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 * Retries tells us how many times we've called reserve_metadata_bytes. The
3409 * idea is if this is the first call (retries == 0) then we will add to our
3410 * reserved count if we can't make the allocation in order to hold our place
3411 * while we go and try and free up space. That way for retries > 1 we don't try
3412 * and add space, we just check to see if the amount of unused space is >= the
3413 * total space, meaning that our reservation is valid.
3415 * However if we don't intend to retry this reservation, pass -1 as retries so
3416 * that it short circuits this logic.
3418 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3419 struct btrfs_root *root,
3420 struct btrfs_block_rsv *block_rsv,
3421 u64 orig_bytes, int flush)
3423 struct btrfs_space_info *space_info = block_rsv->space_info;
3425 u64 num_bytes = orig_bytes;
3428 bool committed = false;
3429 bool flushing = false;
3432 spin_lock(&space_info->lock);
3434 * We only want to wait if somebody other than us is flushing and we are
3435 * actually alloed to flush.
3437 while (flush && !flushing && space_info->flush) {
3438 spin_unlock(&space_info->lock);
3440 * If we have a trans handle we can't wait because the flusher
3441 * may have to commit the transaction, which would mean we would
3442 * deadlock since we are waiting for the flusher to finish, but
3443 * hold the current transaction open.
3447 ret = wait_event_interruptible(space_info->wait,
3448 !space_info->flush);
3449 /* Must have been interrupted, return */
3453 spin_lock(&space_info->lock);
3457 unused = space_info->bytes_used + space_info->bytes_reserved +
3458 space_info->bytes_pinned + space_info->bytes_readonly +
3459 space_info->bytes_may_use;
3462 * The idea here is that we've not already over-reserved the block group
3463 * then we can go ahead and save our reservation first and then start
3464 * flushing if we need to. Otherwise if we've already overcommitted
3465 * lets start flushing stuff first and then come back and try to make
3468 if (unused <= space_info->total_bytes) {
3469 unused = space_info->total_bytes - unused;
3470 if (unused >= num_bytes) {
3471 space_info->bytes_may_use += orig_bytes;
3475 * Ok set num_bytes to orig_bytes since we aren't
3476 * overocmmitted, this way we only try and reclaim what
3479 num_bytes = orig_bytes;
3483 * Ok we're over committed, set num_bytes to the overcommitted
3484 * amount plus the amount of bytes that we need for this
3487 num_bytes = unused - space_info->total_bytes +
3488 (orig_bytes * (retries + 1));
3492 * Couldn't make our reservation, save our place so while we're trying
3493 * to reclaim space we can actually use it instead of somebody else
3494 * stealing it from us.
3498 space_info->flush = 1;
3501 spin_unlock(&space_info->lock);
3507 * We do synchronous shrinking since we don't actually unreserve
3508 * metadata until after the IO is completed.
3510 ret = shrink_delalloc(trans, root, num_bytes, 1);
3517 * So if we were overcommitted it's possible that somebody else flushed
3518 * out enough space and we simply didn't have enough space to reclaim,
3519 * so go back around and try again.
3527 * Not enough space to be reclaimed, don't bother committing the
3530 spin_lock(&space_info->lock);
3531 if (space_info->bytes_pinned < orig_bytes)
3533 spin_unlock(&space_info->lock);
3545 trans = btrfs_join_transaction(root);
3548 ret = btrfs_commit_transaction(trans, root);
3557 spin_lock(&space_info->lock);
3558 space_info->flush = 0;
3559 wake_up_all(&space_info->wait);
3560 spin_unlock(&space_info->lock);
3565 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3566 struct btrfs_root *root)
3568 struct btrfs_block_rsv *block_rsv;
3570 block_rsv = trans->block_rsv;
3572 block_rsv = root->block_rsv;
3575 block_rsv = &root->fs_info->empty_block_rsv;
3580 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3584 spin_lock(&block_rsv->lock);
3585 if (block_rsv->reserved >= num_bytes) {
3586 block_rsv->reserved -= num_bytes;
3587 if (block_rsv->reserved < block_rsv->size)
3588 block_rsv->full = 0;
3591 spin_unlock(&block_rsv->lock);
3595 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3596 u64 num_bytes, int update_size)
3598 spin_lock(&block_rsv->lock);
3599 block_rsv->reserved += num_bytes;
3601 block_rsv->size += num_bytes;
3602 else if (block_rsv->reserved >= block_rsv->size)
3603 block_rsv->full = 1;
3604 spin_unlock(&block_rsv->lock);
3607 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3608 struct btrfs_block_rsv *dest, u64 num_bytes)
3610 struct btrfs_space_info *space_info = block_rsv->space_info;
3612 spin_lock(&block_rsv->lock);
3613 if (num_bytes == (u64)-1)
3614 num_bytes = block_rsv->size;
3615 block_rsv->size -= num_bytes;
3616 if (block_rsv->reserved >= block_rsv->size) {
3617 num_bytes = block_rsv->reserved - block_rsv->size;
3618 block_rsv->reserved = block_rsv->size;
3619 block_rsv->full = 1;
3623 spin_unlock(&block_rsv->lock);
3625 if (num_bytes > 0) {
3627 spin_lock(&dest->lock);
3631 bytes_to_add = dest->size - dest->reserved;
3632 bytes_to_add = min(num_bytes, bytes_to_add);
3633 dest->reserved += bytes_to_add;
3634 if (dest->reserved >= dest->size)
3636 num_bytes -= bytes_to_add;
3638 spin_unlock(&dest->lock);
3641 spin_lock(&space_info->lock);
3642 space_info->bytes_may_use -= num_bytes;
3643 space_info->reservation_progress++;
3644 spin_unlock(&space_info->lock);
3649 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3650 struct btrfs_block_rsv *dst, u64 num_bytes)
3654 ret = block_rsv_use_bytes(src, num_bytes);
3658 block_rsv_add_bytes(dst, num_bytes, 1);
3662 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3664 memset(rsv, 0, sizeof(*rsv));
3665 spin_lock_init(&rsv->lock);
3668 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3670 struct btrfs_block_rsv *block_rsv;
3671 struct btrfs_fs_info *fs_info = root->fs_info;
3673 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3677 btrfs_init_block_rsv(block_rsv);
3678 block_rsv->space_info = __find_space_info(fs_info,
3679 BTRFS_BLOCK_GROUP_METADATA);
3683 void btrfs_free_block_rsv(struct btrfs_root *root,
3684 struct btrfs_block_rsv *rsv)
3686 btrfs_block_rsv_release(root, rsv, (u64)-1);
3690 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3691 struct btrfs_root *root,
3692 struct btrfs_block_rsv *block_rsv,
3700 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3702 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3709 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3710 struct btrfs_root *root,
3711 struct btrfs_block_rsv *block_rsv,
3712 u64 min_reserved, int min_factor, int flush)
3720 spin_lock(&block_rsv->lock);
3722 num_bytes = div_factor(block_rsv->size, min_factor);
3723 if (min_reserved > num_bytes)
3724 num_bytes = min_reserved;
3726 if (block_rsv->reserved >= num_bytes)
3729 num_bytes -= block_rsv->reserved;
3730 spin_unlock(&block_rsv->lock);
3735 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, flush);
3737 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3744 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3745 struct btrfs_block_rsv *dst_rsv,
3748 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3751 void btrfs_block_rsv_release(struct btrfs_root *root,
3752 struct btrfs_block_rsv *block_rsv,
3755 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3756 if (global_rsv->full || global_rsv == block_rsv ||
3757 block_rsv->space_info != global_rsv->space_info)
3759 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3763 * helper to calculate size of global block reservation.
