached != BTRFS_CACHE_NO) {
6789 down_read(&space_info->groups_sem);
6790 if (list_empty(&block_group->list) ||
6793 * someone is removing this block group,
6794 * we can't jump into the have_block_group
6795 * target because our list pointers are not
6798 btrfs_put_block_group(block_group);
6799 up_read(&space_info->groups_sem);
6801 index = get_block_group_index(block_group);
6802 btrfs_lock_block_group(block_group, delalloc);
6803 goto have_block_group;
6805 } else if (block_group) {
6806 btrfs_put_block_group(block_group);
6810 have_caching_bg = false;
6811 down_read(&space_info->groups_sem);
6812 list_for_each_entry(block_group, &space_info->block_groups[index],
6817 btrfs_grab_block_group(block_group, delalloc);
6818 search_start = block_group->key.objectid;
6821 * this can happen if we end up cycling through all the
6822 * raid types, but we want to make sure we only allocate
6823 * for the proper type.
6825 if (!block_group_bits(block_group, flags)) {
6826 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6827 BTRFS_BLOCK_GROUP_RAID1 |
6828 BTRFS_BLOCK_GROUP_RAID5 |
6829 BTRFS_BLOCK_GROUP_RAID6 |
6830 BTRFS_BLOCK_GROUP_RAID10;
6833 * if they asked for extra copies and this block group
6834 * doesn't provide them, bail. This does allow us to
6835 * fill raid0 from raid1.
6837 if ((flags & extra) && !(block_group->flags & extra))
6842 cached = block_group_cache_done(block_group);
6843 if (unlikely(!cached)) {
6844 ret = cache_block_group(block_group, 0);
6849 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6851 if (unlikely(block_group->ro))
6855 * Ok we want to try and use the cluster allocator, so
6859 struct btrfs_block_group_cache *used_block_group;
6860 unsigned long aligned_cluster;
6862 * the refill lock keeps out other
6863 * people trying to start a new cluster
6865 used_block_group = btrfs_lock_cluster(block_group,
6868 if (!used_block_group)
6869 goto refill_cluster;
6871 if (used_block_group != block_group &&
6872 (used_block_group->ro ||
6873 !block_group_bits(used_block_group, flags)))
6874 goto release_cluster;
6876 offset = btrfs_alloc_from_cluster(used_block_group,
6879 used_block_group->key.objectid,
6882 /* we have a block, we're done */
6883 spin_unlock(&last_ptr->refill_lock);
6884 trace_btrfs_reserve_extent_cluster(root,
6886 search_start, num_bytes);
6887 if (used_block_group != block_group) {
6888 btrfs_release_block_group(block_group,
6890 block_group = used_block_group;
6895 WARN_ON(last_ptr->block_group != used_block_group);
6897 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6898 * set up a new clusters, so lets just skip it
6899 * and let the allocator find whatever block
6900 * it can find. If we reach this point, we
6901 * will have tried the cluster allocator
6902 * plenty of times and not have found
6903 * anything, so we are likely way too
6904 * fragmented for the clustering stuff to find
6907 * However, if the cluster is taken from the
6908 * current block group, release the cluster
6909 * first, so that we stand a better chance of
6910 * succeeding in the unclustered
6912 if (loop >= LOOP_NO_EMPTY_SIZE &&
6913 used_block_group != block_group) {
6914 spin_unlock(&last_ptr->refill_lock);
6915 btrfs_release_block_group(used_block_group,
6917 goto unclustered_alloc;
6921 * this cluster didn't work out, free it and
6924 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6926 if (used_block_group != block_group)
6927 btrfs_release_block_group(used_block_group,
6930 if (loop >= LOOP_NO_EMPTY_SIZE) {
6931 spin_unlock(&last_ptr->refill_lock);
6932 goto unclustered_alloc;
6935 aligned_cluster = max_t(unsigned long,
6936 empty_cluster + empty_size,
6937 block_group->full_stripe_len);
6939 /* allocate a cluster in this block group */
6940 ret = btrfs_find_space_cluster(root, block_group,
6941 last_ptr, search_start,
6946 * now pull our allocation out of this
6949 offset = btrfs_alloc_from_cluster(block_group,
6955 /* we found one, proceed */
6956 spin_unlock(&last_ptr->refill_lock);
6957 trace_btrfs_reserve_extent_cluster(root,
6958 block_group, search_start,
6962 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6963 && !failed_cluster_refill) {
6964 spin_unlock(&last_ptr->refill_lock);
6966 failed_cluster_refill = true;
6967 wait_block_group_cache_progress(block_group,
6968 num_bytes + empty_cluster + empty_size);
6969 goto have_block_group;
6973 * at this point we either didn't find a cluster
6974 * or we weren't able to allocate a block from our
6975 * cluster. Free the cluster we've been trying
6976 * to use, and go to the next block group
6978 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6979 spin_unlock(&last_ptr->refill_lock);
6984 spin_lock(&block_group->free_space_ctl->tree_lock);
6986 block_group->free_space_ctl->free_space <
6987 num_bytes + empty_cluster + empty_size) {
6988 if (block_group->free_space_ctl->free_space >
6991 block_group->free_space_ctl->free_space;
6992 spin_unlock(&block_group->free_space_ctl->tree_lock);
6995 spin_unlock(&block_group->free_space_ctl->tree_lock);
6997 offset = btrfs_find_space_for_alloc(block_group, search_start,
6998 num_bytes, empty_size,
7001 * If we didn't find a chunk, and we haven't failed on this
7002 * block group before, and this block group is in the middle of
7003 * caching and we are ok with waiting, then go ahead and wait
7004 * for progress to be made, and set failed_alloc to true.
7006 * If failed_alloc is true then we've already waited on this
7007 * block group once and should move on to the next block group.
7009 if (!offset && !failed_alloc && !cached &&
7010 loop > LOOP_CACHING_NOWAIT) {
7011 wait_block_group_cache_progress(block_group,
7012 num_bytes + empty_size);
7013 failed_alloc = true;
7014 goto have_block_group;
7015 } else if (!offset) {
7017 have_caching_bg = true;
7021 search_start = ALIGN(offset, root->stripesize);
7023 /* move on to the next group */
7024 if (search_start + num_bytes >
7025 block_group->key.objectid + block_group->key.offset) {
7026 btrfs_add_free_space(block_group, offset, num_bytes);
7030 if (offset < search_start)
7031 btrfs_add_free_space(block_group, offset,
7032 search_start - offset);
7033 BUG_ON(offset > search_start);
7035 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7036 alloc_type, delalloc);
7037 if (ret == -EAGAIN) {
7038 btrfs_add_free_space(block_group, offset, num_bytes);
7042 /* we are all good, lets return */
7043 ins->objectid = search_start;
7044 ins->offset = num_bytes;
7046 trace_btrfs_reserve_extent(orig_root, block_group,
7047 search_start, num_bytes);
7048 btrfs_release_block_group(block_group, delalloc);
7051 failed_cluster_refill = false;
7052 failed_alloc = false;
7053 BUG_ON(index != get_block_group_index(block_group));
7054 btrfs_release_block_group(block_group, delalloc);
7056 up_read(&space_info->groups_sem);
7058 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7061 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7065 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7066 * caching kthreads as we move along
7067 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7068 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7069 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7072 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7075 if (loop == LOOP_ALLOC_CHUNK) {
7076 struct btrfs_trans_handle *trans;
7079 trans = current->journal_info;
7083 trans = btrfs_join_transaction(root);
7085 if (IS_ERR(trans)) {
7086 ret = PTR_ERR(trans);
7090 ret = do_chunk_alloc(trans, root, flags,
7093 * Do not bail out on ENOSPC since we
7094 * can do more things.
