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>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE = 0,
52 CHUNK_ALLOC_FORCE = 1,
53 CHUNK_ALLOC_LIMITED = 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT = 2,
71 static int update_block_group(struct btrfs_trans_handle *trans,
72 struct btrfs_root *root,
73 u64 bytenr, u64 num_bytes, int alloc);
74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, u64 parent,
77 u64 root_objectid, u64 owner_objectid,
78 u64 owner_offset, int refs_to_drop,
79 struct btrfs_delayed_extent_op *extra_op);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
81 struct extent_buffer *leaf,
82 struct btrfs_extent_item *ei);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 u64 parent, u64 root_objectid,
86 u64 flags, u64 owner, u64 offset,
87 struct btrfs_key *ins, int ref_mod);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, struct btrfs_disk_key *key,
92 int level, struct btrfs_key *ins);
93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
94 struct btrfs_root *extent_root, u64 alloc_bytes,
95 u64 flags, int force);
96 static int find_next_key(struct btrfs_path *path, int level,
97 struct btrfs_key *key);
98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
99 int dump_block_groups);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
101 u64 num_bytes, int reserve);
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
125 kfree(cache->free_space_ctl);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
135 struct btrfs_block_group_cache *block_group)
138 struct rb_node *parent = NULL;
139 struct btrfs_block_group_cache *cache;
141 spin_lock(&info->block_group_cache_lock);
142 p = &info->block_group_cache_tree.rb_node;
146 cache = rb_entry(parent, struct btrfs_block_group_cache,
148 if (block_group->key.objectid < cache->key.objectid) {
150 } else if (block_group->key.objectid > cache->key.objectid) {
153 spin_unlock(&info->block_group_cache_lock);
158 rb_link_node(&block_group->cache_node, parent, p);
159 rb_insert_color(&block_group->cache_node,
160 &info->block_group_cache_tree);
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_root *root,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&root->fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215 set_extent_bits(&root->fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 static void free_excluded_extents(struct btrfs_root *root,
221 struct btrfs_block_group_cache *cache)
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 clear_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 static int exclude_super_stripes(struct btrfs_root *root,
235 struct btrfs_block_group_cache *cache)
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(root, cache->key.objectid,
250 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
251 bytenr = btrfs_sb_offset(i);
252 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
253 cache->key.objectid, bytenr,
254 0, &logical, &nr, &stripe_len);
258 cache->bytes_super += stripe_len;
259 ret = add_excluded_extent(root, logical[nr],
269 static struct btrfs_caching_control *
270 get_caching_control(struct btrfs_block_group_cache *cache)
272 struct btrfs_caching_control *ctl;
274 spin_lock(&cache->lock);
275 if (cache->cached != BTRFS_CACHE_STARTED) {
276 spin_unlock(&cache->lock);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache->caching_ctl) {
282 spin_unlock(&cache->lock);
286 ctl = cache->caching_ctl;
287 atomic_inc(&ctl->count);
288 spin_unlock(&cache->lock);
292 static void put_caching_control(struct btrfs_caching_control *ctl)
294 if (atomic_dec_and_test(&ctl->count))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
304 struct btrfs_fs_info *info, u64 start, u64 end)
306 u64 extent_start, extent_end, size, total_added = 0;
309 while (start < end) {
310 ret = find_first_extent_bit(info->pinned_extents, start,
311 &extent_start, &extent_end,
312 EXTENT_DIRTY | EXTENT_UPTODATE);
316 if (extent_start <= start) {
317 start = extent_end + 1;
318 } else if (extent_start > start && extent_start < end) {
319 size = extent_start - start;
321 ret = btrfs_add_free_space(block_group, start,
324 start = extent_end + 1;
333 ret = btrfs_add_free_space(block_group, start, size);
340 static noinline void caching_thread(struct btrfs_work *work)
342 struct btrfs_block_group_cache *block_group;
343 struct btrfs_fs_info *fs_info;
344 struct btrfs_caching_control *caching_ctl;
345 struct btrfs_root *extent_root;
346 struct btrfs_path *path;
347 struct extent_buffer *leaf;
348 struct btrfs_key key;
354 caching_ctl = container_of(work, struct btrfs_caching_control, work);
355 block_group = caching_ctl->block_group;
356 fs_info = block_group->fs_info;
357 extent_root = fs_info->extent_root;
359 path = btrfs_alloc_path();
363 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path->skip_locking = 1;
372 path->search_commit_root = 1;
377 key.type = BTRFS_EXTENT_ITEM_KEY;
379 mutex_lock(&caching_ctl->mutex);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info->extent_commit_sem);
383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
387 leaf = path->nodes[0];
388 nritems = btrfs_header_nritems(leaf);
391 if (btrfs_fs_closing(fs_info) > 1) {
396 if (path->slots[0] < nritems) {
397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
399 ret = find_next_key(path, 0, &key);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root, path)) {
405 caching_ctl->progress = last;
406 btrfs_release_path(path);
407 up_read(&fs_info->extent_commit_sem);
408 mutex_unlock(&caching_ctl->mutex);
412 leaf = path->nodes[0];
413 nritems = btrfs_header_nritems(leaf);
417 if (key.objectid < block_group->key.objectid) {
422 if (key.objectid >= block_group->key.objectid +
423 block_group->key.offset)
426 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
427 total_found += add_new_free_space(block_group,
430 last = key.objectid + key.offset;
432 if (total_found > (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl->wait);
441 total_found += add_new_free_space(block_group, fs_info, last,
442 block_group->key.objectid +
443 block_group->key.offset);
444 caching_ctl->progress = (u64)-1;
446 spin_lock(&block_group->lock);
447 block_group->caching_ctl = NULL;
448 block_group->cached = BTRFS_CACHE_FINISHED;
449 spin_unlock(&block_group->lock);
452 btrfs_free_path(path);
453 up_read(&fs_info->extent_commit_sem);
455 free_excluded_extents(extent_root, block_group);
457 mutex_unlock(&caching_ctl->mutex);
459 wake_up(&caching_ctl->wait);
461 put_caching_control(caching_ctl);
462 btrfs_put_block_group(block_group);
465 static int cache_block_group(struct btrfs_block_group_cache *cache,
466 struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
471 struct btrfs_fs_info *fs_info = cache->fs_info;
472 struct btrfs_caching_control *caching_ctl;
475 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
476 BUG_ON(!caching_ctl);
478 INIT_LIST_HEAD(&caching_ctl->list);
479 mutex_init(&caching_ctl->mutex);
480 init_waitqueue_head(&caching_ctl->wait);
481 caching_ctl->block_group = cache;
482 caching_ctl->progress = cache->key.objectid;
483 atomic_set(&caching_ctl->count, 1);
484 caching_ctl->work.func = caching_thread;
486 spin_lock(&cache->lock);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache->cached == BTRFS_CACHE_FAST) {
500 struct btrfs_caching_control *ctl;
502 ctl = cache->caching_ctl;
503 atomic_inc(&ctl->count);
504 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
505 spin_unlock(&cache->lock);
509 finish_wait(&ctl->wait, &wait);
510 put_caching_control(ctl);
511 spin_lock(&cache->lock);
514 if (cache->cached != BTRFS_CACHE_NO) {
515 spin_unlock(&cache->lock);
519 WARN_ON(cache->caching_ctl);
520 cache->caching_ctl = caching_ctl;
521 cache->cached = BTRFS_CACHE_FAST;
522 spin_unlock(&cache->lock);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans && (!trans->transaction->in_commit) &&
531 (root && root != root->fs_info->tree_root) &&
532 btrfs_test_opt(root, SPACE_CACHE)) {
533 ret = load_free_space_cache(fs_info, cache);
535 spin_lock(&cache->lock);
537 cache->caching_ctl = NULL;
538 cache->cached = BTRFS_CACHE_FINISHED;
539 cache->last_byte_to_unpin = (u64)-1;
541 if (load_cache_only) {
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_NO;
545 cache->cached = BTRFS_CACHE_STARTED;
548 spin_unlock(&cache->lock);
549 wake_up(&caching_ctl->wait);
551 put_caching_control(caching_ctl);
552 free_excluded_extents(fs_info->extent_root, cache);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache->lock);
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
567 spin_unlock(&cache->lock);
568 wake_up(&caching_ctl->wait);
571 if (load_cache_only) {
572 put_caching_control(caching_ctl);
576 down_write(&fs_info->extent_commit_sem);
577 atomic_inc(&caching_ctl->count);
578 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
579 up_write(&fs_info->extent_commit_sem);
581 btrfs_get_block_group(cache);
583 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache *
592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
594 struct btrfs_block_group_cache *cache;
596 cache = block_group_cache_tree_search(info, bytenr, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
605 struct btrfs_fs_info *info,
608 struct btrfs_block_group_cache *cache;
610 cache = block_group_cache_tree_search(info, bytenr, 1);
615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 struct list_head *head = &info->space_info;
619 struct btrfs_space_info *found;
621 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
622 BTRFS_BLOCK_GROUP_METADATA;
625 list_for_each_entry_rcu(found, head, list) {
626 if (found->flags & flags) {
636 * after adding space to the filesystem, we need to clear the full flags
637 * on all the space infos.
639 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
641 struct list_head *head = &info->space_info;
642 struct btrfs_space_info *found;
645 list_for_each_entry_rcu(found, head, list)
650 static u64 div_factor(u64 num, int factor)
659 static u64 div_factor_fine(u64 num, int factor)
668 u64 btrfs_find_block_group(struct btrfs_root *root,
669 u64 search_start, u64 search_hint, int owner)
671 struct btrfs_block_group_cache *cache;
673 u64 last = max(search_hint, search_start);
680 cache = btrfs_lookup_first_block_group(root->fs_info, last);
684 spin_lock(&cache->lock);
685 last = cache->key.objectid + cache->key.offset;
686 used = btrfs_block_group_used(&cache->item);
688 if ((full_search || !cache->ro) &&
689 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
690 if (used + cache->pinned + cache->reserved <
691 div_factor(cache->key.offset, factor)) {
692 group_start = cache->key.objectid;
693 spin_unlock(&cache->lock);
694 btrfs_put_block_group(cache);
698 spin_unlock(&cache->lock);
699 btrfs_put_block_group(cache);
707 if (!full_search && factor < 10) {
717 /* simple helper to search for an existing extent at a given offset */
718 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
721 struct btrfs_key key;
722 struct btrfs_path *path;
724 path = btrfs_alloc_path();
728 key.objectid = start;
730 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
731 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
733 btrfs_free_path(path);
738 * helper function to lookup reference count and flags of extent.
740 * the head node for delayed ref is used to store the sum of all the
741 * reference count modifications queued up in the rbtree. the head
742 * node may also store the extent flags to set. This way you can check
743 * to see what the reference count and extent flags would be if all of
744 * the delayed refs are not processed.
746 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
747 struct btrfs_root *root, u64 bytenr,
748 u64 num_bytes, u64 *refs, u64 *flags)
750 struct btrfs_delayed_ref_head *head;
751 struct btrfs_delayed_ref_root *delayed_refs;
752 struct btrfs_path *path;
753 struct btrfs_extent_item *ei;
754 struct extent_buffer *leaf;
755 struct btrfs_key key;
761 path = btrfs_alloc_path();
765 key.objectid = bytenr;
766 key.type = BTRFS_EXTENT_ITEM_KEY;
767 key.offset = num_bytes;
769 path->skip_locking = 1;
770 path->search_commit_root = 1;
773 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
779 leaf = path->nodes[0];
780 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
781 if (item_size >= sizeof(*ei)) {
782 ei = btrfs_item_ptr(leaf, path->slots[0],
783 struct btrfs_extent_item);
784 num_refs = btrfs_extent_refs(leaf, ei);
785 extent_flags = btrfs_extent_flags(leaf, ei);
787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
788 struct btrfs_extent_item_v0 *ei0;
789 BUG_ON(item_size != sizeof(*ei0));
790 ei0 = btrfs_item_ptr(leaf, path->slots[0],
791 struct btrfs_extent_item_v0);
792 num_refs = btrfs_extent_refs_v0(leaf, ei0);
793 /* FIXME: this isn't correct for data */
794 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
799 BUG_ON(num_refs == 0);
809 delayed_refs = &trans->transaction->delayed_refs;
810 spin_lock(&delayed_refs->lock);
811 head = btrfs_find_delayed_ref_head(trans, bytenr);
813 if (!mutex_trylock(&head->mutex)) {
814 atomic_inc(&head->node.refs);
815 spin_unlock(&delayed_refs->lock);
817 btrfs_release_path(path);
820 * Mutex was contended, block until it's released and try
823 mutex_lock(&head->mutex);
824 mutex_unlock(&head->mutex);
825 btrfs_put_delayed_ref(&head->node);
828 if (head->extent_op && head->extent_op->update_flags)
829 extent_flags |= head->extent_op->flags_to_set;
831 BUG_ON(num_refs == 0);
833 num_refs += head->node.ref_mod;
834 mutex_unlock(&head->mutex);
836 spin_unlock(&delayed_refs->lock);
838 WARN_ON(num_refs == 0);
842 *flags = extent_flags;
844 btrfs_free_path(path);
849 * Back reference rules. Back refs have three main goals:
851 * 1) differentiate between all holders of references to an extent so that
852 * when a reference is dropped we can make sure it was a valid reference
853 * before freeing the extent.
855 * 2) Provide enough information to quickly find the holders of an extent
856 * if we notice a given block is corrupted or bad.
858 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
859 * maintenance. This is actually the same as #2, but with a slightly
860 * different use case.
862 * There are two kinds of back refs. The implicit back refs is optimized
863 * for pointers in non-shared tree blocks. For a given pointer in a block,
864 * back refs of this kind provide information about the block's owner tree
865 * and the pointer's key. These information allow us to find the block by
866 * b-tree searching. The full back refs is for pointers in tree blocks not
867 * referenced by their owner trees. The location of tree block is recorded
868 * in the back refs. Actually the full back refs is generic, and can be
869 * used in all cases the implicit back refs is used. The major shortcoming
870 * of the full back refs is its overhead. Every time a tree block gets
871 * COWed, we have to update back refs entry for all pointers in it.
873 * For a newly allocated tree block, we use implicit back refs for
874 * pointers in it. This means most tree related operations only involve
875 * implicit back refs. For a tree block created in old transaction, the
876 * only way to drop a reference to it is COW it. So we can detect the
877 * event that tree block loses its owner tree's reference and do the
878 * back refs conversion.
880 * When a tree block is COW'd through a tree, there are four cases:
882 * The reference count of the block is one and the tree is the block's
883 * owner tree. Nothing to do in this case.
885 * The reference count of the block is one and the tree is not the
886 * block's owner tree. In this case, full back refs is used for pointers
887 * in the block. Remove these full back refs, add implicit back refs for
888 * every pointers in the new block.
890 * The reference count of the block is greater than one and the tree is
891 * the block's owner tree. In this case, implicit back refs is used for
892 * pointers in the block. Add full back refs for every pointers in the
893 * block, increase lower level extents' reference counts. The original
894 * implicit back refs are entailed to the new block.
896 * The reference count of the block is greater than one and the tree is
897 * not the block's owner tree. Add implicit back refs for every pointer in
898 * the new block, increase lower level extents' reference count.
900 * Back Reference Key composing:
902 * The key objectid corresponds to the first byte in the extent,
903 * The key type is used to differentiate between types of back refs.
904 * There are different meanings of the key offset for different types
907 * File extents can be referenced by:
909 * - multiple snapshots, subvolumes, or different generations in one subvol
910 * - different files inside a single subvolume
911 * - different offsets inside a file (bookend extents in file.c)
913 * The extent ref structure for the implicit back refs has fields for:
915 * - Objectid of the subvolume root
916 * - objectid of the file holding the reference
917 * - original offset in the file
918 * - how many bookend extents
920 * The key offset for the implicit back refs is hash of the first
923 * The extent ref structure for the full back refs has field for:
925 * - number of pointers in the tree leaf
927 * The key offset for the implicit back refs is the first byte of
930 * When a file extent is allocated, The implicit back refs is used.
931 * the fields are filled in:
933 * (root_key.objectid, inode objectid, offset in file, 1)
935 * When a file extent is removed file truncation, we find the
936 * corresponding implicit back refs and check the following fields:
938 * (btrfs_header_owner(leaf), inode objectid, offset in file)
940 * Btree extents can be referenced by:
942 * - Different subvolumes
944 * Both the implicit back refs and the full back refs for tree blocks
945 * only consist of key. The key offset for the implicit back refs is
946 * objectid of block's owner tree. The key offset for the full back refs
947 * is the first byte of parent block.
949 * When implicit back refs is used, information about the lowest key and
950 * level of the tree block are required. These information are stored in
951 * tree block info structure.
