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 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1887 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1888 u64 *discarded_bytes)
1891 u64 bytes_left, end;
1892 u64 aligned_start = ALIGN(start, 1 << 9);
1894 if (WARN_ON(start != aligned_start)) {
1895 len -= aligned_start - start;
1896 len = round_down(len, 1 << 9);
1897 start = aligned_start;
1900 *discarded_bytes = 0;
1908 /* Skip any superblocks on this device. */
1909 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1910 u64 sb_start = btrfs_sb_offset(j);
1911 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1912 u64 size = sb_start - start;
1914 if (!in_range(sb_start, start, bytes_left) &&
1915 !in_range(sb_end, start, bytes_left) &&
1916 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1920 * Superblock spans beginning of range. Adjust start and
1923 if (sb_start <= start) {
1924 start += sb_end - start;
1929 bytes_left = end - start;
1934 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1937 *discarded_bytes += size;
1938 else if (ret != -EOPNOTSUPP)
1947 bytes_left = end - start;
1951 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1954 *discarded_bytes += bytes_left;
1959 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1960 u64 num_bytes, u64 *actual_bytes)
1963 u64 discarded_bytes = 0;
1964 struct btrfs_bio *bbio = NULL;
1967 /* Tell the block device(s) that the sectors can be discarded */
1968 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1969 bytenr, &num_bytes, &bbio, 0);
1970 /* Error condition is -ENOMEM */
1972 struct btrfs_bio_stripe *stripe = bbio->stripes;
1976 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1978 if (!stripe->dev->can_discard)
1981 ret = btrfs_issue_discard(stripe->dev->bdev,
1986 discarded_bytes += bytes;
1987 else if (ret != -EOPNOTSUPP)
1988 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1991 * Just in case we get back EOPNOTSUPP for some reason,
1992 * just ignore the return value so we don't screw up
1993 * people calling discard_extent.
1997 btrfs_put_bbio(bbio);
2001 *actual_bytes = discarded_bytes;
2004 if (ret == -EOPNOTSUPP)
2009 /* Can return -ENOMEM */
2010 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2011 struct btrfs_root *root,
2012 u64 bytenr, u64 num_bytes, u64 parent,
2013 u64 root_objectid, u64 owner, u64 offset,
2017 struct btrfs_fs_info *fs_info = root->fs_info;
2019 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2020 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2022 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2023 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2025 parent, root_objectid, (int)owner,
2026 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2028 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2030 parent, root_objectid, owner, offset,
2031 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2036 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2037 struct btrfs_root *root,
2038 struct btrfs_delayed_ref_node *node,
2039 u64 parent, u64 root_objectid,
2040 u64 owner, u64 offset, int refs_to_add,
2041 struct btrfs_delayed_extent_op *extent_op)
2043 struct btrfs_fs_info *fs_info = root->fs_info;
2044 struct btrfs_path *path;
2045 struct extent_buffer *leaf;
2046 struct btrfs_extent_item *item;
2047 struct btrfs_key key;
2048 u64 bytenr = node->bytenr;
2049 u64 num_bytes = node->num_bytes;
2052 int no_quota = node->no_quota;
2054 path = btrfs_alloc_path();
2058 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2062 path->leave_spinning = 1;
2063 /* this will setup the path even if it fails to insert the back ref */
2064 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2065 bytenr, num_bytes, parent,
2066 root_objectid, owner, offset,
2067 refs_to_add, extent_op);
2068 if ((ret < 0 && ret != -EAGAIN) || !ret)
2072 * Ok we had -EAGAIN which means we didn't have space to insert and
2073 * inline extent ref, so just update the reference count and add a
2076 leaf = path->nodes[0];
2077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2078 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2079 refs = btrfs_extent_refs(leaf, item);
2080 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2082 __run_delayed_extent_op(extent_op, leaf, item);
2084 btrfs_mark_buffer_dirty(leaf);
2085 btrfs_release_path(path);
2088 path->leave_spinning = 1;
2089 /* now insert the actual backref */
2090 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2091 path, bytenr, parent, root_objectid,
2092 owner, offset, refs_to_add);
2094 btrfs_abort_transaction(trans, root, ret);
2096 btrfs_free_path(path);
2100 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2101 struct btrfs_root *root,
2102 struct btrfs_delayed_ref_node *node,
2103 struct btrfs_delayed_extent_op *extent_op,
2104 int insert_reserved)
2107 struct btrfs_delayed_data_ref *ref;
2108 struct btrfs_key ins;
2113 ins.objectid = node->bytenr;
2114 ins.offset = node->num_bytes;
2115 ins.type = BTRFS_EXTENT_ITEM_KEY;
2117 ref = btrfs_delayed_node_to_data_ref(node);
2118 trace_run_delayed_data_ref(node, ref, node->action);
2120 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2121 parent = ref->parent;
2122 ref_root = ref->root;
2124 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2126 flags |= extent_op->flags_to_set;
2127 ret = alloc_reserved_file_extent(trans, root,
2128 parent, ref_root, flags,
2129 ref->objectid, ref->offset,
2130 &ins, node->ref_mod);
2131 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2132 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2133 ref_root, ref->objectid,
2134 ref->offset, node->ref_mod,
2136 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2137 ret = __btrfs_free_extent(trans, root, node, parent,
2138 ref_root, ref->objectid,
2139 ref->offset, node->ref_mod,
2147 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2148 struct extent_buffer *leaf,
2149 struct btrfs_extent_item *ei)
2151 u64 flags = btrfs_extent_flags(leaf, ei);
2152 if (extent_op->update_flags) {
2153 flags |= extent_op->flags_to_set;
2154 btrfs_set_extent_flags(leaf, ei, flags);
2157 if (extent_op->update_key) {
2158 struct btrfs_tree_block_info *bi;
2159 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2160 bi = (struct btrfs_tree_block_info *)(ei + 1);
2161 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2165 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2166 struct btrfs_root *root,
2167 struct btrfs_delayed_ref_node *node,
2168 struct btrfs_delayed_extent_op *extent_op)
2170 struct btrfs_key key;
2171 struct btrfs_path *path;
2172 struct btrfs_extent_item *ei;
2173 struct extent_buffer *leaf;
2177 int metadata = !extent_op->is_data;
2182 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2185 path = btrfs_alloc_path();
2189 key.objectid = node->bytenr;
2192 key.type = BTRFS_METADATA_ITEM_KEY;
2193 key.offset = extent_op->level;
2195 key.type = BTRFS_EXTENT_ITEM_KEY;
2196 key.offset = node->num_bytes;
2201 path->leave_spinning = 1;
2202 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2210 if (path->slots[0] > 0) {
2212 btrfs_item_key_to_cpu(path->nodes[0], &key,
2214 if (key.objectid == node->bytenr &&
2215 key.type == BTRFS_EXTENT_ITEM_KEY &&
2216 key.offset == node->num_bytes)
2220 btrfs_release_path(path);
2223 key.objectid = node->bytenr;
2224 key.offset = node->num_bytes;
2225 key.type = BTRFS_EXTENT_ITEM_KEY;
2234 leaf = path->nodes[0];
2235 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2236 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2237 if (item_size < sizeof(*ei)) {
2238 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2244 leaf = path->nodes[0];
2245 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2248 BUG_ON(item_size < sizeof(*ei));
2249 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2250 __run_delayed_extent_op(extent_op, leaf, ei);
2252 btrfs_mark_buffer_dirty(leaf);
2254 btrfs_free_path(path);
2258 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2259 struct btrfs_root *root,
2260 struct btrfs_delayed_ref_node *node,
2261 struct btrfs_delayed_extent_op *extent_op,
2262 int insert_reserved)
2265 struct btrfs_delayed_tree_ref *ref;
2266 struct btrfs_key ins;
2269 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2272 ref = btrfs_delayed_node_to_tree_ref(node);
2273 trace_run_delayed_tree_ref(node, ref, node->action);
2275 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2276 parent = ref->parent;
2277 ref_root = ref->root;
2279 ins.objectid = node->bytenr;
2280 if (skinny_metadata) {
2281 ins.offset = ref->level;
2282 ins.type = BTRFS_METADATA_ITEM_KEY;
2284 ins.offset = node->num_bytes;
2285 ins.type = BTRFS_EXTENT_ITEM_KEY;
2288 BUG_ON(node->ref_mod != 1);
2289 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2290 BUG_ON(!extent_op || !extent_op->update_flags);
2291 ret = alloc_reserved_tree_block(trans, root,
2293 extent_op->flags_to_set,
2297 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2298 ret = __btrfs_inc_extent_ref(trans, root, node,
2302 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2303 ret = __btrfs_free_extent(trans, root, node,
2305 ref->level, 0, 1, extent_op);
2312 /* helper function to actually process a single delayed ref entry */
2313 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2314 struct btrfs_root *root,
2315 struct btrfs_delayed_ref_node *node,
2316 struct btrfs_delayed_extent_op *extent_op,
2317 int insert_reserved)
2321 if (trans->aborted) {
2322 if (insert_reserved)
2323 btrfs_pin_extent(root, node->bytenr,
2324 node->num_bytes, 1);
2328 if (btrfs_delayed_ref_is_head(node)) {
2329 struct btrfs_delayed_ref_head *head;
2331 * we've hit the end of the chain and we were supposed
2332 * to insert this extent into the tree. But, it got
2333 * deleted before we ever needed to insert it, so all
2334 * we have to do is clean up the accounting
2337 head = btrfs_delayed_node_to_head(node);
2338 trace_run_delayed_ref_head(node, head, node->action);
2340 if (insert_reserved) {
2341 btrfs_pin_extent(root, node->bytenr,
2342 node->num_bytes, 1);
2343 if (head->is_data) {
2344 ret = btrfs_del_csums(trans, root,
2352 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2353 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2354 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2356 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2357 node->type == BTRFS_SHARED_DATA_REF_KEY)
2358 ret = run_delayed_data_ref(trans, root, node, extent_op,
2365 static inline struct btrfs_delayed_ref_node *
2366 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2368 struct btrfs_delayed_ref_node *ref;
2370 if (list_empty(&head->ref_list))
2374 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2375 * This is to prevent a ref count from going down to zero, which deletes
2376 * the extent item from the extent tree, when there still are references
2377 * to add, which would fail because they would not find the extent item.
2379 list_for_each_entry(ref, &head->ref_list, list) {
2380 if (ref->action == BTRFS_ADD_DELAYED_REF)
2384 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2389 * Returns 0 on success or if called with an already aborted transaction.
2390 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2392 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2393 struct btrfs_root *root,
2396 struct btrfs_delayed_ref_root *delayed_refs;
2397 struct btrfs_delayed_ref_node *ref;
2398 struct btrfs_delayed_ref_head *locked_ref = NULL;
2399 struct btrfs_delayed_extent_op *extent_op;
2400 struct btrfs_fs_info *fs_info = root->fs_info;
2401 ktime_t start = ktime_get();
2403 unsigned long count = 0;
2404 unsigned long actual_count = 0;
2405 int must_insert_reserved = 0;
2407 delayed_refs = &trans->transaction->delayed_refs;
2413 spin_lock(&delayed_refs->lock);
2414 locked_ref = btrfs_select_ref_head(trans);
2416 spin_unlock(&delayed_refs->lock);
2420 /* grab the lock that says we are going to process
2421 * all the refs for this head */
2422 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2423 spin_unlock(&delayed_refs->lock);
2425 * we may have dropped the spin lock to get the head
2426 * mutex lock, and that might have given someone else
2427 * time to free the head. If that's true, it has been
2428 * removed from our list and we can move on.
2430 if (ret == -EAGAIN) {
2437 spin_lock(&locked_ref->lock);
2440 * locked_ref is the head node, so we have to go one
2441 * node back for any delayed ref updates
2443 ref = select_delayed_ref(locked_ref);
2445 if (ref && ref->seq &&
2446 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2447 spin_unlock(&locked_ref->lock);
2448 btrfs_delayed_ref_unlock(locked_ref);
2449 spin_lock(&delayed_refs->lock);
2450 locked_ref->processing = 0;
2451 delayed_refs->num_heads_ready++;
2452 spin_unlock(&delayed_refs->lock);
2460 * record the must insert reserved flag before we
2461 * drop the spin lock.
2463 must_insert_reserved = locked_ref->must_insert_reserved;
2464 locked_ref->must_insert_reserved = 0;
2466 extent_op = locked_ref->extent_op;
2467 locked_ref->extent_op = NULL;
2472 /* All delayed refs have been processed, Go ahead
2473 * and send the head node to run_one_delayed_ref,
2474 * so that any accounting fixes can happen
2476 ref = &locked_ref->node;
2478 if (extent_op && must_insert_reserved) {
2479 btrfs_free_delayed_extent_op(extent_op);
2484 spin_unlock(&locked_ref->lock);
2485 ret = run_delayed_extent_op(trans, root,
2487 btrfs_free_delayed_extent_op(extent_op);
2491 * Need to reset must_insert_reserved if
2492 * there was an error so the abort stuff
2493 * can cleanup the reserved space
2496 if (must_insert_reserved)
2497 locked_ref->must_insert_reserved = 1;
2498 locked_ref->processing = 0;
2499 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2500 btrfs_delayed_ref_unlock(locked_ref);
2507 * Need to drop our head ref lock and re-aqcuire the
2508 * delayed ref lock and then re-check to make sure
2511 spin_unlock(&locked_ref->lock);
2512 spin_lock(&delayed_refs->lock);
2513 spin_lock(&locked_ref->lock);
2514 if (!list_empty(&locked_ref->ref_list) ||
2515 locked_ref->extent_op) {
2516 spin_unlock(&locked_ref->lock);
2517 spin_unlock(&delayed_refs->lock);
2521 delayed_refs->num_heads--;
2522 rb_erase(&locked_ref->href_node,
2523 &delayed_refs->href_root);
2524 spin_unlock(&delayed_refs->lock);
2528 list_del(&ref->list);
2530 atomic_dec(&delayed_refs->num_entries);
2532 if (!btrfs_delayed_ref_is_head(ref)) {
2534 * when we play the delayed ref, also correct the
2537 switch (ref->action) {
2538 case BTRFS_ADD_DELAYED_REF:
2539 case BTRFS_ADD_DELAYED_EXTENT:
2540 locked_ref->node.ref_mod -= ref->ref_mod;
2542 case BTRFS_DROP_DELAYED_REF:
2543 locked_ref->node.ref_mod += ref->ref_mod;
2549 spin_unlock(&locked_ref->lock);
2551 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2552 must_insert_reserved);
2554 btrfs_free_delayed_extent_op(extent_op);
2556 locked_ref->processing = 0;
2557 btrfs_delayed_ref_unlock(locked_ref);
2558 btrfs_put_delayed_ref(ref);
2559 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2564 * If this node is a head, that means all the refs in this head
2565 * have been dealt with, and we will pick the next head to deal
2566 * with, so we must unlock the head and drop it from the cluster
2567 * list before we release it.
2569 if (btrfs_delayed_ref_is_head(ref)) {
2570 if (locked_ref->is_data &&
2571 locked_ref->total_ref_mod < 0) {
2572 spin_lock(&delayed_refs->lock);
2573 delayed_refs->pending_csums -= ref->num_bytes;
2574 spin_unlock(&delayed_refs->lock);
2576 btrfs_delayed_ref_unlock(locked_ref);
2579 btrfs_put_delayed_ref(ref);
2585 * We don't want to include ref heads since we can have empty ref heads
2586 * and those will drastically skew our runtime down since we just do
2587 * accounting, no actual extent tree updates.
2589 if (actual_count > 0) {
2590 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2594 * We weigh the current average higher than our current runtime
2595 * to avoid large swings in the average.
2597 spin_lock(&delayed_refs->lock);
2598 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2599 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2600 spin_unlock(&delayed_refs->lock);
2605 #ifdef SCRAMBLE_DELAYED_REFS
2607 * Normally delayed refs get processed in ascending bytenr order. This
2608 * correlates in most cases to the order added. To expose dependencies on this
2609 * order, we start to process the tree in the middle instead of the beginning
2611 static u64 find_middle(struct rb_root *root)
2613 struct rb_node *n = root->rb_node;
2614 struct btrfs_delayed_ref_node *entry;
2617 u64 first = 0, last = 0;
2621 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2622 first = entry->bytenr;
2626 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2627 last = entry->bytenr;
2632 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2633 WARN_ON(!entry->in_tree);
2635 middle = entry->bytenr;
2648 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2652 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2653 sizeof(struct btrfs_extent_inline_ref));
2654 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2655 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2658 * We don't ever fill up leaves all the way so multiply by 2 just to be
2659 * closer to what we're really going to want to ouse.
2661 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2665 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2666 * would require to store the csums for that many bytes.
2668 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2671 u64 num_csums_per_leaf;
2674 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2675 num_csums_per_leaf = div64_u64(csum_size,
2676 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2677 num_csums = div64_u64(csum_bytes, root->sectorsize);
2678 num_csums += num_csums_per_leaf - 1;
2679 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2683 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root)
2686 struct btrfs_block_rsv *global_rsv;
2687 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2688 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2689 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2690 u64 num_bytes, num_dirty_bgs_bytes;
2693 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2694 num_heads = heads_to_leaves(root, num_heads);
2696 num_bytes += (num_heads - 1) * root->nodesize;
2698 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2699 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2701 global_rsv = &root->fs_info->global_block_rsv;
2704 * If we can't allocate any more chunks lets make sure we have _lots_ of
2705 * wiggle room since running delayed refs can create more delayed refs.
2707 if (global_rsv->space_info->full) {
2708 num_dirty_bgs_bytes <<= 1;
2712 spin_lock(&global_rsv->lock);
2713 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2715 spin_unlock(&global_rsv->lock);
2719 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2720 struct btrfs_root *root)
2722 struct btrfs_fs_info *fs_info = root->fs_info;
2724 atomic_read(&trans->transaction->delayed_refs.num_entries);
2729 avg_runtime = fs_info->avg_delayed_ref_runtime;
2730 val = num_entries * avg_runtime;
2731 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2733 if (val >= NSEC_PER_SEC / 2)
2736 return btrfs_check_space_for_delayed_refs(trans, root);
2739 struct async_delayed_refs {
2740 struct btrfs_root *root;
2744 struct completion wait;
2745 struct btrfs_work work;
2748 static void delayed_ref_async_start(struct btrfs_work *work)
2750 struct async_delayed_refs *async;
2751 struct btrfs_trans_handle *trans;
2754 async = container_of(work, struct async_delayed_refs, work);
2756 trans = btrfs_join_transaction(async->root);
2757 if (IS_ERR(trans)) {
2758 async->error = PTR_ERR(trans);
2763 * trans->sync means that when we call end_transaciton, we won't
2764 * wait on delayed refs
2767 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2771 ret = btrfs_end_transaction(trans, async->root);
2772 if (ret && !async->error)
2776 complete(&async->wait);
2781 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2782 unsigned long count, int wait)
2784 struct async_delayed_refs *async;
2787 async = kmalloc(sizeof(*async), GFP_NOFS);
2791 async->root = root->fs_info->tree_root;
2792 async->count = count;
2798 init_completion(&async->wait);
2800 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2801 delayed_ref_async_start, NULL, NULL);
2803 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2806 wait_for_completion(&async->wait);
2815 * this starts processing the delayed reference count updates and
2816 * extent insertions we have queued up so far. count can be
2817 * 0, which means to process everything in the tree at the start
2818 * of the run (but not newly added entries), or it can be some target
2819 * number you'd like to process.
2821 * Returns 0 on success or if called with an aborted transaction
2822 * Returns <0 on error and aborts the transaction
2824 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2825 struct btrfs_root *root, unsigned long count)
2827 struct rb_node *node;
2828 struct btrfs_delayed_ref_root *delayed_refs;
2829 struct btrfs_delayed_ref_head *head;
2831 int run_all = count == (unsigned long)-1;
2833 /* We'll clean this up in btrfs_cleanup_transaction */
2837 if (root == root->fs_info->extent_root)
2838 root = root->fs_info->tree_root;
2840 delayed_refs = &trans->transaction->delayed_refs;
2842 count = atomic_read(&delayed_refs->num_entries) * 2;
2845 #ifdef SCRAMBLE_DELAYED_REFS
2846 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2848 ret = __btrfs_run_delayed_refs(trans, root, count);
2850 btrfs_abort_transaction(trans, root, ret);
2855 if (!list_empty(&trans->new_bgs))
2856 btrfs_create_pending_block_groups(trans, root);
2858 spin_lock(&delayed_refs->lock);
2859 node = rb_first(&delayed_refs->href_root);
2861 spin_unlock(&delayed_refs->lock);
2864 count = (unsigned long)-1;
2867 head = rb_entry(node, struct btrfs_delayed_ref_head,
2869 if (btrfs_delayed_ref_is_head(&head->node)) {
2870 struct btrfs_delayed_ref_node *ref;
2873 atomic_inc(&ref->refs);
2875 spin_unlock(&delayed_refs->lock);
2877 * Mutex was contended, block until it's
2878 * released and try again
2880 mutex_lock(&head->mutex);
2881 mutex_unlock(&head->mutex);
2883 btrfs_put_delayed_ref(ref);
2889 node = rb_next(node);
2891 spin_unlock(&delayed_refs->lock);
2896 assert_qgroups_uptodate(trans);
2900 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2901 struct btrfs_root *root,
2902 u64 bytenr, u64 num_bytes, u64 flags,
2903 int level, int is_data)
2905 struct btrfs_delayed_extent_op *extent_op;
2908 extent_op = btrfs_alloc_delayed_extent_op();
2912 extent_op->flags_to_set = flags;
2913 extent_op->update_flags = 1;
2914 extent_op->update_key = 0;
2915 extent_op->is_data = is_data ? 1 : 0;
2916 extent_op->level = level;
2918 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2919 num_bytes, extent_op);
2921 btrfs_free_delayed_extent_op(extent_op);
2925 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2926 struct btrfs_root *root,
2927 struct btrfs_path *path,
2928 u64 objectid, u64 offset, u64 bytenr)
2930 struct btrfs_delayed_ref_head *head;
2931 struct btrfs_delayed_ref_node *ref;
2932 struct btrfs_delayed_data_ref *data_ref;
2933 struct btrfs_delayed_ref_root *delayed_refs;
2936 delayed_refs = &trans->transaction->delayed_refs;
2937 spin_lock(&delayed_refs->lock);
2938 head = btrfs_find_delayed_ref_head(trans, bytenr);
2940 spin_unlock(&delayed_refs->lock);
2944 if (!mutex_trylock(&head->mutex)) {
2945 atomic_inc(&head->node.refs);
2946 spin_unlock(&delayed_refs->lock);
2948 btrfs_release_path(path);
2951 * Mutex was contended, block until it's released and let
2954 mutex_lock(&head->mutex);
2955 mutex_unlock(&head->mutex);
2956 btrfs_put_delayed_ref(&head->node);
2959 spin_unlock(&delayed_refs->lock);
2961 spin_lock(&head->lock);
2962 list_for_each_entry(ref, &head->ref_list, list) {
2963 /* If it's a shared ref we know a cross reference exists */
2964 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2969 data_ref = btrfs_delayed_node_to_data_ref(ref);
2972 * If our ref doesn't match the one we're currently looking at
2973 * then we have a cross reference.