7096 if (ret < 0 && ret != -ENOSPC)
7097 btrfs_abort_transaction(trans,
7102 btrfs_end_transaction(trans, root);
7107 if (loop == LOOP_NO_EMPTY_SIZE) {
7113 } else if (!ins->objectid) {
7115 } else if (ins->objectid) {
7120 ins->offset = max_extent_size;
7124 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7125 int dump_block_groups)
7127 struct btrfs_block_group_cache *cache;
7130 spin_lock(&info->lock);
7131 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7133 info->total_bytes - info->bytes_used - info->bytes_pinned -
7134 info->bytes_reserved - info->bytes_readonly,
7135 (info->full) ? "" : "not ");
7136 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7137 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7138 info->total_bytes, info->bytes_used, info->bytes_pinned,
7139 info->bytes_reserved, info->bytes_may_use,
7140 info->bytes_readonly);
7141 spin_unlock(&info->lock);
7143 if (!dump_block_groups)
7146 down_read(&info->groups_sem);
7148 list_for_each_entry(cache, &info->block_groups[index], list) {
7149 spin_lock(&cache->lock);
7150 printk(KERN_INFO "BTRFS: "
7151 "block group %llu has %llu bytes, "
7152 "%llu used %llu pinned %llu reserved %s\n",
7153 cache->key.objectid, cache->key.offset,
7154 btrfs_block_group_used(&cache->item), cache->pinned,
7155 cache->reserved, cache->ro ? "[readonly]" : "");
7156 btrfs_dump_free_space(cache, bytes);
7157 spin_unlock(&cache->lock);
7159 if (++index < BTRFS_NR_RAID_TYPES)
7161 up_read(&info->groups_sem);
7164 int btrfs_reserve_extent(struct btrfs_root *root,
7165 u64 num_bytes, u64 min_alloc_size,
7166 u64 empty_size, u64 hint_byte,
7167 struct btrfs_key *ins, int is_data, int delalloc)
7169 bool final_tried = false;
7173 flags = btrfs_get_alloc_profile(root, is_data);
7175 WARN_ON(num_bytes < root->sectorsize);
7176 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7179 if (ret == -ENOSPC) {
7180 if (!final_tried && ins->offset) {
7181 num_bytes = min(num_bytes >> 1, ins->offset);
7182 num_bytes = round_down(num_bytes, root->sectorsize);
7183 num_bytes = max(num_bytes, min_alloc_size);
7184 if (num_bytes == min_alloc_size)
7187 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7188 struct btrfs_space_info *sinfo;
7190 sinfo = __find_space_info(root->fs_info, flags);
7191 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7194 dump_space_info(sinfo, num_bytes, 1);
7201 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7203 int pin, int delalloc)
7205 struct btrfs_block_group_cache *cache;
7208 cache = btrfs_lookup_block_group(root->fs_info, start);
7210 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7216 pin_down_extent(root, cache, start, len, 1);
7218 if (btrfs_test_opt(root, DISCARD))
7219 ret = btrfs_discard_extent(root, start, len, NULL);
7220 btrfs_add_free_space(cache, start, len);
7221 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7224 btrfs_put_block_group(cache);
7226 trace_btrfs_reserved_extent_free(root, start, len);
7231 int btrfs_free_reserved_extent(struct btrfs_root *root,
7232 u64 start, u64 len, int delalloc)
7234 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7237 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7240 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7243 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7244 struct btrfs_root *root,
7245 u64 parent, u64 root_objectid,
7246 u64 flags, u64 owner, u64 offset,
7247 struct btrfs_key *ins, int ref_mod)
7250 struct btrfs_fs_info *fs_info = root->fs_info;
7251 struct btrfs_extent_item *extent_item;
7252 struct btrfs_extent_inline_ref *iref;
7253 struct btrfs_path *path;
7254 struct extent_buffer *leaf;
7259 type = BTRFS_SHARED_DATA_REF_KEY;
7261 type = BTRFS_EXTENT_DATA_REF_KEY;
7263 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7265 path = btrfs_alloc_path();
7269 path->leave_spinning = 1;
7270 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7273 btrfs_free_path(path);
7277 leaf = path->nodes[0];
7278 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7279 struct btrfs_extent_item);
7280 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7281 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7282 btrfs_set_extent_flags(leaf, extent_item,
7283 flags | BTRFS_EXTENT_FLAG_DATA);
7285 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7286 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7288 struct btrfs_shared_data_ref *ref;
7289 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7290 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7291 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7293 struct btrfs_extent_data_ref *ref;
7294 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7295 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7296 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7297 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7298 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7301 btrfs_mark_buffer_dirty(path->nodes[0]);
7302 btrfs_free_path(path);
7304 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7305 if (ret) { /* -ENOENT, logic error */
7306 btrfs_err(fs_info, "update block group failed for %llu %llu",
7307 ins->objectid, ins->offset);
7310 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7314 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7315 struct btrfs_root *root,
7316 u64 parent, u64 root_objectid,
7317 u64 flags, struct btrfs_disk_key *key,
7318 int level, struct btrfs_key *ins,
7322 struct btrfs_fs_info *fs_info = root->fs_info;
7323 struct btrfs_extent_item *extent_item;
7324 struct btrfs_tree_block_info *block_info;
7325 struct btrfs_extent_inline_ref *iref;
7326 struct btrfs_path *path;
7327 struct extent_buffer *leaf;
7328 u32 size = sizeof(*extent_item) + sizeof(*iref);
7329 u64 num_bytes = ins->offset;
7330 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7333 if (!skinny_metadata)
7334 size += sizeof(*block_info);
7336 path = btrfs_alloc_path();
7338 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7343 path->leave_spinning = 1;
7344 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7347 btrfs_free_path(path);
7348 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7353 leaf = path->nodes[0];
7354 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7355 struct btrfs_extent_item);
7356 btrfs_set_extent_refs(leaf, extent_item, 1);
7357 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7358 btrfs_set_extent_flags(leaf, extent_item,
7359 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7361 if (skinny_metadata) {
7362 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7363 num_bytes = root->nodesize;
7365 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7366 btrfs_set_tree_block_key(leaf, block_info, key);
7367 btrfs_set_tree_block_level(leaf, block_info, level);
7368 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7372 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7373 btrfs_set_extent_inline_ref_type(leaf, iref,
7374 BTRFS_SHARED_BLOCK_REF_KEY);
7375 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7377 btrfs_set_extent_inline_ref_type(leaf, iref,
7378 BTRFS_TREE_BLOCK_REF_KEY);
7379 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7382 btrfs_mark_buffer_dirty(leaf);
7383 btrfs_free_path(path);
7385 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7387 if (ret) { /* -ENOENT, logic error */
7388 btrfs_err(fs_info, "update block group failed for %llu %llu",
7389 ins->objectid, ins->offset);
7393 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7397 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7398 struct btrfs_root *root,
7399 u64 root_objectid, u64 owner,
7400 u64 offset, struct btrfs_key *ins)
7404 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7406 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7408 root_objectid, owner, offset,
7409 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7414 * this is used by the tree logging recovery code. It records that
7415 * an extent has been allocated and makes sure to clear the free
7416 * space cache bits as well
7418 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7419 struct btrfs_root *root,
7420 u64 root_objectid, u64 owner, u64 offset,
7421 struct btrfs_key *ins)
7424 struct btrfs_block_group_cache *block_group;
7427 * Mixed block groups will exclude before processing the log so we only
7428 * need to do the exlude dance if this fs isn't mixed.
7430 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7431 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7436 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7440 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7441 RESERVE_ALLOC_NO_ACCOUNT, 0);
7442 BUG_ON(ret); /* logic error */
7443 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7444 0, owner, offset, ins, 1);
7445 btrfs_put_block_group(block_group);
7449 static struct extent_buffer *
7450 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7451 u64 bytenr, int level)
7453 struct extent_buffer *buf;
7455 buf = btrfs_find_create_tree_block(root, bytenr);
7457 return ERR_PTR(-ENOMEM);
7458 btrfs_set_header_generation(buf, trans->transid);
7459 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7460 btrfs_tree_lock(buf);
7461 clean_tree_block(trans, root->fs_info, buf);
7462 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7464 btrfs_set_lock_blocking(buf);
7465 btrfs_set_buffer_uptodate(buf);
7467 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7468 buf->log_index = root->log_transid % 2;
7470 * we allow two log transactions at a time, use different
7471 * EXENT bit to differentiate dirty pages.
7473 if (buf->log_index == 0)
7474 set_extent_dirty(&root->dirty_log_pages, buf->start,
7475 buf->start + buf->len - 1, GFP_NOFS);
7477 set_extent_new(&root->dirty_log_pages, buf->start,
7478 buf->start + buf->len - 1, GFP_NOFS);
7480 buf->log_index = -1;
7481 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7482 buf->start + buf->len - 1, GFP_NOFS);
7484 trans->blocks_used++;
7485 /* this returns a buffer locked for blocking */
7489 static struct btrfs_block_rsv *
7490 use_block_rsv(struct btrfs_trans_handle *trans,
7491 struct btrfs_root *root, u32 blocksize)
7493 struct btrfs_block_rsv *block_rsv;
7494 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7496 bool global_updated = false;
7498 block_rsv = get_block_rsv(trans, root);
7500 if (unlikely(block_rsv->size == 0))
7503 ret = block_rsv_use_bytes(block_rsv, blocksize);
7507 if (block_rsv->failfast)
7508 return ERR_PTR(ret);
7510 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7511 global_updated = true;
7512 update_global_block_rsv(root->fs_info);
7516 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7517 static DEFINE_RATELIMIT_STATE(_rs,
7518 DEFAULT_RATELIMIT_INTERVAL * 10,
7519 /*DEFAULT_RATELIMIT_BURST*/ 1);
7520 if (__ratelimit(&_rs))
7522 "BTRFS: block rsv returned %d\n", ret);
7525 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7526 BTRFS_RESERVE_NO_FLUSH);
7530 * If we couldn't reserve metadata bytes try and use some from
7531 * the global reserve if its space type is the same as the global
7534 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7535 block_rsv->space_info == global_rsv->space_info) {
7536 ret = block_rsv_use_bytes(global_rsv, blocksize);
7540 return ERR_PTR(ret);
7543 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7544 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7546 block_rsv_add_bytes(block_rsv, blocksize, 0);
7547 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7551 * finds a free extent and does all the dirty work required for allocation
7552 * returns the key for the extent through ins, and a tree buffer for
7553 * the first block of the extent through buf.
7555 * returns the tree buffer or an ERR_PTR on error.
