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 spin_lock(&cache->lock);
486 if (cache->cached != BTRFS_CACHE_NO) {
487 spin_unlock(&cache->lock);
490 cache->cached = BTRFS_CACHE_STARTED;
491 spin_unlock(&cache->lock);
493 ret = load_free_space_cache(fs_info, cache);
495 spin_lock(&cache->lock);
497 cache->cached = BTRFS_CACHE_FINISHED;
498 cache->last_byte_to_unpin = (u64)-1;
500 cache->cached = BTRFS_CACHE_NO;
502 spin_unlock(&cache->lock);
504 free_excluded_extents(fs_info->extent_root, cache);
512 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
513 BUG_ON(!caching_ctl);
515 INIT_LIST_HEAD(&caching_ctl->list);
516 mutex_init(&caching_ctl->mutex);
517 init_waitqueue_head(&caching_ctl->wait);
518 caching_ctl->block_group = cache;
519 caching_ctl->progress = cache->key.objectid;
520 /* one for caching kthread, one for caching block group list */
521 atomic_set(&caching_ctl->count, 2);
522 caching_ctl->work.func = caching_thread;
524 spin_lock(&cache->lock);
525 if (cache->cached != BTRFS_CACHE_NO) {
526 spin_unlock(&cache->lock);
530 cache->caching_ctl = caching_ctl;
531 cache->cached = BTRFS_CACHE_STARTED;
532 spin_unlock(&cache->lock);
534 down_write(&fs_info->extent_commit_sem);
535 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
536 up_write(&fs_info->extent_commit_sem);
538 btrfs_get_block_group(cache);
540 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
546 * return the block group that starts at or after bytenr
548 static struct btrfs_block_group_cache *
549 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
551 struct btrfs_block_group_cache *cache;
553 cache = block_group_cache_tree_search(info, bytenr, 0);
559 * return the block group that contains the given bytenr
561 struct btrfs_block_group_cache *btrfs_lookup_block_group(
562 struct btrfs_fs_info *info,
565 struct btrfs_block_group_cache *cache;
567 cache = block_group_cache_tree_search(info, bytenr, 1);
572 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
575 struct list_head *head = &info->space_info;
576 struct btrfs_space_info *found;
578 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
579 BTRFS_BLOCK_GROUP_METADATA;
582 list_for_each_entry_rcu(found, head, list) {
583 if (found->flags & flags) {
593 * after adding space to the filesystem, we need to clear the full flags
594 * on all the space infos.
596 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
598 struct list_head *head = &info->space_info;
599 struct btrfs_space_info *found;
602 list_for_each_entry_rcu(found, head, list)
607 static u64 div_factor(u64 num, int factor)
616 static u64 div_factor_fine(u64 num, int factor)
625 u64 btrfs_find_block_group(struct btrfs_root *root,
626 u64 search_start, u64 search_hint, int owner)
628 struct btrfs_block_group_cache *cache;
630 u64 last = max(search_hint, search_start);
637 cache = btrfs_lookup_first_block_group(root->fs_info, last);
641 spin_lock(&cache->lock);
642 last = cache->key.objectid + cache->key.offset;
643 used = btrfs_block_group_used(&cache->item);
645 if ((full_search || !cache->ro) &&
646 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
647 if (used + cache->pinned + cache->reserved <
648 div_factor(cache->key.offset, factor)) {
649 group_start = cache->key.objectid;
650 spin_unlock(&cache->lock);
651 btrfs_put_block_group(cache);
655 spin_unlock(&cache->lock);
656 btrfs_put_block_group(cache);
664 if (!full_search && factor < 10) {
674 /* simple helper to search for an existing extent at a given offset */
675 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
678 struct btrfs_key key;
679 struct btrfs_path *path;
681 path = btrfs_alloc_path();
685 key.objectid = start;
687 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
688 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
690 btrfs_free_path(path);
695 * helper function to lookup reference count and flags of extent.
697 * the head node for delayed ref is used to store the sum of all the
698 * reference count modifications queued up in the rbtree. the head
699 * node may also store the extent flags to set. This way you can check
700 * to see what the reference count and extent flags would be if all of
701 * the delayed refs are not processed.
703 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
704 struct btrfs_root *root, u64 bytenr,
705 u64 num_bytes, u64 *refs, u64 *flags)
707 struct btrfs_delayed_ref_head *head;
708 struct btrfs_delayed_ref_root *delayed_refs;
709 struct btrfs_path *path;
710 struct btrfs_extent_item *ei;
711 struct extent_buffer *leaf;
712 struct btrfs_key key;
718 path = btrfs_alloc_path();
722 key.objectid = bytenr;
723 key.type = BTRFS_EXTENT_ITEM_KEY;
724 key.offset = num_bytes;
726 path->skip_locking = 1;
727 path->search_commit_root = 1;
730 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
736 leaf = path->nodes[0];
737 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
738 if (item_size >= sizeof(*ei)) {
739 ei = btrfs_item_ptr(leaf, path->slots[0],
740 struct btrfs_extent_item);
741 num_refs = btrfs_extent_refs(leaf, ei);
742 extent_flags = btrfs_extent_flags(leaf, ei);
744 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
745 struct btrfs_extent_item_v0 *ei0;
746 BUG_ON(item_size != sizeof(*ei0));
747 ei0 = btrfs_item_ptr(leaf, path->slots[0],
748 struct btrfs_extent_item_v0);
749 num_refs = btrfs_extent_refs_v0(leaf, ei0);
750 /* FIXME: this isn't correct for data */
751 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
756 BUG_ON(num_refs == 0);
766 delayed_refs = &trans->transaction->delayed_refs;
767 spin_lock(&delayed_refs->lock);
768 head = btrfs_find_delayed_ref_head(trans, bytenr);
770 if (!mutex_trylock(&head->mutex)) {
771 atomic_inc(&head->node.refs);
772 spin_unlock(&delayed_refs->lock);
774 btrfs_release_path(path);
777 * Mutex was contended, block until it's released and try
780 mutex_lock(&head->mutex);
781 mutex_unlock(&head->mutex);
782 btrfs_put_delayed_ref(&head->node);
785 if (head->extent_op && head->extent_op->update_flags)
786 extent_flags |= head->extent_op->flags_to_set;
788 BUG_ON(num_refs == 0);
790 num_refs += head->node.ref_mod;
791 mutex_unlock(&head->mutex);
793 spin_unlock(&delayed_refs->lock);
795 WARN_ON(num_refs == 0);
799 *flags = extent_flags;
801 btrfs_free_path(path);
806 * Back reference rules. Back refs have three main goals:
808 * 1) differentiate between all holders of references to an extent so that
809 * when a reference is dropped we can make sure it was a valid reference
810 * before freeing the extent.
812 * 2) Provide enough information to quickly find the holders of an extent
813 * if we notice a given block is corrupted or bad.
815 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
816 * maintenance. This is actually the same as #2, but with a slightly
817 * different use case.
819 * There are two kinds of back refs. The implicit back refs is optimized
820 * for pointers in non-shared tree blocks. For a given pointer in a block,
821 * back refs of this kind provide information about the block's owner tree
822 * and the pointer's key. These information allow us to find the block by
823 * b-tree searching. The full back refs is for pointers in tree blocks not
824 * referenced by their owner trees. The location of tree block is recorded
825 * in the back refs. Actually the full back refs is generic, and can be
826 * used in all cases the implicit back refs is used. The major shortcoming
827 * of the full back refs is its overhead. Every time a tree block gets
828 * COWed, we have to update back refs entry for all pointers in it.
830 * For a newly allocated tree block, we use implicit back refs for
831 * pointers in it. This means most tree related operations only involve
832 * implicit back refs. For a tree block created in old transaction, the
833 * only way to drop a reference to it is COW it. So we can detect the
834 * event that tree block loses its owner tree's reference and do the
835 * back refs conversion.
837 * When a tree block is COW'd through a tree, there are four cases:
839 * The reference count of the block is one and the tree is the block's
840 * owner tree. Nothing to do in this case.
842 * The reference count of the block is one and the tree is not the
843 * block's owner tree. In this case, full back refs is used for pointers
844 * in the block. Remove these full back refs, add implicit back refs for
845 * every pointers in the new block.
847 * The reference count of the block is greater than one and the tree is
848 * the block's owner tree. In this case, implicit back refs is used for
849 * pointers in the block. Add full back refs for every pointers in the
850 * block, increase lower level extents' reference counts. The original
851 * implicit back refs are entailed to the new block.
853 * The reference count of the block is greater than one and the tree is
854 * not the block's owner tree. Add implicit back refs for every pointer in
855 * the new block, increase lower level extents' reference count.
857 * Back Reference Key composing:
859 * The key objectid corresponds to the first byte in the extent,
860 * The key type is used to differentiate between types of back refs.
861 * There are different meanings of the key offset for different types
864 * File extents can be referenced by:
866 * - multiple snapshots, subvolumes, or different generations in one subvol
867 * - different files inside a single subvolume
868 * - different offsets inside a file (bookend extents in file.c)
870 * The extent ref structure for the implicit back refs has fields for:
872 * - Objectid of the subvolume root
873 * - objectid of the file holding the reference
874 * - original offset in the file
875 * - how many bookend extents
877 * The key offset for the implicit back refs is hash of the first
880 * The extent ref structure for the full back refs has field for:
882 * - number of pointers in the tree leaf
884 * The key offset for the implicit back refs is the first byte of
887 * When a file extent is allocated, The implicit back refs is used.
888 * the fields are filled in:
890 * (root_key.objectid, inode objectid, offset in file, 1)
892 * When a file extent is removed file truncation, we find the
893 * corresponding implicit back refs and check the following fields:
895 * (btrfs_header_owner(leaf), inode objectid, offset in file)
897 * Btree extents can be referenced by:
899 * - Different subvolumes
901 * Both the implicit back refs and the full back refs for tree blocks
902 * only consist of key. The key offset for the implicit back refs is
903 * objectid of block's owner tree. The key offset for the full back refs
904 * is the first byte of parent block.
906 * When implicit back refs is used, information about the lowest key and
907 * level of the tree block are required. These information are stored in
908 * tree block info structure.
911 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
912 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
913 struct btrfs_root *root,
914 struct btrfs_path *path,
915 u64 owner, u32 extra_size)
917 struct btrfs_extent_item *item;
918 struct btrfs_extent_item_v0 *ei0;
919 struct btrfs_extent_ref_v0 *ref0;
920 struct btrfs_tree_block_info *bi;
921 struct extent_buffer *leaf;
922 struct btrfs_key key;
923 struct btrfs_key found_key;
924 u32 new_size = sizeof(*item);
928 leaf = path->nodes[0];
929 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
931 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
932 ei0 = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_extent_item_v0);
934 refs = btrfs_extent_refs_v0(leaf, ei0);
936 if (owner == (u64)-1) {
938 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
939 ret = btrfs_next_leaf(root, path);
943 leaf = path->nodes[0];
945 btrfs_item_key_to_cpu(leaf, &found_key,
947 BUG_ON(key.objectid != found_key.objectid);
948 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
952 ref0 = btrfs_item_ptr(leaf, path->slots[0],
953 struct btrfs_extent_ref_v0);
954 owner = btrfs_ref_objectid_v0(leaf, ref0);
958 btrfs_release_path(path);
960 if (owner < BTRFS_FIRST_FREE_OBJECTID)
961 new_size += sizeof(*bi);
963 new_size -= sizeof(*ei0);
964 ret = btrfs_search_slot(trans, root, &key, path,
965 new_size + extra_size, 1);
970 ret = btrfs_extend_item(trans, root, path, new_size);
972 leaf = path->nodes[0];
973 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
974 btrfs_set_extent_refs(leaf, item, refs);
975 /* FIXME: get real generation */
976 btrfs_set_extent_generation(leaf, item, 0);
977 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
978 btrfs_set_extent_flags(leaf, item,
979 BTRFS_EXTENT_FLAG_TREE_BLOCK |
980 BTRFS_BLOCK_FLAG_FULL_BACKREF);
981 bi = (struct btrfs_tree_block_info *)(item + 1);
982 /* FIXME: get first key of the block */
983 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
984 btrfs_set_tree_block_level(leaf, bi, (int)owner);
986 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
988 btrfs_mark_buffer_dirty(leaf);
993 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
995 u32 high_crc = ~(u32)0;
996 u32 low_crc = ~(u32)0;
999 lenum = cpu_to_le64(root_objectid);
1000 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1001 lenum = cpu_to_le64(owner);
1002 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1003 lenum = cpu_to_le64(offset);
1004 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1006 return ((u64)high_crc << 31) ^ (u64)low_crc;
1009 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1010 struct btrfs_extent_data_ref *ref)
1012 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1013 btrfs_extent_data_ref_objectid(leaf, ref),
1014 btrfs_extent_data_ref_offset(leaf, ref));
1017 static int match_extent_data_ref(struct extent_buffer *leaf,
1018 struct btrfs_extent_data_ref *ref,
1019 u64 root_objectid, u64 owner, u64 offset)
1021 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1022 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1023 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1028 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1029 struct btrfs_root *root,
1030 struct btrfs_path *path,
1031 u64 bytenr, u64 parent,
1033 u64 owner, u64 offset)
1035 struct btrfs_key key;
1036 struct btrfs_extent_data_ref *ref;
1037 struct extent_buffer *leaf;
1043 key.objectid = bytenr;
1045 key.type = BTRFS_SHARED_DATA_REF_KEY;
1046 key.offset = parent;
1048 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1049 key.offset = hash_extent_data_ref(root_objectid,
1054 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1063 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1064 key.type = BTRFS_EXTENT_REF_V0_KEY;
1065 btrfs_release_path(path);
1066 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1077 leaf = path->nodes[0];
1078 nritems = btrfs_header_nritems(leaf);
1080 if (path->slots[0] >= nritems) {
1081 ret = btrfs_next_leaf(root, path);
1087 leaf = path->nodes[0];
1088 nritems = btrfs_header_nritems(leaf);
1092 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1093 if (key.objectid != bytenr ||
1094 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1097 ref = btrfs_item_ptr(leaf, path->slots[0],
1098 struct btrfs_extent_data_ref);
1100 if (match_extent_data_ref(leaf, ref, root_objectid,
1103 btrfs_release_path(path);
1115 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1116 struct btrfs_root *root,
1117 struct btrfs_path *path,
1118 u64 bytenr, u64 parent,
1119 u64 root_objectid, u64 owner,
1120 u64 offset, int refs_to_add)
1122 struct btrfs_key key;
1123 struct extent_buffer *leaf;
1128 key.objectid = bytenr;
1130 key.type = BTRFS_SHARED_DATA_REF_KEY;
1131 key.offset = parent;
1132 size = sizeof(struct btrfs_shared_data_ref);
1134 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1135 key.offset = hash_extent_data_ref(root_objectid,
1137 size = sizeof(struct btrfs_extent_data_ref);
1140 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1141 if (ret && ret != -EEXIST)
1144 leaf = path->nodes[0];
1146 struct btrfs_shared_data_ref *ref;
1147 ref = btrfs_item_ptr(leaf, path->slots[0],
1148 struct btrfs_shared_data_ref);
1150 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1152 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1153 num_refs += refs_to_add;
1154 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1157 struct btrfs_extent_data_ref *ref;
1158 while (ret == -EEXIST) {
1159 ref = btrfs_item_ptr(leaf, path->slots[0],
1160 struct btrfs_extent_data_ref);
1161 if (match_extent_data_ref(leaf, ref, root_objectid,
1164 btrfs_release_path(path);
1166 ret = btrfs_insert_empty_item(trans, root, path, &key,
1168 if (ret && ret != -EEXIST)
1171 leaf = path->nodes[0];
1173 ref = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_extent_data_ref);
1176 btrfs_set_extent_data_ref_root(leaf, ref,
1178 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1179 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1180 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1182 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1183 num_refs += refs_to_add;
1184 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1187 btrfs_mark_buffer_dirty(leaf);
1190 btrfs_release_path(path);
1194 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root,
1196 struct btrfs_path *path,
1199 struct btrfs_key key;
1200 struct btrfs_extent_data_ref *ref1 = NULL;
1201 struct btrfs_shared_data_ref *ref2 = NULL;
1202 struct extent_buffer *leaf;
1206 leaf = path->nodes[0];
1207 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1209 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1210 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1211 struct btrfs_extent_data_ref);
1212 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1213 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1214 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1216 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1217 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1218 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1219 struct btrfs_extent_ref_v0 *ref0;
1220 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1221 struct btrfs_extent_ref_v0);
1222 num_refs = btrfs_ref_count_v0(leaf, ref0);
1228 BUG_ON(num_refs < refs_to_drop);
1229 num_refs -= refs_to_drop;
1231 if (num_refs == 0) {
1232 ret = btrfs_del_item(trans, root, path);
1234 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1235 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1236 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1237 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1238 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1240 struct btrfs_extent_ref_v0 *ref0;
1241 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_ref_v0);
1243 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1246 btrfs_mark_buffer_dirty(leaf);
1251 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1252 struct btrfs_path *path,
1253 struct btrfs_extent_inline_ref *iref)
1255 struct btrfs_key key;
1256 struct extent_buffer *leaf;
1257 struct btrfs_extent_data_ref *ref1;
1258 struct btrfs_shared_data_ref *ref2;
1261 leaf = path->nodes[0];
1262 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1264 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1265 BTRFS_EXTENT_DATA_REF_KEY) {
1266 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1267 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1269 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1270 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1272 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1273 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1274 struct btrfs_extent_data_ref);
1275 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1276 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1277 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_shared_data_ref);
1279 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1282 struct btrfs_extent_ref_v0 *ref0;
1283 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_ref_v0);
1285 num_refs = btrfs_ref_count_v0(leaf, ref0);
1293 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1294 struct btrfs_root *root,
1295 struct btrfs_path *path,
1296 u64 bytenr, u64 parent,
1299 struct btrfs_key key;
1302 key.objectid = bytenr;
1304 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1305 key.offset = parent;
1307 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1308 key.offset = root_objectid;
1311 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1315 if (ret == -ENOENT && parent) {
1316 btrfs_release_path(path);
1317 key.type = BTRFS_EXTENT_REF_V0_KEY;
1318 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1326 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1327 struct btrfs_root *root,
1328 struct btrfs_path *path,
1329 u64 bytenr, u64 parent,
1332 struct btrfs_key key;
1335 key.objectid = bytenr;
1337 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1338 key.offset = parent;
1340 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1341 key.offset = root_objectid;
1344 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1345 btrfs_release_path(path);
1349 static inline int extent_ref_type(u64 parent, u64 owner)
1352 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1354 type = BTRFS_SHARED_BLOCK_REF_KEY;
1356 type = BTRFS_TREE_BLOCK_REF_KEY;
1359 type = BTRFS_SHARED_DATA_REF_KEY;
1361 type = BTRFS_EXTENT_DATA_REF_KEY;
1366 static int find_next_key(struct btrfs_path *path, int level,
1367 struct btrfs_key *key)
1370 for (; level < BTRFS_MAX_LEVEL; level++) {
1371 if (!path->nodes[level])
1373 if (path->slots[level] + 1 >=
1374 btrfs_header_nritems(path->nodes[level]))
1377 btrfs_item_key_to_cpu(path->nodes[level], key,
1378 path->slots[level] + 1);
1380 btrfs_node_key_to_cpu(path->nodes[level], key,
1381 path->slots[level] + 1);
1388 * look for inline back ref. if back ref is found, *ref_ret is set
1389 * to the address of inline back ref, and 0 is returned.