954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
955 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
956 struct btrfs_root *root,
957 struct btrfs_path *path,
958 u64 owner, u32 extra_size)
960 struct btrfs_extent_item *item;
961 struct btrfs_extent_item_v0 *ei0;
962 struct btrfs_extent_ref_v0 *ref0;
963 struct btrfs_tree_block_info *bi;
964 struct extent_buffer *leaf;
965 struct btrfs_key key;
966 struct btrfs_key found_key;
967 u32 new_size = sizeof(*item);
971 leaf = path->nodes[0];
972 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
974 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
975 ei0 = btrfs_item_ptr(leaf, path->slots[0],
976 struct btrfs_extent_item_v0);
977 refs = btrfs_extent_refs_v0(leaf, ei0);
979 if (owner == (u64)-1) {
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
986 leaf = path->nodes[0];
988 btrfs_item_key_to_cpu(leaf, &found_key,
990 BUG_ON(key.objectid != found_key.objectid);
991 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
995 ref0 = btrfs_item_ptr(leaf, path->slots[0],
996 struct btrfs_extent_ref_v0);
997 owner = btrfs_ref_objectid_v0(leaf, ref0);
1001 btrfs_release_path(path);
1003 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1004 new_size += sizeof(*bi);
1006 new_size -= sizeof(*ei0);
1007 ret = btrfs_search_slot(trans, root, &key, path,
1008 new_size + extra_size, 1);
1013 ret = btrfs_extend_item(trans, root, path, new_size);
1015 leaf = path->nodes[0];
1016 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017 btrfs_set_extent_refs(leaf, item, refs);
1018 /* FIXME: get real generation */
1019 btrfs_set_extent_generation(leaf, item, 0);
1020 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1021 btrfs_set_extent_flags(leaf, item,
1022 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1023 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1024 bi = (struct btrfs_tree_block_info *)(item + 1);
1025 /* FIXME: get first key of the block */
1026 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1027 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1029 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1031 btrfs_mark_buffer_dirty(leaf);
1036 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1038 u32 high_crc = ~(u32)0;
1039 u32 low_crc = ~(u32)0;
1042 lenum = cpu_to_le64(root_objectid);
1043 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1044 lenum = cpu_to_le64(owner);
1045 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1046 lenum = cpu_to_le64(offset);
1047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1049 return ((u64)high_crc << 31) ^ (u64)low_crc;
1052 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1053 struct btrfs_extent_data_ref *ref)
1055 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1056 btrfs_extent_data_ref_objectid(leaf, ref),
1057 btrfs_extent_data_ref_offset(leaf, ref));
1060 static int match_extent_data_ref(struct extent_buffer *leaf,
1061 struct btrfs_extent_data_ref *ref,
1062 u64 root_objectid, u64 owner, u64 offset)
1064 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1065 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1066 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1071 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1072 struct btrfs_root *root,
1073 struct btrfs_path *path,
1074 u64 bytenr, u64 parent,
1076 u64 owner, u64 offset)
1078 struct btrfs_key key;
1079 struct btrfs_extent_data_ref *ref;
1080 struct extent_buffer *leaf;
1086 key.objectid = bytenr;
1088 key.type = BTRFS_SHARED_DATA_REF_KEY;
1089 key.offset = parent;
1091 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1092 key.offset = hash_extent_data_ref(root_objectid,
1097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107 key.type = BTRFS_EXTENT_REF_V0_KEY;
1108 btrfs_release_path(path);
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1120 leaf = path->nodes[0];
1121 nritems = btrfs_header_nritems(leaf);
1123 if (path->slots[0] >= nritems) {
1124 ret = btrfs_next_leaf(root, path);
1130 leaf = path->nodes[0];
1131 nritems = btrfs_header_nritems(leaf);
1135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1136 if (key.objectid != bytenr ||
1137 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1140 ref = btrfs_item_ptr(leaf, path->slots[0],
1141 struct btrfs_extent_data_ref);
1143 if (match_extent_data_ref(leaf, ref, root_objectid,
1146 btrfs_release_path(path);
1158 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct btrfs_path *path,
1161 u64 bytenr, u64 parent,
1162 u64 root_objectid, u64 owner,
1163 u64 offset, int refs_to_add)
1165 struct btrfs_key key;
1166 struct extent_buffer *leaf;
1171 key.objectid = bytenr;
1173 key.type = BTRFS_SHARED_DATA_REF_KEY;
1174 key.offset = parent;
1175 size = sizeof(struct btrfs_shared_data_ref);
1177 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1178 key.offset = hash_extent_data_ref(root_objectid,
1180 size = sizeof(struct btrfs_extent_data_ref);
1183 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1184 if (ret && ret != -EEXIST)
1187 leaf = path->nodes[0];
1189 struct btrfs_shared_data_ref *ref;
1190 ref = btrfs_item_ptr(leaf, path->slots[0],
1191 struct btrfs_shared_data_ref);
1193 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1195 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1196 num_refs += refs_to_add;
1197 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1200 struct btrfs_extent_data_ref *ref;
1201 while (ret == -EEXIST) {
1202 ref = btrfs_item_ptr(leaf, path->slots[0],
1203 struct btrfs_extent_data_ref);
1204 if (match_extent_data_ref(leaf, ref, root_objectid,
1207 btrfs_release_path(path);
1209 ret = btrfs_insert_empty_item(trans, root, path, &key,
1211 if (ret && ret != -EEXIST)
1214 leaf = path->nodes[0];
1216 ref = btrfs_item_ptr(leaf, path->slots[0],
1217 struct btrfs_extent_data_ref);
1219 btrfs_set_extent_data_ref_root(leaf, ref,
1221 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1222 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1223 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1225 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1226 num_refs += refs_to_add;
1227 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1230 btrfs_mark_buffer_dirty(leaf);
1233 btrfs_release_path(path);
1237 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1238 struct btrfs_root *root,
1239 struct btrfs_path *path,
1242 struct btrfs_key key;
1243 struct btrfs_extent_data_ref *ref1 = NULL;
1244 struct btrfs_shared_data_ref *ref2 = NULL;
1245 struct extent_buffer *leaf;
1249 leaf = path->nodes[0];
1250 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1255 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1259 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262 struct btrfs_extent_ref_v0 *ref0;
1263 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_extent_ref_v0);
1265 num_refs = btrfs_ref_count_v0(leaf, ref0);
1271 BUG_ON(num_refs < refs_to_drop);
1272 num_refs -= refs_to_drop;
1274 if (num_refs == 0) {
1275 ret = btrfs_del_item(trans, root, path);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1278 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1279 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1280 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 struct btrfs_extent_ref_v0 *ref0;
1284 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_ref_v0);
1286 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1289 btrfs_mark_buffer_dirty(leaf);
1294 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1295 struct btrfs_path *path,
1296 struct btrfs_extent_inline_ref *iref)
1298 struct btrfs_key key;
1299 struct extent_buffer *leaf;
1300 struct btrfs_extent_data_ref *ref1;
1301 struct btrfs_shared_data_ref *ref2;
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1307 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1308 BTRFS_EXTENT_DATA_REF_KEY) {
1309 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1310 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1312 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1313 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1315 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1316 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_data_ref);
1318 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1319 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1320 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1321 struct btrfs_shared_data_ref);
1322 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1325 struct btrfs_extent_ref_v0 *ref0;
1326 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1327 struct btrfs_extent_ref_v0);
1328 num_refs = btrfs_ref_count_v0(leaf, ref0);
1336 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1337 struct btrfs_root *root,
1338 struct btrfs_path *path,
1339 u64 bytenr, u64 parent,
1342 struct btrfs_key key;
1345 key.objectid = bytenr;
1347 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1348 key.offset = parent;
1350 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1351 key.offset = root_objectid;
1354 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (ret == -ENOENT && parent) {
1359 btrfs_release_path(path);
1360 key.type = BTRFS_EXTENT_REF_V0_KEY;
1361 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1369 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1375 struct btrfs_key key;
1378 key.objectid = bytenr;
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1388 btrfs_release_path(path);
1392 static inline int extent_ref_type(u64 parent, u64 owner)
1395 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1397 type = BTRFS_SHARED_BLOCK_REF_KEY;
1399 type = BTRFS_TREE_BLOCK_REF_KEY;
1402 type = BTRFS_SHARED_DATA_REF_KEY;
1404 type = BTRFS_EXTENT_DATA_REF_KEY;
1409 static int find_next_key(struct btrfs_path *path, int level,
1410 struct btrfs_key *key)
1413 for (; level < BTRFS_MAX_LEVEL; level++) {
1414 if (!path->nodes[level])
1416 if (path->slots[level] + 1 >=
1417 btrfs_header_nritems(path->nodes[level]))
1420 btrfs_item_key_to_cpu(path->nodes[level], key,
1421 path->slots[level] + 1);
1423 btrfs_node_key_to_cpu(path->nodes[level], key,
1424 path->slots[level] + 1);
1431 * look for inline back ref. if back ref is found, *ref_ret is set
1432 * to the address of inline back ref, and 0 is returned.
1434 * if back ref isn't found, *ref_ret is set to the address where it
1435 * should be inserted, and -ENOENT is returned.
1437 * if insert is true and there are too many inline back refs, the path
1438 * points to the extent item, and -EAGAIN is returned.
1440 * NOTE: inline back refs are ordered in the same way that back ref
1441 * items in the tree are ordered.
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1445 struct btrfs_root *root,
1446 struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref **ref_ret,
1448 u64 bytenr, u64 num_bytes,
1449 u64 parent, u64 root_objectid,
1450 u64 owner, u64 offset, int insert)
1452 struct btrfs_key key;
1453 struct extent_buffer *leaf;
1454 struct btrfs_extent_item *ei;
1455 struct btrfs_extent_inline_ref *iref;
1466 key.objectid = bytenr;
1467 key.type = BTRFS_EXTENT_ITEM_KEY;
1468 key.offset = num_bytes;
1470 want = extent_ref_type(parent, owner);
1472 extra_size = btrfs_extent_inline_ref_size(want);
1473 path->keep_locks = 1;
1476 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1483 leaf = path->nodes[0];
1484 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486 if (item_size < sizeof(*ei)) {
1491 ret = convert_extent_item_v0(trans, root, path, owner,
1497 leaf = path->nodes[0];
1498 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1501 BUG_ON(item_size < sizeof(*ei));
1503 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1504 flags = btrfs_extent_flags(leaf, ei);
1506 ptr = (unsigned long)(ei + 1);
1507 end = (unsigned long)ei + item_size;
1509 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1510 ptr += sizeof(struct btrfs_tree_block_info);
1513 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1522 iref = (struct btrfs_extent_inline_ref *)ptr;
1523 type = btrfs_extent_inline_ref_type(leaf, iref);
1527 ptr += btrfs_extent_inline_ref_size(type);
1531 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1532 struct btrfs_extent_data_ref *dref;
1533 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1534 if (match_extent_data_ref(leaf, dref, root_objectid,
1539 if (hash_extent_data_ref_item(leaf, dref) <
1540 hash_extent_data_ref(root_objectid, owner, offset))
1544 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1546 if (parent == ref_offset) {
1550 if (ref_offset < parent)
1553 if (root_objectid == ref_offset) {
1557 if (ref_offset < root_objectid)
1561 ptr += btrfs_extent_inline_ref_size(type);
1563 if (err == -ENOENT && insert) {
1564 if (item_size + extra_size >=
1565 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1570 * To add new inline back ref, we have to make sure
1571 * there is no corresponding back ref item.
1572 * For simplicity, we just do not add new inline back
1573 * ref if there is any kind of item for this block
1575 if (find_next_key(path, 0, &key) == 0 &&
1576 key.objectid == bytenr &&
1577 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1582 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1585 path->keep_locks = 0;
1586 btrfs_unlock_up_safe(path, 1);
1592 * helper to add new inline back ref
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1596 struct btrfs_root *root,
1597 struct btrfs_path *path,
1598 struct btrfs_extent_inline_ref *iref,
1599 u64 parent, u64 root_objectid,
1600 u64 owner, u64 offset, int refs_to_add,
1601 struct btrfs_delayed_extent_op *extent_op)
1603 struct extent_buffer *leaf;
1604 struct btrfs_extent_item *ei;
1607 unsigned long item_offset;
1613 leaf = path->nodes[0];
1614 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1615 item_offset = (unsigned long)iref - (unsigned long)ei;
1617 type = extent_ref_type(parent, owner);
1618 size = btrfs_extent_inline_ref_size(type);
1620 ret = btrfs_extend_item(trans, root, path, size);
1622 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1623 refs = btrfs_extent_refs(leaf, ei);
1624 refs += refs_to_add;
1625 btrfs_set_extent_refs(leaf, ei, refs);
1627 __run_delayed_extent_op(extent_op, leaf, ei);
1629 ptr = (unsigned long)ei + item_offset;
1630 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1631 if (ptr < end - size)
1632 memmove_extent_buffer(leaf, ptr + size, ptr,
1635 iref = (struct btrfs_extent_inline_ref *)ptr;
1636 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1637 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1638 struct btrfs_extent_data_ref *dref;
1639 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1640 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1641 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1642 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1643 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1644 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645 struct btrfs_shared_data_ref *sref;
1646 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1647 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1648 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1649 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1650 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1654 btrfs_mark_buffer_dirty(leaf);
1658 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root,
1660 struct btrfs_path *path,
1661 struct btrfs_extent_inline_ref **ref_ret,
1662 u64 bytenr, u64 num_bytes, u64 parent,
1663 u64 root_objectid, u64 owner, u64 offset)
1667 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1668 bytenr, num_bytes, parent,
1669 root_objectid, owner, offset, 0);
1673 btrfs_release_path(path);
1676 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1677 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1680 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1681 root_objectid, owner, offset);
1687 * helper to update/remove inline back ref
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct btrfs_path *path,
1693 struct btrfs_extent_inline_ref *iref,
1695 struct btrfs_delayed_extent_op *extent_op)
1697 struct extent_buffer *leaf;
1698 struct btrfs_extent_item *ei;
1699 struct btrfs_extent_data_ref *dref = NULL;
1700 struct btrfs_shared_data_ref *sref = NULL;
1709 leaf = path->nodes[0];
1710 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711 refs = btrfs_extent_refs(leaf, ei);
1712 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713 refs += refs_to_mod;
1714 btrfs_set_extent_refs(leaf, ei, refs);
1716 __run_delayed_extent_op(extent_op, leaf, ei);
1718 type = btrfs_extent_inline_ref_type(leaf, iref);
1720 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722 refs = btrfs_extent_data_ref_count(leaf, dref);
1723 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725 refs = btrfs_shared_data_ref_count(leaf, sref);
1728 BUG_ON(refs_to_mod != -1);
1731 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732 refs += refs_to_mod;
1735 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1738 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1740 size = btrfs_extent_inline_ref_size(type);
1741 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742 ptr = (unsigned long)iref;
1743 end = (unsigned long)ei + item_size;
1744 if (ptr + size < end)
1745 memmove_extent_buffer(leaf, ptr, ptr + size,
1748 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1750 btrfs_mark_buffer_dirty(leaf);
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_path *path,
1758 u64 bytenr, u64 num_bytes, u64 parent,
1759 u64 root_objectid, u64 owner,
1760 u64 offset, int refs_to_add,
1761 struct btrfs_delayed_extent_op *extent_op)
1763 struct btrfs_extent_inline_ref *iref;
1766 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1767 bytenr, num_bytes, parent,
1768 root_objectid, owner, offset, 1);
1770 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1771 ret = update_inline_extent_backref(trans, root, path, iref,
1772 refs_to_add, extent_op);
1773 } else if (ret == -ENOENT) {
1774 ret = setup_inline_extent_backref(trans, root, path, iref,
1775 parent, root_objectid,
1776 owner, offset, refs_to_add,
1782 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1783 struct btrfs_root *root,
1784 struct btrfs_path *path,
1785 u64 bytenr, u64 parent, u64 root_objectid,
1786 u64 owner, u64 offset, int refs_to_add)
1789 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790 BUG_ON(refs_to_add != 1);
1791 ret = insert_tree_block_ref(trans, root, path, bytenr,
1792 parent, root_objectid);
1794 ret = insert_extent_data_ref(trans, root, path, bytenr,
1795 parent, root_objectid,
1796 owner, offset, refs_to_add);
1801 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1802 struct btrfs_root *root,
1803 struct btrfs_path *path,
1804 struct btrfs_extent_inline_ref *iref,
1805 int refs_to_drop, int is_data)
1809 BUG_ON(!is_data && refs_to_drop != 1);
1811 ret = update_inline_extent_backref(trans, root, path, iref,
1812 -refs_to_drop, NULL);
1813 } else if (is_data) {
1814 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1816 ret = btrfs_del_item(trans, root, path);
1821 static int btrfs_issue_discard(struct block_device *bdev,
1824 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1827 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1828 u64 num_bytes, u64 *actual_bytes)
1831 u64 discarded_bytes = 0;
1832 struct btrfs_bio *bbio = NULL;
1835 /* Tell the block device(s) that the sectors can be discarded */
1836 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1837 bytenr, &num_bytes, &bbio, 0);
1839 struct btrfs_bio_stripe *stripe = bbio->stripes;
1843 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844 if (!stripe->dev->can_discard)
1847 ret = btrfs_issue_discard(stripe->dev->bdev,
1851 discarded_bytes += stripe->length;
1852 else if (ret != -EOPNOTSUPP)
1856 * Just in case we get back EOPNOTSUPP for some reason,
1857 * just ignore the return value so we don't screw up
1858 * people calling discard_extent.
1866 *actual_bytes = discarded_bytes;
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1873 struct btrfs_root *root,
1874 u64 bytenr, u64 num_bytes, u64 parent,
1875 u64 root_objectid, u64 owner, u64 offset)
1878 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1879 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1881 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1882 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1883 parent, root_objectid, (int)owner,
1884 BTRFS_ADD_DELAYED_REF, NULL);
1886 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1887 parent, root_objectid, owner, offset,
1888 BTRFS_ADD_DELAYED_REF, NULL);
1893 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1894 struct btrfs_root *root,
1895 u64 bytenr, u64 num_bytes,
1896 u64 parent, u64 root_objectid,
1897 u64 owner, u64 offset, int refs_to_add,
1898 struct btrfs_delayed_extent_op *extent_op)
1900 struct btrfs_path *path;
1901 struct extent_buffer *leaf;
1902 struct btrfs_extent_item *item;
1907 path = btrfs_alloc_path();
1912 path->leave_spinning = 1;
1913 /* this will setup the path even if it fails to insert the back ref */
1914 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1915 path, bytenr, num_bytes, parent,
1916 root_objectid, owner, offset,
1917 refs_to_add, extent_op);
1921 if (ret != -EAGAIN) {
1926 leaf = path->nodes[0];
1927 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1928 refs = btrfs_extent_refs(leaf, item);
1929 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1931 __run_delayed_extent_op(extent_op, leaf, item);
1933 btrfs_mark_buffer_dirty(leaf);
1934 btrfs_release_path(path);
1937 path->leave_spinning = 1;
1939 /* now insert the actual backref */
1940 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1941 path, bytenr, parent, root_objectid,
1942 owner, offset, refs_to_add);
1945 btrfs_free_path(path);
1949 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1950 struct btrfs_root *root,
1951 struct btrfs_delayed_ref_node *node,
1952 struct btrfs_delayed_extent_op *extent_op,
1953 int insert_reserved)
1956 struct btrfs_delayed_data_ref *ref;
1957 struct btrfs_key ins;
1962 ins.objectid = node->bytenr;
1963 ins.offset = node->num_bytes;
1964 ins.type = BTRFS_EXTENT_ITEM_KEY;
1966 ref = btrfs_delayed_node_to_data_ref(node);
1967 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1968 parent = ref->parent;
1970 ref_root = ref->root;
1972 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1974 BUG_ON(extent_op->update_key);
1975 flags |= extent_op->flags_to_set;
1977 ret = alloc_reserved_file_extent(trans, root,
1978 parent, ref_root, flags,
1979 ref->objectid, ref->offset,
1980 &ins, node->ref_mod);
1981 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1982 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1983 node->num_bytes, parent,
1984 ref_root, ref->objectid,
1985 ref->offset, node->ref_mod,
1987 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1988 ret = __btrfs_free_extent(trans, root, node->bytenr,
1989 node->num_bytes, parent,
1990 ref_root, ref->objectid,
1991 ref->offset, node->ref_mod,
1999 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2000 struct extent_buffer *leaf,
2001 struct btrfs_extent_item *ei)
2003 u64 flags = btrfs_extent_flags(leaf, ei);
2004 if (extent_op->update_flags) {
2005 flags |= extent_op->flags_to_set;
2006 btrfs_set_extent_flags(leaf, ei, flags);
2009 if (extent_op->update_key) {
2010 struct btrfs_tree_block_info *bi;
2011 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2012 bi = (struct btrfs_tree_block_info *)(ei + 1);
2013 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2017 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2018 struct btrfs_root *root,
2019 struct btrfs_delayed_ref_node *node,
2020 struct btrfs_delayed_extent_op *extent_op)
2022 struct btrfs_key key;
2023 struct btrfs_path *path;
2024 struct btrfs_extent_item *ei;
2025 struct extent_buffer *leaf;
2030 path = btrfs_alloc_path();
2034 key.objectid = node->bytenr;
2035 key.type = BTRFS_EXTENT_ITEM_KEY;
2036 key.offset = node->num_bytes;
2039 path->leave_spinning = 1;
2040 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2051 leaf = path->nodes[0];
2052 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2054 if (item_size < sizeof(*ei)) {
2055 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2061 leaf = path->nodes[0];
2062 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2065 BUG_ON(item_size < sizeof(*ei));
2066 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2067 __run_delayed_extent_op(extent_op, leaf, ei);
2069 btrfs_mark_buffer_dirty(leaf);
2071 btrfs_free_path(path);
2075 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2076 struct btrfs_root *root,
2077 struct btrfs_delayed_ref_node *node,
2078 struct btrfs_delayed_extent_op *extent_op,
2079 int insert_reserved)
2082 struct btrfs_delayed_tree_ref *ref;
2083 struct btrfs_key ins;
2087 ins.objectid = node->bytenr;
2088 ins.offset = node->num_bytes;
2089 ins.type = BTRFS_EXTENT_ITEM_KEY;
2091 ref = btrfs_delayed_node_to_tree_ref(node);
2092 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2093 parent = ref->parent;
2095 ref_root = ref->root;
2097 BUG_ON(node->ref_mod != 1);
2098 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2099 BUG_ON(!extent_op || !extent_op->update_flags ||
2100 !extent_op->update_key);
2101 ret = alloc_reserved_tree_block(trans, root,
2103 extent_op->flags_to_set,
2106 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2107 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2108 node->num_bytes, parent, ref_root,
2109 ref->level, 0, 1, extent_op);
2110 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2111 ret = __btrfs_free_extent(trans, root, node->bytenr,
2112 node->num_bytes, parent, ref_root,
2113 ref->level, 0, 1, extent_op);
2120 /* helper function to actually process a single delayed ref entry */
2121 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2122 struct btrfs_root *root,
2123 struct btrfs_delayed_ref_node *node,
2124 struct btrfs_delayed_extent_op *extent_op,
2125 int insert_reserved)
2128 if (btrfs_delayed_ref_is_head(node)) {
2129 struct btrfs_delayed_ref_head *head;
2131 * we've hit the end of the chain and we were supposed
2132 * to insert this extent into the tree. But, it got
2133 * deleted before we ever needed to insert it, so all
2134 * we have to do is clean up the accounting
2137 head = btrfs_delayed_node_to_head(node);
2138 if (insert_reserved) {
2139 btrfs_pin_extent(root, node->bytenr,
2140 node->num_bytes, 1);
2141 if (head->is_data) {
2142 ret = btrfs_del_csums(trans, root,
2148 mutex_unlock(&head->mutex);
2152 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2153 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2154 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2156 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2157 node->type == BTRFS_SHARED_DATA_REF_KEY)
2158 ret = run_delayed_data_ref(trans, root, node, extent_op,
2165 static noinline struct btrfs_delayed_ref_node *
2166 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2168 struct rb_node *node;
2169 struct btrfs_delayed_ref_node *ref;
2170 int action = BTRFS_ADD_DELAYED_REF;
2173 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2174 * this prevents ref count from going down to zero when
2175 * there still are pending delayed ref.