7557 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7558 struct btrfs_root *root,
7559 u64 parent, u64 root_objectid,
7560 struct btrfs_disk_key *key, int level,
7561 u64 hint, u64 empty_size)
7563 struct btrfs_key ins;
7564 struct btrfs_block_rsv *block_rsv;
7565 struct extent_buffer *buf;
7566 struct btrfs_delayed_extent_op *extent_op;
7569 u32 blocksize = root->nodesize;
7570 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7573 if (btrfs_test_is_dummy_root(root)) {
7574 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7577 root->alloc_bytenr += blocksize;
7581 block_rsv = use_block_rsv(trans, root, blocksize);
7582 if (IS_ERR(block_rsv))
7583 return ERR_CAST(block_rsv);
7585 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7586 empty_size, hint, &ins, 0, 0);
7590 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7593 goto out_free_reserved;
7596 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7598 parent = ins.objectid;
7599 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7603 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7604 extent_op = btrfs_alloc_delayed_extent_op();
7610 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7612 memset(&extent_op->key, 0, sizeof(extent_op->key));
7613 extent_op->flags_to_set = flags;
7614 if (skinny_metadata)
7615 extent_op->update_key = 0;
7617 extent_op->update_key = 1;
7618 extent_op->update_flags = 1;
7619 extent_op->is_data = 0;
7620 extent_op->level = level;
7622 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7623 ins.objectid, ins.offset,
7624 parent, root_objectid, level,
7625 BTRFS_ADD_DELAYED_EXTENT,
7628 goto out_free_delayed;
7633 btrfs_free_delayed_extent_op(extent_op);
7635 free_extent_buffer(buf);
7637 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7639 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7640 return ERR_PTR(ret);
7643 struct walk_control {
7644 u64 refs[BTRFS_MAX_LEVEL];
7645 u64 flags[BTRFS_MAX_LEVEL];
7646 struct btrfs_key update_progress;
7657 #define DROP_REFERENCE 1
7658 #define UPDATE_BACKREF 2
7660 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7661 struct btrfs_root *root,
7662 struct walk_control *wc,
7663 struct btrfs_path *path)
7671 struct btrfs_key key;
7672 struct extent_buffer *eb;
7677 if (path->slots[wc->level] < wc->reada_slot) {
7678 wc->reada_count = wc->reada_count * 2 / 3;
7679 wc->reada_count = max(wc->reada_count, 2);
7681 wc->reada_count = wc->reada_count * 3 / 2;
7682 wc->reada_count = min_t(int, wc->reada_count,
7683 BTRFS_NODEPTRS_PER_BLOCK(root));
7686 eb = path->nodes[wc->level];
7687 nritems = btrfs_header_nritems(eb);
7688 blocksize = root->nodesize;
7690 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7691 if (nread >= wc->reada_count)
7695 bytenr = btrfs_node_blockptr(eb, slot);
7696 generation = btrfs_node_ptr_generation(eb, slot);
7698 if (slot == path->slots[wc->level])
7701 if (wc->stage == UPDATE_BACKREF &&
7702 generation <= root->root_key.offset)
7705 /* We don't lock the tree block, it's OK to be racy here */
7706 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7707 wc->level - 1, 1, &refs,
7709 /* We don't care about errors in readahead. */
7714 if (wc->stage == DROP_REFERENCE) {
7718 if (wc->level == 1 &&
7719 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7721 if (!wc->update_ref ||
7722 generation <= root->root_key.offset)
7724 btrfs_node_key_to_cpu(eb, &key, slot);
7725 ret = btrfs_comp_cpu_keys(&key,
7726 &wc->update_progress);
7730 if (wc->level == 1 &&
7731 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7735 readahead_tree_block(root, bytenr);
7738 wc->reada_slot = slot;
7742 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7743 * for later qgroup accounting.
7745 * Current, this function does nothing.
7747 static int account_leaf_items(struct btrfs_trans_handle *trans,
7748 struct btrfs_root *root,
7749 struct extent_buffer *eb)
7751 int nr = btrfs_header_nritems(eb);
7753 struct btrfs_key key;
7754 struct btrfs_file_extent_item *fi;
7755 u64 bytenr, num_bytes;
7757 for (i = 0; i < nr; i++) {
7758 btrfs_item_key_to_cpu(eb, &key, i);
7760 if (key.type != BTRFS_EXTENT_DATA_KEY)
7763 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7764 /* filter out non qgroup-accountable extents */
7765 extent_type = btrfs_file_extent_type(eb, fi);
7767 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7770 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7774 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7780 * Walk up the tree from the bottom, freeing leaves and any interior
7781 * nodes which have had all slots visited. If a node (leaf or
7782 * interior) is freed, the node above it will have it's slot
7783 * incremented. The root node will never be freed.
7785 * At the end of this function, we should have a path which has all
7786 * slots incremented to the next position for a search. If we need to
7787 * read a new node it will be NULL and the node above it will have the
7788 * correct slot selected for a later read.
7790 * If we increment the root nodes slot counter past the number of
7791 * elements, 1 is returned to signal completion of the search.
7793 static int adjust_slots_upwards(struct btrfs_root *root,
7794 struct btrfs_path *path, int root_level)
7798 struct extent_buffer *eb;
7800 if (root_level == 0)
7803 while (level <= root_level) {
7804 eb = path->nodes[level];
7805 nr = btrfs_header_nritems(eb);
7806 path->slots[level]++;
7807 slot = path->slots[level];
7808 if (slot >= nr || level == 0) {
7810 * Don't free the root - we will detect this
7811 * condition after our loop and return a
7812 * positive value for caller to stop walking the tree.
7814 if (level != root_level) {
7815 btrfs_tree_unlock_rw(eb, path->locks[level]);
7816 path->locks[level] = 0;
7818 free_extent_buffer(eb);
7819 path->nodes[level] = NULL;
7820 path->slots[level] = 0;
7824 * We have a valid slot to walk back down
7825 * from. Stop here so caller can process these
7834 eb = path->nodes[root_level];
7835 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7842 * root_eb is the subtree root and is locked before this function is called.
7843 * TODO: Modify this function to mark all (including complete shared node)
7844 * to dirty_extent_root to allow it get accounted in qgroup.
7846 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7847 struct btrfs_root *root,
7848 struct extent_buffer *root_eb,
7854 struct extent_buffer *eb = root_eb;
7855 struct btrfs_path *path = NULL;
7857 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7858 BUG_ON(root_eb == NULL);
7860 if (!root->fs_info->quota_enabled)
7863 if (!extent_buffer_uptodate(root_eb)) {
7864 ret = btrfs_read_buffer(root_eb, root_gen);
7869 if (root_level == 0) {
7870 ret = account_leaf_items(trans, root, root_eb);
7874 path = btrfs_alloc_path();
7879 * Walk down the tree. Missing extent blocks are filled in as
7880 * we go. Metadata is accounted every time we read a new
7883 * When we reach a leaf, we account for file extent items in it,
7884 * walk back up the tree (adjusting slot pointers as we go)
7885 * and restart the search process.
7887 extent_buffer_get(root_eb); /* For path */
7888 path->nodes[root_level] = root_eb;
7889 path->slots[root_level] = 0;
7890 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7893 while (level >= 0) {
7894 if (path->nodes[level] == NULL) {
7899 /* We need to get child blockptr/gen from
7900 * parent before we can read it. */
7901 eb = path->nodes[level + 1];
7902 parent_slot = path->slots[level + 1];
7903 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7904 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7906 eb = read_tree_block(root, child_bytenr, child_gen);
7910 } else if (!extent_buffer_uptodate(eb)) {
7911 free_extent_buffer(eb);
7916 path->nodes[level] = eb;
7917 path->slots[level] = 0;
7919 btrfs_tree_read_lock(eb);
7920 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7921 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7925 ret = account_leaf_items(trans, root, path->nodes[level]);
7929 /* Nonzero return here means we completed our search */
7930 ret = adjust_slots_upwards(root, path, root_level);
7934 /* Restart search with new slots */
7943 btrfs_free_path(path);
7949 * helper to process tree block while walking down the tree.
7951 * when wc->stage == UPDATE_BACKREF, this function updates
7952 * back refs for pointers in the block.
7954 * NOTE: return value 1 means we should stop walking down.
7956 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7957 struct btrfs_root *root,
7958 struct btrfs_path *path,
7959 struct walk_control *wc, int lookup_info)
7961 int level = wc->level;
7962 struct extent_buffer *eb = path->nodes[level];
7963 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7966 if (wc->stage == UPDATE_BACKREF &&
7967 btrfs_header_owner(eb) != root->root_key.objectid)
7971 * when reference count of tree block is 1, it won't increase
7972 * again. once full backref flag is set, we never clear it.
7975 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7976 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7977 BUG_ON(!path->locks[level]);
7978 ret = btrfs_lookup_extent_info(trans, root,
7979 eb->start, level, 1,
7982 BUG_ON(ret == -ENOMEM);
7985 BUG_ON(wc->refs[level] == 0);
7988 if (wc->stage == DROP_REFERENCE) {
7989 if (wc->refs[level] > 1)
7992 if (path->locks[level] && !wc->keep_locks) {
7993 btrfs_tree_unlock_rw(eb, path->locks[level]);
7994 path->locks[level] = 0;
7999 /* wc->stage == UPDATE_BACKREF */
8000 if (!(wc->flags[level] & flag)) {
8001 BUG_ON(!path->locks[level]);
8002 ret = btrfs_inc_ref(trans, root, eb, 1);
8003 BUG_ON(ret); /* -ENOMEM */
8004 ret = btrfs_dec_ref(trans, root, eb, 0);
8005 BUG_ON(ret); /* -ENOMEM */
8006 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8008 btrfs_header_level(eb), 0);
8009 BUG_ON(ret); /* -ENOMEM */
8010 wc->flags[level] |= flag;
8014 * the block is shared by multiple trees, so it's not good to
8015 * keep the tree lock
8017 if (path->locks[level] && level > 0) {
8018 btrfs_tree_unlock_rw(eb, path->locks[level]);
8019 path->locks[level] = 0;
8025 * helper to process tree block pointer.
8027 * when wc->stage == DROP_REFERENCE, this function checks
8028 * reference count of the block pointed to. if the block
8029 * is shared and we need update back refs for the subtree
8030 * rooted at the block, this function changes wc->stage to
8031 * UPDATE_BACKREF. if the block is shared and there is no
8032 * need to update back, this function drops the reference
8035 * NOTE: return value 1 means we should stop walking down.