1391 * if back ref isn't found, *ref_ret is set to the address where it
1392 * should be inserted, and -ENOENT is returned.
1394 * if insert is true and there are too many inline back refs, the path
1395 * points to the extent item, and -EAGAIN is returned.
1397 * NOTE: inline back refs are ordered in the same way that back ref
1398 * items in the tree are ordered.
1400 static noinline_for_stack
1401 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 struct btrfs_extent_inline_ref **ref_ret,
1405 u64 bytenr, u64 num_bytes,
1406 u64 parent, u64 root_objectid,
1407 u64 owner, u64 offset, int insert)
1409 struct btrfs_key key;
1410 struct extent_buffer *leaf;
1411 struct btrfs_extent_item *ei;
1412 struct btrfs_extent_inline_ref *iref;
1423 key.objectid = bytenr;
1424 key.type = BTRFS_EXTENT_ITEM_KEY;
1425 key.offset = num_bytes;
1427 want = extent_ref_type(parent, owner);
1429 extra_size = btrfs_extent_inline_ref_size(want);
1430 path->keep_locks = 1;
1433 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1440 leaf = path->nodes[0];
1441 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1442 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1443 if (item_size < sizeof(*ei)) {
1448 ret = convert_extent_item_v0(trans, root, path, owner,
1454 leaf = path->nodes[0];
1455 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1458 BUG_ON(item_size < sizeof(*ei));
1460 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1461 flags = btrfs_extent_flags(leaf, ei);
1463 ptr = (unsigned long)(ei + 1);
1464 end = (unsigned long)ei + item_size;
1466 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1467 ptr += sizeof(struct btrfs_tree_block_info);
1470 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1479 iref = (struct btrfs_extent_inline_ref *)ptr;
1480 type = btrfs_extent_inline_ref_type(leaf, iref);
1484 ptr += btrfs_extent_inline_ref_size(type);
1488 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1489 struct btrfs_extent_data_ref *dref;
1490 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1491 if (match_extent_data_ref(leaf, dref, root_objectid,
1496 if (hash_extent_data_ref_item(leaf, dref) <
1497 hash_extent_data_ref(root_objectid, owner, offset))
1501 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1503 if (parent == ref_offset) {
1507 if (ref_offset < parent)
1510 if (root_objectid == ref_offset) {
1514 if (ref_offset < root_objectid)
1518 ptr += btrfs_extent_inline_ref_size(type);
1520 if (err == -ENOENT && insert) {
1521 if (item_size + extra_size >=
1522 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1527 * To add new inline back ref, we have to make sure
1528 * there is no corresponding back ref item.
1529 * For simplicity, we just do not add new inline back
1530 * ref if there is any kind of item for this block
1532 if (find_next_key(path, 0, &key) == 0 &&
1533 key.objectid == bytenr &&
1534 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1539 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1542 path->keep_locks = 0;
1543 btrfs_unlock_up_safe(path, 1);
1549 * helper to add new inline back ref
1551 static noinline_for_stack
1552 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1553 struct btrfs_root *root,
1554 struct btrfs_path *path,
1555 struct btrfs_extent_inline_ref *iref,
1556 u64 parent, u64 root_objectid,
1557 u64 owner, u64 offset, int refs_to_add,
1558 struct btrfs_delayed_extent_op *extent_op)
1560 struct extent_buffer *leaf;
1561 struct btrfs_extent_item *ei;
1564 unsigned long item_offset;
1570 leaf = path->nodes[0];
1571 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1572 item_offset = (unsigned long)iref - (unsigned long)ei;
1574 type = extent_ref_type(parent, owner);
1575 size = btrfs_extent_inline_ref_size(type);
1577 ret = btrfs_extend_item(trans, root, path, size);
1579 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1580 refs = btrfs_extent_refs(leaf, ei);
1581 refs += refs_to_add;
1582 btrfs_set_extent_refs(leaf, ei, refs);
1584 __run_delayed_extent_op(extent_op, leaf, ei);
1586 ptr = (unsigned long)ei + item_offset;
1587 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1588 if (ptr < end - size)
1589 memmove_extent_buffer(leaf, ptr + size, ptr,
1592 iref = (struct btrfs_extent_inline_ref *)ptr;
1593 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1594 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1595 struct btrfs_extent_data_ref *dref;
1596 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1597 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1598 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1599 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1600 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1601 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1602 struct btrfs_shared_data_ref *sref;
1603 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1604 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1605 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1606 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1607 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1609 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1611 btrfs_mark_buffer_dirty(leaf);
1615 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1616 struct btrfs_root *root,
1617 struct btrfs_path *path,
1618 struct btrfs_extent_inline_ref **ref_ret,
1619 u64 bytenr, u64 num_bytes, u64 parent,
1620 u64 root_objectid, u64 owner, u64 offset)
1624 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1625 bytenr, num_bytes, parent,
1626 root_objectid, owner, offset, 0);
1630 btrfs_release_path(path);
1633 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1634 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1637 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1638 root_objectid, owner, offset);
1644 * helper to update/remove inline back ref
1646 static noinline_for_stack
1647 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1648 struct btrfs_root *root,
1649 struct btrfs_path *path,
1650 struct btrfs_extent_inline_ref *iref,
1652 struct btrfs_delayed_extent_op *extent_op)
1654 struct extent_buffer *leaf;
1655 struct btrfs_extent_item *ei;
1656 struct btrfs_extent_data_ref *dref = NULL;
1657 struct btrfs_shared_data_ref *sref = NULL;
1666 leaf = path->nodes[0];
1667 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1668 refs = btrfs_extent_refs(leaf, ei);
1669 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1670 refs += refs_to_mod;
1671 btrfs_set_extent_refs(leaf, ei, refs);
1673 __run_delayed_extent_op(extent_op, leaf, ei);
1675 type = btrfs_extent_inline_ref_type(leaf, iref);
1677 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1678 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1679 refs = btrfs_extent_data_ref_count(leaf, dref);
1680 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1681 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1682 refs = btrfs_shared_data_ref_count(leaf, sref);
1685 BUG_ON(refs_to_mod != -1);
1688 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1689 refs += refs_to_mod;
1692 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1693 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1695 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1697 size = btrfs_extent_inline_ref_size(type);
1698 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1699 ptr = (unsigned long)iref;
1700 end = (unsigned long)ei + item_size;
1701 if (ptr + size < end)
1702 memmove_extent_buffer(leaf, ptr, ptr + size,
1705 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1707 btrfs_mark_buffer_dirty(leaf);
1711 static noinline_for_stack
1712 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1713 struct btrfs_root *root,
1714 struct btrfs_path *path,
1715 u64 bytenr, u64 num_bytes, u64 parent,
1716 u64 root_objectid, u64 owner,
1717 u64 offset, int refs_to_add,
1718 struct btrfs_delayed_extent_op *extent_op)
1720 struct btrfs_extent_inline_ref *iref;
1723 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1724 bytenr, num_bytes, parent,
1725 root_objectid, owner, offset, 1);
1727 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1728 ret = update_inline_extent_backref(trans, root, path, iref,
1729 refs_to_add, extent_op);
1730 } else if (ret == -ENOENT) {
1731 ret = setup_inline_extent_backref(trans, root, path, iref,
1732 parent, root_objectid,
1733 owner, offset, refs_to_add,
1739 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1740 struct btrfs_root *root,
1741 struct btrfs_path *path,
1742 u64 bytenr, u64 parent, u64 root_objectid,
1743 u64 owner, u64 offset, int refs_to_add)
1746 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1747 BUG_ON(refs_to_add != 1);
1748 ret = insert_tree_block_ref(trans, root, path, bytenr,
1749 parent, root_objectid);
1751 ret = insert_extent_data_ref(trans, root, path, bytenr,
1752 parent, root_objectid,
1753 owner, offset, refs_to_add);
1758 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1759 struct btrfs_root *root,
1760 struct btrfs_path *path,
1761 struct btrfs_extent_inline_ref *iref,
1762 int refs_to_drop, int is_data)
1766 BUG_ON(!is_data && refs_to_drop != 1);
1768 ret = update_inline_extent_backref(trans, root, path, iref,
1769 -refs_to_drop, NULL);
1770 } else if (is_data) {
1771 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1773 ret = btrfs_del_item(trans, root, path);
1778 static int btrfs_issue_discard(struct block_device *bdev,
1781 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1784 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1785 u64 num_bytes, u64 *actual_bytes)
1788 u64 discarded_bytes = 0;
1789 struct btrfs_multi_bio *multi = NULL;
1792 /* Tell the block device(s) that the sectors can be discarded */
1793 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1794 bytenr, &num_bytes, &multi, 0);
1796 struct btrfs_bio_stripe *stripe = multi->stripes;
1800 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1801 if (!stripe->dev->can_discard)
1804 ret = btrfs_issue_discard(stripe->dev->bdev,
1808 discarded_bytes += stripe->length;
1809 else if (ret != -EOPNOTSUPP)
1813 * Just in case we get back EOPNOTSUPP for some reason,
1814 * just ignore the return value so we don't screw up
1815 * people calling discard_extent.
1823 *actual_bytes = discarded_bytes;
1829 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1830 struct btrfs_root *root,
1831 u64 bytenr, u64 num_bytes, u64 parent,
1832 u64 root_objectid, u64 owner, u64 offset)
1835 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1836 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1838 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1839 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1840 parent, root_objectid, (int)owner,
1841 BTRFS_ADD_DELAYED_REF, NULL);
1843 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1844 parent, root_objectid, owner, offset,
1845 BTRFS_ADD_DELAYED_REF, NULL);
1850 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1851 struct btrfs_root *root,
1852 u64 bytenr, u64 num_bytes,
1853 u64 parent, u64 root_objectid,
1854 u64 owner, u64 offset, int refs_to_add,
1855 struct btrfs_delayed_extent_op *extent_op)
1857 struct btrfs_path *path;
1858 struct extent_buffer *leaf;
1859 struct btrfs_extent_item *item;
1864 path = btrfs_alloc_path();
1869 path->leave_spinning = 1;
1870 /* this will setup the path even if it fails to insert the back ref */
1871 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1872 path, bytenr, num_bytes, parent,
1873 root_objectid, owner, offset,
1874 refs_to_add, extent_op);
1878 if (ret != -EAGAIN) {
1883 leaf = path->nodes[0];
1884 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1885 refs = btrfs_extent_refs(leaf, item);
1886 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1888 __run_delayed_extent_op(extent_op, leaf, item);
1890 btrfs_mark_buffer_dirty(leaf);
1891 btrfs_release_path(path);
1894 path->leave_spinning = 1;
1896 /* now insert the actual backref */
1897 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1898 path, bytenr, parent, root_objectid,
1899 owner, offset, refs_to_add);
1902 btrfs_free_path(path);
1906 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1907 struct btrfs_root *root,
1908 struct btrfs_delayed_ref_node *node,
1909 struct btrfs_delayed_extent_op *extent_op,
1910 int insert_reserved)
1913 struct btrfs_delayed_data_ref *ref;
1914 struct btrfs_key ins;
1919 ins.objectid = node->bytenr;
1920 ins.offset = node->num_bytes;
1921 ins.type = BTRFS_EXTENT_ITEM_KEY;
1923 ref = btrfs_delayed_node_to_data_ref(node);
1924 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1925 parent = ref->parent;
1927 ref_root = ref->root;
1929 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1931 BUG_ON(extent_op->update_key);
1932 flags |= extent_op->flags_to_set;
1934 ret = alloc_reserved_file_extent(trans, root,
1935 parent, ref_root, flags,
1936 ref->objectid, ref->offset,
1937 &ins, node->ref_mod);
1938 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1939 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1940 node->num_bytes, parent,
1941 ref_root, ref->objectid,
1942 ref->offset, node->ref_mod,
1944 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1945 ret = __btrfs_free_extent(trans, root, node->bytenr,
1946 node->num_bytes, parent,
1947 ref_root, ref->objectid,
1948 ref->offset, node->ref_mod,
1956 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1957 struct extent_buffer *leaf,
1958 struct btrfs_extent_item *ei)
1960 u64 flags = btrfs_extent_flags(leaf, ei);
1961 if (extent_op->update_flags) {
1962 flags |= extent_op->flags_to_set;
1963 btrfs_set_extent_flags(leaf, ei, flags);
1966 if (extent_op->update_key) {
1967 struct btrfs_tree_block_info *bi;
1968 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1969 bi = (struct btrfs_tree_block_info *)(ei + 1);
1970 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1974 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1975 struct btrfs_root *root,
1976 struct btrfs_delayed_ref_node *node,
1977 struct btrfs_delayed_extent_op *extent_op)
1979 struct btrfs_key key;
1980 struct btrfs_path *path;
1981 struct btrfs_extent_item *ei;
1982 struct extent_buffer *leaf;
1987 path = btrfs_alloc_path();
1991 key.objectid = node->bytenr;
1992 key.type = BTRFS_EXTENT_ITEM_KEY;
1993 key.offset = node->num_bytes;
1996 path->leave_spinning = 1;
1997 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2008 leaf = path->nodes[0];
2009 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2010 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2011 if (item_size < sizeof(*ei)) {
2012 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2018 leaf = path->nodes[0];
2019 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2022 BUG_ON(item_size < sizeof(*ei));
2023 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2024 __run_delayed_extent_op(extent_op, leaf, ei);
2026 btrfs_mark_buffer_dirty(leaf);
2028 btrfs_free_path(path);
2032 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2033 struct btrfs_root *root,
2034 struct btrfs_delayed_ref_node *node,
2035 struct btrfs_delayed_extent_op *extent_op,
2036 int insert_reserved)
2039 struct btrfs_delayed_tree_ref *ref;
2040 struct btrfs_key ins;
2044 ins.objectid = node->bytenr;
2045 ins.offset = node->num_bytes;
2046 ins.type = BTRFS_EXTENT_ITEM_KEY;
2048 ref = btrfs_delayed_node_to_tree_ref(node);
2049 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2050 parent = ref->parent;
2052 ref_root = ref->root;
2054 BUG_ON(node->ref_mod != 1);
2055 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2056 BUG_ON(!extent_op || !extent_op->update_flags ||
2057 !extent_op->update_key);
2058 ret = alloc_reserved_tree_block(trans, root,
2060 extent_op->flags_to_set,
2063 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2064 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2065 node->num_bytes, parent, ref_root,
2066 ref->level, 0, 1, extent_op);
2067 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2068 ret = __btrfs_free_extent(trans, root, node->bytenr,
2069 node->num_bytes, parent, ref_root,
2070 ref->level, 0, 1, extent_op);
2077 /* helper function to actually process a single delayed ref entry */
2078 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2079 struct btrfs_root *root,
2080 struct btrfs_delayed_ref_node *node,
2081 struct btrfs_delayed_extent_op *extent_op,
2082 int insert_reserved)
2085 if (btrfs_delayed_ref_is_head(node)) {
2086 struct btrfs_delayed_ref_head *head;
2088 * we've hit the end of the chain and we were supposed
2089 * to insert this extent into the tree. But, it got
2090 * deleted before we ever needed to insert it, so all
2091 * we have to do is clean up the accounting
2094 head = btrfs_delayed_node_to_head(node);
2095 if (insert_reserved) {
2096 btrfs_pin_extent(root, node->bytenr,
2097 node->num_bytes, 1);
2098 if (head->is_data) {
2099 ret = btrfs_del_csums(trans, root,
2105 mutex_unlock(&head->mutex);
2109 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2110 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2111 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2113 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2114 node->type == BTRFS_SHARED_DATA_REF_KEY)
2115 ret = run_delayed_data_ref(trans, root, node, extent_op,
2122 static noinline struct btrfs_delayed_ref_node *
2123 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2125 struct rb_node *node;
2126 struct btrfs_delayed_ref_node *ref;
2127 int action = BTRFS_ADD_DELAYED_REF;
2130 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2131 * this prevents ref count from going down to zero when
2132 * there still are pending delayed ref.
2134 node = rb_prev(&head->node.rb_node);
2138 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2140 if (ref->bytenr != head->node.bytenr)
2142 if (ref->action == action)
2144 node = rb_prev(node);
2146 if (action == BTRFS_ADD_DELAYED_REF) {
2147 action = BTRFS_DROP_DELAYED_REF;
2153 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2154 struct btrfs_root *root,
2155 struct list_head *cluster)
2157 struct btrfs_delayed_ref_root *delayed_refs;
2158 struct btrfs_delayed_ref_node *ref;
2159 struct btrfs_delayed_ref_head *locked_ref = NULL;
2160 struct btrfs_delayed_extent_op *extent_op;
2163 int must_insert_reserved = 0;
2165 delayed_refs = &trans->transaction->delayed_refs;
2168 /* pick a new head ref from the cluster list */
2169 if (list_empty(cluster))
2172 locked_ref = list_entry(cluster->next,
2173 struct btrfs_delayed_ref_head, cluster);
2175 /* grab the lock that says we are going to process
2176 * all the refs for this head */
2177 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2180 * we may have dropped the spin lock to get the head
2181 * mutex lock, and that might have given someone else
2182 * time to free the head. If that's true, it has been
2183 * removed from our list and we can move on.
2185 if (ret == -EAGAIN) {
2193 * record the must insert reserved flag before we
2194 * drop the spin lock.
2196 must_insert_reserved = locked_ref->must_insert_reserved;
2197 locked_ref->must_insert_reserved = 0;
2199 extent_op = locked_ref->extent_op;
2200 locked_ref->extent_op = NULL;
2203 * locked_ref is the head node, so we have to go one
2204 * node back for any delayed ref updates
2206 ref = select_delayed_ref(locked_ref);
2208 /* All delayed refs have been processed, Go ahead
2209 * and send the head node to run_one_delayed_ref,
2210 * so that any accounting fixes can happen
2212 ref = &locked_ref->node;
2214 if (extent_op && must_insert_reserved) {
2220 spin_unlock(&delayed_refs->lock);
2222 ret = run_delayed_extent_op(trans, root,
2228 spin_lock(&delayed_refs->lock);
2232 list_del_init(&locked_ref->cluster);
2237 rb_erase(&ref->rb_node, &delayed_refs->root);
2238 delayed_refs->num_entries--;
2240 spin_unlock(&delayed_refs->lock);
2242 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2243 must_insert_reserved);
2246 btrfs_put_delayed_ref(ref);
2251 spin_lock(&delayed_refs->lock);
2257 * this starts processing the delayed reference count updates and
2258 * extent insertions we have queued up so far. count can be
2259 * 0, which means to process everything in the tree at the start
2260 * of the run (but not newly added entries), or it can be some target
2261 * number you'd like to process.