2975 if (data_ref->root != root->root_key.objectid ||
2976 data_ref->objectid != objectid ||
2977 data_ref->offset != offset) {
2982 spin_unlock(&head->lock);
2983 mutex_unlock(&head->mutex);
2987 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2988 struct btrfs_root *root,
2989 struct btrfs_path *path,
2990 u64 objectid, u64 offset, u64 bytenr)
2992 struct btrfs_root *extent_root = root->fs_info->extent_root;
2993 struct extent_buffer *leaf;
2994 struct btrfs_extent_data_ref *ref;
2995 struct btrfs_extent_inline_ref *iref;
2996 struct btrfs_extent_item *ei;
2997 struct btrfs_key key;
3001 key.objectid = bytenr;
3002 key.offset = (u64)-1;
3003 key.type = BTRFS_EXTENT_ITEM_KEY;
3005 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3008 BUG_ON(ret == 0); /* Corruption */
3011 if (path->slots[0] == 0)
3015 leaf = path->nodes[0];
3016 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3018 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3022 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3023 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3024 if (item_size < sizeof(*ei)) {
3025 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3029 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3031 if (item_size != sizeof(*ei) +
3032 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3035 if (btrfs_extent_generation(leaf, ei) <=
3036 btrfs_root_last_snapshot(&root->root_item))
3039 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3040 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3041 BTRFS_EXTENT_DATA_REF_KEY)
3044 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3045 if (btrfs_extent_refs(leaf, ei) !=
3046 btrfs_extent_data_ref_count(leaf, ref) ||
3047 btrfs_extent_data_ref_root(leaf, ref) !=
3048 root->root_key.objectid ||
3049 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3050 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3058 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3059 struct btrfs_root *root,
3060 u64 objectid, u64 offset, u64 bytenr)
3062 struct btrfs_path *path;
3066 path = btrfs_alloc_path();
3071 ret = check_committed_ref(trans, root, path, objectid,
3073 if (ret && ret != -ENOENT)
3076 ret2 = check_delayed_ref(trans, root, path, objectid,
3078 } while (ret2 == -EAGAIN);
3080 if (ret2 && ret2 != -ENOENT) {
3085 if (ret != -ENOENT || ret2 != -ENOENT)
3088 btrfs_free_path(path);
3089 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3094 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3095 struct btrfs_root *root,
3096 struct extent_buffer *buf,
3097 int full_backref, int inc)
3104 struct btrfs_key key;
3105 struct btrfs_file_extent_item *fi;
3109 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3110 u64, u64, u64, u64, u64, u64, int);
3113 if (btrfs_test_is_dummy_root(root))
3116 ref_root = btrfs_header_owner(buf);
3117 nritems = btrfs_header_nritems(buf);
3118 level = btrfs_header_level(buf);
3120 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3124 process_func = btrfs_inc_extent_ref;
3126 process_func = btrfs_free_extent;
3129 parent = buf->start;
3133 for (i = 0; i < nritems; i++) {
3135 btrfs_item_key_to_cpu(buf, &key, i);
3136 if (key.type != BTRFS_EXTENT_DATA_KEY)
3138 fi = btrfs_item_ptr(buf, i,
3139 struct btrfs_file_extent_item);
3140 if (btrfs_file_extent_type(buf, fi) ==
3141 BTRFS_FILE_EXTENT_INLINE)
3143 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3147 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3148 key.offset -= btrfs_file_extent_offset(buf, fi);
3149 ret = process_func(trans, root, bytenr, num_bytes,
3150 parent, ref_root, key.objectid,
3155 bytenr = btrfs_node_blockptr(buf, i);
3156 num_bytes = root->nodesize;
3157 ret = process_func(trans, root, bytenr, num_bytes,
3158 parent, ref_root, level - 1, 0,
3169 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3170 struct extent_buffer *buf, int full_backref)
3172 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3175 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3176 struct extent_buffer *buf, int full_backref)
3178 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3181 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3182 struct btrfs_root *root,
3183 struct btrfs_path *path,
3184 struct btrfs_block_group_cache *cache)
3187 struct btrfs_root *extent_root = root->fs_info->extent_root;
3189 struct extent_buffer *leaf;
3191 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3198 leaf = path->nodes[0];
3199 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3200 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3201 btrfs_mark_buffer_dirty(leaf);
3203 btrfs_release_path(path);
3208 static struct btrfs_block_group_cache *
3209 next_block_group(struct btrfs_root *root,
3210 struct btrfs_block_group_cache *cache)
3212 struct rb_node *node;
3214 spin_lock(&root->fs_info->block_group_cache_lock);
3216 /* If our block group was removed, we need a full search. */
3217 if (RB_EMPTY_NODE(&cache->cache_node)) {
3218 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3220 spin_unlock(&root->fs_info->block_group_cache_lock);
3221 btrfs_put_block_group(cache);
3222 cache = btrfs_lookup_first_block_group(root->fs_info,
3226 node = rb_next(&cache->cache_node);
3227 btrfs_put_block_group(cache);
3229 cache = rb_entry(node, struct btrfs_block_group_cache,
3231 btrfs_get_block_group(cache);
3234 spin_unlock(&root->fs_info->block_group_cache_lock);
3238 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3239 struct btrfs_trans_handle *trans,
3240 struct btrfs_path *path)
3242 struct btrfs_root *root = block_group->fs_info->tree_root;
3243 struct inode *inode = NULL;
3245 int dcs = BTRFS_DC_ERROR;
3251 * If this block group is smaller than 100 megs don't bother caching the
3254 if (block_group->key.offset < (100 * 1024 * 1024)) {
3255 spin_lock(&block_group->lock);
3256 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3257 spin_unlock(&block_group->lock);
3264 inode = lookup_free_space_inode(root, block_group, path);
3265 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3266 ret = PTR_ERR(inode);
3267 btrfs_release_path(path);
3271 if (IS_ERR(inode)) {
3275 if (block_group->ro)
3278 ret = create_free_space_inode(root, trans, block_group, path);
3284 /* We've already setup this transaction, go ahead and exit */
3285 if (block_group->cache_generation == trans->transid &&
3286 i_size_read(inode)) {
3287 dcs = BTRFS_DC_SETUP;
3292 * We want to set the generation to 0, that way if anything goes wrong
3293 * from here on out we know not to trust this cache when we load up next
3296 BTRFS_I(inode)->generation = 0;
3297 ret = btrfs_update_inode(trans, root, inode);
3300 * So theoretically we could recover from this, simply set the
3301 * super cache generation to 0 so we know to invalidate the
3302 * cache, but then we'd have to keep track of the block groups
3303 * that fail this way so we know we _have_ to reset this cache
3304 * before the next commit or risk reading stale cache. So to
3305 * limit our exposure to horrible edge cases lets just abort the
3306 * transaction, this only happens in really bad situations
3309 btrfs_abort_transaction(trans, root, ret);
3314 if (i_size_read(inode) > 0) {
3315 ret = btrfs_check_trunc_cache_free_space(root,
3316 &root->fs_info->global_block_rsv);
3320 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3325 spin_lock(&block_group->lock);
3326 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3327 !btrfs_test_opt(root, SPACE_CACHE)) {
3329 * don't bother trying to write stuff out _if_
3330 * a) we're not cached,
3331 * b) we're with nospace_cache mount option.
3333 dcs = BTRFS_DC_WRITTEN;
3334 spin_unlock(&block_group->lock);
3337 spin_unlock(&block_group->lock);
3340 * Try to preallocate enough space based on how big the block group is.
3341 * Keep in mind this has to include any pinned space which could end up
3342 * taking up quite a bit since it's not folded into the other space
3345 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3350 num_pages *= PAGE_CACHE_SIZE;
3352 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3356 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3357 num_pages, num_pages,
3360 dcs = BTRFS_DC_SETUP;
3361 btrfs_free_reserved_data_space(inode, num_pages);
3366 btrfs_release_path(path);
3368 spin_lock(&block_group->lock);
3369 if (!ret && dcs == BTRFS_DC_SETUP)
3370 block_group->cache_generation = trans->transid;
3371 block_group->disk_cache_state = dcs;
3372 spin_unlock(&block_group->lock);
3377 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3378 struct btrfs_root *root)
3380 struct btrfs_block_group_cache *cache, *tmp;
3381 struct btrfs_transaction *cur_trans = trans->transaction;
3382 struct btrfs_path *path;
3384 if (list_empty(&cur_trans->dirty_bgs) ||
3385 !btrfs_test_opt(root, SPACE_CACHE))
3388 path = btrfs_alloc_path();
3392 /* Could add new block groups, use _safe just in case */
3393 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3395 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3396 cache_save_setup(cache, trans, path);
3399 btrfs_free_path(path);
3404 * transaction commit does final block group cache writeback during a
3405 * critical section where nothing is allowed to change the FS. This is
3406 * required in order for the cache to actually match the block group,
3407 * but can introduce a lot of latency into the commit.
3409 * So, btrfs_start_dirty_block_groups is here to kick off block group
3410 * cache IO. There's a chance we'll have to redo some of it if the
3411 * block group changes again during the commit, but it greatly reduces
3412 * the commit latency by getting rid of the easy block groups while
3413 * we're still allowing others to join the commit.
3415 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3416 struct btrfs_root *root)
3418 struct btrfs_block_group_cache *cache;
3419 struct btrfs_transaction *cur_trans = trans->transaction;
3422 struct btrfs_path *path = NULL;
3424 struct list_head *io = &cur_trans->io_bgs;
3425 int num_started = 0;
3428 spin_lock(&cur_trans->dirty_bgs_lock);
3429 if (list_empty(&cur_trans->dirty_bgs)) {
3430 spin_unlock(&cur_trans->dirty_bgs_lock);
3433 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3434 spin_unlock(&cur_trans->dirty_bgs_lock);
3438 * make sure all the block groups on our dirty list actually
3441 btrfs_create_pending_block_groups(trans, root);
3444 path = btrfs_alloc_path();
3450 * cache_write_mutex is here only to save us from balance or automatic
3451 * removal of empty block groups deleting this block group while we are
3452 * writing out the cache
3454 mutex_lock(&trans->transaction->cache_write_mutex);
3455 while (!list_empty(&dirty)) {
3456 cache = list_first_entry(&dirty,
3457 struct btrfs_block_group_cache,
3460 * this can happen if something re-dirties a block
3461 * group that is already under IO. Just wait for it to
3462 * finish and then do it all again
3464 if (!list_empty(&cache->io_list)) {
3465 list_del_init(&cache->io_list);
3466 btrfs_wait_cache_io(root, trans, cache,
3467 &cache->io_ctl, path,
3468 cache->key.objectid);
3469 btrfs_put_block_group(cache);
3474 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3475 * if it should update the cache_state. Don't delete
3476 * until after we wait.
3478 * Since we're not running in the commit critical section
3479 * we need the dirty_bgs_lock to protect from update_block_group
3481 spin_lock(&cur_trans->dirty_bgs_lock);
3482 list_del_init(&cache->dirty_list);
3483 spin_unlock(&cur_trans->dirty_bgs_lock);
3487 cache_save_setup(cache, trans, path);
3489 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3490 cache->io_ctl.inode = NULL;
3491 ret = btrfs_write_out_cache(root, trans, cache, path);
3492 if (ret == 0 && cache->io_ctl.inode) {
3497 * the cache_write_mutex is protecting
3500 list_add_tail(&cache->io_list, io);
3503 * if we failed to write the cache, the
3504 * generation will be bad and life goes on
3510 ret = write_one_cache_group(trans, root, path, cache);
3512 * Our block group might still be attached to the list
3513 * of new block groups in the transaction handle of some
3514 * other task (struct btrfs_trans_handle->new_bgs). This
3515 * means its block group item isn't yet in the extent
3516 * tree. If this happens ignore the error, as we will
3517 * try again later in the critical section of the
3518 * transaction commit.
3520 if (ret == -ENOENT) {
3522 spin_lock(&cur_trans->dirty_bgs_lock);
3523 if (list_empty(&cache->dirty_list)) {
3524 list_add_tail(&cache->dirty_list,
3525 &cur_trans->dirty_bgs);
3526 btrfs_get_block_group(cache);
3528 spin_unlock(&cur_trans->dirty_bgs_lock);
3530 btrfs_abort_transaction(trans, root, ret);
3534 /* if its not on the io list, we need to put the block group */
3536 btrfs_put_block_group(cache);
3542 * Avoid blocking other tasks for too long. It might even save
3543 * us from writing caches for block groups that are going to be
3546 mutex_unlock(&trans->transaction->cache_write_mutex);
3547 mutex_lock(&trans->transaction->cache_write_mutex);
3549 mutex_unlock(&trans->transaction->cache_write_mutex);
3552 * go through delayed refs for all the stuff we've just kicked off
3553 * and then loop back (just once)
3555 ret = btrfs_run_delayed_refs(trans, root, 0);
3556 if (!ret && loops == 0) {
3558 spin_lock(&cur_trans->dirty_bgs_lock);
3559 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3561 * dirty_bgs_lock protects us from concurrent block group
3562 * deletes too (not just cache_write_mutex).
3564 if (!list_empty(&dirty)) {
3565 spin_unlock(&cur_trans->dirty_bgs_lock);
3568 spin_unlock(&cur_trans->dirty_bgs_lock);
3571 btrfs_free_path(path);
3575 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3576 struct btrfs_root *root)
3578 struct btrfs_block_group_cache *cache;
3579 struct btrfs_transaction *cur_trans = trans->transaction;
3582 struct btrfs_path *path;
3583 struct list_head *io = &cur_trans->io_bgs;
3584 int num_started = 0;
3586 path = btrfs_alloc_path();
3591 * We don't need the lock here since we are protected by the transaction
3592 * commit. We want to do the cache_save_setup first and then run the
3593 * delayed refs to make sure we have the best chance at doing this all
3596 while (!list_empty(&cur_trans->dirty_bgs)) {
3597 cache = list_first_entry(&cur_trans->dirty_bgs,
3598 struct btrfs_block_group_cache,
3602 * this can happen if cache_save_setup re-dirties a block
3603 * group that is already under IO. Just wait for it to
3604 * finish and then do it all again
3606 if (!list_empty(&cache->io_list)) {
3607 list_del_init(&cache->io_list);
3608 btrfs_wait_cache_io(root, trans, cache,
3609 &cache->io_ctl, path,
3610 cache->key.objectid);
3611 btrfs_put_block_group(cache);
3615 * don't remove from the dirty list until after we've waited
3618 list_del_init(&cache->dirty_list);
3621 cache_save_setup(cache, trans, path);
3624 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3626 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3627 cache->io_ctl.inode = NULL;
3628 ret = btrfs_write_out_cache(root, trans, cache, path);
3629 if (ret == 0 && cache->io_ctl.inode) {
3632 list_add_tail(&cache->io_list, io);
3635 * if we failed to write the cache, the
3636 * generation will be bad and life goes on
3642 ret = write_one_cache_group(trans, root, path, cache);
3644 btrfs_abort_transaction(trans, root, ret);
3647 /* if its not on the io list, we need to put the block group */
3649 btrfs_put_block_group(cache);
3652 while (!list_empty(io)) {
3653 cache = list_first_entry(io, struct btrfs_block_group_cache,
3655 list_del_init(&cache->io_list);
3656 btrfs_wait_cache_io(root, trans, cache,
3657 &cache->io_ctl, path, cache->key.objectid);
3658 btrfs_put_block_group(cache);
3661 btrfs_free_path(path);
3665 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3667 struct btrfs_block_group_cache *block_group;
3670 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3671 if (!block_group || block_group->ro)
3674 btrfs_put_block_group(block_group);
3678 static const char *alloc_name(u64 flags)
3681 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3683 case BTRFS_BLOCK_GROUP_METADATA:
3685 case BTRFS_BLOCK_GROUP_DATA:
3687 case BTRFS_BLOCK_GROUP_SYSTEM:
3691 return "invalid-combination";
3695 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3696 u64 total_bytes, u64 bytes_used,
3697 struct btrfs_space_info **space_info)
3699 struct btrfs_space_info *found;
3704 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3705 BTRFS_BLOCK_GROUP_RAID10))
3710 found = __find_space_info(info, flags);
3712 spin_lock(&found->lock);
3713 found->total_bytes += total_bytes;
3714 found->disk_total += total_bytes * factor;
3715 found->bytes_used += bytes_used;
3716 found->disk_used += bytes_used * factor;
3717 if (total_bytes > 0)
3719 spin_unlock(&found->lock);
3720 *space_info = found;
3723 found = kzalloc(sizeof(*found), GFP_NOFS);
3727 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3733 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3734 INIT_LIST_HEAD(&found->block_groups[i]);
3735 init_rwsem(&found->groups_sem);
3736 spin_lock_init(&found->lock);
3737 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3738 found->total_bytes = total_bytes;
3739 found->disk_total = total_bytes * factor;
3740 found->bytes_used = bytes_used;
3741 found->disk_used = bytes_used * factor;
3742 found->bytes_pinned = 0;
3743 found->bytes_reserved = 0;
3744 found->bytes_readonly = 0;
3745 found->bytes_may_use = 0;
3746 if (total_bytes > 0)
3750 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3751 found->chunk_alloc = 0;
3753 init_waitqueue_head(&found->wait);
3754 INIT_LIST_HEAD(&found->ro_bgs);
3756 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3757 info->space_info_kobj, "%s",
3758 alloc_name(found->flags));
3764 *space_info = found;
3765 list_add_rcu(&found->list, &info->space_info);
3766 if (flags & BTRFS_BLOCK_GROUP_DATA)
3767 info->data_sinfo = found;
3772 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3774 u64 extra_flags = chunk_to_extended(flags) &
3775 BTRFS_EXTENDED_PROFILE_MASK;
3777 write_seqlock(&fs_info->profiles_lock);
3778 if (flags & BTRFS_BLOCK_GROUP_DATA)
3779 fs_info->avail_data_alloc_bits |= extra_flags;
3780 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3781 fs_info->avail_metadata_alloc_bits |= extra_flags;
3782 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3783 fs_info->avail_system_alloc_bits |= extra_flags;
3784 write_sequnlock(&fs_info->profiles_lock);
3788 * returns target flags in extended format or 0 if restripe for this
3789 * chunk_type is not in progress
3791 * should be called with either volume_mutex or balance_lock held
3793 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3795 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3801 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3802 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3803 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3804 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3805 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3806 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3807 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3808 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3809 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3816 * @flags: available profiles in extended format (see ctree.h)
3818 * Returns reduced profile in chunk format. If profile changing is in
3819 * progress (either running or paused) picks the target profile (if it's
3820 * already available), otherwise falls back to plain reducing.
3822 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3824 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3829 * see if restripe for this chunk_type is in progress, if so
3830 * try to reduce to the target profile
3832 spin_lock(&root->fs_info->balance_lock);
3833 target = get_restripe_target(root->fs_info, flags);
3835 /* pick target profile only if it's already available */
3836 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3837 spin_unlock(&root->fs_info->balance_lock);
3838 return extended_to_chunk(target);
3841 spin_unlock(&root->fs_info->balance_lock);
3843 /* First, mask out the RAID levels which aren't possible */
3844 if (num_devices == 1)
3845 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3846 BTRFS_BLOCK_GROUP_RAID5);
3847 if (num_devices < 3)
3848 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3849 if (num_devices < 4)
3850 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3852 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3853 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3854 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3857 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3858 tmp = BTRFS_BLOCK_GROUP_RAID6;
3859 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3860 tmp = BTRFS_BLOCK_GROUP_RAID5;
3861 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3862 tmp = BTRFS_BLOCK_GROUP_RAID10;
3863 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3864 tmp = BTRFS_BLOCK_GROUP_RAID1;
3865 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3866 tmp = BTRFS_BLOCK_GROUP_RAID0;
3868 return extended_to_chunk(flags | tmp);
3871 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3878 seq = read_seqbegin(&root->fs_info->profiles_lock);
3880 if (flags & BTRFS_BLOCK_GROUP_DATA)
3881 flags |= root->fs_info->avail_data_alloc_bits;
3882 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3883 flags |= root->fs_info->avail_system_alloc_bits;
3884 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3885 flags |= root->fs_info->avail_metadata_alloc_bits;
3886 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3888 return btrfs_reduce_alloc_profile(root, flags);
3891 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3897 flags = BTRFS_BLOCK_GROUP_DATA;
3898 else if (root == root->fs_info->chunk_root)
3899 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3901 flags = BTRFS_BLOCK_GROUP_METADATA;
3903 ret = get_alloc_profile(root, flags);
3908 * This will check the space that the inode allocates from to make sure we have
3909 * enough space for bytes.
3911 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3913 struct btrfs_space_info *data_sinfo;
3914 struct btrfs_root *root = BTRFS_I(inode)->root;
3915 struct btrfs_fs_info *fs_info = root->fs_info;
3918 int need_commit = 2;
3919 int have_pinned_space;
3921 /* make sure bytes are sectorsize aligned */
3922 bytes = ALIGN(bytes, root->sectorsize);
3924 if (btrfs_is_free_space_inode(inode)) {
3926 ASSERT(current->journal_info);
3929 data_sinfo = fs_info->data_sinfo;
3934 /* make sure we have enough space to handle the data first */
3935 spin_lock(&data_sinfo->lock);
3936 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3937 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3938 data_sinfo->bytes_may_use;
3940 if (used + bytes > data_sinfo->total_bytes) {
3941 struct btrfs_trans_handle *trans;
3944 * if we don't have enough free bytes in this space then we need
3945 * to alloc a new chunk.
3947 if (!data_sinfo->full) {
3950 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3951 spin_unlock(&data_sinfo->lock);
3953 alloc_target = btrfs_get_alloc_profile(root, 1);
3955 * It is ugly that we don't call nolock join
3956 * transaction for the free space inode case here.
3957 * But it is safe because we only do the data space
3958 * reservation for the free space cache in the
3959 * transaction context, the common join transaction
3960 * just increase the counter of the current transaction
3961 * handler, doesn't try to acquire the trans_lock of
3964 trans = btrfs_join_transaction(root);
3966 return PTR_ERR(trans);
3968 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3970 CHUNK_ALLOC_NO_FORCE);
3971 btrfs_end_transaction(trans, root);
3976 have_pinned_space = 1;
3982 data_sinfo = fs_info->data_sinfo;
3988 * If we don't have enough pinned space to deal with this
3989 * allocation, and no removed chunk in current transaction,
3990 * don't bother committing the transaction.
3992 have_pinned_space = percpu_counter_compare(
3993 &data_sinfo->total_bytes_pinned,
3994 used + bytes - data_sinfo->total_bytes);
3995 spin_unlock(&data_sinfo->lock);
3997 /* commit the current transaction and try again */
4000 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4003 if (need_commit > 0)
4004 btrfs_wait_ordered_roots(fs_info, -1);
4006 trans = btrfs_join_transaction(root);
4008 return PTR_ERR(trans);
4009 if (have_pinned_space >= 0 ||
4010 trans->transaction->have_free_bgs ||
4012 ret = btrfs_commit_transaction(trans, root);
4016 * make sure that all running delayed iput are
4019 down_write(&root->fs_info->delayed_iput_sem);
4020 up_write(&root->fs_info->delayed_iput_sem);
4023 btrfs_end_transaction(trans, root);
4027 trace_btrfs_space_reservation(root->fs_info,
4028 "space_info:enospc",
4029 data_sinfo->flags, bytes, 1);
4032 ret = btrfs_qgroup_reserve(root, write_bytes);
4035 data_sinfo->bytes_may_use += bytes;
4036 trace_btrfs_space_reservation(root->fs_info, "space_info",
4037 data_sinfo->flags, bytes, 1);
4039 spin_unlock(&data_sinfo->lock);
4045 * Called if we need to clear a data reservation for this inode.
4047 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4049 struct btrfs_root *root = BTRFS_I(inode)->root;
4050 struct btrfs_space_info *data_sinfo;
4052 /* make sure bytes are sectorsize aligned */
4053 bytes = ALIGN(bytes, root->sectorsize);
4055 data_sinfo = root->fs_info->data_sinfo;
4056 spin_lock(&data_sinfo->lock);
4057 WARN_ON(data_sinfo->bytes_may_use < bytes);
4058 data_sinfo->bytes_may_use -= bytes;
4059 trace_btrfs_space_reservation(root->fs_info, "space_info",
4060 data_sinfo->flags, bytes, 0);
4061 spin_unlock(&data_sinfo->lock);
4064 static void force_metadata_allocation(struct btrfs_fs_info *info)
4066 struct list_head *head = &info->space_info;
4067 struct btrfs_space_info *found;
4070 list_for_each_entry_rcu(found, head, list) {
4071 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4072 found->force_alloc = CHUNK_ALLOC_FORCE;
4077 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4079 return (global->size << 1);
4082 static int should_alloc_chunk(struct btrfs_root *root,
4083 struct btrfs_space_info *sinfo, int force)
4085 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4086 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4087 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4090 if (force == CHUNK_ALLOC_FORCE)
4094 * We need to take into account the global rsv because for all intents
4095 * and purposes it's used space. Don't worry about locking the
4096 * global_rsv, it doesn't change except when the transaction commits.