3764 * the desired value is sum of space used by extent tree,
3765 * checksum tree and root tree
3767 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3769 struct btrfs_space_info *sinfo;
3773 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3775 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3776 spin_lock(&sinfo->lock);
3777 data_used = sinfo->bytes_used;
3778 spin_unlock(&sinfo->lock);
3780 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3781 spin_lock(&sinfo->lock);
3782 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3784 meta_used = sinfo->bytes_used;
3785 spin_unlock(&sinfo->lock);
3787 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3789 num_bytes += div64_u64(data_used + meta_used, 50);
3791 if (num_bytes * 3 > meta_used)
3792 num_bytes = div64_u64(meta_used, 3);
3794 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3797 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3799 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3800 struct btrfs_space_info *sinfo = block_rsv->space_info;
3803 num_bytes = calc_global_metadata_size(fs_info);
3805 spin_lock(&block_rsv->lock);
3806 spin_lock(&sinfo->lock);
3808 block_rsv->size = num_bytes;
3810 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3811 sinfo->bytes_reserved + sinfo->bytes_readonly +
3812 sinfo->bytes_may_use;
3814 if (sinfo->total_bytes > num_bytes) {
3815 num_bytes = sinfo->total_bytes - num_bytes;
3816 block_rsv->reserved += num_bytes;
3817 sinfo->bytes_may_use += num_bytes;
3820 if (block_rsv->reserved >= block_rsv->size) {
3821 num_bytes = block_rsv->reserved - block_rsv->size;
3822 sinfo->bytes_may_use -= num_bytes;
3823 sinfo->reservation_progress++;
3824 block_rsv->reserved = block_rsv->size;
3825 block_rsv->full = 1;
3828 spin_unlock(&sinfo->lock);
3829 spin_unlock(&block_rsv->lock);
3832 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3834 struct btrfs_space_info *space_info;
3836 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3837 fs_info->chunk_block_rsv.space_info = space_info;
3839 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3840 fs_info->global_block_rsv.space_info = space_info;
3841 fs_info->delalloc_block_rsv.space_info = space_info;
3842 fs_info->trans_block_rsv.space_info = space_info;
3843 fs_info->empty_block_rsv.space_info = space_info;
3845 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3846 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3847 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3848 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3849 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3851 update_global_block_rsv(fs_info);
3854 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3856 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3857 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3858 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3859 WARN_ON(fs_info->trans_block_rsv.size > 0);
3860 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3861 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3862 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3865 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3866 struct btrfs_root *root)
3868 if (!trans->bytes_reserved)
3871 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3872 btrfs_block_rsv_release(root, trans->block_rsv,
3873 trans->bytes_reserved);
3874 trans->bytes_reserved = 0;
3877 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3878 struct inode *inode)
3880 struct btrfs_root *root = BTRFS_I(inode)->root;
3881 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3882 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3885 * We need to hold space in order to delete our orphan item once we've
3886 * added it, so this takes the reservation so we can release it later
3887 * when we are truly done with the orphan item.
3889 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3890 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3893 void btrfs_orphan_release_metadata(struct inode *inode)
3895 struct btrfs_root *root = BTRFS_I(inode)->root;
3896 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3897 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3900 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3901 struct btrfs_pending_snapshot *pending)
3903 struct btrfs_root *root = pending->root;
3904 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3905 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3907 * two for root back/forward refs, two for directory entries
3908 * and one for root of the snapshot.
3910 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3911 dst_rsv->space_info = src_rsv->space_info;
3912 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3916 * drop_outstanding_extent - drop an outstanding extent
3917 * @inode: the inode we're dropping the extent for
3919 * This is called when we are freeing up an outstanding extent, either called
3920 * after an error or after an extent is written. This will return the number of
3921 * reserved extents that need to be freed. This must be called with
3922 * BTRFS_I(inode)->lock held.
3924 static unsigned drop_outstanding_extent(struct inode *inode)
3926 unsigned dropped_extents = 0;
3928 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3929 BTRFS_I(inode)->outstanding_extents--;
3932 * If we have more or the same amount of outsanding extents than we have
3933 * reserved then we need to leave the reserved extents count alone.
3935 if (BTRFS_I(inode)->outstanding_extents >=
3936 BTRFS_I(inode)->reserved_extents)
3939 dropped_extents = BTRFS_I(inode)->reserved_extents -
3940 BTRFS_I(inode)->outstanding_extents;
3941 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3942 return dropped_extents;
3946 * calc_csum_metadata_size - return the amount of metada space that must be
3947 * reserved/free'd for the given bytes.
3948 * @inode: the inode we're manipulating
3949 * @num_bytes: the number of bytes in question
3950 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3952 * This adjusts the number of csum_bytes in the inode and then returns the
3953 * correct amount of metadata that must either be reserved or freed. We
3954 * calculate how many checksums we can fit into one leaf and then divide the
3955 * number of bytes that will need to be checksumed by this value to figure out
3956 * how many checksums will be required. If we are adding bytes then the number
3957 * may go up and we will return the number of additional bytes that must be
3958 * reserved. If it is going down we will return the number of bytes that must
3961 * This must be called with BTRFS_I(inode)->lock held.
3963 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
3966 struct btrfs_root *root = BTRFS_I(inode)->root;
3968 int num_csums_per_leaf;
3972 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
3973 BTRFS_I(inode)->csum_bytes == 0)
3976 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
3978 BTRFS_I(inode)->csum_bytes += num_bytes;
3980 BTRFS_I(inode)->csum_bytes -= num_bytes;
3981 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
3982 num_csums_per_leaf = (int)div64_u64(csum_size,
3983 sizeof(struct btrfs_csum_item) +
3984 sizeof(struct btrfs_disk_key));
3985 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
3986 num_csums = num_csums + num_csums_per_leaf - 1;
3987 num_csums = num_csums / num_csums_per_leaf;
3989 old_csums = old_csums + num_csums_per_leaf - 1;
3990 old_csums = old_csums / num_csums_per_leaf;
3992 /* No change, no need to reserve more */
3993 if (old_csums == num_csums)
3997 return btrfs_calc_trans_metadata_size(root,
3998 num_csums - old_csums);
4000 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4003 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4005 struct btrfs_root *root = BTRFS_I(inode)->root;
4006 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4008 unsigned nr_extents = 0;
4012 if (btrfs_is_free_space_inode(root, inode))
4015 if (flush && btrfs_transaction_in_commit(root->fs_info))
4016 schedule_timeout(1);
4018 num_bytes = ALIGN(num_bytes, root->sectorsize);
4020 spin_lock(&BTRFS_I(inode)->lock);
4021 BTRFS_I(inode)->outstanding_extents++;
4023 if (BTRFS_I(inode)->outstanding_extents >
4024 BTRFS_I(inode)->reserved_extents) {
4025 nr_extents = BTRFS_I(inode)->outstanding_extents -
4026 BTRFS_I(inode)->reserved_extents;
4027 BTRFS_I(inode)->reserved_extents += nr_extents;
4029 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4031 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4032 spin_unlock(&BTRFS_I(inode)->lock);
4034 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, flush);
4039 spin_lock(&BTRFS_I(inode)->lock);
4040 dropped = drop_outstanding_extent(inode);
4041 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4042 spin_unlock(&BTRFS_I(inode)->lock);
4043 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4046 * Somebody could have come in and twiddled with the
4047 * reservation, so if we have to free more than we would have
4048 * reserved from this reservation go ahead and release those
4051 to_free -= to_reserve;
4053 btrfs_block_rsv_release(root, block_rsv, to_free);
4057 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4063 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4064 * @inode: the inode to release the reservation for
4065 * @num_bytes: the number of bytes we're releasing
4067 * This will release the metadata reservation for an inode. This can be called
4068 * once we complete IO for a given set of bytes to release their metadata
4071 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4073 struct btrfs_root *root = BTRFS_I(inode)->root;
4077 num_bytes = ALIGN(num_bytes, root->sectorsize);
4078 spin_lock(&BTRFS_I(inode)->lock);
4079 dropped = drop_outstanding_extent(inode);
4081 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4082 spin_unlock(&BTRFS_I(inode)->lock);
4084 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4086 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4091 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4092 * @inode: inode we're writing to
4093 * @num_bytes: the number of bytes we want to allocate
4095 * This will do the following things
4097 * o reserve space in the data space info for num_bytes
4098 * o reserve space in the metadata space info based on number of outstanding
4099 * extents and how much csums will be needed
4100 * o add to the inodes ->delalloc_bytes
4101 * o add it to the fs_info's delalloc inodes list.