2263 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2264 struct btrfs_root *root, unsigned long count)
2266 struct rb_node *node;
2267 struct btrfs_delayed_ref_root *delayed_refs;
2268 struct btrfs_delayed_ref_node *ref;
2269 struct list_head cluster;
2271 int run_all = count == (unsigned long)-1;
2274 if (root == root->fs_info->extent_root)
2275 root = root->fs_info->tree_root;
2277 delayed_refs = &trans->transaction->delayed_refs;
2278 INIT_LIST_HEAD(&cluster);
2280 spin_lock(&delayed_refs->lock);
2282 count = delayed_refs->num_entries * 2;
2286 if (!(run_all || run_most) &&
2287 delayed_refs->num_heads_ready < 64)
2291 * go find something we can process in the rbtree. We start at
2292 * the beginning of the tree, and then build a cluster
2293 * of refs to process starting at the first one we are able to
2296 ret = btrfs_find_ref_cluster(trans, &cluster,
2297 delayed_refs->run_delayed_start);
2301 ret = run_clustered_refs(trans, root, &cluster);
2304 count -= min_t(unsigned long, ret, count);
2311 node = rb_first(&delayed_refs->root);
2314 count = (unsigned long)-1;
2317 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2319 if (btrfs_delayed_ref_is_head(ref)) {
2320 struct btrfs_delayed_ref_head *head;
2322 head = btrfs_delayed_node_to_head(ref);
2323 atomic_inc(&ref->refs);
2325 spin_unlock(&delayed_refs->lock);
2327 * Mutex was contended, block until it's
2328 * released and try again
2330 mutex_lock(&head->mutex);
2331 mutex_unlock(&head->mutex);
2333 btrfs_put_delayed_ref(ref);
2337 node = rb_next(node);
2339 spin_unlock(&delayed_refs->lock);
2340 schedule_timeout(1);
2344 spin_unlock(&delayed_refs->lock);
2348 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2349 struct btrfs_root *root,
2350 u64 bytenr, u64 num_bytes, u64 flags,
2353 struct btrfs_delayed_extent_op *extent_op;
2356 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2360 extent_op->flags_to_set = flags;
2361 extent_op->update_flags = 1;
2362 extent_op->update_key = 0;
2363 extent_op->is_data = is_data ? 1 : 0;
2365 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2371 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2372 struct btrfs_root *root,
2373 struct btrfs_path *path,
2374 u64 objectid, u64 offset, u64 bytenr)
2376 struct btrfs_delayed_ref_head *head;
2377 struct btrfs_delayed_ref_node *ref;
2378 struct btrfs_delayed_data_ref *data_ref;
2379 struct btrfs_delayed_ref_root *delayed_refs;
2380 struct rb_node *node;
2384 delayed_refs = &trans->transaction->delayed_refs;
2385 spin_lock(&delayed_refs->lock);
2386 head = btrfs_find_delayed_ref_head(trans, bytenr);
2390 if (!mutex_trylock(&head->mutex)) {
2391 atomic_inc(&head->node.refs);
2392 spin_unlock(&delayed_refs->lock);
2394 btrfs_release_path(path);
2397 * Mutex was contended, block until it's released and let
2400 mutex_lock(&head->mutex);
2401 mutex_unlock(&head->mutex);
2402 btrfs_put_delayed_ref(&head->node);
2406 node = rb_prev(&head->node.rb_node);
2410 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2412 if (ref->bytenr != bytenr)
2416 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2419 data_ref = btrfs_delayed_node_to_data_ref(ref);
2421 node = rb_prev(node);
2423 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2424 if (ref->bytenr == bytenr)
2428 if (data_ref->root != root->root_key.objectid ||
2429 data_ref->objectid != objectid || data_ref->offset != offset)
2434 mutex_unlock(&head->mutex);
2436 spin_unlock(&delayed_refs->lock);
2440 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2441 struct btrfs_root *root,
2442 struct btrfs_path *path,
2443 u64 objectid, u64 offset, u64 bytenr)
2445 struct btrfs_root *extent_root = root->fs_info->extent_root;
2446 struct extent_buffer *leaf;
2447 struct btrfs_extent_data_ref *ref;
2448 struct btrfs_extent_inline_ref *iref;
2449 struct btrfs_extent_item *ei;
2450 struct btrfs_key key;
2454 key.objectid = bytenr;
2455 key.offset = (u64)-1;
2456 key.type = BTRFS_EXTENT_ITEM_KEY;
2458 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2464 if (path->slots[0] == 0)
2468 leaf = path->nodes[0];
2469 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2471 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2475 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2476 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2477 if (item_size < sizeof(*ei)) {
2478 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2482 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2484 if (item_size != sizeof(*ei) +
2485 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2488 if (btrfs_extent_generation(leaf, ei) <=
2489 btrfs_root_last_snapshot(&root->root_item))
2492 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2493 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2494 BTRFS_EXTENT_DATA_REF_KEY)
2497 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2498 if (btrfs_extent_refs(leaf, ei) !=
2499 btrfs_extent_data_ref_count(leaf, ref) ||
2500 btrfs_extent_data_ref_root(leaf, ref) !=
2501 root->root_key.objectid ||
2502 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2503 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2511 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2512 struct btrfs_root *root,
2513 u64 objectid, u64 offset, u64 bytenr)
2515 struct btrfs_path *path;
2519 path = btrfs_alloc_path();
2524 ret = check_committed_ref(trans, root, path, objectid,
2526 if (ret && ret != -ENOENT)
2529 ret2 = check_delayed_ref(trans, root, path, objectid,
2531 } while (ret2 == -EAGAIN);
2533 if (ret2 && ret2 != -ENOENT) {
2538 if (ret != -ENOENT || ret2 != -ENOENT)
2541 btrfs_free_path(path);
2542 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2547 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2548 struct btrfs_root *root,
2549 struct extent_buffer *buf,
2550 int full_backref, int inc)
2557 struct btrfs_key key;
2558 struct btrfs_file_extent_item *fi;
2562 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2563 u64, u64, u64, u64, u64, u64);
2565 ref_root = btrfs_header_owner(buf);
2566 nritems = btrfs_header_nritems(buf);
2567 level = btrfs_header_level(buf);
2569 if (!root->ref_cows && level == 0)
2573 process_func = btrfs_inc_extent_ref;
2575 process_func = btrfs_free_extent;
2578 parent = buf->start;
2582 for (i = 0; i < nritems; i++) {
2584 btrfs_item_key_to_cpu(buf, &key, i);
2585 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2587 fi = btrfs_item_ptr(buf, i,
2588 struct btrfs_file_extent_item);
2589 if (btrfs_file_extent_type(buf, fi) ==
2590 BTRFS_FILE_EXTENT_INLINE)
2592 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2596 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2597 key.offset -= btrfs_file_extent_offset(buf, fi);
2598 ret = process_func(trans, root, bytenr, num_bytes,
2599 parent, ref_root, key.objectid,
2604 bytenr = btrfs_node_blockptr(buf, i);
2605 num_bytes = btrfs_level_size(root, level - 1);
2606 ret = process_func(trans, root, bytenr, num_bytes,
2607 parent, ref_root, level - 1, 0);
2618 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2619 struct extent_buffer *buf, int full_backref)
2621 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2624 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2625 struct extent_buffer *buf, int full_backref)
2627 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2630 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2631 struct btrfs_root *root,
2632 struct btrfs_path *path,
2633 struct btrfs_block_group_cache *cache)
2636 struct btrfs_root *extent_root = root->fs_info->extent_root;
2638 struct extent_buffer *leaf;
2640 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2645 leaf = path->nodes[0];
2646 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2647 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2648 btrfs_mark_buffer_dirty(leaf);
2649 btrfs_release_path(path);
2657 static struct btrfs_block_group_cache *
2658 next_block_group(struct btrfs_root *root,
2659 struct btrfs_block_group_cache *cache)
2661 struct rb_node *node;
2662 spin_lock(&root->fs_info->block_group_cache_lock);
2663 node = rb_next(&cache->cache_node);
2664 btrfs_put_block_group(cache);
2666 cache = rb_entry(node, struct btrfs_block_group_cache,
2668 btrfs_get_block_group(cache);
2671 spin_unlock(&root->fs_info->block_group_cache_lock);
2675 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2676 struct btrfs_trans_handle *trans,
2677 struct btrfs_path *path)
2679 struct btrfs_root *root = block_group->fs_info->tree_root;
2680 struct inode *inode = NULL;
2682 int dcs = BTRFS_DC_ERROR;
2688 * If this block group is smaller than 100 megs don't bother caching the
2691 if (block_group->key.offset < (100 * 1024 * 1024)) {
2692 spin_lock(&block_group->lock);
2693 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2694 spin_unlock(&block_group->lock);
2699 inode = lookup_free_space_inode(root, block_group, path);
2700 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2701 ret = PTR_ERR(inode);
2702 btrfs_release_path(path);
2706 if (IS_ERR(inode)) {
2710 if (block_group->ro)
2713 ret = create_free_space_inode(root, trans, block_group, path);
2720 * We want to set the generation to 0, that way if anything goes wrong
2721 * from here on out we know not to trust this cache when we load up next
2724 BTRFS_I(inode)->generation = 0;
2725 ret = btrfs_update_inode(trans, root, inode);
2728 if (i_size_read(inode) > 0) {
2729 ret = btrfs_truncate_free_space_cache(root, trans, path,
2735 spin_lock(&block_group->lock);
2736 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2737 /* We're not cached, don't bother trying to write stuff out */
2738 dcs = BTRFS_DC_WRITTEN;
2739 spin_unlock(&block_group->lock);
2742 spin_unlock(&block_group->lock);
2744 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2749 * Just to make absolutely sure we have enough space, we're going to
2750 * preallocate 12 pages worth of space for each block group. In
2751 * practice we ought to use at most 8, but we need extra space so we can
2752 * add our header and have a terminator between the extents and the
2756 num_pages *= PAGE_CACHE_SIZE;
2758 ret = btrfs_delalloc_reserve_space(inode, num_pages);
2762 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2763 num_pages, num_pages,
2766 dcs = BTRFS_DC_SETUP;
2767 btrfs_free_reserved_data_space(inode, num_pages);
2769 btrfs_delalloc_release_space(inode, num_pages);
2775 btrfs_release_path(path);
2777 spin_lock(&block_group->lock);
2778 block_group->disk_cache_state = dcs;
2779 spin_unlock(&block_group->lock);
2784 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root)
2787 struct btrfs_block_group_cache *cache;
2789 struct btrfs_path *path;
2792 path = btrfs_alloc_path();
2798 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2800 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2802 cache = next_block_group(root, cache);
2810 err = cache_save_setup(cache, trans, path);
2811 last = cache->key.objectid + cache->key.offset;
2812 btrfs_put_block_group(cache);
2817 err = btrfs_run_delayed_refs(trans, root,
2822 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2824 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2825 btrfs_put_block_group(cache);
2831 cache = next_block_group(root, cache);
2840 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2841 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2843 last = cache->key.objectid + cache->key.offset;
2845 err = write_one_cache_group(trans, root, path, cache);
2847 btrfs_put_block_group(cache);
2852 * I don't think this is needed since we're just marking our
2853 * preallocated extent as written, but just in case it can't
2857 err = btrfs_run_delayed_refs(trans, root,
2862 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2865 * Really this shouldn't happen, but it could if we
2866 * couldn't write the entire preallocated extent and
2867 * splitting the extent resulted in a new block.
2870 btrfs_put_block_group(cache);
2873 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2875 cache = next_block_group(root, cache);
2884 btrfs_write_out_cache(root, trans, cache, path);
2887 * If we didn't have an error then the cache state is still
2888 * NEED_WRITE, so we can set it to WRITTEN.
2890 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2891 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2892 last = cache->key.objectid + cache->key.offset;
2893 btrfs_put_block_group(cache);
2896 btrfs_free_path(path);
2900 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2902 struct btrfs_block_group_cache *block_group;
2905 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2906 if (!block_group || block_group->ro)
2909 btrfs_put_block_group(block_group);
2913 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2914 u64 total_bytes, u64 bytes_used,
2915 struct btrfs_space_info **space_info)
2917 struct btrfs_space_info *found;
2921 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2922 BTRFS_BLOCK_GROUP_RAID10))
2927 found = __find_space_info(info, flags);
2929 spin_lock(&found->lock);
2930 found->total_bytes += total_bytes;
2931 found->disk_total += total_bytes * factor;
2932 found->bytes_used += bytes_used;
2933 found->disk_used += bytes_used * factor;
2935 spin_unlock(&found->lock);
2936 *space_info = found;
2939 found = kzalloc(sizeof(*found), GFP_NOFS);
2943 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2944 INIT_LIST_HEAD(&found->block_groups[i]);
2945 init_rwsem(&found->groups_sem);
2946 spin_lock_init(&found->lock);
2947 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2948 BTRFS_BLOCK_GROUP_SYSTEM |
2949 BTRFS_BLOCK_GROUP_METADATA);
2950 found->total_bytes = total_bytes;
2951 found->disk_total = total_bytes * factor;
2952 found->bytes_used = bytes_used;
2953 found->disk_used = bytes_used * factor;
2954 found->bytes_pinned = 0;
2955 found->bytes_reserved = 0;
2956 found->bytes_readonly = 0;
2957 found->bytes_may_use = 0;
2959 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2960 found->chunk_alloc = 0;
2962 init_waitqueue_head(&found->wait);
2963 *space_info = found;
2964 list_add_rcu(&found->list, &info->space_info);
2968 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2970 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2971 BTRFS_BLOCK_GROUP_RAID1 |
2972 BTRFS_BLOCK_GROUP_RAID10 |
2973 BTRFS_BLOCK_GROUP_DUP);
2975 if (flags & BTRFS_BLOCK_GROUP_DATA)
2976 fs_info->avail_data_alloc_bits |= extra_flags;
2977 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2978 fs_info->avail_metadata_alloc_bits |= extra_flags;
2979 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2980 fs_info->avail_system_alloc_bits |= extra_flags;
2984 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2987 * we add in the count of missing devices because we want
2988 * to make sure that any RAID levels on a degraded FS
2989 * continue to be honored.
2991 u64 num_devices = root->fs_info->fs_devices->rw_devices +
2992 root->fs_info->fs_devices->missing_devices;
2994 if (num_devices == 1)
2995 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2996 if (num_devices < 4)
2997 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2999 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3000 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3001 BTRFS_BLOCK_GROUP_RAID10))) {
3002 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3005 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3006 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3007 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3010 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3011 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3012 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3013 (flags & BTRFS_BLOCK_GROUP_DUP)))
3014 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3018 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3020 if (flags & BTRFS_BLOCK_GROUP_DATA)
3021 flags |= root->fs_info->avail_data_alloc_bits &
3022 root->fs_info->data_alloc_profile;
3023 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3024 flags |= root->fs_info->avail_system_alloc_bits &
3025 root->fs_info->system_alloc_profile;
3026 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3027 flags |= root->fs_info->avail_metadata_alloc_bits &
3028 root->fs_info->metadata_alloc_profile;
3029 return btrfs_reduce_alloc_profile(root, flags);
3032 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3037 flags = BTRFS_BLOCK_GROUP_DATA;
3038 else if (root == root->fs_info->chunk_root)
3039 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3041 flags = BTRFS_BLOCK_GROUP_METADATA;
3043 return get_alloc_profile(root, flags);
3046 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3048 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3049 BTRFS_BLOCK_GROUP_DATA);
3053 * This will check the space that the inode allocates from to make sure we have
3054 * enough space for bytes.
3056 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3058 struct btrfs_space_info *data_sinfo;
3059 struct btrfs_root *root = BTRFS_I(inode)->root;
3061 int ret = 0, committed = 0, alloc_chunk = 1;
3063 /* make sure bytes are sectorsize aligned */
3064 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3066 if (root == root->fs_info->tree_root ||
3067 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3072 data_sinfo = BTRFS_I(inode)->space_info;
3077 /* make sure we have enough space to handle the data first */
3078 spin_lock(&data_sinfo->lock);
3079 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3080 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3081 data_sinfo->bytes_may_use;
3083 if (used + bytes > data_sinfo->total_bytes) {
3084 struct btrfs_trans_handle *trans;
3087 * if we don't have enough free bytes in this space then we need
3088 * to alloc a new chunk.
3090 if (!data_sinfo->full && alloc_chunk) {
3093 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3094 spin_unlock(&data_sinfo->lock);
3096 alloc_target = btrfs_get_alloc_profile(root, 1);
3097 trans = btrfs_join_transaction(root);
3099 return PTR_ERR(trans);
3101 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3102 bytes + 2 * 1024 * 1024,
3104 CHUNK_ALLOC_NO_FORCE);
3105 btrfs_end_transaction(trans, root);
3114 btrfs_set_inode_space_info(root, inode);
3115 data_sinfo = BTRFS_I(inode)->space_info;
3121 * If we have less pinned bytes than we want to allocate then
3122 * don't bother committing the transaction, it won't help us.
3124 if (data_sinfo->bytes_pinned < bytes)
3126 spin_unlock(&data_sinfo->lock);
3128 /* commit the current transaction and try again */
3131 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3133 trans = btrfs_join_transaction(root);
3135 return PTR_ERR(trans);
3136 ret = btrfs_commit_transaction(trans, root);
3144 data_sinfo->bytes_may_use += bytes;
3145 spin_unlock(&data_sinfo->lock);
3151 * Called if we need to clear a data reservation for this inode.
3153 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3155 struct btrfs_root *root = BTRFS_I(inode)->root;
3156 struct btrfs_space_info *data_sinfo;
3158 /* make sure bytes are sectorsize aligned */
3159 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3161 data_sinfo = BTRFS_I(inode)->space_info;
3162 spin_lock(&data_sinfo->lock);
3163 data_sinfo->bytes_may_use -= bytes;
3164 spin_unlock(&data_sinfo->lock);
3167 static void force_metadata_allocation(struct btrfs_fs_info *info)
3169 struct list_head *head = &info->space_info;
3170 struct btrfs_space_info *found;
3173 list_for_each_entry_rcu(found, head, list) {
3174 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3175 found->force_alloc = CHUNK_ALLOC_FORCE;
3180 static int should_alloc_chunk(struct btrfs_root *root,
3181 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3184 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3185 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3186 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3189 if (force == CHUNK_ALLOC_FORCE)
3193 * We need to take into account the global rsv because for all intents
3194 * and purposes it's used space. Don't worry about locking the
3195 * global_rsv, it doesn't change except when the transaction commits.
3197 num_allocated += global_rsv->size;
3200 * in limited mode, we want to have some free space up to
3201 * about 1% of the FS size.
3203 if (force == CHUNK_ALLOC_LIMITED) {
3204 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3205 thresh = max_t(u64, 64 * 1024 * 1024,
3206 div_factor_fine(thresh, 1));
3208 if (num_bytes - num_allocated < thresh)
3213 * we have two similar checks here, one based on percentage
3214 * and once based on a hard number of 256MB. The idea
3215 * is that if we have a good amount of free
3216 * room, don't allocate a chunk. A good mount is
3217 * less than 80% utilized of the chunks we have allocated,
3218 * or more than 256MB free
3220 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3223 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3226 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3228 /* 256MB or 5% of the FS */
3229 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3231 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3236 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3237 struct btrfs_root *extent_root, u64 alloc_bytes,
3238 u64 flags, int force)
3240 struct btrfs_space_info *space_info;
3241 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3242 int wait_for_alloc = 0;
3245 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3247 space_info = __find_space_info(extent_root->fs_info, flags);
3249 ret = update_space_info(extent_root->fs_info, flags,
3253 BUG_ON(!space_info);
3256 spin_lock(&space_info->lock);
3257 if (space_info->force_alloc)
3258 force = space_info->force_alloc;
3259 if (space_info->full) {
3260 spin_unlock(&space_info->lock);
3264 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3265 spin_unlock(&space_info->lock);
3267 } else if (space_info->chunk_alloc) {
3270 space_info->chunk_alloc = 1;
3273 spin_unlock(&space_info->lock);
3275 mutex_lock(&fs_info->chunk_mutex);
3278 * The chunk_mutex is held throughout the entirety of a chunk
3279 * allocation, so once we've acquired the chunk_mutex we know that the
3280 * other guy is done and we need to recheck and see if we should
3283 if (wait_for_alloc) {
3284 mutex_unlock(&fs_info->chunk_mutex);
3290 * If we have mixed data/metadata chunks we want to make sure we keep
3291 * allocating mixed chunks instead of individual chunks.