4098 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4099 num_allocated += calc_global_rsv_need_space(global_rsv);
4102 * in limited mode, we want to have some free space up to
4103 * about 1% of the FS size.
4105 if (force == CHUNK_ALLOC_LIMITED) {
4106 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4107 thresh = max_t(u64, 64 * 1024 * 1024,
4108 div_factor_fine(thresh, 1));
4110 if (num_bytes - num_allocated < thresh)
4114 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4119 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4123 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4124 BTRFS_BLOCK_GROUP_RAID0 |
4125 BTRFS_BLOCK_GROUP_RAID5 |
4126 BTRFS_BLOCK_GROUP_RAID6))
4127 num_dev = root->fs_info->fs_devices->rw_devices;
4128 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4131 num_dev = 1; /* DUP or single */
4137 * If @is_allocation is true, reserve space in the system space info necessary
4138 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4141 void check_system_chunk(struct btrfs_trans_handle *trans,
4142 struct btrfs_root *root,
4145 struct btrfs_space_info *info;
4152 * Needed because we can end up allocating a system chunk and for an
4153 * atomic and race free space reservation in the chunk block reserve.
4155 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4157 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4158 spin_lock(&info->lock);
4159 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4160 info->bytes_reserved - info->bytes_readonly -
4161 info->bytes_may_use;
4162 spin_unlock(&info->lock);
4164 num_devs = get_profile_num_devs(root, type);
4166 /* num_devs device items to update and 1 chunk item to add or remove */
4167 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4168 btrfs_calc_trans_metadata_size(root, 1);
4170 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4171 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4172 left, thresh, type);
4173 dump_space_info(info, 0, 0);
4176 if (left < thresh) {
4179 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4181 * Ignore failure to create system chunk. We might end up not
4182 * needing it, as we might not need to COW all nodes/leafs from
4183 * the paths we visit in the chunk tree (they were already COWed
4184 * or created in the current transaction for example).
4186 ret = btrfs_alloc_chunk(trans, root, flags);
4190 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4191 &root->fs_info->chunk_block_rsv,
4192 thresh, BTRFS_RESERVE_NO_FLUSH);
4194 trans->chunk_bytes_reserved += thresh;
4198 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4199 struct btrfs_root *extent_root, u64 flags, int force)
4201 struct btrfs_space_info *space_info;
4202 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4203 int wait_for_alloc = 0;
4206 /* Don't re-enter if we're already allocating a chunk */
4207 if (trans->allocating_chunk)
4210 space_info = __find_space_info(extent_root->fs_info, flags);
4212 ret = update_space_info(extent_root->fs_info, flags,
4214 BUG_ON(ret); /* -ENOMEM */
4216 BUG_ON(!space_info); /* Logic error */
4219 spin_lock(&space_info->lock);
4220 if (force < space_info->force_alloc)
4221 force = space_info->force_alloc;
4222 if (space_info->full) {
4223 if (should_alloc_chunk(extent_root, space_info, force))
4227 spin_unlock(&space_info->lock);
4231 if (!should_alloc_chunk(extent_root, space_info, force)) {
4232 spin_unlock(&space_info->lock);
4234 } else if (space_info->chunk_alloc) {
4237 space_info->chunk_alloc = 1;
4240 spin_unlock(&space_info->lock);
4242 mutex_lock(&fs_info->chunk_mutex);
4245 * The chunk_mutex is held throughout the entirety of a chunk
4246 * allocation, so once we've acquired the chunk_mutex we know that the
4247 * other guy is done and we need to recheck and see if we should
4250 if (wait_for_alloc) {
4251 mutex_unlock(&fs_info->chunk_mutex);
4256 trans->allocating_chunk = true;
4259 * If we have mixed data/metadata chunks we want to make sure we keep
4260 * allocating mixed chunks instead of individual chunks.
4262 if (btrfs_mixed_space_info(space_info))
4263 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4266 * if we're doing a data chunk, go ahead and make sure that
4267 * we keep a reasonable number of metadata chunks allocated in the
4270 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4271 fs_info->data_chunk_allocations++;
4272 if (!(fs_info->data_chunk_allocations %
4273 fs_info->metadata_ratio))
4274 force_metadata_allocation(fs_info);
4278 * Check if we have enough space in SYSTEM chunk because we may need
4279 * to update devices.
4281 check_system_chunk(trans, extent_root, flags);
4283 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4284 trans->allocating_chunk = false;
4286 spin_lock(&space_info->lock);
4287 if (ret < 0 && ret != -ENOSPC)
4290 space_info->full = 1;
4294 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4296 space_info->chunk_alloc = 0;
4297 spin_unlock(&space_info->lock);
4298 mutex_unlock(&fs_info->chunk_mutex);
4300 * When we allocate a new chunk we reserve space in the chunk block
4301 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4302 * add new nodes/leafs to it if we end up needing to do it when
4303 * inserting the chunk item and updating device items as part of the
4304 * second phase of chunk allocation, performed by
4305 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4306 * large number of new block groups to create in our transaction
4307 * handle's new_bgs list to avoid exhausting the chunk block reserve
4308 * in extreme cases - like having a single transaction create many new
4309 * block groups when starting to write out the free space caches of all
4310 * the block groups that were made dirty during the lifetime of the
4313 if (trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4314 btrfs_create_pending_block_groups(trans, trans->root);
4315 btrfs_trans_release_chunk_metadata(trans);
4320 static int can_overcommit(struct btrfs_root *root,
4321 struct btrfs_space_info *space_info, u64 bytes,
4322 enum btrfs_reserve_flush_enum flush)
4324 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4325 u64 profile = btrfs_get_alloc_profile(root, 0);
4330 used = space_info->bytes_used + space_info->bytes_reserved +
4331 space_info->bytes_pinned + space_info->bytes_readonly;
4334 * We only want to allow over committing if we have lots of actual space
4335 * free, but if we don't have enough space to handle the global reserve
4336 * space then we could end up having a real enospc problem when trying
4337 * to allocate a chunk or some other such important allocation.
4339 spin_lock(&global_rsv->lock);
4340 space_size = calc_global_rsv_need_space(global_rsv);
4341 spin_unlock(&global_rsv->lock);
4342 if (used + space_size >= space_info->total_bytes)
4345 used += space_info->bytes_may_use;
4347 spin_lock(&root->fs_info->free_chunk_lock);
4348 avail = root->fs_info->free_chunk_space;
4349 spin_unlock(&root->fs_info->free_chunk_lock);
4352 * If we have dup, raid1 or raid10 then only half of the free
4353 * space is actually useable. For raid56, the space info used
4354 * doesn't include the parity drive, so we don't have to
4357 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4358 BTRFS_BLOCK_GROUP_RAID1 |
4359 BTRFS_BLOCK_GROUP_RAID10))
4363 * If we aren't flushing all things, let us overcommit up to
4364 * 1/2th of the space. If we can flush, don't let us overcommit
4365 * too much, let it overcommit up to 1/8 of the space.
4367 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4372 if (used + bytes < space_info->total_bytes + avail)
4377 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4378 unsigned long nr_pages, int nr_items)
4380 struct super_block *sb = root->fs_info->sb;
4382 if (down_read_trylock(&sb->s_umount)) {
4383 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4384 up_read(&sb->s_umount);
4387 * We needn't worry the filesystem going from r/w to r/o though
4388 * we don't acquire ->s_umount mutex, because the filesystem
4389 * should guarantee the delalloc inodes list be empty after
4390 * the filesystem is readonly(all dirty pages are written to
4393 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4394 if (!current->journal_info)
4395 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4399 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4404 bytes = btrfs_calc_trans_metadata_size(root, 1);
4405 nr = (int)div64_u64(to_reclaim, bytes);
4411 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4414 * shrink metadata reservation for delalloc
4416 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4419 struct btrfs_block_rsv *block_rsv;
4420 struct btrfs_space_info *space_info;
4421 struct btrfs_trans_handle *trans;
4425 unsigned long nr_pages;
4428 enum btrfs_reserve_flush_enum flush;
4430 /* Calc the number of the pages we need flush for space reservation */
4431 items = calc_reclaim_items_nr(root, to_reclaim);
4432 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4434 trans = (struct btrfs_trans_handle *)current->journal_info;
4435 block_rsv = &root->fs_info->delalloc_block_rsv;
4436 space_info = block_rsv->space_info;
4438 delalloc_bytes = percpu_counter_sum_positive(
4439 &root->fs_info->delalloc_bytes);
4440 if (delalloc_bytes == 0) {
4444 btrfs_wait_ordered_roots(root->fs_info, items);
4449 while (delalloc_bytes && loops < 3) {
4450 max_reclaim = min(delalloc_bytes, to_reclaim);
4451 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4452 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4454 * We need to wait for the async pages to actually start before
4457 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4461 if (max_reclaim <= nr_pages)
4464 max_reclaim -= nr_pages;
4466 wait_event(root->fs_info->async_submit_wait,
4467 atomic_read(&root->fs_info->async_delalloc_pages) <=
4471 flush = BTRFS_RESERVE_FLUSH_ALL;
4473 flush = BTRFS_RESERVE_NO_FLUSH;
4474 spin_lock(&space_info->lock);
4475 if (can_overcommit(root, space_info, orig, flush)) {
4476 spin_unlock(&space_info->lock);
4479 spin_unlock(&space_info->lock);
4482 if (wait_ordered && !trans) {
4483 btrfs_wait_ordered_roots(root->fs_info, items);
4485 time_left = schedule_timeout_killable(1);
4489 delalloc_bytes = percpu_counter_sum_positive(
4490 &root->fs_info->delalloc_bytes);
4495 * maybe_commit_transaction - possibly commit the transaction if its ok to
4496 * @root - the root we're allocating for
4497 * @bytes - the number of bytes we want to reserve
4498 * @force - force the commit
4500 * This will check to make sure that committing the transaction will actually
4501 * get us somewhere and then commit the transaction if it does. Otherwise it
4502 * will return -ENOSPC.
4504 static int may_commit_transaction(struct btrfs_root *root,
4505 struct btrfs_space_info *space_info,
4506 u64 bytes, int force)
4508 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4509 struct btrfs_trans_handle *trans;
4511 trans = (struct btrfs_trans_handle *)current->journal_info;
4518 /* See if there is enough pinned space to make this reservation */
4519 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4524 * See if there is some space in the delayed insertion reservation for
4527 if (space_info != delayed_rsv->space_info)
4530 spin_lock(&delayed_rsv->lock);
4531 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4532 bytes - delayed_rsv->size) >= 0) {
4533 spin_unlock(&delayed_rsv->lock);
4536 spin_unlock(&delayed_rsv->lock);
4539 trans = btrfs_join_transaction(root);
4543 return btrfs_commit_transaction(trans, root);
4547 FLUSH_DELAYED_ITEMS_NR = 1,
4548 FLUSH_DELAYED_ITEMS = 2,
4550 FLUSH_DELALLOC_WAIT = 4,
4555 static int flush_space(struct btrfs_root *root,
4556 struct btrfs_space_info *space_info, u64 num_bytes,
4557 u64 orig_bytes, int state)
4559 struct btrfs_trans_handle *trans;
4564 case FLUSH_DELAYED_ITEMS_NR:
4565 case FLUSH_DELAYED_ITEMS:
4566 if (state == FLUSH_DELAYED_ITEMS_NR)
4567 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4571 trans = btrfs_join_transaction(root);
4572 if (IS_ERR(trans)) {
4573 ret = PTR_ERR(trans);
4576 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4577 btrfs_end_transaction(trans, root);
4579 case FLUSH_DELALLOC:
4580 case FLUSH_DELALLOC_WAIT:
4581 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4582 state == FLUSH_DELALLOC_WAIT);
4585 trans = btrfs_join_transaction(root);
4586 if (IS_ERR(trans)) {
4587 ret = PTR_ERR(trans);
4590 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4591 btrfs_get_alloc_profile(root, 0),
4592 CHUNK_ALLOC_NO_FORCE);
4593 btrfs_end_transaction(trans, root);
4598 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4609 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4610 struct btrfs_space_info *space_info)
4616 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4618 spin_lock(&space_info->lock);
4619 if (can_overcommit(root, space_info, to_reclaim,
4620 BTRFS_RESERVE_FLUSH_ALL)) {
4625 used = space_info->bytes_used + space_info->bytes_reserved +
4626 space_info->bytes_pinned + space_info->bytes_readonly +
4627 space_info->bytes_may_use;
4628 if (can_overcommit(root, space_info, 1024 * 1024,
4629 BTRFS_RESERVE_FLUSH_ALL))
4630 expected = div_factor_fine(space_info->total_bytes, 95);
4632 expected = div_factor_fine(space_info->total_bytes, 90);
4634 if (used > expected)
4635 to_reclaim = used - expected;
4638 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4639 space_info->bytes_reserved);
4641 spin_unlock(&space_info->lock);
4646 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4647 struct btrfs_fs_info *fs_info, u64 used)
4649 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4651 /* If we're just plain full then async reclaim just slows us down. */
4652 if (space_info->bytes_used >= thresh)
4655 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4656 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4659 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4660 struct btrfs_fs_info *fs_info,
4665 spin_lock(&space_info->lock);
4667 * We run out of space and have not got any free space via flush_space,
4668 * so don't bother doing async reclaim.
4670 if (flush_state > COMMIT_TRANS && space_info->full) {
4671 spin_unlock(&space_info->lock);
4675 used = space_info->bytes_used + space_info->bytes_reserved +
4676 space_info->bytes_pinned + space_info->bytes_readonly +
4677 space_info->bytes_may_use;
4678 if (need_do_async_reclaim(space_info, fs_info, used)) {
4679 spin_unlock(&space_info->lock);
4682 spin_unlock(&space_info->lock);
4687 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4689 struct btrfs_fs_info *fs_info;
4690 struct btrfs_space_info *space_info;
4694 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4695 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4697 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4702 flush_state = FLUSH_DELAYED_ITEMS_NR;
4704 flush_space(fs_info->fs_root, space_info, to_reclaim,
4705 to_reclaim, flush_state);
4707 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4710 } while (flush_state < COMMIT_TRANS);
4713 void btrfs_init_async_reclaim_work(struct work_struct *work)
4715 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4719 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4720 * @root - the root we're allocating for
4721 * @block_rsv - the block_rsv we're allocating for
4722 * @orig_bytes - the number of bytes we want
4723 * @flush - whether or not we can flush to make our reservation
4725 * This will reserve orgi_bytes number of bytes from the space info associated
4726 * with the block_rsv. If there is not enough space it will make an attempt to
4727 * flush out space to make room. It will do this by flushing delalloc if
4728 * possible or committing the transaction. If flush is 0 then no attempts to
4729 * regain reservations will be made and this will fail if there is not enough
4732 static int reserve_metadata_bytes(struct btrfs_root *root,
4733 struct btrfs_block_rsv *block_rsv,
4735 enum btrfs_reserve_flush_enum flush)
4737 struct btrfs_space_info *space_info = block_rsv->space_info;
4739 u64 num_bytes = orig_bytes;
4740 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4742 bool flushing = false;
4746 spin_lock(&space_info->lock);
4748 * We only want to wait if somebody other than us is flushing and we
4749 * are actually allowed to flush all things.
4751 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4752 space_info->flush) {
4753 spin_unlock(&space_info->lock);
4755 * If we have a trans handle we can't wait because the flusher
4756 * may have to commit the transaction, which would mean we would
4757 * deadlock since we are waiting for the flusher to finish, but
4758 * hold the current transaction open.
4760 if (current->journal_info)
4762 ret = wait_event_killable(space_info->wait, !space_info->flush);
4763 /* Must have been killed, return */
4767 spin_lock(&space_info->lock);
4771 used = space_info->bytes_used + space_info->bytes_reserved +
4772 space_info->bytes_pinned + space_info->bytes_readonly +
4773 space_info->bytes_may_use;
4776 * The idea here is that we've not already over-reserved the block group
4777 * then we can go ahead and save our reservation first and then start
4778 * flushing if we need to. Otherwise if we've already overcommitted
4779 * lets start flushing stuff first and then come back and try to make
4782 if (used <= space_info->total_bytes) {
4783 if (used + orig_bytes <= space_info->total_bytes) {
4784 space_info->bytes_may_use += orig_bytes;
4785 trace_btrfs_space_reservation(root->fs_info,
4786 "space_info", space_info->flags, orig_bytes, 1);
4790 * Ok set num_bytes to orig_bytes since we aren't
4791 * overocmmitted, this way we only try and reclaim what
4794 num_bytes = orig_bytes;
4798 * Ok we're over committed, set num_bytes to the overcommitted
4799 * amount plus the amount of bytes that we need for this
4802 num_bytes = used - space_info->total_bytes +
4806 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4807 space_info->bytes_may_use += orig_bytes;
4808 trace_btrfs_space_reservation(root->fs_info, "space_info",
4809 space_info->flags, orig_bytes,
4815 * Couldn't make our reservation, save our place so while we're trying
4816 * to reclaim space we can actually use it instead of somebody else
4817 * stealing it from us.
4819 * We make the other tasks wait for the flush only when we can flush
4822 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4824 space_info->flush = 1;
4825 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4828 * We will do the space reservation dance during log replay,
4829 * which means we won't have fs_info->fs_root set, so don't do
4830 * the async reclaim as we will panic.
4832 if (!root->fs_info->log_root_recovering &&
4833 need_do_async_reclaim(space_info, root->fs_info, used) &&
4834 !work_busy(&root->fs_info->async_reclaim_work))
4835 queue_work(system_unbound_wq,
4836 &root->fs_info->async_reclaim_work);
4838 spin_unlock(&space_info->lock);
4840 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4843 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4848 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4849 * would happen. So skip delalloc flush.
4851 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4852 (flush_state == FLUSH_DELALLOC ||
4853 flush_state == FLUSH_DELALLOC_WAIT))
4854 flush_state = ALLOC_CHUNK;
4858 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4859 flush_state < COMMIT_TRANS)
4861 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4862 flush_state <= COMMIT_TRANS)
4866 if (ret == -ENOSPC &&
4867 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4868 struct btrfs_block_rsv *global_rsv =
4869 &root->fs_info->global_block_rsv;
4871 if (block_rsv != global_rsv &&
4872 !block_rsv_use_bytes(global_rsv, orig_bytes))
4876 trace_btrfs_space_reservation(root->fs_info,
4877 "space_info:enospc",
4878 space_info->flags, orig_bytes, 1);
4880 spin_lock(&space_info->lock);
4881 space_info->flush = 0;
4882 wake_up_all(&space_info->wait);
4883 spin_unlock(&space_info->lock);
4888 static struct btrfs_block_rsv *get_block_rsv(
4889 const struct btrfs_trans_handle *trans,
4890 const struct btrfs_root *root)
4892 struct btrfs_block_rsv *block_rsv = NULL;
4894 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4895 block_rsv = trans->block_rsv;
4897 if (root == root->fs_info->csum_root && trans->adding_csums)
4898 block_rsv = trans->block_rsv;
4900 if (root == root->fs_info->uuid_root)
4901 block_rsv = trans->block_rsv;
4904 block_rsv = root->block_rsv;
4907 block_rsv = &root->fs_info->empty_block_rsv;
4912 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4916 spin_lock(&block_rsv->lock);
4917 if (block_rsv->reserved >= num_bytes) {
4918 block_rsv->reserved -= num_bytes;
4919 if (block_rsv->reserved < block_rsv->size)
4920 block_rsv->full = 0;
4923 spin_unlock(&block_rsv->lock);
4927 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4928 u64 num_bytes, int update_size)
4930 spin_lock(&block_rsv->lock);
4931 block_rsv->reserved += num_bytes;
4933 block_rsv->size += num_bytes;
4934 else if (block_rsv->reserved >= block_rsv->size)
4935 block_rsv->full = 1;
4936 spin_unlock(&block_rsv->lock);
4939 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4940 struct btrfs_block_rsv *dest, u64 num_bytes,
4943 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4946 if (global_rsv->space_info != dest->space_info)
4949 spin_lock(&global_rsv->lock);
4950 min_bytes = div_factor(global_rsv->size, min_factor);
4951 if (global_rsv->reserved < min_bytes + num_bytes) {
4952 spin_unlock(&global_rsv->lock);
4955 global_rsv->reserved -= num_bytes;
4956 if (global_rsv->reserved < global_rsv->size)
4957 global_rsv->full = 0;
4958 spin_unlock(&global_rsv->lock);
4960 block_rsv_add_bytes(dest, num_bytes, 1);
4964 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4965 struct btrfs_block_rsv *block_rsv,
4966 struct btrfs_block_rsv *dest, u64 num_bytes)
4968 struct btrfs_space_info *space_info = block_rsv->space_info;
4970 spin_lock(&block_rsv->lock);
4971 if (num_bytes == (u64)-1)
4972 num_bytes = block_rsv->size;
4973 block_rsv->size -= num_bytes;
4974 if (block_rsv->reserved >= block_rsv->size) {
4975 num_bytes = block_rsv->reserved - block_rsv->size;
4976 block_rsv->reserved = block_rsv->size;
4977 block_rsv->full = 1;
4981 spin_unlock(&block_rsv->lock);
4983 if (num_bytes > 0) {
4985 spin_lock(&dest->lock);
4989 bytes_to_add = dest->size - dest->reserved;
4990 bytes_to_add = min(num_bytes, bytes_to_add);
4991 dest->reserved += bytes_to_add;
4992 if (dest->reserved >= dest->size)
4994 num_bytes -= bytes_to_add;
4996 spin_unlock(&dest->lock);
4999 spin_lock(&space_info->lock);
5000 space_info->bytes_may_use -= num_bytes;
5001 trace_btrfs_space_reservation(fs_info, "space_info",
5002 space_info->flags, num_bytes, 0);
5003 spin_unlock(&space_info->lock);
5008 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5009 struct btrfs_block_rsv *dst, u64 num_bytes)
5013 ret = block_rsv_use_bytes(src, num_bytes);
5017 block_rsv_add_bytes(dst, num_bytes, 1);
5021 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5023 memset(rsv, 0, sizeof(*rsv));
5024 spin_lock_init(&rsv->lock);
5028 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5029 unsigned short type)
5031 struct btrfs_block_rsv *block_rsv;
5032 struct btrfs_fs_info *fs_info = root->fs_info;
5034 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5038 btrfs_init_block_rsv(block_rsv, type);
5039 block_rsv->space_info = __find_space_info(fs_info,
5040 BTRFS_BLOCK_GROUP_METADATA);
5044 void btrfs_free_block_rsv(struct btrfs_root *root,
5045 struct btrfs_block_rsv *rsv)
5049 btrfs_block_rsv_release(root, rsv, (u64)-1);
5053 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5058 int btrfs_block_rsv_add(struct btrfs_root *root,
5059 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5060 enum btrfs_reserve_flush_enum flush)
5067 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5069 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5076 int btrfs_block_rsv_check(struct btrfs_root *root,
5077 struct btrfs_block_rsv *block_rsv, int min_factor)
5085 spin_lock(&block_rsv->lock);
5086 num_bytes = div_factor(block_rsv->size, min_factor);
5087 if (block_rsv->reserved >= num_bytes)
5089 spin_unlock(&block_rsv->lock);
5094 int btrfs_block_rsv_refill(struct btrfs_root *root,
5095 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5096 enum btrfs_reserve_flush_enum flush)
5104 spin_lock(&block_rsv->lock);
5105 num_bytes = min_reserved;
5106 if (block_rsv->reserved >= num_bytes)
5109 num_bytes -= block_rsv->reserved;
5110 spin_unlock(&block_rsv->lock);
5115 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5117 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5124 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5125 struct btrfs_block_rsv *dst_rsv,
5128 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5131 void btrfs_block_rsv_release(struct btrfs_root *root,
5132 struct btrfs_block_rsv *block_rsv,
5135 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5136 if (global_rsv == block_rsv ||
5137 block_rsv->space_info != global_rsv->space_info)
5139 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5144 * helper to calculate size of global block reservation.
