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>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
56 * Control how reservations are dealt with.
58 * RESERVE_FREE - freeing a reservation.
59 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
61 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
62 * bytes_may_use as the ENOSPC accounting is done elsewhere
67 RESERVE_ALLOC_NO_ACCOUNT = 2,
70 static int update_block_group(struct btrfs_trans_handle *trans,
71 struct btrfs_root *root,
72 u64 bytenr, u64 num_bytes, int alloc);
73 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
74 struct btrfs_root *root,
75 u64 bytenr, u64 num_bytes, u64 parent,
76 u64 root_objectid, u64 owner_objectid,
77 u64 owner_offset, int refs_to_drop,
78 struct btrfs_delayed_extent_op *extra_op);
79 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
80 struct extent_buffer *leaf,
81 struct btrfs_extent_item *ei);
82 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
83 struct btrfs_root *root,
84 u64 parent, u64 root_objectid,
85 u64 flags, u64 owner, u64 offset,
86 struct btrfs_key *ins, int ref_mod);
87 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, struct btrfs_disk_key *key,
91 int level, struct btrfs_key *ins);
92 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
93 struct btrfs_root *extent_root, u64 alloc_bytes,
94 u64 flags, int force);
95 static int find_next_key(struct btrfs_path *path, int level,
96 struct btrfs_key *key);
97 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
98 int dump_block_groups);
99 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
100 u64 num_bytes, int reserve);
103 block_group_cache_done(struct btrfs_block_group_cache *cache)
106 return cache->cached == BTRFS_CACHE_FINISHED;
109 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
111 return (cache->flags & bits) == bits;
114 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
116 atomic_inc(&cache->count);
119 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
121 if (atomic_dec_and_test(&cache->count)) {
122 WARN_ON(cache->pinned > 0);
123 WARN_ON(cache->reserved > 0);
124 kfree(cache->free_space_ctl);
130 * this adds the block group to the fs_info rb tree for the block group
133 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
134 struct btrfs_block_group_cache *block_group)
137 struct rb_node *parent = NULL;
138 struct btrfs_block_group_cache *cache;
140 spin_lock(&info->block_group_cache_lock);
141 p = &info->block_group_cache_tree.rb_node;
145 cache = rb_entry(parent, struct btrfs_block_group_cache,
147 if (block_group->key.objectid < cache->key.objectid) {
149 } else if (block_group->key.objectid > cache->key.objectid) {
152 spin_unlock(&info->block_group_cache_lock);
157 rb_link_node(&block_group->cache_node, parent, p);
158 rb_insert_color(&block_group->cache_node,
159 &info->block_group_cache_tree);
160 spin_unlock(&info->block_group_cache_lock);
166 * This will return the block group at or after bytenr if contains is 0, else
167 * it will return the block group that contains the bytenr
169 static struct btrfs_block_group_cache *
170 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
173 struct btrfs_block_group_cache *cache, *ret = NULL;
177 spin_lock(&info->block_group_cache_lock);
178 n = info->block_group_cache_tree.rb_node;
181 cache = rb_entry(n, struct btrfs_block_group_cache,
183 end = cache->key.objectid + cache->key.offset - 1;
184 start = cache->key.objectid;
186 if (bytenr < start) {
187 if (!contains && (!ret || start < ret->key.objectid))
190 } else if (bytenr > start) {
191 if (contains && bytenr <= end) {
202 btrfs_get_block_group(ret);
203 spin_unlock(&info->block_group_cache_lock);
208 static int add_excluded_extent(struct btrfs_root *root,
209 u64 start, u64 num_bytes)
211 u64 end = start + num_bytes - 1;
212 set_extent_bits(&root->fs_info->freed_extents[0],
213 start, end, EXTENT_UPTODATE, GFP_NOFS);
214 set_extent_bits(&root->fs_info->freed_extents[1],
215 start, end, EXTENT_UPTODATE, GFP_NOFS);
219 static void free_excluded_extents(struct btrfs_root *root,
220 struct btrfs_block_group_cache *cache)
224 start = cache->key.objectid;
225 end = start + cache->key.offset - 1;
227 clear_extent_bits(&root->fs_info->freed_extents[0],
228 start, end, EXTENT_UPTODATE, GFP_NOFS);
229 clear_extent_bits(&root->fs_info->freed_extents[1],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 static int exclude_super_stripes(struct btrfs_root *root,
234 struct btrfs_block_group_cache *cache)
241 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
242 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
243 cache->bytes_super += stripe_len;
244 ret = add_excluded_extent(root, cache->key.objectid,
249 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
250 bytenr = btrfs_sb_offset(i);
251 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
252 cache->key.objectid, bytenr,
253 0, &logical, &nr, &stripe_len);
257 cache->bytes_super += stripe_len;
258 ret = add_excluded_extent(root, logical[nr],
268 static struct btrfs_caching_control *
269 get_caching_control(struct btrfs_block_group_cache *cache)
271 struct btrfs_caching_control *ctl;
273 spin_lock(&cache->lock);
274 if (cache->cached != BTRFS_CACHE_STARTED) {
275 spin_unlock(&cache->lock);
279 /* We're loading it the fast way, so we don't have a caching_ctl. */
280 if (!cache->caching_ctl) {
281 spin_unlock(&cache->lock);
285 ctl = cache->caching_ctl;
286 atomic_inc(&ctl->count);
287 spin_unlock(&cache->lock);
291 static void put_caching_control(struct btrfs_caching_control *ctl)
293 if (atomic_dec_and_test(&ctl->count))
298 * this is only called by cache_block_group, since we could have freed extents
299 * we need to check the pinned_extents for any extents that can't be used yet
300 * since their free space will be released as soon as the transaction commits.
302 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
303 struct btrfs_fs_info *info, u64 start, u64 end)
305 u64 extent_start, extent_end, size, total_added = 0;
308 while (start < end) {
309 ret = find_first_extent_bit(info->pinned_extents, start,
310 &extent_start, &extent_end,
311 EXTENT_DIRTY | EXTENT_UPTODATE);
315 if (extent_start <= start) {
316 start = extent_end + 1;
317 } else if (extent_start > start && extent_start < end) {
318 size = extent_start - start;
320 ret = btrfs_add_free_space(block_group, start,
323 start = extent_end + 1;
332 ret = btrfs_add_free_space(block_group, start, size);
339 static noinline void caching_thread(struct btrfs_work *work)
341 struct btrfs_block_group_cache *block_group;
342 struct btrfs_fs_info *fs_info;
343 struct btrfs_caching_control *caching_ctl;
344 struct btrfs_root *extent_root;
345 struct btrfs_path *path;
346 struct extent_buffer *leaf;
347 struct btrfs_key key;
353 caching_ctl = container_of(work, struct btrfs_caching_control, work);
354 block_group = caching_ctl->block_group;
355 fs_info = block_group->fs_info;
356 extent_root = fs_info->extent_root;
358 path = btrfs_alloc_path();
362 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
365 * We don't want to deadlock with somebody trying to allocate a new
366 * extent for the extent root while also trying to search the extent
367 * root to add free space. So we skip locking and search the commit
368 * root, since its read-only
370 path->skip_locking = 1;
371 path->search_commit_root = 1;
376 key.type = BTRFS_EXTENT_ITEM_KEY;
378 mutex_lock(&caching_ctl->mutex);
379 /* need to make sure the commit_root doesn't disappear */
380 down_read(&fs_info->extent_commit_sem);
382 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
386 leaf = path->nodes[0];
387 nritems = btrfs_header_nritems(leaf);
390 if (btrfs_fs_closing(fs_info) > 1) {
395 if (path->slots[0] < nritems) {
396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
398 ret = find_next_key(path, 0, &key);
402 if (need_resched() ||
403 btrfs_next_leaf(extent_root, path)) {
404 caching_ctl->progress = last;
405 btrfs_release_path(path);
406 up_read(&fs_info->extent_commit_sem);
407 mutex_unlock(&caching_ctl->mutex);
411 leaf = path->nodes[0];
412 nritems = btrfs_header_nritems(leaf);
416 if (key.objectid < block_group->key.objectid) {
421 if (key.objectid >= block_group->key.objectid +
422 block_group->key.offset)
425 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
426 total_found += add_new_free_space(block_group,
429 last = key.objectid + key.offset;
431 if (total_found > (1024 * 1024 * 2)) {
433 wake_up(&caching_ctl->wait);
440 total_found += add_new_free_space(block_group, fs_info, last,
441 block_group->key.objectid +
442 block_group->key.offset);
443 caching_ctl->progress = (u64)-1;
445 spin_lock(&block_group->lock);
446 block_group->caching_ctl = NULL;
447 block_group->cached = BTRFS_CACHE_FINISHED;
448 spin_unlock(&block_group->lock);
451 btrfs_free_path(path);
452 up_read(&fs_info->extent_commit_sem);
454 free_excluded_extents(extent_root, block_group);
456 mutex_unlock(&caching_ctl->mutex);
458 wake_up(&caching_ctl->wait);
460 put_caching_control(caching_ctl);
461 btrfs_put_block_group(block_group);
464 static int cache_block_group(struct btrfs_block_group_cache *cache,
465 struct btrfs_trans_handle *trans,
466 struct btrfs_root *root,
469 struct btrfs_fs_info *fs_info = cache->fs_info;
470 struct btrfs_caching_control *caching_ctl;
474 if (cache->cached != BTRFS_CACHE_NO)
478 * We can't do the read from on-disk cache during a commit since we need
479 * to have the normal tree locking. Also if we are currently trying to
480 * allocate blocks for the tree root we can't do the fast caching since
481 * we likely hold important locks.
483 if (trans && (!trans->transaction->in_commit) &&
484 (root && root != root->fs_info->tree_root) &&
485 btrfs_test_opt(root, SPACE_CACHE)) {
486 spin_lock(&cache->lock);
487 if (cache->cached != BTRFS_CACHE_NO) {
488 spin_unlock(&cache->lock);
491 cache->cached = BTRFS_CACHE_STARTED;
492 spin_unlock(&cache->lock);
494 ret = load_free_space_cache(fs_info, cache);
496 spin_lock(&cache->lock);
498 cache->cached = BTRFS_CACHE_FINISHED;
499 cache->last_byte_to_unpin = (u64)-1;
501 cache->cached = BTRFS_CACHE_NO;
503 spin_unlock(&cache->lock);
505 free_excluded_extents(fs_info->extent_root, cache);
513 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
514 BUG_ON(!caching_ctl);
516 INIT_LIST_HEAD(&caching_ctl->list);
517 mutex_init(&caching_ctl->mutex);
518 init_waitqueue_head(&caching_ctl->wait);
519 caching_ctl->block_group = cache;
520 caching_ctl->progress = cache->key.objectid;
521 /* one for caching kthread, one for caching block group list */
522 atomic_set(&caching_ctl->count, 2);
523 caching_ctl->work.func = caching_thread;
525 spin_lock(&cache->lock);
526 if (cache->cached != BTRFS_CACHE_NO) {
527 spin_unlock(&cache->lock);
531 cache->caching_ctl = caching_ctl;
532 cache->cached = BTRFS_CACHE_STARTED;
533 spin_unlock(&cache->lock);
535 down_write(&fs_info->extent_commit_sem);
536 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
537 up_write(&fs_info->extent_commit_sem);
539 btrfs_get_block_group(cache);
541 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
547 * return the block group that starts at or after bytenr
549 static struct btrfs_block_group_cache *
550 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
552 struct btrfs_block_group_cache *cache;
554 cache = block_group_cache_tree_search(info, bytenr, 0);
560 * return the block group that contains the given bytenr
562 struct btrfs_block_group_cache *btrfs_lookup_block_group(
563 struct btrfs_fs_info *info,
566 struct btrfs_block_group_cache *cache;
568 cache = block_group_cache_tree_search(info, bytenr, 1);
573 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
576 struct list_head *head = &info->space_info;
577 struct btrfs_space_info *found;
579 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
580 BTRFS_BLOCK_GROUP_METADATA;
583 list_for_each_entry_rcu(found, head, list) {
584 if (found->flags & flags) {
594 * after adding space to the filesystem, we need to clear the full flags
595 * on all the space infos.
597 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
599 struct list_head *head = &info->space_info;
600 struct btrfs_space_info *found;
603 list_for_each_entry_rcu(found, head, list)
608 static u64 div_factor(u64 num, int factor)
617 static u64 div_factor_fine(u64 num, int factor)
626 u64 btrfs_find_block_group(struct btrfs_root *root,
627 u64 search_start, u64 search_hint, int owner)
629 struct btrfs_block_group_cache *cache;
631 u64 last = max(search_hint, search_start);
638 cache = btrfs_lookup_first_block_group(root->fs_info, last);
642 spin_lock(&cache->lock);
643 last = cache->key.objectid + cache->key.offset;
644 used = btrfs_block_group_used(&cache->item);
646 if ((full_search || !cache->ro) &&
647 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
648 if (used + cache->pinned + cache->reserved <
649 div_factor(cache->key.offset, factor)) {
650 group_start = cache->key.objectid;
651 spin_unlock(&cache->lock);
652 btrfs_put_block_group(cache);
656 spin_unlock(&cache->lock);
657 btrfs_put_block_group(cache);
665 if (!full_search && factor < 10) {
675 /* simple helper to search for an existing extent at a given offset */
676 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
679 struct btrfs_key key;
680 struct btrfs_path *path;
682 path = btrfs_alloc_path();
686 key.objectid = start;
688 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
689 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
691 btrfs_free_path(path);
696 * helper function to lookup reference count and flags of extent.
698 * the head node for delayed ref is used to store the sum of all the
699 * reference count modifications queued up in the rbtree. the head
700 * node may also store the extent flags to set. This way you can check
701 * to see what the reference count and extent flags would be if all of
702 * the delayed refs are not processed.
704 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
705 struct btrfs_root *root, u64 bytenr,
706 u64 num_bytes, u64 *refs, u64 *flags)
708 struct btrfs_delayed_ref_head *head;
709 struct btrfs_delayed_ref_root *delayed_refs;
710 struct btrfs_path *path;
711 struct btrfs_extent_item *ei;
712 struct extent_buffer *leaf;
713 struct btrfs_key key;
719 path = btrfs_alloc_path();
723 key.objectid = bytenr;
724 key.type = BTRFS_EXTENT_ITEM_KEY;
725 key.offset = num_bytes;
727 path->skip_locking = 1;
728 path->search_commit_root = 1;
731 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
737 leaf = path->nodes[0];
738 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
739 if (item_size >= sizeof(*ei)) {
740 ei = btrfs_item_ptr(leaf, path->slots[0],
741 struct btrfs_extent_item);
742 num_refs = btrfs_extent_refs(leaf, ei);
743 extent_flags = btrfs_extent_flags(leaf, ei);
745 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
746 struct btrfs_extent_item_v0 *ei0;
747 BUG_ON(item_size != sizeof(*ei0));
748 ei0 = btrfs_item_ptr(leaf, path->slots[0],
749 struct btrfs_extent_item_v0);
750 num_refs = btrfs_extent_refs_v0(leaf, ei0);
751 /* FIXME: this isn't correct for data */
752 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
757 BUG_ON(num_refs == 0);
767 delayed_refs = &trans->transaction->delayed_refs;
768 spin_lock(&delayed_refs->lock);
769 head = btrfs_find_delayed_ref_head(trans, bytenr);
771 if (!mutex_trylock(&head->mutex)) {
772 atomic_inc(&head->node.refs);
773 spin_unlock(&delayed_refs->lock);
775 btrfs_release_path(path);
778 * Mutex was contended, block until it's released and try
781 mutex_lock(&head->mutex);
782 mutex_unlock(&head->mutex);
783 btrfs_put_delayed_ref(&head->node);
786 if (head->extent_op && head->extent_op->update_flags)
787 extent_flags |= head->extent_op->flags_to_set;
789 BUG_ON(num_refs == 0);
791 num_refs += head->node.ref_mod;
792 mutex_unlock(&head->mutex);
794 spin_unlock(&delayed_refs->lock);
796 WARN_ON(num_refs == 0);
800 *flags = extent_flags;
802 btrfs_free_path(path);
807 * Back reference rules. Back refs have three main goals:
809 * 1) differentiate between all holders of references to an extent so that
810 * when a reference is dropped we can make sure it was a valid reference
811 * before freeing the extent.
813 * 2) Provide enough information to quickly find the holders of an extent
814 * if we notice a given block is corrupted or bad.
816 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
817 * maintenance. This is actually the same as #2, but with a slightly
818 * different use case.
820 * There are two kinds of back refs. The implicit back refs is optimized
821 * for pointers in non-shared tree blocks. For a given pointer in a block,
822 * back refs of this kind provide information about the block's owner tree
823 * and the pointer's key. These information allow us to find the block by
824 * b-tree searching. The full back refs is for pointers in tree blocks not
825 * referenced by their owner trees. The location of tree block is recorded
826 * in the back refs. Actually the full back refs is generic, and can be
827 * used in all cases the implicit back refs is used. The major shortcoming
828 * of the full back refs is its overhead. Every time a tree block gets
829 * COWed, we have to update back refs entry for all pointers in it.
831 * For a newly allocated tree block, we use implicit back refs for
832 * pointers in it. This means most tree related operations only involve
833 * implicit back refs. For a tree block created in old transaction, the
834 * only way to drop a reference to it is COW it. So we can detect the
835 * event that tree block loses its owner tree's reference and do the
836 * back refs conversion.
838 * When a tree block is COW'd through a tree, there are four cases:
840 * The reference count of the block is one and the tree is the block's
841 * owner tree. Nothing to do in this case.
843 * The reference count of the block is one and the tree is not the
844 * block's owner tree. In this case, full back refs is used for pointers
845 * in the block. Remove these full back refs, add implicit back refs for
846 * every pointers in the new block.
848 * The reference count of the block is greater than one and the tree is
849 * the block's owner tree. In this case, implicit back refs is used for
850 * pointers in the block. Add full back refs for every pointers in the
851 * block, increase lower level extents' reference counts. The original
852 * implicit back refs are entailed to the new block.
854 * The reference count of the block is greater than one and the tree is
855 * not the block's owner tree. Add implicit back refs for every pointer in
856 * the new block, increase lower level extents' reference count.
858 * Back Reference Key composing:
860 * The key objectid corresponds to the first byte in the extent,
861 * The key type is used to differentiate between types of back refs.
862 * There are different meanings of the key offset for different types
865 * File extents can be referenced by:
867 * - multiple snapshots, subvolumes, or different generations in one subvol
868 * - different files inside a single subvolume
869 * - different offsets inside a file (bookend extents in file.c)
871 * The extent ref structure for the implicit back refs has fields for:
873 * - Objectid of the subvolume root
874 * - objectid of the file holding the reference
875 * - original offset in the file
876 * - how many bookend extents
878 * The key offset for the implicit back refs is hash of the first
881 * The extent ref structure for the full back refs has field for:
883 * - number of pointers in the tree leaf
885 * The key offset for the implicit back refs is the first byte of
888 * When a file extent is allocated, The implicit back refs is used.
889 * the fields are filled in:
891 * (root_key.objectid, inode objectid, offset in file, 1)
893 * When a file extent is removed file truncation, we find the
894 * corresponding implicit back refs and check the following fields:
896 * (btrfs_header_owner(leaf), inode objectid, offset in file)
898 * Btree extents can be referenced by:
900 * - Different subvolumes
902 * Both the implicit back refs and the full back refs for tree blocks
903 * only consist of key. The key offset for the implicit back refs is
904 * objectid of block's owner tree. The key offset for the full back refs
905 * is the first byte of parent block.
907 * When implicit back refs is used, information about the lowest key and
908 * level of the tree block are required. These information are stored in
909 * tree block info structure.