3293 if (btrfs_mixed_space_info(space_info))
3294 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3297 * if we're doing a data chunk, go ahead and make sure that
3298 * we keep a reasonable number of metadata chunks allocated in the
3301 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3302 fs_info->data_chunk_allocations++;
3303 if (!(fs_info->data_chunk_allocations %
3304 fs_info->metadata_ratio))
3305 force_metadata_allocation(fs_info);
3308 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3309 if (ret < 0 && ret != -ENOSPC)
3312 spin_lock(&space_info->lock);
3314 space_info->full = 1;
3318 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3319 space_info->chunk_alloc = 0;
3320 spin_unlock(&space_info->lock);
3322 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3327 * shrink metadata reservation for delalloc
3329 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root, u64 to_reclaim, int sync)
3332 struct btrfs_block_rsv *block_rsv;
3333 struct btrfs_space_info *space_info;
3338 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3340 unsigned long progress;
3342 block_rsv = &root->fs_info->delalloc_block_rsv;
3343 space_info = block_rsv->space_info;
3346 reserved = space_info->bytes_may_use;
3347 progress = space_info->reservation_progress;
3353 if (root->fs_info->delalloc_bytes == 0) {
3356 btrfs_wait_ordered_extents(root, 0, 0);
3360 max_reclaim = min(reserved, to_reclaim);
3362 while (loops < 1024) {
3363 /* have the flusher threads jump in and do some IO */
3365 nr_pages = min_t(unsigned long, nr_pages,
3366 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3367 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3369 spin_lock(&space_info->lock);
3370 if (reserved > space_info->bytes_may_use)
3371 reclaimed += reserved - space_info->bytes_may_use;
3372 reserved = space_info->bytes_may_use;
3373 spin_unlock(&space_info->lock);
3377 if (reserved == 0 || reclaimed >= max_reclaim)
3380 if (trans && trans->transaction->blocked)
3383 time_left = schedule_timeout_interruptible(1);
3385 /* We were interrupted, exit */
3389 /* we've kicked the IO a few times, if anything has been freed,
3390 * exit. There is no sense in looping here for a long time
3391 * when we really need to commit the transaction, or there are
3392 * just too many writers without enough free space
3397 if (progress != space_info->reservation_progress)
3402 if (reclaimed >= to_reclaim && !trans)
3403 btrfs_wait_ordered_extents(root, 0, 0);
3404 return reclaimed >= to_reclaim;
3408 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3409 * @root - the root we're allocating for
3410 * @block_rsv - the block_rsv we're allocating for
3411 * @orig_bytes - the number of bytes we want
3412 * @flush - wether or not we can flush to make our reservation
3413 * @check - wether this is just to check if we have enough space or not
3415 * This will reserve orgi_bytes number of bytes from the space info associated
3416 * with the block_rsv. If there is not enough space it will make an attempt to
3417 * flush out space to make room. It will do this by flushing delalloc if
3418 * possible or committing the transaction. If flush is 0 then no attempts to
3419 * regain reservations will be made and this will fail if there is not enough
3422 static int reserve_metadata_bytes(struct btrfs_root *root,
3423 struct btrfs_block_rsv *block_rsv,
3424 u64 orig_bytes, int flush, int check)
3426 struct btrfs_space_info *space_info = block_rsv->space_info;
3427 struct btrfs_trans_handle *trans;
3429 u64 num_bytes = orig_bytes;
3432 bool committed = false;
3433 bool flushing = false;
3435 trans = (struct btrfs_trans_handle *)current->journal_info;
3438 spin_lock(&space_info->lock);
3440 * We only want to wait if somebody other than us is flushing and we are
3441 * actually alloed to flush.
3443 while (flush && !flushing && space_info->flush) {
3444 spin_unlock(&space_info->lock);
3446 * If we have a trans handle we can't wait because the flusher
3447 * may have to commit the transaction, which would mean we would
3448 * deadlock since we are waiting for the flusher to finish, but
3449 * hold the current transaction open.
3453 ret = wait_event_interruptible(space_info->wait,
3454 !space_info->flush);
3455 /* Must have been interrupted, return */
3459 spin_lock(&space_info->lock);
3463 used = space_info->bytes_used + space_info->bytes_reserved +
3464 space_info->bytes_pinned + space_info->bytes_readonly +
3465 space_info->bytes_may_use;
3468 * The idea here is that we've not already over-reserved the block group
3469 * then we can go ahead and save our reservation first and then start
3470 * flushing if we need to. Otherwise if we've already overcommitted
3471 * lets start flushing stuff first and then come back and try to make
3474 if (used <= space_info->total_bytes) {
3475 if (used + orig_bytes <= space_info->total_bytes) {
3476 space_info->bytes_may_use += orig_bytes;
3480 * Ok set num_bytes to orig_bytes since we aren't
3481 * overocmmitted, this way we only try and reclaim what
3484 num_bytes = orig_bytes;
3488 * Ok we're over committed, set num_bytes to the overcommitted
3489 * amount plus the amount of bytes that we need for this
3492 num_bytes = used - space_info->total_bytes +
3493 (orig_bytes * (retries + 1));
3496 if (ret && !check) {
3497 u64 profile = btrfs_get_alloc_profile(root, 0);
3500 spin_lock(&root->fs_info->free_chunk_lock);
3501 avail = root->fs_info->free_chunk_space;
3504 * If we have dup, raid1 or raid10 then only half of the free
3505 * space is actually useable.
3507 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3508 BTRFS_BLOCK_GROUP_RAID1 |
3509 BTRFS_BLOCK_GROUP_RAID10))
3513 * If we aren't flushing don't let us overcommit too much, say
3514 * 1/8th of the space. If we can flush, let it overcommit up to
3521 spin_unlock(&root->fs_info->free_chunk_lock);
3523 if (used + orig_bytes < space_info->total_bytes + avail) {
3524 space_info->bytes_may_use += orig_bytes;
3530 * Couldn't make our reservation, save our place so while we're trying
3531 * to reclaim space we can actually use it instead of somebody else
3532 * stealing it from us.
3536 space_info->flush = 1;
3539 spin_unlock(&space_info->lock);
3545 * We do synchronous shrinking since we don't actually unreserve
3546 * metadata until after the IO is completed.
3548 ret = shrink_delalloc(trans, root, num_bytes, 1);
3555 * So if we were overcommitted it's possible that somebody else flushed
3556 * out enough space and we simply didn't have enough space to reclaim,
3557 * so go back around and try again.
3565 * Not enough space to be reclaimed, don't bother committing the
3568 spin_lock(&space_info->lock);
3569 if (space_info->bytes_pinned < orig_bytes)
3571 spin_unlock(&space_info->lock);
3583 trans = btrfs_join_transaction(root);
3586 ret = btrfs_commit_transaction(trans, root);
3595 spin_lock(&space_info->lock);
3596 space_info->flush = 0;
3597 wake_up_all(&space_info->wait);
3598 spin_unlock(&space_info->lock);
3603 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3604 struct btrfs_root *root)
3606 struct btrfs_block_rsv *block_rsv = NULL;
3608 if (root->ref_cows || root == root->fs_info->csum_root)
3609 block_rsv = trans->block_rsv;
3612 block_rsv = root->block_rsv;
3615 block_rsv = &root->fs_info->empty_block_rsv;
3620 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3624 spin_lock(&block_rsv->lock);
3625 if (block_rsv->reserved >= num_bytes) {
3626 block_rsv->reserved -= num_bytes;
3627 if (block_rsv->reserved < block_rsv->size)
3628 block_rsv->full = 0;
3631 spin_unlock(&block_rsv->lock);
3635 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3636 u64 num_bytes, int update_size)
3638 spin_lock(&block_rsv->lock);
3639 block_rsv->reserved += num_bytes;
3641 block_rsv->size += num_bytes;
3642 else if (block_rsv->reserved >= block_rsv->size)
3643 block_rsv->full = 1;
3644 spin_unlock(&block_rsv->lock);
3647 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3648 struct btrfs_block_rsv *dest, u64 num_bytes)
3650 struct btrfs_space_info *space_info = block_rsv->space_info;
3652 spin_lock(&block_rsv->lock);
3653 if (num_bytes == (u64)-1)
3654 num_bytes = block_rsv->size;
3655 block_rsv->size -= num_bytes;
3656 if (block_rsv->reserved >= block_rsv->size) {
3657 num_bytes = block_rsv->reserved - block_rsv->size;
3658 block_rsv->reserved = block_rsv->size;
3659 block_rsv->full = 1;
3663 spin_unlock(&block_rsv->lock);
3665 if (num_bytes > 0) {
3667 spin_lock(&dest->lock);
3671 bytes_to_add = dest->size - dest->reserved;
3672 bytes_to_add = min(num_bytes, bytes_to_add);
3673 dest->reserved += bytes_to_add;
3674 if (dest->reserved >= dest->size)
3676 num_bytes -= bytes_to_add;
3678 spin_unlock(&dest->lock);
3681 spin_lock(&space_info->lock);
3682 space_info->bytes_may_use -= num_bytes;
3683 space_info->reservation_progress++;
3684 spin_unlock(&space_info->lock);
3689 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3690 struct btrfs_block_rsv *dst, u64 num_bytes)
3694 ret = block_rsv_use_bytes(src, num_bytes);
3698 block_rsv_add_bytes(dst, num_bytes, 1);
3702 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3704 memset(rsv, 0, sizeof(*rsv));
3705 spin_lock_init(&rsv->lock);
3708 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3710 struct btrfs_block_rsv *block_rsv;
3711 struct btrfs_fs_info *fs_info = root->fs_info;
3713 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3717 btrfs_init_block_rsv(block_rsv);
3718 block_rsv->space_info = __find_space_info(fs_info,
3719 BTRFS_BLOCK_GROUP_METADATA);
3723 void btrfs_free_block_rsv(struct btrfs_root *root,
3724 struct btrfs_block_rsv *rsv)
3726 btrfs_block_rsv_release(root, rsv, (u64)-1);
3730 int btrfs_block_rsv_add(struct btrfs_root *root,
3731 struct btrfs_block_rsv *block_rsv,
3739 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1, 0);
3741 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3748 int btrfs_block_rsv_check(struct btrfs_root *root,
3749 struct btrfs_block_rsv *block_rsv,
3750 u64 min_reserved, int min_factor, int flush)
3758 spin_lock(&block_rsv->lock);
3760 num_bytes = div_factor(block_rsv->size, min_factor);
3761 if (min_reserved > num_bytes)
3762 num_bytes = min_reserved;
3764 if (block_rsv->reserved >= num_bytes)
3767 num_bytes -= block_rsv->reserved;
3768 spin_unlock(&block_rsv->lock);
3773 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush, !flush);
3775 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3782 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3783 struct btrfs_block_rsv *dst_rsv,
3786 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3789 void btrfs_block_rsv_release(struct btrfs_root *root,
3790 struct btrfs_block_rsv *block_rsv,
3793 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3794 if (global_rsv->full || global_rsv == block_rsv ||
3795 block_rsv->space_info != global_rsv->space_info)
3797 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3801 * helper to calculate size of global block reservation.
3802 * the desired value is sum of space used by extent tree,
3803 * checksum tree and root tree
3805 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3807 struct btrfs_space_info *sinfo;
3811 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3813 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3814 spin_lock(&sinfo->lock);
3815 data_used = sinfo->bytes_used;
3816 spin_unlock(&sinfo->lock);
3818 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3819 spin_lock(&sinfo->lock);
3820 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3822 meta_used = sinfo->bytes_used;
3823 spin_unlock(&sinfo->lock);
3825 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3827 num_bytes += div64_u64(data_used + meta_used, 50);
3829 if (num_bytes * 3 > meta_used)
3830 num_bytes = div64_u64(meta_used, 3);
3832 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3835 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3837 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3838 struct btrfs_space_info *sinfo = block_rsv->space_info;
3841 num_bytes = calc_global_metadata_size(fs_info);
3843 spin_lock(&block_rsv->lock);
3844 spin_lock(&sinfo->lock);
3846 block_rsv->size = num_bytes;
3848 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3849 sinfo->bytes_reserved + sinfo->bytes_readonly +
3850 sinfo->bytes_may_use;
3852 if (sinfo->total_bytes > num_bytes) {
3853 num_bytes = sinfo->total_bytes - num_bytes;
3854 block_rsv->reserved += num_bytes;
3855 sinfo->bytes_may_use += num_bytes;
3858 if (block_rsv->reserved >= block_rsv->size) {
3859 num_bytes = block_rsv->reserved - block_rsv->size;
3860 sinfo->bytes_may_use -= num_bytes;
3861 sinfo->reservation_progress++;
3862 block_rsv->reserved = block_rsv->size;
3863 block_rsv->full = 1;
3866 spin_unlock(&sinfo->lock);
3867 spin_unlock(&block_rsv->lock);
3870 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3872 struct btrfs_space_info *space_info;
3874 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3875 fs_info->chunk_block_rsv.space_info = space_info;
3877 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3878 fs_info->global_block_rsv.space_info = space_info;
3879 fs_info->delalloc_block_rsv.space_info = space_info;
3880 fs_info->trans_block_rsv.space_info = space_info;
3881 fs_info->empty_block_rsv.space_info = space_info;
3883 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3884 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3885 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3886 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3887 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3889 update_global_block_rsv(fs_info);
3892 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3894 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3895 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3896 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3897 WARN_ON(fs_info->trans_block_rsv.size > 0);
3898 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3899 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3900 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3903 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3904 struct btrfs_root *root)
3906 struct btrfs_block_rsv *block_rsv;
3908 if (!trans->bytes_reserved)
3911 block_rsv = &root->fs_info->trans_block_rsv;
3912 btrfs_block_rsv_release(root, block_rsv, trans->bytes_reserved);
3913 trans->bytes_reserved = 0;
3916 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3917 struct inode *inode)
3919 struct btrfs_root *root = BTRFS_I(inode)->root;
3920 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3921 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3924 * We need to hold space in order to delete our orphan item once we've
3925 * added it, so this takes the reservation so we can release it later
3926 * when we are truly done with the orphan item.
3928 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3929 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3932 void btrfs_orphan_release_metadata(struct inode *inode)
3934 struct btrfs_root *root = BTRFS_I(inode)->root;
3935 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3936 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3939 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3940 struct btrfs_pending_snapshot *pending)
3942 struct btrfs_root *root = pending->root;
3943 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3944 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3946 * two for root back/forward refs, two for directory entries
3947 * and one for root of the snapshot.
3949 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3950 dst_rsv->space_info = src_rsv->space_info;
3951 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3955 * drop_outstanding_extent - drop an outstanding extent
3956 * @inode: the inode we're dropping the extent for
3958 * This is called when we are freeing up an outstanding extent, either called
3959 * after an error or after an extent is written. This will return the number of
3960 * reserved extents that need to be freed. This must be called with
3961 * BTRFS_I(inode)->lock held.
3963 static unsigned drop_outstanding_extent(struct inode *inode)
3965 unsigned dropped_extents = 0;
3967 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3968 BTRFS_I(inode)->outstanding_extents--;
3971 * If we have more or the same amount of outsanding extents than we have
3972 * reserved then we need to leave the reserved extents count alone.
3974 if (BTRFS_I(inode)->outstanding_extents >=
3975 BTRFS_I(inode)->reserved_extents)
3978 dropped_extents = BTRFS_I(inode)->reserved_extents -
3979 BTRFS_I(inode)->outstanding_extents;
3980 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3981 return dropped_extents;
3985 * calc_csum_metadata_size - return the amount of metada space that must be
3986 * reserved/free'd for the given bytes.
3987 * @inode: the inode we're manipulating
3988 * @num_bytes: the number of bytes in question
3989 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3991 * This adjusts the number of csum_bytes in the inode and then returns the
3992 * correct amount of metadata that must either be reserved or freed. We
3993 * calculate how many checksums we can fit into one leaf and then divide the
3994 * number of bytes that will need to be checksumed by this value to figure out
3995 * how many checksums will be required. If we are adding bytes then the number
3996 * may go up and we will return the number of additional bytes that must be
3997 * reserved. If it is going down we will return the number of bytes that must
4000 * This must be called with BTRFS_I(inode)->lock held.
4002 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4005 struct btrfs_root *root = BTRFS_I(inode)->root;
4007 int num_csums_per_leaf;
4011 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4012 BTRFS_I(inode)->csum_bytes == 0)
4015 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4017 BTRFS_I(inode)->csum_bytes += num_bytes;
4019 BTRFS_I(inode)->csum_bytes -= num_bytes;
4020 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4021 num_csums_per_leaf = (int)div64_u64(csum_size,
4022 sizeof(struct btrfs_csum_item) +
4023 sizeof(struct btrfs_disk_key));
4024 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4025 num_csums = num_csums + num_csums_per_leaf - 1;
4026 num_csums = num_csums / num_csums_per_leaf;
4028 old_csums = old_csums + num_csums_per_leaf - 1;
4029 old_csums = old_csums / num_csums_per_leaf;
4031 /* No change, no need to reserve more */
4032 if (old_csums == num_csums)
4036 return btrfs_calc_trans_metadata_size(root,
4037 num_csums - old_csums);
4039 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4042 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4044 struct btrfs_root *root = BTRFS_I(inode)->root;
4045 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4047 unsigned nr_extents = 0;
4051 if (btrfs_is_free_space_inode(root, inode))
4054 if (flush && btrfs_transaction_in_commit(root->fs_info))
4055 schedule_timeout(1);
4057 num_bytes = ALIGN(num_bytes, root->sectorsize);
4059 spin_lock(&BTRFS_I(inode)->lock);
4060 BTRFS_I(inode)->outstanding_extents++;
4062 if (BTRFS_I(inode)->outstanding_extents >
4063 BTRFS_I(inode)->reserved_extents) {
4064 nr_extents = BTRFS_I(inode)->outstanding_extents -
4065 BTRFS_I(inode)->reserved_extents;
4066 BTRFS_I(inode)->reserved_extents += nr_extents;
4068 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4070 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4071 spin_unlock(&BTRFS_I(inode)->lock);
4073 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush, 0);
4078 spin_lock(&BTRFS_I(inode)->lock);
4079 dropped = drop_outstanding_extent(inode);
4080 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4081 spin_unlock(&BTRFS_I(inode)->lock);
4082 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4085 * Somebody could have come in and twiddled with the
4086 * reservation, so if we have to free more than we would have
4087 * reserved from this reservation go ahead and release those
4090 to_free -= to_reserve;
4092 btrfs_block_rsv_release(root, block_rsv, to_free);
4096 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4102 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4103 * @inode: the inode to release the reservation for
4104 * @num_bytes: the number of bytes we're releasing
4106 * This will release the metadata reservation for an inode. This can be called
4107 * once we complete IO for a given set of bytes to release their metadata
4110 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4112 struct btrfs_root *root = BTRFS_I(inode)->root;
4116 num_bytes = ALIGN(num_bytes, root->sectorsize);
4117 spin_lock(&BTRFS_I(inode)->lock);
4118 dropped = drop_outstanding_extent(inode);
4120 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4121 spin_unlock(&BTRFS_I(inode)->lock);
4123 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4125 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4130 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4131 * @inode: inode we're writing to
4132 * @num_bytes: the number of bytes we want to allocate
4134 * This will do the following things
4136 * o reserve space in the data space info for num_bytes
4137 * o reserve space in the metadata space info based on number of outstanding
4138 * extents and how much csums will be needed
4139 * o add to the inodes ->delalloc_bytes
4140 * o add it to the fs_info's delalloc inodes list.