4103 * This will return 0 for success and -ENOSPC if there is no space left.
4105 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4109 ret = btrfs_check_data_free_space(inode, num_bytes);
4113 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4115 btrfs_free_reserved_data_space(inode, num_bytes);
4123 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4124 * @inode: inode we're releasing space for
4125 * @num_bytes: the number of bytes we want to free up
4127 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4128 * called in the case that we don't need the metadata AND data reservations
4129 * anymore. So if there is an error or we insert an inline extent.
4131 * This function will release the metadata space that was not used and will
4132 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4133 * list if there are no delalloc bytes left.
4135 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4137 btrfs_delalloc_release_metadata(inode, num_bytes);
4138 btrfs_free_reserved_data_space(inode, num_bytes);
4141 static int update_block_group(struct btrfs_trans_handle *trans,
4142 struct btrfs_root *root,
4143 u64 bytenr, u64 num_bytes, int alloc)
4145 struct btrfs_block_group_cache *cache = NULL;
4146 struct btrfs_fs_info *info = root->fs_info;
4147 u64 total = num_bytes;
4152 /* block accounting for super block */
4153 spin_lock(&info->delalloc_lock);
4154 old_val = btrfs_super_bytes_used(&info->super_copy);
4156 old_val += num_bytes;
4158 old_val -= num_bytes;
4159 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4160 spin_unlock(&info->delalloc_lock);
4163 cache = btrfs_lookup_block_group(info, bytenr);
4166 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4167 BTRFS_BLOCK_GROUP_RAID1 |
4168 BTRFS_BLOCK_GROUP_RAID10))
4173 * If this block group has free space cache written out, we
4174 * need to make sure to load it if we are removing space. This
4175 * is because we need the unpinning stage to actually add the
4176 * space back to the block group, otherwise we will leak space.
4178 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4179 cache_block_group(cache, trans, NULL, 1);
4181 byte_in_group = bytenr - cache->key.objectid;
4182 WARN_ON(byte_in_group > cache->key.offset);
4184 spin_lock(&cache->space_info->lock);
4185 spin_lock(&cache->lock);
4187 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4188 cache->disk_cache_state < BTRFS_DC_CLEAR)
4189 cache->disk_cache_state = BTRFS_DC_CLEAR;
4192 old_val = btrfs_block_group_used(&cache->item);
4193 num_bytes = min(total, cache->key.offset - byte_in_group);
4195 old_val += num_bytes;
4196 btrfs_set_block_group_used(&cache->item, old_val);
4197 cache->reserved -= num_bytes;
4198 cache->space_info->bytes_reserved -= num_bytes;
4199 cache->space_info->bytes_used += num_bytes;
4200 cache->space_info->disk_used += num_bytes * factor;
4201 spin_unlock(&cache->lock);
4202 spin_unlock(&cache->space_info->lock);
4204 old_val -= num_bytes;
4205 btrfs_set_block_group_used(&cache->item, old_val);
4206 cache->pinned += num_bytes;
4207 cache->space_info->bytes_pinned += num_bytes;
4208 cache->space_info->bytes_used -= num_bytes;
4209 cache->space_info->disk_used -= num_bytes * factor;
4210 spin_unlock(&cache->lock);
4211 spin_unlock(&cache->space_info->lock);
4213 set_extent_dirty(info->pinned_extents,
4214 bytenr, bytenr + num_bytes - 1,
4215 GFP_NOFS | __GFP_NOFAIL);
4217 btrfs_put_block_group(cache);
4219 bytenr += num_bytes;
4224 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4226 struct btrfs_block_group_cache *cache;
4229 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4233 bytenr = cache->key.objectid;
4234 btrfs_put_block_group(cache);
4239 static int pin_down_extent(struct btrfs_root *root,
4240 struct btrfs_block_group_cache *cache,
4241 u64 bytenr, u64 num_bytes, int reserved)
4243 spin_lock(&cache->space_info->lock);
4244 spin_lock(&cache->lock);
4245 cache->pinned += num_bytes;
4246 cache->space_info->bytes_pinned += num_bytes;
4248 cache->reserved -= num_bytes;
4249 cache->space_info->bytes_reserved -= num_bytes;
4251 spin_unlock(&cache->lock);
4252 spin_unlock(&cache->space_info->lock);
4254 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4255 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4260 * this function must be called within transaction
4262 int btrfs_pin_extent(struct btrfs_root *root,
4263 u64 bytenr, u64 num_bytes, int reserved)
4265 struct btrfs_block_group_cache *cache;
4267 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4270 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4272 btrfs_put_block_group(cache);
4277 * btrfs_update_reserved_bytes - update the block_group and space info counters
4278 * @cache: The cache we are manipulating
4279 * @num_bytes: The number of bytes in question
4280 * @reserve: One of the reservation enums
4282 * This is called by the allocator when it reserves space, or by somebody who is
4283 * freeing space that was never actually used on disk. For example if you
4284 * reserve some space for a new leaf in transaction A and before transaction A
4285 * commits you free that leaf, you call this with reserve set to 0 in order to
4286 * clear the reservation.
4288 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4289 * ENOSPC accounting. For data we handle the reservation through clearing the
4290 * delalloc bits in the io_tree. We have to do this since we could end up
4291 * allocating less disk space for the amount of data we have reserved in the
4292 * case of compression.