912 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
913 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root,
915 struct btrfs_path *path,
916 u64 owner, u32 extra_size)
918 struct btrfs_extent_item *item;
919 struct btrfs_extent_item_v0 *ei0;
920 struct btrfs_extent_ref_v0 *ref0;
921 struct btrfs_tree_block_info *bi;
922 struct extent_buffer *leaf;
923 struct btrfs_key key;
924 struct btrfs_key found_key;
925 u32 new_size = sizeof(*item);
929 leaf = path->nodes[0];
930 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
932 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
933 ei0 = btrfs_item_ptr(leaf, path->slots[0],
934 struct btrfs_extent_item_v0);
935 refs = btrfs_extent_refs_v0(leaf, ei0);
937 if (owner == (u64)-1) {
939 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
940 ret = btrfs_next_leaf(root, path);
944 leaf = path->nodes[0];
946 btrfs_item_key_to_cpu(leaf, &found_key,
948 BUG_ON(key.objectid != found_key.objectid);
949 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
953 ref0 = btrfs_item_ptr(leaf, path->slots[0],
954 struct btrfs_extent_ref_v0);
955 owner = btrfs_ref_objectid_v0(leaf, ref0);
959 btrfs_release_path(path);
961 if (owner < BTRFS_FIRST_FREE_OBJECTID)
962 new_size += sizeof(*bi);
964 new_size -= sizeof(*ei0);
965 ret = btrfs_search_slot(trans, root, &key, path,
966 new_size + extra_size, 1);
971 ret = btrfs_extend_item(trans, root, path, new_size);
973 leaf = path->nodes[0];
974 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
975 btrfs_set_extent_refs(leaf, item, refs);
976 /* FIXME: get real generation */
977 btrfs_set_extent_generation(leaf, item, 0);
978 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
979 btrfs_set_extent_flags(leaf, item,
980 BTRFS_EXTENT_FLAG_TREE_BLOCK |
981 BTRFS_BLOCK_FLAG_FULL_BACKREF);
982 bi = (struct btrfs_tree_block_info *)(item + 1);
983 /* FIXME: get first key of the block */
984 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
985 btrfs_set_tree_block_level(leaf, bi, (int)owner);
987 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
989 btrfs_mark_buffer_dirty(leaf);
994 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
996 u32 high_crc = ~(u32)0;
997 u32 low_crc = ~(u32)0;
1000 lenum = cpu_to_le64(root_objectid);
1001 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1002 lenum = cpu_to_le64(owner);
1003 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1004 lenum = cpu_to_le64(offset);
1005 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1007 return ((u64)high_crc << 31) ^ (u64)low_crc;
1010 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1011 struct btrfs_extent_data_ref *ref)
1013 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1014 btrfs_extent_data_ref_objectid(leaf, ref),
1015 btrfs_extent_data_ref_offset(leaf, ref));
1018 static int match_extent_data_ref(struct extent_buffer *leaf,
1019 struct btrfs_extent_data_ref *ref,
1020 u64 root_objectid, u64 owner, u64 offset)
1022 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1023 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1024 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1029 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1030 struct btrfs_root *root,
1031 struct btrfs_path *path,
1032 u64 bytenr, u64 parent,
1034 u64 owner, u64 offset)
1036 struct btrfs_key key;
1037 struct btrfs_extent_data_ref *ref;
1038 struct extent_buffer *leaf;
1044 key.objectid = bytenr;
1046 key.type = BTRFS_SHARED_DATA_REF_KEY;
1047 key.offset = parent;
1049 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1050 key.offset = hash_extent_data_ref(root_objectid,
1055 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1065 key.type = BTRFS_EXTENT_REF_V0_KEY;
1066 btrfs_release_path(path);
1067 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1078 leaf = path->nodes[0];
1079 nritems = btrfs_header_nritems(leaf);
1081 if (path->slots[0] >= nritems) {
1082 ret = btrfs_next_leaf(root, path);
1088 leaf = path->nodes[0];
1089 nritems = btrfs_header_nritems(leaf);
1093 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1094 if (key.objectid != bytenr ||
1095 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1098 ref = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_data_ref);
1101 if (match_extent_data_ref(leaf, ref, root_objectid,
1104 btrfs_release_path(path);
1116 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1117 struct btrfs_root *root,
1118 struct btrfs_path *path,
1119 u64 bytenr, u64 parent,
1120 u64 root_objectid, u64 owner,
1121 u64 offset, int refs_to_add)
1123 struct btrfs_key key;
1124 struct extent_buffer *leaf;
1129 key.objectid = bytenr;
1131 key.type = BTRFS_SHARED_DATA_REF_KEY;
1132 key.offset = parent;
1133 size = sizeof(struct btrfs_shared_data_ref);
1135 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1136 key.offset = hash_extent_data_ref(root_objectid,
1138 size = sizeof(struct btrfs_extent_data_ref);
1141 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1142 if (ret && ret != -EEXIST)
1145 leaf = path->nodes[0];
1147 struct btrfs_shared_data_ref *ref;
1148 ref = btrfs_item_ptr(leaf, path->slots[0],
1149 struct btrfs_shared_data_ref);
1151 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1153 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1154 num_refs += refs_to_add;
1155 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1158 struct btrfs_extent_data_ref *ref;
1159 while (ret == -EEXIST) {
1160 ref = btrfs_item_ptr(leaf, path->slots[0],
1161 struct btrfs_extent_data_ref);
1162 if (match_extent_data_ref(leaf, ref, root_objectid,
1165 btrfs_release_path(path);
1167 ret = btrfs_insert_empty_item(trans, root, path, &key,
1169 if (ret && ret != -EEXIST)
1172 leaf = path->nodes[0];
1174 ref = btrfs_item_ptr(leaf, path->slots[0],
1175 struct btrfs_extent_data_ref);
1177 btrfs_set_extent_data_ref_root(leaf, ref,
1179 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1180 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1181 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1183 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1184 num_refs += refs_to_add;
1185 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1188 btrfs_mark_buffer_dirty(leaf);
1191 btrfs_release_path(path);
1195 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1196 struct btrfs_root *root,
1197 struct btrfs_path *path,
1200 struct btrfs_key key;
1201 struct btrfs_extent_data_ref *ref1 = NULL;
1202 struct btrfs_shared_data_ref *ref2 = NULL;
1203 struct extent_buffer *leaf;
1207 leaf = path->nodes[0];
1208 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1210 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1211 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1212 struct btrfs_extent_data_ref);
1213 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1214 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1215 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1217 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1220 struct btrfs_extent_ref_v0 *ref0;
1221 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_extent_ref_v0);
1223 num_refs = btrfs_ref_count_v0(leaf, ref0);
1229 BUG_ON(num_refs < refs_to_drop);
1230 num_refs -= refs_to_drop;
1232 if (num_refs == 0) {
1233 ret = btrfs_del_item(trans, root, path);
1235 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1236 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1237 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1238 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1239 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1241 struct btrfs_extent_ref_v0 *ref0;
1242 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_extent_ref_v0);
1244 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1247 btrfs_mark_buffer_dirty(leaf);
1252 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1253 struct btrfs_path *path,
1254 struct btrfs_extent_inline_ref *iref)
1256 struct btrfs_key key;
1257 struct extent_buffer *leaf;
1258 struct btrfs_extent_data_ref *ref1;
1259 struct btrfs_shared_data_ref *ref2;
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1265 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1266 BTRFS_EXTENT_DATA_REF_KEY) {
1267 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1268 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1270 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1271 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1273 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1274 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1275 struct btrfs_extent_data_ref);
1276 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1277 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1278 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_shared_data_ref);
1280 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1283 struct btrfs_extent_ref_v0 *ref0;
1284 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_ref_v0);
1286 num_refs = btrfs_ref_count_v0(leaf, ref0);
1294 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1295 struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 u64 bytenr, u64 parent,
1300 struct btrfs_key key;
1303 key.objectid = bytenr;
1305 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1306 key.offset = parent;
1308 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1309 key.offset = root_objectid;
1312 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 if (ret == -ENOENT && parent) {
1317 btrfs_release_path(path);
1318 key.type = BTRFS_EXTENT_REF_V0_KEY;
1319 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1327 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root,
1329 struct btrfs_path *path,
1330 u64 bytenr, u64 parent,
1333 struct btrfs_key key;
1336 key.objectid = bytenr;
1338 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1339 key.offset = parent;
1341 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1342 key.offset = root_objectid;
1345 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1346 btrfs_release_path(path);
1350 static inline int extent_ref_type(u64 parent, u64 owner)
1353 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1355 type = BTRFS_SHARED_BLOCK_REF_KEY;
1357 type = BTRFS_TREE_BLOCK_REF_KEY;
1360 type = BTRFS_SHARED_DATA_REF_KEY;
1362 type = BTRFS_EXTENT_DATA_REF_KEY;
1367 static int find_next_key(struct btrfs_path *path, int level,
1368 struct btrfs_key *key)
1371 for (; level < BTRFS_MAX_LEVEL; level++) {
1372 if (!path->nodes[level])
1374 if (path->slots[level] + 1 >=
1375 btrfs_header_nritems(path->nodes[level]))
1378 btrfs_item_key_to_cpu(path->nodes[level], key,
1379 path->slots[level] + 1);
1381 btrfs_node_key_to_cpu(path->nodes[level], key,
1382 path->slots[level] + 1);
1389 * look for inline back ref. if back ref is found, *ref_ret is set
1390 * to the address of inline back ref, and 0 is returned.
1392 * if back ref isn't found, *ref_ret is set to the address where it
1393 * should be inserted, and -ENOENT is returned.
1395 * if insert is true and there are too many inline back refs, the path
1396 * points to the extent item, and -EAGAIN is returned.
1398 * NOTE: inline back refs are ordered in the same way that back ref
1399 * items in the tree are ordered.
1401 static noinline_for_stack
1402 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 struct btrfs_path *path,
1405 struct btrfs_extent_inline_ref **ref_ret,
1406 u64 bytenr, u64 num_bytes,
1407 u64 parent, u64 root_objectid,
1408 u64 owner, u64 offset, int insert)
1410 struct btrfs_key key;
1411 struct extent_buffer *leaf;
1412 struct btrfs_extent_item *ei;
1413 struct btrfs_extent_inline_ref *iref;
1424 key.objectid = bytenr;
1425 key.type = BTRFS_EXTENT_ITEM_KEY;
1426 key.offset = num_bytes;
1428 want = extent_ref_type(parent, owner);
1430 extra_size = btrfs_extent_inline_ref_size(want);
1431 path->keep_locks = 1;
1434 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1441 leaf = path->nodes[0];
1442 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1444 if (item_size < sizeof(*ei)) {
1449 ret = convert_extent_item_v0(trans, root, path, owner,
1455 leaf = path->nodes[0];
1456 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1459 BUG_ON(item_size < sizeof(*ei));
1461 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1462 flags = btrfs_extent_flags(leaf, ei);
1464 ptr = (unsigned long)(ei + 1);
1465 end = (unsigned long)ei + item_size;
1467 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1468 ptr += sizeof(struct btrfs_tree_block_info);
1471 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1480 iref = (struct btrfs_extent_inline_ref *)ptr;
1481 type = btrfs_extent_inline_ref_type(leaf, iref);
1485 ptr += btrfs_extent_inline_ref_size(type);
1489 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1490 struct btrfs_extent_data_ref *dref;
1491 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1492 if (match_extent_data_ref(leaf, dref, root_objectid,
1497 if (hash_extent_data_ref_item(leaf, dref) <
1498 hash_extent_data_ref(root_objectid, owner, offset))
1502 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1504 if (parent == ref_offset) {
1508 if (ref_offset < parent)
1511 if (root_objectid == ref_offset) {
1515 if (ref_offset < root_objectid)
1519 ptr += btrfs_extent_inline_ref_size(type);
1521 if (err == -ENOENT && insert) {
1522 if (item_size + extra_size >=
1523 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1528 * To add new inline back ref, we have to make sure
1529 * there is no corresponding back ref item.
1530 * For simplicity, we just do not add new inline back
1531 * ref if there is any kind of item for this block
1533 if (find_next_key(path, 0, &key) == 0 &&
1534 key.objectid == bytenr &&
1535 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1540 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1543 path->keep_locks = 0;
1544 btrfs_unlock_up_safe(path, 1);
1550 * helper to add new inline back ref
1552 static noinline_for_stack
1553 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1554 struct btrfs_root *root,
1555 struct btrfs_path *path,
1556 struct btrfs_extent_inline_ref *iref,
1557 u64 parent, u64 root_objectid,
1558 u64 owner, u64 offset, int refs_to_add,
1559 struct btrfs_delayed_extent_op *extent_op)
1561 struct extent_buffer *leaf;
1562 struct btrfs_extent_item *ei;
1565 unsigned long item_offset;
1571 leaf = path->nodes[0];
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 item_offset = (unsigned long)iref - (unsigned long)ei;
1575 type = extent_ref_type(parent, owner);
1576 size = btrfs_extent_inline_ref_size(type);
1578 ret = btrfs_extend_item(trans, root, path, size);
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 refs = btrfs_extent_refs(leaf, ei);
1582 refs += refs_to_add;
1583 btrfs_set_extent_refs(leaf, ei, refs);
1585 __run_delayed_extent_op(extent_op, leaf, ei);
1587 ptr = (unsigned long)ei + item_offset;
1588 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1589 if (ptr < end - size)
1590 memmove_extent_buffer(leaf, ptr + size, ptr,
1593 iref = (struct btrfs_extent_inline_ref *)ptr;
1594 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1595 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1596 struct btrfs_extent_data_ref *dref;
1597 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1598 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1599 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1600 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1601 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1602 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1603 struct btrfs_shared_data_ref *sref;
1604 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1605 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1606 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1607 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1608 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1610 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1612 btrfs_mark_buffer_dirty(leaf);
1616 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1617 struct btrfs_root *root,
1618 struct btrfs_path *path,
1619 struct btrfs_extent_inline_ref **ref_ret,
1620 u64 bytenr, u64 num_bytes, u64 parent,
1621 u64 root_objectid, u64 owner, u64 offset)
1625 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1626 bytenr, num_bytes, parent,
1627 root_objectid, owner, offset, 0);
1631 btrfs_release_path(path);
1634 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1635 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1638 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1639 root_objectid, owner, offset);
1645 * helper to update/remove inline back ref
1647 static noinline_for_stack
1648 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1649 struct btrfs_root *root,
1650 struct btrfs_path *path,
1651 struct btrfs_extent_inline_ref *iref,
1653 struct btrfs_delayed_extent_op *extent_op)
1655 struct extent_buffer *leaf;
1656 struct btrfs_extent_item *ei;
1657 struct btrfs_extent_data_ref *dref = NULL;
1658 struct btrfs_shared_data_ref *sref = NULL;
1667 leaf = path->nodes[0];
1668 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1669 refs = btrfs_extent_refs(leaf, ei);
1670 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1671 refs += refs_to_mod;
1672 btrfs_set_extent_refs(leaf, ei, refs);
1674 __run_delayed_extent_op(extent_op, leaf, ei);
1676 type = btrfs_extent_inline_ref_type(leaf, iref);
1678 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1679 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1680 refs = btrfs_extent_data_ref_count(leaf, dref);
1681 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1682 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1683 refs = btrfs_shared_data_ref_count(leaf, sref);
1686 BUG_ON(refs_to_mod != -1);
1689 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1690 refs += refs_to_mod;
1693 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1694 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1696 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1698 size = btrfs_extent_inline_ref_size(type);
1699 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1700 ptr = (unsigned long)iref;
1701 end = (unsigned long)ei + item_size;
1702 if (ptr + size < end)
1703 memmove_extent_buffer(leaf, ptr, ptr + size,
1706 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1708 btrfs_mark_buffer_dirty(leaf);
1712 static noinline_for_stack
1713 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1714 struct btrfs_root *root,
1715 struct btrfs_path *path,
1716 u64 bytenr, u64 num_bytes, u64 parent,
1717 u64 root_objectid, u64 owner,
1718 u64 offset, int refs_to_add,
1719 struct btrfs_delayed_extent_op *extent_op)
1721 struct btrfs_extent_inline_ref *iref;
1724 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1725 bytenr, num_bytes, parent,
1726 root_objectid, owner, offset, 1);
1728 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1729 ret = update_inline_extent_backref(trans, root, path, iref,
1730 refs_to_add, extent_op);
1731 } else if (ret == -ENOENT) {
1732 ret = setup_inline_extent_backref(trans, root, path, iref,
1733 parent, root_objectid,
1734 owner, offset, refs_to_add,
1740 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 u64 bytenr, u64 parent, u64 root_objectid,
1744 u64 owner, u64 offset, int refs_to_add)
1747 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1748 BUG_ON(refs_to_add != 1);
1749 ret = insert_tree_block_ref(trans, root, path, bytenr,
1750 parent, root_objectid);
1752 ret = insert_extent_data_ref(trans, root, path, bytenr,
1753 parent, root_objectid,
1754 owner, offset, refs_to_add);
1759 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1760 struct btrfs_root *root,
1761 struct btrfs_path *path,
1762 struct btrfs_extent_inline_ref *iref,
1763 int refs_to_drop, int is_data)
1767 BUG_ON(!is_data && refs_to_drop != 1);
1769 ret = update_inline_extent_backref(trans, root, path, iref,
1770 -refs_to_drop, NULL);
1771 } else if (is_data) {
1772 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1774 ret = btrfs_del_item(trans, root, path);
1779 static int btrfs_issue_discard(struct block_device *bdev,
1782 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1785 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1786 u64 num_bytes, u64 *actual_bytes)
1789 u64 discarded_bytes = 0;
1790 struct btrfs_multi_bio *multi = NULL;
1793 /* Tell the block device(s) that the sectors can be discarded */
1794 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1795 bytenr, &num_bytes, &multi, 0);
1797 struct btrfs_bio_stripe *stripe = multi->stripes;
1801 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1802 if (!stripe->dev->can_discard)
1805 ret = btrfs_issue_discard(stripe->dev->bdev,
1809 discarded_bytes += stripe->length;
1810 else if (ret != -EOPNOTSUPP)
1814 * Just in case we get back EOPNOTSUPP for some reason,
1815 * just ignore the return value so we don't screw up
1816 * people calling discard_extent.