4142 * This will return 0 for success and -ENOSPC if there is no space left.
4144 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4148 ret = btrfs_check_data_free_space(inode, num_bytes);
4152 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4154 btrfs_free_reserved_data_space(inode, num_bytes);
4162 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4163 * @inode: inode we're releasing space for
4164 * @num_bytes: the number of bytes we want to free up
4166 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4167 * called in the case that we don't need the metadata AND data reservations
4168 * anymore. So if there is an error or we insert an inline extent.
4170 * This function will release the metadata space that was not used and will
4171 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4172 * list if there are no delalloc bytes left.
4174 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4176 btrfs_delalloc_release_metadata(inode, num_bytes);
4177 btrfs_free_reserved_data_space(inode, num_bytes);
4180 static int update_block_group(struct btrfs_trans_handle *trans,
4181 struct btrfs_root *root,
4182 u64 bytenr, u64 num_bytes, int alloc)
4184 struct btrfs_block_group_cache *cache = NULL;
4185 struct btrfs_fs_info *info = root->fs_info;
4186 u64 total = num_bytes;
4191 /* block accounting for super block */
4192 spin_lock(&info->delalloc_lock);
4193 old_val = btrfs_super_bytes_used(&info->super_copy);
4195 old_val += num_bytes;
4197 old_val -= num_bytes;
4198 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4199 spin_unlock(&info->delalloc_lock);
4202 cache = btrfs_lookup_block_group(info, bytenr);
4205 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4206 BTRFS_BLOCK_GROUP_RAID1 |
4207 BTRFS_BLOCK_GROUP_RAID10))
4212 * If this block group has free space cache written out, we
4213 * need to make sure to load it if we are removing space. This
4214 * is because we need the unpinning stage to actually add the
4215 * space back to the block group, otherwise we will leak space.
4217 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4218 cache_block_group(cache, trans, NULL, 1);
4220 byte_in_group = bytenr - cache->key.objectid;
4221 WARN_ON(byte_in_group > cache->key.offset);
4223 spin_lock(&cache->space_info->lock);
4224 spin_lock(&cache->lock);
4226 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4227 cache->disk_cache_state < BTRFS_DC_CLEAR)
4228 cache->disk_cache_state = BTRFS_DC_CLEAR;
4231 old_val = btrfs_block_group_used(&cache->item);
4232 num_bytes = min(total, cache->key.offset - byte_in_group);
4234 old_val += num_bytes;
4235 btrfs_set_block_group_used(&cache->item, old_val);
4236 cache->reserved -= num_bytes;
4237 cache->space_info->bytes_reserved -= num_bytes;
4238 cache->space_info->bytes_used += num_bytes;
4239 cache->space_info->disk_used += num_bytes * factor;
4240 spin_unlock(&cache->lock);
4241 spin_unlock(&cache->space_info->lock);
4243 old_val -= num_bytes;
4244 btrfs_set_block_group_used(&cache->item, old_val);
4245 cache->pinned += num_bytes;
4246 cache->space_info->bytes_pinned += num_bytes;
4247 cache->space_info->bytes_used -= num_bytes;
4248 cache->space_info->disk_used -= num_bytes * factor;
4249 spin_unlock(&cache->lock);
4250 spin_unlock(&cache->space_info->lock);
4252 set_extent_dirty(info->pinned_extents,
4253 bytenr, bytenr + num_bytes - 1,
4254 GFP_NOFS | __GFP_NOFAIL);
4256 btrfs_put_block_group(cache);
4258 bytenr += num_bytes;
4263 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4265 struct btrfs_block_group_cache *cache;
4268 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4272 bytenr = cache->key.objectid;
4273 btrfs_put_block_group(cache);
4278 static int pin_down_extent(struct btrfs_root *root,
4279 struct btrfs_block_group_cache *cache,
4280 u64 bytenr, u64 num_bytes, int reserved)
4282 spin_lock(&cache->space_info->lock);
4283 spin_lock(&cache->lock);
4284 cache->pinned += num_bytes;
4285 cache->space_info->bytes_pinned += num_bytes;
4287 cache->reserved -= num_bytes;
4288 cache->space_info->bytes_reserved -= num_bytes;
4290 spin_unlock(&cache->lock);
4291 spin_unlock(&cache->space_info->lock);
4293 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4294 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4299 * this function must be called within transaction
4301 int btrfs_pin_extent(struct btrfs_root *root,
4302 u64 bytenr, u64 num_bytes, int reserved)
4304 struct btrfs_block_group_cache *cache;
4306 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4309 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4311 btrfs_put_block_group(cache);
4316 * btrfs_update_reserved_bytes - update the block_group and space info counters
4317 * @cache: The cache we are manipulating
4318 * @num_bytes: The number of bytes in question
4319 * @reserve: One of the reservation enums
4321 * This is called by the allocator when it reserves space, or by somebody who is
4322 * freeing space that was never actually used on disk. For example if you
4323 * reserve some space for a new leaf in transaction A and before transaction A
4324 * commits you free that leaf, you call this with reserve set to 0 in order to
4325 * clear the reservation.
4327 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4328 * ENOSPC accounting. For data we handle the reservation through clearing the
4329 * delalloc bits in the io_tree. We have to do this since we could end up
4330 * allocating less disk space for the amount of data we have reserved in the
4331 * case of compression.
4333 * If this is a reservation and the block group has become read only we cannot
4334 * make the reservation and return -EAGAIN, otherwise this function always
4337 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4338 u64 num_bytes, int reserve)
4340 struct btrfs_space_info *space_info = cache->space_info;
4342 spin_lock(&space_info->lock);
4343 spin_lock(&cache->lock);
4344 if (reserve != RESERVE_FREE) {
4348 cache->reserved += num_bytes;
4349 space_info->bytes_reserved += num_bytes;
4350 if (reserve == RESERVE_ALLOC) {
4351 BUG_ON(space_info->bytes_may_use < num_bytes);
4352 space_info->bytes_may_use -= num_bytes;
4357 space_info->bytes_readonly += num_bytes;
4358 cache->reserved -= num_bytes;
4359 space_info->bytes_reserved -= num_bytes;
4360 space_info->reservation_progress++;
4362 spin_unlock(&cache->lock);
4363 spin_unlock(&space_info->lock);
4367 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4368 struct btrfs_root *root)
4370 struct btrfs_fs_info *fs_info = root->fs_info;
4371 struct btrfs_caching_control *next;
4372 struct btrfs_caching_control *caching_ctl;
4373 struct btrfs_block_group_cache *cache;
4375 down_write(&fs_info->extent_commit_sem);
4377 list_for_each_entry_safe(caching_ctl, next,
4378 &fs_info->caching_block_groups, list) {
4379 cache = caching_ctl->block_group;
4380 if (block_group_cache_done(cache)) {
4381 cache->last_byte_to_unpin = (u64)-1;
4382 list_del_init(&caching_ctl->list);
4383 put_caching_control(caching_ctl);
4385 cache->last_byte_to_unpin = caching_ctl->progress;
4389 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4390 fs_info->pinned_extents = &fs_info->freed_extents[1];
4392 fs_info->pinned_extents = &fs_info->freed_extents[0];
4394 up_write(&fs_info->extent_commit_sem);
4396 update_global_block_rsv(fs_info);
4400 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4402 struct btrfs_fs_info *fs_info = root->fs_info;
4403 struct btrfs_block_group_cache *cache = NULL;
4406 while (start <= end) {
4408 start >= cache->key.objectid + cache->key.offset) {
4410 btrfs_put_block_group(cache);
4411 cache = btrfs_lookup_block_group(fs_info, start);
4415 len = cache->key.objectid + cache->key.offset - start;
4416 len = min(len, end + 1 - start);
4418 if (start < cache->last_byte_to_unpin) {
4419 len = min(len, cache->last_byte_to_unpin - start);
4420 btrfs_add_free_space(cache, start, len);
4425 spin_lock(&cache->space_info->lock);
4426 spin_lock(&cache->lock);
4427 cache->pinned -= len;
4428 cache->space_info->bytes_pinned -= len;
4430 cache->space_info->bytes_readonly += len;
4431 spin_unlock(&cache->lock);
4432 spin_unlock(&cache->space_info->lock);
4436 btrfs_put_block_group(cache);
4440 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4441 struct btrfs_root *root)
4443 struct btrfs_fs_info *fs_info = root->fs_info;
4444 struct extent_io_tree *unpin;
4449 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4450 unpin = &fs_info->freed_extents[1];
4452 unpin = &fs_info->freed_extents[0];
4455 ret = find_first_extent_bit(unpin, 0, &start, &end,
4460 if (btrfs_test_opt(root, DISCARD))
4461 ret = btrfs_discard_extent(root, start,
4462 end + 1 - start, NULL);
4464 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4465 unpin_extent_range(root, start, end);
4472 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4473 struct btrfs_root *root,
4474 u64 bytenr, u64 num_bytes, u64 parent,
4475 u64 root_objectid, u64 owner_objectid,
4476 u64 owner_offset, int refs_to_drop,
4477 struct btrfs_delayed_extent_op *extent_op)
4479 struct btrfs_key key;
4480 struct btrfs_path *path;
4481 struct btrfs_fs_info *info = root->fs_info;
4482 struct btrfs_root *extent_root = info->extent_root;
4483 struct extent_buffer *leaf;
4484 struct btrfs_extent_item *ei;
4485 struct btrfs_extent_inline_ref *iref;
4488 int extent_slot = 0;
4489 int found_extent = 0;
4494 path = btrfs_alloc_path();
4499 path->leave_spinning = 1;
4501 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4502 BUG_ON(!is_data && refs_to_drop != 1);
4504 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4505 bytenr, num_bytes, parent,
4506 root_objectid, owner_objectid,
4509 extent_slot = path->slots[0];
4510 while (extent_slot >= 0) {
4511 btrfs_item_key_to_cpu(path->nodes[0], &key,
4513 if (key.objectid != bytenr)
4515 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4516 key.offset == num_bytes) {
4520 if (path->slots[0] - extent_slot > 5)
4524 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4525 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4526 if (found_extent && item_size < sizeof(*ei))
4529 if (!found_extent) {
4531 ret = remove_extent_backref(trans, extent_root, path,
4535 btrfs_release_path(path);
4536 path->leave_spinning = 1;
4538 key.objectid = bytenr;
4539 key.type = BTRFS_EXTENT_ITEM_KEY;
4540 key.offset = num_bytes;
4542 ret = btrfs_search_slot(trans, extent_root,
4545 printk(KERN_ERR "umm, got %d back from search"
4546 ", was looking for %llu\n", ret,
4547 (unsigned long long)bytenr);
4549 btrfs_print_leaf(extent_root,
4553 extent_slot = path->slots[0];
4556 btrfs_print_leaf(extent_root, path->nodes[0]);
4558 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4559 "parent %llu root %llu owner %llu offset %llu\n",
4560 (unsigned long long)bytenr,
4561 (unsigned long long)parent,
4562 (unsigned long long)root_objectid,
4563 (unsigned long long)owner_objectid,
4564 (unsigned long long)owner_offset);
4567 leaf = path->nodes[0];
4568 item_size = btrfs_item_size_nr(leaf, extent_slot);
4569 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4570 if (item_size < sizeof(*ei)) {
4571 BUG_ON(found_extent || extent_slot != path->slots[0]);
4572 ret = convert_extent_item_v0(trans, extent_root, path,
4576 btrfs_release_path(path);
4577 path->leave_spinning = 1;
4579 key.objectid = bytenr;
4580 key.type = BTRFS_EXTENT_ITEM_KEY;
4581 key.offset = num_bytes;
4583 ret = btrfs_search_slot(trans, extent_root, &key, path,
4586 printk(KERN_ERR "umm, got %d back from search"
4587 ", was looking for %llu\n", ret,
4588 (unsigned long long)bytenr);
4589 btrfs_print_leaf(extent_root, path->nodes[0]);
4592 extent_slot = path->slots[0];
4593 leaf = path->nodes[0];
4594 item_size = btrfs_item_size_nr(leaf, extent_slot);
4597 BUG_ON(item_size < sizeof(*ei));
4598 ei = btrfs_item_ptr(leaf, extent_slot,
4599 struct btrfs_extent_item);
4600 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4601 struct btrfs_tree_block_info *bi;
4602 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4603 bi = (struct btrfs_tree_block_info *)(ei + 1);
4604 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4607 refs = btrfs_extent_refs(leaf, ei);
4608 BUG_ON(refs < refs_to_drop);
4609 refs -= refs_to_drop;
4613 __run_delayed_extent_op(extent_op, leaf, ei);
4615 * In the case of inline back ref, reference count will
4616 * be updated by remove_extent_backref
4619 BUG_ON(!found_extent);
4621 btrfs_set_extent_refs(leaf, ei, refs);
4622 btrfs_mark_buffer_dirty(leaf);
4625 ret = remove_extent_backref(trans, extent_root, path,
4632 BUG_ON(is_data && refs_to_drop !=
4633 extent_data_ref_count(root, path, iref));
4635 BUG_ON(path->slots[0] != extent_slot);
4637 BUG_ON(path->slots[0] != extent_slot + 1);
4638 path->slots[0] = extent_slot;
4643 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4646 btrfs_release_path(path);
4649 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4652 invalidate_mapping_pages(info->btree_inode->i_mapping,
4653 bytenr >> PAGE_CACHE_SHIFT,
4654 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4657 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4660 btrfs_free_path(path);
4665 * when we free an block, it is possible (and likely) that we free the last
4666 * delayed ref for that extent as well. This searches the delayed ref tree for
4667 * a given extent, and if there are no other delayed refs to be processed, it
4668 * removes it from the tree.
4670 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4671 struct btrfs_root *root, u64 bytenr)
4673 struct btrfs_delayed_ref_head *head;
4674 struct btrfs_delayed_ref_root *delayed_refs;
4675 struct btrfs_delayed_ref_node *ref;
4676 struct rb_node *node;
4679 delayed_refs = &trans->transaction->delayed_refs;
4680 spin_lock(&delayed_refs->lock);
4681 head = btrfs_find_delayed_ref_head(trans, bytenr);
4685 node = rb_prev(&head->node.rb_node);
4689 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4691 /* there are still entries for this ref, we can't drop it */
4692 if (ref->bytenr == bytenr)
4695 if (head->extent_op) {
4696 if (!head->must_insert_reserved)
4698 kfree(head->extent_op);
4699 head->extent_op = NULL;
4703 * waiting for the lock here would deadlock. If someone else has it
4704 * locked they are already in the process of dropping it anyway
4706 if (!mutex_trylock(&head->mutex))
4710 * at this point we have a head with no other entries. Go
4711 * ahead and process it.
4713 head->node.in_tree = 0;
4714 rb_erase(&head->node.rb_node, &delayed_refs->root);
4716 delayed_refs->num_entries--;
4719 * we don't take a ref on the node because we're removing it from the
4720 * tree, so we just steal the ref the tree was holding.
4722 delayed_refs->num_heads--;
4723 if (list_empty(&head->cluster))
4724 delayed_refs->num_heads_ready--;
4726 list_del_init(&head->cluster);
4727 spin_unlock(&delayed_refs->lock);
4729 BUG_ON(head->extent_op);
4730 if (head->must_insert_reserved)
4733 mutex_unlock(&head->mutex);
4734 btrfs_put_delayed_ref(&head->node);
4737 spin_unlock(&delayed_refs->lock);
4741 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4742 struct btrfs_root *root,
4743 struct extent_buffer *buf,
4744 u64 parent, int last_ref)
4746 struct btrfs_block_group_cache *cache = NULL;
4749 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4750 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4751 parent, root->root_key.objectid,
4752 btrfs_header_level(buf),
4753 BTRFS_DROP_DELAYED_REF, NULL);
4760 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4762 if (btrfs_header_generation(buf) == trans->transid) {
4763 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4764 ret = check_ref_cleanup(trans, root, buf->start);
4769 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4770 pin_down_extent(root, cache, buf->start, buf->len, 1);
4774 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4776 btrfs_add_free_space(cache, buf->start, buf->len);
4777 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4781 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4784 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4785 btrfs_put_block_group(cache);
4788 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4789 struct btrfs_root *root,
4790 u64 bytenr, u64 num_bytes, u64 parent,
4791 u64 root_objectid, u64 owner, u64 offset)
4796 * tree log blocks never actually go into the extent allocation
4797 * tree, just update pinning info and exit early.
4799 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4800 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4801 /* unlocks the pinned mutex */
4802 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4804 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4805 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4806 parent, root_objectid, (int)owner,
4807 BTRFS_DROP_DELAYED_REF, NULL);
4810 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4811 parent, root_objectid, owner,
4812 offset, BTRFS_DROP_DELAYED_REF, NULL);
4818 static u64 stripe_align(struct btrfs_root *root, u64 val)
4820 u64 mask = ((u64)root->stripesize - 1);
4821 u64 ret = (val + mask) & ~mask;
4826 * when we wait for progress in the block group caching, its because
4827 * our allocation attempt failed at least once. So, we must sleep
4828 * and let some progress happen before we try again.
4830 * This function will sleep at least once waiting for new free space to
4831 * show up, and then it will check the block group free space numbers
4832 * for our min num_bytes. Another option is to have it go ahead
4833 * and look in the rbtree for a free extent of a given size, but this
4837 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4840 struct btrfs_caching_control *caching_ctl;
4843 caching_ctl = get_caching_control(cache);
4847 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4848 (cache->free_space_ctl->free_space >= num_bytes));
4850 put_caching_control(caching_ctl);
4855 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4857 struct btrfs_caching_control *caching_ctl;
4860 caching_ctl = get_caching_control(cache);
4864 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4866 put_caching_control(caching_ctl);
4870 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4873 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4875 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4877 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4879 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4886 enum btrfs_loop_type {
4887 LOOP_FIND_IDEAL = 0,
4888 LOOP_CACHING_NOWAIT = 1,
4889 LOOP_CACHING_WAIT = 2,
4890 LOOP_ALLOC_CHUNK = 3,
4891 LOOP_NO_EMPTY_SIZE = 4,
4895 * walks the btree of allocated extents and find a hole of a given size.
4896 * The key ins is changed to record the hole:
4897 * ins->objectid == block start
4898 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4899 * ins->offset == number of blocks
4900 * Any available blocks before search_start are skipped.