5145 * the desired value is sum of space used by extent tree,
5146 * checksum tree and root tree
5148 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5150 struct btrfs_space_info *sinfo;
5154 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5156 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5157 spin_lock(&sinfo->lock);
5158 data_used = sinfo->bytes_used;
5159 spin_unlock(&sinfo->lock);
5161 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5162 spin_lock(&sinfo->lock);
5163 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5165 meta_used = sinfo->bytes_used;
5166 spin_unlock(&sinfo->lock);
5168 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5170 num_bytes += div_u64(data_used + meta_used, 50);
5172 if (num_bytes * 3 > meta_used)
5173 num_bytes = div_u64(meta_used, 3);
5175 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5178 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5180 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5181 struct btrfs_space_info *sinfo = block_rsv->space_info;
5184 num_bytes = calc_global_metadata_size(fs_info);
5186 spin_lock(&sinfo->lock);
5187 spin_lock(&block_rsv->lock);
5189 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5191 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5192 sinfo->bytes_reserved + sinfo->bytes_readonly +
5193 sinfo->bytes_may_use;
5195 if (sinfo->total_bytes > num_bytes) {
5196 num_bytes = sinfo->total_bytes - num_bytes;
5197 block_rsv->reserved += num_bytes;
5198 sinfo->bytes_may_use += num_bytes;
5199 trace_btrfs_space_reservation(fs_info, "space_info",
5200 sinfo->flags, num_bytes, 1);
5203 if (block_rsv->reserved >= block_rsv->size) {
5204 num_bytes = block_rsv->reserved - block_rsv->size;
5205 sinfo->bytes_may_use -= num_bytes;
5206 trace_btrfs_space_reservation(fs_info, "space_info",
5207 sinfo->flags, num_bytes, 0);
5208 block_rsv->reserved = block_rsv->size;
5209 block_rsv->full = 1;
5212 spin_unlock(&block_rsv->lock);
5213 spin_unlock(&sinfo->lock);
5216 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5218 struct btrfs_space_info *space_info;
5220 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5221 fs_info->chunk_block_rsv.space_info = space_info;
5223 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5224 fs_info->global_block_rsv.space_info = space_info;
5225 fs_info->delalloc_block_rsv.space_info = space_info;
5226 fs_info->trans_block_rsv.space_info = space_info;
5227 fs_info->empty_block_rsv.space_info = space_info;
5228 fs_info->delayed_block_rsv.space_info = space_info;
5230 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5231 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5232 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5233 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5234 if (fs_info->quota_root)
5235 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5236 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5238 update_global_block_rsv(fs_info);
5241 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5243 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5245 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5246 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5247 WARN_ON(fs_info->trans_block_rsv.size > 0);
5248 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5249 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5250 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5251 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5252 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5255 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5256 struct btrfs_root *root)
5258 if (!trans->block_rsv)
5261 if (!trans->bytes_reserved)
5264 trace_btrfs_space_reservation(root->fs_info, "transaction",
5265 trans->transid, trans->bytes_reserved, 0);
5266 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5267 trans->bytes_reserved = 0;
5271 * To be called after all the new block groups attached to the transaction
5272 * handle have been created (btrfs_create_pending_block_groups()).
5274 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5276 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5278 if (!trans->chunk_bytes_reserved)
5281 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5283 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5284 trans->chunk_bytes_reserved);
5285 trans->chunk_bytes_reserved = 0;
5288 /* Can only return 0 or -ENOSPC */
5289 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5290 struct inode *inode)
5292 struct btrfs_root *root = BTRFS_I(inode)->root;
5293 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5294 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5297 * We need to hold space in order to delete our orphan item once we've
5298 * added it, so this takes the reservation so we can release it later
5299 * when we are truly done with the orphan item.
5301 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5302 trace_btrfs_space_reservation(root->fs_info, "orphan",
5303 btrfs_ino(inode), num_bytes, 1);
5304 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5307 void btrfs_orphan_release_metadata(struct inode *inode)
5309 struct btrfs_root *root = BTRFS_I(inode)->root;
5310 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5311 trace_btrfs_space_reservation(root->fs_info, "orphan",
5312 btrfs_ino(inode), num_bytes, 0);
5313 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5317 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5318 * root: the root of the parent directory
5319 * rsv: block reservation
5320 * items: the number of items that we need do reservation
5321 * qgroup_reserved: used to return the reserved size in qgroup
5323 * This function is used to reserve the space for snapshot/subvolume
5324 * creation and deletion. Those operations are different with the
5325 * common file/directory operations, they change two fs/file trees
5326 * and root tree, the number of items that the qgroup reserves is
5327 * different with the free space reservation. So we can not use
5328 * the space reseravtion mechanism in start_transaction().
5330 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5331 struct btrfs_block_rsv *rsv,
5333 u64 *qgroup_reserved,
5334 bool use_global_rsv)
5338 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5340 if (root->fs_info->quota_enabled) {
5341 /* One for parent inode, two for dir entries */
5342 num_bytes = 3 * root->nodesize;
5343 ret = btrfs_qgroup_reserve(root, num_bytes);
5350 *qgroup_reserved = num_bytes;
5352 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5353 rsv->space_info = __find_space_info(root->fs_info,
5354 BTRFS_BLOCK_GROUP_METADATA);
5355 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5356 BTRFS_RESERVE_FLUSH_ALL);
5358 if (ret == -ENOSPC && use_global_rsv)
5359 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5362 if (*qgroup_reserved)
5363 btrfs_qgroup_free(root, *qgroup_reserved);
5369 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5370 struct btrfs_block_rsv *rsv,
5371 u64 qgroup_reserved)
5373 btrfs_block_rsv_release(root, rsv, (u64)-1);
5377 * drop_outstanding_extent - drop an outstanding extent
5378 * @inode: the inode we're dropping the extent for
5379 * @num_bytes: the number of bytes we're relaseing.
5381 * This is called when we are freeing up an outstanding extent, either called
5382 * after an error or after an extent is written. This will return the number of
5383 * reserved extents that need to be freed. This must be called with
5384 * BTRFS_I(inode)->lock held.
5386 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5388 unsigned drop_inode_space = 0;
5389 unsigned dropped_extents = 0;
5390 unsigned num_extents = 0;
5392 num_extents = (unsigned)div64_u64(num_bytes +
5393 BTRFS_MAX_EXTENT_SIZE - 1,
5394 BTRFS_MAX_EXTENT_SIZE);
5395 ASSERT(num_extents);
5396 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5397 BTRFS_I(inode)->outstanding_extents -= num_extents;
5399 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5400 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5401 &BTRFS_I(inode)->runtime_flags))
5402 drop_inode_space = 1;
5405 * If we have more or the same amount of outsanding extents than we have
5406 * reserved then we need to leave the reserved extents count alone.
5408 if (BTRFS_I(inode)->outstanding_extents >=
5409 BTRFS_I(inode)->reserved_extents)
5410 return drop_inode_space;
5412 dropped_extents = BTRFS_I(inode)->reserved_extents -
5413 BTRFS_I(inode)->outstanding_extents;
5414 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5415 return dropped_extents + drop_inode_space;
5419 * calc_csum_metadata_size - return the amount of metada space that must be
5420 * reserved/free'd for the given bytes.
5421 * @inode: the inode we're manipulating
5422 * @num_bytes: the number of bytes in question
5423 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5425 * This adjusts the number of csum_bytes in the inode and then returns the
5426 * correct amount of metadata that must either be reserved or freed. We
5427 * calculate how many checksums we can fit into one leaf and then divide the
5428 * number of bytes that will need to be checksumed by this value to figure out
5429 * how many checksums will be required. If we are adding bytes then the number
5430 * may go up and we will return the number of additional bytes that must be
5431 * reserved. If it is going down we will return the number of bytes that must
5434 * This must be called with BTRFS_I(inode)->lock held.
5436 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5439 struct btrfs_root *root = BTRFS_I(inode)->root;
5440 u64 old_csums, num_csums;
5442 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5443 BTRFS_I(inode)->csum_bytes == 0)
5446 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5448 BTRFS_I(inode)->csum_bytes += num_bytes;
5450 BTRFS_I(inode)->csum_bytes -= num_bytes;
5451 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5453 /* No change, no need to reserve more */
5454 if (old_csums == num_csums)
5458 return btrfs_calc_trans_metadata_size(root,
5459 num_csums - old_csums);
5461 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5464 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5466 struct btrfs_root *root = BTRFS_I(inode)->root;
5467 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5470 unsigned nr_extents = 0;
5471 int extra_reserve = 0;
5472 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5474 bool delalloc_lock = true;
5478 /* If we are a free space inode we need to not flush since we will be in
5479 * the middle of a transaction commit. We also don't need the delalloc
5480 * mutex since we won't race with anybody. We need this mostly to make
5481 * lockdep shut its filthy mouth.
5483 if (btrfs_is_free_space_inode(inode)) {
5484 flush = BTRFS_RESERVE_NO_FLUSH;
5485 delalloc_lock = false;
5488 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5489 btrfs_transaction_in_commit(root->fs_info))
5490 schedule_timeout(1);
5493 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5495 num_bytes = ALIGN(num_bytes, root->sectorsize);
5497 spin_lock(&BTRFS_I(inode)->lock);
5498 nr_extents = (unsigned)div64_u64(num_bytes +
5499 BTRFS_MAX_EXTENT_SIZE - 1,
5500 BTRFS_MAX_EXTENT_SIZE);
5501 BTRFS_I(inode)->outstanding_extents += nr_extents;
5504 if (BTRFS_I(inode)->outstanding_extents >
5505 BTRFS_I(inode)->reserved_extents)
5506 nr_extents = BTRFS_I(inode)->outstanding_extents -
5507 BTRFS_I(inode)->reserved_extents;
5510 * Add an item to reserve for updating the inode when we complete the
5513 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5514 &BTRFS_I(inode)->runtime_flags)) {
5519 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5520 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5521 csum_bytes = BTRFS_I(inode)->csum_bytes;
5522 spin_unlock(&BTRFS_I(inode)->lock);
5524 if (root->fs_info->quota_enabled) {
5525 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5530 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5531 if (unlikely(ret)) {
5532 if (root->fs_info->quota_enabled)
5533 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5537 spin_lock(&BTRFS_I(inode)->lock);
5538 if (extra_reserve) {
5539 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5540 &BTRFS_I(inode)->runtime_flags);
5543 BTRFS_I(inode)->reserved_extents += nr_extents;
5544 spin_unlock(&BTRFS_I(inode)->lock);
5547 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5550 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5551 btrfs_ino(inode), to_reserve, 1);
5552 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5557 spin_lock(&BTRFS_I(inode)->lock);
5558 dropped = drop_outstanding_extent(inode, num_bytes);
5560 * If the inodes csum_bytes is the same as the original
5561 * csum_bytes then we know we haven't raced with any free()ers
5562 * so we can just reduce our inodes csum bytes and carry on.
5564 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5565 calc_csum_metadata_size(inode, num_bytes, 0);
5567 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5571 * This is tricky, but first we need to figure out how much we
5572 * free'd from any free-ers that occured during this
5573 * reservation, so we reset ->csum_bytes to the csum_bytes
5574 * before we dropped our lock, and then call the free for the
5575 * number of bytes that were freed while we were trying our
5578 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5579 BTRFS_I(inode)->csum_bytes = csum_bytes;
5580 to_free = calc_csum_metadata_size(inode, bytes, 0);
5584 * Now we need to see how much we would have freed had we not
5585 * been making this reservation and our ->csum_bytes were not
5586 * artificially inflated.
5588 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5589 bytes = csum_bytes - orig_csum_bytes;
5590 bytes = calc_csum_metadata_size(inode, bytes, 0);
5593 * Now reset ->csum_bytes to what it should be. If bytes is
5594 * more than to_free then we would have free'd more space had we
5595 * not had an artificially high ->csum_bytes, so we need to free
5596 * the remainder. If bytes is the same or less then we don't
5597 * need to do anything, the other free-ers did the correct
5600 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5601 if (bytes > to_free)
5602 to_free = bytes - to_free;
5606 spin_unlock(&BTRFS_I(inode)->lock);
5608 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5611 btrfs_block_rsv_release(root, block_rsv, to_free);
5612 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5613 btrfs_ino(inode), to_free, 0);
5616 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5621 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5622 * @inode: the inode to release the reservation for
5623 * @num_bytes: the number of bytes we're releasing
5625 * This will release the metadata reservation for an inode. This can be called
5626 * once we complete IO for a given set of bytes to release their metadata
5629 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5631 struct btrfs_root *root = BTRFS_I(inode)->root;
5635 num_bytes = ALIGN(num_bytes, root->sectorsize);
5636 spin_lock(&BTRFS_I(inode)->lock);
5637 dropped = drop_outstanding_extent(inode, num_bytes);
5640 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5641 spin_unlock(&BTRFS_I(inode)->lock);
5643 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5645 if (btrfs_test_is_dummy_root(root))
5648 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5649 btrfs_ino(inode), to_free, 0);
5651 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5656 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5657 * @inode: inode we're writing to
5658 * @num_bytes: the number of bytes we want to allocate
5660 * This will do the following things
5662 * o reserve space in the data space info for num_bytes
5663 * o reserve space in the metadata space info based on number of outstanding
5664 * extents and how much csums will be needed
5665 * o add to the inodes ->delalloc_bytes
5666 * o add it to the fs_info's delalloc inodes list.
5668 * This will return 0 for success and -ENOSPC if there is no space left.
5670 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5674 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5678 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5680 btrfs_free_reserved_data_space(inode, num_bytes);
5688 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5689 * @inode: inode we're releasing space for
5690 * @num_bytes: the number of bytes we want to free up
5692 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5693 * called in the case that we don't need the metadata AND data reservations
5694 * anymore. So if there is an error or we insert an inline extent.
5696 * This function will release the metadata space that was not used and will
5697 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5698 * list if there are no delalloc bytes left.
5700 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5702 btrfs_delalloc_release_metadata(inode, num_bytes);
5703 btrfs_free_reserved_data_space(inode, num_bytes);
5706 static int update_block_group(struct btrfs_trans_handle *trans,
5707 struct btrfs_root *root, u64 bytenr,
5708 u64 num_bytes, int alloc)
5710 struct btrfs_block_group_cache *cache = NULL;
5711 struct btrfs_fs_info *info = root->fs_info;
5712 u64 total = num_bytes;
5717 /* block accounting for super block */
5718 spin_lock(&info->delalloc_root_lock);
5719 old_val = btrfs_super_bytes_used(info->super_copy);
5721 old_val += num_bytes;
5723 old_val -= num_bytes;
5724 btrfs_set_super_bytes_used(info->super_copy, old_val);
5725 spin_unlock(&info->delalloc_root_lock);
5728 cache = btrfs_lookup_block_group(info, bytenr);
5731 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5732 BTRFS_BLOCK_GROUP_RAID1 |
5733 BTRFS_BLOCK_GROUP_RAID10))
5738 * If this block group has free space cache written out, we
5739 * need to make sure to load it if we are removing space. This
5740 * is because we need the unpinning stage to actually add the
5741 * space back to the block group, otherwise we will leak space.
5743 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5744 cache_block_group(cache, 1);
5746 byte_in_group = bytenr - cache->key.objectid;
5747 WARN_ON(byte_in_group > cache->key.offset);
5749 spin_lock(&cache->space_info->lock);
5750 spin_lock(&cache->lock);
5752 if (btrfs_test_opt(root, SPACE_CACHE) &&
5753 cache->disk_cache_state < BTRFS_DC_CLEAR)
5754 cache->disk_cache_state = BTRFS_DC_CLEAR;
5756 old_val = btrfs_block_group_used(&cache->item);
5757 num_bytes = min(total, cache->key.offset - byte_in_group);
5759 old_val += num_bytes;
5760 btrfs_set_block_group_used(&cache->item, old_val);
5761 cache->reserved -= num_bytes;
5762 cache->space_info->bytes_reserved -= num_bytes;
5763 cache->space_info->bytes_used += num_bytes;
5764 cache->space_info->disk_used += num_bytes * factor;
5765 spin_unlock(&cache->lock);
5766 spin_unlock(&cache->space_info->lock);
5768 old_val -= num_bytes;
5769 btrfs_set_block_group_used(&cache->item, old_val);
5770 cache->pinned += num_bytes;
5771 cache->space_info->bytes_pinned += num_bytes;
5772 cache->space_info->bytes_used -= num_bytes;
5773 cache->space_info->disk_used -= num_bytes * factor;
5774 spin_unlock(&cache->lock);
5775 spin_unlock(&cache->space_info->lock);
5777 set_extent_dirty(info->pinned_extents,
5778 bytenr, bytenr + num_bytes - 1,
5779 GFP_NOFS | __GFP_NOFAIL);
5781 * No longer have used bytes in this block group, queue
5785 spin_lock(&info->unused_bgs_lock);
5786 if (list_empty(&cache->bg_list)) {
5787 btrfs_get_block_group(cache);
5788 list_add_tail(&cache->bg_list,
5791 spin_unlock(&info->unused_bgs_lock);
5795 spin_lock(&trans->transaction->dirty_bgs_lock);
5796 if (list_empty(&cache->dirty_list)) {
5797 list_add_tail(&cache->dirty_list,
5798 &trans->transaction->dirty_bgs);
5799 trans->transaction->num_dirty_bgs++;
5800 btrfs_get_block_group(cache);
5802 spin_unlock(&trans->transaction->dirty_bgs_lock);
5804 btrfs_put_block_group(cache);
5806 bytenr += num_bytes;
5811 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5813 struct btrfs_block_group_cache *cache;
5816 spin_lock(&root->fs_info->block_group_cache_lock);
5817 bytenr = root->fs_info->first_logical_byte;
5818 spin_unlock(&root->fs_info->block_group_cache_lock);
5820 if (bytenr < (u64)-1)
5823 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5827 bytenr = cache->key.objectid;
5828 btrfs_put_block_group(cache);
5833 static int pin_down_extent(struct btrfs_root *root,
5834 struct btrfs_block_group_cache *cache,
5835 u64 bytenr, u64 num_bytes, int reserved)
5837 spin_lock(&cache->space_info->lock);
5838 spin_lock(&cache->lock);
5839 cache->pinned += num_bytes;
5840 cache->space_info->bytes_pinned += num_bytes;
5842 cache->reserved -= num_bytes;
5843 cache->space_info->bytes_reserved -= num_bytes;
5845 spin_unlock(&cache->lock);
5846 spin_unlock(&cache->space_info->lock);
5848 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5849 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5851 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5856 * this function must be called within transaction
5858 int btrfs_pin_extent(struct btrfs_root *root,
5859 u64 bytenr, u64 num_bytes, int reserved)
5861 struct btrfs_block_group_cache *cache;
5863 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5864 BUG_ON(!cache); /* Logic error */
5866 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5868 btrfs_put_block_group(cache);
5873 * this function must be called within transaction
5875 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5876 u64 bytenr, u64 num_bytes)
5878 struct btrfs_block_group_cache *cache;
5881 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5886 * pull in the free space cache (if any) so that our pin
5887 * removes the free space from the cache. We have load_only set
5888 * to one because the slow code to read in the free extents does check
5889 * the pinned extents.
5891 cache_block_group(cache, 1);
5893 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5895 /* remove us from the free space cache (if we're there at all) */
5896 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5897 btrfs_put_block_group(cache);
5901 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5904 struct btrfs_block_group_cache *block_group;
5905 struct btrfs_caching_control *caching_ctl;
5907 block_group = btrfs_lookup_block_group(root->fs_info, start);
5911 cache_block_group(block_group, 0);
5912 caching_ctl = get_caching_control(block_group);
5916 BUG_ON(!block_group_cache_done(block_group));
5917 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5919 mutex_lock(&caching_ctl->mutex);
5921 if (start >= caching_ctl->progress) {
5922 ret = add_excluded_extent(root, start, num_bytes);
5923 } else if (start + num_bytes <= caching_ctl->progress) {
5924 ret = btrfs_remove_free_space(block_group,
5927 num_bytes = caching_ctl->progress - start;
5928 ret = btrfs_remove_free_space(block_group,
5933 num_bytes = (start + num_bytes) -
5934 caching_ctl->progress;
5935 start = caching_ctl->progress;
5936 ret = add_excluded_extent(root, start, num_bytes);
5939 mutex_unlock(&caching_ctl->mutex);
5940 put_caching_control(caching_ctl);
5942 btrfs_put_block_group(block_group);
5946 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5947 struct extent_buffer *eb)
5949 struct btrfs_file_extent_item *item;
5950 struct btrfs_key key;
5954 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5957 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5958 btrfs_item_key_to_cpu(eb, &key, i);
5959 if (key.type != BTRFS_EXTENT_DATA_KEY)
5961 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5962 found_type = btrfs_file_extent_type(eb, item);
5963 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5965 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5967 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5968 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5969 __exclude_logged_extent(log, key.objectid, key.offset);
5976 * btrfs_update_reserved_bytes - update the block_group and space info counters
5977 * @cache: The cache we are manipulating
5978 * @num_bytes: The number of bytes in question
5979 * @reserve: One of the reservation enums
5980 * @delalloc: The blocks are allocated for the delalloc write
5982 * This is called by the allocator when it reserves space, or by somebody who is
5983 * freeing space that was never actually used on disk. For example if you
5984 * reserve some space for a new leaf in transaction A and before transaction A
5985 * commits you free that leaf, you call this with reserve set to 0 in order to
5986 * clear the reservation.
5988 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5989 * ENOSPC accounting. For data we handle the reservation through clearing the
5990 * delalloc bits in the io_tree. We have to do this since we could end up
5991 * allocating less disk space for the amount of data we have reserved in the
5992 * case of compression.