8037 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8038 struct btrfs_root *root,
8039 struct btrfs_path *path,
8040 struct walk_control *wc, int *lookup_info)
8046 struct btrfs_key key;
8047 struct extent_buffer *next;
8048 int level = wc->level;
8051 bool need_account = false;
8053 generation = btrfs_node_ptr_generation(path->nodes[level],
8054 path->slots[level]);
8056 * if the lower level block was created before the snapshot
8057 * was created, we know there is no need to update back refs
8060 if (wc->stage == UPDATE_BACKREF &&
8061 generation <= root->root_key.offset) {
8066 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8067 blocksize = root->nodesize;
8069 next = btrfs_find_tree_block(root->fs_info, bytenr);
8071 next = btrfs_find_create_tree_block(root, bytenr);
8074 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8078 btrfs_tree_lock(next);
8079 btrfs_set_lock_blocking(next);
8081 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8082 &wc->refs[level - 1],
8083 &wc->flags[level - 1]);
8085 btrfs_tree_unlock(next);
8089 if (unlikely(wc->refs[level - 1] == 0)) {
8090 btrfs_err(root->fs_info, "Missing references.");
8095 if (wc->stage == DROP_REFERENCE) {
8096 if (wc->refs[level - 1] > 1) {
8097 need_account = true;
8099 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8102 if (!wc->update_ref ||
8103 generation <= root->root_key.offset)
8106 btrfs_node_key_to_cpu(path->nodes[level], &key,
8107 path->slots[level]);
8108 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8112 wc->stage = UPDATE_BACKREF;
8113 wc->shared_level = level - 1;
8117 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8121 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8122 btrfs_tree_unlock(next);
8123 free_extent_buffer(next);
8129 if (reada && level == 1)
8130 reada_walk_down(trans, root, wc, path);
8131 next = read_tree_block(root, bytenr, generation);
8133 return PTR_ERR(next);
8134 } else if (!extent_buffer_uptodate(next)) {
8135 free_extent_buffer(next);
8138 btrfs_tree_lock(next);
8139 btrfs_set_lock_blocking(next);
8143 BUG_ON(level != btrfs_header_level(next));
8144 path->nodes[level] = next;
8145 path->slots[level] = 0;
8146 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8152 wc->refs[level - 1] = 0;
8153 wc->flags[level - 1] = 0;
8154 if (wc->stage == DROP_REFERENCE) {
8155 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8156 parent = path->nodes[level]->start;
8158 BUG_ON(root->root_key.objectid !=
8159 btrfs_header_owner(path->nodes[level]));
8164 ret = account_shared_subtree(trans, root, next,
8165 generation, level - 1);
8167 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8168 "%d accounting shared subtree. Quota "
8169 "is out of sync, rescan required.\n",
8170 root->fs_info->sb->s_id, ret);
8173 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8174 root->root_key.objectid, level - 1, 0, 0);
8175 BUG_ON(ret); /* -ENOMEM */
8177 btrfs_tree_unlock(next);
8178 free_extent_buffer(next);
8184 * helper to process tree block while walking up the tree.
8186 * when wc->stage == DROP_REFERENCE, this function drops
8187 * reference count on the block.
8189 * when wc->stage == UPDATE_BACKREF, this function changes
8190 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8191 * to UPDATE_BACKREF previously while processing the block.
8193 * NOTE: return value 1 means we should stop walking up.
8195 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8196 struct btrfs_root *root,
8197 struct btrfs_path *path,
8198 struct walk_control *wc)
8201 int level = wc->level;
8202 struct extent_buffer *eb = path->nodes[level];
8205 if (wc->stage == UPDATE_BACKREF) {
8206 BUG_ON(wc->shared_level < level);
8207 if (level < wc->shared_level)
8210 ret = find_next_key(path, level + 1, &wc->update_progress);
8214 wc->stage = DROP_REFERENCE;
8215 wc->shared_level = -1;
8216 path->slots[level] = 0;
8219 * check reference count again if the block isn't locked.
8220 * we should start walking down the tree again if reference
8223 if (!path->locks[level]) {
8225 btrfs_tree_lock(eb);
8226 btrfs_set_lock_blocking(eb);
8227 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8229 ret = btrfs_lookup_extent_info(trans, root,
8230 eb->start, level, 1,
8234 btrfs_tree_unlock_rw(eb, path->locks[level]);
8235 path->locks[level] = 0;
8238 BUG_ON(wc->refs[level] == 0);
8239 if (wc->refs[level] == 1) {
8240 btrfs_tree_unlock_rw(eb, path->locks[level]);
8241 path->locks[level] = 0;
8247 /* wc->stage == DROP_REFERENCE */
8248 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8250 if (wc->refs[level] == 1) {
8252 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8253 ret = btrfs_dec_ref(trans, root, eb, 1);
8255 ret = btrfs_dec_ref(trans, root, eb, 0);
8256 BUG_ON(ret); /* -ENOMEM */
8257 ret = account_leaf_items(trans, root, eb);
8259 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8260 "%d accounting leaf items. Quota "
8261 "is out of sync, rescan required.\n",
8262 root->fs_info->sb->s_id, ret);
8265 /* make block locked assertion in clean_tree_block happy */
8266 if (!path->locks[level] &&
8267 btrfs_header_generation(eb) == trans->transid) {
8268 btrfs_tree_lock(eb);
8269 btrfs_set_lock_blocking(eb);
8270 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8272 clean_tree_block(trans, root->fs_info, eb);
8275 if (eb == root->node) {
8276 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8279 BUG_ON(root->root_key.objectid !=
8280 btrfs_header_owner(eb));
8282 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8283 parent = path->nodes[level + 1]->start;
8285 BUG_ON(root->root_key.objectid !=
8286 btrfs_header_owner(path->nodes[level + 1]));
8289 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8291 wc->refs[level] = 0;
8292 wc->flags[level] = 0;
8296 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8297 struct btrfs_root *root,
8298 struct btrfs_path *path,
8299 struct walk_control *wc)
8301 int level = wc->level;
8302 int lookup_info = 1;
8305 while (level >= 0) {
8306 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8313 if (path->slots[level] >=
8314 btrfs_header_nritems(path->nodes[level]))
8317 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8319 path->slots[level]++;
8328 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8329 struct btrfs_root *root,
8330 struct btrfs_path *path,
8331 struct walk_control *wc, int max_level)
8333 int level = wc->level;
8336 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8337 while (level < max_level && path->nodes[level]) {
8339 if (path->slots[level] + 1 <
8340 btrfs_header_nritems(path->nodes[level])) {
8341 path->slots[level]++;
8344 ret = walk_up_proc(trans, root, path, wc);
8348 if (path->locks[level]) {
8349 btrfs_tree_unlock_rw(path->nodes[level],
8350 path->locks[level]);
8351 path->locks[level] = 0;
8353 free_extent_buffer(path->nodes[level]);
8354 path->nodes[level] = NULL;
8362 * drop a subvolume tree.
8364 * this function traverses the tree freeing any blocks that only
8365 * referenced by the tree.
8367 * when a shared tree block is found. this function decreases its
8368 * reference count by one. if update_ref is true, this function
8369 * also make sure backrefs for the shared block and all lower level
8370 * blocks are properly updated.
8372 * If called with for_reloc == 0, may exit early with -EAGAIN
8374 int btrfs_drop_snapshot(struct btrfs_root *root,
8375 struct btrfs_block_rsv *block_rsv, int update_ref,
8378 struct btrfs_path *path;
8379 struct btrfs_trans_handle *trans;
8380 struct btrfs_root *tree_root = root->fs_info->tree_root;
8381 struct btrfs_root_item *root_item = &root->root_item;
8382 struct walk_control *wc;
8383 struct btrfs_key key;
8387 bool root_dropped = false;
8389 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8391 path = btrfs_alloc_path();
8397 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8399 btrfs_free_path(path);
8404 trans = btrfs_start_transaction(tree_root, 0);
8405 if (IS_ERR(trans)) {
8406 err = PTR_ERR(trans);
8411 trans->block_rsv = block_rsv;
8413 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8414 level = btrfs_header_level(root->node);
8415 path->nodes[level] = btrfs_lock_root_node(root);
8416 btrfs_set_lock_blocking(path->nodes[level]);
8417 path->slots[level] = 0;
8418 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8419 memset(&wc->update_progress, 0,
8420 sizeof(wc->update_progress));
8422 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8423 memcpy(&wc->update_progress, &key,
8424 sizeof(wc->update_progress));
8426 level = root_item->drop_level;
8428 path->lowest_level = level;
8429 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8430 path->lowest_level = 0;
8438 * unlock our path, this is safe because only this
8439 * function is allowed to delete this snapshot
8441 btrfs_unlock_up_safe(path, 0);
8443 level = btrfs_header_level(root->node);
8445 btrfs_tree_lock(path->nodes[level]);
8446 btrfs_set_lock_blocking(path->nodes[level]);
8447 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8449 ret = btrfs_lookup_extent_info(trans, root,
8450 path->nodes[level]->start,
8451 level, 1, &wc->refs[level],
8457 BUG_ON(wc->refs[level] == 0);
8459 if (level == root_item->drop_level)
8462 btrfs_tree_unlock(path->nodes[level]);
8463 path->locks[level] = 0;
8464 WARN_ON(wc->refs[level] != 1);
8470 wc->shared_level = -1;
8471 wc->stage = DROP_REFERENCE;
8472 wc->update_ref = update_ref;
8474 wc->for_reloc = for_reloc;
8475 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8479 ret = walk_down_tree(trans, root, path, wc);
8485 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8492 BUG_ON(wc->stage != DROP_REFERENCE);
8496 if (wc->stage == DROP_REFERENCE) {
8498 btrfs_node_key(path->nodes[level],
8499 &root_item->drop_progress,
8500 path->slots[level]);
8501 root_item->drop_level = level;
8504 BUG_ON(wc->level == 0);
8505 if (btrfs_should_end_transaction(trans, tree_root) ||
8506 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8507 ret = btrfs_update_root(trans, tree_root,
8511 btrfs_abort_transaction(trans, tree_root, ret);
8516 btrfs_end_transaction_throttle(trans, tree_root);
8517 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8518 pr_debug("BTRFS: drop snapshot early exit\n");
8523 trans = btrfs_start_transaction(tree_root, 0);
8524 if (IS_ERR(trans)) {
8525 err = PTR_ERR(trans);
8529 trans->block_rsv = block_rsv;
8532 btrfs_release_path(path);
8536 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8538 btrfs_abort_transaction(trans, tree_root, ret);
8542 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8543 ret = btrfs_find_root(tree_root, &root->root_key, path,
8546 btrfs_abort_transaction(trans, tree_root, ret);
8549 } else if (ret > 0) {
8550 /* if we fail to delete the orphan item this time
8551 * around, it'll get picked up the next time.