4902 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4903 struct btrfs_root *orig_root,
4904 u64 num_bytes, u64 empty_size,
4905 u64 search_start, u64 search_end,
4906 u64 hint_byte, struct btrfs_key *ins,
4910 struct btrfs_root *root = orig_root->fs_info->extent_root;
4911 struct btrfs_free_cluster *last_ptr = NULL;
4912 struct btrfs_block_group_cache *block_group = NULL;
4913 int empty_cluster = 2 * 1024 * 1024;
4914 int allowed_chunk_alloc = 0;
4915 int done_chunk_alloc = 0;
4916 struct btrfs_space_info *space_info;
4917 int last_ptr_loop = 0;
4920 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
4921 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
4922 bool found_uncached_bg = false;
4923 bool failed_cluster_refill = false;
4924 bool failed_alloc = false;
4925 bool use_cluster = true;
4926 u64 ideal_cache_percent = 0;
4927 u64 ideal_cache_offset = 0;
4929 WARN_ON(num_bytes < root->sectorsize);
4930 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4934 space_info = __find_space_info(root->fs_info, data);
4936 printk(KERN_ERR "No space info for %llu\n", data);
4941 * If the space info is for both data and metadata it means we have a
4942 * small filesystem and we can't use the clustering stuff.
4944 if (btrfs_mixed_space_info(space_info))
4945 use_cluster = false;
4947 if (orig_root->ref_cows || empty_size)
4948 allowed_chunk_alloc = 1;
4950 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4951 last_ptr = &root->fs_info->meta_alloc_cluster;
4952 if (!btrfs_test_opt(root, SSD))
4953 empty_cluster = 64 * 1024;
4956 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4957 btrfs_test_opt(root, SSD)) {
4958 last_ptr = &root->fs_info->data_alloc_cluster;
4962 spin_lock(&last_ptr->lock);
4963 if (last_ptr->block_group)
4964 hint_byte = last_ptr->window_start;
4965 spin_unlock(&last_ptr->lock);
4968 search_start = max(search_start, first_logical_byte(root, 0));
4969 search_start = max(search_start, hint_byte);
4974 if (search_start == hint_byte) {
4976 block_group = btrfs_lookup_block_group(root->fs_info,
4979 * we don't want to use the block group if it doesn't match our
4980 * allocation bits, or if its not cached.
4982 * However if we are re-searching with an ideal block group
4983 * picked out then we don't care that the block group is cached.
4985 if (block_group && block_group_bits(block_group, data) &&
4986 (block_group->cached != BTRFS_CACHE_NO ||
4987 search_start == ideal_cache_offset)) {
4988 down_read(&space_info->groups_sem);
4989 if (list_empty(&block_group->list) ||
4992 * someone is removing this block group,
4993 * we can't jump into the have_block_group
4994 * target because our list pointers are not
4997 btrfs_put_block_group(block_group);
4998 up_read(&space_info->groups_sem);
5000 index = get_block_group_index(block_group);
5001 goto have_block_group;
5003 } else if (block_group) {
5004 btrfs_put_block_group(block_group);
5008 down_read(&space_info->groups_sem);
5009 list_for_each_entry(block_group, &space_info->block_groups[index],
5014 btrfs_get_block_group(block_group);
5015 search_start = block_group->key.objectid;
5018 * this can happen if we end up cycling through all the
5019 * raid types, but we want to make sure we only allocate
5020 * for the proper type.
5022 if (!block_group_bits(block_group, data)) {
5023 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5024 BTRFS_BLOCK_GROUP_RAID1 |
5025 BTRFS_BLOCK_GROUP_RAID10;
5028 * if they asked for extra copies and this block group
5029 * doesn't provide them, bail. This does allow us to
5030 * fill raid0 from raid1.
5032 if ((data & extra) && !(block_group->flags & extra))
5037 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5040 ret = cache_block_group(block_group, trans,
5042 if (block_group->cached == BTRFS_CACHE_FINISHED)
5043 goto have_block_group;
5045 free_percent = btrfs_block_group_used(&block_group->item);
5046 free_percent *= 100;
5047 free_percent = div64_u64(free_percent,
5048 block_group->key.offset);
5049 free_percent = 100 - free_percent;
5050 if (free_percent > ideal_cache_percent &&
5051 likely(!block_group->ro)) {
5052 ideal_cache_offset = block_group->key.objectid;
5053 ideal_cache_percent = free_percent;
5057 * The caching workers are limited to 2 threads, so we
5058 * can queue as much work as we care to.
5060 if (loop > LOOP_FIND_IDEAL) {
5061 ret = cache_block_group(block_group, trans,
5065 found_uncached_bg = true;
5068 * If loop is set for cached only, try the next block
5071 if (loop == LOOP_FIND_IDEAL)
5075 cached = block_group_cache_done(block_group);
5076 if (unlikely(!cached))
5077 found_uncached_bg = true;
5079 if (unlikely(block_group->ro))
5082 spin_lock(&block_group->free_space_ctl->tree_lock);
5084 block_group->free_space_ctl->free_space <
5085 num_bytes + empty_size) {
5086 spin_unlock(&block_group->free_space_ctl->tree_lock);
5089 spin_unlock(&block_group->free_space_ctl->tree_lock);
5092 * Ok we want to try and use the cluster allocator, so lets look
5093 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5094 * have tried the cluster allocator plenty of times at this
5095 * point and not have found anything, so we are likely way too
5096 * fragmented for the clustering stuff to find anything, so lets
5097 * just skip it and let the allocator find whatever block it can
5100 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5102 * the refill lock keeps out other
5103 * people trying to start a new cluster
5105 spin_lock(&last_ptr->refill_lock);
5106 if (last_ptr->block_group &&
5107 (last_ptr->block_group->ro ||
5108 !block_group_bits(last_ptr->block_group, data))) {
5110 goto refill_cluster;
5113 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5114 num_bytes, search_start);
5116 /* we have a block, we're done */
5117 spin_unlock(&last_ptr->refill_lock);
5121 spin_lock(&last_ptr->lock);
5123 * whoops, this cluster doesn't actually point to
5124 * this block group. Get a ref on the block
5125 * group is does point to and try again
5127 if (!last_ptr_loop && last_ptr->block_group &&
5128 last_ptr->block_group != block_group &&
5130 get_block_group_index(last_ptr->block_group)) {
5132 btrfs_put_block_group(block_group);
5133 block_group = last_ptr->block_group;
5134 btrfs_get_block_group(block_group);
5135 spin_unlock(&last_ptr->lock);
5136 spin_unlock(&last_ptr->refill_lock);
5139 search_start = block_group->key.objectid;
5141 * we know this block group is properly
5142 * in the list because
5143 * btrfs_remove_block_group, drops the
5144 * cluster before it removes the block
5145 * group from the list
5147 goto have_block_group;
5149 spin_unlock(&last_ptr->lock);
5152 * this cluster didn't work out, free it and
5155 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5159 /* allocate a cluster in this block group */
5160 ret = btrfs_find_space_cluster(trans, root,
5161 block_group, last_ptr,
5163 empty_cluster + empty_size);
5166 * now pull our allocation out of this
5169 offset = btrfs_alloc_from_cluster(block_group,
5170 last_ptr, num_bytes,
5173 /* we found one, proceed */
5174 spin_unlock(&last_ptr->refill_lock);
5177 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5178 && !failed_cluster_refill) {
5179 spin_unlock(&last_ptr->refill_lock);
5181 failed_cluster_refill = true;
5182 wait_block_group_cache_progress(block_group,
5183 num_bytes + empty_cluster + empty_size);
5184 goto have_block_group;
5188 * at this point we either didn't find a cluster
5189 * or we weren't able to allocate a block from our
5190 * cluster. Free the cluster we've been trying
5191 * to use, and go to the next block group
5193 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5194 spin_unlock(&last_ptr->refill_lock);
5198 offset = btrfs_find_space_for_alloc(block_group, search_start,
5199 num_bytes, empty_size);
5201 * If we didn't find a chunk, and we haven't failed on this
5202 * block group before, and this block group is in the middle of
5203 * caching and we are ok with waiting, then go ahead and wait
5204 * for progress to be made, and set failed_alloc to true.
5206 * If failed_alloc is true then we've already waited on this
5207 * block group once and should move on to the next block group.
5209 if (!offset && !failed_alloc && !cached &&
5210 loop > LOOP_CACHING_NOWAIT) {
5211 wait_block_group_cache_progress(block_group,
5212 num_bytes + empty_size);
5213 failed_alloc = true;
5214 goto have_block_group;
5215 } else if (!offset) {
5219 search_start = stripe_align(root, offset);
5220 /* move on to the next group */
5221 if (search_start + num_bytes >= search_end) {
5222 btrfs_add_free_space(block_group, offset, num_bytes);
5226 /* move on to the next group */
5227 if (search_start + num_bytes >
5228 block_group->key.objectid + block_group->key.offset) {
5229 btrfs_add_free_space(block_group, offset, num_bytes);
5233 ins->objectid = search_start;
5234 ins->offset = num_bytes;
5236 if (offset < search_start)
5237 btrfs_add_free_space(block_group, offset,
5238 search_start - offset);
5239 BUG_ON(offset > search_start);
5241 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5243 if (ret == -EAGAIN) {
5244 btrfs_add_free_space(block_group, offset, num_bytes);
5248 /* we are all good, lets return */
5249 ins->objectid = search_start;
5250 ins->offset = num_bytes;
5252 if (offset < search_start)
5253 btrfs_add_free_space(block_group, offset,
5254 search_start - offset);
5255 BUG_ON(offset > search_start);
5256 btrfs_put_block_group(block_group);
5259 failed_cluster_refill = false;
5260 failed_alloc = false;
5261 BUG_ON(index != get_block_group_index(block_group));
5262 btrfs_put_block_group(block_group);
5264 up_read(&space_info->groups_sem);
5266 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5269 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5270 * for them to make caching progress. Also
5271 * determine the best possible bg to cache
5272 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5273 * caching kthreads as we move along
5274 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5275 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5276 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5279 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5281 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5282 found_uncached_bg = false;
5284 if (!ideal_cache_percent)
5288 * 1 of the following 2 things have happened so far
5290 * 1) We found an ideal block group for caching that
5291 * is mostly full and will cache quickly, so we might
5292 * as well wait for it.
5294 * 2) We searched for cached only and we didn't find
5295 * anything, and we didn't start any caching kthreads
5296 * either, so chances are we will loop through and
5297 * start a couple caching kthreads, and then come back
5298 * around and just wait for them. This will be slower
5299 * because we will have 2 caching kthreads reading at
5300 * the same time when we could have just started one
5301 * and waited for it to get far enough to give us an
5302 * allocation, so go ahead and go to the wait caching
5305 loop = LOOP_CACHING_WAIT;
5306 search_start = ideal_cache_offset;
5307 ideal_cache_percent = 0;
5309 } else if (loop == LOOP_FIND_IDEAL) {
5311 * Didn't find a uncached bg, wait on anything we find
5314 loop = LOOP_CACHING_WAIT;
5320 if (loop == LOOP_ALLOC_CHUNK) {
5321 if (allowed_chunk_alloc) {
5322 ret = do_chunk_alloc(trans, root, num_bytes +
5323 2 * 1024 * 1024, data,
5324 CHUNK_ALLOC_LIMITED);
5325 allowed_chunk_alloc = 0;
5327 done_chunk_alloc = 1;
5328 } else if (!done_chunk_alloc &&
5329 space_info->force_alloc ==
5330 CHUNK_ALLOC_NO_FORCE) {
5331 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5335 * We didn't allocate a chunk, go ahead and drop the
5336 * empty size and loop again.
5338 if (!done_chunk_alloc)
5339 loop = LOOP_NO_EMPTY_SIZE;
5342 if (loop == LOOP_NO_EMPTY_SIZE) {
5348 } else if (!ins->objectid) {
5350 } else if (ins->objectid) {
5357 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5358 int dump_block_groups)
5360 struct btrfs_block_group_cache *cache;
5363 spin_lock(&info->lock);
5364 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5365 (unsigned long long)info->flags,
5366 (unsigned long long)(info->total_bytes - info->bytes_used -
5367 info->bytes_pinned - info->bytes_reserved -
5368 info->bytes_readonly),
5369 (info->full) ? "" : "not ");
5370 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5371 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5372 (unsigned long long)info->total_bytes,
5373 (unsigned long long)info->bytes_used,
5374 (unsigned long long)info->bytes_pinned,
5375 (unsigned long long)info->bytes_reserved,
5376 (unsigned long long)info->bytes_may_use,
5377 (unsigned long long)info->bytes_readonly);
5378 spin_unlock(&info->lock);
5380 if (!dump_block_groups)
5383 down_read(&info->groups_sem);
5385 list_for_each_entry(cache, &info->block_groups[index], list) {
5386 spin_lock(&cache->lock);
5387 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5388 "%llu pinned %llu reserved\n",
5389 (unsigned long long)cache->key.objectid,
5390 (unsigned long long)cache->key.offset,
5391 (unsigned long long)btrfs_block_group_used(&cache->item),
5392 (unsigned long long)cache->pinned,
5393 (unsigned long long)cache->reserved);
5394 btrfs_dump_free_space(cache, bytes);
5395 spin_unlock(&cache->lock);
5397 if (++index < BTRFS_NR_RAID_TYPES)
5399 up_read(&info->groups_sem);
5402 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5403 struct btrfs_root *root,
5404 u64 num_bytes, u64 min_alloc_size,
5405 u64 empty_size, u64 hint_byte,
5406 u64 search_end, struct btrfs_key *ins,
5410 u64 search_start = 0;
5412 data = btrfs_get_alloc_profile(root, data);
5415 * the only place that sets empty_size is btrfs_realloc_node, which
5416 * is not called recursively on allocations
5418 if (empty_size || root->ref_cows)
5419 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5420 num_bytes + 2 * 1024 * 1024, data,
5421 CHUNK_ALLOC_NO_FORCE);
5423 WARN_ON(num_bytes < root->sectorsize);
5424 ret = find_free_extent(trans, root, num_bytes, empty_size,
5425 search_start, search_end, hint_byte,
5428 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5429 num_bytes = num_bytes >> 1;
5430 num_bytes = num_bytes & ~(root->sectorsize - 1);
5431 num_bytes = max(num_bytes, min_alloc_size);
5432 do_chunk_alloc(trans, root->fs_info->extent_root,
5433 num_bytes, data, CHUNK_ALLOC_FORCE);
5436 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5437 struct btrfs_space_info *sinfo;
5439 sinfo = __find_space_info(root->fs_info, data);
5440 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5441 "wanted %llu\n", (unsigned long long)data,
5442 (unsigned long long)num_bytes);
5443 dump_space_info(sinfo, num_bytes, 1);
5446 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5451 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5453 struct btrfs_block_group_cache *cache;
5456 cache = btrfs_lookup_block_group(root->fs_info, start);
5458 printk(KERN_ERR "Unable to find block group for %llu\n",
5459 (unsigned long long)start);
5463 if (btrfs_test_opt(root, DISCARD))
5464 ret = btrfs_discard_extent(root, start, len, NULL);
5466 btrfs_add_free_space(cache, start, len);
5467 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5468 btrfs_put_block_group(cache);
5470 trace_btrfs_reserved_extent_free(root, start, len);
5475 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5476 struct btrfs_root *root,
5477 u64 parent, u64 root_objectid,
5478 u64 flags, u64 owner, u64 offset,
5479 struct btrfs_key *ins, int ref_mod)
5482 struct btrfs_fs_info *fs_info = root->fs_info;
5483 struct btrfs_extent_item *extent_item;
5484 struct btrfs_extent_inline_ref *iref;
5485 struct btrfs_path *path;
5486 struct extent_buffer *leaf;
5491 type = BTRFS_SHARED_DATA_REF_KEY;
5493 type = BTRFS_EXTENT_DATA_REF_KEY;
5495 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5497 path = btrfs_alloc_path();
5501 path->leave_spinning = 1;
5502 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5506 leaf = path->nodes[0];
5507 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5508 struct btrfs_extent_item);
5509 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5510 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5511 btrfs_set_extent_flags(leaf, extent_item,
5512 flags | BTRFS_EXTENT_FLAG_DATA);
5514 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5515 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5517 struct btrfs_shared_data_ref *ref;
5518 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5519 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5520 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5522 struct btrfs_extent_data_ref *ref;
5523 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5524 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5525 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5526 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5527 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5530 btrfs_mark_buffer_dirty(path->nodes[0]);
5531 btrfs_free_path(path);
5533 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5535 printk(KERN_ERR "btrfs update block group failed for %llu "
5536 "%llu\n", (unsigned long long)ins->objectid,
5537 (unsigned long long)ins->offset);
5543 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5544 struct btrfs_root *root,
5545 u64 parent, u64 root_objectid,
5546 u64 flags, struct btrfs_disk_key *key,
5547 int level, struct btrfs_key *ins)
5550 struct btrfs_fs_info *fs_info = root->fs_info;
5551 struct btrfs_extent_item *extent_item;
5552 struct btrfs_tree_block_info *block_info;
5553 struct btrfs_extent_inline_ref *iref;
5554 struct btrfs_path *path;
5555 struct extent_buffer *leaf;
5556 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5558 path = btrfs_alloc_path();
5562 path->leave_spinning = 1;
5563 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5567 leaf = path->nodes[0];
5568 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5569 struct btrfs_extent_item);
5570 btrfs_set_extent_refs(leaf, extent_item, 1);
5571 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5572 btrfs_set_extent_flags(leaf, extent_item,
5573 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5574 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5576 btrfs_set_tree_block_key(leaf, block_info, key);
5577 btrfs_set_tree_block_level(leaf, block_info, level);
5579 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5581 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5582 btrfs_set_extent_inline_ref_type(leaf, iref,
5583 BTRFS_SHARED_BLOCK_REF_KEY);
5584 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5586 btrfs_set_extent_inline_ref_type(leaf, iref,
5587 BTRFS_TREE_BLOCK_REF_KEY);
5588 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5591 btrfs_mark_buffer_dirty(leaf);
5592 btrfs_free_path(path);
5594 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5596 printk(KERN_ERR "btrfs update block group failed for %llu "
5597 "%llu\n", (unsigned long long)ins->objectid,
5598 (unsigned long long)ins->offset);
5604 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5605 struct btrfs_root *root,
5606 u64 root_objectid, u64 owner,
5607 u64 offset, struct btrfs_key *ins)
5611 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5613 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5614 0, root_objectid, owner, offset,
5615 BTRFS_ADD_DELAYED_EXTENT, NULL);
5620 * this is used by the tree logging recovery code. It records that
5621 * an extent has been allocated and makes sure to clear the free
5622 * space cache bits as well
5624 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5625 struct btrfs_root *root,
5626 u64 root_objectid, u64 owner, u64 offset,
5627 struct btrfs_key *ins)
5630 struct btrfs_block_group_cache *block_group;
5631 struct btrfs_caching_control *caching_ctl;
5632 u64 start = ins->objectid;
5633 u64 num_bytes = ins->offset;
5635 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5636 cache_block_group(block_group, trans, NULL, 0);
5637 caching_ctl = get_caching_control(block_group);
5640 BUG_ON(!block_group_cache_done(block_group));
5641 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5644 mutex_lock(&caching_ctl->mutex);
5646 if (start >= caching_ctl->progress) {
5647 ret = add_excluded_extent(root, start, num_bytes);
5649 } else if (start + num_bytes <= caching_ctl->progress) {
5650 ret = btrfs_remove_free_space(block_group,
5654 num_bytes = caching_ctl->progress - start;
5655 ret = btrfs_remove_free_space(block_group,
5659 start = caching_ctl->progress;
5660 num_bytes = ins->objectid + ins->offset -
5661 caching_ctl->progress;
5662 ret = add_excluded_extent(root, start, num_bytes);
5666 mutex_unlock(&caching_ctl->mutex);
5667 put_caching_control(caching_ctl);
5670 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5671 RESERVE_ALLOC_NO_ACCOUNT);
5673 btrfs_put_block_group(block_group);
5674 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5675 0, owner, offset, ins, 1);
5679 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5680 struct btrfs_root *root,
5681 u64 bytenr, u32 blocksize,
5684 struct extent_buffer *buf;
5686 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5688 return ERR_PTR(-ENOMEM);
5689 btrfs_set_header_generation(buf, trans->transid);
5690 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5691 btrfs_tree_lock(buf);
5692 clean_tree_block(trans, root, buf);
5694 btrfs_set_lock_blocking(buf);
5695 btrfs_set_buffer_uptodate(buf);
5697 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5699 * we allow two log transactions at a time, use different
5700 * EXENT bit to differentiate dirty pages.