5994 * If this is a reservation and the block group has become read only we cannot
5995 * make the reservation and return -EAGAIN, otherwise this function always
5998 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5999 u64 num_bytes, int reserve, int delalloc)
6001 struct btrfs_space_info *space_info = cache->space_info;
6004 spin_lock(&space_info->lock);
6005 spin_lock(&cache->lock);
6006 if (reserve != RESERVE_FREE) {
6010 cache->reserved += num_bytes;
6011 space_info->bytes_reserved += num_bytes;
6012 if (reserve == RESERVE_ALLOC) {
6013 trace_btrfs_space_reservation(cache->fs_info,
6014 "space_info", space_info->flags,
6016 space_info->bytes_may_use -= num_bytes;
6020 cache->delalloc_bytes += num_bytes;
6024 space_info->bytes_readonly += num_bytes;
6025 cache->reserved -= num_bytes;
6026 space_info->bytes_reserved -= num_bytes;
6029 cache->delalloc_bytes -= num_bytes;
6031 spin_unlock(&cache->lock);
6032 spin_unlock(&space_info->lock);
6036 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6037 struct btrfs_root *root)
6039 struct btrfs_fs_info *fs_info = root->fs_info;
6040 struct btrfs_caching_control *next;
6041 struct btrfs_caching_control *caching_ctl;
6042 struct btrfs_block_group_cache *cache;
6044 down_write(&fs_info->commit_root_sem);
6046 list_for_each_entry_safe(caching_ctl, next,
6047 &fs_info->caching_block_groups, list) {
6048 cache = caching_ctl->block_group;
6049 if (block_group_cache_done(cache)) {
6050 cache->last_byte_to_unpin = (u64)-1;
6051 list_del_init(&caching_ctl->list);
6052 put_caching_control(caching_ctl);
6054 cache->last_byte_to_unpin = caching_ctl->progress;
6058 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6059 fs_info->pinned_extents = &fs_info->freed_extents[1];
6061 fs_info->pinned_extents = &fs_info->freed_extents[0];
6063 up_write(&fs_info->commit_root_sem);
6065 update_global_block_rsv(fs_info);
6068 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6069 const bool return_free_space)
6071 struct btrfs_fs_info *fs_info = root->fs_info;
6072 struct btrfs_block_group_cache *cache = NULL;
6073 struct btrfs_space_info *space_info;
6074 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6078 while (start <= end) {
6081 start >= cache->key.objectid + cache->key.offset) {
6083 btrfs_put_block_group(cache);
6084 cache = btrfs_lookup_block_group(fs_info, start);
6085 BUG_ON(!cache); /* Logic error */
6088 len = cache->key.objectid + cache->key.offset - start;
6089 len = min(len, end + 1 - start);
6091 if (start < cache->last_byte_to_unpin) {
6092 len = min(len, cache->last_byte_to_unpin - start);
6093 if (return_free_space)
6094 btrfs_add_free_space(cache, start, len);
6098 space_info = cache->space_info;
6100 spin_lock(&space_info->lock);
6101 spin_lock(&cache->lock);
6102 cache->pinned -= len;
6103 space_info->bytes_pinned -= len;
6104 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6106 space_info->bytes_readonly += len;
6109 spin_unlock(&cache->lock);
6110 if (!readonly && global_rsv->space_info == space_info) {
6111 spin_lock(&global_rsv->lock);
6112 if (!global_rsv->full) {
6113 len = min(len, global_rsv->size -
6114 global_rsv->reserved);
6115 global_rsv->reserved += len;
6116 space_info->bytes_may_use += len;
6117 if (global_rsv->reserved >= global_rsv->size)
6118 global_rsv->full = 1;
6120 spin_unlock(&global_rsv->lock);
6122 spin_unlock(&space_info->lock);
6126 btrfs_put_block_group(cache);
6130 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6131 struct btrfs_root *root)
6133 struct btrfs_fs_info *fs_info = root->fs_info;
6134 struct extent_io_tree *unpin;
6142 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6143 unpin = &fs_info->freed_extents[1];
6145 unpin = &fs_info->freed_extents[0];
6148 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6149 ret = find_first_extent_bit(unpin, 0, &start, &end,
6150 EXTENT_DIRTY, NULL);
6152 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6156 if (btrfs_test_opt(root, DISCARD))
6157 ret = btrfs_discard_extent(root, start,
6158 end + 1 - start, NULL);
6160 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6161 unpin_extent_range(root, start, end, true);
6162 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6169 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6170 u64 owner, u64 root_objectid)
6172 struct btrfs_space_info *space_info;
6175 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6176 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6177 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6179 flags = BTRFS_BLOCK_GROUP_METADATA;
6181 flags = BTRFS_BLOCK_GROUP_DATA;
6184 space_info = __find_space_info(fs_info, flags);
6185 BUG_ON(!space_info); /* Logic bug */
6186 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6190 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6191 struct btrfs_root *root,
6192 struct btrfs_delayed_ref_node *node, u64 parent,
6193 u64 root_objectid, u64 owner_objectid,
6194 u64 owner_offset, int refs_to_drop,
6195 struct btrfs_delayed_extent_op *extent_op)
6197 struct btrfs_key key;
6198 struct btrfs_path *path;
6199 struct btrfs_fs_info *info = root->fs_info;
6200 struct btrfs_root *extent_root = info->extent_root;
6201 struct extent_buffer *leaf;
6202 struct btrfs_extent_item *ei;
6203 struct btrfs_extent_inline_ref *iref;
6206 int extent_slot = 0;
6207 int found_extent = 0;
6209 int no_quota = node->no_quota;
6212 u64 bytenr = node->bytenr;
6213 u64 num_bytes = node->num_bytes;
6215 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6218 if (!info->quota_enabled || !is_fstree(root_objectid))
6221 path = btrfs_alloc_path();
6226 path->leave_spinning = 1;
6228 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6229 BUG_ON(!is_data && refs_to_drop != 1);
6232 skinny_metadata = 0;
6234 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6235 bytenr, num_bytes, parent,
6236 root_objectid, owner_objectid,
6239 extent_slot = path->slots[0];
6240 while (extent_slot >= 0) {
6241 btrfs_item_key_to_cpu(path->nodes[0], &key,
6243 if (key.objectid != bytenr)
6245 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6246 key.offset == num_bytes) {
6250 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6251 key.offset == owner_objectid) {
6255 if (path->slots[0] - extent_slot > 5)
6259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6260 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6261 if (found_extent && item_size < sizeof(*ei))
6264 if (!found_extent) {
6266 ret = remove_extent_backref(trans, extent_root, path,
6268 is_data, &last_ref);
6270 btrfs_abort_transaction(trans, extent_root, ret);
6273 btrfs_release_path(path);
6274 path->leave_spinning = 1;
6276 key.objectid = bytenr;
6277 key.type = BTRFS_EXTENT_ITEM_KEY;
6278 key.offset = num_bytes;
6280 if (!is_data && skinny_metadata) {
6281 key.type = BTRFS_METADATA_ITEM_KEY;
6282 key.offset = owner_objectid;
6285 ret = btrfs_search_slot(trans, extent_root,
6287 if (ret > 0 && skinny_metadata && path->slots[0]) {
6289 * Couldn't find our skinny metadata item,
6290 * see if we have ye olde extent item.
6293 btrfs_item_key_to_cpu(path->nodes[0], &key,
6295 if (key.objectid == bytenr &&
6296 key.type == BTRFS_EXTENT_ITEM_KEY &&
6297 key.offset == num_bytes)
6301 if (ret > 0 && skinny_metadata) {
6302 skinny_metadata = false;
6303 key.objectid = bytenr;
6304 key.type = BTRFS_EXTENT_ITEM_KEY;
6305 key.offset = num_bytes;
6306 btrfs_release_path(path);
6307 ret = btrfs_search_slot(trans, extent_root,
6312 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6315 btrfs_print_leaf(extent_root,
6319 btrfs_abort_transaction(trans, extent_root, ret);
6322 extent_slot = path->slots[0];
6324 } else if (WARN_ON(ret == -ENOENT)) {
6325 btrfs_print_leaf(extent_root, path->nodes[0]);
6327 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6328 bytenr, parent, root_objectid, owner_objectid,
6330 btrfs_abort_transaction(trans, extent_root, ret);
6333 btrfs_abort_transaction(trans, extent_root, ret);
6337 leaf = path->nodes[0];
6338 item_size = btrfs_item_size_nr(leaf, extent_slot);
6339 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6340 if (item_size < sizeof(*ei)) {
6341 BUG_ON(found_extent || extent_slot != path->slots[0]);
6342 ret = convert_extent_item_v0(trans, extent_root, path,
6345 btrfs_abort_transaction(trans, extent_root, ret);
6349 btrfs_release_path(path);
6350 path->leave_spinning = 1;
6352 key.objectid = bytenr;
6353 key.type = BTRFS_EXTENT_ITEM_KEY;
6354 key.offset = num_bytes;
6356 ret = btrfs_search_slot(trans, extent_root, &key, path,
6359 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6361 btrfs_print_leaf(extent_root, path->nodes[0]);
6364 btrfs_abort_transaction(trans, extent_root, ret);
6368 extent_slot = path->slots[0];
6369 leaf = path->nodes[0];
6370 item_size = btrfs_item_size_nr(leaf, extent_slot);
6373 BUG_ON(item_size < sizeof(*ei));
6374 ei = btrfs_item_ptr(leaf, extent_slot,
6375 struct btrfs_extent_item);
6376 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6377 key.type == BTRFS_EXTENT_ITEM_KEY) {
6378 struct btrfs_tree_block_info *bi;
6379 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6380 bi = (struct btrfs_tree_block_info *)(ei + 1);
6381 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6384 refs = btrfs_extent_refs(leaf, ei);
6385 if (refs < refs_to_drop) {
6386 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6387 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6389 btrfs_abort_transaction(trans, extent_root, ret);
6392 refs -= refs_to_drop;
6396 __run_delayed_extent_op(extent_op, leaf, ei);
6398 * In the case of inline back ref, reference count will
6399 * be updated by remove_extent_backref
6402 BUG_ON(!found_extent);
6404 btrfs_set_extent_refs(leaf, ei, refs);
6405 btrfs_mark_buffer_dirty(leaf);
6408 ret = remove_extent_backref(trans, extent_root, path,
6410 is_data, &last_ref);
6412 btrfs_abort_transaction(trans, extent_root, ret);
6416 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6420 BUG_ON(is_data && refs_to_drop !=
6421 extent_data_ref_count(root, path, iref));
6423 BUG_ON(path->slots[0] != extent_slot);
6425 BUG_ON(path->slots[0] != extent_slot + 1);
6426 path->slots[0] = extent_slot;
6432 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6435 btrfs_abort_transaction(trans, extent_root, ret);
6438 btrfs_release_path(path);
6441 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6443 btrfs_abort_transaction(trans, extent_root, ret);
6448 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6450 btrfs_abort_transaction(trans, extent_root, ret);
6454 btrfs_release_path(path);
6457 btrfs_free_path(path);
6462 * when we free an block, it is possible (and likely) that we free the last
6463 * delayed ref for that extent as well. This searches the delayed ref tree for
6464 * a given extent, and if there are no other delayed refs to be processed, it
6465 * removes it from the tree.
6467 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6468 struct btrfs_root *root, u64 bytenr)
6470 struct btrfs_delayed_ref_head *head;
6471 struct btrfs_delayed_ref_root *delayed_refs;
6474 delayed_refs = &trans->transaction->delayed_refs;
6475 spin_lock(&delayed_refs->lock);
6476 head = btrfs_find_delayed_ref_head(trans, bytenr);
6478 goto out_delayed_unlock;
6480 spin_lock(&head->lock);
6481 if (!list_empty(&head->ref_list))
6484 if (head->extent_op) {
6485 if (!head->must_insert_reserved)
6487 btrfs_free_delayed_extent_op(head->extent_op);
6488 head->extent_op = NULL;
6492 * waiting for the lock here would deadlock. If someone else has it
6493 * locked they are already in the process of dropping it anyway
6495 if (!mutex_trylock(&head->mutex))
6499 * at this point we have a head with no other entries. Go
6500 * ahead and process it.
6502 head->node.in_tree = 0;
6503 rb_erase(&head->href_node, &delayed_refs->href_root);
6505 atomic_dec(&delayed_refs->num_entries);
6508 * we don't take a ref on the node because we're removing it from the
6509 * tree, so we just steal the ref the tree was holding.
6511 delayed_refs->num_heads--;
6512 if (head->processing == 0)
6513 delayed_refs->num_heads_ready--;
6514 head->processing = 0;
6515 spin_unlock(&head->lock);
6516 spin_unlock(&delayed_refs->lock);
6518 BUG_ON(head->extent_op);
6519 if (head->must_insert_reserved)
6522 mutex_unlock(&head->mutex);
6523 btrfs_put_delayed_ref(&head->node);
6526 spin_unlock(&head->lock);
6529 spin_unlock(&delayed_refs->lock);
6533 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6534 struct btrfs_root *root,
6535 struct extent_buffer *buf,
6536 u64 parent, int last_ref)
6541 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6542 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6543 buf->start, buf->len,
6544 parent, root->root_key.objectid,
6545 btrfs_header_level(buf),
6546 BTRFS_DROP_DELAYED_REF, NULL, 0);
6547 BUG_ON(ret); /* -ENOMEM */
6553 if (btrfs_header_generation(buf) == trans->transid) {
6554 struct btrfs_block_group_cache *cache;
6556 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6557 ret = check_ref_cleanup(trans, root, buf->start);
6562 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6564 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6565 pin_down_extent(root, cache, buf->start, buf->len, 1);
6566 btrfs_put_block_group(cache);
6570 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6572 btrfs_add_free_space(cache, buf->start, buf->len);
6573 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6574 btrfs_put_block_group(cache);
6575 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6580 add_pinned_bytes(root->fs_info, buf->len,
6581 btrfs_header_level(buf),
6582 root->root_key.objectid);
6585 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6588 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6591 /* Can return -ENOMEM */
6592 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6593 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6594 u64 owner, u64 offset, int no_quota)
6597 struct btrfs_fs_info *fs_info = root->fs_info;
6599 if (btrfs_test_is_dummy_root(root))
6602 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6605 * tree log blocks never actually go into the extent allocation
6606 * tree, just update pinning info and exit early.
6608 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6609 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6610 /* unlocks the pinned mutex */
6611 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6613 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6614 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6616 parent, root_objectid, (int)owner,
6617 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6619 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6621 parent, root_objectid, owner,
6622 offset, BTRFS_DROP_DELAYED_REF,
6629 * when we wait for progress in the block group caching, its because
6630 * our allocation attempt failed at least once. So, we must sleep
6631 * and let some progress happen before we try again.
6633 * This function will sleep at least once waiting for new free space to
6634 * show up, and then it will check the block group free space numbers
6635 * for our min num_bytes. Another option is to have it go ahead
6636 * and look in the rbtree for a free extent of a given size, but this
6639 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6640 * any of the information in this block group.
6642 static noinline void
6643 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6646 struct btrfs_caching_control *caching_ctl;
6648 caching_ctl = get_caching_control(cache);
6652 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6653 (cache->free_space_ctl->free_space >= num_bytes));
6655 put_caching_control(caching_ctl);
6659 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6661 struct btrfs_caching_control *caching_ctl;
6664 caching_ctl = get_caching_control(cache);
6666 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6668 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6669 if (cache->cached == BTRFS_CACHE_ERROR)
6671 put_caching_control(caching_ctl);
6675 int __get_raid_index(u64 flags)
6677 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6678 return BTRFS_RAID_RAID10;
6679 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6680 return BTRFS_RAID_RAID1;
6681 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6682 return BTRFS_RAID_DUP;
6683 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6684 return BTRFS_RAID_RAID0;
6685 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6686 return BTRFS_RAID_RAID5;
6687 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6688 return BTRFS_RAID_RAID6;
6690 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6693 int get_block_group_index(struct btrfs_block_group_cache *cache)
6695 return __get_raid_index(cache->flags);
6698 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6699 [BTRFS_RAID_RAID10] = "raid10",
6700 [BTRFS_RAID_RAID1] = "raid1",
6701 [BTRFS_RAID_DUP] = "dup",
6702 [BTRFS_RAID_RAID0] = "raid0",
6703 [BTRFS_RAID_SINGLE] = "single",
6704 [BTRFS_RAID_RAID5] = "raid5",
6705 [BTRFS_RAID_RAID6] = "raid6",
6708 static const char *get_raid_name(enum btrfs_raid_types type)
6710 if (type >= BTRFS_NR_RAID_TYPES)
6713 return btrfs_raid_type_names[type];
6716 enum btrfs_loop_type {
6717 LOOP_CACHING_NOWAIT = 0,
6718 LOOP_CACHING_WAIT = 1,
6719 LOOP_ALLOC_CHUNK = 2,
6720 LOOP_NO_EMPTY_SIZE = 3,
6724 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6728 down_read(&cache->data_rwsem);
6732 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6735 btrfs_get_block_group(cache);
6737 down_read(&cache->data_rwsem);
6740 static struct btrfs_block_group_cache *
6741 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6742 struct btrfs_free_cluster *cluster,
6745 struct btrfs_block_group_cache *used_bg;
6746 bool locked = false;
6748 spin_lock(&cluster->refill_lock);
6750 if (used_bg == cluster->block_group)
6753 up_read(&used_bg->data_rwsem);
6754 btrfs_put_block_group(used_bg);
6757 used_bg = cluster->block_group;
6761 if (used_bg == block_group)
6764 btrfs_get_block_group(used_bg);
6769 if (down_read_trylock(&used_bg->data_rwsem))
6772 spin_unlock(&cluster->refill_lock);
6773 down_read(&used_bg->data_rwsem);
6779 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6783 up_read(&cache->data_rwsem);
6784 btrfs_put_block_group(cache);
6788 * walks the btree of allocated extents and find a hole of a given size.
6789 * The key ins is changed to record the hole:
6790 * ins->objectid == start position
6791 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6792 * ins->offset == the size of the hole.
6793 * Any available blocks before search_start are skipped.
6795 * If there is no suitable free space, we will record the max size of
6796 * the free space extent currently.
6798 static noinline int find_free_extent(struct btrfs_root *orig_root,
6799 u64 num_bytes, u64 empty_size,
6800 u64 hint_byte, struct btrfs_key *ins,
6801 u64 flags, int delalloc)
6804 struct btrfs_root *root = orig_root->fs_info->extent_root;
6805 struct btrfs_free_cluster *last_ptr = NULL;
6806 struct btrfs_block_group_cache *block_group = NULL;
6807 u64 search_start = 0;
6808 u64 max_extent_size = 0;
6809 int empty_cluster = 2 * 1024 * 1024;
6810 struct btrfs_space_info *space_info;
6812 int index = __get_raid_index(flags);
6813 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6814 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6815 bool failed_cluster_refill = false;
6816 bool failed_alloc = false;
6817 bool use_cluster = true;
6818 bool have_caching_bg = false;
6820 WARN_ON(num_bytes < root->sectorsize);
6821 ins->type = BTRFS_EXTENT_ITEM_KEY;
6825 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6827 space_info = __find_space_info(root->fs_info, flags);
6829 btrfs_err(root->fs_info, "No space info for %llu", flags);
6834 * If the space info is for both data and metadata it means we have a
6835 * small filesystem and we can't use the clustering stuff.
6837 if (btrfs_mixed_space_info(space_info))
6838 use_cluster = false;
6840 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6841 last_ptr = &root->fs_info->meta_alloc_cluster;
6842 if (!btrfs_test_opt(root, SSD))
6843 empty_cluster = 64 * 1024;
6846 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6847 btrfs_test_opt(root, SSD)) {
6848 last_ptr = &root->fs_info->data_alloc_cluster;
6852 spin_lock(&last_ptr->lock);
6853 if (last_ptr->block_group)
6854 hint_byte = last_ptr->window_start;
6855 spin_unlock(&last_ptr->lock);
6858 search_start = max(search_start, first_logical_byte(root, 0));
6859 search_start = max(search_start, hint_byte);
6864 if (search_start == hint_byte) {
6865 block_group = btrfs_lookup_block_group(root->fs_info,
6868 * we don't want to use the block group if it doesn't match our
6869 * allocation bits, or if its not cached.
6871 * However if we are re-searching with an ideal block group
6872 * picked out then we don't care that the block group is cached.
6874 if (block_group && block_group_bits(block_group, flags) &&
6875 block_group->cached != BTRFS_CACHE_NO) {
6876 down_read(&space_info->groups_sem);
6877 if (list_empty(&block_group->list) ||
6880 * someone is removing this block group,
6881 * we can't jump into the have_block_group
6882 * target because our list pointers are not
6885 btrfs_put_block_group(block_group);
6886 up_read(&space_info->groups_sem);
6888 index = get_block_group_index(block_group);
6889 btrfs_lock_block_group(block_group, delalloc);
6890 goto have_block_group;
6892 } else if (block_group) {
6893 btrfs_put_block_group(block_group);
6897 have_caching_bg = false;
6898 down_read(&space_info->groups_sem);
6899 list_for_each_entry(block_group, &space_info->block_groups[index],
6904 btrfs_grab_block_group(block_group, delalloc);
6905 search_start = block_group->key.objectid;
6908 * this can happen if we end up cycling through all the
6909 * raid types, but we want to make sure we only allocate
6910 * for the proper type.
6912 if (!block_group_bits(block_group, flags)) {
6913 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6914 BTRFS_BLOCK_GROUP_RAID1 |
6915 BTRFS_BLOCK_GROUP_RAID5 |
6916 BTRFS_BLOCK_GROUP_RAID6 |
6917 BTRFS_BLOCK_GROUP_RAID10;
6920 * if they asked for extra copies and this block group
6921 * doesn't provide them, bail. This does allow us to
6922 * fill raid0 from raid1.
6924 if ((flags & extra) && !(block_group->flags & extra))
6929 cached = block_group_cache_done(block_group);
6930 if (unlikely(!cached)) {
6931 ret = cache_block_group(block_group, 0);
6936 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6938 if (unlikely(block_group->ro))
6942 * Ok we want to try and use the cluster allocator, so
6946 struct btrfs_block_group_cache *used_block_group;
6947 unsigned long aligned_cluster;
6949 * the refill lock keeps out other
6950 * people trying to start a new cluster
6952 used_block_group = btrfs_lock_cluster(block_group,
6955 if (!used_block_group)
6956 goto refill_cluster;
6958 if (used_block_group != block_group &&
6959 (used_block_group->ro ||
6960 !block_group_bits(used_block_group, flags)))
6961 goto release_cluster;
6963 offset = btrfs_alloc_from_cluster(used_block_group,
6966 used_block_group->key.objectid,
6969 /* we have a block, we're done */
6970 spin_unlock(&last_ptr->refill_lock);
6971 trace_btrfs_reserve_extent_cluster(root,
6973 search_start, num_bytes);
6974 if (used_block_group != block_group) {
6975 btrfs_release_block_group(block_group,
6977 block_group = used_block_group;
6982 WARN_ON(last_ptr->block_group != used_block_group);
6984 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6985 * set up a new clusters, so lets just skip it
6986 * and let the allocator find whatever block
6987 * it can find. If we reach this point, we
6988 * will have tried the cluster allocator
6989 * plenty of times and not have found
6990 * anything, so we are likely way too
6991 * fragmented for the clustering stuff to find
6994 * However, if the cluster is taken from the
6995 * current block group, release the cluster
6996 * first, so that we stand a better chance of
6997 * succeeding in the unclustered
6999 if (loop >= LOOP_NO_EMPTY_SIZE &&
7000 used_block_group != block_group) {
7001 spin_unlock(&last_ptr->refill_lock);
7002 btrfs_release_block_group(used_block_group,
7004 goto unclustered_alloc;
7008 * this cluster didn't work out, free it and
7011 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7013 if (used_block_group != block_group)
7014 btrfs_release_block_group(used_block_group,
7017 if (loop >= LOOP_NO_EMPTY_SIZE) {
7018 spin_unlock(&last_ptr->refill_lock);
7019 goto unclustered_alloc;
7022 aligned_cluster = max_t(unsigned long,
7023 empty_cluster + empty_size,
7024 block_group->full_stripe_len);
7026 /* allocate a cluster in this block group */
7027 ret = btrfs_find_space_cluster(root, block_group,
7028 last_ptr, search_start,
7033 * now pull our allocation out of this
7036 offset = btrfs_alloc_from_cluster(block_group,
7042 /* we found one, proceed */
7043 spin_unlock(&last_ptr->refill_lock);
7044 trace_btrfs_reserve_extent_cluster(root,
7045 block_group, search_start,
7049 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7050 && !failed_cluster_refill) {
7051 spin_unlock(&last_ptr->refill_lock);
7053 failed_cluster_refill = true;
7054 wait_block_group_cache_progress(block_group,
7055 num_bytes + empty_cluster + empty_size);
7056 goto have_block_group;
7060 * at this point we either didn't find a cluster
7061 * or we weren't able to allocate a block from our
7062 * cluster. Free the cluster we've been trying
7063 * to use, and go to the next block group
7065 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7066 spin_unlock(&last_ptr->refill_lock);
7071 spin_lock(&block_group->free_space_ctl->tree_lock);
7073 block_group->free_space_ctl->free_space <
7074 num_bytes + empty_cluster + empty_size) {
7075 if (block_group->free_space_ctl->free_space >
7078 block_group->free_space_ctl->free_space;
7079 spin_unlock(&block_group->free_space_ctl->tree_lock);
7082 spin_unlock(&block_group->free_space_ctl->tree_lock);
7084 offset = btrfs_find_space_for_alloc(block_group, search_start,
7085 num_bytes, empty_size,
7088 * If we didn't find a chunk, and we haven't failed on this
7089 * block group before, and this block group is in the middle of
7090 * caching and we are ok with waiting, then go ahead and wait
7091 * for progress to be made, and set failed_alloc to true.
7093 * If failed_alloc is true then we've already waited on this
7094 * block group once and should move on to the next block group.