8553 * The most common failure here is just -ENOENT.
8555 btrfs_del_orphan_item(trans, tree_root,
8556 root->root_key.objectid);
8560 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8561 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8563 free_extent_buffer(root->node);
8564 free_extent_buffer(root->commit_root);
8565 btrfs_put_fs_root(root);
8567 root_dropped = true;
8569 btrfs_end_transaction_throttle(trans, tree_root);
8572 btrfs_free_path(path);
8575 * So if we need to stop dropping the snapshot for whatever reason we
8576 * need to make sure to add it back to the dead root list so that we
8577 * keep trying to do the work later. This also cleans up roots if we
8578 * don't have it in the radix (like when we recover after a power fail
8579 * or unmount) so we don't leak memory.
8581 if (!for_reloc && root_dropped == false)
8582 btrfs_add_dead_root(root);
8583 if (err && err != -EAGAIN)
8584 btrfs_std_error(root->fs_info, err);
8589 * drop subtree rooted at tree block 'node'.
8591 * NOTE: this function will unlock and release tree block 'node'
8592 * only used by relocation code
8594 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8595 struct btrfs_root *root,
8596 struct extent_buffer *node,
8597 struct extent_buffer *parent)
8599 struct btrfs_path *path;
8600 struct walk_control *wc;
8606 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8608 path = btrfs_alloc_path();
8612 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8614 btrfs_free_path(path);
8618 btrfs_assert_tree_locked(parent);
8619 parent_level = btrfs_header_level(parent);
8620 extent_buffer_get(parent);
8621 path->nodes[parent_level] = parent;
8622 path->slots[parent_level] = btrfs_header_nritems(parent);
8624 btrfs_assert_tree_locked(node);
8625 level = btrfs_header_level(node);
8626 path->nodes[level] = node;
8627 path->slots[level] = 0;
8628 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8630 wc->refs[parent_level] = 1;
8631 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8633 wc->shared_level = -1;
8634 wc->stage = DROP_REFERENCE;
8638 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8641 wret = walk_down_tree(trans, root, path, wc);
8647 wret = walk_up_tree(trans, root, path, wc, parent_level);
8655 btrfs_free_path(path);
8659 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8665 * if restripe for this chunk_type is on pick target profile and
8666 * return, otherwise do the usual balance
8668 stripped = get_restripe_target(root->fs_info, flags);
8670 return extended_to_chunk(stripped);
8672 num_devices = root->fs_info->fs_devices->rw_devices;
8674 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8675 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8676 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8678 if (num_devices == 1) {
8679 stripped |= BTRFS_BLOCK_GROUP_DUP;
8680 stripped = flags & ~stripped;
8682 /* turn raid0 into single device chunks */
8683 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8686 /* turn mirroring into duplication */
8687 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8688 BTRFS_BLOCK_GROUP_RAID10))
8689 return stripped | BTRFS_BLOCK_GROUP_DUP;
8691 /* they already had raid on here, just return */
8692 if (flags & stripped)
8695 stripped |= BTRFS_BLOCK_GROUP_DUP;
8696 stripped = flags & ~stripped;
8698 /* switch duplicated blocks with raid1 */
8699 if (flags & BTRFS_BLOCK_GROUP_DUP)
8700 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8702 /* this is drive concat, leave it alone */
8708 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8710 struct btrfs_space_info *sinfo = cache->space_info;
8712 u64 min_allocable_bytes;
8717 * We need some metadata space and system metadata space for
8718 * allocating chunks in some corner cases until we force to set
8719 * it to be readonly.
8722 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8724 min_allocable_bytes = 1 * 1024 * 1024;
8726 min_allocable_bytes = 0;
8728 spin_lock(&sinfo->lock);
8729 spin_lock(&cache->lock);
8736 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8737 cache->bytes_super - btrfs_block_group_used(&cache->item);
8739 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8740 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8741 min_allocable_bytes <= sinfo->total_bytes) {
8742 sinfo->bytes_readonly += num_bytes;
8744 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8748 spin_unlock(&cache->lock);
8749 spin_unlock(&sinfo->lock);
8753 int btrfs_set_block_group_ro(struct btrfs_root *root,
8754 struct btrfs_block_group_cache *cache)
8757 struct btrfs_trans_handle *trans;
8764 trans = btrfs_join_transaction(root);
8766 return PTR_ERR(trans);
8769 * we're not allowed to set block groups readonly after the dirty
8770 * block groups cache has started writing. If it already started,
8771 * back off and let this transaction commit
8773 mutex_lock(&root->fs_info->ro_block_group_mutex);
8774 if (trans->transaction->dirty_bg_run) {
8775 u64 transid = trans->transid;
8777 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8778 btrfs_end_transaction(trans, root);
8780 ret = btrfs_wait_for_commit(root, transid);
8787 * if we are changing raid levels, try to allocate a corresponding
8788 * block group with the new raid level.
8790 alloc_flags = update_block_group_flags(root, cache->flags);
8791 if (alloc_flags != cache->flags) {
8792 ret = do_chunk_alloc(trans, root, alloc_flags,
8795 * ENOSPC is allowed here, we may have enough space
8796 * already allocated at the new raid level to
8805 ret = set_block_group_ro(cache, 0);
8808 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8809 ret = do_chunk_alloc(trans, root, alloc_flags,
8813 ret = set_block_group_ro(cache, 0);
8815 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8816 alloc_flags = update_block_group_flags(root, cache->flags);
8817 lock_chunks(root->fs_info->chunk_root);
8818 check_system_chunk(trans, root, alloc_flags);
8819 unlock_chunks(root->fs_info->chunk_root);
8821 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8823 btrfs_end_transaction(trans, root);
8827 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8828 struct btrfs_root *root, u64 type)
8830 u64 alloc_flags = get_alloc_profile(root, type);
8831 return do_chunk_alloc(trans, root, alloc_flags,
8836 * helper to account the unused space of all the readonly block group in the
8837 * space_info. takes mirrors into account.
8839 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8841 struct btrfs_block_group_cache *block_group;
8845 /* It's df, we don't care if it's racey */
8846 if (list_empty(&sinfo->ro_bgs))
8849 spin_lock(&sinfo->lock);
8850 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8851 spin_lock(&block_group->lock);
8853 if (!block_group->ro) {
8854 spin_unlock(&block_group->lock);
8858 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8859 BTRFS_BLOCK_GROUP_RAID10 |
8860 BTRFS_BLOCK_GROUP_DUP))
8865 free_bytes += (block_group->key.offset -
8866 btrfs_block_group_used(&block_group->item)) *
8869 spin_unlock(&block_group->lock);
8871 spin_unlock(&sinfo->lock);
8876 void btrfs_set_block_group_rw(struct btrfs_root *root,
8877 struct btrfs_block_group_cache *cache)
8879 struct btrfs_space_info *sinfo = cache->space_info;
8884 spin_lock(&sinfo->lock);
8885 spin_lock(&cache->lock);
8886 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8887 cache->bytes_super - btrfs_block_group_used(&cache->item);
8888 sinfo->bytes_readonly -= num_bytes;
8890 list_del_init(&cache->ro_list);
8891 spin_unlock(&cache->lock);
8892 spin_unlock(&sinfo->lock);
8896 * checks to see if its even possible to relocate this block group.
8898 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8899 * ok to go ahead and try.
8901 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8903 struct btrfs_block_group_cache *block_group;
8904 struct btrfs_space_info *space_info;
8905 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8906 struct btrfs_device *device;
8907 struct btrfs_trans_handle *trans;
8916 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8918 /* odd, couldn't find the block group, leave it alone */
8922 min_free = btrfs_block_group_used(&block_group->item);
8924 /* no bytes used, we're good */
8928 space_info = block_group->space_info;
8929 spin_lock(&space_info->lock);
8931 full = space_info->full;
8934 * if this is the last block group we have in this space, we can't
8935 * relocate it unless we're able to allocate a new chunk below.
8937 * Otherwise, we need to make sure we have room in the space to handle
8938 * all of the extents from this block group. If we can, we're good
8940 if ((space_info->total_bytes != block_group->key.offset) &&
8941 (space_info->bytes_used + space_info->bytes_reserved +
8942 space_info->bytes_pinned + space_info->bytes_readonly +
8943 min_free < space_info->total_bytes)) {
8944 spin_unlock(&space_info->lock);
8947 spin_unlock(&space_info->lock);
8950 * ok we don't have enough space, but maybe we have free space on our
8951 * devices to allocate new chunks for relocation, so loop through our
8952 * alloc devices and guess if we have enough space. if this block
8953 * group is going to be restriped, run checks against the target
8954 * profile instead of the current one.
8966 target = get_restripe_target(root->fs_info, block_group->flags);
8968 index = __get_raid_index(extended_to_chunk(target));
8971 * this is just a balance, so if we were marked as full
8972 * we know there is no space for a new chunk
8977 index = get_block_group_index(block_group);
8980 if (index == BTRFS_RAID_RAID10) {
8984 } else if (index == BTRFS_RAID_RAID1) {
8986 } else if (index == BTRFS_RAID_DUP) {
8989 } else if (index == BTRFS_RAID_RAID0) {
8990 dev_min = fs_devices->rw_devices;
8991 min_free = div64_u64(min_free, dev_min);
8994 /* We need to do this so that we can look at pending chunks */
8995 trans = btrfs_join_transaction(root);
8996 if (IS_ERR(trans)) {
8997 ret = PTR_ERR(trans);
9001 mutex_lock(&root->fs_info->chunk_mutex);
9002 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9006 * check to make sure we can actually find a chunk with enough
9007 * space to fit our block group in.