4294 * If this is a reservation and the block group has become read only we cannot
4295 * make the reservation and return -EAGAIN, otherwise this function always
4298 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4299 u64 num_bytes, int reserve)
4301 struct btrfs_space_info *space_info = cache->space_info;
4303 spin_lock(&space_info->lock);
4304 spin_lock(&cache->lock);
4305 if (reserve != RESERVE_FREE) {
4309 cache->reserved += num_bytes;
4310 space_info->bytes_reserved += num_bytes;
4311 if (reserve == RESERVE_ALLOC) {
4312 BUG_ON(space_info->bytes_may_use < num_bytes);
4313 space_info->bytes_may_use -= num_bytes;
4318 space_info->bytes_readonly += num_bytes;
4319 cache->reserved -= num_bytes;
4320 space_info->bytes_reserved -= num_bytes;
4321 space_info->reservation_progress++;
4323 spin_unlock(&cache->lock);
4324 spin_unlock(&space_info->lock);
4328 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4329 struct btrfs_root *root)
4331 struct btrfs_fs_info *fs_info = root->fs_info;
4332 struct btrfs_caching_control *next;
4333 struct btrfs_caching_control *caching_ctl;
4334 struct btrfs_block_group_cache *cache;
4336 down_write(&fs_info->extent_commit_sem);
4338 list_for_each_entry_safe(caching_ctl, next,
4339 &fs_info->caching_block_groups, list) {
4340 cache = caching_ctl->block_group;
4341 if (block_group_cache_done(cache)) {
4342 cache->last_byte_to_unpin = (u64)-1;
4343 list_del_init(&caching_ctl->list);
4344 put_caching_control(caching_ctl);
4346 cache->last_byte_to_unpin = caching_ctl->progress;
4350 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4351 fs_info->pinned_extents = &fs_info->freed_extents[1];
4353 fs_info->pinned_extents = &fs_info->freed_extents[0];
4355 up_write(&fs_info->extent_commit_sem);
4357 update_global_block_rsv(fs_info);
4361 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4363 struct btrfs_fs_info *fs_info = root->fs_info;
4364 struct btrfs_block_group_cache *cache = NULL;
4367 while (start <= end) {
4369 start >= cache->key.objectid + cache->key.offset) {
4371 btrfs_put_block_group(cache);
4372 cache = btrfs_lookup_block_group(fs_info, start);
4376 len = cache->key.objectid + cache->key.offset - start;
4377 len = min(len, end + 1 - start);
4379 if (start < cache->last_byte_to_unpin) {
4380 len = min(len, cache->last_byte_to_unpin - start);
4381 btrfs_add_free_space(cache, start, len);
4386 spin_lock(&cache->space_info->lock);
4387 spin_lock(&cache->lock);
4388 cache->pinned -= len;
4389 cache->space_info->bytes_pinned -= len;
4391 cache->space_info->bytes_readonly += len;
4392 spin_unlock(&cache->lock);
4393 spin_unlock(&cache->space_info->lock);
4397 btrfs_put_block_group(cache);
4401 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4402 struct btrfs_root *root)
4404 struct btrfs_fs_info *fs_info = root->fs_info;
4405 struct extent_io_tree *unpin;
4410 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4411 unpin = &fs_info->freed_extents[1];
4413 unpin = &fs_info->freed_extents[0];
4416 ret = find_first_extent_bit(unpin, 0, &start, &end,
4421 if (btrfs_test_opt(root, DISCARD))
4422 ret = btrfs_discard_extent(root, start,
4423 end + 1 - start, NULL);
4425 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4426 unpin_extent_range(root, start, end);
4433 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4434 struct btrfs_root *root,
4435 u64 bytenr, u64 num_bytes, u64 parent,
4436 u64 root_objectid, u64 owner_objectid,
4437 u64 owner_offset, int refs_to_drop,
4438 struct btrfs_delayed_extent_op *extent_op)
4440 struct btrfs_key key;
4441 struct btrfs_path *path;
4442 struct btrfs_fs_info *info = root->fs_info;
4443 struct btrfs_root *extent_root = info->extent_root;
4444 struct extent_buffer *leaf;
4445 struct btrfs_extent_item *ei;
4446 struct btrfs_extent_inline_ref *iref;
4449 int extent_slot = 0;
4450 int found_extent = 0;
4455 path = btrfs_alloc_path();
4460 path->leave_spinning = 1;
4462 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4463 BUG_ON(!is_data && refs_to_drop != 1);
4465 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4466 bytenr, num_bytes, parent,
4467 root_objectid, owner_objectid,
4470 extent_slot = path->slots[0];
4471 while (extent_slot >= 0) {
4472 btrfs_item_key_to_cpu(path->nodes[0], &key,
4474 if (key.objectid != bytenr)
4476 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4477 key.offset == num_bytes) {
4481 if (path->slots[0] - extent_slot > 5)
4485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4486 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4487 if (found_extent && item_size < sizeof(*ei))
4490 if (!found_extent) {
4492 ret = remove_extent_backref(trans, extent_root, path,
4496 btrfs_release_path(path);
4497 path->leave_spinning = 1;
4499 key.objectid = bytenr;
4500 key.type = BTRFS_EXTENT_ITEM_KEY;
4501 key.offset = num_bytes;
4503 ret = btrfs_search_slot(trans, extent_root,
4506 printk(KERN_ERR "umm, got %d back from search"
4507 ", was looking for %llu\n", ret,
4508 (unsigned long long)bytenr);
4510 btrfs_print_leaf(extent_root,
4514 extent_slot = path->slots[0];
4517 btrfs_print_leaf(extent_root, path->nodes[0]);
4519 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4520 "parent %llu root %llu owner %llu offset %llu\n",
4521 (unsigned long long)bytenr,
4522 (unsigned long long)parent,
4523 (unsigned long long)root_objectid,
4524 (unsigned long long)owner_objectid,
4525 (unsigned long long)owner_offset);
4528 leaf = path->nodes[0];
4529 item_size = btrfs_item_size_nr(leaf, extent_slot);
4530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4531 if (item_size < sizeof(*ei)) {
4532 BUG_ON(found_extent || extent_slot != path->slots[0]);
4533 ret = convert_extent_item_v0(trans, extent_root, path,
4537 btrfs_release_path(path);
4538 path->leave_spinning = 1;
4540 key.objectid = bytenr;
4541 key.type = BTRFS_EXTENT_ITEM_KEY;
4542 key.offset = num_bytes;
4544 ret = btrfs_search_slot(trans, extent_root, &key, path,
4547 printk(KERN_ERR "umm, got %d back from search"
4548 ", was looking for %llu\n", ret,
4549 (unsigned long long)bytenr);
4550 btrfs_print_leaf(extent_root, path->nodes[0]);
4553 extent_slot = path->slots[0];
4554 leaf = path->nodes[0];
4555 item_size = btrfs_item_size_nr(leaf, extent_slot);
4558 BUG_ON(item_size < sizeof(*ei));
4559 ei = btrfs_item_ptr(leaf, extent_slot,
4560 struct btrfs_extent_item);
4561 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4562 struct btrfs_tree_block_info *bi;
4563 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4564 bi = (struct btrfs_tree_block_info *)(ei + 1);
4565 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4568 refs = btrfs_extent_refs(leaf, ei);
4569 BUG_ON(refs < refs_to_drop);
4570 refs -= refs_to_drop;
4574 __run_delayed_extent_op(extent_op, leaf, ei);
4576 * In the case of inline back ref, reference count will
4577 * be updated by remove_extent_backref
4580 BUG_ON(!found_extent);
4582 btrfs_set_extent_refs(leaf, ei, refs);
4583 btrfs_mark_buffer_dirty(leaf);
4586 ret = remove_extent_backref(trans, extent_root, path,
4593 BUG_ON(is_data && refs_to_drop !=
4594 extent_data_ref_count(root, path, iref));
4596 BUG_ON(path->slots[0] != extent_slot);
4598 BUG_ON(path->slots[0] != extent_slot + 1);
4599 path->slots[0] = extent_slot;
4604 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4607 btrfs_release_path(path);
4610 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4613 invalidate_mapping_pages(info->btree_inode->i_mapping,
4614 bytenr >> PAGE_CACHE_SHIFT,
4615 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4618 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4621 btrfs_free_path(path);
4626 * when we free an block, it is possible (and likely) that we free the last
4627 * delayed ref for that extent as well. This searches the delayed ref tree for
4628 * a given extent, and if there are no other delayed refs to be processed, it
4629 * removes it from the tree.