7096 if (!offset && !failed_alloc && !cached &&
7097 loop > LOOP_CACHING_NOWAIT) {
7098 wait_block_group_cache_progress(block_group,
7099 num_bytes + empty_size);
7100 failed_alloc = true;
7101 goto have_block_group;
7102 } else if (!offset) {
7104 have_caching_bg = true;
7108 search_start = ALIGN(offset, root->stripesize);
7110 /* move on to the next group */
7111 if (search_start + num_bytes >
7112 block_group->key.objectid + block_group->key.offset) {
7113 btrfs_add_free_space(block_group, offset, num_bytes);
7117 if (offset < search_start)
7118 btrfs_add_free_space(block_group, offset,
7119 search_start - offset);
7120 BUG_ON(offset > search_start);
7122 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7123 alloc_type, delalloc);
7124 if (ret == -EAGAIN) {
7125 btrfs_add_free_space(block_group, offset, num_bytes);
7129 /* we are all good, lets return */
7130 ins->objectid = search_start;
7131 ins->offset = num_bytes;
7133 trace_btrfs_reserve_extent(orig_root, block_group,
7134 search_start, num_bytes);
7135 btrfs_release_block_group(block_group, delalloc);
7138 failed_cluster_refill = false;
7139 failed_alloc = false;
7140 BUG_ON(index != get_block_group_index(block_group));
7141 btrfs_release_block_group(block_group, delalloc);
7143 up_read(&space_info->groups_sem);
7145 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7148 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7152 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7153 * caching kthreads as we move along
7154 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7155 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7156 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7159 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7162 if (loop == LOOP_ALLOC_CHUNK) {
7163 struct btrfs_trans_handle *trans;
7166 trans = current->journal_info;
7170 trans = btrfs_join_transaction(root);
7172 if (IS_ERR(trans)) {
7173 ret = PTR_ERR(trans);
7177 ret = do_chunk_alloc(trans, root, flags,
7180 * Do not bail out on ENOSPC since we
7181 * can do more things.
7183 if (ret < 0 && ret != -ENOSPC)
7184 btrfs_abort_transaction(trans,
7189 btrfs_end_transaction(trans, root);
7194 if (loop == LOOP_NO_EMPTY_SIZE) {
7200 } else if (!ins->objectid) {
7202 } else if (ins->objectid) {
7207 ins->offset = max_extent_size;
7211 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7212 int dump_block_groups)
7214 struct btrfs_block_group_cache *cache;
7217 spin_lock(&info->lock);
7218 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7220 info->total_bytes - info->bytes_used - info->bytes_pinned -
7221 info->bytes_reserved - info->bytes_readonly,
7222 (info->full) ? "" : "not ");
7223 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7224 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7225 info->total_bytes, info->bytes_used, info->bytes_pinned,
7226 info->bytes_reserved, info->bytes_may_use,
7227 info->bytes_readonly);
7228 spin_unlock(&info->lock);
7230 if (!dump_block_groups)
7233 down_read(&info->groups_sem);
7235 list_for_each_entry(cache, &info->block_groups[index], list) {
7236 spin_lock(&cache->lock);
7237 printk(KERN_INFO "BTRFS: "
7238 "block group %llu has %llu bytes, "
7239 "%llu used %llu pinned %llu reserved %s\n",
7240 cache->key.objectid, cache->key.offset,
7241 btrfs_block_group_used(&cache->item), cache->pinned,
7242 cache->reserved, cache->ro ? "[readonly]" : "");
7243 btrfs_dump_free_space(cache, bytes);
7244 spin_unlock(&cache->lock);
7246 if (++index < BTRFS_NR_RAID_TYPES)
7248 up_read(&info->groups_sem);
7251 int btrfs_reserve_extent(struct btrfs_root *root,
7252 u64 num_bytes, u64 min_alloc_size,
7253 u64 empty_size, u64 hint_byte,
7254 struct btrfs_key *ins, int is_data, int delalloc)
7256 bool final_tried = false;
7260 flags = btrfs_get_alloc_profile(root, is_data);
7262 WARN_ON(num_bytes < root->sectorsize);
7263 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7266 if (ret == -ENOSPC) {
7267 if (!final_tried && ins->offset) {
7268 num_bytes = min(num_bytes >> 1, ins->offset);
7269 num_bytes = round_down(num_bytes, root->sectorsize);
7270 num_bytes = max(num_bytes, min_alloc_size);
7271 if (num_bytes == min_alloc_size)
7274 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7275 struct btrfs_space_info *sinfo;
7277 sinfo = __find_space_info(root->fs_info, flags);
7278 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7281 dump_space_info(sinfo, num_bytes, 1);
7288 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7290 int pin, int delalloc)
7292 struct btrfs_block_group_cache *cache;
7295 cache = btrfs_lookup_block_group(root->fs_info, start);
7297 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7303 pin_down_extent(root, cache, start, len, 1);
7305 if (btrfs_test_opt(root, DISCARD))
7306 ret = btrfs_discard_extent(root, start, len, NULL);
7307 btrfs_add_free_space(cache, start, len);
7308 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7311 btrfs_put_block_group(cache);
7313 trace_btrfs_reserved_extent_free(root, start, len);
7318 int btrfs_free_reserved_extent(struct btrfs_root *root,
7319 u64 start, u64 len, int delalloc)
7321 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7324 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7327 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7330 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7331 struct btrfs_root *root,
7332 u64 parent, u64 root_objectid,
7333 u64 flags, u64 owner, u64 offset,
7334 struct btrfs_key *ins, int ref_mod)
7337 struct btrfs_fs_info *fs_info = root->fs_info;
7338 struct btrfs_extent_item *extent_item;
7339 struct btrfs_extent_inline_ref *iref;
7340 struct btrfs_path *path;
7341 struct extent_buffer *leaf;
7346 type = BTRFS_SHARED_DATA_REF_KEY;
7348 type = BTRFS_EXTENT_DATA_REF_KEY;
7350 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7352 path = btrfs_alloc_path();
7356 path->leave_spinning = 1;
7357 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7360 btrfs_free_path(path);
7364 leaf = path->nodes[0];
7365 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7366 struct btrfs_extent_item);
7367 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7368 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7369 btrfs_set_extent_flags(leaf, extent_item,
7370 flags | BTRFS_EXTENT_FLAG_DATA);
7372 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7373 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7375 struct btrfs_shared_data_ref *ref;
7376 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7377 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7378 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7380 struct btrfs_extent_data_ref *ref;
7381 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7382 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7383 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7384 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7385 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7388 btrfs_mark_buffer_dirty(path->nodes[0]);
7389 btrfs_free_path(path);
7391 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7392 if (ret) { /* -ENOENT, logic error */
7393 btrfs_err(fs_info, "update block group failed for %llu %llu",
7394 ins->objectid, ins->offset);
7397 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7401 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7402 struct btrfs_root *root,
7403 u64 parent, u64 root_objectid,
7404 u64 flags, struct btrfs_disk_key *key,
7405 int level, struct btrfs_key *ins,
7409 struct btrfs_fs_info *fs_info = root->fs_info;
7410 struct btrfs_extent_item *extent_item;
7411 struct btrfs_tree_block_info *block_info;
7412 struct btrfs_extent_inline_ref *iref;
7413 struct btrfs_path *path;
7414 struct extent_buffer *leaf;
7415 u32 size = sizeof(*extent_item) + sizeof(*iref);
7416 u64 num_bytes = ins->offset;
7417 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7420 if (!skinny_metadata)
7421 size += sizeof(*block_info);
7423 path = btrfs_alloc_path();
7425 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7430 path->leave_spinning = 1;
7431 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7434 btrfs_free_path(path);
7435 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7440 leaf = path->nodes[0];
7441 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7442 struct btrfs_extent_item);
7443 btrfs_set_extent_refs(leaf, extent_item, 1);
7444 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7445 btrfs_set_extent_flags(leaf, extent_item,
7446 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7448 if (skinny_metadata) {
7449 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7450 num_bytes = root->nodesize;
7452 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7453 btrfs_set_tree_block_key(leaf, block_info, key);
7454 btrfs_set_tree_block_level(leaf, block_info, level);
7455 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7459 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7460 btrfs_set_extent_inline_ref_type(leaf, iref,
7461 BTRFS_SHARED_BLOCK_REF_KEY);
7462 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7464 btrfs_set_extent_inline_ref_type(leaf, iref,
7465 BTRFS_TREE_BLOCK_REF_KEY);
7466 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7469 btrfs_mark_buffer_dirty(leaf);
7470 btrfs_free_path(path);
7472 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7474 if (ret) { /* -ENOENT, logic error */
7475 btrfs_err(fs_info, "update block group failed for %llu %llu",
7476 ins->objectid, ins->offset);
7480 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7484 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7485 struct btrfs_root *root,
7486 u64 root_objectid, u64 owner,
7487 u64 offset, struct btrfs_key *ins)
7491 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7493 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7495 root_objectid, owner, offset,
7496 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7501 * this is used by the tree logging recovery code. It records that
7502 * an extent has been allocated and makes sure to clear the free
7503 * space cache bits as well
7505 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7506 struct btrfs_root *root,
7507 u64 root_objectid, u64 owner, u64 offset,
7508 struct btrfs_key *ins)
7511 struct btrfs_block_group_cache *block_group;
7514 * Mixed block groups will exclude before processing the log so we only
7515 * need to do the exlude dance if this fs isn't mixed.
7517 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7518 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7523 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7527 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7528 RESERVE_ALLOC_NO_ACCOUNT, 0);
7529 BUG_ON(ret); /* logic error */
7530 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7531 0, owner, offset, ins, 1);
7532 btrfs_put_block_group(block_group);
7536 static struct extent_buffer *
7537 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7538 u64 bytenr, int level)
7540 struct extent_buffer *buf;
7542 buf = btrfs_find_create_tree_block(root, bytenr);
7544 return ERR_PTR(-ENOMEM);
7545 btrfs_set_header_generation(buf, trans->transid);
7546 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7547 btrfs_tree_lock(buf);
7548 clean_tree_block(trans, root->fs_info, buf);
7549 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7551 btrfs_set_lock_blocking(buf);
7552 btrfs_set_buffer_uptodate(buf);
7554 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7555 buf->log_index = root->log_transid % 2;
7557 * we allow two log transactions at a time, use different
7558 * EXENT bit to differentiate dirty pages.
7560 if (buf->log_index == 0)
7561 set_extent_dirty(&root->dirty_log_pages, buf->start,
7562 buf->start + buf->len - 1, GFP_NOFS);
7564 set_extent_new(&root->dirty_log_pages, buf->start,
7565 buf->start + buf->len - 1, GFP_NOFS);
7567 buf->log_index = -1;
7568 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7569 buf->start + buf->len - 1, GFP_NOFS);
7571 trans->blocks_used++;
7572 /* this returns a buffer locked for blocking */
7576 static struct btrfs_block_rsv *
7577 use_block_rsv(struct btrfs_trans_handle *trans,
7578 struct btrfs_root *root, u32 blocksize)
7580 struct btrfs_block_rsv *block_rsv;
7581 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7583 bool global_updated = false;
7585 block_rsv = get_block_rsv(trans, root);
7587 if (unlikely(block_rsv->size == 0))
7590 ret = block_rsv_use_bytes(block_rsv, blocksize);
7594 if (block_rsv->failfast)
7595 return ERR_PTR(ret);
7597 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7598 global_updated = true;
7599 update_global_block_rsv(root->fs_info);
7603 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7604 static DEFINE_RATELIMIT_STATE(_rs,
7605 DEFAULT_RATELIMIT_INTERVAL * 10,
7606 /*DEFAULT_RATELIMIT_BURST*/ 1);
7607 if (__ratelimit(&_rs))
7609 "BTRFS: block rsv returned %d\n", ret);
7612 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7613 BTRFS_RESERVE_NO_FLUSH);
7617 * If we couldn't reserve metadata bytes try and use some from
7618 * the global reserve if its space type is the same as the global
7621 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7622 block_rsv->space_info == global_rsv->space_info) {
7623 ret = block_rsv_use_bytes(global_rsv, blocksize);
7627 return ERR_PTR(ret);
7630 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7631 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7633 block_rsv_add_bytes(block_rsv, blocksize, 0);
7634 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7638 * finds a free extent and does all the dirty work required for allocation
7639 * returns the key for the extent through ins, and a tree buffer for
7640 * the first block of the extent through buf.
7642 * returns the tree buffer or an ERR_PTR on error.
7644 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7645 struct btrfs_root *root,
7646 u64 parent, u64 root_objectid,
7647 struct btrfs_disk_key *key, int level,
7648 u64 hint, u64 empty_size)
7650 struct btrfs_key ins;
7651 struct btrfs_block_rsv *block_rsv;
7652 struct extent_buffer *buf;
7653 struct btrfs_delayed_extent_op *extent_op;
7656 u32 blocksize = root->nodesize;
7657 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7660 if (btrfs_test_is_dummy_root(root)) {
7661 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7664 root->alloc_bytenr += blocksize;
7668 block_rsv = use_block_rsv(trans, root, blocksize);
7669 if (IS_ERR(block_rsv))
7670 return ERR_CAST(block_rsv);
7672 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7673 empty_size, hint, &ins, 0, 0);
7677 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7680 goto out_free_reserved;
7683 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7685 parent = ins.objectid;
7686 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7690 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7691 extent_op = btrfs_alloc_delayed_extent_op();
7697 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7699 memset(&extent_op->key, 0, sizeof(extent_op->key));
7700 extent_op->flags_to_set = flags;
7701 if (skinny_metadata)
7702 extent_op->update_key = 0;
7704 extent_op->update_key = 1;
7705 extent_op->update_flags = 1;
7706 extent_op->is_data = 0;
7707 extent_op->level = level;
7709 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7710 ins.objectid, ins.offset,
7711 parent, root_objectid, level,
7712 BTRFS_ADD_DELAYED_EXTENT,
7715 goto out_free_delayed;
7720 btrfs_free_delayed_extent_op(extent_op);
7722 free_extent_buffer(buf);
7724 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7726 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7727 return ERR_PTR(ret);
7730 struct walk_control {
7731 u64 refs[BTRFS_MAX_LEVEL];
7732 u64 flags[BTRFS_MAX_LEVEL];
7733 struct btrfs_key update_progress;
7744 #define DROP_REFERENCE 1
7745 #define UPDATE_BACKREF 2
7747 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7748 struct btrfs_root *root,
7749 struct walk_control *wc,
7750 struct btrfs_path *path)
7758 struct btrfs_key key;
7759 struct extent_buffer *eb;
7764 if (path->slots[wc->level] < wc->reada_slot) {
7765 wc->reada_count = wc->reada_count * 2 / 3;
7766 wc->reada_count = max(wc->reada_count, 2);
7768 wc->reada_count = wc->reada_count * 3 / 2;
7769 wc->reada_count = min_t(int, wc->reada_count,
7770 BTRFS_NODEPTRS_PER_BLOCK(root));
7773 eb = path->nodes[wc->level];
7774 nritems = btrfs_header_nritems(eb);
7775 blocksize = root->nodesize;
7777 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7778 if (nread >= wc->reada_count)
7782 bytenr = btrfs_node_blockptr(eb, slot);
7783 generation = btrfs_node_ptr_generation(eb, slot);
7785 if (slot == path->slots[wc->level])
7788 if (wc->stage == UPDATE_BACKREF &&
7789 generation <= root->root_key.offset)
7792 /* We don't lock the tree block, it's OK to be racy here */
7793 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7794 wc->level - 1, 1, &refs,
7796 /* We don't care about errors in readahead. */
7801 if (wc->stage == DROP_REFERENCE) {
7805 if (wc->level == 1 &&
7806 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7808 if (!wc->update_ref ||
7809 generation <= root->root_key.offset)
7811 btrfs_node_key_to_cpu(eb, &key, slot);
7812 ret = btrfs_comp_cpu_keys(&key,
7813 &wc->update_progress);
7817 if (wc->level == 1 &&
7818 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7822 readahead_tree_block(root, bytenr);
7825 wc->reada_slot = slot;
7829 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7830 * for later qgroup accounting.
7832 * Current, this function does nothing.
7834 static int account_leaf_items(struct btrfs_trans_handle *trans,
7835 struct btrfs_root *root,
7836 struct extent_buffer *eb)
7838 int nr = btrfs_header_nritems(eb);
7840 struct btrfs_key key;
7841 struct btrfs_file_extent_item *fi;
7842 u64 bytenr, num_bytes;
7844 for (i = 0; i < nr; i++) {
7845 btrfs_item_key_to_cpu(eb, &key, i);
7847 if (key.type != BTRFS_EXTENT_DATA_KEY)
7850 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7851 /* filter out non qgroup-accountable extents */
7852 extent_type = btrfs_file_extent_type(eb, fi);
7854 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7857 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7861 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7867 * Walk up the tree from the bottom, freeing leaves and any interior
7868 * nodes which have had all slots visited. If a node (leaf or
7869 * interior) is freed, the node above it will have it's slot
7870 * incremented. The root node will never be freed.
7872 * At the end of this function, we should have a path which has all
7873 * slots incremented to the next position for a search. If we need to
7874 * read a new node it will be NULL and the node above it will have the
7875 * correct slot selected for a later read.
7877 * If we increment the root nodes slot counter past the number of
7878 * elements, 1 is returned to signal completion of the search.
7880 static int adjust_slots_upwards(struct btrfs_root *root,
7881 struct btrfs_path *path, int root_level)
7885 struct extent_buffer *eb;
7887 if (root_level == 0)
7890 while (level <= root_level) {
7891 eb = path->nodes[level];
7892 nr = btrfs_header_nritems(eb);
7893 path->slots[level]++;
7894 slot = path->slots[level];
7895 if (slot >= nr || level == 0) {
7897 * Don't free the root - we will detect this
7898 * condition after our loop and return a
7899 * positive value for caller to stop walking the tree.
7901 if (level != root_level) {
7902 btrfs_tree_unlock_rw(eb, path->locks[level]);
7903 path->locks[level] = 0;
7905 free_extent_buffer(eb);
7906 path->nodes[level] = NULL;
7907 path->slots[level] = 0;
7911 * We have a valid slot to walk back down
7912 * from. Stop here so caller can process these
7921 eb = path->nodes[root_level];
7922 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7929 * root_eb is the subtree root and is locked before this function is called.
7930 * TODO: Modify this function to mark all (including complete shared node)
7931 * to dirty_extent_root to allow it get accounted in qgroup.
7933 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7934 struct btrfs_root *root,
7935 struct extent_buffer *root_eb,
7941 struct extent_buffer *eb = root_eb;
7942 struct btrfs_path *path = NULL;
7944 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7945 BUG_ON(root_eb == NULL);
7947 if (!root->fs_info->quota_enabled)
7950 if (!extent_buffer_uptodate(root_eb)) {
7951 ret = btrfs_read_buffer(root_eb, root_gen);
7956 if (root_level == 0) {
7957 ret = account_leaf_items(trans, root, root_eb);
7961 path = btrfs_alloc_path();
7966 * Walk down the tree. Missing extent blocks are filled in as
7967 * we go. Metadata is accounted every time we read a new
7970 * When we reach a leaf, we account for file extent items in it,
7971 * walk back up the tree (adjusting slot pointers as we go)
7972 * and restart the search process.
7974 extent_buffer_get(root_eb); /* For path */
7975 path->nodes[root_level] = root_eb;
7976 path->slots[root_level] = 0;
7977 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7980 while (level >= 0) {
7981 if (path->nodes[level] == NULL) {
7986 /* We need to get child blockptr/gen from
7987 * parent before we can read it. */
7988 eb = path->nodes[level + 1];
7989 parent_slot = path->slots[level + 1];
7990 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7991 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7993 eb = read_tree_block(root, child_bytenr, child_gen);
7997 } else if (!extent_buffer_uptodate(eb)) {
7998 free_extent_buffer(eb);
8003 path->nodes[level] = eb;
8004 path->slots[level] = 0;
8006 btrfs_tree_read_lock(eb);
8007 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8008 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8012 ret = account_leaf_items(trans, root, path->nodes[level]);
8016 /* Nonzero return here means we completed our search */
8017 ret = adjust_slots_upwards(root, path, root_level);
8021 /* Restart search with new slots */
8030 btrfs_free_path(path);
8036 * helper to process tree block while walking down the tree.
8038 * when wc->stage == UPDATE_BACKREF, this function updates
8039 * back refs for pointers in the block.
8041 * NOTE: return value 1 means we should stop walking down.
8043 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8044 struct btrfs_root *root,
8045 struct btrfs_path *path,
8046 struct walk_control *wc, int lookup_info)
8048 int level = wc->level;
8049 struct extent_buffer *eb = path->nodes[level];
8050 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8053 if (wc->stage == UPDATE_BACKREF &&
8054 btrfs_header_owner(eb) != root->root_key.objectid)
8058 * when reference count of tree block is 1, it won't increase
8059 * again. once full backref flag is set, we never clear it.
8062 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8063 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8064 BUG_ON(!path->locks[level]);
8065 ret = btrfs_lookup_extent_info(trans, root,
8066 eb->start, level, 1,
8069 BUG_ON(ret == -ENOMEM);
8072 BUG_ON(wc->refs[level] == 0);
8075 if (wc->stage == DROP_REFERENCE) {
8076 if (wc->refs[level] > 1)
8079 if (path->locks[level] && !wc->keep_locks) {
8080 btrfs_tree_unlock_rw(eb, path->locks[level]);
8081 path->locks[level] = 0;
8086 /* wc->stage == UPDATE_BACKREF */
8087 if (!(wc->flags[level] & flag)) {
8088 BUG_ON(!path->locks[level]);
8089 ret = btrfs_inc_ref(trans, root, eb, 1);
8090 BUG_ON(ret); /* -ENOMEM */
8091 ret = btrfs_dec_ref(trans, root, eb, 0);
8092 BUG_ON(ret); /* -ENOMEM */
8093 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8095 btrfs_header_level(eb), 0);
8096 BUG_ON(ret); /* -ENOMEM */
8097 wc->flags[level] |= flag;
8101 * the block is shared by multiple trees, so it's not good to
8102 * keep the tree lock
8104 if (path->locks[level] && level > 0) {
8105 btrfs_tree_unlock_rw(eb, path->locks[level]);
8106 path->locks[level] = 0;
8112 * helper to process tree block pointer.
8114 * when wc->stage == DROP_REFERENCE, this function checks
8115 * reference count of the block pointed to. if the block
8116 * is shared and we need update back refs for the subtree
8117 * rooted at the block, this function changes wc->stage to
8118 * UPDATE_BACKREF. if the block is shared and there is no
8119 * need to update back, this function drops the reference
8122 * NOTE: return value 1 means we should stop walking down.