2177 node = rb_prev(&head->node.rb_node);
2181 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2183 if (ref->bytenr != head->node.bytenr)
2185 if (ref->action == action)
2187 node = rb_prev(node);
2189 if (action == BTRFS_ADD_DELAYED_REF) {
2190 action = BTRFS_DROP_DELAYED_REF;
2196 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 struct list_head *cluster)
2200 struct btrfs_delayed_ref_root *delayed_refs;
2201 struct btrfs_delayed_ref_node *ref;
2202 struct btrfs_delayed_ref_head *locked_ref = NULL;
2203 struct btrfs_delayed_extent_op *extent_op;
2206 int must_insert_reserved = 0;
2208 delayed_refs = &trans->transaction->delayed_refs;
2211 /* pick a new head ref from the cluster list */
2212 if (list_empty(cluster))
2215 locked_ref = list_entry(cluster->next,
2216 struct btrfs_delayed_ref_head, cluster);
2218 /* grab the lock that says we are going to process
2219 * all the refs for this head */
2220 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2223 * we may have dropped the spin lock to get the head
2224 * mutex lock, and that might have given someone else
2225 * time to free the head. If that's true, it has been
2226 * removed from our list and we can move on.
2228 if (ret == -EAGAIN) {
2236 * record the must insert reserved flag before we
2237 * drop the spin lock.
2239 must_insert_reserved = locked_ref->must_insert_reserved;
2240 locked_ref->must_insert_reserved = 0;
2242 extent_op = locked_ref->extent_op;
2243 locked_ref->extent_op = NULL;
2246 * locked_ref is the head node, so we have to go one
2247 * node back for any delayed ref updates
2249 ref = select_delayed_ref(locked_ref);
2251 /* All delayed refs have been processed, Go ahead
2252 * and send the head node to run_one_delayed_ref,
2253 * so that any accounting fixes can happen
2255 ref = &locked_ref->node;
2257 if (extent_op && must_insert_reserved) {
2263 spin_unlock(&delayed_refs->lock);
2265 ret = run_delayed_extent_op(trans, root,
2273 list_del_init(&locked_ref->cluster);
2278 rb_erase(&ref->rb_node, &delayed_refs->root);
2279 delayed_refs->num_entries--;
2281 spin_unlock(&delayed_refs->lock);
2283 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2284 must_insert_reserved);
2287 btrfs_put_delayed_ref(ref);
2291 do_chunk_alloc(trans, root->fs_info->extent_root,
2293 btrfs_get_alloc_profile(root, 0),
2294 CHUNK_ALLOC_NO_FORCE);
2296 spin_lock(&delayed_refs->lock);
2302 * this starts processing the delayed reference count updates and
2303 * extent insertions we have queued up so far. count can be
2304 * 0, which means to process everything in the tree at the start
2305 * of the run (but not newly added entries), or it can be some target
2306 * number you'd like to process.
2308 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2309 struct btrfs_root *root, unsigned long count)
2311 struct rb_node *node;
2312 struct btrfs_delayed_ref_root *delayed_refs;
2313 struct btrfs_delayed_ref_node *ref;
2314 struct list_head cluster;
2316 int run_all = count == (unsigned long)-1;
2319 if (root == root->fs_info->extent_root)
2320 root = root->fs_info->tree_root;
2322 do_chunk_alloc(trans, root->fs_info->extent_root,
2323 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2324 CHUNK_ALLOC_NO_FORCE);
2326 delayed_refs = &trans->transaction->delayed_refs;
2327 INIT_LIST_HEAD(&cluster);
2329 spin_lock(&delayed_refs->lock);
2331 count = delayed_refs->num_entries * 2;
2335 if (!(run_all || run_most) &&
2336 delayed_refs->num_heads_ready < 64)
2340 * go find something we can process in the rbtree. We start at
2341 * the beginning of the tree, and then build a cluster
2342 * of refs to process starting at the first one we are able to
2345 ret = btrfs_find_ref_cluster(trans, &cluster,
2346 delayed_refs->run_delayed_start);
2350 ret = run_clustered_refs(trans, root, &cluster);
2353 count -= min_t(unsigned long, ret, count);
2360 node = rb_first(&delayed_refs->root);
2363 count = (unsigned long)-1;
2366 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2368 if (btrfs_delayed_ref_is_head(ref)) {
2369 struct btrfs_delayed_ref_head *head;
2371 head = btrfs_delayed_node_to_head(ref);
2372 atomic_inc(&ref->refs);
2374 spin_unlock(&delayed_refs->lock);
2376 * Mutex was contended, block until it's
2377 * released and try again
2379 mutex_lock(&head->mutex);
2380 mutex_unlock(&head->mutex);
2382 btrfs_put_delayed_ref(ref);
2386 node = rb_next(node);
2388 spin_unlock(&delayed_refs->lock);
2389 schedule_timeout(1);
2393 spin_unlock(&delayed_refs->lock);
2397 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2398 struct btrfs_root *root,
2399 u64 bytenr, u64 num_bytes, u64 flags,
2402 struct btrfs_delayed_extent_op *extent_op;
2405 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2409 extent_op->flags_to_set = flags;
2410 extent_op->update_flags = 1;
2411 extent_op->update_key = 0;
2412 extent_op->is_data = is_data ? 1 : 0;
2414 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2420 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2421 struct btrfs_root *root,
2422 struct btrfs_path *path,
2423 u64 objectid, u64 offset, u64 bytenr)
2425 struct btrfs_delayed_ref_head *head;
2426 struct btrfs_delayed_ref_node *ref;
2427 struct btrfs_delayed_data_ref *data_ref;
2428 struct btrfs_delayed_ref_root *delayed_refs;
2429 struct rb_node *node;
2433 delayed_refs = &trans->transaction->delayed_refs;
2434 spin_lock(&delayed_refs->lock);
2435 head = btrfs_find_delayed_ref_head(trans, bytenr);
2439 if (!mutex_trylock(&head->mutex)) {
2440 atomic_inc(&head->node.refs);
2441 spin_unlock(&delayed_refs->lock);
2443 btrfs_release_path(path);
2446 * Mutex was contended, block until it's released and let
2449 mutex_lock(&head->mutex);
2450 mutex_unlock(&head->mutex);
2451 btrfs_put_delayed_ref(&head->node);
2455 node = rb_prev(&head->node.rb_node);
2459 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2461 if (ref->bytenr != bytenr)
2465 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2468 data_ref = btrfs_delayed_node_to_data_ref(ref);
2470 node = rb_prev(node);
2472 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2473 if (ref->bytenr == bytenr)
2477 if (data_ref->root != root->root_key.objectid ||
2478 data_ref->objectid != objectid || data_ref->offset != offset)
2483 mutex_unlock(&head->mutex);
2485 spin_unlock(&delayed_refs->lock);
2489 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2490 struct btrfs_root *root,
2491 struct btrfs_path *path,
2492 u64 objectid, u64 offset, u64 bytenr)
2494 struct btrfs_root *extent_root = root->fs_info->extent_root;
2495 struct extent_buffer *leaf;
2496 struct btrfs_extent_data_ref *ref;
2497 struct btrfs_extent_inline_ref *iref;
2498 struct btrfs_extent_item *ei;
2499 struct btrfs_key key;
2503 key.objectid = bytenr;
2504 key.offset = (u64)-1;
2505 key.type = BTRFS_EXTENT_ITEM_KEY;
2507 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2513 if (path->slots[0] == 0)
2517 leaf = path->nodes[0];
2518 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2520 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2524 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2525 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2526 if (item_size < sizeof(*ei)) {
2527 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2531 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2533 if (item_size != sizeof(*ei) +
2534 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2537 if (btrfs_extent_generation(leaf, ei) <=
2538 btrfs_root_last_snapshot(&root->root_item))
2541 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2542 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2543 BTRFS_EXTENT_DATA_REF_KEY)
2546 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2547 if (btrfs_extent_refs(leaf, ei) !=
2548 btrfs_extent_data_ref_count(leaf, ref) ||
2549 btrfs_extent_data_ref_root(leaf, ref) !=
2550 root->root_key.objectid ||
2551 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2552 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2560 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2561 struct btrfs_root *root,
2562 u64 objectid, u64 offset, u64 bytenr)
2564 struct btrfs_path *path;
2568 path = btrfs_alloc_path();
2573 ret = check_committed_ref(trans, root, path, objectid,
2575 if (ret && ret != -ENOENT)
2578 ret2 = check_delayed_ref(trans, root, path, objectid,
2580 } while (ret2 == -EAGAIN);
2582 if (ret2 && ret2 != -ENOENT) {
2587 if (ret != -ENOENT || ret2 != -ENOENT)
2590 btrfs_free_path(path);
2591 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2596 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2597 struct btrfs_root *root,
2598 struct extent_buffer *buf,
2599 int full_backref, int inc)
2606 struct btrfs_key key;
2607 struct btrfs_file_extent_item *fi;
2611 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2612 u64, u64, u64, u64, u64, u64);
2614 ref_root = btrfs_header_owner(buf);
2615 nritems = btrfs_header_nritems(buf);
2616 level = btrfs_header_level(buf);
2618 if (!root->ref_cows && level == 0)
2622 process_func = btrfs_inc_extent_ref;
2624 process_func = btrfs_free_extent;
2627 parent = buf->start;
2631 for (i = 0; i < nritems; i++) {
2633 btrfs_item_key_to_cpu(buf, &key, i);
2634 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2636 fi = btrfs_item_ptr(buf, i,
2637 struct btrfs_file_extent_item);
2638 if (btrfs_file_extent_type(buf, fi) ==
2639 BTRFS_FILE_EXTENT_INLINE)
2641 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2645 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2646 key.offset -= btrfs_file_extent_offset(buf, fi);
2647 ret = process_func(trans, root, bytenr, num_bytes,
2648 parent, ref_root, key.objectid,
2653 bytenr = btrfs_node_blockptr(buf, i);
2654 num_bytes = btrfs_level_size(root, level - 1);
2655 ret = process_func(trans, root, bytenr, num_bytes,
2656 parent, ref_root, level - 1, 0);
2667 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2668 struct extent_buffer *buf, int full_backref)
2670 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2673 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2674 struct extent_buffer *buf, int full_backref)
2676 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2679 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2680 struct btrfs_root *root,
2681 struct btrfs_path *path,
2682 struct btrfs_block_group_cache *cache)
2685 struct btrfs_root *extent_root = root->fs_info->extent_root;
2687 struct extent_buffer *leaf;
2689 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2694 leaf = path->nodes[0];
2695 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2696 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2697 btrfs_mark_buffer_dirty(leaf);
2698 btrfs_release_path(path);
2706 static struct btrfs_block_group_cache *
2707 next_block_group(struct btrfs_root *root,
2708 struct btrfs_block_group_cache *cache)
2710 struct rb_node *node;
2711 spin_lock(&root->fs_info->block_group_cache_lock);
2712 node = rb_next(&cache->cache_node);
2713 btrfs_put_block_group(cache);
2715 cache = rb_entry(node, struct btrfs_block_group_cache,
2717 btrfs_get_block_group(cache);
2720 spin_unlock(&root->fs_info->block_group_cache_lock);
2724 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2725 struct btrfs_trans_handle *trans,
2726 struct btrfs_path *path)
2728 struct btrfs_root *root = block_group->fs_info->tree_root;
2729 struct inode *inode = NULL;
2731 int dcs = BTRFS_DC_ERROR;
2737 * If this block group is smaller than 100 megs don't bother caching the
2740 if (block_group->key.offset < (100 * 1024 * 1024)) {
2741 spin_lock(&block_group->lock);
2742 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2743 spin_unlock(&block_group->lock);
2748 inode = lookup_free_space_inode(root, block_group, path);
2749 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2750 ret = PTR_ERR(inode);
2751 btrfs_release_path(path);
2755 if (IS_ERR(inode)) {
2759 if (block_group->ro)
2762 ret = create_free_space_inode(root, trans, block_group, path);
2768 /* We've already setup this transaction, go ahead and exit */
2769 if (block_group->cache_generation == trans->transid &&
2770 i_size_read(inode)) {
2771 dcs = BTRFS_DC_SETUP;
2776 * We want to set the generation to 0, that way if anything goes wrong
2777 * from here on out we know not to trust this cache when we load up next
2780 BTRFS_I(inode)->generation = 0;
2781 ret = btrfs_update_inode(trans, root, inode);
2784 if (i_size_read(inode) > 0) {
2785 ret = btrfs_truncate_free_space_cache(root, trans, path,
2791 spin_lock(&block_group->lock);
2792 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2793 /* We're not cached, don't bother trying to write stuff out */
2794 dcs = BTRFS_DC_WRITTEN;
2795 spin_unlock(&block_group->lock);
2798 spin_unlock(&block_group->lock);
2800 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2805 * Just to make absolutely sure we have enough space, we're going to
2806 * preallocate 12 pages worth of space for each block group. In
2807 * practice we ought to use at most 8, but we need extra space so we can
2808 * add our header and have a terminator between the extents and the
2812 num_pages *= PAGE_CACHE_SIZE;
2814 ret = btrfs_check_data_free_space(inode, num_pages);
2818 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2819 num_pages, num_pages,
2822 dcs = BTRFS_DC_SETUP;
2823 btrfs_free_reserved_data_space(inode, num_pages);
2828 btrfs_release_path(path);
2830 spin_lock(&block_group->lock);
2831 if (!ret && dcs == BTRFS_DC_SETUP)
2832 block_group->cache_generation = trans->transid;
2833 block_group->disk_cache_state = dcs;
2834 spin_unlock(&block_group->lock);
2839 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2840 struct btrfs_root *root)
2842 struct btrfs_block_group_cache *cache;
2844 struct btrfs_path *path;
2847 path = btrfs_alloc_path();
2853 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2855 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2857 cache = next_block_group(root, cache);
2865 err = cache_save_setup(cache, trans, path);
2866 last = cache->key.objectid + cache->key.offset;
2867 btrfs_put_block_group(cache);
2872 err = btrfs_run_delayed_refs(trans, root,
2877 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2879 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2880 btrfs_put_block_group(cache);
2886 cache = next_block_group(root, cache);
2895 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2896 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2898 last = cache->key.objectid + cache->key.offset;
2900 err = write_one_cache_group(trans, root, path, cache);
2902 btrfs_put_block_group(cache);
2907 * I don't think this is needed since we're just marking our
2908 * preallocated extent as written, but just in case it can't
2912 err = btrfs_run_delayed_refs(trans, root,
2917 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2920 * Really this shouldn't happen, but it could if we
2921 * couldn't write the entire preallocated extent and
2922 * splitting the extent resulted in a new block.
2925 btrfs_put_block_group(cache);
2928 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2930 cache = next_block_group(root, cache);
2939 btrfs_write_out_cache(root, trans, cache, path);
2942 * If we didn't have an error then the cache state is still
2943 * NEED_WRITE, so we can set it to WRITTEN.
2945 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2946 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2947 last = cache->key.objectid + cache->key.offset;
2948 btrfs_put_block_group(cache);
2951 btrfs_free_path(path);
2955 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2957 struct btrfs_block_group_cache *block_group;
2960 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2961 if (!block_group || block_group->ro)
2964 btrfs_put_block_group(block_group);
2968 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2969 u64 total_bytes, u64 bytes_used,
2970 struct btrfs_space_info **space_info)
2972 struct btrfs_space_info *found;
2976 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2977 BTRFS_BLOCK_GROUP_RAID10))
2982 found = __find_space_info(info, flags);
2984 spin_lock(&found->lock);
2985 found->total_bytes += total_bytes;
2986 found->disk_total += total_bytes * factor;
2987 found->bytes_used += bytes_used;
2988 found->disk_used += bytes_used * factor;
2990 spin_unlock(&found->lock);
2991 *space_info = found;
2994 found = kzalloc(sizeof(*found), GFP_NOFS);
2998 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2999 INIT_LIST_HEAD(&found->block_groups[i]);
3000 init_rwsem(&found->groups_sem);
3001 spin_lock_init(&found->lock);
3002 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3003 BTRFS_BLOCK_GROUP_SYSTEM |
3004 BTRFS_BLOCK_GROUP_METADATA);
3005 found->total_bytes = total_bytes;
3006 found->disk_total = total_bytes * factor;
3007 found->bytes_used = bytes_used;
3008 found->disk_used = bytes_used * factor;
3009 found->bytes_pinned = 0;
3010 found->bytes_reserved = 0;
3011 found->bytes_readonly = 0;
3012 found->bytes_may_use = 0;
3014 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3015 found->chunk_alloc = 0;
3017 init_waitqueue_head(&found->wait);
3018 *space_info = found;
3019 list_add_rcu(&found->list, &info->space_info);
3023 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3025 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3026 BTRFS_BLOCK_GROUP_RAID1 |
3027 BTRFS_BLOCK_GROUP_RAID10 |
3028 BTRFS_BLOCK_GROUP_DUP);
3030 if (flags & BTRFS_BLOCK_GROUP_DATA)
3031 fs_info->avail_data_alloc_bits |= extra_flags;
3032 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3033 fs_info->avail_metadata_alloc_bits |= extra_flags;
3034 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3035 fs_info->avail_system_alloc_bits |= extra_flags;
3039 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3042 * we add in the count of missing devices because we want
3043 * to make sure that any RAID levels on a degraded FS
3044 * continue to be honored.
3046 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3047 root->fs_info->fs_devices->missing_devices;
3049 if (num_devices == 1)
3050 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3051 if (num_devices < 4)
3052 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3054 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3055 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3056 BTRFS_BLOCK_GROUP_RAID10))) {
3057 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3060 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3061 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3062 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3065 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3066 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3067 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3068 (flags & BTRFS_BLOCK_GROUP_DUP)))
3069 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3073 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3075 if (flags & BTRFS_BLOCK_GROUP_DATA)
3076 flags |= root->fs_info->avail_data_alloc_bits &
3077 root->fs_info->data_alloc_profile;
3078 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3079 flags |= root->fs_info->avail_system_alloc_bits &
3080 root->fs_info->system_alloc_profile;
3081 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3082 flags |= root->fs_info->avail_metadata_alloc_bits &
3083 root->fs_info->metadata_alloc_profile;
3084 return btrfs_reduce_alloc_profile(root, flags);
3087 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3092 flags = BTRFS_BLOCK_GROUP_DATA;
3093 else if (root == root->fs_info->chunk_root)
3094 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3096 flags = BTRFS_BLOCK_GROUP_METADATA;
3098 return get_alloc_profile(root, flags);
3101 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3103 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3104 BTRFS_BLOCK_GROUP_DATA);
3108 * This will check the space that the inode allocates from to make sure we have
3109 * enough space for bytes.
3111 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3113 struct btrfs_space_info *data_sinfo;
3114 struct btrfs_root *root = BTRFS_I(inode)->root;
3116 int ret = 0, committed = 0, alloc_chunk = 1;
3118 /* make sure bytes are sectorsize aligned */
3119 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3121 if (root == root->fs_info->tree_root ||
3122 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3127 data_sinfo = BTRFS_I(inode)->space_info;
3132 /* make sure we have enough space to handle the data first */
3133 spin_lock(&data_sinfo->lock);
3134 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3135 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3136 data_sinfo->bytes_may_use;
3138 if (used + bytes > data_sinfo->total_bytes) {
3139 struct btrfs_trans_handle *trans;
3142 * if we don't have enough free bytes in this space then we need
3143 * to alloc a new chunk.
3145 if (!data_sinfo->full && alloc_chunk) {
3148 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3149 spin_unlock(&data_sinfo->lock);
3151 alloc_target = btrfs_get_alloc_profile(root, 1);
3152 trans = btrfs_join_transaction(root);
3154 return PTR_ERR(trans);
3156 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3157 bytes + 2 * 1024 * 1024,
3159 CHUNK_ALLOC_NO_FORCE);
3160 btrfs_end_transaction(trans, root);
3169 btrfs_set_inode_space_info(root, inode);
3170 data_sinfo = BTRFS_I(inode)->space_info;
3176 * If we have less pinned bytes than we want to allocate then
3177 * don't bother committing the transaction, it won't help us.
3179 if (data_sinfo->bytes_pinned < bytes)
3181 spin_unlock(&data_sinfo->lock);
3183 /* commit the current transaction and try again */
3186 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3188 trans = btrfs_join_transaction(root);
3190 return PTR_ERR(trans);
3191 ret = btrfs_commit_transaction(trans, root);
3199 data_sinfo->bytes_may_use += bytes;
3200 spin_unlock(&data_sinfo->lock);
3206 * Called if we need to clear a data reservation for this inode.