5702 if (root->log_transid % 2 == 0)
5703 set_extent_dirty(&root->dirty_log_pages, buf->start,
5704 buf->start + buf->len - 1, GFP_NOFS);
5706 set_extent_new(&root->dirty_log_pages, buf->start,
5707 buf->start + buf->len - 1, GFP_NOFS);
5709 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5710 buf->start + buf->len - 1, GFP_NOFS);
5712 trans->blocks_used++;
5713 /* this returns a buffer locked for blocking */
5717 static struct btrfs_block_rsv *
5718 use_block_rsv(struct btrfs_trans_handle *trans,
5719 struct btrfs_root *root, u32 blocksize)
5721 struct btrfs_block_rsv *block_rsv;
5722 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5725 block_rsv = get_block_rsv(trans, root);
5727 if (block_rsv->size == 0) {
5728 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5730 * If we couldn't reserve metadata bytes try and use some from
5731 * the global reserve.
5733 if (ret && block_rsv != global_rsv) {
5734 ret = block_rsv_use_bytes(global_rsv, blocksize);
5737 return ERR_PTR(ret);
5739 return ERR_PTR(ret);
5744 ret = block_rsv_use_bytes(block_rsv, blocksize);
5749 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5752 } else if (ret && block_rsv != global_rsv) {
5753 ret = block_rsv_use_bytes(global_rsv, blocksize);
5759 return ERR_PTR(-ENOSPC);
5762 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5764 block_rsv_add_bytes(block_rsv, blocksize, 0);
5765 block_rsv_release_bytes(block_rsv, NULL, 0);
5769 * finds a free extent and does all the dirty work required for allocation
5770 * returns the key for the extent through ins, and a tree buffer for
5771 * the first block of the extent through buf.
5773 * returns the tree buffer or NULL.
5775 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5776 struct btrfs_root *root, u32 blocksize,
5777 u64 parent, u64 root_objectid,
5778 struct btrfs_disk_key *key, int level,
5779 u64 hint, u64 empty_size)
5781 struct btrfs_key ins;
5782 struct btrfs_block_rsv *block_rsv;
5783 struct extent_buffer *buf;
5788 block_rsv = use_block_rsv(trans, root, blocksize);
5789 if (IS_ERR(block_rsv))
5790 return ERR_CAST(block_rsv);
5792 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5793 empty_size, hint, (u64)-1, &ins, 0);
5795 unuse_block_rsv(block_rsv, blocksize);
5796 return ERR_PTR(ret);
5799 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5801 BUG_ON(IS_ERR(buf));
5803 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5805 parent = ins.objectid;
5806 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5810 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5811 struct btrfs_delayed_extent_op *extent_op;
5812 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5815 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5817 memset(&extent_op->key, 0, sizeof(extent_op->key));
5818 extent_op->flags_to_set = flags;
5819 extent_op->update_key = 1;
5820 extent_op->update_flags = 1;
5821 extent_op->is_data = 0;
5823 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5824 ins.offset, parent, root_objectid,
5825 level, BTRFS_ADD_DELAYED_EXTENT,
5832 struct walk_control {
5833 u64 refs[BTRFS_MAX_LEVEL];
5834 u64 flags[BTRFS_MAX_LEVEL];
5835 struct btrfs_key update_progress;
5845 #define DROP_REFERENCE 1
5846 #define UPDATE_BACKREF 2
5848 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5849 struct btrfs_root *root,
5850 struct walk_control *wc,
5851 struct btrfs_path *path)
5859 struct btrfs_key key;
5860 struct extent_buffer *eb;
5865 if (path->slots[wc->level] < wc->reada_slot) {
5866 wc->reada_count = wc->reada_count * 2 / 3;
5867 wc->reada_count = max(wc->reada_count, 2);
5869 wc->reada_count = wc->reada_count * 3 / 2;
5870 wc->reada_count = min_t(int, wc->reada_count,
5871 BTRFS_NODEPTRS_PER_BLOCK(root));
5874 eb = path->nodes[wc->level];
5875 nritems = btrfs_header_nritems(eb);
5876 blocksize = btrfs_level_size(root, wc->level - 1);
5878 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5879 if (nread >= wc->reada_count)
5883 bytenr = btrfs_node_blockptr(eb, slot);
5884 generation = btrfs_node_ptr_generation(eb, slot);
5886 if (slot == path->slots[wc->level])
5889 if (wc->stage == UPDATE_BACKREF &&
5890 generation <= root->root_key.offset)
5893 /* We don't lock the tree block, it's OK to be racy here */
5894 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5899 if (wc->stage == DROP_REFERENCE) {
5903 if (wc->level == 1 &&
5904 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5906 if (!wc->update_ref ||
5907 generation <= root->root_key.offset)
5909 btrfs_node_key_to_cpu(eb, &key, slot);
5910 ret = btrfs_comp_cpu_keys(&key,
5911 &wc->update_progress);
5915 if (wc->level == 1 &&
5916 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5920 ret = readahead_tree_block(root, bytenr, blocksize,
5926 wc->reada_slot = slot;
5930 * hepler to process tree block while walking down the tree.
5932 * when wc->stage == UPDATE_BACKREF, this function updates
5933 * back refs for pointers in the block.
5935 * NOTE: return value 1 means we should stop walking down.
5937 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5938 struct btrfs_root *root,
5939 struct btrfs_path *path,
5940 struct walk_control *wc, int lookup_info)
5942 int level = wc->level;
5943 struct extent_buffer *eb = path->nodes[level];
5944 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5947 if (wc->stage == UPDATE_BACKREF &&
5948 btrfs_header_owner(eb) != root->root_key.objectid)
5952 * when reference count of tree block is 1, it won't increase
5953 * again. once full backref flag is set, we never clear it.
5956 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5957 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5958 BUG_ON(!path->locks[level]);
5959 ret = btrfs_lookup_extent_info(trans, root,
5964 BUG_ON(wc->refs[level] == 0);
5967 if (wc->stage == DROP_REFERENCE) {
5968 if (wc->refs[level] > 1)
5971 if (path->locks[level] && !wc->keep_locks) {
5972 btrfs_tree_unlock_rw(eb, path->locks[level]);
5973 path->locks[level] = 0;
5978 /* wc->stage == UPDATE_BACKREF */
5979 if (!(wc->flags[level] & flag)) {
5980 BUG_ON(!path->locks[level]);
5981 ret = btrfs_inc_ref(trans, root, eb, 1);
5983 ret = btrfs_dec_ref(trans, root, eb, 0);
5985 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5988 wc->flags[level] |= flag;
5992 * the block is shared by multiple trees, so it's not good to
5993 * keep the tree lock
5995 if (path->locks[level] && level > 0) {
5996 btrfs_tree_unlock_rw(eb, path->locks[level]);
5997 path->locks[level] = 0;
6003 * hepler to process tree block pointer.
6005 * when wc->stage == DROP_REFERENCE, this function checks
6006 * reference count of the block pointed to. if the block
6007 * is shared and we need update back refs for the subtree
6008 * rooted at the block, this function changes wc->stage to
6009 * UPDATE_BACKREF. if the block is shared and there is no
6010 * need to update back, this function drops the reference
6013 * NOTE: return value 1 means we should stop walking down.
6015 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6016 struct btrfs_root *root,
6017 struct btrfs_path *path,
6018 struct walk_control *wc, int *lookup_info)
6024 struct btrfs_key key;
6025 struct extent_buffer *next;
6026 int level = wc->level;
6030 generation = btrfs_node_ptr_generation(path->nodes[level],
6031 path->slots[level]);
6033 * if the lower level block was created before the snapshot
6034 * was created, we know there is no need to update back refs
6037 if (wc->stage == UPDATE_BACKREF &&
6038 generation <= root->root_key.offset) {
6043 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6044 blocksize = btrfs_level_size(root, level - 1);
6046 next = btrfs_find_tree_block(root, bytenr, blocksize);
6048 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6053 btrfs_tree_lock(next);
6054 btrfs_set_lock_blocking(next);
6056 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6057 &wc->refs[level - 1],
6058 &wc->flags[level - 1]);
6060 BUG_ON(wc->refs[level - 1] == 0);
6063 if (wc->stage == DROP_REFERENCE) {
6064 if (wc->refs[level - 1] > 1) {
6066 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6069 if (!wc->update_ref ||
6070 generation <= root->root_key.offset)
6073 btrfs_node_key_to_cpu(path->nodes[level], &key,
6074 path->slots[level]);
6075 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6079 wc->stage = UPDATE_BACKREF;
6080 wc->shared_level = level - 1;
6084 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6088 if (!btrfs_buffer_uptodate(next, generation)) {
6089 btrfs_tree_unlock(next);
6090 free_extent_buffer(next);
6096 if (reada && level == 1)
6097 reada_walk_down(trans, root, wc, path);
6098 next = read_tree_block(root, bytenr, blocksize, generation);
6101 btrfs_tree_lock(next);
6102 btrfs_set_lock_blocking(next);
6106 BUG_ON(level != btrfs_header_level(next));
6107 path->nodes[level] = next;
6108 path->slots[level] = 0;
6109 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6115 wc->refs[level - 1] = 0;
6116 wc->flags[level - 1] = 0;
6117 if (wc->stage == DROP_REFERENCE) {
6118 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6119 parent = path->nodes[level]->start;
6121 BUG_ON(root->root_key.objectid !=
6122 btrfs_header_owner(path->nodes[level]));
6126 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6127 root->root_key.objectid, level - 1, 0);
6130 btrfs_tree_unlock(next);
6131 free_extent_buffer(next);
6137 * hepler to process tree block while walking up the tree.
6139 * when wc->stage == DROP_REFERENCE, this function drops
6140 * reference count on the block.
6142 * when wc->stage == UPDATE_BACKREF, this function changes
6143 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6144 * to UPDATE_BACKREF previously while processing the block.
6146 * NOTE: return value 1 means we should stop walking up.
6148 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6149 struct btrfs_root *root,
6150 struct btrfs_path *path,
6151 struct walk_control *wc)
6154 int level = wc->level;
6155 struct extent_buffer *eb = path->nodes[level];
6158 if (wc->stage == UPDATE_BACKREF) {
6159 BUG_ON(wc->shared_level < level);
6160 if (level < wc->shared_level)
6163 ret = find_next_key(path, level + 1, &wc->update_progress);
6167 wc->stage = DROP_REFERENCE;
6168 wc->shared_level = -1;
6169 path->slots[level] = 0;
6172 * check reference count again if the block isn't locked.
6173 * we should start walking down the tree again if reference
6176 if (!path->locks[level]) {
6178 btrfs_tree_lock(eb);
6179 btrfs_set_lock_blocking(eb);
6180 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6182 ret = btrfs_lookup_extent_info(trans, root,
6187 BUG_ON(wc->refs[level] == 0);
6188 if (wc->refs[level] == 1) {
6189 btrfs_tree_unlock_rw(eb, path->locks[level]);
6195 /* wc->stage == DROP_REFERENCE */
6196 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6198 if (wc->refs[level] == 1) {
6200 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6201 ret = btrfs_dec_ref(trans, root, eb, 1);
6203 ret = btrfs_dec_ref(trans, root, eb, 0);
6206 /* make block locked assertion in clean_tree_block happy */
6207 if (!path->locks[level] &&
6208 btrfs_header_generation(eb) == trans->transid) {
6209 btrfs_tree_lock(eb);
6210 btrfs_set_lock_blocking(eb);
6211 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6213 clean_tree_block(trans, root, eb);
6216 if (eb == root->node) {
6217 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6220 BUG_ON(root->root_key.objectid !=
6221 btrfs_header_owner(eb));
6223 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6224 parent = path->nodes[level + 1]->start;
6226 BUG_ON(root->root_key.objectid !=
6227 btrfs_header_owner(path->nodes[level + 1]));
6230 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6232 wc->refs[level] = 0;
6233 wc->flags[level] = 0;
6237 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6238 struct btrfs_root *root,
6239 struct btrfs_path *path,
6240 struct walk_control *wc)
6242 int level = wc->level;
6243 int lookup_info = 1;
6246 while (level >= 0) {
6247 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6254 if (path->slots[level] >=
6255 btrfs_header_nritems(path->nodes[level]))
6258 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6260 path->slots[level]++;
6269 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6270 struct btrfs_root *root,
6271 struct btrfs_path *path,
6272 struct walk_control *wc, int max_level)
6274 int level = wc->level;
6277 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6278 while (level < max_level && path->nodes[level]) {
6280 if (path->slots[level] + 1 <
6281 btrfs_header_nritems(path->nodes[level])) {
6282 path->slots[level]++;
6285 ret = walk_up_proc(trans, root, path, wc);
6289 if (path->locks[level]) {
6290 btrfs_tree_unlock_rw(path->nodes[level],
6291 path->locks[level]);
6292 path->locks[level] = 0;
6294 free_extent_buffer(path->nodes[level]);
6295 path->nodes[level] = NULL;
6303 * drop a subvolume tree.
6305 * this function traverses the tree freeing any blocks that only
6306 * referenced by the tree.
6308 * when a shared tree block is found. this function decreases its
6309 * reference count by one. if update_ref is true, this function
6310 * also make sure backrefs for the shared block and all lower level
6311 * blocks are properly updated.
6313 void btrfs_drop_snapshot(struct btrfs_root *root,
6314 struct btrfs_block_rsv *block_rsv, int update_ref)
6316 struct btrfs_path *path;
6317 struct btrfs_trans_handle *trans;
6318 struct btrfs_root *tree_root = root->fs_info->tree_root;
6319 struct btrfs_root_item *root_item = &root->root_item;
6320 struct walk_control *wc;
6321 struct btrfs_key key;
6326 path = btrfs_alloc_path();
6332 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6334 btrfs_free_path(path);
6339 trans = btrfs_start_transaction(tree_root, 0);
6340 BUG_ON(IS_ERR(trans));
6343 trans->block_rsv = block_rsv;
6345 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6346 level = btrfs_header_level(root->node);
6347 path->nodes[level] = btrfs_lock_root_node(root);
6348 btrfs_set_lock_blocking(path->nodes[level]);
6349 path->slots[level] = 0;
6350 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6351 memset(&wc->update_progress, 0,
6352 sizeof(wc->update_progress));
6354 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6355 memcpy(&wc->update_progress, &key,
6356 sizeof(wc->update_progress));
6358 level = root_item->drop_level;
6360 path->lowest_level = level;
6361 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6362 path->lowest_level = 0;
6370 * unlock our path, this is safe because only this
6371 * function is allowed to delete this snapshot
6373 btrfs_unlock_up_safe(path, 0);
6375 level = btrfs_header_level(root->node);
6377 btrfs_tree_lock(path->nodes[level]);
6378 btrfs_set_lock_blocking(path->nodes[level]);
6380 ret = btrfs_lookup_extent_info(trans, root,
6381 path->nodes[level]->start,
6382 path->nodes[level]->len,
6386 BUG_ON(wc->refs[level] == 0);
6388 if (level == root_item->drop_level)
6391 btrfs_tree_unlock(path->nodes[level]);
6392 WARN_ON(wc->refs[level] != 1);
6398 wc->shared_level = -1;
6399 wc->stage = DROP_REFERENCE;
6400 wc->update_ref = update_ref;
6402 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6405 ret = walk_down_tree(trans, root, path, wc);
6411 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6418 BUG_ON(wc->stage != DROP_REFERENCE);
6422 if (wc->stage == DROP_REFERENCE) {
6424 btrfs_node_key(path->nodes[level],
6425 &root_item->drop_progress,
6426 path->slots[level]);
6427 root_item->drop_level = level;
6430 BUG_ON(wc->level == 0);
6431 if (btrfs_should_end_transaction(trans, tree_root)) {
6432 ret = btrfs_update_root(trans, tree_root,
6437 btrfs_end_transaction_throttle(trans, tree_root);
6438 trans = btrfs_start_transaction(tree_root, 0);
6439 BUG_ON(IS_ERR(trans));
6441 trans->block_rsv = block_rsv;
6444 btrfs_release_path(path);
6447 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6450 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6451 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6455 /* if we fail to delete the orphan item this time
6456 * around, it'll get picked up the next time.
6458 * The most common failure here is just -ENOENT.
6460 btrfs_del_orphan_item(trans, tree_root,
6461 root->root_key.objectid);
6465 if (root->in_radix) {
6466 btrfs_free_fs_root(tree_root->fs_info, root);
6468 free_extent_buffer(root->node);
6469 free_extent_buffer(root->commit_root);
6473 btrfs_end_transaction_throttle(trans, tree_root);
6475 btrfs_free_path(path);
6478 btrfs_std_error(root->fs_info, err);
6483 * drop subtree rooted at tree block 'node'.
6485 * NOTE: this function will unlock and release tree block 'node'
6487 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6488 struct btrfs_root *root,
6489 struct extent_buffer *node,
6490 struct extent_buffer *parent)
6492 struct btrfs_path *path;
6493 struct walk_control *wc;
6499 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6501 path = btrfs_alloc_path();
6505 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6507 btrfs_free_path(path);
6511 btrfs_assert_tree_locked(parent);
6512 parent_level = btrfs_header_level(parent);
6513 extent_buffer_get(parent);
6514 path->nodes[parent_level] = parent;
6515 path->slots[parent_level] = btrfs_header_nritems(parent);
6517 btrfs_assert_tree_locked(node);
6518 level = btrfs_header_level(node);
6519 path->nodes[level] = node;
6520 path->slots[level] = 0;
6521 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6523 wc->refs[parent_level] = 1;
6524 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6526 wc->shared_level = -1;
6527 wc->stage = DROP_REFERENCE;
6530 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6533 wret = walk_down_tree(trans, root, path, wc);
6539 wret = walk_up_tree(trans, root, path, wc, parent_level);
6547 btrfs_free_path(path);
6551 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6554 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6555 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6558 * we add in the count of missing devices because we want
6559 * to make sure that any RAID levels on a degraded FS
6560 * continue to be honored.
6562 num_devices = root->fs_info->fs_devices->rw_devices +
6563 root->fs_info->fs_devices->missing_devices;
6565 if (num_devices == 1) {
6566 stripped |= BTRFS_BLOCK_GROUP_DUP;
6567 stripped = flags & ~stripped;
6569 /* turn raid0 into single device chunks */
6570 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6573 /* turn mirroring into duplication */
6574 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6575 BTRFS_BLOCK_GROUP_RAID10))
6576 return stripped | BTRFS_BLOCK_GROUP_DUP;
6579 /* they already had raid on here, just return */
6580 if (flags & stripped)
6583 stripped |= BTRFS_BLOCK_GROUP_DUP;
6584 stripped = flags & ~stripped;
6586 /* switch duplicated blocks with raid1 */
6587 if (flags & BTRFS_BLOCK_GROUP_DUP)
6588 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6590 /* turn single device chunks into raid0 */
6591 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6596 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6598 struct btrfs_space_info *sinfo = cache->space_info;
6600 u64 min_allocable_bytes;
6605 * We need some metadata space and system metadata space for
6606 * allocating chunks in some corner cases until we force to set
6607 * it to be readonly.