8124 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8125 struct btrfs_root *root,
8126 struct btrfs_path *path,
8127 struct walk_control *wc, int *lookup_info)
8133 struct btrfs_key key;
8134 struct extent_buffer *next;
8135 int level = wc->level;
8138 bool need_account = false;
8140 generation = btrfs_node_ptr_generation(path->nodes[level],
8141 path->slots[level]);
8143 * if the lower level block was created before the snapshot
8144 * was created, we know there is no need to update back refs
8147 if (wc->stage == UPDATE_BACKREF &&
8148 generation <= root->root_key.offset) {
8153 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8154 blocksize = root->nodesize;
8156 next = btrfs_find_tree_block(root->fs_info, bytenr);
8158 next = btrfs_find_create_tree_block(root, bytenr);
8161 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8165 btrfs_tree_lock(next);
8166 btrfs_set_lock_blocking(next);
8168 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8169 &wc->refs[level - 1],
8170 &wc->flags[level - 1]);
8172 btrfs_tree_unlock(next);
8176 if (unlikely(wc->refs[level - 1] == 0)) {
8177 btrfs_err(root->fs_info, "Missing references.");
8182 if (wc->stage == DROP_REFERENCE) {
8183 if (wc->refs[level - 1] > 1) {
8184 need_account = true;
8186 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8189 if (!wc->update_ref ||
8190 generation <= root->root_key.offset)
8193 btrfs_node_key_to_cpu(path->nodes[level], &key,
8194 path->slots[level]);
8195 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8199 wc->stage = UPDATE_BACKREF;
8200 wc->shared_level = level - 1;
8204 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8208 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8209 btrfs_tree_unlock(next);
8210 free_extent_buffer(next);
8216 if (reada && level == 1)
8217 reada_walk_down(trans, root, wc, path);
8218 next = read_tree_block(root, bytenr, generation);
8220 return PTR_ERR(next);
8221 } else if (!extent_buffer_uptodate(next)) {
8222 free_extent_buffer(next);
8225 btrfs_tree_lock(next);
8226 btrfs_set_lock_blocking(next);
8230 BUG_ON(level != btrfs_header_level(next));
8231 path->nodes[level] = next;
8232 path->slots[level] = 0;
8233 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8239 wc->refs[level - 1] = 0;
8240 wc->flags[level - 1] = 0;
8241 if (wc->stage == DROP_REFERENCE) {
8242 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8243 parent = path->nodes[level]->start;
8245 BUG_ON(root->root_key.objectid !=
8246 btrfs_header_owner(path->nodes[level]));
8251 ret = account_shared_subtree(trans, root, next,
8252 generation, level - 1);
8254 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8255 "%d accounting shared subtree. Quota "
8256 "is out of sync, rescan required.\n",
8257 root->fs_info->sb->s_id, ret);
8260 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8261 root->root_key.objectid, level - 1, 0, 0);
8262 BUG_ON(ret); /* -ENOMEM */
8264 btrfs_tree_unlock(next);
8265 free_extent_buffer(next);
8271 * helper to process tree block while walking up the tree.
8273 * when wc->stage == DROP_REFERENCE, this function drops
8274 * reference count on the block.
8276 * when wc->stage == UPDATE_BACKREF, this function changes
8277 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8278 * to UPDATE_BACKREF previously while processing the block.
8280 * NOTE: return value 1 means we should stop walking up.
8282 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8283 struct btrfs_root *root,
8284 struct btrfs_path *path,
8285 struct walk_control *wc)
8288 int level = wc->level;
8289 struct extent_buffer *eb = path->nodes[level];
8292 if (wc->stage == UPDATE_BACKREF) {
8293 BUG_ON(wc->shared_level < level);
8294 if (level < wc->shared_level)
8297 ret = find_next_key(path, level + 1, &wc->update_progress);
8301 wc->stage = DROP_REFERENCE;
8302 wc->shared_level = -1;
8303 path->slots[level] = 0;
8306 * check reference count again if the block isn't locked.
8307 * we should start walking down the tree again if reference
8310 if (!path->locks[level]) {
8312 btrfs_tree_lock(eb);
8313 btrfs_set_lock_blocking(eb);
8314 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8316 ret = btrfs_lookup_extent_info(trans, root,
8317 eb->start, level, 1,
8321 btrfs_tree_unlock_rw(eb, path->locks[level]);
8322 path->locks[level] = 0;
8325 BUG_ON(wc->refs[level] == 0);
8326 if (wc->refs[level] == 1) {
8327 btrfs_tree_unlock_rw(eb, path->locks[level]);
8328 path->locks[level] = 0;
8334 /* wc->stage == DROP_REFERENCE */
8335 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8337 if (wc->refs[level] == 1) {
8339 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8340 ret = btrfs_dec_ref(trans, root, eb, 1);
8342 ret = btrfs_dec_ref(trans, root, eb, 0);
8343 BUG_ON(ret); /* -ENOMEM */
8344 ret = account_leaf_items(trans, root, eb);
8346 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8347 "%d accounting leaf items. Quota "
8348 "is out of sync, rescan required.\n",
8349 root->fs_info->sb->s_id, ret);
8352 /* make block locked assertion in clean_tree_block happy */
8353 if (!path->locks[level] &&
8354 btrfs_header_generation(eb) == trans->transid) {
8355 btrfs_tree_lock(eb);
8356 btrfs_set_lock_blocking(eb);
8357 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8359 clean_tree_block(trans, root->fs_info, eb);
8362 if (eb == root->node) {
8363 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8366 BUG_ON(root->root_key.objectid !=
8367 btrfs_header_owner(eb));
8369 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8370 parent = path->nodes[level + 1]->start;
8372 BUG_ON(root->root_key.objectid !=
8373 btrfs_header_owner(path->nodes[level + 1]));
8376 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8378 wc->refs[level] = 0;
8379 wc->flags[level] = 0;
8383 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8384 struct btrfs_root *root,
8385 struct btrfs_path *path,
8386 struct walk_control *wc)
8388 int level = wc->level;
8389 int lookup_info = 1;
8392 while (level >= 0) {
8393 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8400 if (path->slots[level] >=
8401 btrfs_header_nritems(path->nodes[level]))
8404 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8406 path->slots[level]++;
8415 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8416 struct btrfs_root *root,
8417 struct btrfs_path *path,
8418 struct walk_control *wc, int max_level)
8420 int level = wc->level;
8423 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8424 while (level < max_level && path->nodes[level]) {
8426 if (path->slots[level] + 1 <
8427 btrfs_header_nritems(path->nodes[level])) {
8428 path->slots[level]++;
8431 ret = walk_up_proc(trans, root, path, wc);
8435 if (path->locks[level]) {
8436 btrfs_tree_unlock_rw(path->nodes[level],
8437 path->locks[level]);
8438 path->locks[level] = 0;
8440 free_extent_buffer(path->nodes[level]);
8441 path->nodes[level] = NULL;
8449 * drop a subvolume tree.
8451 * this function traverses the tree freeing any blocks that only
8452 * referenced by the tree.
8454 * when a shared tree block is found. this function decreases its
8455 * reference count by one. if update_ref is true, this function
8456 * also make sure backrefs for the shared block and all lower level
8457 * blocks are properly updated.
8459 * If called with for_reloc == 0, may exit early with -EAGAIN
8461 int btrfs_drop_snapshot(struct btrfs_root *root,
8462 struct btrfs_block_rsv *block_rsv, int update_ref,
8465 struct btrfs_path *path;
8466 struct btrfs_trans_handle *trans;
8467 struct btrfs_root *tree_root = root->fs_info->tree_root;
8468 struct btrfs_root_item *root_item = &root->root_item;
8469 struct walk_control *wc;
8470 struct btrfs_key key;
8474 bool root_dropped = false;
8476 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8478 path = btrfs_alloc_path();
8484 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8486 btrfs_free_path(path);
8491 trans = btrfs_start_transaction(tree_root, 0);
8492 if (IS_ERR(trans)) {
8493 err = PTR_ERR(trans);
8498 trans->block_rsv = block_rsv;
8500 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8501 level = btrfs_header_level(root->node);
8502 path->nodes[level] = btrfs_lock_root_node(root);
8503 btrfs_set_lock_blocking(path->nodes[level]);
8504 path->slots[level] = 0;
8505 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8506 memset(&wc->update_progress, 0,
8507 sizeof(wc->update_progress));
8509 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8510 memcpy(&wc->update_progress, &key,
8511 sizeof(wc->update_progress));
8513 level = root_item->drop_level;
8515 path->lowest_level = level;
8516 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8517 path->lowest_level = 0;
8525 * unlock our path, this is safe because only this
8526 * function is allowed to delete this snapshot
8528 btrfs_unlock_up_safe(path, 0);
8530 level = btrfs_header_level(root->node);
8532 btrfs_tree_lock(path->nodes[level]);
8533 btrfs_set_lock_blocking(path->nodes[level]);
8534 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8536 ret = btrfs_lookup_extent_info(trans, root,
8537 path->nodes[level]->start,
8538 level, 1, &wc->refs[level],
8544 BUG_ON(wc->refs[level] == 0);
8546 if (level == root_item->drop_level)
8549 btrfs_tree_unlock(path->nodes[level]);
8550 path->locks[level] = 0;
8551 WARN_ON(wc->refs[level] != 1);
8557 wc->shared_level = -1;
8558 wc->stage = DROP_REFERENCE;
8559 wc->update_ref = update_ref;
8561 wc->for_reloc = for_reloc;
8562 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8566 ret = walk_down_tree(trans, root, path, wc);
8572 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8579 BUG_ON(wc->stage != DROP_REFERENCE);
8583 if (wc->stage == DROP_REFERENCE) {
8585 btrfs_node_key(path->nodes[level],
8586 &root_item->drop_progress,
8587 path->slots[level]);
8588 root_item->drop_level = level;
8591 BUG_ON(wc->level == 0);
8592 if (btrfs_should_end_transaction(trans, tree_root) ||
8593 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8594 ret = btrfs_update_root(trans, tree_root,
8598 btrfs_abort_transaction(trans, tree_root, ret);
8603 btrfs_end_transaction_throttle(trans, tree_root);
8604 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8605 pr_debug("BTRFS: drop snapshot early exit\n");
8610 trans = btrfs_start_transaction(tree_root, 0);
8611 if (IS_ERR(trans)) {
8612 err = PTR_ERR(trans);
8616 trans->block_rsv = block_rsv;
8619 btrfs_release_path(path);
8623 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8625 btrfs_abort_transaction(trans, tree_root, ret);
8629 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8630 ret = btrfs_find_root(tree_root, &root->root_key, path,
8633 btrfs_abort_transaction(trans, tree_root, ret);
8636 } else if (ret > 0) {
8637 /* if we fail to delete the orphan item this time
8638 * around, it'll get picked up the next time.
8640 * The most common failure here is just -ENOENT.
8642 btrfs_del_orphan_item(trans, tree_root,
8643 root->root_key.objectid);
8647 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8648 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8650 free_extent_buffer(root->node);
8651 free_extent_buffer(root->commit_root);
8652 btrfs_put_fs_root(root);
8654 root_dropped = true;
8656 btrfs_end_transaction_throttle(trans, tree_root);
8659 btrfs_free_path(path);
8662 * So if we need to stop dropping the snapshot for whatever reason we
8663 * need to make sure to add it back to the dead root list so that we
8664 * keep trying to do the work later. This also cleans up roots if we
8665 * don't have it in the radix (like when we recover after a power fail
8666 * or unmount) so we don't leak memory.
8668 if (!for_reloc && root_dropped == false)
8669 btrfs_add_dead_root(root);
8670 if (err && err != -EAGAIN)
8671 btrfs_std_error(root->fs_info, err);
8676 * drop subtree rooted at tree block 'node'.
8678 * NOTE: this function will unlock and release tree block 'node'
8679 * only used by relocation code
8681 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8682 struct btrfs_root *root,
8683 struct extent_buffer *node,
8684 struct extent_buffer *parent)
8686 struct btrfs_path *path;
8687 struct walk_control *wc;
8693 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8695 path = btrfs_alloc_path();
8699 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8701 btrfs_free_path(path);
8705 btrfs_assert_tree_locked(parent);
8706 parent_level = btrfs_header_level(parent);
8707 extent_buffer_get(parent);
8708 path->nodes[parent_level] = parent;
8709 path->slots[parent_level] = btrfs_header_nritems(parent);
8711 btrfs_assert_tree_locked(node);
8712 level = btrfs_header_level(node);
8713 path->nodes[level] = node;
8714 path->slots[level] = 0;
8715 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8717 wc->refs[parent_level] = 1;
8718 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8720 wc->shared_level = -1;
8721 wc->stage = DROP_REFERENCE;
8725 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8728 wret = walk_down_tree(trans, root, path, wc);
8734 wret = walk_up_tree(trans, root, path, wc, parent_level);
8742 btrfs_free_path(path);
8746 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8752 * if restripe for this chunk_type is on pick target profile and
8753 * return, otherwise do the usual balance
8755 stripped = get_restripe_target(root->fs_info, flags);
8757 return extended_to_chunk(stripped);
8759 num_devices = root->fs_info->fs_devices->rw_devices;
8761 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8762 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8763 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8765 if (num_devices == 1) {
8766 stripped |= BTRFS_BLOCK_GROUP_DUP;
8767 stripped = flags & ~stripped;
8769 /* turn raid0 into single device chunks */
8770 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8773 /* turn mirroring into duplication */
8774 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8775 BTRFS_BLOCK_GROUP_RAID10))
8776 return stripped | BTRFS_BLOCK_GROUP_DUP;
8778 /* they already had raid on here, just return */
8779 if (flags & stripped)
8782 stripped |= BTRFS_BLOCK_GROUP_DUP;
8783 stripped = flags & ~stripped;
8785 /* switch duplicated blocks with raid1 */
8786 if (flags & BTRFS_BLOCK_GROUP_DUP)
8787 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8789 /* this is drive concat, leave it alone */
8795 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8797 struct btrfs_space_info *sinfo = cache->space_info;
8799 u64 min_allocable_bytes;
8804 * We need some metadata space and system metadata space for
8805 * allocating chunks in some corner cases until we force to set
8806 * it to be readonly.
8809 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8811 min_allocable_bytes = 1 * 1024 * 1024;
8813 min_allocable_bytes = 0;
8815 spin_lock(&sinfo->lock);
8816 spin_lock(&cache->lock);
8823 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8824 cache->bytes_super - btrfs_block_group_used(&cache->item);
8826 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8827 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8828 min_allocable_bytes <= sinfo->total_bytes) {
8829 sinfo->bytes_readonly += num_bytes;
8831 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8835 spin_unlock(&cache->lock);
8836 spin_unlock(&sinfo->lock);
8840 int btrfs_set_block_group_ro(struct btrfs_root *root,
8841 struct btrfs_block_group_cache *cache)
8844 struct btrfs_trans_handle *trans;
8851 trans = btrfs_join_transaction(root);
8853 return PTR_ERR(trans);
8856 * we're not allowed to set block groups readonly after the dirty
8857 * block groups cache has started writing. If it already started,
8858 * back off and let this transaction commit
8860 mutex_lock(&root->fs_info->ro_block_group_mutex);
8861 if (trans->transaction->dirty_bg_run) {
8862 u64 transid = trans->transid;
8864 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8865 btrfs_end_transaction(trans, root);
8867 ret = btrfs_wait_for_commit(root, transid);
8874 * if we are changing raid levels, try to allocate a corresponding
8875 * block group with the new raid level.
8877 alloc_flags = update_block_group_flags(root, cache->flags);
8878 if (alloc_flags != cache->flags) {
8879 ret = do_chunk_alloc(trans, root, alloc_flags,
8882 * ENOSPC is allowed here, we may have enough space
8883 * already allocated at the new raid level to
8892 ret = set_block_group_ro(cache, 0);
8895 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8896 ret = do_chunk_alloc(trans, root, alloc_flags,
8900 ret = set_block_group_ro(cache, 0);
8902 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8903 alloc_flags = update_block_group_flags(root, cache->flags);
8904 lock_chunks(root->fs_info->chunk_root);
8905 check_system_chunk(trans, root, alloc_flags);
8906 unlock_chunks(root->fs_info->chunk_root);
8908 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8910 btrfs_end_transaction(trans, root);
8914 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8915 struct btrfs_root *root, u64 type)
8917 u64 alloc_flags = get_alloc_profile(root, type);
8918 return do_chunk_alloc(trans, root, alloc_flags,
8923 * helper to account the unused space of all the readonly block group in the
8924 * space_info. takes mirrors into account.
8926 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8928 struct btrfs_block_group_cache *block_group;
8932 /* It's df, we don't care if it's racey */
8933 if (list_empty(&sinfo->ro_bgs))
8936 spin_lock(&sinfo->lock);
8937 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8938 spin_lock(&block_group->lock);
8940 if (!block_group->ro) {
8941 spin_unlock(&block_group->lock);
8945 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8946 BTRFS_BLOCK_GROUP_RAID10 |
8947 BTRFS_BLOCK_GROUP_DUP))
8952 free_bytes += (block_group->key.offset -
8953 btrfs_block_group_used(&block_group->item)) *
8956 spin_unlock(&block_group->lock);
8958 spin_unlock(&sinfo->lock);
8963 void btrfs_set_block_group_rw(struct btrfs_root *root,
8964 struct btrfs_block_group_cache *cache)
8966 struct btrfs_space_info *sinfo = cache->space_info;
8971 spin_lock(&sinfo->lock);
8972 spin_lock(&cache->lock);
8973 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8974 cache->bytes_super - btrfs_block_group_used(&cache->item);
8975 sinfo->bytes_readonly -= num_bytes;
8977 list_del_init(&cache->ro_list);
8978 spin_unlock(&cache->lock);
8979 spin_unlock(&sinfo->lock);
8983 * checks to see if its even possible to relocate this block group.
8985 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8986 * ok to go ahead and try.
8988 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8990 struct btrfs_block_group_cache *block_group;
8991 struct btrfs_space_info *space_info;
8992 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8993 struct btrfs_device *device;
8994 struct btrfs_trans_handle *trans;
9003 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9005 /* odd, couldn't find the block group, leave it alone */
9009 min_free = btrfs_block_group_used(&block_group->item);
9011 /* no bytes used, we're good */
9015 space_info = block_group->space_info;
9016 spin_lock(&space_info->lock);
9018 full = space_info->full;
9021 * if this is the last block group we have in this space, we can't
9022 * relocate it unless we're able to allocate a new chunk below.
9024 * Otherwise, we need to make sure we have room in the space to handle
9025 * all of the extents from this block group. If we can, we're good
9027 if ((space_info->total_bytes != block_group->key.offset) &&
9028 (space_info->bytes_used + space_info->bytes_reserved +
9029 space_info->bytes_pinned + space_info->bytes_readonly +
9030 min_free < space_info->total_bytes)) {
9031 spin_unlock(&space_info->lock);
9034 spin_unlock(&space_info->lock);
9037 * ok we don't have enough space, but maybe we have free space on our
9038 * devices to allocate new chunks for relocation, so loop through our
9039 * alloc devices and guess if we have enough space. if this block
9040 * group is going to be restriped, run checks against the target
9041 * profile instead of the current one.
9053 target = get_restripe_target(root->fs_info, block_group->flags);
9055 index = __get_raid_index(extended_to_chunk(target));
9058 * this is just a balance, so if we were marked as full
9059 * we know there is no space for a new chunk
9064 index = get_block_group_index(block_group);
9067 if (index == BTRFS_RAID_RAID10) {
9071 } else if (index == BTRFS_RAID_RAID1) {
9073 } else if (index == BTRFS_RAID_DUP) {
9076 } else if (index == BTRFS_RAID_RAID0) {
9077 dev_min = fs_devices->rw_devices;
9078 min_free = div64_u64(min_free, dev_min);
9081 /* We need to do this so that we can look at pending chunks */
9082 trans = btrfs_join_transaction(root);
9083 if (IS_ERR(trans)) {
9084 ret = PTR_ERR(trans);
9088 mutex_lock(&root->fs_info->chunk_mutex);
9089 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9093 * check to make sure we can actually find a chunk with enough
9094 * space to fit our block group in.
9096 if (device->total_bytes > device->bytes_used + min_free &&
9097 !device->is_tgtdev_for_dev_replace) {
9098 ret = find_free_dev_extent(trans, device, min_free,
9103 if (dev_nr >= dev_min)
9109 mutex_unlock(&root->fs_info->chunk_mutex);
9110 btrfs_end_transaction(trans, root);
9112 btrfs_put_block_group(block_group);
9116 static int find_first_block_group(struct btrfs_root *root,
9117 struct btrfs_path *path, struct btrfs_key *key)
9120 struct btrfs_key found_key;
9121 struct extent_buffer *leaf;
9124 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9129 slot = path->slots[0];
9130 leaf = path->nodes[0];
9131 if (slot >= btrfs_header_nritems(leaf)) {
9132 ret = btrfs_next_leaf(root, path);
9139 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9141 if (found_key.objectid >= key->objectid &&
9142 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9152 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9154 struct btrfs_block_group_cache *block_group;
9158 struct inode *inode;
9160 block_group = btrfs_lookup_first_block_group(info, last);
9161 while (block_group) {
9162 spin_lock(&block_group->lock);
9163 if (block_group->iref)
9165 spin_unlock(&block_group->lock);
9166 block_group = next_block_group(info->tree_root,
9176 inode = block_group->inode;
9177 block_group->iref = 0;
9178 block_group->inode = NULL;
9179 spin_unlock(&block_group->lock);
9181 last = block_group->key.objectid + block_group->key.offset;
9182 btrfs_put_block_group(block_group);
9186 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9188 struct btrfs_block_group_cache *block_group;
9189 struct btrfs_space_info *space_info;
9190 struct btrfs_caching_control *caching_ctl;
9193 down_write(&info->commit_root_sem);
9194 while (!list_empty(&info->caching_block_groups)) {
9195 caching_ctl = list_entry(info->caching_block_groups.next,
9196 struct btrfs_caching_control, list);
9197 list_del(&caching_ctl->list);
9198 put_caching_control(caching_ctl);
9200 up_write(&info->commit_root_sem);
9202 spin_lock(&info->unused_bgs_lock);
9203 while (!list_empty(&info->unused_bgs)) {
9204 block_group = list_first_entry(&info->unused_bgs,
9205 struct btrfs_block_group_cache,
9207 list_del_init(&block_group->bg_list);
9208 btrfs_put_block_group(block_group);
9210 spin_unlock(&info->unused_bgs_lock);
9212 spin_lock(&info->block_group_cache_lock);
9213 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9214 block_group = rb_entry(n, struct btrfs_block_group_cache,
9216 rb_erase(&block_group->cache_node,
9217 &info->block_group_cache_tree);
9218 RB_CLEAR_NODE(&block_group->cache_node);
9219 spin_unlock(&info->block_group_cache_lock);
9221 down_write(&block_group->space_info->groups_sem);
9222 list_del(&block_group->list);
9223 up_write(&block_group->space_info->groups_sem);
9225 if (block_group->cached == BTRFS_CACHE_STARTED)
9226 wait_block_group_cache_done(block_group);
9229 * We haven't cached this block group, which means we could
9230 * possibly have excluded extents on this block group.
9232 if (block_group->cached == BTRFS_CACHE_NO ||
9233 block_group->cached == BTRFS_CACHE_ERROR)
9234 free_excluded_extents(info->extent_root, block_group);
9236 btrfs_remove_free_space_cache(block_group);
9237 btrfs_put_block_group(block_group);
9239 spin_lock(&info->block_group_cache_lock);
9241 spin_unlock(&info->block_group_cache_lock);
9243 /* now that all the block groups are freed, go through and
9244 * free all the space_info structs. This is only called during
9245 * the final stages of unmount, and so we know nobody is
9246 * using them. We call synchronize_rcu() once before we start,
9247 * just to be on the safe side.