3208 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3210 struct btrfs_root *root = BTRFS_I(inode)->root;
3211 struct btrfs_space_info *data_sinfo;
3213 /* make sure bytes are sectorsize aligned */
3214 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3216 data_sinfo = BTRFS_I(inode)->space_info;
3217 spin_lock(&data_sinfo->lock);
3218 data_sinfo->bytes_may_use -= bytes;
3219 spin_unlock(&data_sinfo->lock);
3222 static void force_metadata_allocation(struct btrfs_fs_info *info)
3224 struct list_head *head = &info->space_info;
3225 struct btrfs_space_info *found;
3228 list_for_each_entry_rcu(found, head, list) {
3229 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3230 found->force_alloc = CHUNK_ALLOC_FORCE;
3235 static int should_alloc_chunk(struct btrfs_root *root,
3236 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3239 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3240 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3241 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3244 if (force == CHUNK_ALLOC_FORCE)
3248 * We need to take into account the global rsv because for all intents
3249 * and purposes it's used space. Don't worry about locking the
3250 * global_rsv, it doesn't change except when the transaction commits.
3252 num_allocated += global_rsv->size;
3255 * in limited mode, we want to have some free space up to
3256 * about 1% of the FS size.
3258 if (force == CHUNK_ALLOC_LIMITED) {
3259 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3260 thresh = max_t(u64, 64 * 1024 * 1024,
3261 div_factor_fine(thresh, 1));
3263 if (num_bytes - num_allocated < thresh)
3266 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3268 /* 256MB or 2% of the FS */
3269 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3271 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3276 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3277 struct btrfs_root *extent_root, u64 alloc_bytes,
3278 u64 flags, int force)
3280 struct btrfs_space_info *space_info;
3281 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3282 int wait_for_alloc = 0;
3285 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3287 space_info = __find_space_info(extent_root->fs_info, flags);
3289 ret = update_space_info(extent_root->fs_info, flags,
3293 BUG_ON(!space_info);
3296 spin_lock(&space_info->lock);
3297 if (space_info->force_alloc)
3298 force = space_info->force_alloc;
3299 if (space_info->full) {
3300 spin_unlock(&space_info->lock);
3304 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3305 spin_unlock(&space_info->lock);
3307 } else if (space_info->chunk_alloc) {
3310 space_info->chunk_alloc = 1;
3313 spin_unlock(&space_info->lock);
3315 mutex_lock(&fs_info->chunk_mutex);
3318 * The chunk_mutex is held throughout the entirety of a chunk
3319 * allocation, so once we've acquired the chunk_mutex we know that the
3320 * other guy is done and we need to recheck and see if we should
3323 if (wait_for_alloc) {
3324 mutex_unlock(&fs_info->chunk_mutex);
3330 * If we have mixed data/metadata chunks we want to make sure we keep
3331 * allocating mixed chunks instead of individual chunks.
3333 if (btrfs_mixed_space_info(space_info))
3334 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3337 * if we're doing a data chunk, go ahead and make sure that
3338 * we keep a reasonable number of metadata chunks allocated in the
3341 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3342 fs_info->data_chunk_allocations++;
3343 if (!(fs_info->data_chunk_allocations %
3344 fs_info->metadata_ratio))
3345 force_metadata_allocation(fs_info);
3348 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3349 if (ret < 0 && ret != -ENOSPC)
3352 spin_lock(&space_info->lock);
3354 space_info->full = 1;
3358 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3359 space_info->chunk_alloc = 0;
3360 spin_unlock(&space_info->lock);
3362 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3367 * shrink metadata reservation for delalloc
3369 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3372 struct btrfs_block_rsv *block_rsv;
3373 struct btrfs_space_info *space_info;
3374 struct btrfs_trans_handle *trans;
3379 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3381 unsigned long progress;
3383 trans = (struct btrfs_trans_handle *)current->journal_info;
3384 block_rsv = &root->fs_info->delalloc_block_rsv;
3385 space_info = block_rsv->space_info;
3388 reserved = space_info->bytes_may_use;
3389 progress = space_info->reservation_progress;
3395 if (root->fs_info->delalloc_bytes == 0) {
3398 btrfs_wait_ordered_extents(root, 0, 0);
3402 max_reclaim = min(reserved, to_reclaim);
3403 nr_pages = max_t(unsigned long, nr_pages,
3404 max_reclaim >> PAGE_CACHE_SHIFT);
3405 while (loops < 1024) {
3406 /* have the flusher threads jump in and do some IO */
3408 nr_pages = min_t(unsigned long, nr_pages,
3409 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3410 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3411 WB_REASON_FS_FREE_SPACE);
3413 spin_lock(&space_info->lock);
3414 if (reserved > space_info->bytes_may_use)
3415 reclaimed += reserved - space_info->bytes_may_use;
3416 reserved = space_info->bytes_may_use;
3417 spin_unlock(&space_info->lock);
3421 if (reserved == 0 || reclaimed >= max_reclaim)
3424 if (trans && trans->transaction->blocked)
3427 if (wait_ordered && !trans) {
3428 btrfs_wait_ordered_extents(root, 0, 0);
3430 time_left = schedule_timeout_interruptible(1);
3432 /* We were interrupted, exit */
3437 /* we've kicked the IO a few times, if anything has been freed,
3438 * exit. There is no sense in looping here for a long time
3439 * when we really need to commit the transaction, or there are
3440 * just too many writers without enough free space
3445 if (progress != space_info->reservation_progress)
3451 return reclaimed >= to_reclaim;
3455 * maybe_commit_transaction - possibly commit the transaction if its ok to
3456 * @root - the root we're allocating for
3457 * @bytes - the number of bytes we want to reserve
3458 * @force - force the commit
3460 * This will check to make sure that committing the transaction will actually
3461 * get us somewhere and then commit the transaction if it does. Otherwise it
3462 * will return -ENOSPC.
3464 static int may_commit_transaction(struct btrfs_root *root,
3465 struct btrfs_space_info *space_info,
3466 u64 bytes, int force)
3468 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3469 struct btrfs_trans_handle *trans;
3471 trans = (struct btrfs_trans_handle *)current->journal_info;
3478 /* See if there is enough pinned space to make this reservation */
3479 spin_lock(&space_info->lock);
3480 if (space_info->bytes_pinned >= bytes) {
3481 spin_unlock(&space_info->lock);
3484 spin_unlock(&space_info->lock);
3487 * See if there is some space in the delayed insertion reservation for
3490 if (space_info != delayed_rsv->space_info)
3493 spin_lock(&delayed_rsv->lock);
3494 if (delayed_rsv->size < bytes) {
3495 spin_unlock(&delayed_rsv->lock);
3498 spin_unlock(&delayed_rsv->lock);
3501 trans = btrfs_join_transaction(root);
3505 return btrfs_commit_transaction(trans, root);
3509 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3510 * @root - the root we're allocating for
3511 * @block_rsv - the block_rsv we're allocating for
3512 * @orig_bytes - the number of bytes we want
3513 * @flush - wether or not we can flush to make our reservation
3515 * This will reserve orgi_bytes number of bytes from the space info associated
3516 * with the block_rsv. If there is not enough space it will make an attempt to
3517 * flush out space to make room. It will do this by flushing delalloc if
3518 * possible or committing the transaction. If flush is 0 then no attempts to
3519 * regain reservations will be made and this will fail if there is not enough
3522 static int reserve_metadata_bytes(struct btrfs_root *root,
3523 struct btrfs_block_rsv *block_rsv,
3524 u64 orig_bytes, int flush)
3526 struct btrfs_space_info *space_info = block_rsv->space_info;
3528 u64 num_bytes = orig_bytes;
3531 bool committed = false;
3532 bool flushing = false;
3533 bool wait_ordered = false;
3537 spin_lock(&space_info->lock);
3539 * We only want to wait if somebody other than us is flushing and we are
3540 * actually alloed to flush.
3542 while (flush && !flushing && space_info->flush) {
3543 spin_unlock(&space_info->lock);
3545 * If we have a trans handle we can't wait because the flusher
3546 * may have to commit the transaction, which would mean we would
3547 * deadlock since we are waiting for the flusher to finish, but
3548 * hold the current transaction open.
3550 if (current->journal_info)
3552 ret = wait_event_interruptible(space_info->wait,
3553 !space_info->flush);
3554 /* Must have been interrupted, return */
3558 spin_lock(&space_info->lock);
3562 used = space_info->bytes_used + space_info->bytes_reserved +
3563 space_info->bytes_pinned + space_info->bytes_readonly +
3564 space_info->bytes_may_use;
3567 * The idea here is that we've not already over-reserved the block group
3568 * then we can go ahead and save our reservation first and then start
3569 * flushing if we need to. Otherwise if we've already overcommitted
3570 * lets start flushing stuff first and then come back and try to make
3573 if (used <= space_info->total_bytes) {
3574 if (used + orig_bytes <= space_info->total_bytes) {
3575 space_info->bytes_may_use += orig_bytes;
3579 * Ok set num_bytes to orig_bytes since we aren't
3580 * overocmmitted, this way we only try and reclaim what
3583 num_bytes = orig_bytes;
3587 * Ok we're over committed, set num_bytes to the overcommitted
3588 * amount plus the amount of bytes that we need for this
3591 wait_ordered = true;
3592 num_bytes = used - space_info->total_bytes +
3593 (orig_bytes * (retries + 1));
3597 u64 profile = btrfs_get_alloc_profile(root, 0);
3601 * If we have a lot of space that's pinned, don't bother doing
3602 * the overcommit dance yet and just commit the transaction.
3604 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3606 if (space_info->bytes_pinned >= avail && flush && !committed) {
3607 space_info->flush = 1;
3609 spin_unlock(&space_info->lock);
3610 ret = may_commit_transaction(root, space_info,
3618 spin_lock(&root->fs_info->free_chunk_lock);
3619 avail = root->fs_info->free_chunk_space;
3622 * If we have dup, raid1 or raid10 then only half of the free
3623 * space is actually useable.
3625 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3626 BTRFS_BLOCK_GROUP_RAID1 |
3627 BTRFS_BLOCK_GROUP_RAID10))
3631 * If we aren't flushing don't let us overcommit too much, say
3632 * 1/8th of the space. If we can flush, let it overcommit up to
3639 spin_unlock(&root->fs_info->free_chunk_lock);
3641 if (used + num_bytes < space_info->total_bytes + avail) {
3642 space_info->bytes_may_use += orig_bytes;
3645 wait_ordered = true;
3650 * Couldn't make our reservation, save our place so while we're trying
3651 * to reclaim space we can actually use it instead of somebody else
3652 * stealing it from us.
3656 space_info->flush = 1;
3659 spin_unlock(&space_info->lock);
3665 * We do synchronous shrinking since we don't actually unreserve
3666 * metadata until after the IO is completed.
3668 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3675 * So if we were overcommitted it's possible that somebody else flushed
3676 * out enough space and we simply didn't have enough space to reclaim,
3677 * so go back around and try again.
3680 wait_ordered = true;
3689 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3697 spin_lock(&space_info->lock);
3698 space_info->flush = 0;
3699 wake_up_all(&space_info->wait);
3700 spin_unlock(&space_info->lock);
3705 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3706 struct btrfs_root *root)
3708 struct btrfs_block_rsv *block_rsv = NULL;
3710 if (root->ref_cows || root == root->fs_info->csum_root)
3711 block_rsv = trans->block_rsv;
3714 block_rsv = root->block_rsv;
3717 block_rsv = &root->fs_info->empty_block_rsv;
3722 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3726 spin_lock(&block_rsv->lock);
3727 if (block_rsv->reserved >= num_bytes) {
3728 block_rsv->reserved -= num_bytes;
3729 if (block_rsv->reserved < block_rsv->size)
3730 block_rsv->full = 0;
3733 spin_unlock(&block_rsv->lock);
3737 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3738 u64 num_bytes, int update_size)
3740 spin_lock(&block_rsv->lock);
3741 block_rsv->reserved += num_bytes;
3743 block_rsv->size += num_bytes;
3744 else if (block_rsv->reserved >= block_rsv->size)
3745 block_rsv->full = 1;
3746 spin_unlock(&block_rsv->lock);
3749 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3750 struct btrfs_block_rsv *dest, u64 num_bytes)
3752 struct btrfs_space_info *space_info = block_rsv->space_info;
3754 spin_lock(&block_rsv->lock);
3755 if (num_bytes == (u64)-1)
3756 num_bytes = block_rsv->size;
3757 block_rsv->size -= num_bytes;
3758 if (block_rsv->reserved >= block_rsv->size) {
3759 num_bytes = block_rsv->reserved - block_rsv->size;
3760 block_rsv->reserved = block_rsv->size;
3761 block_rsv->full = 1;
3765 spin_unlock(&block_rsv->lock);
3767 if (num_bytes > 0) {
3769 spin_lock(&dest->lock);
3773 bytes_to_add = dest->size - dest->reserved;
3774 bytes_to_add = min(num_bytes, bytes_to_add);
3775 dest->reserved += bytes_to_add;
3776 if (dest->reserved >= dest->size)
3778 num_bytes -= bytes_to_add;
3780 spin_unlock(&dest->lock);
3783 spin_lock(&space_info->lock);
3784 space_info->bytes_may_use -= num_bytes;
3785 space_info->reservation_progress++;
3786 spin_unlock(&space_info->lock);
3791 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3792 struct btrfs_block_rsv *dst, u64 num_bytes)
3796 ret = block_rsv_use_bytes(src, num_bytes);
3800 block_rsv_add_bytes(dst, num_bytes, 1);
3804 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3806 memset(rsv, 0, sizeof(*rsv));
3807 spin_lock_init(&rsv->lock);
3810 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3812 struct btrfs_block_rsv *block_rsv;
3813 struct btrfs_fs_info *fs_info = root->fs_info;
3815 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3819 btrfs_init_block_rsv(block_rsv);
3820 block_rsv->space_info = __find_space_info(fs_info,
3821 BTRFS_BLOCK_GROUP_METADATA);
3825 void btrfs_free_block_rsv(struct btrfs_root *root,
3826 struct btrfs_block_rsv *rsv)
3828 btrfs_block_rsv_release(root, rsv, (u64)-1);
3832 static inline int __block_rsv_add(struct btrfs_root *root,
3833 struct btrfs_block_rsv *block_rsv,
3834 u64 num_bytes, int flush)
3841 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3843 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3850 int btrfs_block_rsv_add(struct btrfs_root *root,
3851 struct btrfs_block_rsv *block_rsv,
3854 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3857 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3858 struct btrfs_block_rsv *block_rsv,
3861 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3864 int btrfs_block_rsv_check(struct btrfs_root *root,
3865 struct btrfs_block_rsv *block_rsv, int min_factor)
3873 spin_lock(&block_rsv->lock);
3874 num_bytes = div_factor(block_rsv->size, min_factor);
3875 if (block_rsv->reserved >= num_bytes)
3877 spin_unlock(&block_rsv->lock);
3882 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3883 struct btrfs_block_rsv *block_rsv,
3884 u64 min_reserved, int flush)
3892 spin_lock(&block_rsv->lock);
3893 num_bytes = min_reserved;
3894 if (block_rsv->reserved >= num_bytes)
3897 num_bytes -= block_rsv->reserved;
3898 spin_unlock(&block_rsv->lock);
3903 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3905 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3912 int btrfs_block_rsv_refill(struct btrfs_root *root,
3913 struct btrfs_block_rsv *block_rsv,
3916 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3919 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3920 struct btrfs_block_rsv *block_rsv,
3923 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3926 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3927 struct btrfs_block_rsv *dst_rsv,
3930 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3933 void btrfs_block_rsv_release(struct btrfs_root *root,
3934 struct btrfs_block_rsv *block_rsv,
3937 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3938 if (global_rsv->full || global_rsv == block_rsv ||
3939 block_rsv->space_info != global_rsv->space_info)
3941 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3945 * helper to calculate size of global block reservation.
3946 * the desired value is sum of space used by extent tree,
3947 * checksum tree and root tree
3949 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3951 struct btrfs_space_info *sinfo;
3955 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3957 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3958 spin_lock(&sinfo->lock);
3959 data_used = sinfo->bytes_used;
3960 spin_unlock(&sinfo->lock);
3962 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3963 spin_lock(&sinfo->lock);
3964 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3966 meta_used = sinfo->bytes_used;
3967 spin_unlock(&sinfo->lock);
3969 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3971 num_bytes += div64_u64(data_used + meta_used, 50);
3973 if (num_bytes * 3 > meta_used)
3974 num_bytes = div64_u64(meta_used, 3);
3976 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3979 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3981 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3982 struct btrfs_space_info *sinfo = block_rsv->space_info;
3985 num_bytes = calc_global_metadata_size(fs_info);
3987 spin_lock(&block_rsv->lock);
3988 spin_lock(&sinfo->lock);
3990 block_rsv->size = num_bytes;
3992 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3993 sinfo->bytes_reserved + sinfo->bytes_readonly +
3994 sinfo->bytes_may_use;
3996 if (sinfo->total_bytes > num_bytes) {
3997 num_bytes = sinfo->total_bytes - num_bytes;
3998 block_rsv->reserved += num_bytes;
3999 sinfo->bytes_may_use += num_bytes;
4002 if (block_rsv->reserved >= block_rsv->size) {
4003 num_bytes = block_rsv->reserved - block_rsv->size;
4004 sinfo->bytes_may_use -= num_bytes;
4005 sinfo->reservation_progress++;
4006 block_rsv->reserved = block_rsv->size;
4007 block_rsv->full = 1;
4010 spin_unlock(&sinfo->lock);
4011 spin_unlock(&block_rsv->lock);
4014 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4016 struct btrfs_space_info *space_info;
4018 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4019 fs_info->chunk_block_rsv.space_info = space_info;
4021 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4022 fs_info->global_block_rsv.space_info = space_info;
4023 fs_info->delalloc_block_rsv.space_info = space_info;
4024 fs_info->trans_block_rsv.space_info = space_info;
4025 fs_info->empty_block_rsv.space_info = space_info;
4026 fs_info->delayed_block_rsv.space_info = space_info;
4028 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4029 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4030 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4031 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4032 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4034 update_global_block_rsv(fs_info);
4037 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4039 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4040 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4041 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4042 WARN_ON(fs_info->trans_block_rsv.size > 0);
4043 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4044 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4045 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4046 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4047 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4050 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4051 struct btrfs_root *root)
4053 if (!trans->bytes_reserved)
4056 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4057 trans->bytes_reserved = 0;
4060 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4061 struct inode *inode)
4063 struct btrfs_root *root = BTRFS_I(inode)->root;
4064 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4065 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4068 * We need to hold space in order to delete our orphan item once we've
4069 * added it, so this takes the reservation so we can release it later
4070 * when we are truly done with the orphan item.
4072 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4073 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4076 void btrfs_orphan_release_metadata(struct inode *inode)
4078 struct btrfs_root *root = BTRFS_I(inode)->root;
4079 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4080 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4083 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4084 struct btrfs_pending_snapshot *pending)
4086 struct btrfs_root *root = pending->root;
4087 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4088 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4090 * two for root back/forward refs, two for directory entries
4091 * and one for root of the snapshot.
4093 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4094 dst_rsv->space_info = src_rsv->space_info;
4095 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4099 * drop_outstanding_extent - drop an outstanding extent
4100 * @inode: the inode we're dropping the extent for
4102 * This is called when we are freeing up an outstanding extent, either called
4103 * after an error or after an extent is written. This will return the number of
4104 * reserved extents that need to be freed. This must be called with
4105 * BTRFS_I(inode)->lock held.
4107 static unsigned drop_outstanding_extent(struct inode *inode)
4109 unsigned drop_inode_space = 0;
4110 unsigned dropped_extents = 0;
4112 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4113 BTRFS_I(inode)->outstanding_extents--;
4115 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4116 BTRFS_I(inode)->delalloc_meta_reserved) {
4117 drop_inode_space = 1;
4118 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4122 * If we have more or the same amount of outsanding extents than we have
4123 * reserved then we need to leave the reserved extents count alone.
4125 if (BTRFS_I(inode)->outstanding_extents >=
4126 BTRFS_I(inode)->reserved_extents)
4127 return drop_inode_space;
4129 dropped_extents = BTRFS_I(inode)->reserved_extents -
4130 BTRFS_I(inode)->outstanding_extents;
4131 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4132 return dropped_extents + drop_inode_space;
4136 * calc_csum_metadata_size - return the amount of metada space that must be
4137 * reserved/free'd for the given bytes.
4138 * @inode: the inode we're manipulating
4139 * @num_bytes: the number of bytes in question
4140 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4142 * This adjusts the number of csum_bytes in the inode and then returns the
4143 * correct amount of metadata that must either be reserved or freed. We
4144 * calculate how many checksums we can fit into one leaf and then divide the
4145 * number of bytes that will need to be checksumed by this value to figure out
4146 * how many checksums will be required. If we are adding bytes then the number
4147 * may go up and we will return the number of additional bytes that must be
4148 * reserved. If it is going down we will return the number of bytes that must
4151 * This must be called with BTRFS_I(inode)->lock held.