1824 *actual_bytes = discarded_bytes;
1830 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1831 struct btrfs_root *root,
1832 u64 bytenr, u64 num_bytes, u64 parent,
1833 u64 root_objectid, u64 owner, u64 offset)
1836 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1837 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1839 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1840 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1841 parent, root_objectid, (int)owner,
1842 BTRFS_ADD_DELAYED_REF, NULL);
1844 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1845 parent, root_objectid, owner, offset,
1846 BTRFS_ADD_DELAYED_REF, NULL);
1851 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 u64 bytenr, u64 num_bytes,
1854 u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add,
1856 struct btrfs_delayed_extent_op *extent_op)
1858 struct btrfs_path *path;
1859 struct extent_buffer *leaf;
1860 struct btrfs_extent_item *item;
1865 path = btrfs_alloc_path();
1870 path->leave_spinning = 1;
1871 /* this will setup the path even if it fails to insert the back ref */
1872 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1873 path, bytenr, num_bytes, parent,
1874 root_objectid, owner, offset,
1875 refs_to_add, extent_op);
1879 if (ret != -EAGAIN) {
1884 leaf = path->nodes[0];
1885 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1886 refs = btrfs_extent_refs(leaf, item);
1887 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1889 __run_delayed_extent_op(extent_op, leaf, item);
1891 btrfs_mark_buffer_dirty(leaf);
1892 btrfs_release_path(path);
1895 path->leave_spinning = 1;
1897 /* now insert the actual backref */
1898 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1899 path, bytenr, parent, root_objectid,
1900 owner, offset, refs_to_add);
1903 btrfs_free_path(path);
1907 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root,
1909 struct btrfs_delayed_ref_node *node,
1910 struct btrfs_delayed_extent_op *extent_op,
1911 int insert_reserved)
1914 struct btrfs_delayed_data_ref *ref;
1915 struct btrfs_key ins;
1920 ins.objectid = node->bytenr;
1921 ins.offset = node->num_bytes;
1922 ins.type = BTRFS_EXTENT_ITEM_KEY;
1924 ref = btrfs_delayed_node_to_data_ref(node);
1925 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1926 parent = ref->parent;
1928 ref_root = ref->root;
1930 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1932 BUG_ON(extent_op->update_key);
1933 flags |= extent_op->flags_to_set;
1935 ret = alloc_reserved_file_extent(trans, root,
1936 parent, ref_root, flags,
1937 ref->objectid, ref->offset,
1938 &ins, node->ref_mod);
1939 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1940 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1941 node->num_bytes, parent,
1942 ref_root, ref->objectid,
1943 ref->offset, node->ref_mod,
1945 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1946 ret = __btrfs_free_extent(trans, root, node->bytenr,
1947 node->num_bytes, parent,
1948 ref_root, ref->objectid,
1949 ref->offset, node->ref_mod,
1957 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1958 struct extent_buffer *leaf,
1959 struct btrfs_extent_item *ei)
1961 u64 flags = btrfs_extent_flags(leaf, ei);
1962 if (extent_op->update_flags) {
1963 flags |= extent_op->flags_to_set;
1964 btrfs_set_extent_flags(leaf, ei, flags);
1967 if (extent_op->update_key) {
1968 struct btrfs_tree_block_info *bi;
1969 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1970 bi = (struct btrfs_tree_block_info *)(ei + 1);
1971 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1975 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1976 struct btrfs_root *root,
1977 struct btrfs_delayed_ref_node *node,
1978 struct btrfs_delayed_extent_op *extent_op)
1980 struct btrfs_key key;
1981 struct btrfs_path *path;
1982 struct btrfs_extent_item *ei;
1983 struct extent_buffer *leaf;
1988 path = btrfs_alloc_path();
1992 key.objectid = node->bytenr;
1993 key.type = BTRFS_EXTENT_ITEM_KEY;
1994 key.offset = node->num_bytes;
1997 path->leave_spinning = 1;
1998 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2009 leaf = path->nodes[0];
2010 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2011 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2012 if (item_size < sizeof(*ei)) {
2013 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2019 leaf = path->nodes[0];
2020 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2023 BUG_ON(item_size < sizeof(*ei));
2024 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2025 __run_delayed_extent_op(extent_op, leaf, ei);
2027 btrfs_mark_buffer_dirty(leaf);
2029 btrfs_free_path(path);
2033 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2034 struct btrfs_root *root,
2035 struct btrfs_delayed_ref_node *node,
2036 struct btrfs_delayed_extent_op *extent_op,
2037 int insert_reserved)
2040 struct btrfs_delayed_tree_ref *ref;
2041 struct btrfs_key ins;
2045 ins.objectid = node->bytenr;
2046 ins.offset = node->num_bytes;
2047 ins.type = BTRFS_EXTENT_ITEM_KEY;
2049 ref = btrfs_delayed_node_to_tree_ref(node);
2050 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2051 parent = ref->parent;
2053 ref_root = ref->root;
2055 BUG_ON(node->ref_mod != 1);
2056 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2057 BUG_ON(!extent_op || !extent_op->update_flags ||
2058 !extent_op->update_key);
2059 ret = alloc_reserved_tree_block(trans, root,
2061 extent_op->flags_to_set,
2064 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2065 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2066 node->num_bytes, parent, ref_root,
2067 ref->level, 0, 1, extent_op);
2068 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2069 ret = __btrfs_free_extent(trans, root, node->bytenr,
2070 node->num_bytes, parent, ref_root,
2071 ref->level, 0, 1, extent_op);
2078 /* helper function to actually process a single delayed ref entry */
2079 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2080 struct btrfs_root *root,
2081 struct btrfs_delayed_ref_node *node,
2082 struct btrfs_delayed_extent_op *extent_op,
2083 int insert_reserved)
2086 if (btrfs_delayed_ref_is_head(node)) {
2087 struct btrfs_delayed_ref_head *head;
2089 * we've hit the end of the chain and we were supposed
2090 * to insert this extent into the tree. But, it got
2091 * deleted before we ever needed to insert it, so all
2092 * we have to do is clean up the accounting
2095 head = btrfs_delayed_node_to_head(node);
2096 if (insert_reserved) {
2097 btrfs_pin_extent(root, node->bytenr,
2098 node->num_bytes, 1);
2099 if (head->is_data) {
2100 ret = btrfs_del_csums(trans, root,
2106 mutex_unlock(&head->mutex);
2110 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2111 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2112 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2114 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2115 node->type == BTRFS_SHARED_DATA_REF_KEY)
2116 ret = run_delayed_data_ref(trans, root, node, extent_op,
2123 static noinline struct btrfs_delayed_ref_node *
2124 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2126 struct rb_node *node;
2127 struct btrfs_delayed_ref_node *ref;
2128 int action = BTRFS_ADD_DELAYED_REF;
2131 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2132 * this prevents ref count from going down to zero when
2133 * there still are pending delayed ref.
2135 node = rb_prev(&head->node.rb_node);
2139 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2141 if (ref->bytenr != head->node.bytenr)
2143 if (ref->action == action)
2145 node = rb_prev(node);
2147 if (action == BTRFS_ADD_DELAYED_REF) {
2148 action = BTRFS_DROP_DELAYED_REF;
2154 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2155 struct btrfs_root *root,
2156 struct list_head *cluster)
2158 struct btrfs_delayed_ref_root *delayed_refs;
2159 struct btrfs_delayed_ref_node *ref;
2160 struct btrfs_delayed_ref_head *locked_ref = NULL;
2161 struct btrfs_delayed_extent_op *extent_op;
2164 int must_insert_reserved = 0;
2166 delayed_refs = &trans->transaction->delayed_refs;
2169 /* pick a new head ref from the cluster list */
2170 if (list_empty(cluster))
2173 locked_ref = list_entry(cluster->next,
2174 struct btrfs_delayed_ref_head, cluster);
2176 /* grab the lock that says we are going to process
2177 * all the refs for this head */
2178 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2181 * we may have dropped the spin lock to get the head
2182 * mutex lock, and that might have given someone else
2183 * time to free the head. If that's true, it has been
2184 * removed from our list and we can move on.
2186 if (ret == -EAGAIN) {
2194 * record the must insert reserved flag before we
2195 * drop the spin lock.
2197 must_insert_reserved = locked_ref->must_insert_reserved;
2198 locked_ref->must_insert_reserved = 0;
2200 extent_op = locked_ref->extent_op;
2201 locked_ref->extent_op = NULL;
2204 * locked_ref is the head node, so we have to go one
2205 * node back for any delayed ref updates
2207 ref = select_delayed_ref(locked_ref);
2209 /* All delayed refs have been processed, Go ahead
2210 * and send the head node to run_one_delayed_ref,
2211 * so that any accounting fixes can happen
2213 ref = &locked_ref->node;
2215 if (extent_op && must_insert_reserved) {
2221 spin_unlock(&delayed_refs->lock);
2223 ret = run_delayed_extent_op(trans, root,
2229 spin_lock(&delayed_refs->lock);
2233 list_del_init(&locked_ref->cluster);
2238 rb_erase(&ref->rb_node, &delayed_refs->root);
2239 delayed_refs->num_entries--;
2241 spin_unlock(&delayed_refs->lock);
2243 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2244 must_insert_reserved);
2247 btrfs_put_delayed_ref(ref);
2252 spin_lock(&delayed_refs->lock);
2258 * this starts processing the delayed reference count updates and
2259 * extent insertions we have queued up so far. count can be
2260 * 0, which means to process everything in the tree at the start
2261 * of the run (but not newly added entries), or it can be some target
2262 * number you'd like to process.
2264 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2265 struct btrfs_root *root, unsigned long count)
2267 struct rb_node *node;
2268 struct btrfs_delayed_ref_root *delayed_refs;
2269 struct btrfs_delayed_ref_node *ref;
2270 struct list_head cluster;
2272 int run_all = count == (unsigned long)-1;
2275 if (root == root->fs_info->extent_root)
2276 root = root->fs_info->tree_root;
2278 delayed_refs = &trans->transaction->delayed_refs;
2279 INIT_LIST_HEAD(&cluster);
2281 spin_lock(&delayed_refs->lock);
2283 count = delayed_refs->num_entries * 2;
2287 if (!(run_all || run_most) &&
2288 delayed_refs->num_heads_ready < 64)
2292 * go find something we can process in the rbtree. We start at
2293 * the beginning of the tree, and then build a cluster
2294 * of refs to process starting at the first one we are able to
2297 ret = btrfs_find_ref_cluster(trans, &cluster,
2298 delayed_refs->run_delayed_start);
2302 ret = run_clustered_refs(trans, root, &cluster);
2305 count -= min_t(unsigned long, ret, count);
2312 node = rb_first(&delayed_refs->root);
2315 count = (unsigned long)-1;
2318 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2320 if (btrfs_delayed_ref_is_head(ref)) {
2321 struct btrfs_delayed_ref_head *head;
2323 head = btrfs_delayed_node_to_head(ref);
2324 atomic_inc(&ref->refs);
2326 spin_unlock(&delayed_refs->lock);
2328 * Mutex was contended, block until it's
2329 * released and try again
2331 mutex_lock(&head->mutex);
2332 mutex_unlock(&head->mutex);
2334 btrfs_put_delayed_ref(ref);
2338 node = rb_next(node);
2340 spin_unlock(&delayed_refs->lock);
2341 schedule_timeout(1);
2345 spin_unlock(&delayed_refs->lock);
2349 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2350 struct btrfs_root *root,
2351 u64 bytenr, u64 num_bytes, u64 flags,
2354 struct btrfs_delayed_extent_op *extent_op;
2357 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2361 extent_op->flags_to_set = flags;
2362 extent_op->update_flags = 1;
2363 extent_op->update_key = 0;
2364 extent_op->is_data = is_data ? 1 : 0;
2366 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2372 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2373 struct btrfs_root *root,
2374 struct btrfs_path *path,
2375 u64 objectid, u64 offset, u64 bytenr)
2377 struct btrfs_delayed_ref_head *head;
2378 struct btrfs_delayed_ref_node *ref;
2379 struct btrfs_delayed_data_ref *data_ref;
2380 struct btrfs_delayed_ref_root *delayed_refs;
2381 struct rb_node *node;
2385 delayed_refs = &trans->transaction->delayed_refs;
2386 spin_lock(&delayed_refs->lock);
2387 head = btrfs_find_delayed_ref_head(trans, bytenr);
2391 if (!mutex_trylock(&head->mutex)) {
2392 atomic_inc(&head->node.refs);
2393 spin_unlock(&delayed_refs->lock);
2395 btrfs_release_path(path);
2398 * Mutex was contended, block until it's released and let
2401 mutex_lock(&head->mutex);
2402 mutex_unlock(&head->mutex);
2403 btrfs_put_delayed_ref(&head->node);
2407 node = rb_prev(&head->node.rb_node);
2411 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2413 if (ref->bytenr != bytenr)
2417 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2420 data_ref = btrfs_delayed_node_to_data_ref(ref);
2422 node = rb_prev(node);
2424 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2425 if (ref->bytenr == bytenr)
2429 if (data_ref->root != root->root_key.objectid ||
2430 data_ref->objectid != objectid || data_ref->offset != offset)
2435 mutex_unlock(&head->mutex);
2437 spin_unlock(&delayed_refs->lock);
2441 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2442 struct btrfs_root *root,
2443 struct btrfs_path *path,
2444 u64 objectid, u64 offset, u64 bytenr)
2446 struct btrfs_root *extent_root = root->fs_info->extent_root;
2447 struct extent_buffer *leaf;
2448 struct btrfs_extent_data_ref *ref;
2449 struct btrfs_extent_inline_ref *iref;
2450 struct btrfs_extent_item *ei;
2451 struct btrfs_key key;
2455 key.objectid = bytenr;
2456 key.offset = (u64)-1;
2457 key.type = BTRFS_EXTENT_ITEM_KEY;
2459 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2465 if (path->slots[0] == 0)
2469 leaf = path->nodes[0];
2470 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2472 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2476 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2478 if (item_size < sizeof(*ei)) {
2479 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2483 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2485 if (item_size != sizeof(*ei) +
2486 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2489 if (btrfs_extent_generation(leaf, ei) <=
2490 btrfs_root_last_snapshot(&root->root_item))
2493 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2494 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2495 BTRFS_EXTENT_DATA_REF_KEY)
2498 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2499 if (btrfs_extent_refs(leaf, ei) !=
2500 btrfs_extent_data_ref_count(leaf, ref) ||
2501 btrfs_extent_data_ref_root(leaf, ref) !=
2502 root->root_key.objectid ||
2503 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2504 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2512 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2513 struct btrfs_root *root,
2514 u64 objectid, u64 offset, u64 bytenr)
2516 struct btrfs_path *path;
2520 path = btrfs_alloc_path();
2525 ret = check_committed_ref(trans, root, path, objectid,
2527 if (ret && ret != -ENOENT)
2530 ret2 = check_delayed_ref(trans, root, path, objectid,
2532 } while (ret2 == -EAGAIN);
2534 if (ret2 && ret2 != -ENOENT) {
2539 if (ret != -ENOENT || ret2 != -ENOENT)
2542 btrfs_free_path(path);
2543 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2548 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2549 struct btrfs_root *root,
2550 struct extent_buffer *buf,
2551 int full_backref, int inc)
2558 struct btrfs_key key;
2559 struct btrfs_file_extent_item *fi;
2563 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2564 u64, u64, u64, u64, u64, u64);
2566 ref_root = btrfs_header_owner(buf);
2567 nritems = btrfs_header_nritems(buf);
2568 level = btrfs_header_level(buf);
2570 if (!root->ref_cows && level == 0)
2574 process_func = btrfs_inc_extent_ref;
2576 process_func = btrfs_free_extent;
2579 parent = buf->start;
2583 for (i = 0; i < nritems; i++) {
2585 btrfs_item_key_to_cpu(buf, &key, i);
2586 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2588 fi = btrfs_item_ptr(buf, i,
2589 struct btrfs_file_extent_item);
2590 if (btrfs_file_extent_type(buf, fi) ==
2591 BTRFS_FILE_EXTENT_INLINE)
2593 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2597 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2598 key.offset -= btrfs_file_extent_offset(buf, fi);
2599 ret = process_func(trans, root, bytenr, num_bytes,
2600 parent, ref_root, key.objectid,
2605 bytenr = btrfs_node_blockptr(buf, i);
2606 num_bytes = btrfs_level_size(root, level - 1);
2607 ret = process_func(trans, root, bytenr, num_bytes,
2608 parent, ref_root, level - 1, 0);
2619 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2620 struct extent_buffer *buf, int full_backref)
2622 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2625 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2626 struct extent_buffer *buf, int full_backref)
2628 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2631 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root,
2633 struct btrfs_path *path,
2634 struct btrfs_block_group_cache *cache)
2637 struct btrfs_root *extent_root = root->fs_info->extent_root;
2639 struct extent_buffer *leaf;
2641 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2646 leaf = path->nodes[0];
2647 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2648 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2649 btrfs_mark_buffer_dirty(leaf);
2650 btrfs_release_path(path);
2658 static struct btrfs_block_group_cache *
2659 next_block_group(struct btrfs_root *root,
2660 struct btrfs_block_group_cache *cache)
2662 struct rb_node *node;
2663 spin_lock(&root->fs_info->block_group_cache_lock);
2664 node = rb_next(&cache->cache_node);
2665 btrfs_put_block_group(cache);
2667 cache = rb_entry(node, struct btrfs_block_group_cache,
2669 btrfs_get_block_group(cache);
2672 spin_unlock(&root->fs_info->block_group_cache_lock);
2676 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2677 struct btrfs_trans_handle *trans,
2678 struct btrfs_path *path)
2680 struct btrfs_root *root = block_group->fs_info->tree_root;
2681 struct inode *inode = NULL;
2683 int dcs = BTRFS_DC_ERROR;
2689 * If this block group is smaller than 100 megs don't bother caching the
2692 if (block_group->key.offset < (100 * 1024 * 1024)) {
2693 spin_lock(&block_group->lock);
2694 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2695 spin_unlock(&block_group->lock);
2700 inode = lookup_free_space_inode(root, block_group, path);
2701 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2702 ret = PTR_ERR(inode);
2703 btrfs_release_path(path);
2707 if (IS_ERR(inode)) {
2711 if (block_group->ro)
2714 ret = create_free_space_inode(root, trans, block_group, path);
2720 /* We've already setup this transaction, go ahead and exit */
2721 if (block_group->cache_generation == trans->transid &&
2722 i_size_read(inode)) {
2723 dcs = BTRFS_DC_SETUP;
2728 * We want to set the generation to 0, that way if anything goes wrong
2729 * from here on out we know not to trust this cache when we load up next
2732 BTRFS_I(inode)->generation = 0;
2733 ret = btrfs_update_inode(trans, root, inode);
2736 if (i_size_read(inode) > 0) {
2737 ret = btrfs_truncate_free_space_cache(root, trans, path,
2743 spin_lock(&block_group->lock);
2744 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2745 /* We're not cached, don't bother trying to write stuff out */
2746 dcs = BTRFS_DC_WRITTEN;
2747 spin_unlock(&block_group->lock);
2750 spin_unlock(&block_group->lock);
2752 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2757 * Just to make absolutely sure we have enough space, we're going to
2758 * preallocate 12 pages worth of space for each block group. In
2759 * practice we ought to use at most 8, but we need extra space so we can
2760 * add our header and have a terminator between the extents and the
2764 num_pages *= PAGE_CACHE_SIZE;
2766 ret = btrfs_check_data_free_space(inode, num_pages);
2770 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2771 num_pages, num_pages,
2774 dcs = BTRFS_DC_SETUP;
2775 btrfs_free_reserved_data_space(inode, num_pages);
2780 btrfs_release_path(path);
2782 spin_lock(&block_group->lock);
2784 block_group->cache_generation = trans->transid;
2785 block_group->disk_cache_state = dcs;
2786 spin_unlock(&block_group->lock);
2791 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2792 struct btrfs_root *root)
2794 struct btrfs_block_group_cache *cache;
2796 struct btrfs_path *path;
2799 path = btrfs_alloc_path();
2805 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2807 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2809 cache = next_block_group(root, cache);
2817 err = cache_save_setup(cache, trans, path);
2818 last = cache->key.objectid + cache->key.offset;
2819 btrfs_put_block_group(cache);
2824 err = btrfs_run_delayed_refs(trans, root,
2829 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2831 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2832 btrfs_put_block_group(cache);
2838 cache = next_block_group(root, cache);
2847 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2848 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2850 last = cache->key.objectid + cache->key.offset;
2852 err = write_one_cache_group(trans, root, path, cache);
2854 btrfs_put_block_group(cache);
2859 * I don't think this is needed since we're just marking our
2860 * preallocated extent as written, but just in case it can't
2864 err = btrfs_run_delayed_refs(trans, root,
2869 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2872 * Really this shouldn't happen, but it could if we
2873 * couldn't write the entire preallocated extent and
2874 * splitting the extent resulted in a new block.
2877 btrfs_put_block_group(cache);
2880 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2882 cache = next_block_group(root, cache);
2891 btrfs_write_out_cache(root, trans, cache, path);
2894 * If we didn't have an error then the cache state is still
2895 * NEED_WRITE, so we can set it to WRITTEN.
2897 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2898 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2899 last = cache->key.objectid + cache->key.offset;
2900 btrfs_put_block_group(cache);
2903 btrfs_free_path(path);
2907 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2909 struct btrfs_block_group_cache *block_group;
2912 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2913 if (!block_group || block_group->ro)
2916 btrfs_put_block_group(block_group);
2920 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2921 u64 total_bytes, u64 bytes_used,
2922 struct btrfs_space_info **space_info)
2924 struct btrfs_space_info *found;
2928 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2929 BTRFS_BLOCK_GROUP_RAID10))
2934 found = __find_space_info(info, flags);
2936 spin_lock(&found->lock);
2937 found->total_bytes += total_bytes;
2938 found->disk_total += total_bytes * factor;
2939 found->bytes_used += bytes_used;
2940 found->disk_used += bytes_used * factor;
2942 spin_unlock(&found->lock);
2943 *space_info = found;
2946 found = kzalloc(sizeof(*found), GFP_NOFS);
2950 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2951 INIT_LIST_HEAD(&found->block_groups[i]);
2952 init_rwsem(&found->groups_sem);
2953 spin_lock_init(&found->lock);
2954 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2955 BTRFS_BLOCK_GROUP_SYSTEM |
2956 BTRFS_BLOCK_GROUP_METADATA);
2957 found->total_bytes = total_bytes;
2958 found->disk_total = total_bytes * factor;
2959 found->bytes_used = bytes_used;
2960 found->disk_used = bytes_used * factor;
2961 found->bytes_pinned = 0;
2962 found->bytes_reserved = 0;
2963 found->bytes_readonly = 0;
2964 found->bytes_may_use = 0;
2966 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2967 found->chunk_alloc = 0;
2969 init_waitqueue_head(&found->wait);
2970 *space_info = found;
2971 list_add_rcu(&found->list, &info->space_info);
2975 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2977 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2978 BTRFS_BLOCK_GROUP_RAID1 |
2979 BTRFS_BLOCK_GROUP_RAID10 |
2980 BTRFS_BLOCK_GROUP_DUP);
2982 if (flags & BTRFS_BLOCK_GROUP_DATA)
2983 fs_info->avail_data_alloc_bits |= extra_flags;
2984 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2985 fs_info->avail_metadata_alloc_bits |= extra_flags;
2986 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2987 fs_info->avail_system_alloc_bits |= extra_flags;
2991 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2994 * we add in the count of missing devices because we want
2995 * to make sure that any RAID levels on a degraded FS
2996 * continue to be honored.