4631 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4632 struct btrfs_root *root, u64 bytenr)
4634 struct btrfs_delayed_ref_head *head;
4635 struct btrfs_delayed_ref_root *delayed_refs;
4636 struct btrfs_delayed_ref_node *ref;
4637 struct rb_node *node;
4640 delayed_refs = &trans->transaction->delayed_refs;
4641 spin_lock(&delayed_refs->lock);
4642 head = btrfs_find_delayed_ref_head(trans, bytenr);
4646 node = rb_prev(&head->node.rb_node);
4650 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4652 /* there are still entries for this ref, we can't drop it */
4653 if (ref->bytenr == bytenr)
4656 if (head->extent_op) {
4657 if (!head->must_insert_reserved)
4659 kfree(head->extent_op);
4660 head->extent_op = NULL;
4664 * waiting for the lock here would deadlock. If someone else has it
4665 * locked they are already in the process of dropping it anyway
4667 if (!mutex_trylock(&head->mutex))
4671 * at this point we have a head with no other entries. Go
4672 * ahead and process it.
4674 head->node.in_tree = 0;
4675 rb_erase(&head->node.rb_node, &delayed_refs->root);
4677 delayed_refs->num_entries--;
4680 * we don't take a ref on the node because we're removing it from the
4681 * tree, so we just steal the ref the tree was holding.
4683 delayed_refs->num_heads--;
4684 if (list_empty(&head->cluster))
4685 delayed_refs->num_heads_ready--;
4687 list_del_init(&head->cluster);
4688 spin_unlock(&delayed_refs->lock);
4690 BUG_ON(head->extent_op);
4691 if (head->must_insert_reserved)
4694 mutex_unlock(&head->mutex);
4695 btrfs_put_delayed_ref(&head->node);
4698 spin_unlock(&delayed_refs->lock);
4702 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4703 struct btrfs_root *root,
4704 struct extent_buffer *buf,
4705 u64 parent, int last_ref)
4707 struct btrfs_block_rsv *block_rsv;
4708 struct btrfs_block_group_cache *cache = NULL;
4711 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4712 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4713 parent, root->root_key.objectid,
4714 btrfs_header_level(buf),
4715 BTRFS_DROP_DELAYED_REF, NULL);
4722 block_rsv = get_block_rsv(trans, root);
4723 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4724 if (block_rsv->space_info != cache->space_info)
4727 if (btrfs_header_generation(buf) == trans->transid) {
4728 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4729 ret = check_ref_cleanup(trans, root, buf->start);
4734 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4735 pin_down_extent(root, cache, buf->start, buf->len, 1);
4739 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4741 btrfs_add_free_space(cache, buf->start, buf->len);
4742 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4746 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4749 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4750 btrfs_put_block_group(cache);
4753 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4754 struct btrfs_root *root,
4755 u64 bytenr, u64 num_bytes, u64 parent,
4756 u64 root_objectid, u64 owner, u64 offset)
4761 * tree log blocks never actually go into the extent allocation
4762 * tree, just update pinning info and exit early.
4764 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4765 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4766 /* unlocks the pinned mutex */
4767 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4769 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4770 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4771 parent, root_objectid, (int)owner,
4772 BTRFS_DROP_DELAYED_REF, NULL);
4775 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4776 parent, root_objectid, owner,
4777 offset, BTRFS_DROP_DELAYED_REF, NULL);
4783 static u64 stripe_align(struct btrfs_root *root, u64 val)
4785 u64 mask = ((u64)root->stripesize - 1);
4786 u64 ret = (val + mask) & ~mask;
4791 * when we wait for progress in the block group caching, its because
4792 * our allocation attempt failed at least once. So, we must sleep
4793 * and let some progress happen before we try again.
4795 * This function will sleep at least once waiting for new free space to
4796 * show up, and then it will check the block group free space numbers
4797 * for our min num_bytes. Another option is to have it go ahead
4798 * and look in the rbtree for a free extent of a given size, but this
4802 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4805 struct btrfs_caching_control *caching_ctl;
4808 caching_ctl = get_caching_control(cache);
4812 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4813 (cache->free_space_ctl->free_space >= num_bytes));
4815 put_caching_control(caching_ctl);
4820 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4822 struct btrfs_caching_control *caching_ctl;
4825 caching_ctl = get_caching_control(cache);
4829 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4831 put_caching_control(caching_ctl);
4835 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4838 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4840 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4842 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4844 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4851 enum btrfs_loop_type {
4852 LOOP_FIND_IDEAL = 0,
4853 LOOP_CACHING_NOWAIT = 1,
4854 LOOP_CACHING_WAIT = 2,
4855 LOOP_ALLOC_CHUNK = 3,
4856 LOOP_NO_EMPTY_SIZE = 4,
4860 * walks the btree of allocated extents and find a hole of a given size.
4861 * The key ins is changed to record the hole:
4862 * ins->objectid == block start
4863 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4864 * ins->offset == number of blocks
4865 * Any available blocks before search_start are skipped.
4867 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4868 struct btrfs_root *orig_root,
4869 u64 num_bytes, u64 empty_size,
4870 u64 search_start, u64 search_end,
4871 u64 hint_byte, struct btrfs_key *ins,
4875 struct btrfs_root *root = orig_root->fs_info->extent_root;
4876 struct btrfs_free_cluster *last_ptr = NULL;
4877 struct btrfs_block_group_cache *block_group = NULL;
4878 int empty_cluster = 2 * 1024 * 1024;
4879 int allowed_chunk_alloc = 0;
4880 int done_chunk_alloc = 0;
4881 struct btrfs_space_info *space_info;
4882 int last_ptr_loop = 0;
4885 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
4886 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
4887 bool found_uncached_bg = false;
4888 bool failed_cluster_refill = false;
4889 bool failed_alloc = false;
4890 bool use_cluster = true;
4891 u64 ideal_cache_percent = 0;
4892 u64 ideal_cache_offset = 0;
4894 WARN_ON(num_bytes < root->sectorsize);
4895 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4899 space_info = __find_space_info(root->fs_info, data);
4901 printk(KERN_ERR "No space info for %llu\n", data);
4906 * If the space info is for both data and metadata it means we have a
4907 * small filesystem and we can't use the clustering stuff.
4909 if (btrfs_mixed_space_info(space_info))
4910 use_cluster = false;
4912 if (orig_root->ref_cows || empty_size)
4913 allowed_chunk_alloc = 1;
4915 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4916 last_ptr = &root->fs_info->meta_alloc_cluster;
4917 if (!btrfs_test_opt(root, SSD))
4918 empty_cluster = 64 * 1024;
4921 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4922 btrfs_test_opt(root, SSD)) {
4923 last_ptr = &root->fs_info->data_alloc_cluster;
4927 spin_lock(&last_ptr->lock);
4928 if (last_ptr->block_group)
4929 hint_byte = last_ptr->window_start;
4930 spin_unlock(&last_ptr->lock);
4933 search_start = max(search_start, first_logical_byte(root, 0));
4934 search_start = max(search_start, hint_byte);
4939 if (search_start == hint_byte) {
4941 block_group = btrfs_lookup_block_group(root->fs_info,
4944 * we don't want to use the block group if it doesn't match our
4945 * allocation bits, or if its not cached.
4947 * However if we are re-searching with an ideal block group
4948 * picked out then we don't care that the block group is cached.
4950 if (block_group && block_group_bits(block_group, data) &&
4951 (block_group->cached != BTRFS_CACHE_NO ||
4952 search_start == ideal_cache_offset)) {
4953 down_read(&space_info->groups_sem);
4954 if (list_empty(&block_group->list) ||
4957 * someone is removing this block group,
4958 * we can't jump into the have_block_group
4959 * target because our list pointers are not
4962 btrfs_put_block_group(block_group);
4963 up_read(&space_info->groups_sem);
4965 index = get_block_group_index(block_group);
4966 goto have_block_group;
4968 } else if (block_group) {
4969 btrfs_put_block_group(block_group);
4973 down_read(&space_info->groups_sem);
4974 list_for_each_entry(block_group, &space_info->block_groups[index],
4979 btrfs_get_block_group(block_group);
4980 search_start = block_group->key.objectid;
4983 * this can happen if we end up cycling through all the
4984 * raid types, but we want to make sure we only allocate
4985 * for the proper type.