4153 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4156 struct btrfs_root *root = BTRFS_I(inode)->root;
4158 int num_csums_per_leaf;
4162 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4163 BTRFS_I(inode)->csum_bytes == 0)
4166 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4168 BTRFS_I(inode)->csum_bytes += num_bytes;
4170 BTRFS_I(inode)->csum_bytes -= num_bytes;
4171 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4172 num_csums_per_leaf = (int)div64_u64(csum_size,
4173 sizeof(struct btrfs_csum_item) +
4174 sizeof(struct btrfs_disk_key));
4175 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4176 num_csums = num_csums + num_csums_per_leaf - 1;
4177 num_csums = num_csums / num_csums_per_leaf;
4179 old_csums = old_csums + num_csums_per_leaf - 1;
4180 old_csums = old_csums / num_csums_per_leaf;
4182 /* No change, no need to reserve more */
4183 if (old_csums == num_csums)
4187 return btrfs_calc_trans_metadata_size(root,
4188 num_csums - old_csums);
4190 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4193 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4195 struct btrfs_root *root = BTRFS_I(inode)->root;
4196 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4199 unsigned nr_extents = 0;
4200 int extra_reserve = 0;
4204 /* Need to be holding the i_mutex here if we aren't free space cache */
4205 if (btrfs_is_free_space_inode(root, inode))
4208 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4210 if (flush && btrfs_transaction_in_commit(root->fs_info))
4211 schedule_timeout(1);
4213 num_bytes = ALIGN(num_bytes, root->sectorsize);
4215 spin_lock(&BTRFS_I(inode)->lock);
4216 BTRFS_I(inode)->outstanding_extents++;
4218 if (BTRFS_I(inode)->outstanding_extents >
4219 BTRFS_I(inode)->reserved_extents)
4220 nr_extents = BTRFS_I(inode)->outstanding_extents -
4221 BTRFS_I(inode)->reserved_extents;
4224 * Add an item to reserve for updating the inode when we complete the
4227 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4232 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4233 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4234 csum_bytes = BTRFS_I(inode)->csum_bytes;
4235 spin_unlock(&BTRFS_I(inode)->lock);
4237 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4242 spin_lock(&BTRFS_I(inode)->lock);
4243 dropped = drop_outstanding_extent(inode);
4245 * If the inodes csum_bytes is the same as the original
4246 * csum_bytes then we know we haven't raced with any free()ers
4247 * so we can just reduce our inodes csum bytes and carry on.
4248 * Otherwise we have to do the normal free thing to account for
4249 * the case that the free side didn't free up its reserve
4250 * because of this outstanding reservation.
4252 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4253 calc_csum_metadata_size(inode, num_bytes, 0);
4255 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4256 spin_unlock(&BTRFS_I(inode)->lock);
4258 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4261 btrfs_block_rsv_release(root, block_rsv, to_free);
4265 spin_lock(&BTRFS_I(inode)->lock);
4266 if (extra_reserve) {
4267 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4270 BTRFS_I(inode)->reserved_extents += nr_extents;
4271 spin_unlock(&BTRFS_I(inode)->lock);
4273 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4279 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4280 * @inode: the inode to release the reservation for
4281 * @num_bytes: the number of bytes we're releasing
4283 * This will release the metadata reservation for an inode. This can be called
4284 * once we complete IO for a given set of bytes to release their metadata
4287 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4289 struct btrfs_root *root = BTRFS_I(inode)->root;
4293 num_bytes = ALIGN(num_bytes, root->sectorsize);
4294 spin_lock(&BTRFS_I(inode)->lock);
4295 dropped = drop_outstanding_extent(inode);
4297 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4298 spin_unlock(&BTRFS_I(inode)->lock);
4300 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4302 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4307 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4308 * @inode: inode we're writing to
4309 * @num_bytes: the number of bytes we want to allocate
4311 * This will do the following things
4313 * o reserve space in the data space info for num_bytes
4314 * o reserve space in the metadata space info based on number of outstanding
4315 * extents and how much csums will be needed
4316 * o add to the inodes ->delalloc_bytes
4317 * o add it to the fs_info's delalloc inodes list.
4319 * This will return 0 for success and -ENOSPC if there is no space left.
4321 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4325 ret = btrfs_check_data_free_space(inode, num_bytes);
4329 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4331 btrfs_free_reserved_data_space(inode, num_bytes);
4339 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4340 * @inode: inode we're releasing space for
4341 * @num_bytes: the number of bytes we want to free up
4343 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4344 * called in the case that we don't need the metadata AND data reservations
4345 * anymore. So if there is an error or we insert an inline extent.
4347 * This function will release the metadata space that was not used and will
4348 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4349 * list if there are no delalloc bytes left.
4351 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4353 btrfs_delalloc_release_metadata(inode, num_bytes);
4354 btrfs_free_reserved_data_space(inode, num_bytes);
4357 static int update_block_group(struct btrfs_trans_handle *trans,
4358 struct btrfs_root *root,
4359 u64 bytenr, u64 num_bytes, int alloc)
4361 struct btrfs_block_group_cache *cache = NULL;
4362 struct btrfs_fs_info *info = root->fs_info;
4363 u64 total = num_bytes;
4368 /* block accounting for super block */
4369 spin_lock(&info->delalloc_lock);
4370 old_val = btrfs_super_bytes_used(info->super_copy);
4372 old_val += num_bytes;
4374 old_val -= num_bytes;
4375 btrfs_set_super_bytes_used(info->super_copy, old_val);
4376 spin_unlock(&info->delalloc_lock);
4379 cache = btrfs_lookup_block_group(info, bytenr);
4382 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4383 BTRFS_BLOCK_GROUP_RAID1 |
4384 BTRFS_BLOCK_GROUP_RAID10))
4389 * If this block group has free space cache written out, we
4390 * need to make sure to load it if we are removing space. This
4391 * is because we need the unpinning stage to actually add the
4392 * space back to the block group, otherwise we will leak space.
4394 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4395 cache_block_group(cache, trans, NULL, 1);
4397 byte_in_group = bytenr - cache->key.objectid;
4398 WARN_ON(byte_in_group > cache->key.offset);
4400 spin_lock(&cache->space_info->lock);
4401 spin_lock(&cache->lock);
4403 if (btrfs_test_opt(root, SPACE_CACHE) &&
4404 cache->disk_cache_state < BTRFS_DC_CLEAR)
4405 cache->disk_cache_state = BTRFS_DC_CLEAR;
4408 old_val = btrfs_block_group_used(&cache->item);
4409 num_bytes = min(total, cache->key.offset - byte_in_group);
4411 old_val += num_bytes;
4412 btrfs_set_block_group_used(&cache->item, old_val);
4413 cache->reserved -= num_bytes;
4414 cache->space_info->bytes_reserved -= num_bytes;
4415 cache->space_info->bytes_used += num_bytes;
4416 cache->space_info->disk_used += num_bytes * factor;
4417 spin_unlock(&cache->lock);
4418 spin_unlock(&cache->space_info->lock);
4420 old_val -= num_bytes;
4421 btrfs_set_block_group_used(&cache->item, old_val);
4422 cache->pinned += num_bytes;
4423 cache->space_info->bytes_pinned += num_bytes;
4424 cache->space_info->bytes_used -= num_bytes;
4425 cache->space_info->disk_used -= num_bytes * factor;
4426 spin_unlock(&cache->lock);
4427 spin_unlock(&cache->space_info->lock);
4429 set_extent_dirty(info->pinned_extents,
4430 bytenr, bytenr + num_bytes - 1,
4431 GFP_NOFS | __GFP_NOFAIL);
4433 btrfs_put_block_group(cache);
4435 bytenr += num_bytes;
4440 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4442 struct btrfs_block_group_cache *cache;
4445 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4449 bytenr = cache->key.objectid;
4450 btrfs_put_block_group(cache);
4455 static int pin_down_extent(struct btrfs_root *root,
4456 struct btrfs_block_group_cache *cache,
4457 u64 bytenr, u64 num_bytes, int reserved)
4459 spin_lock(&cache->space_info->lock);
4460 spin_lock(&cache->lock);
4461 cache->pinned += num_bytes;
4462 cache->space_info->bytes_pinned += num_bytes;
4464 cache->reserved -= num_bytes;
4465 cache->space_info->bytes_reserved -= num_bytes;
4467 spin_unlock(&cache->lock);
4468 spin_unlock(&cache->space_info->lock);
4470 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4471 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4476 * this function must be called within transaction
4478 int btrfs_pin_extent(struct btrfs_root *root,
4479 u64 bytenr, u64 num_bytes, int reserved)
4481 struct btrfs_block_group_cache *cache;
4483 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4486 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4488 btrfs_put_block_group(cache);
4493 * this function must be called within transaction
4495 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
4497 u64 bytenr, u64 num_bytes)
4499 struct btrfs_block_group_cache *cache;
4501 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4505 * pull in the free space cache (if any) so that our pin
4506 * removes the free space from the cache. We have load_only set
4507 * to one because the slow code to read in the free extents does check
4508 * the pinned extents.
4510 cache_block_group(cache, trans, root, 1);
4512 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4514 /* remove us from the free space cache (if we're there at all) */
4515 btrfs_remove_free_space(cache, bytenr, num_bytes);
4516 btrfs_put_block_group(cache);
4521 * btrfs_update_reserved_bytes - update the block_group and space info counters
4522 * @cache: The cache we are manipulating
4523 * @num_bytes: The number of bytes in question
4524 * @reserve: One of the reservation enums
4526 * This is called by the allocator when it reserves space, or by somebody who is
4527 * freeing space that was never actually used on disk. For example if you
4528 * reserve some space for a new leaf in transaction A and before transaction A
4529 * commits you free that leaf, you call this with reserve set to 0 in order to
4530 * clear the reservation.
4532 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4533 * ENOSPC accounting. For data we handle the reservation through clearing the
4534 * delalloc bits in the io_tree. We have to do this since we could end up
4535 * allocating less disk space for the amount of data we have reserved in the
4536 * case of compression.
4538 * If this is a reservation and the block group has become read only we cannot
4539 * make the reservation and return -EAGAIN, otherwise this function always
4542 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4543 u64 num_bytes, int reserve)
4545 struct btrfs_space_info *space_info = cache->space_info;
4547 spin_lock(&space_info->lock);
4548 spin_lock(&cache->lock);
4549 if (reserve != RESERVE_FREE) {
4553 cache->reserved += num_bytes;
4554 space_info->bytes_reserved += num_bytes;
4555 if (reserve == RESERVE_ALLOC) {
4556 BUG_ON(space_info->bytes_may_use < num_bytes);
4557 space_info->bytes_may_use -= num_bytes;
4562 space_info->bytes_readonly += num_bytes;
4563 cache->reserved -= num_bytes;
4564 space_info->bytes_reserved -= num_bytes;
4565 space_info->reservation_progress++;
4567 spin_unlock(&cache->lock);
4568 spin_unlock(&space_info->lock);
4572 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4573 struct btrfs_root *root)
4575 struct btrfs_fs_info *fs_info = root->fs_info;
4576 struct btrfs_caching_control *next;
4577 struct btrfs_caching_control *caching_ctl;
4578 struct btrfs_block_group_cache *cache;
4580 down_write(&fs_info->extent_commit_sem);
4582 list_for_each_entry_safe(caching_ctl, next,
4583 &fs_info->caching_block_groups, list) {
4584 cache = caching_ctl->block_group;
4585 if (block_group_cache_done(cache)) {
4586 cache->last_byte_to_unpin = (u64)-1;
4587 list_del_init(&caching_ctl->list);
4588 put_caching_control(caching_ctl);
4590 cache->last_byte_to_unpin = caching_ctl->progress;
4594 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4595 fs_info->pinned_extents = &fs_info->freed_extents[1];
4597 fs_info->pinned_extents = &fs_info->freed_extents[0];
4599 up_write(&fs_info->extent_commit_sem);
4601 update_global_block_rsv(fs_info);
4605 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4607 struct btrfs_fs_info *fs_info = root->fs_info;
4608 struct btrfs_block_group_cache *cache = NULL;
4611 while (start <= end) {
4613 start >= cache->key.objectid + cache->key.offset) {
4615 btrfs_put_block_group(cache);
4616 cache = btrfs_lookup_block_group(fs_info, start);
4620 len = cache->key.objectid + cache->key.offset - start;
4621 len = min(len, end + 1 - start);
4623 if (start < cache->last_byte_to_unpin) {
4624 len = min(len, cache->last_byte_to_unpin - start);
4625 btrfs_add_free_space(cache, start, len);
4630 spin_lock(&cache->space_info->lock);
4631 spin_lock(&cache->lock);
4632 cache->pinned -= len;
4633 cache->space_info->bytes_pinned -= len;
4635 cache->space_info->bytes_readonly += len;
4636 spin_unlock(&cache->lock);
4637 spin_unlock(&cache->space_info->lock);
4641 btrfs_put_block_group(cache);
4645 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4646 struct btrfs_root *root)
4648 struct btrfs_fs_info *fs_info = root->fs_info;
4649 struct extent_io_tree *unpin;
4654 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4655 unpin = &fs_info->freed_extents[1];
4657 unpin = &fs_info->freed_extents[0];
4660 ret = find_first_extent_bit(unpin, 0, &start, &end,
4665 if (btrfs_test_opt(root, DISCARD))
4666 ret = btrfs_discard_extent(root, start,
4667 end + 1 - start, NULL);
4669 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4670 unpin_extent_range(root, start, end);
4677 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4678 struct btrfs_root *root,
4679 u64 bytenr, u64 num_bytes, u64 parent,
4680 u64 root_objectid, u64 owner_objectid,
4681 u64 owner_offset, int refs_to_drop,
4682 struct btrfs_delayed_extent_op *extent_op)
4684 struct btrfs_key key;
4685 struct btrfs_path *path;
4686 struct btrfs_fs_info *info = root->fs_info;
4687 struct btrfs_root *extent_root = info->extent_root;
4688 struct extent_buffer *leaf;
4689 struct btrfs_extent_item *ei;
4690 struct btrfs_extent_inline_ref *iref;
4693 int extent_slot = 0;
4694 int found_extent = 0;
4699 path = btrfs_alloc_path();
4704 path->leave_spinning = 1;
4706 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4707 BUG_ON(!is_data && refs_to_drop != 1);
4709 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4710 bytenr, num_bytes, parent,
4711 root_objectid, owner_objectid,
4714 extent_slot = path->slots[0];
4715 while (extent_slot >= 0) {
4716 btrfs_item_key_to_cpu(path->nodes[0], &key,
4718 if (key.objectid != bytenr)
4720 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4721 key.offset == num_bytes) {
4725 if (path->slots[0] - extent_slot > 5)
4729 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4730 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4731 if (found_extent && item_size < sizeof(*ei))
4734 if (!found_extent) {
4736 ret = remove_extent_backref(trans, extent_root, path,
4740 btrfs_release_path(path);
4741 path->leave_spinning = 1;
4743 key.objectid = bytenr;
4744 key.type = BTRFS_EXTENT_ITEM_KEY;
4745 key.offset = num_bytes;
4747 ret = btrfs_search_slot(trans, extent_root,
4750 printk(KERN_ERR "umm, got %d back from search"
4751 ", was looking for %llu\n", ret,
4752 (unsigned long long)bytenr);
4754 btrfs_print_leaf(extent_root,
4758 extent_slot = path->slots[0];
4761 btrfs_print_leaf(extent_root, path->nodes[0]);
4763 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4764 "parent %llu root %llu owner %llu offset %llu\n",
4765 (unsigned long long)bytenr,
4766 (unsigned long long)parent,
4767 (unsigned long long)root_objectid,
4768 (unsigned long long)owner_objectid,
4769 (unsigned long long)owner_offset);
4772 leaf = path->nodes[0];
4773 item_size = btrfs_item_size_nr(leaf, extent_slot);
4774 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4775 if (item_size < sizeof(*ei)) {
4776 BUG_ON(found_extent || extent_slot != path->slots[0]);
4777 ret = convert_extent_item_v0(trans, extent_root, path,
4781 btrfs_release_path(path);
4782 path->leave_spinning = 1;
4784 key.objectid = bytenr;
4785 key.type = BTRFS_EXTENT_ITEM_KEY;
4786 key.offset = num_bytes;
4788 ret = btrfs_search_slot(trans, extent_root, &key, path,
4791 printk(KERN_ERR "umm, got %d back from search"
4792 ", was looking for %llu\n", ret,
4793 (unsigned long long)bytenr);
4794 btrfs_print_leaf(extent_root, path->nodes[0]);
4797 extent_slot = path->slots[0];
4798 leaf = path->nodes[0];
4799 item_size = btrfs_item_size_nr(leaf, extent_slot);
4802 BUG_ON(item_size < sizeof(*ei));
4803 ei = btrfs_item_ptr(leaf, extent_slot,
4804 struct btrfs_extent_item);
4805 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4806 struct btrfs_tree_block_info *bi;
4807 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4808 bi = (struct btrfs_tree_block_info *)(ei + 1);
4809 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4812 refs = btrfs_extent_refs(leaf, ei);
4813 BUG_ON(refs < refs_to_drop);
4814 refs -= refs_to_drop;
4818 __run_delayed_extent_op(extent_op, leaf, ei);
4820 * In the case of inline back ref, reference count will
4821 * be updated by remove_extent_backref
4824 BUG_ON(!found_extent);
4826 btrfs_set_extent_refs(leaf, ei, refs);
4827 btrfs_mark_buffer_dirty(leaf);
4830 ret = remove_extent_backref(trans, extent_root, path,
4837 BUG_ON(is_data && refs_to_drop !=
4838 extent_data_ref_count(root, path, iref));
4840 BUG_ON(path->slots[0] != extent_slot);
4842 BUG_ON(path->slots[0] != extent_slot + 1);
4843 path->slots[0] = extent_slot;
4848 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4851 btrfs_release_path(path);
4854 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4857 invalidate_mapping_pages(info->btree_inode->i_mapping,
4858 bytenr >> PAGE_CACHE_SHIFT,
4859 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4862 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4865 btrfs_free_path(path);
4870 * when we free an block, it is possible (and likely) that we free the last
4871 * delayed ref for that extent as well. This searches the delayed ref tree for
4872 * a given extent, and if there are no other delayed refs to be processed, it
4873 * removes it from the tree.
4875 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4876 struct btrfs_root *root, u64 bytenr)
4878 struct btrfs_delayed_ref_head *head;
4879 struct btrfs_delayed_ref_root *delayed_refs;
4880 struct btrfs_delayed_ref_node *ref;
4881 struct rb_node *node;
4884 delayed_refs = &trans->transaction->delayed_refs;
4885 spin_lock(&delayed_refs->lock);
4886 head = btrfs_find_delayed_ref_head(trans, bytenr);
4890 node = rb_prev(&head->node.rb_node);
4894 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4896 /* there are still entries for this ref, we can't drop it */
4897 if (ref->bytenr == bytenr)
4900 if (head->extent_op) {
4901 if (!head->must_insert_reserved)
4903 kfree(head->extent_op);
4904 head->extent_op = NULL;
4908 * waiting for the lock here would deadlock. If someone else has it
4909 * locked they are already in the process of dropping it anyway
4911 if (!mutex_trylock(&head->mutex))
4915 * at this point we have a head with no other entries. Go
4916 * ahead and process it.
4918 head->node.in_tree = 0;
4919 rb_erase(&head->node.rb_node, &delayed_refs->root);
4921 delayed_refs->num_entries--;
4924 * we don't take a ref on the node because we're removing it from the
4925 * tree, so we just steal the ref the tree was holding.
4927 delayed_refs->num_heads--;
4928 if (list_empty(&head->cluster))
4929 delayed_refs->num_heads_ready--;
4931 list_del_init(&head->cluster);
4932 spin_unlock(&delayed_refs->lock);
4934 BUG_ON(head->extent_op);
4935 if (head->must_insert_reserved)
4938 mutex_unlock(&head->mutex);
4939 btrfs_put_delayed_ref(&head->node);
4942 spin_unlock(&delayed_refs->lock);
4946 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4947 struct btrfs_root *root,
4948 struct extent_buffer *buf,
4949 u64 parent, int last_ref)
4951 struct btrfs_block_group_cache *cache = NULL;
4954 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4955 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4956 parent, root->root_key.objectid,
4957 btrfs_header_level(buf),
4958 BTRFS_DROP_DELAYED_REF, NULL);
4965 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4967 if (btrfs_header_generation(buf) == trans->transid) {
4968 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4969 ret = check_ref_cleanup(trans, root, buf->start);
4974 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4975 pin_down_extent(root, cache, buf->start, buf->len, 1);
4979 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4981 btrfs_add_free_space(cache, buf->start, buf->len);
4982 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4986 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4989 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4990 btrfs_put_block_group(cache);
4993 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4994 struct btrfs_root *root,
4995 u64 bytenr, u64 num_bytes, u64 parent,
4996 u64 root_objectid, u64 owner, u64 offset)
5001 * tree log blocks never actually go into the extent allocation
5002 * tree, just update pinning info and exit early.