9009 if (device->total_bytes > device->bytes_used + min_free &&
9010 !device->is_tgtdev_for_dev_replace) {
9011 ret = find_free_dev_extent(trans, device, min_free,
9016 if (dev_nr >= dev_min)
9022 mutex_unlock(&root->fs_info->chunk_mutex);
9023 btrfs_end_transaction(trans, root);
9025 btrfs_put_block_group(block_group);
9029 static int find_first_block_group(struct btrfs_root *root,
9030 struct btrfs_path *path, struct btrfs_key *key)
9033 struct btrfs_key found_key;
9034 struct extent_buffer *leaf;
9037 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9042 slot = path->slots[0];
9043 leaf = path->nodes[0];
9044 if (slot >= btrfs_header_nritems(leaf)) {
9045 ret = btrfs_next_leaf(root, path);
9052 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9054 if (found_key.objectid >= key->objectid &&
9055 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9065 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9067 struct btrfs_block_group_cache *block_group;
9071 struct inode *inode;
9073 block_group = btrfs_lookup_first_block_group(info, last);
9074 while (block_group) {
9075 spin_lock(&block_group->lock);
9076 if (block_group->iref)
9078 spin_unlock(&block_group->lock);
9079 block_group = next_block_group(info->tree_root,
9089 inode = block_group->inode;
9090 block_group->iref = 0;
9091 block_group->inode = NULL;
9092 spin_unlock(&block_group->lock);
9094 last = block_group->key.objectid + block_group->key.offset;
9095 btrfs_put_block_group(block_group);
9099 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9101 struct btrfs_block_group_cache *block_group;
9102 struct btrfs_space_info *space_info;
9103 struct btrfs_caching_control *caching_ctl;
9106 down_write(&info->commit_root_sem);
9107 while (!list_empty(&info->caching_block_groups)) {
9108 caching_ctl = list_entry(info->caching_block_groups.next,
9109 struct btrfs_caching_control, list);
9110 list_del(&caching_ctl->list);
9111 put_caching_control(caching_ctl);
9113 up_write(&info->commit_root_sem);
9115 spin_lock(&info->unused_bgs_lock);
9116 while (!list_empty(&info->unused_bgs)) {
9117 block_group = list_first_entry(&info->unused_bgs,
9118 struct btrfs_block_group_cache,
9120 list_del_init(&block_group->bg_list);
9121 btrfs_put_block_group(block_group);
9123 spin_unlock(&info->unused_bgs_lock);
9125 spin_lock(&info->block_group_cache_lock);
9126 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9127 block_group = rb_entry(n, struct btrfs_block_group_cache,
9129 rb_erase(&block_group->cache_node,
9130 &info->block_group_cache_tree);
9131 RB_CLEAR_NODE(&block_group->cache_node);
9132 spin_unlock(&info->block_group_cache_lock);
9134 down_write(&block_group->space_info->groups_sem);
9135 list_del(&block_group->list);
9136 up_write(&block_group->space_info->groups_sem);
9138 if (block_group->cached == BTRFS_CACHE_STARTED)
9139 wait_block_group_cache_done(block_group);
9142 * We haven't cached this block group, which means we could
9143 * possibly have excluded extents on this block group.
9145 if (block_group->cached == BTRFS_CACHE_NO ||
9146 block_group->cached == BTRFS_CACHE_ERROR)
9147 free_excluded_extents(info->extent_root, block_group);
9149 btrfs_remove_free_space_cache(block_group);
9150 btrfs_put_block_group(block_group);
9152 spin_lock(&info->block_group_cache_lock);
9154 spin_unlock(&info->block_group_cache_lock);
9156 /* now that all the block groups are freed, go through and
9157 * free all the space_info structs. This is only called during
9158 * the final stages of unmount, and so we know nobody is
9159 * using them. We call synchronize_rcu() once before we start,
9160 * just to be on the safe side.
9164 release_global_block_rsv(info);
9166 while (!list_empty(&info->space_info)) {
9169 space_info = list_entry(info->space_info.next,
9170 struct btrfs_space_info,
9172 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9173 if (WARN_ON(space_info->bytes_pinned > 0 ||
9174 space_info->bytes_reserved > 0 ||
9175 space_info->bytes_may_use > 0)) {
9176 dump_space_info(space_info, 0, 0);
9179 list_del(&space_info->list);
9180 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9181 struct kobject *kobj;
9182 kobj = space_info->block_group_kobjs[i];
9183 space_info->block_group_kobjs[i] = NULL;
9189 kobject_del(&space_info->kobj);
9190 kobject_put(&space_info->kobj);
9195 static void __link_block_group(struct btrfs_space_info *space_info,
9196 struct btrfs_block_group_cache *cache)
9198 int index = get_block_group_index(cache);
9201 down_write(&space_info->groups_sem);
9202 if (list_empty(&space_info->block_groups[index]))
9204 list_add_tail(&cache->list, &space_info->block_groups[index]);
9205 up_write(&space_info->groups_sem);
9208 struct raid_kobject *rkobj;
9211 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9214 rkobj->raid_type = index;
9215 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9216 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9217 "%s", get_raid_name(index));
9219 kobject_put(&rkobj->kobj);
9222 space_info->block_group_kobjs[index] = &rkobj->kobj;
9227 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9230 static struct btrfs_block_group_cache *
9231 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9233 struct btrfs_block_group_cache *cache;
9235 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9239 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9241 if (!cache->free_space_ctl) {
9246 cache->key.objectid = start;
9247 cache->key.offset = size;
9248 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9250 cache->sectorsize = root->sectorsize;
9251 cache->fs_info = root->fs_info;
9252 cache->full_stripe_len = btrfs_full_stripe_len(root,
9253 &root->fs_info->mapping_tree,
9255 atomic_set(&cache->count, 1);
9256 spin_lock_init(&cache->lock);
9257 init_rwsem(&cache->data_rwsem);
9258 INIT_LIST_HEAD(&cache->list);
9259 INIT_LIST_HEAD(&cache->cluster_list);
9260 INIT_LIST_HEAD(&cache->bg_list);
9261 INIT_LIST_HEAD(&cache->ro_list);
9262 INIT_LIST_HEAD(&cache->dirty_list);
9263 INIT_LIST_HEAD(&cache->io_list);
9264 btrfs_init_free_space_ctl(cache);
9265 atomic_set(&cache->trimming, 0);
9270 int btrfs_read_block_groups(struct btrfs_root *root)
9272 struct btrfs_path *path;
9274 struct btrfs_block_group_cache *cache;
9275 struct btrfs_fs_info *info = root->fs_info;
9276 struct btrfs_space_info *space_info;
9277 struct btrfs_key key;
9278 struct btrfs_key found_key;
9279 struct extent_buffer *leaf;
9283 root = info->extent_root;
9286 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9287 path = btrfs_alloc_path();
9292 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9293 if (btrfs_test_opt(root, SPACE_CACHE) &&
9294 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9296 if (btrfs_test_opt(root, CLEAR_CACHE))
9300 ret = find_first_block_group(root, path, &key);
9306 leaf = path->nodes[0];
9307 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9309 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9318 * When we mount with old space cache, we need to
9319 * set BTRFS_DC_CLEAR and set dirty flag.
9321 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9322 * truncate the old free space cache inode and
9324 * b) Setting 'dirty flag' makes sure that we flush
9325 * the new space cache info onto disk.
9327 if (btrfs_test_opt(root, SPACE_CACHE))
9328 cache->disk_cache_state = BTRFS_DC_CLEAR;
9331 read_extent_buffer(leaf, &cache->item,
9332 btrfs_item_ptr_offset(leaf, path->slots[0]),
9333 sizeof(cache->item));
9334 cache->flags = btrfs_block_group_flags(&cache->item);
9336 key.objectid = found_key.objectid + found_key.offset;
9337 btrfs_release_path(path);
9340 * We need to exclude the super stripes now so that the space
9341 * info has super bytes accounted for, otherwise we'll think
9342 * we have more space than we actually do.
9344 ret = exclude_super_stripes(root, cache);
9347 * We may have excluded something, so call this just in
9350 free_excluded_extents(root, cache);
9351 btrfs_put_block_group(cache);
9356 * check for two cases, either we are full, and therefore
9357 * don't need to bother with the caching work since we won't
9358 * find any space, or we are empty, and we can just add all
9359 * the space in and be done with it. This saves us _alot_ of
9360 * time, particularly in the full case.
9362 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9363 cache->last_byte_to_unpin = (u64)-1;
9364 cache->cached = BTRFS_CACHE_FINISHED;
9365 free_excluded_extents(root, cache);
9366 } else if (btrfs_block_group_used(&cache->item) == 0) {
9367 cache->last_byte_to_unpin = (u64)-1;
9368 cache->cached = BTRFS_CACHE_FINISHED;
9369 add_new_free_space(cache, root->fs_info,
9371 found_key.objectid +
9373 free_excluded_extents(root, cache);
9376 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9378 btrfs_remove_free_space_cache(cache);
9379 btrfs_put_block_group(cache);
9383 ret = update_space_info(info, cache->flags, found_key.offset,
9384 btrfs_block_group_used(&cache->item),
9387 btrfs_remove_free_space_cache(cache);
9388 spin_lock(&info->block_group_cache_lock);
9389 rb_erase(&cache->cache_node,
9390 &info->block_group_cache_tree);
9391 RB_CLEAR_NODE(&cache->cache_node);
9392 spin_unlock(&info->block_group_cache_lock);
9393 btrfs_put_block_group(cache);
9397 cache->space_info = space_info;
9398 spin_lock(&cache->space_info->lock);
9399 cache->space_info->bytes_readonly += cache->bytes_super;
9400 spin_unlock(&cache->space_info->lock);
9402 __link_block_group(space_info, cache);
9404 set_avail_alloc_bits(root->fs_info, cache->flags);
9405 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9406 set_block_group_ro(cache, 1);
9407 } else if (btrfs_block_group_used(&cache->item) == 0) {
9408 spin_lock(&info->unused_bgs_lock);
9409 /* Should always be true but just in case. */
9410 if (list_empty(&cache->bg_list)) {
9411 btrfs_get_block_group(cache);
9412 list_add_tail(&cache->bg_list,
9415 spin_unlock(&info->unused_bgs_lock);
9419 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9420 if (!(get_alloc_profile(root, space_info->flags) &
9421 (BTRFS_BLOCK_GROUP_RAID10 |
9422 BTRFS_BLOCK_GROUP_RAID1 |
9423 BTRFS_BLOCK_GROUP_RAID5 |
9424 BTRFS_BLOCK_GROUP_RAID6 |
9425 BTRFS_BLOCK_GROUP_DUP)))
9428 * avoid allocating from un-mirrored block group if there are
9429 * mirrored block groups.