4987 if (!block_group_bits(block_group, data)) {
4988 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4989 BTRFS_BLOCK_GROUP_RAID1 |
4990 BTRFS_BLOCK_GROUP_RAID10;
4993 * if they asked for extra copies and this block group
4994 * doesn't provide them, bail. This does allow us to
4995 * fill raid0 from raid1.
4997 if ((data & extra) && !(block_group->flags & extra))
5002 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5005 ret = cache_block_group(block_group, trans,
5007 if (block_group->cached == BTRFS_CACHE_FINISHED)
5008 goto have_block_group;
5010 free_percent = btrfs_block_group_used(&block_group->item);
5011 free_percent *= 100;
5012 free_percent = div64_u64(free_percent,
5013 block_group->key.offset);
5014 free_percent = 100 - free_percent;
5015 if (free_percent > ideal_cache_percent &&
5016 likely(!block_group->ro)) {
5017 ideal_cache_offset = block_group->key.objectid;
5018 ideal_cache_percent = free_percent;
5022 * The caching workers are limited to 2 threads, so we
5023 * can queue as much work as we care to.
5025 if (loop > LOOP_FIND_IDEAL) {
5026 ret = cache_block_group(block_group, trans,
5030 found_uncached_bg = true;
5033 * If loop is set for cached only, try the next block
5036 if (loop == LOOP_FIND_IDEAL)
5040 cached = block_group_cache_done(block_group);
5041 if (unlikely(!cached))
5042 found_uncached_bg = true;
5044 if (unlikely(block_group->ro))
5047 spin_lock(&block_group->free_space_ctl->tree_lock);
5049 block_group->free_space_ctl->free_space <
5050 num_bytes + empty_size) {
5051 spin_unlock(&block_group->free_space_ctl->tree_lock);
5054 spin_unlock(&block_group->free_space_ctl->tree_lock);
5057 * Ok we want to try and use the cluster allocator, so lets look
5058 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5059 * have tried the cluster allocator plenty of times at this
5060 * point and not have found anything, so we are likely way too
5061 * fragmented for the clustering stuff to find anything, so lets
5062 * just skip it and let the allocator find whatever block it can
5065 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5067 * the refill lock keeps out other
5068 * people trying to start a new cluster
5070 spin_lock(&last_ptr->refill_lock);
5071 if (last_ptr->block_group &&
5072 (last_ptr->block_group->ro ||
5073 !block_group_bits(last_ptr->block_group, data))) {
5075 goto refill_cluster;
5078 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5079 num_bytes, search_start);
5081 /* we have a block, we're done */
5082 spin_unlock(&last_ptr->refill_lock);
5086 spin_lock(&last_ptr->lock);
5088 * whoops, this cluster doesn't actually point to
5089 * this block group. Get a ref on the block
5090 * group is does point to and try again
5092 if (!last_ptr_loop && last_ptr->block_group &&
5093 last_ptr->block_group != block_group &&
5095 get_block_group_index(last_ptr->block_group)) {
5097 btrfs_put_block_group(block_group);
5098 block_group = last_ptr->block_group;
5099 btrfs_get_block_group(block_group);
5100 spin_unlock(&last_ptr->lock);
5101 spin_unlock(&last_ptr->refill_lock);
5104 search_start = block_group->key.objectid;
5106 * we know this block group is properly
5107 * in the list because
5108 * btrfs_remove_block_group, drops the
5109 * cluster before it removes the block
5110 * group from the list
5112 goto have_block_group;
5114 spin_unlock(&last_ptr->lock);
5117 * this cluster didn't work out, free it and
5120 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5124 /* allocate a cluster in this block group */
5125 ret = btrfs_find_space_cluster(trans, root,
5126 block_group, last_ptr,
5128 empty_cluster + empty_size);
5131 * now pull our allocation out of this
5134 offset = btrfs_alloc_from_cluster(block_group,
5135 last_ptr, num_bytes,
5138 /* we found one, proceed */
5139 spin_unlock(&last_ptr->refill_lock);
5142 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5143 && !failed_cluster_refill) {
5144 spin_unlock(&last_ptr->refill_lock);
5146 failed_cluster_refill = true;
5147 wait_block_group_cache_progress(block_group,
5148 num_bytes + empty_cluster + empty_size);
5149 goto have_block_group;
5153 * at this point we either didn't find a cluster
5154 * or we weren't able to allocate a block from our
5155 * cluster. Free the cluster we've been trying
5156 * to use, and go to the next block group
5158 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5159 spin_unlock(&last_ptr->refill_lock);
5163 offset = btrfs_find_space_for_alloc(block_group, search_start,
5164 num_bytes, empty_size);
5166 * If we didn't find a chunk, and we haven't failed on this
5167 * block group before, and this block group is in the middle of
5168 * caching and we are ok with waiting, then go ahead and wait
5169 * for progress to be made, and set failed_alloc to true.
5171 * If failed_alloc is true then we've already waited on this
5172 * block group once and should move on to the next block group.
5174 if (!offset && !failed_alloc && !cached &&
5175 loop > LOOP_CACHING_NOWAIT) {
5176 wait_block_group_cache_progress(block_group,
5177 num_bytes + empty_size);
5178 failed_alloc = true;
5179 goto have_block_group;
5180 } else if (!offset) {
5184 search_start = stripe_align(root, offset);
5185 /* move on to the next group */
5186 if (search_start + num_bytes >= search_end) {
5187 btrfs_add_free_space(block_group, offset, num_bytes);
5191 /* move on to the next group */
5192 if (search_start + num_bytes >
5193 block_group->key.objectid + block_group->key.offset) {
5194 btrfs_add_free_space(block_group, offset, num_bytes);
5198 ins->objectid = search_start;
5199 ins->offset = num_bytes;
5201 if (offset < search_start)
5202 btrfs_add_free_space(block_group, offset,
5203 search_start - offset);
5204 BUG_ON(offset > search_start);
5206 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5208 if (ret == -EAGAIN) {
5209 btrfs_add_free_space(block_group, offset, num_bytes);
5213 /* we are all good, lets return */
5214 ins->objectid = search_start;
5215 ins->offset = num_bytes;
5217 if (offset < search_start)
5218 btrfs_add_free_space(block_group, offset,
5219 search_start - offset);
5220 BUG_ON(offset > search_start);
5221 btrfs_put_block_group(block_group);
5224 failed_cluster_refill = false;
5225 failed_alloc = false;
5226 BUG_ON(index != get_block_group_index(block_group));
5227 btrfs_put_block_group(block_group);
5229 up_read(&space_info->groups_sem);
5231 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5234 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5235 * for them to make caching progress. Also
5236 * determine the best possible bg to cache
5237 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5238 * caching kthreads as we move along
5239 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5240 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5241 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5244 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5246 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5247 found_uncached_bg = false;
5249 if (!ideal_cache_percent)
5253 * 1 of the following 2 things have happened so far
5255 * 1) We found an ideal block group for caching that
5256 * is mostly full and will cache quickly, so we might
5257 * as well wait for it.