9251 release_global_block_rsv(info);
9253 while (!list_empty(&info->space_info)) {
9256 space_info = list_entry(info->space_info.next,
9257 struct btrfs_space_info,
9259 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9260 if (WARN_ON(space_info->bytes_pinned > 0 ||
9261 space_info->bytes_reserved > 0 ||
9262 space_info->bytes_may_use > 0)) {
9263 dump_space_info(space_info, 0, 0);
9266 list_del(&space_info->list);
9267 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9268 struct kobject *kobj;
9269 kobj = space_info->block_group_kobjs[i];
9270 space_info->block_group_kobjs[i] = NULL;
9276 kobject_del(&space_info->kobj);
9277 kobject_put(&space_info->kobj);
9282 static void __link_block_group(struct btrfs_space_info *space_info,
9283 struct btrfs_block_group_cache *cache)
9285 int index = get_block_group_index(cache);
9288 down_write(&space_info->groups_sem);
9289 if (list_empty(&space_info->block_groups[index]))
9291 list_add_tail(&cache->list, &space_info->block_groups[index]);
9292 up_write(&space_info->groups_sem);
9295 struct raid_kobject *rkobj;
9298 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9301 rkobj->raid_type = index;
9302 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9303 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9304 "%s", get_raid_name(index));
9306 kobject_put(&rkobj->kobj);
9309 space_info->block_group_kobjs[index] = &rkobj->kobj;
9314 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9317 static struct btrfs_block_group_cache *
9318 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9320 struct btrfs_block_group_cache *cache;
9322 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9326 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9328 if (!cache->free_space_ctl) {
9333 cache->key.objectid = start;
9334 cache->key.offset = size;
9335 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9337 cache->sectorsize = root->sectorsize;
9338 cache->fs_info = root->fs_info;
9339 cache->full_stripe_len = btrfs_full_stripe_len(root,
9340 &root->fs_info->mapping_tree,
9342 atomic_set(&cache->count, 1);
9343 spin_lock_init(&cache->lock);
9344 init_rwsem(&cache->data_rwsem);
9345 INIT_LIST_HEAD(&cache->list);
9346 INIT_LIST_HEAD(&cache->cluster_list);
9347 INIT_LIST_HEAD(&cache->bg_list);
9348 INIT_LIST_HEAD(&cache->ro_list);
9349 INIT_LIST_HEAD(&cache->dirty_list);
9350 INIT_LIST_HEAD(&cache->io_list);
9351 btrfs_init_free_space_ctl(cache);
9352 atomic_set(&cache->trimming, 0);
9357 int btrfs_read_block_groups(struct btrfs_root *root)
9359 struct btrfs_path *path;
9361 struct btrfs_block_group_cache *cache;
9362 struct btrfs_fs_info *info = root->fs_info;
9363 struct btrfs_space_info *space_info;
9364 struct btrfs_key key;
9365 struct btrfs_key found_key;
9366 struct extent_buffer *leaf;
9370 root = info->extent_root;
9373 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9374 path = btrfs_alloc_path();
9379 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9380 if (btrfs_test_opt(root, SPACE_CACHE) &&
9381 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9383 if (btrfs_test_opt(root, CLEAR_CACHE))
9387 ret = find_first_block_group(root, path, &key);
9393 leaf = path->nodes[0];
9394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9396 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9405 * When we mount with old space cache, we need to
9406 * set BTRFS_DC_CLEAR and set dirty flag.
9408 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9409 * truncate the old free space cache inode and
9411 * b) Setting 'dirty flag' makes sure that we flush
9412 * the new space cache info onto disk.
9414 if (btrfs_test_opt(root, SPACE_CACHE))
9415 cache->disk_cache_state = BTRFS_DC_CLEAR;
9418 read_extent_buffer(leaf, &cache->item,
9419 btrfs_item_ptr_offset(leaf, path->slots[0]),
9420 sizeof(cache->item));
9421 cache->flags = btrfs_block_group_flags(&cache->item);
9423 key.objectid = found_key.objectid + found_key.offset;
9424 btrfs_release_path(path);
9427 * We need to exclude the super stripes now so that the space
9428 * info has super bytes accounted for, otherwise we'll think
9429 * we have more space than we actually do.
9431 ret = exclude_super_stripes(root, cache);
9434 * We may have excluded something, so call this just in
9437 free_excluded_extents(root, cache);
9438 btrfs_put_block_group(cache);
9443 * check for two cases, either we are full, and therefore
9444 * don't need to bother with the caching work since we won't
9445 * find any space, or we are empty, and we can just add all
9446 * the space in and be done with it. This saves us _alot_ of
9447 * time, particularly in the full case.
9449 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9450 cache->last_byte_to_unpin = (u64)-1;
9451 cache->cached = BTRFS_CACHE_FINISHED;
9452 free_excluded_extents(root, cache);
9453 } else if (btrfs_block_group_used(&cache->item) == 0) {
9454 cache->last_byte_to_unpin = (u64)-1;
9455 cache->cached = BTRFS_CACHE_FINISHED;
9456 add_new_free_space(cache, root->fs_info,
9458 found_key.objectid +
9460 free_excluded_extents(root, cache);
9463 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9465 btrfs_remove_free_space_cache(cache);
9466 btrfs_put_block_group(cache);
9470 ret = update_space_info(info, cache->flags, found_key.offset,
9471 btrfs_block_group_used(&cache->item),
9474 btrfs_remove_free_space_cache(cache);
9475 spin_lock(&info->block_group_cache_lock);
9476 rb_erase(&cache->cache_node,
9477 &info->block_group_cache_tree);
9478 RB_CLEAR_NODE(&cache->cache_node);
9479 spin_unlock(&info->block_group_cache_lock);
9480 btrfs_put_block_group(cache);
9484 cache->space_info = space_info;
9485 spin_lock(&cache->space_info->lock);
9486 cache->space_info->bytes_readonly += cache->bytes_super;
9487 spin_unlock(&cache->space_info->lock);
9489 __link_block_group(space_info, cache);
9491 set_avail_alloc_bits(root->fs_info, cache->flags);
9492 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9493 set_block_group_ro(cache, 1);
9494 } else if (btrfs_block_group_used(&cache->item) == 0) {
9495 spin_lock(&info->unused_bgs_lock);
9496 /* Should always be true but just in case. */
9497 if (list_empty(&cache->bg_list)) {
9498 btrfs_get_block_group(cache);
9499 list_add_tail(&cache->bg_list,
9502 spin_unlock(&info->unused_bgs_lock);
9506 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9507 if (!(get_alloc_profile(root, space_info->flags) &
9508 (BTRFS_BLOCK_GROUP_RAID10 |
9509 BTRFS_BLOCK_GROUP_RAID1 |
9510 BTRFS_BLOCK_GROUP_RAID5 |
9511 BTRFS_BLOCK_GROUP_RAID6 |
9512 BTRFS_BLOCK_GROUP_DUP)))
9515 * avoid allocating from un-mirrored block group if there are
9516 * mirrored block groups.
9518 list_for_each_entry(cache,
9519 &space_info->block_groups[BTRFS_RAID_RAID0],
9521 set_block_group_ro(cache, 1);
9522 list_for_each_entry(cache,
9523 &space_info->block_groups[BTRFS_RAID_SINGLE],
9525 set_block_group_ro(cache, 1);
9528 init_global_block_rsv(info);
9531 btrfs_free_path(path);
9535 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9536 struct btrfs_root *root)
9538 struct btrfs_block_group_cache *block_group, *tmp;
9539 struct btrfs_root *extent_root = root->fs_info->extent_root;
9540 struct btrfs_block_group_item item;
9541 struct btrfs_key key;
9544 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9548 spin_lock(&block_group->lock);
9549 memcpy(&item, &block_group->item, sizeof(item));
9550 memcpy(&key, &block_group->key, sizeof(key));
9551 spin_unlock(&block_group->lock);
9553 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9556 btrfs_abort_transaction(trans, extent_root, ret);
9557 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9558 key.objectid, key.offset);
9560 btrfs_abort_transaction(trans, extent_root, ret);
9562 list_del_init(&block_group->bg_list);
9566 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9567 struct btrfs_root *root, u64 bytes_used,
9568 u64 type, u64 chunk_objectid, u64 chunk_offset,
9572 struct btrfs_root *extent_root;
9573 struct btrfs_block_group_cache *cache;
9575 extent_root = root->fs_info->extent_root;
9577 btrfs_set_log_full_commit(root->fs_info, trans);
9579 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9583 btrfs_set_block_group_used(&cache->item, bytes_used);
9584 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9585 btrfs_set_block_group_flags(&cache->item, type);
9587 cache->flags = type;
9588 cache->last_byte_to_unpin = (u64)-1;
9589 cache->cached = BTRFS_CACHE_FINISHED;
9590 ret = exclude_super_stripes(root, cache);
9593 * We may have excluded something, so call this just in
9596 free_excluded_extents(root, cache);
9597 btrfs_put_block_group(cache);
9601 add_new_free_space(cache, root->fs_info, chunk_offset,
9602 chunk_offset + size);
9604 free_excluded_extents(root, cache);
9607 * Call to ensure the corresponding space_info object is created and
9608 * assigned to our block group, but don't update its counters just yet.
9609 * We want our bg to be added to the rbtree with its ->space_info set.
9611 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9612 &cache->space_info);
9614 btrfs_remove_free_space_cache(cache);
9615 btrfs_put_block_group(cache);
9619 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9621 btrfs_remove_free_space_cache(cache);
9622 btrfs_put_block_group(cache);
9627 * Now that our block group has its ->space_info set and is inserted in
9628 * the rbtree, update the space info's counters.
9630 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9631 &cache->space_info);
9633 btrfs_remove_free_space_cache(cache);
9634 spin_lock(&root->fs_info->block_group_cache_lock);
9635 rb_erase(&cache->cache_node,
9636 &root->fs_info->block_group_cache_tree);
9637 RB_CLEAR_NODE(&cache->cache_node);
9638 spin_unlock(&root->fs_info->block_group_cache_lock);
9639 btrfs_put_block_group(cache);
9642 update_global_block_rsv(root->fs_info);
9644 spin_lock(&cache->space_info->lock);
9645 cache->space_info->bytes_readonly += cache->bytes_super;
9646 spin_unlock(&cache->space_info->lock);
9648 __link_block_group(cache->space_info, cache);
9650 list_add_tail(&cache->bg_list, &trans->new_bgs);
9652 set_avail_alloc_bits(extent_root->fs_info, type);
9657 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9659 u64 extra_flags = chunk_to_extended(flags) &
9660 BTRFS_EXTENDED_PROFILE_MASK;
9662 write_seqlock(&fs_info->profiles_lock);
9663 if (flags & BTRFS_BLOCK_GROUP_DATA)
9664 fs_info->avail_data_alloc_bits &= ~extra_flags;
9665 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9666 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9667 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9668 fs_info->avail_system_alloc_bits &= ~extra_flags;
9669 write_sequnlock(&fs_info->profiles_lock);
9672 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9673 struct btrfs_root *root, u64 group_start,
9674 struct extent_map *em)
9676 struct btrfs_path *path;
9677 struct btrfs_block_group_cache *block_group;
9678 struct btrfs_free_cluster *cluster;
9679 struct btrfs_root *tree_root = root->fs_info->tree_root;
9680 struct btrfs_key key;
9681 struct inode *inode;
9682 struct kobject *kobj = NULL;
9686 struct btrfs_caching_control *caching_ctl = NULL;
9689 root = root->fs_info->extent_root;
9691 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9692 BUG_ON(!block_group);
9693 BUG_ON(!block_group->ro);
9696 * Free the reserved super bytes from this block group before
9699 free_excluded_extents(root, block_group);
9701 memcpy(&key, &block_group->key, sizeof(key));
9702 index = get_block_group_index(block_group);
9703 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9704 BTRFS_BLOCK_GROUP_RAID1 |
9705 BTRFS_BLOCK_GROUP_RAID10))
9710 /* make sure this block group isn't part of an allocation cluster */
9711 cluster = &root->fs_info->data_alloc_cluster;
9712 spin_lock(&cluster->refill_lock);
9713 btrfs_return_cluster_to_free_space(block_group, cluster);
9714 spin_unlock(&cluster->refill_lock);
9717 * make sure this block group isn't part of a metadata
9718 * allocation cluster
9720 cluster = &root->fs_info->meta_alloc_cluster;
9721 spin_lock(&cluster->refill_lock);
9722 btrfs_return_cluster_to_free_space(block_group, cluster);
9723 spin_unlock(&cluster->refill_lock);
9725 path = btrfs_alloc_path();
9732 * get the inode first so any iput calls done for the io_list
9733 * aren't the final iput (no unlinks allowed now)
9735 inode = lookup_free_space_inode(tree_root, block_group, path);
9737 mutex_lock(&trans->transaction->cache_write_mutex);
9739 * make sure our free spache cache IO is done before remove the
9742 spin_lock(&trans->transaction->dirty_bgs_lock);
9743 if (!list_empty(&block_group->io_list)) {
9744 list_del_init(&block_group->io_list);
9746 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9748 spin_unlock(&trans->transaction->dirty_bgs_lock);
9749 btrfs_wait_cache_io(root, trans, block_group,
9750 &block_group->io_ctl, path,
9751 block_group->key.objectid);
9752 btrfs_put_block_group(block_group);
9753 spin_lock(&trans->transaction->dirty_bgs_lock);
9756 if (!list_empty(&block_group->dirty_list)) {
9757 list_del_init(&block_group->dirty_list);
9758 btrfs_put_block_group(block_group);
9760 spin_unlock(&trans->transaction->dirty_bgs_lock);
9761 mutex_unlock(&trans->transaction->cache_write_mutex);
9763 if (!IS_ERR(inode)) {
9764 ret = btrfs_orphan_add(trans, inode);
9766 btrfs_add_delayed_iput(inode);
9770 /* One for the block groups ref */
9771 spin_lock(&block_group->lock);
9772 if (block_group->iref) {
9773 block_group->iref = 0;
9774 block_group->inode = NULL;
9775 spin_unlock(&block_group->lock);
9778 spin_unlock(&block_group->lock);
9780 /* One for our lookup ref */
9781 btrfs_add_delayed_iput(inode);
9784 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9785 key.offset = block_group->key.objectid;
9788 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9792 btrfs_release_path(path);
9794 ret = btrfs_del_item(trans, tree_root, path);
9797 btrfs_release_path(path);
9800 spin_lock(&root->fs_info->block_group_cache_lock);
9801 rb_erase(&block_group->cache_node,
9802 &root->fs_info->block_group_cache_tree);
9803 RB_CLEAR_NODE(&block_group->cache_node);
9805 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9806 root->fs_info->first_logical_byte = (u64)-1;
9807 spin_unlock(&root->fs_info->block_group_cache_lock);
9809 down_write(&block_group->space_info->groups_sem);
9811 * we must use list_del_init so people can check to see if they
9812 * are still on the list after taking the semaphore
9814 list_del_init(&block_group->list);
9815 if (list_empty(&block_group->space_info->block_groups[index])) {
9816 kobj = block_group->space_info->block_group_kobjs[index];
9817 block_group->space_info->block_group_kobjs[index] = NULL;
9818 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9820 up_write(&block_group->space_info->groups_sem);
9826 if (block_group->has_caching_ctl)
9827 caching_ctl = get_caching_control(block_group);
9828 if (block_group->cached == BTRFS_CACHE_STARTED)
9829 wait_block_group_cache_done(block_group);
9830 if (block_group->has_caching_ctl) {
9831 down_write(&root->fs_info->commit_root_sem);
9833 struct btrfs_caching_control *ctl;
9835 list_for_each_entry(ctl,
9836 &root->fs_info->caching_block_groups, list)
9837 if (ctl->block_group == block_group) {
9839 atomic_inc(&caching_ctl->count);
9844 list_del_init(&caching_ctl->list);
9845 up_write(&root->fs_info->commit_root_sem);
9847 /* Once for the caching bgs list and once for us. */
9848 put_caching_control(caching_ctl);
9849 put_caching_control(caching_ctl);
9853 spin_lock(&trans->transaction->dirty_bgs_lock);
9854 if (!list_empty(&block_group->dirty_list)) {
9857 if (!list_empty(&block_group->io_list)) {
9860 spin_unlock(&trans->transaction->dirty_bgs_lock);
9861 btrfs_remove_free_space_cache(block_group);
9863 spin_lock(&block_group->space_info->lock);
9864 list_del_init(&block_group->ro_list);
9866 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9867 WARN_ON(block_group->space_info->total_bytes
9868 < block_group->key.offset);
9869 WARN_ON(block_group->space_info->bytes_readonly
9870 < block_group->key.offset);
9871 WARN_ON(block_group->space_info->disk_total
9872 < block_group->key.offset * factor);
9874 block_group->space_info->total_bytes -= block_group->key.offset;
9875 block_group->space_info->bytes_readonly -= block_group->key.offset;
9876 block_group->space_info->disk_total -= block_group->key.offset * factor;
9878 spin_unlock(&block_group->space_info->lock);
9880 memcpy(&key, &block_group->key, sizeof(key));
9883 if (!list_empty(&em->list)) {
9884 /* We're in the transaction->pending_chunks list. */
9885 free_extent_map(em);
9887 spin_lock(&block_group->lock);
9888 block_group->removed = 1;
9890 * At this point trimming can't start on this block group, because we
9891 * removed the block group from the tree fs_info->block_group_cache_tree
9892 * so no one can't find it anymore and even if someone already got this
9893 * block group before we removed it from the rbtree, they have already
9894 * incremented block_group->trimming - if they didn't, they won't find
9895 * any free space entries because we already removed them all when we
9896 * called btrfs_remove_free_space_cache().
9898 * And we must not remove the extent map from the fs_info->mapping_tree
9899 * to prevent the same logical address range and physical device space
9900 * ranges from being reused for a new block group. This is because our
9901 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9902 * completely transactionless, so while it is trimming a range the
9903 * currently running transaction might finish and a new one start,
9904 * allowing for new block groups to be created that can reuse the same
9905 * physical device locations unless we take this special care.
9907 remove_em = (atomic_read(&block_group->trimming) == 0);
9909 * Make sure a trimmer task always sees the em in the pinned_chunks list
9910 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9911 * before checking block_group->removed).
9915 * Our em might be in trans->transaction->pending_chunks which
9916 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9917 * and so is the fs_info->pinned_chunks list.
9919 * So at this point we must be holding the chunk_mutex to avoid
9920 * any races with chunk allocation (more specifically at
9921 * volumes.c:contains_pending_extent()), to ensure it always
9922 * sees the em, either in the pending_chunks list or in the
9923 * pinned_chunks list.
9925 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9927 spin_unlock(&block_group->lock);
9930 struct extent_map_tree *em_tree;
9932 em_tree = &root->fs_info->mapping_tree.map_tree;
9933 write_lock(&em_tree->lock);
9935 * The em might be in the pending_chunks list, so make sure the
9936 * chunk mutex is locked, since remove_extent_mapping() will
9937 * delete us from that list.
9939 remove_extent_mapping(em_tree, em);
9940 write_unlock(&em_tree->lock);
9941 /* once for the tree */
9942 free_extent_map(em);
9945 unlock_chunks(root);
9947 btrfs_put_block_group(block_group);
9948 btrfs_put_block_group(block_group);
9950 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9956 ret = btrfs_del_item(trans, root, path);
9958 btrfs_free_path(path);
9963 * Process the unused_bgs list and remove any that don't have any allocated
9964 * space inside of them.
9966 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9968 struct btrfs_block_group_cache *block_group;
9969 struct btrfs_space_info *space_info;
9970 struct btrfs_root *root = fs_info->extent_root;
9971 struct btrfs_trans_handle *trans;
9977 spin_lock(&fs_info->unused_bgs_lock);
9978 while (!list_empty(&fs_info->unused_bgs)) {
9981 block_group = list_first_entry(&fs_info->unused_bgs,
9982 struct btrfs_block_group_cache,
9984 space_info = block_group->space_info;
9985 list_del_init(&block_group->bg_list);
9986 if (ret || btrfs_mixed_space_info(space_info)) {
9987 btrfs_put_block_group(block_group);
9990 spin_unlock(&fs_info->unused_bgs_lock);
9992 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
9994 /* Don't want to race with allocators so take the groups_sem */
9995 down_write(&space_info->groups_sem);
9996 spin_lock(&block_group->lock);
9997 if (block_group->reserved ||
9998 btrfs_block_group_used(&block_group->item) ||
10001 * We want to bail if we made new allocations or have
10002 * outstanding allocations in this block group. We do
10003 * the ro check in case balance is currently acting on
10004 * this block group.
10006 spin_unlock(&block_group->lock);
10007 up_write(&space_info->groups_sem);
10010 spin_unlock(&block_group->lock);
10012 /* We don't want to force the issue, only flip if it's ok. */
10013 ret = set_block_group_ro(block_group, 0);
10014 up_write(&space_info->groups_sem);
10021 * Want to do this before we do anything else so we can recover
10022 * properly if we fail to join the transaction.
10024 /* 1 for btrfs_orphan_reserve_metadata() */
10025 trans = btrfs_start_transaction(root, 1);
10026 if (IS_ERR(trans)) {
10027 btrfs_set_block_group_rw(root, block_group);
10028 ret = PTR_ERR(trans);
10033 * We could have pending pinned extents for this block group,
10034 * just delete them, we don't care about them anymore.
10036 start = block_group->key.objectid;
10037 end = start + block_group->key.offset - 1;
10039 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10040 * btrfs_finish_extent_commit(). If we are at transaction N,
10041 * another task might be running finish_extent_commit() for the
10042 * previous transaction N - 1, and have seen a range belonging
10043 * to the block group in freed_extents[] before we were able to
10044 * clear the whole block group range from freed_extents[]. This
10045 * means that task can lookup for the block group after we
10046 * unpinned it from freed_extents[] and removed it, leading to
10047 * a BUG_ON() at btrfs_unpin_extent_range().
10049 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10050 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10051 EXTENT_DIRTY, GFP_NOFS);
10053 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10054 btrfs_set_block_group_rw(root, block_group);
10057 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10058 EXTENT_DIRTY, GFP_NOFS);
10060 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10061 btrfs_set_block_group_rw(root, block_group);
10064 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10066 /* Reset pinned so btrfs_put_block_group doesn't complain */
10067 spin_lock(&space_info->lock);
10068 spin_lock(&block_group->lock);
10070 space_info->bytes_pinned -= block_group->pinned;
10071 space_info->bytes_readonly += block_group->pinned;
10072 percpu_counter_add(&space_info->total_bytes_pinned,
10073 -block_group->pinned);
10074 block_group->pinned = 0;
10076 spin_unlock(&block_group->lock);
10077 spin_unlock(&space_info->lock);
10080 * Btrfs_remove_chunk will abort the transaction if things go
10083 ret = btrfs_remove_chunk(trans, root,
10084 block_group->key.objectid);
10086 btrfs_end_transaction(trans, root);
10088 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10089 btrfs_put_block_group(block_group);
10090 spin_lock(&fs_info->unused_bgs_lock);
10092 spin_unlock(&fs_info->unused_bgs_lock);
10095 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10097 struct btrfs_space_info *space_info;
10098 struct btrfs_super_block *disk_super;
10104 disk_super = fs_info->super_copy;
10105 if (!btrfs_super_root(disk_super))
10108 features = btrfs_super_incompat_flags(disk_super);
10109 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10112 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10113 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10118 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10119 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10121 flags = BTRFS_BLOCK_GROUP_METADATA;
10122 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10126 flags = BTRFS_BLOCK_GROUP_DATA;
10127 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10133 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10135 return unpin_extent_range(root, start, end, false);
10138 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10140 struct btrfs_fs_info *fs_info = root->fs_info;
10141 struct btrfs_block_group_cache *cache = NULL;
10146 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10150 * try to trim all FS space, our block group may start from non-zero.
10152 if (range->len == total_bytes)
10153 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10155 cache = btrfs_lookup_block_group(fs_info, range->start);
10158 if (cache->key.objectid >= (range->start + range->len)) {
10159 btrfs_put_block_group(cache);
10163 start = max(range->start, cache->key.objectid);
10164 end = min(range->start + range->len,
10165 cache->key.objectid + cache->key.offset);
10167 if (end - start >= range->minlen) {
10168 if (!block_group_cache_done(cache)) {
10169 ret = cache_block_group(cache, 0);
10171 btrfs_put_block_group(cache);
10174 ret = wait_block_group_cache_done(cache);
10176 btrfs_put_block_group(cache);
10180 ret = btrfs_trim_block_group(cache,
10186 trimmed += group_trimmed;
10188 btrfs_put_block_group(cache);
10193 cache = next_block_group(fs_info->tree_root, cache);
10196 range->len = trimmed;
10201 * btrfs_{start,end}_write_no_snapshoting() are similar to
10202 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10203 * data into the page cache through nocow before the subvolume is snapshoted,
10204 * but flush the data into disk after the snapshot creation, or to prevent
10205 * operations while snapshoting is ongoing and that cause the snapshot to be
10206 * inconsistent (writes followed by expanding truncates for example).
10208 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10210 percpu_counter_dec(&root->subv_writers->counter);
10212 * Make sure counter is updated before we wake up
10216 if (waitqueue_active(&root->subv_writers->wait))
10217 wake_up(&root->subv_writers->wait);
10220 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10222 if (atomic_read(&root->will_be_snapshoted))
10225 percpu_counter_inc(&root->subv_writers->counter);
10227 * Make sure counter is updated before we check for snapshot creation.
10230 if (atomic_read(&root->will_be_snapshoted)) {
10231 btrfs_end_write_no_snapshoting(root);