9431 list_for_each_entry(cache,
9432 &space_info->block_groups[BTRFS_RAID_RAID0],
9434 set_block_group_ro(cache, 1);
9435 list_for_each_entry(cache,
9436 &space_info->block_groups[BTRFS_RAID_SINGLE],
9438 set_block_group_ro(cache, 1);
9441 init_global_block_rsv(info);
9444 btrfs_free_path(path);
9448 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9449 struct btrfs_root *root)
9451 struct btrfs_block_group_cache *block_group, *tmp;
9452 struct btrfs_root *extent_root = root->fs_info->extent_root;
9453 struct btrfs_block_group_item item;
9454 struct btrfs_key key;
9457 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9461 spin_lock(&block_group->lock);
9462 memcpy(&item, &block_group->item, sizeof(item));
9463 memcpy(&key, &block_group->key, sizeof(key));
9464 spin_unlock(&block_group->lock);
9466 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9469 btrfs_abort_transaction(trans, extent_root, ret);
9470 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9471 key.objectid, key.offset);
9473 btrfs_abort_transaction(trans, extent_root, ret);
9475 list_del_init(&block_group->bg_list);
9479 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9480 struct btrfs_root *root, u64 bytes_used,
9481 u64 type, u64 chunk_objectid, u64 chunk_offset,
9485 struct btrfs_root *extent_root;
9486 struct btrfs_block_group_cache *cache;
9488 extent_root = root->fs_info->extent_root;
9490 btrfs_set_log_full_commit(root->fs_info, trans);
9492 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9496 btrfs_set_block_group_used(&cache->item, bytes_used);
9497 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9498 btrfs_set_block_group_flags(&cache->item, type);
9500 cache->flags = type;
9501 cache->last_byte_to_unpin = (u64)-1;
9502 cache->cached = BTRFS_CACHE_FINISHED;
9503 ret = exclude_super_stripes(root, cache);
9506 * We may have excluded something, so call this just in
9509 free_excluded_extents(root, cache);
9510 btrfs_put_block_group(cache);
9514 add_new_free_space(cache, root->fs_info, chunk_offset,
9515 chunk_offset + size);
9517 free_excluded_extents(root, cache);
9520 * Call to ensure the corresponding space_info object is created and
9521 * assigned to our block group, but don't update its counters just yet.
9522 * We want our bg to be added to the rbtree with its ->space_info set.
9524 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9525 &cache->space_info);
9527 btrfs_remove_free_space_cache(cache);
9528 btrfs_put_block_group(cache);
9532 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9534 btrfs_remove_free_space_cache(cache);
9535 btrfs_put_block_group(cache);
9540 * Now that our block group has its ->space_info set and is inserted in
9541 * the rbtree, update the space info's counters.
9543 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9544 &cache->space_info);
9546 btrfs_remove_free_space_cache(cache);
9547 spin_lock(&root->fs_info->block_group_cache_lock);
9548 rb_erase(&cache->cache_node,
9549 &root->fs_info->block_group_cache_tree);
9550 RB_CLEAR_NODE(&cache->cache_node);
9551 spin_unlock(&root->fs_info->block_group_cache_lock);
9552 btrfs_put_block_group(cache);
9555 update_global_block_rsv(root->fs_info);
9557 spin_lock(&cache->space_info->lock);
9558 cache->space_info->bytes_readonly += cache->bytes_super;
9559 spin_unlock(&cache->space_info->lock);
9561 __link_block_group(cache->space_info, cache);
9563 list_add_tail(&cache->bg_list, &trans->new_bgs);
9565 set_avail_alloc_bits(extent_root->fs_info, type);
9570 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9572 u64 extra_flags = chunk_to_extended(flags) &
9573 BTRFS_EXTENDED_PROFILE_MASK;
9575 write_seqlock(&fs_info->profiles_lock);
9576 if (flags & BTRFS_BLOCK_GROUP_DATA)
9577 fs_info->avail_data_alloc_bits &= ~extra_flags;
9578 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9579 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9580 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9581 fs_info->avail_system_alloc_bits &= ~extra_flags;
9582 write_sequnlock(&fs_info->profiles_lock);
9585 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9586 struct btrfs_root *root, u64 group_start,
9587 struct extent_map *em)
9589 struct btrfs_path *path;
9590 struct btrfs_block_group_cache *block_group;
9591 struct btrfs_free_cluster *cluster;
9592 struct btrfs_root *tree_root = root->fs_info->tree_root;
9593 struct btrfs_key key;
9594 struct inode *inode;
9595 struct kobject *kobj = NULL;
9599 struct btrfs_caching_control *caching_ctl = NULL;
9602 root = root->fs_info->extent_root;
9604 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9605 BUG_ON(!block_group);
9606 BUG_ON(!block_group->ro);
9609 * Free the reserved super bytes from this block group before
9612 free_excluded_extents(root, block_group);
9614 memcpy(&key, &block_group->key, sizeof(key));
9615 index = get_block_group_index(block_group);
9616 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9617 BTRFS_BLOCK_GROUP_RAID1 |
9618 BTRFS_BLOCK_GROUP_RAID10))
9623 /* make sure this block group isn't part of an allocation cluster */
9624 cluster = &root->fs_info->data_alloc_cluster;
9625 spin_lock(&cluster->refill_lock);
9626 btrfs_return_cluster_to_free_space(block_group, cluster);
9627 spin_unlock(&cluster->refill_lock);
9630 * make sure this block group isn't part of a metadata
9631 * allocation cluster
9633 cluster = &root->fs_info->meta_alloc_cluster;
9634 spin_lock(&cluster->refill_lock);
9635 btrfs_return_cluster_to_free_space(block_group, cluster);
9636 spin_unlock(&cluster->refill_lock);
9638 path = btrfs_alloc_path();
9645 * get the inode first so any iput calls done for the io_list
9646 * aren't the final iput (no unlinks allowed now)
9648 inode = lookup_free_space_inode(tree_root, block_group, path);
9650 mutex_lock(&trans->transaction->cache_write_mutex);
9652 * make sure our free spache cache IO is done before remove the
9655 spin_lock(&trans->transaction->dirty_bgs_lock);
9656 if (!list_empty(&block_group->io_list)) {
9657 list_del_init(&block_group->io_list);
9659 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9661 spin_unlock(&trans->transaction->dirty_bgs_lock);
9662 btrfs_wait_cache_io(root, trans, block_group,
9663 &block_group->io_ctl, path,
9664 block_group->key.objectid);
9665 btrfs_put_block_group(block_group);
9666 spin_lock(&trans->transaction->dirty_bgs_lock);
9669 if (!list_empty(&block_group->dirty_list)) {
9670 list_del_init(&block_group->dirty_list);
9671 btrfs_put_block_group(block_group);
9673 spin_unlock(&trans->transaction->dirty_bgs_lock);
9674 mutex_unlock(&trans->transaction->cache_write_mutex);
9676 if (!IS_ERR(inode)) {
9677 ret = btrfs_orphan_add(trans, inode);
9679 btrfs_add_delayed_iput(inode);
9683 /* One for the block groups ref */
9684 spin_lock(&block_group->lock);
9685 if (block_group->iref) {
9686 block_group->iref = 0;
9687 block_group->inode = NULL;
9688 spin_unlock(&block_group->lock);
9691 spin_unlock(&block_group->lock);
9693 /* One for our lookup ref */
9694 btrfs_add_delayed_iput(inode);
9697 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9698 key.offset = block_group->key.objectid;
9701 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9705 btrfs_release_path(path);
9707 ret = btrfs_del_item(trans, tree_root, path);
9710 btrfs_release_path(path);
9713 spin_lock(&root->fs_info->block_group_cache_lock);
9714 rb_erase(&block_group->cache_node,
9715 &root->fs_info->block_group_cache_tree);
9716 RB_CLEAR_NODE(&block_group->cache_node);
9718 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9719 root->fs_info->first_logical_byte = (u64)-1;
9720 spin_unlock(&root->fs_info->block_group_cache_lock);
9722 down_write(&block_group->space_info->groups_sem);
9724 * we must use list_del_init so people can check to see if they
9725 * are still on the list after taking the semaphore
9727 list_del_init(&block_group->list);
9728 if (list_empty(&block_group->space_info->block_groups[index])) {
9729 kobj = block_group->space_info->block_group_kobjs[index];
9730 block_group->space_info->block_group_kobjs[index] = NULL;
9731 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9733 up_write(&block_group->space_info->groups_sem);
9739 if (block_group->has_caching_ctl)
9740 caching_ctl = get_caching_control(block_group);
9741 if (block_group->cached == BTRFS_CACHE_STARTED)
9742 wait_block_group_cache_done(block_group);
9743 if (block_group->has_caching_ctl) {
9744 down_write(&root->fs_info->commit_root_sem);
9746 struct btrfs_caching_control *ctl;
9748 list_for_each_entry(ctl,
9749 &root->fs_info->caching_block_groups, list)
9750 if (ctl->block_group == block_group) {
9752 atomic_inc(&caching_ctl->count);
9757 list_del_init(&caching_ctl->list);
9758 up_write(&root->fs_info->commit_root_sem);
9760 /* Once for the caching bgs list and once for us. */
9761 put_caching_control(caching_ctl);
9762 put_caching_control(caching_ctl);
9766 spin_lock(&trans->transaction->dirty_bgs_lock);
9767 if (!list_empty(&block_group->dirty_list)) {
9770 if (!list_empty(&block_group->io_list)) {
9773 spin_unlock(&trans->transaction->dirty_bgs_lock);
9774 btrfs_remove_free_space_cache(block_group);
9776 spin_lock(&block_group->space_info->lock);
9777 list_del_init(&block_group->ro_list);
9779 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9780 WARN_ON(block_group->space_info->total_bytes
9781 < block_group->key.offset);
9782 WARN_ON(block_group->space_info->bytes_readonly
9783 < block_group->key.offset);
9784 WARN_ON(block_group->space_info->disk_total
9785 < block_group->key.offset * factor);
9787 block_group->space_info->total_bytes -= block_group->key.offset;
9788 block_group->space_info->bytes_readonly -= block_group->key.offset;
9789 block_group->space_info->disk_total -= block_group->key.offset * factor;
9791 spin_unlock(&block_group->space_info->lock);
9793 memcpy(&key, &block_group->key, sizeof(key));
9796 if (!list_empty(&em->list)) {
9797 /* We're in the transaction->pending_chunks list. */
9798 free_extent_map(em);
9800 spin_lock(&block_group->lock);
9801 block_group->removed = 1;
9803 * At this point trimming can't start on this block group, because we
9804 * removed the block group from the tree fs_info->block_group_cache_tree
9805 * so no one can't find it anymore and even if someone already got this
9806 * block group before we removed it from the rbtree, they have already
9807 * incremented block_group->trimming - if they didn't, they won't find
9808 * any free space entries because we already removed them all when we
9809 * called btrfs_remove_free_space_cache().