2998 u64 num_devices = root->fs_info->fs_devices->rw_devices +
2999 root->fs_info->fs_devices->missing_devices;
3001 if (num_devices == 1)
3002 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3003 if (num_devices < 4)
3004 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3006 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3007 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3008 BTRFS_BLOCK_GROUP_RAID10))) {
3009 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3012 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3013 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3014 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3017 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3018 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3019 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3020 (flags & BTRFS_BLOCK_GROUP_DUP)))
3021 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3025 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3027 if (flags & BTRFS_BLOCK_GROUP_DATA)
3028 flags |= root->fs_info->avail_data_alloc_bits &
3029 root->fs_info->data_alloc_profile;
3030 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3031 flags |= root->fs_info->avail_system_alloc_bits &
3032 root->fs_info->system_alloc_profile;
3033 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3034 flags |= root->fs_info->avail_metadata_alloc_bits &
3035 root->fs_info->metadata_alloc_profile;
3036 return btrfs_reduce_alloc_profile(root, flags);
3039 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3044 flags = BTRFS_BLOCK_GROUP_DATA;
3045 else if (root == root->fs_info->chunk_root)
3046 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3048 flags = BTRFS_BLOCK_GROUP_METADATA;
3050 return get_alloc_profile(root, flags);
3053 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3055 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3056 BTRFS_BLOCK_GROUP_DATA);
3060 * This will check the space that the inode allocates from to make sure we have
3061 * enough space for bytes.
3063 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3065 struct btrfs_space_info *data_sinfo;
3066 struct btrfs_root *root = BTRFS_I(inode)->root;
3068 int ret = 0, committed = 0, alloc_chunk = 1;
3070 /* make sure bytes are sectorsize aligned */
3071 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3073 if (root == root->fs_info->tree_root ||
3074 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3079 data_sinfo = BTRFS_I(inode)->space_info;
3084 /* make sure we have enough space to handle the data first */
3085 spin_lock(&data_sinfo->lock);
3086 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3087 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3088 data_sinfo->bytes_may_use;
3090 if (used + bytes > data_sinfo->total_bytes) {
3091 struct btrfs_trans_handle *trans;
3094 * if we don't have enough free bytes in this space then we need
3095 * to alloc a new chunk.
3097 if (!data_sinfo->full && alloc_chunk) {
3100 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3101 spin_unlock(&data_sinfo->lock);
3103 alloc_target = btrfs_get_alloc_profile(root, 1);
3104 trans = btrfs_join_transaction(root);
3106 return PTR_ERR(trans);
3108 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3109 bytes + 2 * 1024 * 1024,
3111 CHUNK_ALLOC_NO_FORCE);
3112 btrfs_end_transaction(trans, root);
3121 btrfs_set_inode_space_info(root, inode);
3122 data_sinfo = BTRFS_I(inode)->space_info;
3128 * If we have less pinned bytes than we want to allocate then
3129 * don't bother committing the transaction, it won't help us.
3131 if (data_sinfo->bytes_pinned < bytes)
3133 spin_unlock(&data_sinfo->lock);
3135 /* commit the current transaction and try again */
3138 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3140 trans = btrfs_join_transaction(root);
3142 return PTR_ERR(trans);
3143 ret = btrfs_commit_transaction(trans, root);
3151 data_sinfo->bytes_may_use += bytes;
3152 spin_unlock(&data_sinfo->lock);
3158 * Called if we need to clear a data reservation for this inode.
3160 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3162 struct btrfs_root *root = BTRFS_I(inode)->root;
3163 struct btrfs_space_info *data_sinfo;
3165 /* make sure bytes are sectorsize aligned */
3166 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3168 data_sinfo = BTRFS_I(inode)->space_info;
3169 spin_lock(&data_sinfo->lock);
3170 data_sinfo->bytes_may_use -= bytes;
3171 spin_unlock(&data_sinfo->lock);
3174 static void force_metadata_allocation(struct btrfs_fs_info *info)
3176 struct list_head *head = &info->space_info;
3177 struct btrfs_space_info *found;
3180 list_for_each_entry_rcu(found, head, list) {
3181 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3182 found->force_alloc = CHUNK_ALLOC_FORCE;
3187 static int should_alloc_chunk(struct btrfs_root *root,
3188 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3191 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3192 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3193 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3196 if (force == CHUNK_ALLOC_FORCE)
3200 * We need to take into account the global rsv because for all intents
3201 * and purposes it's used space. Don't worry about locking the
3202 * global_rsv, it doesn't change except when the transaction commits.
3204 num_allocated += global_rsv->size;
3207 * in limited mode, we want to have some free space up to
3208 * about 1% of the FS size.
3210 if (force == CHUNK_ALLOC_LIMITED) {
3211 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3212 thresh = max_t(u64, 64 * 1024 * 1024,
3213 div_factor_fine(thresh, 1));
3215 if (num_bytes - num_allocated < thresh)
3220 * we have two similar checks here, one based on percentage
3221 * and once based on a hard number of 256MB. The idea
3222 * is that if we have a good amount of free
3223 * room, don't allocate a chunk. A good mount is
3224 * less than 80% utilized of the chunks we have allocated,
3225 * or more than 256MB free
3227 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3230 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3233 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3235 /* 256MB or 5% of the FS */
3236 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3238 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3243 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *extent_root, u64 alloc_bytes,
3245 u64 flags, int force)
3247 struct btrfs_space_info *space_info;
3248 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3249 int wait_for_alloc = 0;
3252 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3254 space_info = __find_space_info(extent_root->fs_info, flags);
3256 ret = update_space_info(extent_root->fs_info, flags,
3260 BUG_ON(!space_info);
3263 spin_lock(&space_info->lock);
3264 if (space_info->force_alloc)
3265 force = space_info->force_alloc;
3266 if (space_info->full) {
3267 spin_unlock(&space_info->lock);
3271 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3272 spin_unlock(&space_info->lock);
3274 } else if (space_info->chunk_alloc) {
3277 space_info->chunk_alloc = 1;
3280 spin_unlock(&space_info->lock);
3282 mutex_lock(&fs_info->chunk_mutex);
3285 * The chunk_mutex is held throughout the entirety of a chunk
3286 * allocation, so once we've acquired the chunk_mutex we know that the
3287 * other guy is done and we need to recheck and see if we should
3290 if (wait_for_alloc) {
3291 mutex_unlock(&fs_info->chunk_mutex);
3297 * If we have mixed data/metadata chunks we want to make sure we keep
3298 * allocating mixed chunks instead of individual chunks.
3300 if (btrfs_mixed_space_info(space_info))
3301 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3304 * if we're doing a data chunk, go ahead and make sure that
3305 * we keep a reasonable number of metadata chunks allocated in the
3308 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3309 fs_info->data_chunk_allocations++;
3310 if (!(fs_info->data_chunk_allocations %
3311 fs_info->metadata_ratio))
3312 force_metadata_allocation(fs_info);
3315 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3316 if (ret < 0 && ret != -ENOSPC)
3319 spin_lock(&space_info->lock);
3321 space_info->full = 1;
3325 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3326 space_info->chunk_alloc = 0;
3327 spin_unlock(&space_info->lock);
3329 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3334 * shrink metadata reservation for delalloc
3336 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3337 struct btrfs_root *root, u64 to_reclaim,
3340 struct btrfs_block_rsv *block_rsv;
3341 struct btrfs_space_info *space_info;
3346 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3348 unsigned long progress;
3350 block_rsv = &root->fs_info->delalloc_block_rsv;
3351 space_info = block_rsv->space_info;
3354 reserved = space_info->bytes_may_use;
3355 progress = space_info->reservation_progress;
3361 if (root->fs_info->delalloc_bytes == 0) {
3364 btrfs_wait_ordered_extents(root, 0, 0);
3368 max_reclaim = min(reserved, to_reclaim);
3370 while (loops < 1024) {
3371 /* have the flusher threads jump in and do some IO */
3373 nr_pages = min_t(unsigned long, nr_pages,
3374 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3375 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3377 spin_lock(&space_info->lock);
3378 if (reserved > space_info->bytes_may_use)
3379 reclaimed += reserved - space_info->bytes_may_use;
3380 reserved = space_info->bytes_may_use;
3381 spin_unlock(&space_info->lock);
3385 if (reserved == 0 || reclaimed >= max_reclaim)
3388 if (trans && trans->transaction->blocked)
3391 if (wait_ordered && !trans) {
3392 btrfs_wait_ordered_extents(root, 0, 0);
3394 time_left = schedule_timeout_interruptible(1);
3396 /* We were interrupted, exit */
3401 /* we've kicked the IO a few times, if anything has been freed,
3402 * exit. There is no sense in looping here for a long time
3403 * when we really need to commit the transaction, or there are
3404 * just too many writers without enough free space
3409 if (progress != space_info->reservation_progress)
3415 return reclaimed >= to_reclaim;
3419 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3420 * @root - the root we're allocating for
3421 * @block_rsv - the block_rsv we're allocating for
3422 * @orig_bytes - the number of bytes we want
3423 * @flush - wether or not we can flush to make our reservation
3424 * @check - wether this is just to check if we have enough space or not
3426 * This will reserve orgi_bytes number of bytes from the space info associated
3427 * with the block_rsv. If there is not enough space it will make an attempt to
3428 * flush out space to make room. It will do this by flushing delalloc if
3429 * possible or committing the transaction. If flush is 0 then no attempts to
3430 * regain reservations will be made and this will fail if there is not enough
3433 static int reserve_metadata_bytes(struct btrfs_root *root,
3434 struct btrfs_block_rsv *block_rsv,
3435 u64 orig_bytes, int flush, int check)
3437 struct btrfs_space_info *space_info = block_rsv->space_info;
3438 struct btrfs_trans_handle *trans;
3440 u64 num_bytes = orig_bytes;
3443 bool committed = false;
3444 bool flushing = false;
3445 bool wait_ordered = false;
3447 trans = (struct btrfs_trans_handle *)current->journal_info;
3450 spin_lock(&space_info->lock);
3452 * We only want to wait if somebody other than us is flushing and we are
3453 * actually alloed to flush.
3455 while (flush && !flushing && space_info->flush) {
3456 spin_unlock(&space_info->lock);
3458 * If we have a trans handle we can't wait because the flusher
3459 * may have to commit the transaction, which would mean we would
3460 * deadlock since we are waiting for the flusher to finish, but
3461 * hold the current transaction open.
3465 ret = wait_event_interruptible(space_info->wait,
3466 !space_info->flush);
3467 /* Must have been interrupted, return */
3471 spin_lock(&space_info->lock);
3475 used = space_info->bytes_used + space_info->bytes_reserved +
3476 space_info->bytes_pinned + space_info->bytes_readonly +
3477 space_info->bytes_may_use;
3480 * The idea here is that we've not already over-reserved the block group
3481 * then we can go ahead and save our reservation first and then start
3482 * flushing if we need to. Otherwise if we've already overcommitted
3483 * lets start flushing stuff first and then come back and try to make
3486 if (used <= space_info->total_bytes) {
3487 if (used + orig_bytes <= space_info->total_bytes) {
3488 space_info->bytes_may_use += orig_bytes;
3492 * Ok set num_bytes to orig_bytes since we aren't
3493 * overocmmitted, this way we only try and reclaim what
3496 num_bytes = orig_bytes;
3500 * Ok we're over committed, set num_bytes to the overcommitted
3501 * amount plus the amount of bytes that we need for this
3504 wait_ordered = true;
3505 num_bytes = used - space_info->total_bytes +
3506 (orig_bytes * (retries + 1));
3509 if (ret && !check) {
3510 u64 profile = btrfs_get_alloc_profile(root, 0);
3513 spin_lock(&root->fs_info->free_chunk_lock);
3514 avail = root->fs_info->free_chunk_space;
3517 * If we have dup, raid1 or raid10 then only half of the free
3518 * space is actually useable.
3520 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3521 BTRFS_BLOCK_GROUP_RAID1 |
3522 BTRFS_BLOCK_GROUP_RAID10))
3526 * If we aren't flushing don't let us overcommit too much, say
3527 * 1/8th of the space. If we can flush, let it overcommit up to
3534 spin_unlock(&root->fs_info->free_chunk_lock);
3536 if (used + num_bytes < space_info->total_bytes + avail) {
3537 space_info->bytes_may_use += orig_bytes;
3540 wait_ordered = true;
3545 * Couldn't make our reservation, save our place so while we're trying
3546 * to reclaim space we can actually use it instead of somebody else
3547 * stealing it from us.
3551 space_info->flush = 1;
3554 spin_unlock(&space_info->lock);
3560 * We do synchronous shrinking since we don't actually unreserve
3561 * metadata until after the IO is completed.
3563 ret = shrink_delalloc(trans, root, num_bytes, wait_ordered);
3570 * So if we were overcommitted it's possible that somebody else flushed
3571 * out enough space and we simply didn't have enough space to reclaim,
3572 * so go back around and try again.
3575 wait_ordered = true;
3588 trans = btrfs_join_transaction(root);
3591 ret = btrfs_commit_transaction(trans, root);
3600 spin_lock(&space_info->lock);
3601 space_info->flush = 0;
3602 wake_up_all(&space_info->wait);
3603 spin_unlock(&space_info->lock);
3608 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3609 struct btrfs_root *root)
3611 struct btrfs_block_rsv *block_rsv = NULL;
3613 if (root->ref_cows || root == root->fs_info->csum_root)
3614 block_rsv = trans->block_rsv;
3617 block_rsv = root->block_rsv;
3620 block_rsv = &root->fs_info->empty_block_rsv;
3625 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3629 spin_lock(&block_rsv->lock);
3630 if (block_rsv->reserved >= num_bytes) {
3631 block_rsv->reserved -= num_bytes;
3632 if (block_rsv->reserved < block_rsv->size)
3633 block_rsv->full = 0;
3636 spin_unlock(&block_rsv->lock);
3640 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3641 u64 num_bytes, int update_size)
3643 spin_lock(&block_rsv->lock);
3644 block_rsv->reserved += num_bytes;
3646 block_rsv->size += num_bytes;
3647 else if (block_rsv->reserved >= block_rsv->size)
3648 block_rsv->full = 1;
3649 spin_unlock(&block_rsv->lock);
3652 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3653 struct btrfs_block_rsv *dest, u64 num_bytes)
3655 struct btrfs_space_info *space_info = block_rsv->space_info;
3657 spin_lock(&block_rsv->lock);
3658 if (num_bytes == (u64)-1)
3659 num_bytes = block_rsv->size;
3660 block_rsv->size -= num_bytes;
3661 if (block_rsv->reserved >= block_rsv->size) {
3662 num_bytes = block_rsv->reserved - block_rsv->size;
3663 block_rsv->reserved = block_rsv->size;
3664 block_rsv->full = 1;
3668 spin_unlock(&block_rsv->lock);
3670 if (num_bytes > 0) {
3672 spin_lock(&dest->lock);
3676 bytes_to_add = dest->size - dest->reserved;
3677 bytes_to_add = min(num_bytes, bytes_to_add);
3678 dest->reserved += bytes_to_add;
3679 if (dest->reserved >= dest->size)
3681 num_bytes -= bytes_to_add;
3683 spin_unlock(&dest->lock);
3686 spin_lock(&space_info->lock);
3687 space_info->bytes_may_use -= num_bytes;
3688 space_info->reservation_progress++;
3689 spin_unlock(&space_info->lock);
3694 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3695 struct btrfs_block_rsv *dst, u64 num_bytes)
3699 ret = block_rsv_use_bytes(src, num_bytes);
3703 block_rsv_add_bytes(dst, num_bytes, 1);
3707 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3709 memset(rsv, 0, sizeof(*rsv));
3710 spin_lock_init(&rsv->lock);
3713 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3715 struct btrfs_block_rsv *block_rsv;
3716 struct btrfs_fs_info *fs_info = root->fs_info;
3718 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3722 btrfs_init_block_rsv(block_rsv);
3723 block_rsv->space_info = __find_space_info(fs_info,
3724 BTRFS_BLOCK_GROUP_METADATA);
3728 void btrfs_free_block_rsv(struct btrfs_root *root,
3729 struct btrfs_block_rsv *rsv)
3731 btrfs_block_rsv_release(root, rsv, (u64)-1);
3735 int btrfs_block_rsv_add(struct btrfs_root *root,
3736 struct btrfs_block_rsv *block_rsv,
3744 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1, 0);
3746 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3753 int btrfs_block_rsv_check(struct btrfs_root *root,
3754 struct btrfs_block_rsv *block_rsv,
3755 u64 min_reserved, int min_factor, int flush)
3763 spin_lock(&block_rsv->lock);
3765 num_bytes = div_factor(block_rsv->size, min_factor);
3766 if (min_reserved > num_bytes)
3767 num_bytes = min_reserved;
3769 if (block_rsv->reserved >= num_bytes)
3772 num_bytes -= block_rsv->reserved;
3773 spin_unlock(&block_rsv->lock);
3778 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush, !flush);
3780 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3787 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3788 struct btrfs_block_rsv *dst_rsv,
3791 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3794 void btrfs_block_rsv_release(struct btrfs_root *root,
3795 struct btrfs_block_rsv *block_rsv,
3798 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3799 if (global_rsv->full || global_rsv == block_rsv ||
3800 block_rsv->space_info != global_rsv->space_info)
3802 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3806 * helper to calculate size of global block reservation.
3807 * the desired value is sum of space used by extent tree,
3808 * checksum tree and root tree
3810 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3812 struct btrfs_space_info *sinfo;
3816 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3818 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3819 spin_lock(&sinfo->lock);
3820 data_used = sinfo->bytes_used;
3821 spin_unlock(&sinfo->lock);
3823 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3824 spin_lock(&sinfo->lock);
3825 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3827 meta_used = sinfo->bytes_used;
3828 spin_unlock(&sinfo->lock);
3830 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3832 num_bytes += div64_u64(data_used + meta_used, 50);
3834 if (num_bytes * 3 > meta_used)
3835 num_bytes = div64_u64(meta_used, 3);
3837 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3840 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3842 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3843 struct btrfs_space_info *sinfo = block_rsv->space_info;
3846 num_bytes = calc_global_metadata_size(fs_info);
3848 spin_lock(&block_rsv->lock);
3849 spin_lock(&sinfo->lock);
3851 block_rsv->size = num_bytes;
3853 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3854 sinfo->bytes_reserved + sinfo->bytes_readonly +
3855 sinfo->bytes_may_use;
3857 if (sinfo->total_bytes > num_bytes) {
3858 num_bytes = sinfo->total_bytes - num_bytes;
3859 block_rsv->reserved += num_bytes;
3860 sinfo->bytes_may_use += num_bytes;
3863 if (block_rsv->reserved >= block_rsv->size) {
3864 num_bytes = block_rsv->reserved - block_rsv->size;
3865 sinfo->bytes_may_use -= num_bytes;
3866 sinfo->reservation_progress++;
3867 block_rsv->reserved = block_rsv->size;
3868 block_rsv->full = 1;
3871 spin_unlock(&sinfo->lock);
3872 spin_unlock(&block_rsv->lock);
3875 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3877 struct btrfs_space_info *space_info;
3879 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3880 fs_info->chunk_block_rsv.space_info = space_info;
3882 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3883 fs_info->global_block_rsv.space_info = space_info;
3884 fs_info->delalloc_block_rsv.space_info = space_info;
3885 fs_info->trans_block_rsv.space_info = space_info;
3886 fs_info->empty_block_rsv.space_info = space_info;
3888 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3889 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3890 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3891 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3892 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3894 update_global_block_rsv(fs_info);
3897 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3899 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3900 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3901 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3902 WARN_ON(fs_info->trans_block_rsv.size > 0);
3903 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3904 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3905 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3908 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3909 struct btrfs_root *root)
3911 struct btrfs_block_rsv *block_rsv;
3913 if (!trans->bytes_reserved)
3916 block_rsv = &root->fs_info->trans_block_rsv;
3917 btrfs_block_rsv_release(root, block_rsv, trans->bytes_reserved);
3918 trans->bytes_reserved = 0;
3921 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3922 struct inode *inode)
3924 struct btrfs_root *root = BTRFS_I(inode)->root;
3925 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3926 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3929 * We need to hold space in order to delete our orphan item once we've
3930 * added it, so this takes the reservation so we can release it later
3931 * when we are truly done with the orphan item.
3933 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3934 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3937 void btrfs_orphan_release_metadata(struct inode *inode)
3939 struct btrfs_root *root = BTRFS_I(inode)->root;
3940 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3941 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3944 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3945 struct btrfs_pending_snapshot *pending)
3947 struct btrfs_root *root = pending->root;
3948 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3949 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3951 * two for root back/forward refs, two for directory entries
3952 * and one for root of the snapshot.