6610 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6612 min_allocable_bytes = 1 * 1024 * 1024;
6614 min_allocable_bytes = 0;
6616 spin_lock(&sinfo->lock);
6617 spin_lock(&cache->lock);
6624 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6625 cache->bytes_super - btrfs_block_group_used(&cache->item);
6627 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6628 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6629 min_allocable_bytes <= sinfo->total_bytes) {
6630 sinfo->bytes_readonly += num_bytes;
6635 spin_unlock(&cache->lock);
6636 spin_unlock(&sinfo->lock);
6640 int btrfs_set_block_group_ro(struct btrfs_root *root,
6641 struct btrfs_block_group_cache *cache)
6644 struct btrfs_trans_handle *trans;
6650 trans = btrfs_join_transaction(root);
6651 BUG_ON(IS_ERR(trans));
6653 alloc_flags = update_block_group_flags(root, cache->flags);
6654 if (alloc_flags != cache->flags)
6655 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6658 ret = set_block_group_ro(cache, 0);
6661 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6662 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6666 ret = set_block_group_ro(cache, 0);
6668 btrfs_end_transaction(trans, root);
6672 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6673 struct btrfs_root *root, u64 type)
6675 u64 alloc_flags = get_alloc_profile(root, type);
6676 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6681 * helper to account the unused space of all the readonly block group in the
6682 * list. takes mirrors into account.
6684 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6686 struct btrfs_block_group_cache *block_group;
6690 list_for_each_entry(block_group, groups_list, list) {
6691 spin_lock(&block_group->lock);
6693 if (!block_group->ro) {
6694 spin_unlock(&block_group->lock);
6698 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6699 BTRFS_BLOCK_GROUP_RAID10 |
6700 BTRFS_BLOCK_GROUP_DUP))
6705 free_bytes += (block_group->key.offset -
6706 btrfs_block_group_used(&block_group->item)) *
6709 spin_unlock(&block_group->lock);
6716 * helper to account the unused space of all the readonly block group in the
6717 * space_info. takes mirrors into account.
6719 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6724 spin_lock(&sinfo->lock);
6726 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6727 if (!list_empty(&sinfo->block_groups[i]))
6728 free_bytes += __btrfs_get_ro_block_group_free_space(
6729 &sinfo->block_groups[i]);
6731 spin_unlock(&sinfo->lock);
6736 int btrfs_set_block_group_rw(struct btrfs_root *root,
6737 struct btrfs_block_group_cache *cache)
6739 struct btrfs_space_info *sinfo = cache->space_info;
6744 spin_lock(&sinfo->lock);
6745 spin_lock(&cache->lock);
6746 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6747 cache->bytes_super - btrfs_block_group_used(&cache->item);
6748 sinfo->bytes_readonly -= num_bytes;
6750 spin_unlock(&cache->lock);
6751 spin_unlock(&sinfo->lock);
6756 * checks to see if its even possible to relocate this block group.
6758 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6759 * ok to go ahead and try.
6761 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6763 struct btrfs_block_group_cache *block_group;
6764 struct btrfs_space_info *space_info;
6765 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6766 struct btrfs_device *device;
6774 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6776 /* odd, couldn't find the block group, leave it alone */
6780 min_free = btrfs_block_group_used(&block_group->item);
6782 /* no bytes used, we're good */
6786 space_info = block_group->space_info;
6787 spin_lock(&space_info->lock);
6789 full = space_info->full;
6792 * if this is the last block group we have in this space, we can't
6793 * relocate it unless we're able to allocate a new chunk below.
6795 * Otherwise, we need to make sure we have room in the space to handle
6796 * all of the extents from this block group. If we can, we're good
6798 if ((space_info->total_bytes != block_group->key.offset) &&
6799 (space_info->bytes_used + space_info->bytes_reserved +
6800 space_info->bytes_pinned + space_info->bytes_readonly +
6801 min_free < space_info->total_bytes)) {
6802 spin_unlock(&space_info->lock);
6805 spin_unlock(&space_info->lock);
6808 * ok we don't have enough space, but maybe we have free space on our
6809 * devices to allocate new chunks for relocation, so loop through our
6810 * alloc devices and guess if we have enough space. However, if we
6811 * were marked as full, then we know there aren't enough chunks, and we
6826 index = get_block_group_index(block_group);
6831 } else if (index == 1) {
6833 } else if (index == 2) {
6836 } else if (index == 3) {
6837 dev_min = fs_devices->rw_devices;
6838 do_div(min_free, dev_min);
6841 mutex_lock(&root->fs_info->chunk_mutex);
6842 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6846 * check to make sure we can actually find a chunk with enough
6847 * space to fit our block group in.
6849 if (device->total_bytes > device->bytes_used + min_free) {
6850 ret = find_free_dev_extent(NULL, device, min_free,
6855 if (dev_nr >= dev_min)
6861 mutex_unlock(&root->fs_info->chunk_mutex);
6863 btrfs_put_block_group(block_group);
6867 static int find_first_block_group(struct btrfs_root *root,
6868 struct btrfs_path *path, struct btrfs_key *key)
6871 struct btrfs_key found_key;
6872 struct extent_buffer *leaf;
6875 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6880 slot = path->slots[0];
6881 leaf = path->nodes[0];
6882 if (slot >= btrfs_header_nritems(leaf)) {
6883 ret = btrfs_next_leaf(root, path);
6890 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6892 if (found_key.objectid >= key->objectid &&
6893 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6903 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6905 struct btrfs_block_group_cache *block_group;
6909 struct inode *inode;
6911 block_group = btrfs_lookup_first_block_group(info, last);
6912 while (block_group) {
6913 spin_lock(&block_group->lock);
6914 if (block_group->iref)
6916 spin_unlock(&block_group->lock);
6917 block_group = next_block_group(info->tree_root,
6927 inode = block_group->inode;
6928 block_group->iref = 0;
6929 block_group->inode = NULL;
6930 spin_unlock(&block_group->lock);
6932 last = block_group->key.objectid + block_group->key.offset;
6933 btrfs_put_block_group(block_group);
6937 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6939 struct btrfs_block_group_cache *block_group;
6940 struct btrfs_space_info *space_info;
6941 struct btrfs_caching_control *caching_ctl;
6944 down_write(&info->extent_commit_sem);
6945 while (!list_empty(&info->caching_block_groups)) {
6946 caching_ctl = list_entry(info->caching_block_groups.next,
6947 struct btrfs_caching_control, list);
6948 list_del(&caching_ctl->list);
6949 put_caching_control(caching_ctl);
6951 up_write(&info->extent_commit_sem);
6953 spin_lock(&info->block_group_cache_lock);
6954 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6955 block_group = rb_entry(n, struct btrfs_block_group_cache,
6957 rb_erase(&block_group->cache_node,
6958 &info->block_group_cache_tree);
6959 spin_unlock(&info->block_group_cache_lock);
6961 down_write(&block_group->space_info->groups_sem);
6962 list_del(&block_group->list);
6963 up_write(&block_group->space_info->groups_sem);
6965 if (block_group->cached == BTRFS_CACHE_STARTED)
6966 wait_block_group_cache_done(block_group);
6969 * We haven't cached this block group, which means we could
6970 * possibly have excluded extents on this block group.
6972 if (block_group->cached == BTRFS_CACHE_NO)
6973 free_excluded_extents(info->extent_root, block_group);
6975 btrfs_remove_free_space_cache(block_group);
6976 btrfs_put_block_group(block_group);
6978 spin_lock(&info->block_group_cache_lock);
6980 spin_unlock(&info->block_group_cache_lock);
6982 /* now that all the block groups are freed, go through and
6983 * free all the space_info structs. This is only called during
6984 * the final stages of unmount, and so we know nobody is
6985 * using them. We call synchronize_rcu() once before we start,
6986 * just to be on the safe side.
6990 release_global_block_rsv(info);
6992 while(!list_empty(&info->space_info)) {
6993 space_info = list_entry(info->space_info.next,
6994 struct btrfs_space_info,
6996 if (space_info->bytes_pinned > 0 ||
6997 space_info->bytes_reserved > 0 ||
6998 space_info->bytes_may_use > 0) {
7000 dump_space_info(space_info, 0, 0);
7002 list_del(&space_info->list);
7008 static void __link_block_group(struct btrfs_space_info *space_info,
7009 struct btrfs_block_group_cache *cache)
7011 int index = get_block_group_index(cache);
7013 down_write(&space_info->groups_sem);
7014 list_add_tail(&cache->list, &space_info->block_groups[index]);
7015 up_write(&space_info->groups_sem);
7018 int btrfs_read_block_groups(struct btrfs_root *root)
7020 struct btrfs_path *path;
7022 struct btrfs_block_group_cache *cache;
7023 struct btrfs_fs_info *info = root->fs_info;
7024 struct btrfs_space_info *space_info;
7025 struct btrfs_key key;
7026 struct btrfs_key found_key;
7027 struct extent_buffer *leaf;
7031 root = info->extent_root;
7034 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7035 path = btrfs_alloc_path();
7040 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7041 if (cache_gen != 0 &&
7042 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7044 if (btrfs_test_opt(root, CLEAR_CACHE))
7046 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
7047 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
7050 ret = find_first_block_group(root, path, &key);
7055 leaf = path->nodes[0];
7056 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7057 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7062 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7064 if (!cache->free_space_ctl) {
7070 atomic_set(&cache->count, 1);
7071 spin_lock_init(&cache->lock);
7072 cache->fs_info = info;
7073 INIT_LIST_HEAD(&cache->list);
7074 INIT_LIST_HEAD(&cache->cluster_list);
7077 cache->disk_cache_state = BTRFS_DC_CLEAR;
7079 read_extent_buffer(leaf, &cache->item,
7080 btrfs_item_ptr_offset(leaf, path->slots[0]),
7081 sizeof(cache->item));
7082 memcpy(&cache->key, &found_key, sizeof(found_key));
7084 key.objectid = found_key.objectid + found_key.offset;
7085 btrfs_release_path(path);
7086 cache->flags = btrfs_block_group_flags(&cache->item);
7087 cache->sectorsize = root->sectorsize;
7089 btrfs_init_free_space_ctl(cache);
7092 * We need to exclude the super stripes now so that the space
7093 * info has super bytes accounted for, otherwise we'll think
7094 * we have more space than we actually do.
7096 exclude_super_stripes(root, cache);
7099 * check for two cases, either we are full, and therefore
7100 * don't need to bother with the caching work since we won't
7101 * find any space, or we are empty, and we can just add all
7102 * the space in and be done with it. This saves us _alot_ of
7103 * time, particularly in the full case.
7105 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7106 cache->last_byte_to_unpin = (u64)-1;
7107 cache->cached = BTRFS_CACHE_FINISHED;
7108 free_excluded_extents(root, cache);
7109 } else if (btrfs_block_group_used(&cache->item) == 0) {
7110 cache->last_byte_to_unpin = (u64)-1;
7111 cache->cached = BTRFS_CACHE_FINISHED;
7112 add_new_free_space(cache, root->fs_info,
7114 found_key.objectid +
7116 free_excluded_extents(root, cache);
7119 ret = update_space_info(info, cache->flags, found_key.offset,
7120 btrfs_block_group_used(&cache->item),
7123 cache->space_info = space_info;
7124 spin_lock(&cache->space_info->lock);
7125 cache->space_info->bytes_readonly += cache->bytes_super;
7126 spin_unlock(&cache->space_info->lock);
7128 __link_block_group(space_info, cache);
7130 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7133 set_avail_alloc_bits(root->fs_info, cache->flags);
7134 if (btrfs_chunk_readonly(root, cache->key.objectid))
7135 set_block_group_ro(cache, 1);
7138 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7139 if (!(get_alloc_profile(root, space_info->flags) &
7140 (BTRFS_BLOCK_GROUP_RAID10 |
7141 BTRFS_BLOCK_GROUP_RAID1 |
7142 BTRFS_BLOCK_GROUP_DUP)))
7145 * avoid allocating from un-mirrored block group if there are
7146 * mirrored block groups.
7148 list_for_each_entry(cache, &space_info->block_groups[3], list)
7149 set_block_group_ro(cache, 1);
7150 list_for_each_entry(cache, &space_info->block_groups[4], list)
7151 set_block_group_ro(cache, 1);
7154 init_global_block_rsv(info);
7157 btrfs_free_path(path);
7161 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7162 struct btrfs_root *root, u64 bytes_used,
7163 u64 type, u64 chunk_objectid, u64 chunk_offset,
7167 struct btrfs_root *extent_root;
7168 struct btrfs_block_group_cache *cache;
7170 extent_root = root->fs_info->extent_root;
7172 root->fs_info->last_trans_log_full_commit = trans->transid;
7174 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7177 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7179 if (!cache->free_space_ctl) {
7184 cache->key.objectid = chunk_offset;
7185 cache->key.offset = size;
7186 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7187 cache->sectorsize = root->sectorsize;
7188 cache->fs_info = root->fs_info;
7190 atomic_set(&cache->count, 1);
7191 spin_lock_init(&cache->lock);
7192 INIT_LIST_HEAD(&cache->list);
7193 INIT_LIST_HEAD(&cache->cluster_list);
7195 btrfs_init_free_space_ctl(cache);
7197 btrfs_set_block_group_used(&cache->item, bytes_used);
7198 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7199 cache->flags = type;
7200 btrfs_set_block_group_flags(&cache->item, type);
7202 cache->last_byte_to_unpin = (u64)-1;
7203 cache->cached = BTRFS_CACHE_FINISHED;
7204 exclude_super_stripes(root, cache);
7206 add_new_free_space(cache, root->fs_info, chunk_offset,
7207 chunk_offset + size);
7209 free_excluded_extents(root, cache);
7211 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7212 &cache->space_info);
7215 spin_lock(&cache->space_info->lock);
7216 cache->space_info->bytes_readonly += cache->bytes_super;
7217 spin_unlock(&cache->space_info->lock);
7219 __link_block_group(cache->space_info, cache);
7221 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7224 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7225 sizeof(cache->item));
7228 set_avail_alloc_bits(extent_root->fs_info, type);
7233 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7234 struct btrfs_root *root, u64 group_start)
7236 struct btrfs_path *path;
7237 struct btrfs_block_group_cache *block_group;
7238 struct btrfs_free_cluster *cluster;
7239 struct btrfs_root *tree_root = root->fs_info->tree_root;
7240 struct btrfs_key key;
7241 struct inode *inode;
7245 root = root->fs_info->extent_root;
7247 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7248 BUG_ON(!block_group);
7249 BUG_ON(!block_group->ro);
7252 * Free the reserved super bytes from this block group before
7255 free_excluded_extents(root, block_group);
7257 memcpy(&key, &block_group->key, sizeof(key));
7258 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7259 BTRFS_BLOCK_GROUP_RAID1 |
7260 BTRFS_BLOCK_GROUP_RAID10))
7265 /* make sure this block group isn't part of an allocation cluster */
7266 cluster = &root->fs_info->data_alloc_cluster;
7267 spin_lock(&cluster->refill_lock);
7268 btrfs_return_cluster_to_free_space(block_group, cluster);
7269 spin_unlock(&cluster->refill_lock);
7272 * make sure this block group isn't part of a metadata
7273 * allocation cluster
7275 cluster = &root->fs_info->meta_alloc_cluster;
7276 spin_lock(&cluster->refill_lock);
7277 btrfs_return_cluster_to_free_space(block_group, cluster);
7278 spin_unlock(&cluster->refill_lock);
7280 path = btrfs_alloc_path();
7286 inode = lookup_free_space_inode(root, block_group, path);
7287 if (!IS_ERR(inode)) {
7288 ret = btrfs_orphan_add(trans, inode);
7291 /* One for the block groups ref */
7292 spin_lock(&block_group->lock);
7293 if (block_group->iref) {
7294 block_group->iref = 0;
7295 block_group->inode = NULL;
7296 spin_unlock(&block_group->lock);
7299 spin_unlock(&block_group->lock);
7301 /* One for our lookup ref */
7302 btrfs_add_delayed_iput(inode);
7305 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7306 key.offset = block_group->key.objectid;
7309 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7313 btrfs_release_path(path);
7315 ret = btrfs_del_item(trans, tree_root, path);
7318 btrfs_release_path(path);
7321 spin_lock(&root->fs_info->block_group_cache_lock);
7322 rb_erase(&block_group->cache_node,
7323 &root->fs_info->block_group_cache_tree);
7324 spin_unlock(&root->fs_info->block_group_cache_lock);
7326 down_write(&block_group->space_info->groups_sem);
7328 * we must use list_del_init so people can check to see if they
7329 * are still on the list after taking the semaphore
7331 list_del_init(&block_group->list);
7332 up_write(&block_group->space_info->groups_sem);
7334 if (block_group->cached == BTRFS_CACHE_STARTED)
7335 wait_block_group_cache_done(block_group);
7337 btrfs_remove_free_space_cache(block_group);
7339 spin_lock(&block_group->space_info->lock);
7340 block_group->space_info->total_bytes -= block_group->key.offset;
7341 block_group->space_info->bytes_readonly -= block_group->key.offset;
7342 block_group->space_info->disk_total -= block_group->key.offset * factor;
7343 spin_unlock(&block_group->space_info->lock);
7345 memcpy(&key, &block_group->key, sizeof(key));
7347 btrfs_clear_space_info_full(root->fs_info);
7349 btrfs_put_block_group(block_group);
7350 btrfs_put_block_group(block_group);
7352 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7358 ret = btrfs_del_item(trans, root, path);
7360 btrfs_free_path(path);
7364 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7366 struct btrfs_space_info *space_info;
7367 struct btrfs_super_block *disk_super;
7373 disk_super = &fs_info->super_copy;
7374 if (!btrfs_super_root(disk_super))
7377 features = btrfs_super_incompat_flags(disk_super);
7378 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7381 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7382 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7387 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7388 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7390 flags = BTRFS_BLOCK_GROUP_METADATA;
7391 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7395 flags = BTRFS_BLOCK_GROUP_DATA;
7396 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7402 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7404 return unpin_extent_range(root, start, end);
7407 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7408 u64 num_bytes, u64 *actual_bytes)
7410 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7413 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7415 struct btrfs_fs_info *fs_info = root->fs_info;
7416 struct btrfs_block_group_cache *cache = NULL;
7423 cache = btrfs_lookup_block_group(fs_info, range->start);
7426 if (cache->key.objectid >= (range->start + range->len)) {
7427 btrfs_put_block_group(cache);
7431 start = max(range->start, cache->key.objectid);
7432 end = min(range->start + range->len,
7433 cache->key.objectid + cache->key.offset);
7435 if (end - start >= range->minlen) {
7436 if (!block_group_cache_done(cache)) {
7437 ret = cache_block_group(cache, NULL, root, 0);
7439 wait_block_group_cache_done(cache);
7441 ret = btrfs_trim_block_group(cache,
7447 trimmed += group_trimmed;
7449 btrfs_put_block_group(cache);
7454 cache = next_block_group(fs_info->tree_root, cache);
7457 range->len = trimmed;