9811 * And we must not remove the extent map from the fs_info->mapping_tree
9812 * to prevent the same logical address range and physical device space
9813 * ranges from being reused for a new block group. This is because our
9814 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9815 * completely transactionless, so while it is trimming a range the
9816 * currently running transaction might finish and a new one start,
9817 * allowing for new block groups to be created that can reuse the same
9818 * physical device locations unless we take this special care.
9820 remove_em = (atomic_read(&block_group->trimming) == 0);
9822 * Make sure a trimmer task always sees the em in the pinned_chunks list
9823 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9824 * before checking block_group->removed).
9828 * Our em might be in trans->transaction->pending_chunks which
9829 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9830 * and so is the fs_info->pinned_chunks list.
9832 * So at this point we must be holding the chunk_mutex to avoid
9833 * any races with chunk allocation (more specifically at
9834 * volumes.c:contains_pending_extent()), to ensure it always
9835 * sees the em, either in the pending_chunks list or in the
9836 * pinned_chunks list.
9838 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9840 spin_unlock(&block_group->lock);
9843 struct extent_map_tree *em_tree;
9845 em_tree = &root->fs_info->mapping_tree.map_tree;
9846 write_lock(&em_tree->lock);
9848 * The em might be in the pending_chunks list, so make sure the
9849 * chunk mutex is locked, since remove_extent_mapping() will
9850 * delete us from that list.
9852 remove_extent_mapping(em_tree, em);
9853 write_unlock(&em_tree->lock);
9854 /* once for the tree */
9855 free_extent_map(em);
9858 unlock_chunks(root);
9860 btrfs_put_block_group(block_group);
9861 btrfs_put_block_group(block_group);
9863 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9869 ret = btrfs_del_item(trans, root, path);
9871 btrfs_free_path(path);
9876 * Process the unused_bgs list and remove any that don't have any allocated
9877 * space inside of them.
9879 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9881 struct btrfs_block_group_cache *block_group;
9882 struct btrfs_space_info *space_info;
9883 struct btrfs_root *root = fs_info->extent_root;
9884 struct btrfs_trans_handle *trans;
9890 spin_lock(&fs_info->unused_bgs_lock);
9891 while (!list_empty(&fs_info->unused_bgs)) {
9894 block_group = list_first_entry(&fs_info->unused_bgs,
9895 struct btrfs_block_group_cache,
9897 space_info = block_group->space_info;
9898 list_del_init(&block_group->bg_list);
9899 if (ret || btrfs_mixed_space_info(space_info)) {
9900 btrfs_put_block_group(block_group);
9903 spin_unlock(&fs_info->unused_bgs_lock);
9905 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
9907 /* Don't want to race with allocators so take the groups_sem */
9908 down_write(&space_info->groups_sem);
9909 spin_lock(&block_group->lock);
9910 if (block_group->reserved ||
9911 btrfs_block_group_used(&block_group->item) ||
9914 * We want to bail if we made new allocations or have
9915 * outstanding allocations in this block group. We do
9916 * the ro check in case balance is currently acting on
9919 spin_unlock(&block_group->lock);
9920 up_write(&space_info->groups_sem);
9923 spin_unlock(&block_group->lock);
9925 /* We don't want to force the issue, only flip if it's ok. */
9926 ret = set_block_group_ro(block_group, 0);
9927 up_write(&space_info->groups_sem);
9934 * Want to do this before we do anything else so we can recover
9935 * properly if we fail to join the transaction.
9937 /* 1 for btrfs_orphan_reserve_metadata() */
9938 trans = btrfs_start_transaction(root, 1);
9939 if (IS_ERR(trans)) {
9940 btrfs_set_block_group_rw(root, block_group);
9941 ret = PTR_ERR(trans);
9946 * We could have pending pinned extents for this block group,
9947 * just delete them, we don't care about them anymore.
9949 start = block_group->key.objectid;
9950 end = start + block_group->key.offset - 1;
9952 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9953 * btrfs_finish_extent_commit(). If we are at transaction N,
9954 * another task might be running finish_extent_commit() for the
9955 * previous transaction N - 1, and have seen a range belonging
9956 * to the block group in freed_extents[] before we were able to
9957 * clear the whole block group range from freed_extents[]. This
9958 * means that task can lookup for the block group after we
9959 * unpinned it from freed_extents[] and removed it, leading to
9960 * a BUG_ON() at btrfs_unpin_extent_range().
9962 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9963 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9964 EXTENT_DIRTY, GFP_NOFS);
9966 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9967 btrfs_set_block_group_rw(root, block_group);
9970 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9971 EXTENT_DIRTY, GFP_NOFS);
9973 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9974 btrfs_set_block_group_rw(root, block_group);
9977 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9979 /* Reset pinned so btrfs_put_block_group doesn't complain */
9980 spin_lock(&space_info->lock);
9981 spin_lock(&block_group->lock);
9983 space_info->bytes_pinned -= block_group->pinned;
9984 space_info->bytes_readonly += block_group->pinned;
9985 percpu_counter_add(&space_info->total_bytes_pinned,
9986 -block_group->pinned);
9987 block_group->pinned = 0;
9989 spin_unlock(&block_group->lock);
9990 spin_unlock(&space_info->lock);
9993 * Btrfs_remove_chunk will abort the transaction if things go
9996 ret = btrfs_remove_chunk(trans, root,
9997 block_group->key.objectid);
9999 btrfs_end_transaction(trans, root);
10001 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10002 btrfs_put_block_group(block_group);
10003 spin_lock(&fs_info->unused_bgs_lock);
10005 spin_unlock(&fs_info->unused_bgs_lock);
10008 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10010 struct btrfs_space_info *space_info;
10011 struct btrfs_super_block *disk_super;
10017 disk_super = fs_info->super_copy;
10018 if (!btrfs_super_root(disk_super))
10021 features = btrfs_super_incompat_flags(disk_super);
10022 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10025 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10026 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10031 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10032 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10034 flags = BTRFS_BLOCK_GROUP_METADATA;
10035 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10039 flags = BTRFS_BLOCK_GROUP_DATA;
10040 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10046 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10048 return unpin_extent_range(root, start, end, false);
10051 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10053 struct btrfs_fs_info *fs_info = root->fs_info;
10054 struct btrfs_block_group_cache *cache = NULL;
10059 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10063 * try to trim all FS space, our block group may start from non-zero.
10065 if (range->len == total_bytes)
10066 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10068 cache = btrfs_lookup_block_group(fs_info, range->start);
10071 if (cache->key.objectid >= (range->start + range->len)) {
10072 btrfs_put_block_group(cache);
10076 start = max(range->start, cache->key.objectid);
10077 end = min(range->start + range->len,
10078 cache->key.objectid + cache->key.offset);
10080 if (end - start >= range->minlen) {
10081 if (!block_group_cache_done(cache)) {
10082 ret = cache_block_group(cache, 0);
10084 btrfs_put_block_group(cache);
10087 ret = wait_block_group_cache_done(cache);
10089 btrfs_put_block_group(cache);
10093 ret = btrfs_trim_block_group(cache,
10099 trimmed += group_trimmed;
10101 btrfs_put_block_group(cache);
10106 cache = next_block_group(fs_info->tree_root, cache);
10109 range->len = trimmed;
10114 * btrfs_{start,end}_write_no_snapshoting() are similar to
10115 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10116 * data into the page cache through nocow before the subvolume is snapshoted,
10117 * but flush the data into disk after the snapshot creation, or to prevent
10118 * operations while snapshoting is ongoing and that cause the snapshot to be
10119 * inconsistent (writes followed by expanding truncates for example).
10121 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10123 percpu_counter_dec(&root->subv_writers->counter);
10125 * Make sure counter is updated before we wake up
10129 if (waitqueue_active(&root->subv_writers->wait))
10130 wake_up(&root->subv_writers->wait);
10133 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10135 if (atomic_read(&root->will_be_snapshoted))
10138 percpu_counter_inc(&root->subv_writers->counter);
10140 * Make sure counter is updated before we check for snapshot creation.
10143 if (atomic_read(&root->will_be_snapshoted)) {
10144 btrfs_end_write_no_snapshoting(root);