5259 * 2) We searched for cached only and we didn't find
5260 * anything, and we didn't start any caching kthreads
5261 * either, so chances are we will loop through and
5262 * start a couple caching kthreads, and then come back
5263 * around and just wait for them. This will be slower
5264 * because we will have 2 caching kthreads reading at
5265 * the same time when we could have just started one
5266 * and waited for it to get far enough to give us an
5267 * allocation, so go ahead and go to the wait caching
5270 loop = LOOP_CACHING_WAIT;
5271 search_start = ideal_cache_offset;
5272 ideal_cache_percent = 0;
5274 } else if (loop == LOOP_FIND_IDEAL) {
5276 * Didn't find a uncached bg, wait on anything we find
5279 loop = LOOP_CACHING_WAIT;
5285 if (loop == LOOP_ALLOC_CHUNK) {
5286 if (allowed_chunk_alloc) {
5287 ret = do_chunk_alloc(trans, root, num_bytes +
5288 2 * 1024 * 1024, data,
5289 CHUNK_ALLOC_LIMITED);
5290 allowed_chunk_alloc = 0;
5292 done_chunk_alloc = 1;
5293 } else if (!done_chunk_alloc &&
5294 space_info->force_alloc ==
5295 CHUNK_ALLOC_NO_FORCE) {
5296 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5300 * We didn't allocate a chunk, go ahead and drop the
5301 * empty size and loop again.
5303 if (!done_chunk_alloc)
5304 loop = LOOP_NO_EMPTY_SIZE;
5307 if (loop == LOOP_NO_EMPTY_SIZE) {
5313 } else if (!ins->objectid) {
5315 } else if (ins->objectid) {
5322 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5323 int dump_block_groups)
5325 struct btrfs_block_group_cache *cache;
5328 spin_lock(&info->lock);
5329 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5330 (unsigned long long)info->flags,
5331 (unsigned long long)(info->total_bytes - info->bytes_used -
5332 info->bytes_pinned - info->bytes_reserved -
5333 info->bytes_readonly),
5334 (info->full) ? "" : "not ");
5335 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5336 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5337 (unsigned long long)info->total_bytes,
5338 (unsigned long long)info->bytes_used,
5339 (unsigned long long)info->bytes_pinned,
5340 (unsigned long long)info->bytes_reserved,
5341 (unsigned long long)info->bytes_may_use,
5342 (unsigned long long)info->bytes_readonly);
5343 spin_unlock(&info->lock);
5345 if (!dump_block_groups)
5348 down_read(&info->groups_sem);
5350 list_for_each_entry(cache, &info->block_groups[index], list) {
5351 spin_lock(&cache->lock);
5352 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5353 "%llu pinned %llu reserved\n",
5354 (unsigned long long)cache->key.objectid,
5355 (unsigned long long)cache->key.offset,
5356 (unsigned long long)btrfs_block_group_used(&cache->item),
5357 (unsigned long long)cache->pinned,
5358 (unsigned long long)cache->reserved);
5359 btrfs_dump_free_space(cache, bytes);
5360 spin_unlock(&cache->lock);
5362 if (++index < BTRFS_NR_RAID_TYPES)
5364 up_read(&info->groups_sem);
5367 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5368 struct btrfs_root *root,
5369 u64 num_bytes, u64 min_alloc_size,
5370 u64 empty_size, u64 hint_byte,
5371 u64 search_end, struct btrfs_key *ins,
5375 u64 search_start = 0;
5377 data = btrfs_get_alloc_profile(root, data);
5380 * the only place that sets empty_size is btrfs_realloc_node, which
5381 * is not called recursively on allocations
5383 if (empty_size || root->ref_cows)
5384 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5385 num_bytes + 2 * 1024 * 1024, data,
5386 CHUNK_ALLOC_NO_FORCE);
5388 WARN_ON(num_bytes < root->sectorsize);
5389 ret = find_free_extent(trans, root, num_bytes, empty_size,
5390 search_start, search_end, hint_byte,
5393 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5394 num_bytes = num_bytes >> 1;
5395 num_bytes = num_bytes & ~(root->sectorsize - 1);
5396 num_bytes = max(num_bytes, min_alloc_size);
5397 do_chunk_alloc(trans, root->fs_info->extent_root,
5398 num_bytes, data, CHUNK_ALLOC_FORCE);
5401 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5402 struct btrfs_space_info *sinfo;
5404 sinfo = __find_space_info(root->fs_info, data);
5405 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5406 "wanted %llu\n", (unsigned long long)data,
5407 (unsigned long long)num_bytes);
5408 dump_space_info(sinfo, num_bytes, 1);
5411 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5416 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5418 struct btrfs_block_group_cache *cache;
5421 cache = btrfs_lookup_block_group(root->fs_info, start);
5423 printk(KERN_ERR "Unable to find block group for %llu\n",
5424 (unsigned long long)start);
5428 if (btrfs_test_opt(root, DISCARD))
5429 ret = btrfs_discard_extent(root, start, len, NULL);
5431 btrfs_add_free_space(cache, start, len);
5432 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5433 btrfs_put_block_group(cache);
5435 trace_btrfs_reserved_extent_free(root, start, len);
5440 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5441 struct btrfs_root *root,
5442 u64 parent, u64 root_objectid,
5443 u64 flags, u64 owner, u64 offset,
5444 struct btrfs_key *ins, int ref_mod)
5447 struct btrfs_fs_info *fs_info = root->fs_info;
5448 struct btrfs_extent_item *extent_item;
5449 struct btrfs_extent_inline_ref *iref;
5450 struct btrfs_path *path;
5451 struct extent_buffer *leaf;
5456 type = BTRFS_SHARED_DATA_REF_KEY;
5458 type = BTRFS_EXTENT_DATA_REF_KEY;
5460 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5462 path = btrfs_alloc_path();
5466 path->leave_spinning = 1;
5467 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5471 leaf = path->nodes[0];
5472 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5473 struct btrfs_extent_item);
5474 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5475 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5476 btrfs_set_extent_flags(leaf, extent_item,
5477 flags | BTRFS_EXTENT_FLAG_DATA);
5479 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5480 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5482 struct btrfs_shared_data_ref *ref;
5483 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5484 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5485 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5487 struct btrfs_extent_data_ref *ref;
5488 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5489 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5490 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5491 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5492 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5495 btrfs_mark_buffer_dirty(path->nodes[0]);
5496 btrfs_free_path(path);
5498 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5500 printk(KERN_ERR "btrfs update block group failed for %llu "
5501 "%llu\n", (unsigned long long)ins->objectid,
5502 (unsigned long long)ins->offset);
5508 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5509 struct btrfs_root *root,
5510 u64 parent, u64 root_objectid,
5511 u64 flags, struct btrfs_disk_key *key,
5512 int level, struct btrfs_key *ins)
5515 struct btrfs_fs_info *fs_info = root->fs_info;
5516 struct btrfs_extent_item *extent_item;
5517 struct btrfs_tree_block_info *block_info;
5518 struct btrfs_extent_inline_ref *iref;
5519 struct btrfs_path *path;
5520 struct extent_buffer *leaf;
5521 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5523 path = btrfs_alloc_path();
5527 path->leave_spinning = 1;
5528 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5532 leaf = path->nodes[0];
5533 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5534 struct btrfs_extent_item);
5535 btrfs_set_extent_refs(leaf, extent_item, 1);
5536 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5537 btrfs_set_extent_flags(leaf, extent_item,
5538 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5539 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5541 btrfs_set_tree_block_key(leaf, block_info, key);
5542 btrfs_set_tree_block_level(leaf, block_info, level);