5004 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5005 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5006 /* unlocks the pinned mutex */
5007 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5009 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5010 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
5011 parent, root_objectid, (int)owner,
5012 BTRFS_DROP_DELAYED_REF, NULL);
5015 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
5016 parent, root_objectid, owner,
5017 offset, BTRFS_DROP_DELAYED_REF, NULL);
5023 static u64 stripe_align(struct btrfs_root *root, u64 val)
5025 u64 mask = ((u64)root->stripesize - 1);
5026 u64 ret = (val + mask) & ~mask;
5031 * when we wait for progress in the block group caching, its because
5032 * our allocation attempt failed at least once. So, we must sleep
5033 * and let some progress happen before we try again.
5035 * This function will sleep at least once waiting for new free space to
5036 * show up, and then it will check the block group free space numbers
5037 * for our min num_bytes. Another option is to have it go ahead
5038 * and look in the rbtree for a free extent of a given size, but this
5042 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5045 struct btrfs_caching_control *caching_ctl;
5048 caching_ctl = get_caching_control(cache);
5052 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5053 (cache->free_space_ctl->free_space >= num_bytes));
5055 put_caching_control(caching_ctl);
5060 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5062 struct btrfs_caching_control *caching_ctl;
5065 caching_ctl = get_caching_control(cache);
5069 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5071 put_caching_control(caching_ctl);
5075 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5078 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5080 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5082 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5084 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5091 enum btrfs_loop_type {
5092 LOOP_FIND_IDEAL = 0,
5093 LOOP_CACHING_NOWAIT = 1,
5094 LOOP_CACHING_WAIT = 2,
5095 LOOP_ALLOC_CHUNK = 3,
5096 LOOP_NO_EMPTY_SIZE = 4,
5100 * walks the btree of allocated extents and find a hole of a given size.
5101 * The key ins is changed to record the hole:
5102 * ins->objectid == block start
5103 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5104 * ins->offset == number of blocks
5105 * Any available blocks before search_start are skipped.
5107 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5108 struct btrfs_root *orig_root,
5109 u64 num_bytes, u64 empty_size,
5110 u64 search_start, u64 search_end,
5111 u64 hint_byte, struct btrfs_key *ins,
5115 struct btrfs_root *root = orig_root->fs_info->extent_root;
5116 struct btrfs_free_cluster *last_ptr = NULL;
5117 struct btrfs_block_group_cache *block_group = NULL;
5118 struct btrfs_block_group_cache *used_block_group;
5119 int empty_cluster = 2 * 1024 * 1024;
5120 int allowed_chunk_alloc = 0;
5121 int done_chunk_alloc = 0;
5122 struct btrfs_space_info *space_info;
5125 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5126 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5127 bool found_uncached_bg = false;
5128 bool failed_cluster_refill = false;
5129 bool failed_alloc = false;
5130 bool use_cluster = true;
5131 bool have_caching_bg = false;
5132 u64 ideal_cache_percent = 0;
5133 u64 ideal_cache_offset = 0;
5135 WARN_ON(num_bytes < root->sectorsize);
5136 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5140 space_info = __find_space_info(root->fs_info, data);
5142 printk(KERN_ERR "No space info for %llu\n", data);
5147 * If the space info is for both data and metadata it means we have a
5148 * small filesystem and we can't use the clustering stuff.
5150 if (btrfs_mixed_space_info(space_info))
5151 use_cluster = false;
5153 if (orig_root->ref_cows || empty_size)
5154 allowed_chunk_alloc = 1;
5156 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5157 last_ptr = &root->fs_info->meta_alloc_cluster;
5158 if (!btrfs_test_opt(root, SSD))
5159 empty_cluster = 64 * 1024;
5162 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5163 btrfs_test_opt(root, SSD)) {
5164 last_ptr = &root->fs_info->data_alloc_cluster;
5168 spin_lock(&last_ptr->lock);
5169 if (last_ptr->block_group)
5170 hint_byte = last_ptr->window_start;
5171 spin_unlock(&last_ptr->lock);
5174 search_start = max(search_start, first_logical_byte(root, 0));
5175 search_start = max(search_start, hint_byte);
5180 if (search_start == hint_byte) {
5182 block_group = btrfs_lookup_block_group(root->fs_info,
5184 used_block_group = block_group;
5186 * we don't want to use the block group if it doesn't match our
5187 * allocation bits, or if its not cached.
5189 * However if we are re-searching with an ideal block group
5190 * picked out then we don't care that the block group is cached.
5192 if (block_group && block_group_bits(block_group, data) &&
5193 (block_group->cached != BTRFS_CACHE_NO ||
5194 search_start == ideal_cache_offset)) {
5195 down_read(&space_info->groups_sem);
5196 if (list_empty(&block_group->list) ||
5199 * someone is removing this block group,
5200 * we can't jump into the have_block_group
5201 * target because our list pointers are not
5204 btrfs_put_block_group(block_group);
5205 up_read(&space_info->groups_sem);
5207 index = get_block_group_index(block_group);
5208 goto have_block_group;
5210 } else if (block_group) {
5211 btrfs_put_block_group(block_group);
5215 have_caching_bg = false;
5216 down_read(&space_info->groups_sem);
5217 list_for_each_entry(block_group, &space_info->block_groups[index],
5222 used_block_group = block_group;
5223 btrfs_get_block_group(block_group);
5224 search_start = block_group->key.objectid;
5227 * this can happen if we end up cycling through all the
5228 * raid types, but we want to make sure we only allocate
5229 * for the proper type.
5231 if (!block_group_bits(block_group, data)) {
5232 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5233 BTRFS_BLOCK_GROUP_RAID1 |
5234 BTRFS_BLOCK_GROUP_RAID10;
5237 * if they asked for extra copies and this block group
5238 * doesn't provide them, bail. This does allow us to
5239 * fill raid0 from raid1.
5241 if ((data & extra) && !(block_group->flags & extra))
5246 cached = block_group_cache_done(block_group);
5247 if (unlikely(!cached)) {
5250 found_uncached_bg = true;
5251 ret = cache_block_group(block_group, trans,
5253 if (block_group->cached == BTRFS_CACHE_FINISHED)
5256 free_percent = btrfs_block_group_used(&block_group->item);
5257 free_percent *= 100;
5258 free_percent = div64_u64(free_percent,
5259 block_group->key.offset);
5260 free_percent = 100 - free_percent;
5261 if (free_percent > ideal_cache_percent &&
5262 likely(!block_group->ro)) {
5263 ideal_cache_offset = block_group->key.objectid;
5264 ideal_cache_percent = free_percent;
5268 * The caching workers are limited to 2 threads, so we
5269 * can queue as much work as we care to.
5271 if (loop > LOOP_FIND_IDEAL) {
5272 ret = cache_block_group(block_group, trans,
5278 * If loop is set for cached only, try the next block
5281 if (loop == LOOP_FIND_IDEAL)
5286 if (unlikely(block_group->ro))
5290 * Ok we want to try and use the cluster allocator, so
5295 * the refill lock keeps out other
5296 * people trying to start a new cluster
5298 spin_lock(&last_ptr->refill_lock);
5299 used_block_group = last_ptr->block_group;
5300 if (used_block_group != block_group &&
5301 (!used_block_group ||
5302 used_block_group->ro ||
5303 !block_group_bits(used_block_group, data))) {
5304 used_block_group = block_group;
5305 goto refill_cluster;
5308 if (used_block_group != block_group)
5309 btrfs_get_block_group(used_block_group);
5311 offset = btrfs_alloc_from_cluster(used_block_group,
5312 last_ptr, num_bytes, used_block_group->key.objectid);
5314 /* we have a block, we're done */
5315 spin_unlock(&last_ptr->refill_lock);
5319 WARN_ON(last_ptr->block_group != used_block_group);
5320 if (used_block_group != block_group) {
5321 btrfs_put_block_group(used_block_group);
5322 used_block_group = block_group;
5325 BUG_ON(used_block_group != block_group);
5326 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5327 * set up a new clusters, so lets just skip it
5328 * and let the allocator find whatever block
5329 * it can find. If we reach this point, we
5330 * will have tried the cluster allocator
5331 * plenty of times and not have found
5332 * anything, so we are likely way too
5333 * fragmented for the clustering stuff to find
5336 * However, if the cluster is taken from the
5337 * current block group, release the cluster
5338 * first, so that we stand a better chance of
5339 * succeeding in the unclustered
5341 if (loop >= LOOP_NO_EMPTY_SIZE &&
5342 last_ptr->block_group != block_group) {
5343 spin_unlock(&last_ptr->refill_lock);
5344 goto unclustered_alloc;
5348 * this cluster didn't work out, free it and
5351 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5353 if (loop >= LOOP_NO_EMPTY_SIZE) {
5354 spin_unlock(&last_ptr->refill_lock);
5355 goto unclustered_alloc;
5358 /* allocate a cluster in this block group */
5359 ret = btrfs_find_space_cluster(trans, root,
5360 block_group, last_ptr,
5361 search_start, num_bytes,
5362 empty_cluster + empty_size);
5365 * now pull our allocation out of this
5368 offset = btrfs_alloc_from_cluster(block_group,
5369 last_ptr, num_bytes,
5372 /* we found one, proceed */
5373 spin_unlock(&last_ptr->refill_lock);
5376 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5377 && !failed_cluster_refill) {
5378 spin_unlock(&last_ptr->refill_lock);
5380 failed_cluster_refill = true;
5381 wait_block_group_cache_progress(block_group,
5382 num_bytes + empty_cluster + empty_size);
5383 goto have_block_group;
5387 * at this point we either didn't find a cluster
5388 * or we weren't able to allocate a block from our
5389 * cluster. Free the cluster we've been trying
5390 * to use, and go to the next block group
5392 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5393 spin_unlock(&last_ptr->refill_lock);
5398 spin_lock(&block_group->free_space_ctl->tree_lock);
5400 block_group->free_space_ctl->free_space <
5401 num_bytes + empty_cluster + empty_size) {
5402 spin_unlock(&block_group->free_space_ctl->tree_lock);
5405 spin_unlock(&block_group->free_space_ctl->tree_lock);
5407 offset = btrfs_find_space_for_alloc(block_group, search_start,
5408 num_bytes, empty_size);
5410 * If we didn't find a chunk, and we haven't failed on this
5411 * block group before, and this block group is in the middle of
5412 * caching and we are ok with waiting, then go ahead and wait
5413 * for progress to be made, and set failed_alloc to true.
5415 * If failed_alloc is true then we've already waited on this
5416 * block group once and should move on to the next block group.
5418 if (!offset && !failed_alloc && !cached &&
5419 loop > LOOP_CACHING_NOWAIT) {
5420 wait_block_group_cache_progress(block_group,
5421 num_bytes + empty_size);
5422 failed_alloc = true;
5423 goto have_block_group;
5424 } else if (!offset) {
5426 have_caching_bg = true;
5430 search_start = stripe_align(root, offset);
5431 /* move on to the next group */
5432 if (search_start + num_bytes >= search_end) {
5433 btrfs_add_free_space(used_block_group, offset, num_bytes);
5437 /* move on to the next group */
5438 if (search_start + num_bytes >
5439 used_block_group->key.objectid + used_block_group->key.offset) {
5440 btrfs_add_free_space(used_block_group, offset, num_bytes);
5444 if (offset < search_start)
5445 btrfs_add_free_space(used_block_group, offset,
5446 search_start - offset);
5447 BUG_ON(offset > search_start);
5449 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5451 if (ret == -EAGAIN) {
5452 btrfs_add_free_space(used_block_group, offset, num_bytes);
5456 /* we are all good, lets return */
5457 ins->objectid = search_start;
5458 ins->offset = num_bytes;
5460 if (offset < search_start)
5461 btrfs_add_free_space(used_block_group, offset,
5462 search_start - offset);
5463 BUG_ON(offset > search_start);
5464 if (used_block_group != block_group)
5465 btrfs_put_block_group(used_block_group);
5466 btrfs_put_block_group(block_group);
5469 failed_cluster_refill = false;
5470 failed_alloc = false;
5471 BUG_ON(index != get_block_group_index(block_group));
5472 if (used_block_group != block_group)
5473 btrfs_put_block_group(used_block_group);
5474 btrfs_put_block_group(block_group);
5476 up_read(&space_info->groups_sem);
5478 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5481 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5484 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5485 * for them to make caching progress. Also
5486 * determine the best possible bg to cache
5487 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5488 * caching kthreads as we move along
5489 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5490 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5491 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5494 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5496 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5497 found_uncached_bg = false;
5499 if (!ideal_cache_percent)
5503 * 1 of the following 2 things have happened so far
5505 * 1) We found an ideal block group for caching that
5506 * is mostly full and will cache quickly, so we might
5507 * as well wait for it.
5509 * 2) We searched for cached only and we didn't find
5510 * anything, and we didn't start any caching kthreads
5511 * either, so chances are we will loop through and
5512 * start a couple caching kthreads, and then come back
5513 * around and just wait for them. This will be slower
5514 * because we will have 2 caching kthreads reading at
5515 * the same time when we could have just started one
5516 * and waited for it to get far enough to give us an
5517 * allocation, so go ahead and go to the wait caching
5520 loop = LOOP_CACHING_WAIT;
5521 search_start = ideal_cache_offset;
5522 ideal_cache_percent = 0;
5524 } else if (loop == LOOP_FIND_IDEAL) {
5526 * Didn't find a uncached bg, wait on anything we find
5529 loop = LOOP_CACHING_WAIT;
5535 if (loop == LOOP_ALLOC_CHUNK) {
5536 if (allowed_chunk_alloc) {
5537 ret = do_chunk_alloc(trans, root, num_bytes +
5538 2 * 1024 * 1024, data,
5539 CHUNK_ALLOC_LIMITED);
5540 allowed_chunk_alloc = 0;
5542 done_chunk_alloc = 1;
5543 } else if (!done_chunk_alloc &&
5544 space_info->force_alloc ==
5545 CHUNK_ALLOC_NO_FORCE) {
5546 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5550 * We didn't allocate a chunk, go ahead and drop the
5551 * empty size and loop again.
5553 if (!done_chunk_alloc)
5554 loop = LOOP_NO_EMPTY_SIZE;
5557 if (loop == LOOP_NO_EMPTY_SIZE) {
5563 } else if (!ins->objectid) {
5565 } else if (ins->objectid) {
5572 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5573 int dump_block_groups)
5575 struct btrfs_block_group_cache *cache;
5578 spin_lock(&info->lock);
5579 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5580 (unsigned long long)info->flags,
5581 (unsigned long long)(info->total_bytes - info->bytes_used -
5582 info->bytes_pinned - info->bytes_reserved -
5583 info->bytes_readonly),
5584 (info->full) ? "" : "not ");
5585 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5586 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5587 (unsigned long long)info->total_bytes,
5588 (unsigned long long)info->bytes_used,
5589 (unsigned long long)info->bytes_pinned,
5590 (unsigned long long)info->bytes_reserved,
5591 (unsigned long long)info->bytes_may_use,
5592 (unsigned long long)info->bytes_readonly);
5593 spin_unlock(&info->lock);
5595 if (!dump_block_groups)
5598 down_read(&info->groups_sem);
5600 list_for_each_entry(cache, &info->block_groups[index], list) {
5601 spin_lock(&cache->lock);
5602 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5603 "%llu pinned %llu reserved\n",
5604 (unsigned long long)cache->key.objectid,
5605 (unsigned long long)cache->key.offset,
5606 (unsigned long long)btrfs_block_group_used(&cache->item),
5607 (unsigned long long)cache->pinned,
5608 (unsigned long long)cache->reserved);
5609 btrfs_dump_free_space(cache, bytes);
5610 spin_unlock(&cache->lock);
5612 if (++index < BTRFS_NR_RAID_TYPES)
5614 up_read(&info->groups_sem);
5617 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5618 struct btrfs_root *root,
5619 u64 num_bytes, u64 min_alloc_size,
5620 u64 empty_size, u64 hint_byte,
5621 u64 search_end, struct btrfs_key *ins,
5625 u64 search_start = 0;
5627 data = btrfs_get_alloc_profile(root, data);
5630 * the only place that sets empty_size is btrfs_realloc_node, which
5631 * is not called recursively on allocations
5633 if (empty_size || root->ref_cows)
5634 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5635 num_bytes + 2 * 1024 * 1024, data,
5636 CHUNK_ALLOC_NO_FORCE);
5638 WARN_ON(num_bytes < root->sectorsize);
5639 ret = find_free_extent(trans, root, num_bytes, empty_size,
5640 search_start, search_end, hint_byte,
5643 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5644 num_bytes = num_bytes >> 1;
5645 num_bytes = num_bytes & ~(root->sectorsize - 1);
5646 num_bytes = max(num_bytes, min_alloc_size);
5647 do_chunk_alloc(trans, root->fs_info->extent_root,
5648 num_bytes, data, CHUNK_ALLOC_FORCE);
5651 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5652 struct btrfs_space_info *sinfo;
5654 sinfo = __find_space_info(root->fs_info, data);
5655 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5656 "wanted %llu\n", (unsigned long long)data,
5657 (unsigned long long)num_bytes);
5658 dump_space_info(sinfo, num_bytes, 1);
5661 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5666 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5667 u64 start, u64 len, int pin)
5669 struct btrfs_block_group_cache *cache;
5672 cache = btrfs_lookup_block_group(root->fs_info, start);
5674 printk(KERN_ERR "Unable to find block group for %llu\n",
5675 (unsigned long long)start);
5679 if (btrfs_test_opt(root, DISCARD))
5680 ret = btrfs_discard_extent(root, start, len, NULL);
5683 pin_down_extent(root, cache, start, len, 1);
5685 btrfs_add_free_space(cache, start, len);
5686 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5688 btrfs_put_block_group(cache);
5690 trace_btrfs_reserved_extent_free(root, start, len);
5695 int btrfs_free_reserved_extent(struct btrfs_root *root,
5698 return __btrfs_free_reserved_extent(root, start, len, 0);
5701 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5704 return __btrfs_free_reserved_extent(root, start, len, 1);
5707 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5708 struct btrfs_root *root,
5709 u64 parent, u64 root_objectid,
5710 u64 flags, u64 owner, u64 offset,
5711 struct btrfs_key *ins, int ref_mod)
5714 struct btrfs_fs_info *fs_info = root->fs_info;
5715 struct btrfs_extent_item *extent_item;
5716 struct btrfs_extent_inline_ref *iref;
5717 struct btrfs_path *path;
5718 struct extent_buffer *leaf;
5723 type = BTRFS_SHARED_DATA_REF_KEY;
5725 type = BTRFS_EXTENT_DATA_REF_KEY;
5727 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5729 path = btrfs_alloc_path();
5733 path->leave_spinning = 1;
5734 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5738 leaf = path->nodes[0];
5739 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5740 struct btrfs_extent_item);
5741 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5742 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5743 btrfs_set_extent_flags(leaf, extent_item,
5744 flags | BTRFS_EXTENT_FLAG_DATA);
5746 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5747 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5749 struct btrfs_shared_data_ref *ref;
5750 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5751 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5752 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5754 struct btrfs_extent_data_ref *ref;
5755 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5756 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5757 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5758 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5759 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5762 btrfs_mark_buffer_dirty(path->nodes[0]);
5763 btrfs_free_path(path);
5765 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5767 printk(KERN_ERR "btrfs update block group failed for %llu "
5768 "%llu\n", (unsigned long long)ins->objectid,
5769 (unsigned long long)ins->offset);
5775 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5776 struct btrfs_root *root,
5777 u64 parent, u64 root_objectid,
5778 u64 flags, struct btrfs_disk_key *key,
5779 int level, struct btrfs_key *ins)
5782 struct btrfs_fs_info *fs_info = root->fs_info;
5783 struct btrfs_extent_item *extent_item;
5784 struct btrfs_tree_block_info *block_info;
5785 struct btrfs_extent_inline_ref *iref;
5786 struct btrfs_path *path;
5787 struct extent_buffer *leaf;
5788 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5790 path = btrfs_alloc_path();
5794 path->leave_spinning = 1;
5795 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5799 leaf = path->nodes[0];
5800 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5801 struct btrfs_extent_item);
5802 btrfs_set_extent_refs(leaf, extent_item, 1);
5803 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5804 btrfs_set_extent_flags(leaf, extent_item,
5805 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5806 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5808 btrfs_set_tree_block_key(leaf, block_info, key);
5809 btrfs_set_tree_block_level(leaf, block_info, level);
5811 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5813 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5814 btrfs_set_extent_inline_ref_type(leaf, iref,
5815 BTRFS_SHARED_BLOCK_REF_KEY);
5816 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5818 btrfs_set_extent_inline_ref_type(leaf, iref,
5819 BTRFS_TREE_BLOCK_REF_KEY);
5820 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5823 btrfs_mark_buffer_dirty(leaf);
5824 btrfs_free_path(path);
5826 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5828 printk(KERN_ERR "btrfs update block group failed for %llu "
5829 "%llu\n", (unsigned long long)ins->objectid,
5830 (unsigned long long)ins->offset);
5836 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5837 struct btrfs_root *root,
5838 u64 root_objectid, u64 owner,
5839 u64 offset, struct btrfs_key *ins)
5843 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5845 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5846 0, root_objectid, owner, offset,
5847 BTRFS_ADD_DELAYED_EXTENT, NULL);
5852 * this is used by the tree logging recovery code. It records that
5853 * an extent has been allocated and makes sure to clear the free
5854 * space cache bits as well
5856 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5857 struct btrfs_root *root,
5858 u64 root_objectid, u64 owner, u64 offset,
5859 struct btrfs_key *ins)
5862 struct btrfs_block_group_cache *block_group;
5863 struct btrfs_caching_control *caching_ctl;
5864 u64 start = ins->objectid;
5865 u64 num_bytes = ins->offset;
5867 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5868 cache_block_group(block_group, trans, NULL, 0);
5869 caching_ctl = get_caching_control(block_group);
5872 BUG_ON(!block_group_cache_done(block_group));
5873 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5876 mutex_lock(&caching_ctl->mutex);
5878 if (start >= caching_ctl->progress) {
5879 ret = add_excluded_extent(root, start, num_bytes);
5881 } else if (start + num_bytes <= caching_ctl->progress) {
5882 ret = btrfs_remove_free_space(block_group,
5886 num_bytes = caching_ctl->progress - start;
5887 ret = btrfs_remove_free_space(block_group,
5891 start = caching_ctl->progress;
5892 num_bytes = ins->objectid + ins->offset -
5893 caching_ctl->progress;
5894 ret = add_excluded_extent(root, start, num_bytes);
5898 mutex_unlock(&caching_ctl->mutex);
5899 put_caching_control(caching_ctl);
5902 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5903 RESERVE_ALLOC_NO_ACCOUNT);
5905 btrfs_put_block_group(block_group);
5906 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5907 0, owner, offset, ins, 1);
5911 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5912 struct btrfs_root *root,
5913 u64 bytenr, u32 blocksize,
5916 struct extent_buffer *buf;
5918 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5920 return ERR_PTR(-ENOMEM);
5921 btrfs_set_header_generation(buf, trans->transid);
5922 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5923 btrfs_tree_lock(buf);
5924 clean_tree_block(trans, root, buf);
5926 btrfs_set_lock_blocking(buf);
5927 btrfs_set_buffer_uptodate(buf);
5929 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5931 * we allow two log transactions at a time, use different
5932 * EXENT bit to differentiate dirty pages.