3954 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3955 dst_rsv->space_info = src_rsv->space_info;
3956 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3960 * drop_outstanding_extent - drop an outstanding extent
3961 * @inode: the inode we're dropping the extent for
3963 * This is called when we are freeing up an outstanding extent, either called
3964 * after an error or after an extent is written. This will return the number of
3965 * reserved extents that need to be freed. This must be called with
3966 * BTRFS_I(inode)->lock held.
3968 static unsigned drop_outstanding_extent(struct inode *inode)
3970 unsigned dropped_extents = 0;
3972 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3973 BTRFS_I(inode)->outstanding_extents--;
3976 * If we have more or the same amount of outsanding extents than we have
3977 * reserved then we need to leave the reserved extents count alone.
3979 if (BTRFS_I(inode)->outstanding_extents >=
3980 BTRFS_I(inode)->reserved_extents)
3983 dropped_extents = BTRFS_I(inode)->reserved_extents -
3984 BTRFS_I(inode)->outstanding_extents;
3985 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3986 return dropped_extents;
3990 * calc_csum_metadata_size - return the amount of metada space that must be
3991 * reserved/free'd for the given bytes.
3992 * @inode: the inode we're manipulating
3993 * @num_bytes: the number of bytes in question
3994 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3996 * This adjusts the number of csum_bytes in the inode and then returns the
3997 * correct amount of metadata that must either be reserved or freed. We
3998 * calculate how many checksums we can fit into one leaf and then divide the
3999 * number of bytes that will need to be checksumed by this value to figure out
4000 * how many checksums will be required. If we are adding bytes then the number
4001 * may go up and we will return the number of additional bytes that must be
4002 * reserved. If it is going down we will return the number of bytes that must
4005 * This must be called with BTRFS_I(inode)->lock held.
4007 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4010 struct btrfs_root *root = BTRFS_I(inode)->root;
4012 int num_csums_per_leaf;
4016 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4017 BTRFS_I(inode)->csum_bytes == 0)
4020 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4022 BTRFS_I(inode)->csum_bytes += num_bytes;
4024 BTRFS_I(inode)->csum_bytes -= num_bytes;
4025 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4026 num_csums_per_leaf = (int)div64_u64(csum_size,
4027 sizeof(struct btrfs_csum_item) +
4028 sizeof(struct btrfs_disk_key));
4029 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4030 num_csums = num_csums + num_csums_per_leaf - 1;
4031 num_csums = num_csums / num_csums_per_leaf;
4033 old_csums = old_csums + num_csums_per_leaf - 1;
4034 old_csums = old_csums / num_csums_per_leaf;
4036 /* No change, no need to reserve more */
4037 if (old_csums == num_csums)
4041 return btrfs_calc_trans_metadata_size(root,
4042 num_csums - old_csums);
4044 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4047 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4049 struct btrfs_root *root = BTRFS_I(inode)->root;
4050 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4052 unsigned nr_extents = 0;
4056 if (btrfs_is_free_space_inode(root, inode))
4059 if (flush && btrfs_transaction_in_commit(root->fs_info))
4060 schedule_timeout(1);
4062 num_bytes = ALIGN(num_bytes, root->sectorsize);
4064 spin_lock(&BTRFS_I(inode)->lock);
4065 BTRFS_I(inode)->outstanding_extents++;
4067 if (BTRFS_I(inode)->outstanding_extents >
4068 BTRFS_I(inode)->reserved_extents) {
4069 nr_extents = BTRFS_I(inode)->outstanding_extents -
4070 BTRFS_I(inode)->reserved_extents;
4071 BTRFS_I(inode)->reserved_extents += nr_extents;
4073 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4075 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4076 spin_unlock(&BTRFS_I(inode)->lock);
4078 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush, 0);
4083 spin_lock(&BTRFS_I(inode)->lock);
4084 dropped = drop_outstanding_extent(inode);
4085 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4086 spin_unlock(&BTRFS_I(inode)->lock);
4087 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4090 * Somebody could have come in and twiddled with the
4091 * reservation, so if we have to free more than we would have
4092 * reserved from this reservation go ahead and release those
4095 to_free -= to_reserve;
4097 btrfs_block_rsv_release(root, block_rsv, to_free);
4101 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4107 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4108 * @inode: the inode to release the reservation for
4109 * @num_bytes: the number of bytes we're releasing
4111 * This will release the metadata reservation for an inode. This can be called
4112 * once we complete IO for a given set of bytes to release their metadata
4115 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4117 struct btrfs_root *root = BTRFS_I(inode)->root;
4121 num_bytes = ALIGN(num_bytes, root->sectorsize);
4122 spin_lock(&BTRFS_I(inode)->lock);
4123 dropped = drop_outstanding_extent(inode);
4125 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4126 spin_unlock(&BTRFS_I(inode)->lock);
4128 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4130 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4135 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4136 * @inode: inode we're writing to
4137 * @num_bytes: the number of bytes we want to allocate
4139 * This will do the following things
4141 * o reserve space in the data space info for num_bytes
4142 * o reserve space in the metadata space info based on number of outstanding
4143 * extents and how much csums will be needed
4144 * o add to the inodes ->delalloc_bytes
4145 * o add it to the fs_info's delalloc inodes list.
4147 * This will return 0 for success and -ENOSPC if there is no space left.
4149 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4153 ret = btrfs_check_data_free_space(inode, num_bytes);
4157 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4159 btrfs_free_reserved_data_space(inode, num_bytes);
4167 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4168 * @inode: inode we're releasing space for
4169 * @num_bytes: the number of bytes we want to free up
4171 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4172 * called in the case that we don't need the metadata AND data reservations
4173 * anymore. So if there is an error or we insert an inline extent.
4175 * This function will release the metadata space that was not used and will
4176 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4177 * list if there are no delalloc bytes left.
4179 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4181 btrfs_delalloc_release_metadata(inode, num_bytes);
4182 btrfs_free_reserved_data_space(inode, num_bytes);
4185 static int update_block_group(struct btrfs_trans_handle *trans,
4186 struct btrfs_root *root,
4187 u64 bytenr, u64 num_bytes, int alloc)
4189 struct btrfs_block_group_cache *cache = NULL;
4190 struct btrfs_fs_info *info = root->fs_info;
4191 u64 total = num_bytes;
4196 /* block accounting for super block */
4197 spin_lock(&info->delalloc_lock);
4198 old_val = btrfs_super_bytes_used(&info->super_copy);
4200 old_val += num_bytes;
4202 old_val -= num_bytes;
4203 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4204 spin_unlock(&info->delalloc_lock);
4207 cache = btrfs_lookup_block_group(info, bytenr);
4210 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4211 BTRFS_BLOCK_GROUP_RAID1 |
4212 BTRFS_BLOCK_GROUP_RAID10))
4217 * If this block group has free space cache written out, we
4218 * need to make sure to load it if we are removing space. This
4219 * is because we need the unpinning stage to actually add the
4220 * space back to the block group, otherwise we will leak space.
4222 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4223 cache_block_group(cache, trans, NULL, 1);
4225 byte_in_group = bytenr - cache->key.objectid;
4226 WARN_ON(byte_in_group > cache->key.offset);
4228 spin_lock(&cache->space_info->lock);
4229 spin_lock(&cache->lock);
4231 if (btrfs_test_opt(root, SPACE_CACHE) &&
4232 cache->disk_cache_state < BTRFS_DC_CLEAR)
4233 cache->disk_cache_state = BTRFS_DC_CLEAR;
4236 old_val = btrfs_block_group_used(&cache->item);
4237 num_bytes = min(total, cache->key.offset - byte_in_group);
4239 old_val += num_bytes;
4240 btrfs_set_block_group_used(&cache->item, old_val);
4241 cache->reserved -= num_bytes;
4242 cache->space_info->bytes_reserved -= num_bytes;
4243 cache->space_info->bytes_used += num_bytes;
4244 cache->space_info->disk_used += num_bytes * factor;
4245 spin_unlock(&cache->lock);
4246 spin_unlock(&cache->space_info->lock);
4248 old_val -= num_bytes;
4249 btrfs_set_block_group_used(&cache->item, old_val);
4250 cache->pinned += num_bytes;
4251 cache->space_info->bytes_pinned += num_bytes;
4252 cache->space_info->bytes_used -= num_bytes;
4253 cache->space_info->disk_used -= num_bytes * factor;
4254 spin_unlock(&cache->lock);
4255 spin_unlock(&cache->space_info->lock);
4257 set_extent_dirty(info->pinned_extents,
4258 bytenr, bytenr + num_bytes - 1,
4259 GFP_NOFS | __GFP_NOFAIL);
4261 btrfs_put_block_group(cache);
4263 bytenr += num_bytes;
4268 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4270 struct btrfs_block_group_cache *cache;
4273 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4277 bytenr = cache->key.objectid;
4278 btrfs_put_block_group(cache);
4283 static int pin_down_extent(struct btrfs_root *root,
4284 struct btrfs_block_group_cache *cache,
4285 u64 bytenr, u64 num_bytes, int reserved)
4287 spin_lock(&cache->space_info->lock);
4288 spin_lock(&cache->lock);
4289 cache->pinned += num_bytes;
4290 cache->space_info->bytes_pinned += num_bytes;
4292 cache->reserved -= num_bytes;
4293 cache->space_info->bytes_reserved -= num_bytes;
4295 spin_unlock(&cache->lock);
4296 spin_unlock(&cache->space_info->lock);
4298 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4299 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4304 * this function must be called within transaction
4306 int btrfs_pin_extent(struct btrfs_root *root,
4307 u64 bytenr, u64 num_bytes, int reserved)
4309 struct btrfs_block_group_cache *cache;
4311 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4314 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4316 btrfs_put_block_group(cache);
4321 * btrfs_update_reserved_bytes - update the block_group and space info counters
4322 * @cache: The cache we are manipulating
4323 * @num_bytes: The number of bytes in question
4324 * @reserve: One of the reservation enums
4326 * This is called by the allocator when it reserves space, or by somebody who is
4327 * freeing space that was never actually used on disk. For example if you
4328 * reserve some space for a new leaf in transaction A and before transaction A
4329 * commits you free that leaf, you call this with reserve set to 0 in order to
4330 * clear the reservation.
4332 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4333 * ENOSPC accounting. For data we handle the reservation through clearing the
4334 * delalloc bits in the io_tree. We have to do this since we could end up
4335 * allocating less disk space for the amount of data we have reserved in the
4336 * case of compression.
4338 * If this is a reservation and the block group has become read only we cannot
4339 * make the reservation and return -EAGAIN, otherwise this function always
4342 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4343 u64 num_bytes, int reserve)
4345 struct btrfs_space_info *space_info = cache->space_info;
4347 spin_lock(&space_info->lock);
4348 spin_lock(&cache->lock);
4349 if (reserve != RESERVE_FREE) {
4353 cache->reserved += num_bytes;
4354 space_info->bytes_reserved += num_bytes;
4355 if (reserve == RESERVE_ALLOC) {
4356 BUG_ON(space_info->bytes_may_use < num_bytes);
4357 space_info->bytes_may_use -= num_bytes;
4362 space_info->bytes_readonly += num_bytes;
4363 cache->reserved -= num_bytes;
4364 space_info->bytes_reserved -= num_bytes;
4365 space_info->reservation_progress++;
4367 spin_unlock(&cache->lock);
4368 spin_unlock(&space_info->lock);
4372 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4373 struct btrfs_root *root)
4375 struct btrfs_fs_info *fs_info = root->fs_info;
4376 struct btrfs_caching_control *next;
4377 struct btrfs_caching_control *caching_ctl;
4378 struct btrfs_block_group_cache *cache;
4380 down_write(&fs_info->extent_commit_sem);
4382 list_for_each_entry_safe(caching_ctl, next,
4383 &fs_info->caching_block_groups, list) {
4384 cache = caching_ctl->block_group;
4385 if (block_group_cache_done(cache)) {
4386 cache->last_byte_to_unpin = (u64)-1;
4387 list_del_init(&caching_ctl->list);
4388 put_caching_control(caching_ctl);
4390 cache->last_byte_to_unpin = caching_ctl->progress;
4394 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4395 fs_info->pinned_extents = &fs_info->freed_extents[1];
4397 fs_info->pinned_extents = &fs_info->freed_extents[0];
4399 up_write(&fs_info->extent_commit_sem);
4401 update_global_block_rsv(fs_info);
4405 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4407 struct btrfs_fs_info *fs_info = root->fs_info;
4408 struct btrfs_block_group_cache *cache = NULL;
4411 while (start <= end) {
4413 start >= cache->key.objectid + cache->key.offset) {
4415 btrfs_put_block_group(cache);
4416 cache = btrfs_lookup_block_group(fs_info, start);
4420 len = cache->key.objectid + cache->key.offset - start;
4421 len = min(len, end + 1 - start);
4423 if (start < cache->last_byte_to_unpin) {
4424 len = min(len, cache->last_byte_to_unpin - start);
4425 btrfs_add_free_space(cache, start, len);
4430 spin_lock(&cache->space_info->lock);
4431 spin_lock(&cache->lock);
4432 cache->pinned -= len;
4433 cache->space_info->bytes_pinned -= len;
4435 cache->space_info->bytes_readonly += len;
4436 spin_unlock(&cache->lock);
4437 spin_unlock(&cache->space_info->lock);
4441 btrfs_put_block_group(cache);
4445 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4446 struct btrfs_root *root)
4448 struct btrfs_fs_info *fs_info = root->fs_info;
4449 struct extent_io_tree *unpin;
4454 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4455 unpin = &fs_info->freed_extents[1];
4457 unpin = &fs_info->freed_extents[0];
4460 ret = find_first_extent_bit(unpin, 0, &start, &end,
4465 if (btrfs_test_opt(root, DISCARD))
4466 ret = btrfs_discard_extent(root, start,
4467 end + 1 - start, NULL);
4469 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4470 unpin_extent_range(root, start, end);
4477 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4478 struct btrfs_root *root,
4479 u64 bytenr, u64 num_bytes, u64 parent,
4480 u64 root_objectid, u64 owner_objectid,
4481 u64 owner_offset, int refs_to_drop,
4482 struct btrfs_delayed_extent_op *extent_op)
4484 struct btrfs_key key;
4485 struct btrfs_path *path;
4486 struct btrfs_fs_info *info = root->fs_info;
4487 struct btrfs_root *extent_root = info->extent_root;
4488 struct extent_buffer *leaf;
4489 struct btrfs_extent_item *ei;
4490 struct btrfs_extent_inline_ref *iref;
4493 int extent_slot = 0;
4494 int found_extent = 0;
4499 path = btrfs_alloc_path();
4504 path->leave_spinning = 1;
4506 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4507 BUG_ON(!is_data && refs_to_drop != 1);
4509 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4510 bytenr, num_bytes, parent,
4511 root_objectid, owner_objectid,
4514 extent_slot = path->slots[0];
4515 while (extent_slot >= 0) {
4516 btrfs_item_key_to_cpu(path->nodes[0], &key,
4518 if (key.objectid != bytenr)
4520 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4521 key.offset == num_bytes) {
4525 if (path->slots[0] - extent_slot > 5)
4529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4530 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4531 if (found_extent && item_size < sizeof(*ei))
4534 if (!found_extent) {
4536 ret = remove_extent_backref(trans, extent_root, path,
4540 btrfs_release_path(path);
4541 path->leave_spinning = 1;
4543 key.objectid = bytenr;
4544 key.type = BTRFS_EXTENT_ITEM_KEY;
4545 key.offset = num_bytes;
4547 ret = btrfs_search_slot(trans, extent_root,
4550 printk(KERN_ERR "umm, got %d back from search"
4551 ", was looking for %llu\n", ret,
4552 (unsigned long long)bytenr);
4554 btrfs_print_leaf(extent_root,
4558 extent_slot = path->slots[0];
4561 btrfs_print_leaf(extent_root, path->nodes[0]);
4563 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4564 "parent %llu root %llu owner %llu offset %llu\n",
4565 (unsigned long long)bytenr,
4566 (unsigned long long)parent,
4567 (unsigned long long)root_objectid,
4568 (unsigned long long)owner_objectid,
4569 (unsigned long long)owner_offset);
4572 leaf = path->nodes[0];
4573 item_size = btrfs_item_size_nr(leaf, extent_slot);
4574 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4575 if (item_size < sizeof(*ei)) {
4576 BUG_ON(found_extent || extent_slot != path->slots[0]);
4577 ret = convert_extent_item_v0(trans, extent_root, path,
4581 btrfs_release_path(path);
4582 path->leave_spinning = 1;
4584 key.objectid = bytenr;
4585 key.type = BTRFS_EXTENT_ITEM_KEY;
4586 key.offset = num_bytes;
4588 ret = btrfs_search_slot(trans, extent_root, &key, path,
4591 printk(KERN_ERR "umm, got %d back from search"
4592 ", was looking for %llu\n", ret,
4593 (unsigned long long)bytenr);
4594 btrfs_print_leaf(extent_root, path->nodes[0]);
4597 extent_slot = path->slots[0];
4598 leaf = path->nodes[0];
4599 item_size = btrfs_item_size_nr(leaf, extent_slot);
4602 BUG_ON(item_size < sizeof(*ei));
4603 ei = btrfs_item_ptr(leaf, extent_slot,
4604 struct btrfs_extent_item);
4605 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4606 struct btrfs_tree_block_info *bi;
4607 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4608 bi = (struct btrfs_tree_block_info *)(ei + 1);
4609 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4612 refs = btrfs_extent_refs(leaf, ei);
4613 BUG_ON(refs < refs_to_drop);
4614 refs -= refs_to_drop;
4618 __run_delayed_extent_op(extent_op, leaf, ei);
4620 * In the case of inline back ref, reference count will
4621 * be updated by remove_extent_backref
4624 BUG_ON(!found_extent);
4626 btrfs_set_extent_refs(leaf, ei, refs);
4627 btrfs_mark_buffer_dirty(leaf);
4630 ret = remove_extent_backref(trans, extent_root, path,
4637 BUG_ON(is_data && refs_to_drop !=
4638 extent_data_ref_count(root, path, iref));
4640 BUG_ON(path->slots[0] != extent_slot);
4642 BUG_ON(path->slots[0] != extent_slot + 1);
4643 path->slots[0] = extent_slot;
4648 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4651 btrfs_release_path(path);
4654 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4657 invalidate_mapping_pages(info->btree_inode->i_mapping,
4658 bytenr >> PAGE_CACHE_SHIFT,
4659 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4662 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4665 btrfs_free_path(path);
4670 * when we free an block, it is possible (and likely) that we free the last
4671 * delayed ref for that extent as well. This searches the delayed ref tree for
4672 * a given extent, and if there are no other delayed refs to be processed, it
4673 * removes it from the tree.
4675 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4676 struct btrfs_root *root, u64 bytenr)
4678 struct btrfs_delayed_ref_head *head;
4679 struct btrfs_delayed_ref_root *delayed_refs;
4680 struct btrfs_delayed_ref_node *ref;
4681 struct rb_node *node;
4684 delayed_refs = &trans->transaction->delayed_refs;
4685 spin_lock(&delayed_refs->lock);
4686 head = btrfs_find_delayed_ref_head(trans, bytenr);
4690 node = rb_prev(&head->node.rb_node);
4694 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4696 /* there are still entries for this ref, we can't drop it */
4697 if (ref->bytenr == bytenr)
4700 if (head->extent_op) {
4701 if (!head->must_insert_reserved)
4703 kfree(head->extent_op);
4704 head->extent_op = NULL;
4708 * waiting for the lock here would deadlock. If someone else has it
4709 * locked they are already in the process of dropping it anyway
4711 if (!mutex_trylock(&head->mutex))
4715 * at this point we have a head with no other entries. Go
4716 * ahead and process it.
4718 head->node.in_tree = 0;
4719 rb_erase(&head->node.rb_node, &delayed_refs->root);
4721 delayed_refs->num_entries--;
4724 * we don't take a ref on the node because we're removing it from the
4725 * tree, so we just steal the ref the tree was holding.
4727 delayed_refs->num_heads--;
4728 if (list_empty(&head->cluster))
4729 delayed_refs->num_heads_ready--;
4731 list_del_init(&head->cluster);
4732 spin_unlock(&delayed_refs->lock);
4734 BUG_ON(head->extent_op);
4735 if (head->must_insert_reserved)
4738 mutex_unlock(&head->mutex);
4739 btrfs_put_delayed_ref(&head->node);
4742 spin_unlock(&delayed_refs->lock);
4746 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4747 struct btrfs_root *root,
4748 struct extent_buffer *buf,
4749 u64 parent, int last_ref)
4751 struct btrfs_block_group_cache *cache = NULL;
4754 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4755 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4756 parent, root->root_key.objectid,
4757 btrfs_header_level(buf),
4758 BTRFS_DROP_DELAYED_REF, NULL);
4765 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4767 if (btrfs_header_generation(buf) == trans->transid) {
4768 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4769 ret = check_ref_cleanup(trans, root, buf->start);
4774 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4775 pin_down_extent(root, cache, buf->start, buf->len, 1);
4779 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4781 btrfs_add_free_space(cache, buf->start, buf->len);
4782 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4786 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4789 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4790 btrfs_put_block_group(cache);
4793 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4794 struct btrfs_root *root,
4795 u64 bytenr, u64 num_bytes, u64 parent,
4796 u64 root_objectid, u64 owner, u64 offset)
4801 * tree log blocks never actually go into the extent allocation
4802 * tree, just update pinning info and exit early.
4804 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4805 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4806 /* unlocks the pinned mutex */
4807 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4809 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4810 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4811 parent, root_objectid, (int)owner,
4812 BTRFS_DROP_DELAYED_REF, NULL);
4815 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4816 parent, root_objectid, owner,
4817 offset, BTRFS_DROP_DELAYED_REF, NULL);
4823 static u64 stripe_align(struct btrfs_root *root, u64 val)
4825 u64 mask = ((u64)root->stripesize - 1);
4826 u64 ret = (val + mask) & ~mask;
4831 * when we wait for progress in the block group caching, its because
4832 * our allocation attempt failed at least once. So, we must sleep
4833 * and let some progress happen before we try again.
4835 * This function will sleep at least once waiting for new free space to
4836 * show up, and then it will check the block group free space numbers
4837 * for our min num_bytes. Another option is to have it go ahead
4838 * and look in the rbtree for a free extent of a given size, but this
4842 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4845 struct btrfs_caching_control *caching_ctl;
4848 caching_ctl = get_caching_control(cache);
4852 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4853 (cache->free_space_ctl->free_space >= num_bytes));
4855 put_caching_control(caching_ctl);
4860 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4862 struct btrfs_caching_control *caching_ctl;
4865 caching_ctl = get_caching_control(cache);
4869 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4871 put_caching_control(caching_ctl);
4875 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4878 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4880 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4882 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4884 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4891 enum btrfs_loop_type {
4892 LOOP_FIND_IDEAL = 0,
4893 LOOP_CACHING_NOWAIT = 1,
4894 LOOP_CACHING_WAIT = 2,
4895 LOOP_ALLOC_CHUNK = 3,
4896 LOOP_NO_EMPTY_SIZE = 4,
4900 * walks the btree of allocated extents and find a hole of a given size.
4901 * The key ins is changed to record the hole:
4902 * ins->objectid == block start
4903 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4904 * ins->offset == number of blocks
4905 * Any available blocks before search_start are skipped.
4907 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4908 struct btrfs_root *orig_root,
4909 u64 num_bytes, u64 empty_size,
4910 u64 search_start, u64 search_end,
4911 u64 hint_byte, struct btrfs_key *ins,
4915 struct btrfs_root *root = orig_root->fs_info->extent_root;
4916 struct btrfs_free_cluster *last_ptr = NULL;
4917 struct btrfs_block_group_cache *block_group = NULL;
4918 int empty_cluster = 2 * 1024 * 1024;
4919 int allowed_chunk_alloc = 0;
4920 int done_chunk_alloc = 0;
4921 struct btrfs_space_info *space_info;
4922 int last_ptr_loop = 0;
4925 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
4926 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
4927 bool found_uncached_bg = false;
4928 bool failed_cluster_refill = false;
4929 bool failed_alloc = false;
4930 bool use_cluster = true;
4931 u64 ideal_cache_percent = 0;
4932 u64 ideal_cache_offset = 0;
4934 WARN_ON(num_bytes < root->sectorsize);
4935 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4939 space_info = __find_space_info(root->fs_info, data);
4941 printk(KERN_ERR "No space info for %llu\n", data);
4946 * If the space info is for both data and metadata it means we have a
4947 * small filesystem and we can't use the clustering stuff.
4949 if (btrfs_mixed_space_info(space_info))
4950 use_cluster = false;
4952 if (orig_root->ref_cows || empty_size)
4953 allowed_chunk_alloc = 1;
4955 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4956 last_ptr = &root->fs_info->meta_alloc_cluster;
4957 if (!btrfs_test_opt(root, SSD))
4958 empty_cluster = 64 * 1024;
4961 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4962 btrfs_test_opt(root, SSD)) {
4963 last_ptr = &root->fs_info->data_alloc_cluster;
4967 spin_lock(&last_ptr->lock);
4968 if (last_ptr->block_group)
4969 hint_byte = last_ptr->window_start;
4970 spin_unlock(&last_ptr->lock);
4973 search_start = max(search_start, first_logical_byte(root, 0));
4974 search_start = max(search_start, hint_byte);
4979 if (search_start == hint_byte) {
4981 block_group = btrfs_lookup_block_group(root->fs_info,
4984 * we don't want to use the block group if it doesn't match our
4985 * allocation bits, or if its not cached.
4987 * However if we are re-searching with an ideal block group
4988 * picked out then we don't care that the block group is cached.
4990 if (block_group && block_group_bits(block_group, data) &&
4991 (block_group->cached != BTRFS_CACHE_NO ||
4992 search_start == ideal_cache_offset)) {
4993 down_read(&space_info->groups_sem);
4994 if (list_empty(&block_group->list) ||
4997 * someone is removing this block group,
4998 * we can't jump into the have_block_group
4999 * target because our list pointers are not
5002 btrfs_put_block_group(block_group);
5003 up_read(&space_info->groups_sem);
5005 index = get_block_group_index(block_group);
5006 goto have_block_group;
5008 } else if (block_group) {
5009 btrfs_put_block_group(block_group);
5013 down_read(&space_info->groups_sem);
5014 list_for_each_entry(block_group, &space_info->block_groups[index],
5019 btrfs_get_block_group(block_group);
5020 search_start = block_group->key.objectid;
5023 * this can happen if we end up cycling through all the
5024 * raid types, but we want to make sure we only allocate
5025 * for the proper type.
5027 if (!block_group_bits(block_group, data)) {
5028 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5029 BTRFS_BLOCK_GROUP_RAID1 |
5030 BTRFS_BLOCK_GROUP_RAID10;
5033 * if they asked for extra copies and this block group
5034 * doesn't provide them, bail. This does allow us to
5035 * fill raid0 from raid1.
5037 if ((data & extra) && !(block_group->flags & extra))
5042 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5045 ret = cache_block_group(block_group, trans,
5047 if (block_group->cached == BTRFS_CACHE_FINISHED)
5048 goto have_block_group;
5050 free_percent = btrfs_block_group_used(&block_group->item);
5051 free_percent *= 100;
5052 free_percent = div64_u64(free_percent,
5053 block_group->key.offset);
5054 free_percent = 100 - free_percent;
5055 if (free_percent > ideal_cache_percent &&
5056 likely(!block_group->ro)) {
5057 ideal_cache_offset = block_group->key.objectid;
5058 ideal_cache_percent = free_percent;
5062 * The caching workers are limited to 2 threads, so we
5063 * can queue as much work as we care to.
5065 if (loop > LOOP_FIND_IDEAL) {
5066 ret = cache_block_group(block_group, trans,
5070 found_uncached_bg = true;
5073 * If loop is set for cached only, try the next block
5076 if (loop == LOOP_FIND_IDEAL)
5080 cached = block_group_cache_done(block_group);
5081 if (unlikely(!cached))
5082 found_uncached_bg = true;
5084 if (unlikely(block_group->ro))
5087 spin_lock(&block_group->free_space_ctl->tree_lock);
5089 block_group->free_space_ctl->free_space <
5090 num_bytes + empty_size) {
5091 spin_unlock(&block_group->free_space_ctl->tree_lock);
5094 spin_unlock(&block_group->free_space_ctl->tree_lock);
5097 * Ok we want to try and use the cluster allocator, so lets look
5098 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5099 * have tried the cluster allocator plenty of times at this
5100 * point and not have found anything, so we are likely way too
5101 * fragmented for the clustering stuff to find anything, so lets
5102 * just skip it and let the allocator find whatever block it can
5105 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5107 * the refill lock keeps out other
5108 * people trying to start a new cluster
5110 spin_lock(&last_ptr->refill_lock);
5111 if (last_ptr->block_group &&
5112 (last_ptr->block_group->ro ||
5113 !block_group_bits(last_ptr->block_group, data))) {
5115 goto refill_cluster;
5118 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5119 num_bytes, search_start);
5121 /* we have a block, we're done */
5122 spin_unlock(&last_ptr->refill_lock);
5126 spin_lock(&last_ptr->lock);
5128 * whoops, this cluster doesn't actually point to
5129 * this block group. Get a ref on the block
5130 * group is does point to and try again
5132 if (!last_ptr_loop && last_ptr->block_group &&
5133 last_ptr->block_group != block_group &&
5135 get_block_group_index(last_ptr->block_group)) {
5137 btrfs_put_block_group(block_group);
5138 block_group = last_ptr->block_group;
5139 btrfs_get_block_group(block_group);
5140 spin_unlock(&last_ptr->lock);
5141 spin_unlock(&last_ptr->refill_lock);
5144 search_start = block_group->key.objectid;
5146 * we know this block group is properly
5147 * in the list because
5148 * btrfs_remove_block_group, drops the
5149 * cluster before it removes the block
5150 * group from the list
5152 goto have_block_group;
5154 spin_unlock(&last_ptr->lock);
5157 * this cluster didn't work out, free it and
5160 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5164 /* allocate a cluster in this block group */
5165 ret = btrfs_find_space_cluster(trans, root,
5166 block_group, last_ptr,
5168 empty_cluster + empty_size);
5171 * now pull our allocation out of this
5174 offset = btrfs_alloc_from_cluster(block_group,
5175 last_ptr, num_bytes,
5178 /* we found one, proceed */
5179 spin_unlock(&last_ptr->refill_lock);
5182 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5183 && !failed_cluster_refill) {
5184 spin_unlock(&last_ptr->refill_lock);
5186 failed_cluster_refill = true;
5187 wait_block_group_cache_progress(block_group,
5188 num_bytes + empty_cluster + empty_size);
5189 goto have_block_group;
5193 * at this point we either didn't find a cluster
5194 * or we weren't able to allocate a block from our
5195 * cluster. Free the cluster we've been trying
5196 * to use, and go to the next block group
5198 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5199 spin_unlock(&last_ptr->refill_lock);
5203 offset = btrfs_find_space_for_alloc(block_group, search_start,
5204 num_bytes, empty_size);
5206 * If we didn't find a chunk, and we haven't failed on this
5207 * block group before, and this block group is in the middle of
5208 * caching and we are ok with waiting, then go ahead and wait
5209 * for progress to be made, and set failed_alloc to true.
5211 * If failed_alloc is true then we've already waited on this
5212 * block group once and should move on to the next block group.
5214 if (!offset && !failed_alloc && !cached &&
5215 loop > LOOP_CACHING_NOWAIT) {
5216 wait_block_group_cache_progress(block_group,
5217 num_bytes + empty_size);
5218 failed_alloc = true;
5219 goto have_block_group;
5220 } else if (!offset) {
5224 search_start = stripe_align(root, offset);
5225 /* move on to the next group */
5226 if (search_start + num_bytes >= search_end) {
5227 btrfs_add_free_space(block_group, offset, num_bytes);
5231 /* move on to the next group */
5232 if (search_start + num_bytes >
5233 block_group->key.objectid + block_group->key.offset) {
5234 btrfs_add_free_space(block_group, offset, num_bytes);
5238 ins->objectid = search_start;
5239 ins->offset = num_bytes;
5241 if (offset < search_start)
5242 btrfs_add_free_space(block_group, offset,
5243 search_start - offset);
5244 BUG_ON(offset > search_start);
5246 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5248 if (ret == -EAGAIN) {
5249 btrfs_add_free_space(block_group, offset, num_bytes);
5253 /* we are all good, lets return */
5254 ins->objectid = search_start;
5255 ins->offset = num_bytes;
5257 if (offset < search_start)
5258 btrfs_add_free_space(block_group, offset,
5259 search_start - offset);
5260 BUG_ON(offset > search_start);
5261 btrfs_put_block_group(block_group);
5264 failed_cluster_refill = false;
5265 failed_alloc = false;
5266 BUG_ON(index != get_block_group_index(block_group));
5267 btrfs_put_block_group(block_group);
5269 up_read(&space_info->groups_sem);
5271 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5274 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5275 * for them to make caching progress. Also
5276 * determine the best possible bg to cache
5277 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5278 * caching kthreads as we move along
5279 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5280 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5281 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5284 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5286 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5287 found_uncached_bg = false;
5289 if (!ideal_cache_percent)
5293 * 1 of the following 2 things have happened so far
5295 * 1) We found an ideal block group for caching that
5296 * is mostly full and will cache quickly, so we might
5297 * as well wait for it.
5299 * 2) We searched for cached only and we didn't find
5300 * anything, and we didn't start any caching kthreads
5301 * either, so chances are we will loop through and
5302 * start a couple caching kthreads, and then come back
5303 * around and just wait for them. This will be slower
5304 * because we will have 2 caching kthreads reading at
5305 * the same time when we could have just started one
5306 * and waited for it to get far enough to give us an
5307 * allocation, so go ahead and go to the wait caching
5310 loop = LOOP_CACHING_WAIT;
5311 search_start = ideal_cache_offset;
5312 ideal_cache_percent = 0;
5314 } else if (loop == LOOP_FIND_IDEAL) {
5316 * Didn't find a uncached bg, wait on anything we find
5319 loop = LOOP_CACHING_WAIT;
5325 if (loop == LOOP_ALLOC_CHUNK) {
5326 if (allowed_chunk_alloc) {
5327 ret = do_chunk_alloc(trans, root, num_bytes +
5328 2 * 1024 * 1024, data,
5329 CHUNK_ALLOC_LIMITED);
5330 allowed_chunk_alloc = 0;
5332 done_chunk_alloc = 1;
5333 } else if (!done_chunk_alloc &&
5334 space_info->force_alloc ==
5335 CHUNK_ALLOC_NO_FORCE) {
5336 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5340 * We didn't allocate a chunk, go ahead and drop the
5341 * empty size and loop again.
5343 if (!done_chunk_alloc)
5344 loop = LOOP_NO_EMPTY_SIZE;
5347 if (loop == LOOP_NO_EMPTY_SIZE) {
5353 } else if (!ins->objectid) {
5355 } else if (ins->objectid) {
5362 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5363 int dump_block_groups)
5365 struct btrfs_block_group_cache *cache;
5368 spin_lock(&info->lock);
5369 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5370 (unsigned long long)info->flags,
5371 (unsigned long long)(info->total_bytes - info->bytes_used -
5372 info->bytes_pinned - info->bytes_reserved -
5373 info->bytes_readonly),
5374 (info->full) ? "" : "not ");
5375 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5376 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5377 (unsigned long long)info->total_bytes,
5378 (unsigned long long)info->bytes_used,
5379 (unsigned long long)info->bytes_pinned,
5380 (unsigned long long)info->bytes_reserved,
5381 (unsigned long long)info->bytes_may_use,
5382 (unsigned long long)info->bytes_readonly);
5383 spin_unlock(&info->lock);
5385 if (!dump_block_groups)
5388 down_read(&info->groups_sem);
5390 list_for_each_entry(cache, &info->block_groups[index], list) {
5391 spin_lock(&cache->lock);
5392 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5393 "%llu pinned %llu reserved\n",
5394 (unsigned long long)cache->key.objectid,
5395 (unsigned long long)cache->key.offset,
5396 (unsigned long long)btrfs_block_group_used(&cache->item),
5397 (unsigned long long)cache->pinned,
5398 (unsigned long long)cache->reserved);
5399 btrfs_dump_free_space(cache, bytes);
5400 spin_unlock(&cache->lock);
5402 if (++index < BTRFS_NR_RAID_TYPES)
5404 up_read(&info->groups_sem);
5407 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5408 struct btrfs_root *root,
5409 u64 num_bytes, u64 min_alloc_size,
5410 u64 empty_size, u64 hint_byte,
5411 u64 search_end, struct btrfs_key *ins,
5415 u64 search_start = 0;
5417 data = btrfs_get_alloc_profile(root, data);
5420 * the only place that sets empty_size is btrfs_realloc_node, which
5421 * is not called recursively on allocations
5423 if (empty_size || root->ref_cows)
5424 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5425 num_bytes + 2 * 1024 * 1024, data,
5426 CHUNK_ALLOC_NO_FORCE);
5428 WARN_ON(num_bytes < root->sectorsize);
5429 ret = find_free_extent(trans, root, num_bytes, empty_size,
5430 search_start, search_end, hint_byte,
5433 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5434 num_bytes = num_bytes >> 1;
5435 num_bytes = num_bytes & ~(root->sectorsize - 1);
5436 num_bytes = max(num_bytes, min_alloc_size);
5437 do_chunk_alloc(trans, root->fs_info->extent_root,
5438 num_bytes, data, CHUNK_ALLOC_FORCE);
5441 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5442 struct btrfs_space_info *sinfo;
5444 sinfo = __find_space_info(root->fs_info, data);
5445 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5446 "wanted %llu\n", (unsigned long long)data,
5447 (unsigned long long)num_bytes);
5448 dump_space_info(sinfo, num_bytes, 1);
5451 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5456 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5458 struct btrfs_block_group_cache *cache;
5461 cache = btrfs_lookup_block_group(root->fs_info, start);
5463 printk(KERN_ERR "Unable to find block group for %llu\n",
5464 (unsigned long long)start);
5468 if (btrfs_test_opt(root, DISCARD))
5469 ret = btrfs_discard_extent(root, start, len, NULL);
5471 btrfs_add_free_space(cache, start, len);
5472 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5473 btrfs_put_block_group(cache);
5475 trace_btrfs_reserved_extent_free(root, start, len);
5480 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5481 struct btrfs_root *root,
5482 u64 parent, u64 root_objectid,
5483 u64 flags, u64 owner, u64 offset,
5484 struct btrfs_key *ins, int ref_mod)
5487 struct btrfs_fs_info *fs_info = root->fs_info;
5488 struct btrfs_extent_item *extent_item;
5489 struct btrfs_extent_inline_ref *iref;
5490 struct btrfs_path *path;
5491 struct extent_buffer *leaf;
5496 type = BTRFS_SHARED_DATA_REF_KEY;
5498 type = BTRFS_EXTENT_DATA_REF_KEY;
5500 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5502 path = btrfs_alloc_path();
5506 path->leave_spinning = 1;
5507 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5511 leaf = path->nodes[0];
5512 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5513 struct btrfs_extent_item);
5514 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5515 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5516 btrfs_set_extent_flags(leaf, extent_item,
5517 flags | BTRFS_EXTENT_FLAG_DATA);
5519 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5520 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5522 struct btrfs_shared_data_ref *ref;
5523 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5524 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5525 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5527 struct btrfs_extent_data_ref *ref;
5528 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5529 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5530 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5531 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5532 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5535 btrfs_mark_buffer_dirty(path->nodes[0]);
5536 btrfs_free_path(path);
5538 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5540 printk(KERN_ERR "btrfs update block group failed for %llu "
5541 "%llu\n", (unsigned long long)ins->objectid,
5542 (unsigned long long)ins->offset);
5548 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5549 struct btrfs_root *root,
5550 u64 parent, u64 root_objectid,
5551 u64 flags, struct btrfs_disk_key *key,
5552 int level, struct btrfs_key *ins)
5555 struct btrfs_fs_info *fs_info = root->fs_info;
5556 struct btrfs_extent_item *extent_item;
5557 struct btrfs_tree_block_info *block_info;
5558 struct btrfs_extent_inline_ref *iref;
5559 struct btrfs_path *path;
5560 struct extent_buffer *leaf;
5561 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5563 path = btrfs_alloc_path();
5567 path->leave_spinning = 1;
5568 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5572 leaf = path->nodes[0];
5573 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5574 struct btrfs_extent_item);
5575 btrfs_set_extent_refs(leaf, extent_item, 1);
5576 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5577 btrfs_set_extent_flags(leaf, extent_item,
5578 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5579 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5581 btrfs_set_tree_block_key(leaf, block_info, key);
5582 btrfs_set_tree_block_level(leaf, block_info, level);
5584 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5586 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5587 btrfs_set_extent_inline_ref_type(leaf, iref,
5588 BTRFS_SHARED_BLOCK_REF_KEY);
5589 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5591 btrfs_set_extent_inline_ref_type(leaf, iref,
5592 BTRFS_TREE_BLOCK_REF_KEY);
5593 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5596 btrfs_mark_buffer_dirty(leaf);
5597 btrfs_free_path(path);
5599 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5601 printk(KERN_ERR "btrfs update block group failed for %llu "
5602 "%llu\n", (unsigned long long)ins->objectid,
5603 (unsigned long long)ins->offset);
5609 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5610 struct btrfs_root *root,
5611 u64 root_objectid, u64 owner,
5612 u64 offset, struct btrfs_key *ins)
5616 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5618 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5619 0, root_objectid, owner, offset,
5620 BTRFS_ADD_DELAYED_EXTENT, NULL);
5625 * this is used by the tree logging recovery code. It records that
5626 * an extent has been allocated and makes sure to clear the free
5627 * space cache bits as well
5629 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5630 struct btrfs_root *root,
5631 u64 root_objectid, u64 owner, u64 offset,
5632 struct btrfs_key *ins)
5635 struct btrfs_block_group_cache *block_group;
5636 struct btrfs_caching_control *caching_ctl;
5637 u64 start = ins->objectid;
5638 u64 num_bytes = ins->offset;
5640 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5641 cache_block_group(block_group, trans, NULL, 0);
5642 caching_ctl = get_caching_control(block_group);
5645 BUG_ON(!block_group_cache_done(block_group));
5646 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5649 mutex_lock(&caching_ctl->mutex);
5651 if (start >= caching_ctl->progress) {
5652 ret = add_excluded_extent(root, start, num_bytes);
5654 } else if (start + num_bytes <= caching_ctl->progress) {
5655 ret = btrfs_remove_free_space(block_group,
5659 num_bytes = caching_ctl->progress - start;
5660 ret = btrfs_remove_free_space(block_group,
5664 start = caching_ctl->progress;
5665 num_bytes = ins->objectid + ins->offset -
5666 caching_ctl->progress;
5667 ret = add_excluded_extent(root, start, num_bytes);
5671 mutex_unlock(&caching_ctl->mutex);
5672 put_caching_control(caching_ctl);
5675 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5676 RESERVE_ALLOC_NO_ACCOUNT);
5678 btrfs_put_block_group(block_group);
5679 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5680 0, owner, offset, ins, 1);
5684 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5685 struct btrfs_root *root,
5686 u64 bytenr, u32 blocksize,
5689 struct extent_buffer *buf;
5691 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5693 return ERR_PTR(-ENOMEM);
5694 btrfs_set_header_generation(buf, trans->transid);
5695 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5696 btrfs_tree_lock(buf);
5697 clean_tree_block(trans, root, buf);
5699 btrfs_set_lock_blocking(buf);
5700 btrfs_set_buffer_uptodate(buf);
5702 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5704 * we allow two log transactions at a time, use different
5705 * EXENT bit to differentiate dirty pages.