5934 if (root->log_transid % 2 == 0)
5935 set_extent_dirty(&root->dirty_log_pages, buf->start,
5936 buf->start + buf->len - 1, GFP_NOFS);
5938 set_extent_new(&root->dirty_log_pages, buf->start,
5939 buf->start + buf->len - 1, GFP_NOFS);
5941 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5942 buf->start + buf->len - 1, GFP_NOFS);
5944 trans->blocks_used++;
5945 /* this returns a buffer locked for blocking */
5949 static struct btrfs_block_rsv *
5950 use_block_rsv(struct btrfs_trans_handle *trans,
5951 struct btrfs_root *root, u32 blocksize)
5953 struct btrfs_block_rsv *block_rsv;
5954 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5957 block_rsv = get_block_rsv(trans, root);
5959 if (block_rsv->size == 0) {
5960 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5962 * If we couldn't reserve metadata bytes try and use some from
5963 * the global reserve.
5965 if (ret && block_rsv != global_rsv) {
5966 ret = block_rsv_use_bytes(global_rsv, blocksize);
5969 return ERR_PTR(ret);
5971 return ERR_PTR(ret);
5976 ret = block_rsv_use_bytes(block_rsv, blocksize);
5980 static DEFINE_RATELIMIT_STATE(_rs,
5981 DEFAULT_RATELIMIT_INTERVAL,
5982 /*DEFAULT_RATELIMIT_BURST*/ 2);
5983 if (__ratelimit(&_rs)) {
5984 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5987 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5990 } else if (ret && block_rsv != global_rsv) {
5991 ret = block_rsv_use_bytes(global_rsv, blocksize);
5997 return ERR_PTR(-ENOSPC);
6000 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6002 block_rsv_add_bytes(block_rsv, blocksize, 0);
6003 block_rsv_release_bytes(block_rsv, NULL, 0);
6007 * finds a free extent and does all the dirty work required for allocation
6008 * returns the key for the extent through ins, and a tree buffer for
6009 * the first block of the extent through buf.
6011 * returns the tree buffer or NULL.
6013 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6014 struct btrfs_root *root, u32 blocksize,
6015 u64 parent, u64 root_objectid,
6016 struct btrfs_disk_key *key, int level,
6017 u64 hint, u64 empty_size)
6019 struct btrfs_key ins;
6020 struct btrfs_block_rsv *block_rsv;
6021 struct extent_buffer *buf;
6026 block_rsv = use_block_rsv(trans, root, blocksize);
6027 if (IS_ERR(block_rsv))
6028 return ERR_CAST(block_rsv);
6030 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6031 empty_size, hint, (u64)-1, &ins, 0);
6033 unuse_block_rsv(block_rsv, blocksize);
6034 return ERR_PTR(ret);
6037 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6039 BUG_ON(IS_ERR(buf));
6041 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6043 parent = ins.objectid;
6044 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6048 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6049 struct btrfs_delayed_extent_op *extent_op;
6050 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6053 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6055 memset(&extent_op->key, 0, sizeof(extent_op->key));
6056 extent_op->flags_to_set = flags;
6057 extent_op->update_key = 1;
6058 extent_op->update_flags = 1;
6059 extent_op->is_data = 0;
6061 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
6062 ins.offset, parent, root_objectid,
6063 level, BTRFS_ADD_DELAYED_EXTENT,
6070 struct walk_control {
6071 u64 refs[BTRFS_MAX_LEVEL];
6072 u64 flags[BTRFS_MAX_LEVEL];
6073 struct btrfs_key update_progress;
6083 #define DROP_REFERENCE 1
6084 #define UPDATE_BACKREF 2
6086 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6087 struct btrfs_root *root,
6088 struct walk_control *wc,
6089 struct btrfs_path *path)
6097 struct btrfs_key key;
6098 struct extent_buffer *eb;
6103 if (path->slots[wc->level] < wc->reada_slot) {
6104 wc->reada_count = wc->reada_count * 2 / 3;
6105 wc->reada_count = max(wc->reada_count, 2);
6107 wc->reada_count = wc->reada_count * 3 / 2;
6108 wc->reada_count = min_t(int, wc->reada_count,
6109 BTRFS_NODEPTRS_PER_BLOCK(root));
6112 eb = path->nodes[wc->level];
6113 nritems = btrfs_header_nritems(eb);
6114 blocksize = btrfs_level_size(root, wc->level - 1);
6116 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6117 if (nread >= wc->reada_count)
6121 bytenr = btrfs_node_blockptr(eb, slot);
6122 generation = btrfs_node_ptr_generation(eb, slot);
6124 if (slot == path->slots[wc->level])
6127 if (wc->stage == UPDATE_BACKREF &&
6128 generation <= root->root_key.offset)
6131 /* We don't lock the tree block, it's OK to be racy here */
6132 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6137 if (wc->stage == DROP_REFERENCE) {
6141 if (wc->level == 1 &&
6142 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6144 if (!wc->update_ref ||
6145 generation <= root->root_key.offset)
6147 btrfs_node_key_to_cpu(eb, &key, slot);
6148 ret = btrfs_comp_cpu_keys(&key,
6149 &wc->update_progress);
6153 if (wc->level == 1 &&
6154 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6158 ret = readahead_tree_block(root, bytenr, blocksize,
6164 wc->reada_slot = slot;
6168 * hepler to process tree block while walking down the tree.
6170 * when wc->stage == UPDATE_BACKREF, this function updates
6171 * back refs for pointers in the block.
6173 * NOTE: return value 1 means we should stop walking down.
6175 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6176 struct btrfs_root *root,
6177 struct btrfs_path *path,
6178 struct walk_control *wc, int lookup_info)
6180 int level = wc->level;
6181 struct extent_buffer *eb = path->nodes[level];
6182 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6185 if (wc->stage == UPDATE_BACKREF &&
6186 btrfs_header_owner(eb) != root->root_key.objectid)
6190 * when reference count of tree block is 1, it won't increase
6191 * again. once full backref flag is set, we never clear it.
6194 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6195 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6196 BUG_ON(!path->locks[level]);
6197 ret = btrfs_lookup_extent_info(trans, root,
6202 BUG_ON(wc->refs[level] == 0);
6205 if (wc->stage == DROP_REFERENCE) {
6206 if (wc->refs[level] > 1)
6209 if (path->locks[level] && !wc->keep_locks) {
6210 btrfs_tree_unlock_rw(eb, path->locks[level]);
6211 path->locks[level] = 0;
6216 /* wc->stage == UPDATE_BACKREF */
6217 if (!(wc->flags[level] & flag)) {
6218 BUG_ON(!path->locks[level]);
6219 ret = btrfs_inc_ref(trans, root, eb, 1);
6221 ret = btrfs_dec_ref(trans, root, eb, 0);
6223 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6226 wc->flags[level] |= flag;
6230 * the block is shared by multiple trees, so it's not good to
6231 * keep the tree lock
6233 if (path->locks[level] && level > 0) {
6234 btrfs_tree_unlock_rw(eb, path->locks[level]);
6235 path->locks[level] = 0;
6241 * hepler to process tree block pointer.
6243 * when wc->stage == DROP_REFERENCE, this function checks
6244 * reference count of the block pointed to. if the block
6245 * is shared and we need update back refs for the subtree
6246 * rooted at the block, this function changes wc->stage to
6247 * UPDATE_BACKREF. if the block is shared and there is no
6248 * need to update back, this function drops the reference
6251 * NOTE: return value 1 means we should stop walking down.
6253 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6254 struct btrfs_root *root,
6255 struct btrfs_path *path,
6256 struct walk_control *wc, int *lookup_info)
6262 struct btrfs_key key;
6263 struct extent_buffer *next;
6264 int level = wc->level;
6268 generation = btrfs_node_ptr_generation(path->nodes[level],
6269 path->slots[level]);
6271 * if the lower level block was created before the snapshot
6272 * was created, we know there is no need to update back refs
6275 if (wc->stage == UPDATE_BACKREF &&
6276 generation <= root->root_key.offset) {
6281 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6282 blocksize = btrfs_level_size(root, level - 1);
6284 next = btrfs_find_tree_block(root, bytenr, blocksize);
6286 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6291 btrfs_tree_lock(next);
6292 btrfs_set_lock_blocking(next);
6294 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6295 &wc->refs[level - 1],
6296 &wc->flags[level - 1]);
6298 BUG_ON(wc->refs[level - 1] == 0);
6301 if (wc->stage == DROP_REFERENCE) {
6302 if (wc->refs[level - 1] > 1) {
6304 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6307 if (!wc->update_ref ||
6308 generation <= root->root_key.offset)
6311 btrfs_node_key_to_cpu(path->nodes[level], &key,
6312 path->slots[level]);
6313 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6317 wc->stage = UPDATE_BACKREF;
6318 wc->shared_level = level - 1;
6322 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6326 if (!btrfs_buffer_uptodate(next, generation)) {
6327 btrfs_tree_unlock(next);
6328 free_extent_buffer(next);
6334 if (reada && level == 1)
6335 reada_walk_down(trans, root, wc, path);
6336 next = read_tree_block(root, bytenr, blocksize, generation);
6339 btrfs_tree_lock(next);
6340 btrfs_set_lock_blocking(next);
6344 BUG_ON(level != btrfs_header_level(next));
6345 path->nodes[level] = next;
6346 path->slots[level] = 0;
6347 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6353 wc->refs[level - 1] = 0;
6354 wc->flags[level - 1] = 0;
6355 if (wc->stage == DROP_REFERENCE) {
6356 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6357 parent = path->nodes[level]->start;
6359 BUG_ON(root->root_key.objectid !=
6360 btrfs_header_owner(path->nodes[level]));
6364 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6365 root->root_key.objectid, level - 1, 0);
6368 btrfs_tree_unlock(next);
6369 free_extent_buffer(next);
6375 * hepler to process tree block while walking up the tree.
6377 * when wc->stage == DROP_REFERENCE, this function drops
6378 * reference count on the block.
6380 * when wc->stage == UPDATE_BACKREF, this function changes
6381 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6382 * to UPDATE_BACKREF previously while processing the block.
6384 * NOTE: return value 1 means we should stop walking up.
6386 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6387 struct btrfs_root *root,
6388 struct btrfs_path *path,
6389 struct walk_control *wc)
6392 int level = wc->level;
6393 struct extent_buffer *eb = path->nodes[level];
6396 if (wc->stage == UPDATE_BACKREF) {
6397 BUG_ON(wc->shared_level < level);
6398 if (level < wc->shared_level)
6401 ret = find_next_key(path, level + 1, &wc->update_progress);
6405 wc->stage = DROP_REFERENCE;
6406 wc->shared_level = -1;
6407 path->slots[level] = 0;
6410 * check reference count again if the block isn't locked.
6411 * we should start walking down the tree again if reference
6414 if (!path->locks[level]) {
6416 btrfs_tree_lock(eb);
6417 btrfs_set_lock_blocking(eb);
6418 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6420 ret = btrfs_lookup_extent_info(trans, root,
6425 BUG_ON(wc->refs[level] == 0);
6426 if (wc->refs[level] == 1) {
6427 btrfs_tree_unlock_rw(eb, path->locks[level]);
6433 /* wc->stage == DROP_REFERENCE */
6434 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6436 if (wc->refs[level] == 1) {
6438 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6439 ret = btrfs_dec_ref(trans, root, eb, 1);
6441 ret = btrfs_dec_ref(trans, root, eb, 0);
6444 /* make block locked assertion in clean_tree_block happy */
6445 if (!path->locks[level] &&
6446 btrfs_header_generation(eb) == trans->transid) {
6447 btrfs_tree_lock(eb);
6448 btrfs_set_lock_blocking(eb);
6449 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6451 clean_tree_block(trans, root, eb);
6454 if (eb == root->node) {
6455 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6458 BUG_ON(root->root_key.objectid !=
6459 btrfs_header_owner(eb));
6461 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6462 parent = path->nodes[level + 1]->start;
6464 BUG_ON(root->root_key.objectid !=
6465 btrfs_header_owner(path->nodes[level + 1]));
6468 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6470 wc->refs[level] = 0;
6471 wc->flags[level] = 0;
6475 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6476 struct btrfs_root *root,
6477 struct btrfs_path *path,
6478 struct walk_control *wc)
6480 int level = wc->level;
6481 int lookup_info = 1;
6484 while (level >= 0) {
6485 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6492 if (path->slots[level] >=
6493 btrfs_header_nritems(path->nodes[level]))
6496 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6498 path->slots[level]++;
6507 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6508 struct btrfs_root *root,
6509 struct btrfs_path *path,
6510 struct walk_control *wc, int max_level)
6512 int level = wc->level;
6515 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6516 while (level < max_level && path->nodes[level]) {
6518 if (path->slots[level] + 1 <
6519 btrfs_header_nritems(path->nodes[level])) {
6520 path->slots[level]++;
6523 ret = walk_up_proc(trans, root, path, wc);
6527 if (path->locks[level]) {
6528 btrfs_tree_unlock_rw(path->nodes[level],
6529 path->locks[level]);
6530 path->locks[level] = 0;
6532 free_extent_buffer(path->nodes[level]);
6533 path->nodes[level] = NULL;
6541 * drop a subvolume tree.
6543 * this function traverses the tree freeing any blocks that only
6544 * referenced by the tree.
6546 * when a shared tree block is found. this function decreases its
6547 * reference count by one. if update_ref is true, this function
6548 * also make sure backrefs for the shared block and all lower level
6549 * blocks are properly updated.
6551 void btrfs_drop_snapshot(struct btrfs_root *root,
6552 struct btrfs_block_rsv *block_rsv, int update_ref)
6554 struct btrfs_path *path;
6555 struct btrfs_trans_handle *trans;
6556 struct btrfs_root *tree_root = root->fs_info->tree_root;
6557 struct btrfs_root_item *root_item = &root->root_item;
6558 struct walk_control *wc;
6559 struct btrfs_key key;
6564 path = btrfs_alloc_path();
6570 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6572 btrfs_free_path(path);
6577 trans = btrfs_start_transaction(tree_root, 0);
6578 BUG_ON(IS_ERR(trans));
6581 trans->block_rsv = block_rsv;
6583 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6584 level = btrfs_header_level(root->node);
6585 path->nodes[level] = btrfs_lock_root_node(root);
6586 btrfs_set_lock_blocking(path->nodes[level]);
6587 path->slots[level] = 0;
6588 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6589 memset(&wc->update_progress, 0,
6590 sizeof(wc->update_progress));
6592 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6593 memcpy(&wc->update_progress, &key,
6594 sizeof(wc->update_progress));
6596 level = root_item->drop_level;
6598 path->lowest_level = level;
6599 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6600 path->lowest_level = 0;
6608 * unlock our path, this is safe because only this
6609 * function is allowed to delete this snapshot
6611 btrfs_unlock_up_safe(path, 0);
6613 level = btrfs_header_level(root->node);
6615 btrfs_tree_lock(path->nodes[level]);
6616 btrfs_set_lock_blocking(path->nodes[level]);
6618 ret = btrfs_lookup_extent_info(trans, root,
6619 path->nodes[level]->start,
6620 path->nodes[level]->len,
6624 BUG_ON(wc->refs[level] == 0);
6626 if (level == root_item->drop_level)
6629 btrfs_tree_unlock(path->nodes[level]);
6630 WARN_ON(wc->refs[level] != 1);
6636 wc->shared_level = -1;
6637 wc->stage = DROP_REFERENCE;
6638 wc->update_ref = update_ref;
6640 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6643 ret = walk_down_tree(trans, root, path, wc);
6649 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6656 BUG_ON(wc->stage != DROP_REFERENCE);
6660 if (wc->stage == DROP_REFERENCE) {
6662 btrfs_node_key(path->nodes[level],
6663 &root_item->drop_progress,
6664 path->slots[level]);
6665 root_item->drop_level = level;
6668 BUG_ON(wc->level == 0);
6669 if (btrfs_should_end_transaction(trans, tree_root)) {
6670 ret = btrfs_update_root(trans, tree_root,
6675 btrfs_end_transaction_throttle(trans, tree_root);
6676 trans = btrfs_start_transaction(tree_root, 0);
6677 BUG_ON(IS_ERR(trans));
6679 trans->block_rsv = block_rsv;
6682 btrfs_release_path(path);
6685 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6688 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6689 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6693 /* if we fail to delete the orphan item this time
6694 * around, it'll get picked up the next time.
6696 * The most common failure here is just -ENOENT.
6698 btrfs_del_orphan_item(trans, tree_root,
6699 root->root_key.objectid);
6703 if (root->in_radix) {
6704 btrfs_free_fs_root(tree_root->fs_info, root);
6706 free_extent_buffer(root->node);
6707 free_extent_buffer(root->commit_root);
6711 btrfs_end_transaction_throttle(trans, tree_root);
6713 btrfs_free_path(path);
6716 btrfs_std_error(root->fs_info, err);
6721 * drop subtree rooted at tree block 'node'.
6723 * NOTE: this function will unlock and release tree block 'node'
6725 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6726 struct btrfs_root *root,
6727 struct extent_buffer *node,
6728 struct extent_buffer *parent)
6730 struct btrfs_path *path;
6731 struct walk_control *wc;
6737 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6739 path = btrfs_alloc_path();
6743 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6745 btrfs_free_path(path);
6749 btrfs_assert_tree_locked(parent);
6750 parent_level = btrfs_header_level(parent);
6751 extent_buffer_get(parent);
6752 path->nodes[parent_level] = parent;
6753 path->slots[parent_level] = btrfs_header_nritems(parent);
6755 btrfs_assert_tree_locked(node);
6756 level = btrfs_header_level(node);
6757 path->nodes[level] = node;
6758 path->slots[level] = 0;
6759 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6761 wc->refs[parent_level] = 1;
6762 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6764 wc->shared_level = -1;
6765 wc->stage = DROP_REFERENCE;
6768 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6771 wret = walk_down_tree(trans, root, path, wc);
6777 wret = walk_up_tree(trans, root, path, wc, parent_level);
6785 btrfs_free_path(path);
6789 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6792 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6793 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6796 * we add in the count of missing devices because we want
6797 * to make sure that any RAID levels on a degraded FS
6798 * continue to be honored.