5707 if (root->log_transid % 2 == 0)
5708 set_extent_dirty(&root->dirty_log_pages, buf->start,
5709 buf->start + buf->len - 1, GFP_NOFS);
5711 set_extent_new(&root->dirty_log_pages, buf->start,
5712 buf->start + buf->len - 1, GFP_NOFS);
5714 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5715 buf->start + buf->len - 1, GFP_NOFS);
5717 trans->blocks_used++;
5718 /* this returns a buffer locked for blocking */
5722 static struct btrfs_block_rsv *
5723 use_block_rsv(struct btrfs_trans_handle *trans,
5724 struct btrfs_root *root, u32 blocksize)
5726 struct btrfs_block_rsv *block_rsv;
5727 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5730 block_rsv = get_block_rsv(trans, root);
5732 if (block_rsv->size == 0) {
5733 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5735 * If we couldn't reserve metadata bytes try and use some from
5736 * the global reserve.
5738 if (ret && block_rsv != global_rsv) {
5739 ret = block_rsv_use_bytes(global_rsv, blocksize);
5742 return ERR_PTR(ret);
5744 return ERR_PTR(ret);
5749 ret = block_rsv_use_bytes(block_rsv, blocksize);
5754 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5757 } else if (ret && block_rsv != global_rsv) {
5758 ret = block_rsv_use_bytes(global_rsv, blocksize);
5764 return ERR_PTR(-ENOSPC);
5767 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5769 block_rsv_add_bytes(block_rsv, blocksize, 0);
5770 block_rsv_release_bytes(block_rsv, NULL, 0);
5774 * finds a free extent and does all the dirty work required for allocation
5775 * returns the key for the extent through ins, and a tree buffer for
5776 * the first block of the extent through buf.
5778 * returns the tree buffer or NULL.
5780 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5781 struct btrfs_root *root, u32 blocksize,
5782 u64 parent, u64 root_objectid,
5783 struct btrfs_disk_key *key, int level,
5784 u64 hint, u64 empty_size)
5786 struct btrfs_key ins;
5787 struct btrfs_block_rsv *block_rsv;
5788 struct extent_buffer *buf;
5793 block_rsv = use_block_rsv(trans, root, blocksize);
5794 if (IS_ERR(block_rsv))
5795 return ERR_CAST(block_rsv);
5797 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5798 empty_size, hint, (u64)-1, &ins, 0);
5800 unuse_block_rsv(block_rsv, blocksize);
5801 return ERR_PTR(ret);
5804 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5806 BUG_ON(IS_ERR(buf));
5808 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5810 parent = ins.objectid;
5811 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5815 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5816 struct btrfs_delayed_extent_op *extent_op;
5817 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5820 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5822 memset(&extent_op->key, 0, sizeof(extent_op->key));
5823 extent_op->flags_to_set = flags;
5824 extent_op->update_key = 1;
5825 extent_op->update_flags = 1;
5826 extent_op->is_data = 0;
5828 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5829 ins.offset, parent, root_objectid,
5830 level, BTRFS_ADD_DELAYED_EXTENT,
5837 struct walk_control {
5838 u64 refs[BTRFS_MAX_LEVEL];
5839 u64 flags[BTRFS_MAX_LEVEL];
5840 struct btrfs_key update_progress;
5850 #define DROP_REFERENCE 1
5851 #define UPDATE_BACKREF 2
5853 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5854 struct btrfs_root *root,
5855 struct walk_control *wc,
5856 struct btrfs_path *path)
5864 struct btrfs_key key;
5865 struct extent_buffer *eb;
5870 if (path->slots[wc->level] < wc->reada_slot) {
5871 wc->reada_count = wc->reada_count * 2 / 3;
5872 wc->reada_count = max(wc->reada_count, 2);
5874 wc->reada_count = wc->reada_count * 3 / 2;
5875 wc->reada_count = min_t(int, wc->reada_count,
5876 BTRFS_NODEPTRS_PER_BLOCK(root));
5879 eb = path->nodes[wc->level];
5880 nritems = btrfs_header_nritems(eb);
5881 blocksize = btrfs_level_size(root, wc->level - 1);
5883 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5884 if (nread >= wc->reada_count)
5888 bytenr = btrfs_node_blockptr(eb, slot);
5889 generation = btrfs_node_ptr_generation(eb, slot);
5891 if (slot == path->slots[wc->level])
5894 if (wc->stage == UPDATE_BACKREF &&
5895 generation <= root->root_key.offset)
5898 /* We don't lock the tree block, it's OK to be racy here */
5899 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5904 if (wc->stage == DROP_REFERENCE) {
5908 if (wc->level == 1 &&
5909 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5911 if (!wc->update_ref ||
5912 generation <= root->root_key.offset)
5914 btrfs_node_key_to_cpu(eb, &key, slot);
5915 ret = btrfs_comp_cpu_keys(&key,
5916 &wc->update_progress);
5920 if (wc->level == 1 &&
5921 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5925 ret = readahead_tree_block(root, bytenr, blocksize,
5931 wc->reada_slot = slot;
5935 * hepler to process tree block while walking down the tree.
5937 * when wc->stage == UPDATE_BACKREF, this function updates
5938 * back refs for pointers in the block.
5940 * NOTE: return value 1 means we should stop walking down.
5942 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5943 struct btrfs_root *root,
5944 struct btrfs_path *path,
5945 struct walk_control *wc, int lookup_info)
5947 int level = wc->level;
5948 struct extent_buffer *eb = path->nodes[level];
5949 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5952 if (wc->stage == UPDATE_BACKREF &&
5953 btrfs_header_owner(eb) != root->root_key.objectid)
5957 * when reference count of tree block is 1, it won't increase
5958 * again. once full backref flag is set, we never clear it.
5961 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5962 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5963 BUG_ON(!path->locks[level]);
5964 ret = btrfs_lookup_extent_info(trans, root,
5969 BUG_ON(wc->refs[level] == 0);
5972 if (wc->stage == DROP_REFERENCE) {
5973 if (wc->refs[level] > 1)
5976 if (path->locks[level] && !wc->keep_locks) {
5977 btrfs_tree_unlock_rw(eb, path->locks[level]);
5978 path->locks[level] = 0;
5983 /* wc->stage == UPDATE_BACKREF */
5984 if (!(wc->flags[level] & flag)) {
5985 BUG_ON(!path->locks[level]);
5986 ret = btrfs_inc_ref(trans, root, eb, 1);
5988 ret = btrfs_dec_ref(trans, root, eb, 0);
5990 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5993 wc->flags[level] |= flag;
5997 * the block is shared by multiple trees, so it's not good to
5998 * keep the tree lock
6000 if (path->locks[level] && level > 0) {
6001 btrfs_tree_unlock_rw(eb, path->locks[level]);
6002 path->locks[level] = 0;
6008 * hepler to process tree block pointer.
6010 * when wc->stage == DROP_REFERENCE, this function checks
6011 * reference count of the block pointed to. if the block
6012 * is shared and we need update back refs for the subtree
6013 * rooted at the block, this function changes wc->stage to
6014 * UPDATE_BACKREF. if the block is shared and there is no
6015 * need to update back, this function drops the reference
6018 * NOTE: return value 1 means we should stop walking down.
6020 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6021 struct btrfs_root *root,
6022 struct btrfs_path *path,
6023 struct walk_control *wc, int *lookup_info)
6029 struct btrfs_key key;
6030 struct extent_buffer *next;
6031 int level = wc->level;
6035 generation = btrfs_node_ptr_generation(path->nodes[level],
6036 path->slots[level]);
6038 * if the lower level block was created before the snapshot
6039 * was created, we know there is no need to update back refs
6042 if (wc->stage == UPDATE_BACKREF &&
6043 generation <= root->root_key.offset) {
6048 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6049 blocksize = btrfs_level_size(root, level - 1);
6051 next = btrfs_find_tree_block(root, bytenr, blocksize);
6053 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6058 btrfs_tree_lock(next);
6059 btrfs_set_lock_blocking(next);
6061 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6062 &wc->refs[level - 1],
6063 &wc->flags[level - 1]);
6065 BUG_ON(wc->refs[level - 1] == 0);
6068 if (wc->stage == DROP_REFERENCE) {
6069 if (wc->refs[level - 1] > 1) {
6071 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6074 if (!wc->update_ref ||
6075 generation <= root->root_key.offset)
6078 btrfs_node_key_to_cpu(path->nodes[level], &key,
6079 path->slots[level]);
6080 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6084 wc->stage = UPDATE_BACKREF;
6085 wc->shared_level = level - 1;
6089 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6093 if (!btrfs_buffer_uptodate(next, generation)) {
6094 btrfs_tree_unlock(next);
6095 free_extent_buffer(next);
6101 if (reada && level == 1)
6102 reada_walk_down(trans, root, wc, path);
6103 next = read_tree_block(root, bytenr, blocksize, generation);
6106 btrfs_tree_lock(next);
6107 btrfs_set_lock_blocking(next);
6111 BUG_ON(level != btrfs_header_level(next));
6112 path->nodes[level] = next;
6113 path->slots[level] = 0;
6114 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6120 wc->refs[level - 1] = 0;
6121 wc->flags[level - 1] = 0;
6122 if (wc->stage == DROP_REFERENCE) {
6123 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6124 parent = path->nodes[level]->start;
6126 BUG_ON(root->root_key.objectid !=
6127 btrfs_header_owner(path->nodes[level]));
6131 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6132 root->root_key.objectid, level - 1, 0);
6135 btrfs_tree_unlock(next);
6136 free_extent_buffer(next);
6142 * hepler to process tree block while walking up the tree.
6144 * when wc->stage == DROP_REFERENCE, this function drops
6145 * reference count on the block.
6147 * when wc->stage == UPDATE_BACKREF, this function changes
6148 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6149 * to UPDATE_BACKREF previously while processing the block.
6151 * NOTE: return value 1 means we should stop walking up.
6153 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6154 struct btrfs_root *root,
6155 struct btrfs_path *path,
6156 struct walk_control *wc)
6159 int level = wc->level;
6160 struct extent_buffer *eb = path->nodes[level];
6163 if (wc->stage == UPDATE_BACKREF) {
6164 BUG_ON(wc->shared_level < level);
6165 if (level < wc->shared_level)
6168 ret = find_next_key(path, level + 1, &wc->update_progress);
6172 wc->stage = DROP_REFERENCE;
6173 wc->shared_level = -1;
6174 path->slots[level] = 0;
6177 * check reference count again if the block isn't locked.
6178 * we should start walking down the tree again if reference
6181 if (!path->locks[level]) {
6183 btrfs_tree_lock(eb);
6184 btrfs_set_lock_blocking(eb);
6185 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6187 ret = btrfs_lookup_extent_info(trans, root,
6192 BUG_ON(wc->refs[level] == 0);
6193 if (wc->refs[level] == 1) {
6194 btrfs_tree_unlock_rw(eb, path->locks[level]);
6200 /* wc->stage == DROP_REFERENCE */
6201 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6203 if (wc->refs[level] == 1) {
6205 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6206 ret = btrfs_dec_ref(trans, root, eb, 1);
6208 ret = btrfs_dec_ref(trans, root, eb, 0);
6211 /* make block locked assertion in clean_tree_block happy */
6212 if (!path->locks[level] &&
6213 btrfs_header_generation(eb) == trans->transid) {
6214 btrfs_tree_lock(eb);
6215 btrfs_set_lock_blocking(eb);
6216 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6218 clean_tree_block(trans, root, eb);
6221 if (eb == root->node) {
6222 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6225 BUG_ON(root->root_key.objectid !=
6226 btrfs_header_owner(eb));
6228 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6229 parent = path->nodes[level + 1]->start;
6231 BUG_ON(root->root_key.objectid !=
6232 btrfs_header_owner(path->nodes[level + 1]));
6235 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6237 wc->refs[level] = 0;
6238 wc->flags[level] = 0;
6242 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6243 struct btrfs_root *root,
6244 struct btrfs_path *path,
6245 struct walk_control *wc)
6247 int level = wc->level;
6248 int lookup_info = 1;
6251 while (level >= 0) {
6252 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6259 if (path->slots[level] >=
6260 btrfs_header_nritems(path->nodes[level]))
6263 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6265 path->slots[level]++;
6274 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6275 struct btrfs_root *root,
6276 struct btrfs_path *path,
6277 struct walk_control *wc, int max_level)
6279 int level = wc->level;
6282 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6283 while (level < max_level && path->nodes[level]) {
6285 if (path->slots[level] + 1 <
6286 btrfs_header_nritems(path->nodes[level])) {
6287 path->slots[level]++;
6290 ret = walk_up_proc(trans, root, path, wc);
6294 if (path->locks[level]) {
6295 btrfs_tree_unlock_rw(path->nodes[level],
6296 path->locks[level]);
6297 path->locks[level] = 0;
6299 free_extent_buffer(path->nodes[level]);
6300 path->nodes[level] = NULL;
6308 * drop a subvolume tree.
6310 * this function traverses the tree freeing any blocks that only
6311 * referenced by the tree.
6313 * when a shared tree block is found. this function decreases its
6314 * reference count by one. if update_ref is true, this function
6315 * also make sure backrefs for the shared block and all lower level
6316 * blocks are properly updated.
6318 void btrfs_drop_snapshot(struct btrfs_root *root,
6319 struct btrfs_block_rsv *block_rsv, int update_ref)
6321 struct btrfs_path *path;
6322 struct btrfs_trans_handle *trans;
6323 struct btrfs_root *tree_root = root->fs_info->tree_root;
6324 struct btrfs_root_item *root_item = &root->root_item;
6325 struct walk_control *wc;
6326 struct btrfs_key key;
6331 path = btrfs_alloc_path();
6337 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6339 btrfs_free_path(path);
6344 trans = btrfs_start_transaction(tree_root, 0);
6345 BUG_ON(IS_ERR(trans));
6348 trans->block_rsv = block_rsv;
6350 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6351 level = btrfs_header_level(root->node);
6352 path->nodes[level] = btrfs_lock_root_node(root);
6353 btrfs_set_lock_blocking(path->nodes[level]);
6354 path->slots[level] = 0;
6355 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6356 memset(&wc->update_progress, 0,
6357 sizeof(wc->update_progress));
6359 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6360 memcpy(&wc->update_progress, &key,
6361 sizeof(wc->update_progress));
6363 level = root_item->drop_level;
6365 path->lowest_level = level;
6366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6367 path->lowest_level = 0;
6375 * unlock our path, this is safe because only this
6376 * function is allowed to delete this snapshot
6378 btrfs_unlock_up_safe(path, 0);
6380 level = btrfs_header_level(root->node);
6382 btrfs_tree_lock(path->nodes[level]);
6383 btrfs_set_lock_blocking(path->nodes[level]);
6385 ret = btrfs_lookup_extent_info(trans, root,
6386 path->nodes[level]->start,
6387 path->nodes[level]->len,
6391 BUG_ON(wc->refs[level] == 0);
6393 if (level == root_item->drop_level)
6396 btrfs_tree_unlock(path->nodes[level]);
6397 WARN_ON(wc->refs[level] != 1);
6403 wc->shared_level = -1;
6404 wc->stage = DROP_REFERENCE;
6405 wc->update_ref = update_ref;
6407 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6410 ret = walk_down_tree(trans, root, path, wc);
6416 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6423 BUG_ON(wc->stage != DROP_REFERENCE);
6427 if (wc->stage == DROP_REFERENCE) {
6429 btrfs_node_key(path->nodes[level],
6430 &root_item->drop_progress,
6431 path->slots[level]);
6432 root_item->drop_level = level;
6435 BUG_ON(wc->level == 0);
6436 if (btrfs_should_end_transaction(trans, tree_root)) {
6437 ret = btrfs_update_root(trans, tree_root,
6442 btrfs_end_transaction_throttle(trans, tree_root);
6443 trans = btrfs_start_transaction(tree_root, 0);
6444 BUG_ON(IS_ERR(trans));
6446 trans->block_rsv = block_rsv;
6449 btrfs_release_path(path);
6452 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6455 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6456 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6460 /* if we fail to delete the orphan item this time
6461 * around, it'll get picked up the next time.
6463 * The most common failure here is just -ENOENT.
6465 btrfs_del_orphan_item(trans, tree_root,
6466 root->root_key.objectid);
6470 if (root->in_radix) {
6471 btrfs_free_fs_root(tree_root->fs_info, root);
6473 free_extent_buffer(root->node);
6474 free_extent_buffer(root->commit_root);
6478 btrfs_end_transaction_throttle(trans, tree_root);
6480 btrfs_free_path(path);
6483 btrfs_std_error(root->fs_info, err);
6488 * drop subtree rooted at tree block 'node'.
6490 * NOTE: this function will unlock and release tree block 'node'
6492 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6493 struct btrfs_root *root,
6494 struct extent_buffer *node,
6495 struct extent_buffer *parent)
6497 struct btrfs_path *path;
6498 struct walk_control *wc;
6504 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6506 path = btrfs_alloc_path();
6510 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6512 btrfs_free_path(path);
6516 btrfs_assert_tree_locked(parent);
6517 parent_level = btrfs_header_level(parent);
6518 extent_buffer_get(parent);
6519 path->nodes[parent_level] = parent;
6520 path->slots[parent_level] = btrfs_header_nritems(parent);
6522 btrfs_assert_tree_locked(node);
6523 level = btrfs_header_level(node);
6524 path->nodes[level] = node;
6525 path->slots[level] = 0;
6526 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6528 wc->refs[parent_level] = 1;
6529 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6531 wc->shared_level = -1;
6532 wc->stage = DROP_REFERENCE;
6535 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6538 wret = walk_down_tree(trans, root, path, wc);
6544 wret = walk_up_tree(trans, root, path, wc, parent_level);
6552 btrfs_free_path(path);
6556 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6559 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6560 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6563 * we add in the count of missing devices because we want
6564 * to make sure that any RAID levels on a degraded FS
6565 * continue to be honored.
6567 num_devices = root->fs_info->fs_devices->rw_devices +
6568 root->fs_info->fs_devices->missing_devices;
6570 if (num_devices == 1) {
6571 stripped |= BTRFS_BLOCK_GROUP_DUP;
6572 stripped = flags & ~stripped;
6574 /* turn raid0 into single device chunks */
6575 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6578 /* turn mirroring into duplication */
6579 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6580 BTRFS_BLOCK_GROUP_RAID10))
6581 return stripped | BTRFS_BLOCK_GROUP_DUP;
6584 /* they already had raid on here, just return */
6585 if (flags & stripped)
6588 stripped |= BTRFS_BLOCK_GROUP_DUP;
6589 stripped = flags & ~stripped;
6591 /* switch duplicated blocks with raid1 */
6592 if (flags & BTRFS_BLOCK_GROUP_DUP)
6593 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6595 /* turn single device chunks into raid0 */
6596 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6601 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6603 struct btrfs_space_info *sinfo = cache->space_info;
6605 u64 min_allocable_bytes;
6610 * We need some metadata space and system metadata space for
6611 * allocating chunks in some corner cases until we force to set
6612 * it to be readonly.
6615 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6617 min_allocable_bytes = 1 * 1024 * 1024;
6619 min_allocable_bytes = 0;
6621 spin_lock(&sinfo->lock);
6622 spin_lock(&cache->lock);
6629 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6630 cache->bytes_super - btrfs_block_group_used(&cache->item);
6632 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6633 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6634 min_allocable_bytes <= sinfo->total_bytes) {
6635 sinfo->bytes_readonly += num_bytes;
6640 spin_unlock(&cache->lock);
6641 spin_unlock(&sinfo->lock);
6645 int btrfs_set_block_group_ro(struct btrfs_root *root,
6646 struct btrfs_block_group_cache *cache)
6649 struct btrfs_trans_handle *trans;
6655 trans = btrfs_join_transaction(root);
6656 BUG_ON(IS_ERR(trans));
6658 alloc_flags = update_block_group_flags(root, cache->flags);
6659 if (alloc_flags != cache->flags)
6660 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6663 ret = set_block_group_ro(cache, 0);
6666 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6667 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6671 ret = set_block_group_ro(cache, 0);
6673 btrfs_end_transaction(trans, root);
6677 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6678 struct btrfs_root *root, u64 type)
6680 u64 alloc_flags = get_alloc_profile(root, type);
6681 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6686 * helper to account the unused space of all the readonly block group in the
6687 * list. takes mirrors into account.
6689 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6691 struct btrfs_block_group_cache *block_group;
6695 list_for_each_entry(block_group, groups_list, list) {
6696 spin_lock(&block_group->lock);
6698 if (!block_group->ro) {
6699 spin_unlock(&block_group->lock);
6703 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6704 BTRFS_BLOCK_GROUP_RAID10 |
6705 BTRFS_BLOCK_GROUP_DUP))
6710 free_bytes += (block_group->key.offset -
6711 btrfs_block_group_used(&block_group->item)) *
6714 spin_unlock(&block_group->lock);
6721 * helper to account the unused space of all the readonly block group in the
6722 * space_info. takes mirrors into account.
6724 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6729 spin_lock(&sinfo->lock);
6731 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6732 if (!list_empty(&sinfo->block_groups[i]))
6733 free_bytes += __btrfs_get_ro_block_group_free_space(
6734 &sinfo->block_groups[i]);
6736 spin_unlock(&sinfo->lock);
6741 int btrfs_set_block_group_rw(struct btrfs_root *root,
6742 struct btrfs_block_group_cache *cache)
6744 struct btrfs_space_info *sinfo = cache->space_info;
6749 spin_lock(&sinfo->lock);
6750 spin_lock(&cache->lock);
6751 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6752 cache->bytes_super - btrfs_block_group_used(&cache->item);
6753 sinfo->bytes_readonly -= num_bytes;
6755 spin_unlock(&cache->lock);
6756 spin_unlock(&sinfo->lock);
6761 * checks to see if its even possible to relocate this block group.
6763 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6764 * ok to go ahead and try.
6766 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6768 struct btrfs_block_group_cache *block_group;
6769 struct btrfs_space_info *space_info;
6770 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6771 struct btrfs_device *device;
6779 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6781 /* odd, couldn't find the block group, leave it alone */
6785 min_free = btrfs_block_group_used(&block_group->item);
6787 /* no bytes used, we're good */
6791 space_info = block_group->space_info;
6792 spin_lock(&space_info->lock);
6794 full = space_info->full;
6797 * if this is the last block group we have in this space, we can't
6798 * relocate it unless we're able to allocate a new chunk below.
6800 * Otherwise, we need to make sure we have room in the space to handle
6801 * all of the extents from this block group. If we can, we're good
6803 if ((space_info->total_bytes != block_group->key.offset) &&
6804 (space_info->bytes_used + space_info->bytes_reserved +
6805 space_info->bytes_pinned + space_info->bytes_readonly +
6806 min_free < space_info->total_bytes)) {
6807 spin_unlock(&space_info->lock);
6810 spin_unlock(&space_info->lock);
6813 * ok we don't have enough space, but maybe we have free space on our
6814 * devices to allocate new chunks for relocation, so loop through our
6815 * alloc devices and guess if we have enough space. However, if we
6816 * were marked as full, then we know there aren't enough chunks, and we
6831 index = get_block_group_index(block_group);
6836 } else if (index == 1) {
6838 } else if (index == 2) {
6841 } else if (index == 3) {
6842 dev_min = fs_devices->rw_devices;
6843 do_div(min_free, dev_min);
6846 mutex_lock(&root->fs_info->chunk_mutex);
6847 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6851 * check to make sure we can actually find a chunk with enough
6852 * space to fit our block group in.
6854 if (device->total_bytes > device->bytes_used + min_free) {
6855 ret = find_free_dev_extent(NULL, device, min_free,
6860 if (dev_nr >= dev_min)
6866 mutex_unlock(&root->fs_info->chunk_mutex);
6868 btrfs_put_block_group(block_group);
6872 static int find_first_block_group(struct btrfs_root *root,
6873 struct btrfs_path *path, struct btrfs_key *key)
6876 struct btrfs_key found_key;
6877 struct extent_buffer *leaf;
6880 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6885 slot = path->slots[0];
6886 leaf = path->nodes[0];
6887 if (slot >= btrfs_header_nritems(leaf)) {
6888 ret = btrfs_next_leaf(root, path);
6895 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6897 if (found_key.objectid >= key->objectid &&
6898 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6908 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6910 struct btrfs_block_group_cache *block_group;
6914 struct inode *inode;
6916 block_group = btrfs_lookup_first_block_group(info, last);
6917 while (block_group) {
6918 spin_lock(&block_group->lock);
6919 if (block_group->iref)
6921 spin_unlock(&block_group->lock);
6922 block_group = next_block_group(info->tree_root,
6932 inode = block_group->inode;
6933 block_group->iref = 0;
6934 block_group->inode = NULL;
6935 spin_unlock(&block_group->lock);
6937 last = block_group->key.objectid + block_group->key.offset;
6938 btrfs_put_block_group(block_group);
6942 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6944 struct btrfs_block_group_cache *block_group;
6945 struct btrfs_space_info *space_info;
6946 struct btrfs_caching_control *caching_ctl;
6949 down_write(&info->extent_commit_sem);
6950 while (!list_empty(&info->caching_block_groups)) {
6951 caching_ctl = list_entry(info->caching_block_groups.next,
6952 struct btrfs_caching_control, list);
6953 list_del(&caching_ctl->list);
6954 put_caching_control(caching_ctl);
6956 up_write(&info->extent_commit_sem);
6958 spin_lock(&info->block_group_cache_lock);
6959 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6960 block_group = rb_entry(n, struct btrfs_block_group_cache,
6962 rb_erase(&block_group->cache_node,
6963 &info->block_group_cache_tree);
6964 spin_unlock(&info->block_group_cache_lock);
6966 down_write(&block_group->space_info->groups_sem);
6967 list_del(&block_group->list);
6968 up_write(&block_group->space_info->groups_sem);
6970 if (block_group->cached == BTRFS_CACHE_STARTED)
6971 wait_block_group_cache_done(block_group);
6974 * We haven't cached this block group, which means we could
6975 * possibly have excluded extents on this block group.
6977 if (block_group->cached == BTRFS_CACHE_NO)
6978 free_excluded_extents(info->extent_root, block_group);
6980 btrfs_remove_free_space_cache(block_group);
6981 btrfs_put_block_group(block_group);
6983 spin_lock(&info->block_group_cache_lock);
6985 spin_unlock(&info->block_group_cache_lock);
6987 /* now that all the block groups are freed, go through and
6988 * free all the space_info structs. This is only called during
6989 * the final stages of unmount, and so we know nobody is
6990 * using them. We call synchronize_rcu() once before we start,
6991 * just to be on the safe side.
6995 release_global_block_rsv(info);
6997 while(!list_empty(&info->space_info)) {
6998 space_info = list_entry(info->space_info.next,
6999 struct btrfs_space_info,
7001 if (space_info->bytes_pinned > 0 ||
7002 space_info->bytes_reserved > 0 ||
7003 space_info->bytes_may_use > 0) {
7005 dump_space_info(space_info, 0, 0);
7007 list_del(&space_info->list);
7013 static void __link_block_group(struct btrfs_space_info *space_info,
7014 struct btrfs_block_group_cache *cache)
7016 int index = get_block_group_index(cache);
7018 down_write(&space_info->groups_sem);
7019 list_add_tail(&cache->list, &space_info->block_groups[index]);
7020 up_write(&space_info->groups_sem);
7023 int btrfs_read_block_groups(struct btrfs_root *root)
7025 struct btrfs_path *path;
7027 struct btrfs_block_group_cache *cache;
7028 struct btrfs_fs_info *info = root->fs_info;
7029 struct btrfs_space_info *space_info;
7030 struct btrfs_key key;
7031 struct btrfs_key found_key;
7032 struct extent_buffer *leaf;
7036 root = info->extent_root;
7039 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7040 path = btrfs_alloc_path();
7045 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7046 if (btrfs_test_opt(root, SPACE_CACHE) &&
7047 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7049 if (btrfs_test_opt(root, CLEAR_CACHE))
7053 ret = find_first_block_group(root, path, &key);
7058 leaf = path->nodes[0];
7059 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7060 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7065 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7067 if (!cache->free_space_ctl) {
7073 atomic_set(&cache->count, 1);
7074 spin_lock_init(&cache->lock);
7075 cache->fs_info = info;
7076 INIT_LIST_HEAD(&cache->list);
7077 INIT_LIST_HEAD(&cache->cluster_list);
7080 cache->disk_cache_state = BTRFS_DC_CLEAR;
7082 read_extent_buffer(leaf, &cache->item,
7083 btrfs_item_ptr_offset(leaf, path->slots[0]),
7084 sizeof(cache->item));
7085 memcpy(&cache->key, &found_key, sizeof(found_key));
7087 key.objectid = found_key.objectid + found_key.offset;
7088 btrfs_release_path(path);
7089 cache->flags = btrfs_block_group_flags(&cache->item);
7090 cache->sectorsize = root->sectorsize;
7092 btrfs_init_free_space_ctl(cache);
7095 * We need to exclude the super stripes now so that the space
7096 * info has super bytes accounted for, otherwise we'll think
7097 * we have more space than we actually do.
7099 exclude_super_stripes(root, cache);
7102 * check for two cases, either we are full, and therefore
7103 * don't need to bother with the caching work since we won't
7104 * find any space, or we are empty, and we can just add all
7105 * the space in and be done with it. This saves us _alot_ of
7106 * time, particularly in the full case.
7108 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7109 cache->last_byte_to_unpin = (u64)-1;
7110 cache->cached = BTRFS_CACHE_FINISHED;
7111 free_excluded_extents(root, cache);
7112 } else if (btrfs_block_group_used(&cache->item) == 0) {
7113 cache->last_byte_to_unpin = (u64)-1;
7114 cache->cached = BTRFS_CACHE_FINISHED;
7115 add_new_free_space(cache, root->fs_info,
7117 found_key.objectid +
7119 free_excluded_extents(root, cache);
7122 ret = update_space_info(info, cache->flags, found_key.offset,
7123 btrfs_block_group_used(&cache->item),
7126 cache->space_info = space_info;
7127 spin_lock(&cache->space_info->lock);
7128 cache->space_info->bytes_readonly += cache->bytes_super;
7129 spin_unlock(&cache->space_info->lock);
7131 __link_block_group(space_info, cache);
7133 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7136 set_avail_alloc_bits(root->fs_info, cache->flags);
7137 if (btrfs_chunk_readonly(root, cache->key.objectid))
7138 set_block_group_ro(cache, 1);
7141 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7142 if (!(get_alloc_profile(root, space_info->flags) &
7143 (BTRFS_BLOCK_GROUP_RAID10 |
7144 BTRFS_BLOCK_GROUP_RAID1 |
7145 BTRFS_BLOCK_GROUP_DUP)))
7148 * avoid allocating from un-mirrored block group if there are
7149 * mirrored block groups.
7151 list_for_each_entry(cache, &space_info->block_groups[3], list)
7152 set_block_group_ro(cache, 1);
7153 list_for_each_entry(cache, &space_info->block_groups[4], list)
7154 set_block_group_ro(cache, 1);
7157 init_global_block_rsv(info);
7160 btrfs_free_path(path);
7164 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7165 struct btrfs_root *root, u64 bytes_used,
7166 u64 type, u64 chunk_objectid, u64 chunk_offset,
7170 struct btrfs_root *extent_root;
7171 struct btrfs_block_group_cache *cache;
7173 extent_root = root->fs_info->extent_root;
7175 root->fs_info->last_trans_log_full_commit = trans->transid;
7177 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7180 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7182 if (!cache->free_space_ctl) {
7187 cache->key.objectid = chunk_offset;
7188 cache->key.offset = size;
7189 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7190 cache->sectorsize = root->sectorsize;
7191 cache->fs_info = root->fs_info;
7193 atomic_set(&cache->count, 1);
7194 spin_lock_init(&cache->lock);
7195 INIT_LIST_HEAD(&cache->list);
7196 INIT_LIST_HEAD(&cache->cluster_list);
7198 btrfs_init_free_space_ctl(cache);
7200 btrfs_set_block_group_used(&cache->item, bytes_used);
7201 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7202 cache->flags = type;
7203 btrfs_set_block_group_flags(&cache->item, type);
7205 cache->last_byte_to_unpin = (u64)-1;
7206 cache->cached = BTRFS_CACHE_FINISHED;
7207 exclude_super_stripes(root, cache);
7209 add_new_free_space(cache, root->fs_info, chunk_offset,
7210 chunk_offset + size);
7212 free_excluded_extents(root, cache);
7214 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7215 &cache->space_info);
7218 spin_lock(&cache->space_info->lock);
7219 cache->space_info->bytes_readonly += cache->bytes_super;
7220 spin_unlock(&cache->space_info->lock);
7222 __link_block_group(cache->space_info, cache);
7224 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7227 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7228 sizeof(cache->item));
7231 set_avail_alloc_bits(extent_root->fs_info, type);
7236 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7237 struct btrfs_root *root, u64 group_start)
7239 struct btrfs_path *path;
7240 struct btrfs_block_group_cache *block_group;
7241 struct btrfs_free_cluster *cluster;
7242 struct btrfs_root *tree_root = root->fs_info->tree_root;
7243 struct btrfs_key key;
7244 struct inode *inode;
7248 root = root->fs_info->extent_root;
7250 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7251 BUG_ON(!block_group);
7252 BUG_ON(!block_group->ro);
7255 * Free the reserved super bytes from this block group before
7258 free_excluded_extents(root, block_group);
7260 memcpy(&key, &block_group->key, sizeof(key));
7261 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7262 BTRFS_BLOCK_GROUP_RAID1 |
7263 BTRFS_BLOCK_GROUP_RAID10))
7268 /* make sure this block group isn't part of an allocation cluster */
7269 cluster = &root->fs_info->data_alloc_cluster;
7270 spin_lock(&cluster->refill_lock);
7271 btrfs_return_cluster_to_free_space(block_group, cluster);
7272 spin_unlock(&cluster->refill_lock);
7275 * make sure this block group isn't part of a metadata
7276 * allocation cluster
7278 cluster = &root->fs_info->meta_alloc_cluster;
7279 spin_lock(&cluster->refill_lock);
7280 btrfs_return_cluster_to_free_space(block_group, cluster);
7281 spin_unlock(&cluster->refill_lock);
7283 path = btrfs_alloc_path();
7289 inode = lookup_free_space_inode(root, block_group, path);
7290 if (!IS_ERR(inode)) {
7291 ret = btrfs_orphan_add(trans, inode);
7294 /* One for the block groups ref */
7295 spin_lock(&block_group->lock);
7296 if (block_group->iref) {
7297 block_group->iref = 0;
7298 block_group->inode = NULL;
7299 spin_unlock(&block_group->lock);
7302 spin_unlock(&block_group->lock);
7304 /* One for our lookup ref */
7305 btrfs_add_delayed_iput(inode);
7308 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7309 key.offset = block_group->key.objectid;
7312 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7316 btrfs_release_path(path);
7318 ret = btrfs_del_item(trans, tree_root, path);
7321 btrfs_release_path(path);
7324 spin_lock(&root->fs_info->block_group_cache_lock);
7325 rb_erase(&block_group->cache_node,
7326 &root->fs_info->block_group_cache_tree);
7327 spin_unlock(&root->fs_info->block_group_cache_lock);
7329 down_write(&block_group->space_info->groups_sem);
7331 * we must use list_del_init so people can check to see if they
7332 * are still on the list after taking the semaphore
7334 list_del_init(&block_group->list);
7335 up_write(&block_group->space_info->groups_sem);
7337 if (block_group->cached == BTRFS_CACHE_STARTED)
7338 wait_block_group_cache_done(block_group);
7340 btrfs_remove_free_space_cache(block_group);
7342 spin_lock(&block_group->space_info->lock);
7343 block_group->space_info->total_bytes -= block_group->key.offset;
7344 block_group->space_info->bytes_readonly -= block_group->key.offset;
7345 block_group->space_info->disk_total -= block_group->key.offset * factor;
7346 spin_unlock(&block_group->space_info->lock);
7348 memcpy(&key, &block_group->key, sizeof(key));
7350 btrfs_clear_space_info_full(root->fs_info);
7352 btrfs_put_block_group(block_group);
7353 btrfs_put_block_group(block_group);
7355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7361 ret = btrfs_del_item(trans, root, path);
7363 btrfs_free_path(path);
7367 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7369 struct btrfs_space_info *space_info;
7370 struct btrfs_super_block *disk_super;
7376 disk_super = &fs_info->super_copy;
7377 if (!btrfs_super_root(disk_super))
7380 features = btrfs_super_incompat_flags(disk_super);
7381 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7384 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7385 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7390 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7391 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7393 flags = BTRFS_BLOCK_GROUP_METADATA;
7394 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7398 flags = BTRFS_BLOCK_GROUP_DATA;
7399 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7405 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7407 return unpin_extent_range(root, start, end);
7410 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7411 u64 num_bytes, u64 *actual_bytes)
7413 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7416 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7418 struct btrfs_fs_info *fs_info = root->fs_info;
7419 struct btrfs_block_group_cache *cache = NULL;
7426 cache = btrfs_lookup_block_group(fs_info, range->start);
7429 if (cache->key.objectid >= (range->start + range->len)) {
7430 btrfs_put_block_group(cache);
7434 start = max(range->start, cache->key.objectid);
7435 end = min(range->start + range->len,
7436 cache->key.objectid + cache->key.offset);
7438 if (end - start >= range->minlen) {
7439 if (!block_group_cache_done(cache)) {
7440 ret = cache_block_group(cache, NULL, root, 0);
7442 wait_block_group_cache_done(cache);
7444 ret = btrfs_trim_block_group(cache,
7450 trimmed += group_trimmed;
7452 btrfs_put_block_group(cache);
7457 cache = next_block_group(fs_info->tree_root, cache);
7460 range->len = trimmed;