6800 num_devices = root->fs_info->fs_devices->rw_devices +
6801 root->fs_info->fs_devices->missing_devices;
6803 if (num_devices == 1) {
6804 stripped |= BTRFS_BLOCK_GROUP_DUP;
6805 stripped = flags & ~stripped;
6807 /* turn raid0 into single device chunks */
6808 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6811 /* turn mirroring into duplication */
6812 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6813 BTRFS_BLOCK_GROUP_RAID10))
6814 return stripped | BTRFS_BLOCK_GROUP_DUP;
6817 /* they already had raid on here, just return */
6818 if (flags & stripped)
6821 stripped |= BTRFS_BLOCK_GROUP_DUP;
6822 stripped = flags & ~stripped;
6824 /* switch duplicated blocks with raid1 */
6825 if (flags & BTRFS_BLOCK_GROUP_DUP)
6826 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6828 /* turn single device chunks into raid0 */
6829 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6834 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6836 struct btrfs_space_info *sinfo = cache->space_info;
6838 u64 min_allocable_bytes;
6843 * We need some metadata space and system metadata space for
6844 * allocating chunks in some corner cases until we force to set
6845 * it to be readonly.
6848 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6850 min_allocable_bytes = 1 * 1024 * 1024;
6852 min_allocable_bytes = 0;
6854 spin_lock(&sinfo->lock);
6855 spin_lock(&cache->lock);
6862 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6863 cache->bytes_super - btrfs_block_group_used(&cache->item);
6865 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6866 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6867 min_allocable_bytes <= sinfo->total_bytes) {
6868 sinfo->bytes_readonly += num_bytes;
6873 spin_unlock(&cache->lock);
6874 spin_unlock(&sinfo->lock);
6878 int btrfs_set_block_group_ro(struct btrfs_root *root,
6879 struct btrfs_block_group_cache *cache)
6882 struct btrfs_trans_handle *trans;
6888 trans = btrfs_join_transaction(root);
6889 BUG_ON(IS_ERR(trans));
6891 alloc_flags = update_block_group_flags(root, cache->flags);
6892 if (alloc_flags != cache->flags)
6893 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6896 ret = set_block_group_ro(cache, 0);
6899 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6900 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6904 ret = set_block_group_ro(cache, 0);
6906 btrfs_end_transaction(trans, root);
6910 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6911 struct btrfs_root *root, u64 type)
6913 u64 alloc_flags = get_alloc_profile(root, type);
6914 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6919 * helper to account the unused space of all the readonly block group in the
6920 * list. takes mirrors into account.
6922 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6924 struct btrfs_block_group_cache *block_group;
6928 list_for_each_entry(block_group, groups_list, list) {
6929 spin_lock(&block_group->lock);
6931 if (!block_group->ro) {
6932 spin_unlock(&block_group->lock);
6936 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6937 BTRFS_BLOCK_GROUP_RAID10 |
6938 BTRFS_BLOCK_GROUP_DUP))
6943 free_bytes += (block_group->key.offset -
6944 btrfs_block_group_used(&block_group->item)) *
6947 spin_unlock(&block_group->lock);
6954 * helper to account the unused space of all the readonly block group in the
6955 * space_info. takes mirrors into account.
6957 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6962 spin_lock(&sinfo->lock);
6964 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6965 if (!list_empty(&sinfo->block_groups[i]))
6966 free_bytes += __btrfs_get_ro_block_group_free_space(
6967 &sinfo->block_groups[i]);
6969 spin_unlock(&sinfo->lock);
6974 int btrfs_set_block_group_rw(struct btrfs_root *root,
6975 struct btrfs_block_group_cache *cache)
6977 struct btrfs_space_info *sinfo = cache->space_info;
6982 spin_lock(&sinfo->lock);
6983 spin_lock(&cache->lock);
6984 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6985 cache->bytes_super - btrfs_block_group_used(&cache->item);
6986 sinfo->bytes_readonly -= num_bytes;
6988 spin_unlock(&cache->lock);
6989 spin_unlock(&sinfo->lock);
6994 * checks to see if its even possible to relocate this block group.
6996 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6997 * ok to go ahead and try.
6999 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7001 struct btrfs_block_group_cache *block_group;
7002 struct btrfs_space_info *space_info;
7003 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7004 struct btrfs_device *device;
7012 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7014 /* odd, couldn't find the block group, leave it alone */
7018 min_free = btrfs_block_group_used(&block_group->item);
7020 /* no bytes used, we're good */
7024 space_info = block_group->space_info;
7025 spin_lock(&space_info->lock);
7027 full = space_info->full;
7030 * if this is the last block group we have in this space, we can't
7031 * relocate it unless we're able to allocate a new chunk below.
7033 * Otherwise, we need to make sure we have room in the space to handle
7034 * all of the extents from this block group. If we can, we're good
7036 if ((space_info->total_bytes != block_group->key.offset) &&
7037 (space_info->bytes_used + space_info->bytes_reserved +
7038 space_info->bytes_pinned + space_info->bytes_readonly +
7039 min_free < space_info->total_bytes)) {
7040 spin_unlock(&space_info->lock);
7043 spin_unlock(&space_info->lock);
7046 * ok we don't have enough space, but maybe we have free space on our
7047 * devices to allocate new chunks for relocation, so loop through our
7048 * alloc devices and guess if we have enough space. However, if we
7049 * were marked as full, then we know there aren't enough chunks, and we
7064 index = get_block_group_index(block_group);
7069 } else if (index == 1) {
7071 } else if (index == 2) {
7074 } else if (index == 3) {
7075 dev_min = fs_devices->rw_devices;
7076 do_div(min_free, dev_min);
7079 mutex_lock(&root->fs_info->chunk_mutex);
7080 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7084 * check to make sure we can actually find a chunk with enough
7085 * space to fit our block group in.
7087 if (device->total_bytes > device->bytes_used + min_free) {
7088 ret = find_free_dev_extent(NULL, device, min_free,
7093 if (dev_nr >= dev_min)
7099 mutex_unlock(&root->fs_info->chunk_mutex);
7101 btrfs_put_block_group(block_group);
7105 static int find_first_block_group(struct btrfs_root *root,
7106 struct btrfs_path *path, struct btrfs_key *key)
7109 struct btrfs_key found_key;
7110 struct extent_buffer *leaf;
7113 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7118 slot = path->slots[0];
7119 leaf = path->nodes[0];
7120 if (slot >= btrfs_header_nritems(leaf)) {
7121 ret = btrfs_next_leaf(root, path);
7128 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7130 if (found_key.objectid >= key->objectid &&
7131 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7141 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7143 struct btrfs_block_group_cache *block_group;
7147 struct inode *inode;
7149 block_group = btrfs_lookup_first_block_group(info, last);
7150 while (block_group) {
7151 spin_lock(&block_group->lock);
7152 if (block_group->iref)
7154 spin_unlock(&block_group->lock);
7155 block_group = next_block_group(info->tree_root,
7165 inode = block_group->inode;
7166 block_group->iref = 0;
7167 block_group->inode = NULL;
7168 spin_unlock(&block_group->lock);
7170 last = block_group->key.objectid + block_group->key.offset;
7171 btrfs_put_block_group(block_group);
7175 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7177 struct btrfs_block_group_cache *block_group;
7178 struct btrfs_space_info *space_info;
7179 struct btrfs_caching_control *caching_ctl;
7182 down_write(&info->extent_commit_sem);
7183 while (!list_empty(&info->caching_block_groups)) {
7184 caching_ctl = list_entry(info->caching_block_groups.next,
7185 struct btrfs_caching_control, list);
7186 list_del(&caching_ctl->list);
7187 put_caching_control(caching_ctl);
7189 up_write(&info->extent_commit_sem);
7191 spin_lock(&info->block_group_cache_lock);
7192 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7193 block_group = rb_entry(n, struct btrfs_block_group_cache,
7195 rb_erase(&block_group->cache_node,
7196 &info->block_group_cache_tree);
7197 spin_unlock(&info->block_group_cache_lock);
7199 down_write(&block_group->space_info->groups_sem);
7200 list_del(&block_group->list);
7201 up_write(&block_group->space_info->groups_sem);
7203 if (block_group->cached == BTRFS_CACHE_STARTED)
7204 wait_block_group_cache_done(block_group);
7207 * We haven't cached this block group, which means we could
7208 * possibly have excluded extents on this block group.
7210 if (block_group->cached == BTRFS_CACHE_NO)
7211 free_excluded_extents(info->extent_root, block_group);
7213 btrfs_remove_free_space_cache(block_group);
7214 btrfs_put_block_group(block_group);
7216 spin_lock(&info->block_group_cache_lock);
7218 spin_unlock(&info->block_group_cache_lock);
7220 /* now that all the block groups are freed, go through and
7221 * free all the space_info structs. This is only called during
7222 * the final stages of unmount, and so we know nobody is
7223 * using them. We call synchronize_rcu() once before we start,
7224 * just to be on the safe side.
7228 release_global_block_rsv(info);
7230 while(!list_empty(&info->space_info)) {
7231 space_info = list_entry(info->space_info.next,
7232 struct btrfs_space_info,
7234 if (space_info->bytes_pinned > 0 ||
7235 space_info->bytes_reserved > 0 ||
7236 space_info->bytes_may_use > 0) {
7238 dump_space_info(space_info, 0, 0);
7240 list_del(&space_info->list);
7246 static void __link_block_group(struct btrfs_space_info *space_info,
7247 struct btrfs_block_group_cache *cache)
7249 int index = get_block_group_index(cache);
7251 down_write(&space_info->groups_sem);
7252 list_add_tail(&cache->list, &space_info->block_groups[index]);
7253 up_write(&space_info->groups_sem);
7256 int btrfs_read_block_groups(struct btrfs_root *root)
7258 struct btrfs_path *path;
7260 struct btrfs_block_group_cache *cache;
7261 struct btrfs_fs_info *info = root->fs_info;
7262 struct btrfs_space_info *space_info;
7263 struct btrfs_key key;
7264 struct btrfs_key found_key;
7265 struct extent_buffer *leaf;
7269 root = info->extent_root;
7272 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7273 path = btrfs_alloc_path();
7278 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7279 if (btrfs_test_opt(root, SPACE_CACHE) &&
7280 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7282 if (btrfs_test_opt(root, CLEAR_CACHE))
7286 ret = find_first_block_group(root, path, &key);
7291 leaf = path->nodes[0];
7292 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7293 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7298 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7300 if (!cache->free_space_ctl) {
7306 atomic_set(&cache->count, 1);
7307 spin_lock_init(&cache->lock);
7308 cache->fs_info = info;
7309 INIT_LIST_HEAD(&cache->list);
7310 INIT_LIST_HEAD(&cache->cluster_list);
7313 cache->disk_cache_state = BTRFS_DC_CLEAR;
7315 read_extent_buffer(leaf, &cache->item,
7316 btrfs_item_ptr_offset(leaf, path->slots[0]),
7317 sizeof(cache->item));
7318 memcpy(&cache->key, &found_key, sizeof(found_key));
7320 key.objectid = found_key.objectid + found_key.offset;
7321 btrfs_release_path(path);
7322 cache->flags = btrfs_block_group_flags(&cache->item);
7323 cache->sectorsize = root->sectorsize;
7325 btrfs_init_free_space_ctl(cache);
7328 * We need to exclude the super stripes now so that the space
7329 * info has super bytes accounted for, otherwise we'll think
7330 * we have more space than we actually do.
7332 exclude_super_stripes(root, cache);
7335 * check for two cases, either we are full, and therefore
7336 * don't need to bother with the caching work since we won't
7337 * find any space, or we are empty, and we can just add all
7338 * the space in and be done with it. This saves us _alot_ of
7339 * time, particularly in the full case.
7341 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7342 cache->last_byte_to_unpin = (u64)-1;
7343 cache->cached = BTRFS_CACHE_FINISHED;
7344 free_excluded_extents(root, cache);
7345 } else if (btrfs_block_group_used(&cache->item) == 0) {
7346 cache->last_byte_to_unpin = (u64)-1;
7347 cache->cached = BTRFS_CACHE_FINISHED;
7348 add_new_free_space(cache, root->fs_info,
7350 found_key.objectid +
7352 free_excluded_extents(root, cache);
7355 ret = update_space_info(info, cache->flags, found_key.offset,
7356 btrfs_block_group_used(&cache->item),
7359 cache->space_info = space_info;
7360 spin_lock(&cache->space_info->lock);
7361 cache->space_info->bytes_readonly += cache->bytes_super;
7362 spin_unlock(&cache->space_info->lock);
7364 __link_block_group(space_info, cache);
7366 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7369 set_avail_alloc_bits(root->fs_info, cache->flags);
7370 if (btrfs_chunk_readonly(root, cache->key.objectid))
7371 set_block_group_ro(cache, 1);
7374 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7375 if (!(get_alloc_profile(root, space_info->flags) &
7376 (BTRFS_BLOCK_GROUP_RAID10 |
7377 BTRFS_BLOCK_GROUP_RAID1 |
7378 BTRFS_BLOCK_GROUP_DUP)))
7381 * avoid allocating from un-mirrored block group if there are
7382 * mirrored block groups.
7384 list_for_each_entry(cache, &space_info->block_groups[3], list)
7385 set_block_group_ro(cache, 1);
7386 list_for_each_entry(cache, &space_info->block_groups[4], list)
7387 set_block_group_ro(cache, 1);
7390 init_global_block_rsv(info);
7393 btrfs_free_path(path);
7397 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7398 struct btrfs_root *root, u64 bytes_used,
7399 u64 type, u64 chunk_objectid, u64 chunk_offset,
7403 struct btrfs_root *extent_root;
7404 struct btrfs_block_group_cache *cache;
7406 extent_root = root->fs_info->extent_root;
7408 root->fs_info->last_trans_log_full_commit = trans->transid;
7410 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7413 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7415 if (!cache->free_space_ctl) {
7420 cache->key.objectid = chunk_offset;
7421 cache->key.offset = size;
7422 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7423 cache->sectorsize = root->sectorsize;
7424 cache->fs_info = root->fs_info;
7426 atomic_set(&cache->count, 1);
7427 spin_lock_init(&cache->lock);
7428 INIT_LIST_HEAD(&cache->list);
7429 INIT_LIST_HEAD(&cache->cluster_list);
7431 btrfs_init_free_space_ctl(cache);
7433 btrfs_set_block_group_used(&cache->item, bytes_used);
7434 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7435 cache->flags = type;
7436 btrfs_set_block_group_flags(&cache->item, type);
7438 cache->last_byte_to_unpin = (u64)-1;
7439 cache->cached = BTRFS_CACHE_FINISHED;
7440 exclude_super_stripes(root, cache);
7442 add_new_free_space(cache, root->fs_info, chunk_offset,
7443 chunk_offset + size);
7445 free_excluded_extents(root, cache);
7447 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7448 &cache->space_info);
7451 spin_lock(&cache->space_info->lock);
7452 cache->space_info->bytes_readonly += cache->bytes_super;
7453 spin_unlock(&cache->space_info->lock);
7455 __link_block_group(cache->space_info, cache);
7457 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7460 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7461 sizeof(cache->item));
7464 set_avail_alloc_bits(extent_root->fs_info, type);
7469 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7470 struct btrfs_root *root, u64 group_start)
7472 struct btrfs_path *path;
7473 struct btrfs_block_group_cache *block_group;
7474 struct btrfs_free_cluster *cluster;
7475 struct btrfs_root *tree_root = root->fs_info->tree_root;
7476 struct btrfs_key key;
7477 struct inode *inode;
7481 root = root->fs_info->extent_root;
7483 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7484 BUG_ON(!block_group);
7485 BUG_ON(!block_group->ro);
7488 * Free the reserved super bytes from this block group before
7491 free_excluded_extents(root, block_group);
7493 memcpy(&key, &block_group->key, sizeof(key));
7494 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7495 BTRFS_BLOCK_GROUP_RAID1 |
7496 BTRFS_BLOCK_GROUP_RAID10))
7501 /* make sure this block group isn't part of an allocation cluster */
7502 cluster = &root->fs_info->data_alloc_cluster;
7503 spin_lock(&cluster->refill_lock);
7504 btrfs_return_cluster_to_free_space(block_group, cluster);
7505 spin_unlock(&cluster->refill_lock);
7508 * make sure this block group isn't part of a metadata
7509 * allocation cluster
7511 cluster = &root->fs_info->meta_alloc_cluster;
7512 spin_lock(&cluster->refill_lock);
7513 btrfs_return_cluster_to_free_space(block_group, cluster);
7514 spin_unlock(&cluster->refill_lock);
7516 path = btrfs_alloc_path();
7522 inode = lookup_free_space_inode(tree_root, block_group, path);
7523 if (!IS_ERR(inode)) {
7524 ret = btrfs_orphan_add(trans, inode);
7527 /* One for the block groups ref */
7528 spin_lock(&block_group->lock);
7529 if (block_group->iref) {
7530 block_group->iref = 0;
7531 block_group->inode = NULL;
7532 spin_unlock(&block_group->lock);
7535 spin_unlock(&block_group->lock);
7537 /* One for our lookup ref */
7538 btrfs_add_delayed_iput(inode);
7541 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7542 key.offset = block_group->key.objectid;
7545 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7549 btrfs_release_path(path);
7551 ret = btrfs_del_item(trans, tree_root, path);
7554 btrfs_release_path(path);
7557 spin_lock(&root->fs_info->block_group_cache_lock);
7558 rb_erase(&block_group->cache_node,
7559 &root->fs_info->block_group_cache_tree);
7560 spin_unlock(&root->fs_info->block_group_cache_lock);
7562 down_write(&block_group->space_info->groups_sem);
7564 * we must use list_del_init so people can check to see if they
7565 * are still on the list after taking the semaphore
7567 list_del_init(&block_group->list);
7568 up_write(&block_group->space_info->groups_sem);
7570 if (block_group->cached == BTRFS_CACHE_STARTED)
7571 wait_block_group_cache_done(block_group);
7573 btrfs_remove_free_space_cache(block_group);
7575 spin_lock(&block_group->space_info->lock);
7576 block_group->space_info->total_bytes -= block_group->key.offset;
7577 block_group->space_info->bytes_readonly -= block_group->key.offset;
7578 block_group->space_info->disk_total -= block_group->key.offset * factor;
7579 spin_unlock(&block_group->space_info->lock);
7581 memcpy(&key, &block_group->key, sizeof(key));
7583 btrfs_clear_space_info_full(root->fs_info);
7585 btrfs_put_block_group(block_group);
7586 btrfs_put_block_group(block_group);
7588 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7594 ret = btrfs_del_item(trans, root, path);
7596 btrfs_free_path(path);
7600 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7602 struct btrfs_space_info *space_info;
7603 struct btrfs_super_block *disk_super;
7609 disk_super = fs_info->super_copy;
7610 if (!btrfs_super_root(disk_super))
7613 features = btrfs_super_incompat_flags(disk_super);
7614 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7617 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7618 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7623 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7624 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7626 flags = BTRFS_BLOCK_GROUP_METADATA;
7627 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7631 flags = BTRFS_BLOCK_GROUP_DATA;
7632 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7638 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7640 return unpin_extent_range(root, start, end);
7643 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7644 u64 num_bytes, u64 *actual_bytes)
7646 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7649 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7651 struct btrfs_fs_info *fs_info = root->fs_info;
7652 struct btrfs_block_group_cache *cache = NULL;
7659 cache = btrfs_lookup_block_group(fs_info, range->start);
7662 if (cache->key.objectid >= (range->start + range->len)) {
7663 btrfs_put_block_group(cache);
7667 start = max(range->start, cache->key.objectid);
7668 end = min(range->start + range->len,
7669 cache->key.objectid + cache->key.offset);
7671 if (end - start >= range->minlen) {
7672 if (!block_group_cache_done(cache)) {
7673 ret = cache_block_group(cache, NULL, root, 0);
7675 wait_block_group_cache_done(cache);
7677 ret = btrfs_trim_block_group(cache,
7683 trimmed += group_trimmed;
7685 btrfs_put_block_group(cache);
7690 cache = next_block_group(fs_info->tree_root, cache);
7693 range->len = trimmed;