5544 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5546 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5547 btrfs_set_extent_inline_ref_type(leaf, iref,
5548 BTRFS_SHARED_BLOCK_REF_KEY);
5549 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5551 btrfs_set_extent_inline_ref_type(leaf, iref,
5552 BTRFS_TREE_BLOCK_REF_KEY);
5553 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5556 btrfs_mark_buffer_dirty(leaf);
5557 btrfs_free_path(path);
5559 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5561 printk(KERN_ERR "btrfs update block group failed for %llu "
5562 "%llu\n", (unsigned long long)ins->objectid,
5563 (unsigned long long)ins->offset);
5569 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5570 struct btrfs_root *root,
5571 u64 root_objectid, u64 owner,
5572 u64 offset, struct btrfs_key *ins)
5576 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5578 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5579 0, root_objectid, owner, offset,
5580 BTRFS_ADD_DELAYED_EXTENT, NULL);
5585 * this is used by the tree logging recovery code. It records that
5586 * an extent has been allocated and makes sure to clear the free
5587 * space cache bits as well
5589 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5590 struct btrfs_root *root,
5591 u64 root_objectid, u64 owner, u64 offset,
5592 struct btrfs_key *ins)
5595 struct btrfs_block_group_cache *block_group;
5596 struct btrfs_caching_control *caching_ctl;
5597 u64 start = ins->objectid;
5598 u64 num_bytes = ins->offset;
5600 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5601 cache_block_group(block_group, trans, NULL, 0);
5602 caching_ctl = get_caching_control(block_group);
5605 BUG_ON(!block_group_cache_done(block_group));
5606 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5609 mutex_lock(&caching_ctl->mutex);
5611 if (start >= caching_ctl->progress) {
5612 ret = add_excluded_extent(root, start, num_bytes);
5614 } else if (start + num_bytes <= caching_ctl->progress) {
5615 ret = btrfs_remove_free_space(block_group,
5619 num_bytes = caching_ctl->progress - start;
5620 ret = btrfs_remove_free_space(block_group,
5624 start = caching_ctl->progress;
5625 num_bytes = ins->objectid + ins->offset -
5626 caching_ctl->progress;
5627 ret = add_excluded_extent(root, start, num_bytes);
5631 mutex_unlock(&caching_ctl->mutex);
5632 put_caching_control(caching_ctl);
5635 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5636 RESERVE_ALLOC_NO_ACCOUNT);
5638 btrfs_put_block_group(block_group);
5639 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5640 0, owner, offset, ins, 1);
5644 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5645 struct btrfs_root *root,
5646 u64 bytenr, u32 blocksize,
5649 struct extent_buffer *buf;
5651 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5653 return ERR_PTR(-ENOMEM);
5654 btrfs_set_header_generation(buf, trans->transid);
5655 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5656 btrfs_tree_lock(buf);
5657 clean_tree_block(trans, root, buf);
5659 btrfs_set_lock_blocking(buf);
5660 btrfs_set_buffer_uptodate(buf);
5662 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5664 * we allow two log transactions at a time, use different
5665 * EXENT bit to differentiate dirty pages.
5667 if (root->log_transid % 2 == 0)
5668 set_extent_dirty(&root->dirty_log_pages, buf->start,
5669 buf->start + buf->len - 1, GFP_NOFS);
5671 set_extent_new(&root->dirty_log_pages, buf->start,
5672 buf->start + buf->len - 1, GFP_NOFS);
5674 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5675 buf->start + buf->len - 1, GFP_NOFS);
5677 trans->blocks_used++;
5678 /* this returns a buffer locked for blocking */
5682 static struct btrfs_block_rsv *
5683 use_block_rsv(struct btrfs_trans_handle *trans,
5684 struct btrfs_root *root, u32 blocksize)
5686 struct btrfs_block_rsv *block_rsv;
5687 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5690 block_rsv = get_block_rsv(trans, root);
5692 if (block_rsv->size == 0) {
5693 ret = reserve_metadata_bytes(trans, root, block_rsv,
5696 * If we couldn't reserve metadata bytes try and use some from
5697 * the global reserve.
5699 if (ret && block_rsv != global_rsv) {
5700 ret = block_rsv_use_bytes(global_rsv, blocksize);
5703 return ERR_PTR(ret);
5705 return ERR_PTR(ret);
5710 ret = block_rsv_use_bytes(block_rsv, blocksize);
5715 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5719 } else if (ret && block_rsv != global_rsv) {
5720 ret = block_rsv_use_bytes(global_rsv, blocksize);
5726 return ERR_PTR(-ENOSPC);
5729 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5731 block_rsv_add_bytes(block_rsv, blocksize, 0);
5732 block_rsv_release_bytes(block_rsv, NULL, 0);
5736 * finds a free extent and does all the dirty work required for allocation
5737 * returns the key for the extent through ins, and a tree buffer for
5738 * the first block of the extent through buf.
5740 * returns the tree buffer or NULL.
5742 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5743 struct btrfs_root *root, u32 blocksize,
5744 u64 parent, u64 root_objectid,
5745 struct btrfs_disk_key *key, int level,
5746 u64 hint, u64 empty_size)
5748 struct btrfs_key ins;
5749 struct btrfs_block_rsv *block_rsv;
5750 struct extent_buffer *buf;
5755 block_rsv = use_block_rsv(trans, root, blocksize);
5756 if (IS_ERR(block_rsv))
5757 return ERR_CAST(block_rsv);
5759 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5760 empty_size, hint, (u64)-1, &ins, 0);
5762 unuse_block_rsv(block_rsv, blocksize);
5763 return ERR_PTR(ret);
5766 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5768 BUG_ON(IS_ERR(buf));
5770 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5772 parent = ins.objectid;
5773 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5777 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5778 struct btrfs_delayed_extent_op *extent_op;
5779 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5782 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5784 memset(&extent_op->key, 0, sizeof(extent_op->key));
5785 extent_op->flags_to_set = flags;
5786 extent_op->update_key = 1;
5787 extent_op->update_flags = 1;
5788 extent_op->is_data = 0;
5790 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5791 ins.offset, parent, root_objectid,
5792 level, BTRFS_ADD_DELAYED_EXTENT,
5799 struct walk_control {
5800 u64 refs[BTRFS_MAX_LEVEL];
5801 u64 flags[BTRFS_MAX_LEVEL];
5802 struct btrfs_key update_progress;
5812 #define DROP_REFERENCE 1
5813 #define UPDATE_BACKREF 2
5815 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5816 struct btrfs_root *root,
5817 struct walk_control *wc,
5818 struct btrfs_path *path)
5826 struct btrfs_key key;
5827 struct extent_buffer *eb;
5832 if (path->slots[wc->level] < wc->reada_slot) {
5833 wc->reada_count = wc->reada_count * 2 / 3;
5834 wc->reada_count = max(wc->reada_count, 2);
5836 wc->reada_count = wc->reada_count * 3 / 2;
5837 wc->reada_count = min_t(int, wc->reada_count,
5838 BTRFS_NODEPTRS_PER_BLOCK(root));
5841 eb = path->nodes[wc->level];
5842 nritems = btrfs_header_nritems(eb);
5843 blocksize = btrfs_level_size(root, wc->level - 1);
5845 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5846 if (nread >= wc->reada_count)
5850 bytenr = btrfs_node_blockptr(eb, slot);
5851 generation = btrfs_node_ptr_generation(eb, slot);
5853 if (slot == path->slots[wc->level])
5856 if (wc->stage == UPDATE_BACKREF &&
5857 generation <= root->root_key.offset)
5860 /* We don't lock the tree block, it's OK to be racy here */
5861 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5866 if (wc->stage == DROP_REFERENCE) {
5870 if (wc->level == 1 &&
5871 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5873 if (!wc->update_ref ||
5874 generation <= root->root_key.offset)
5876 btrfs_node_key_to_cpu(eb, &key, slot);
5877 ret = btrfs_comp_cpu_keys(&key,
5878 &wc->update_progress);
5882 if (wc->level == 1 &&
5883 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5887 ret = readahead_tree_block(root, bytenr, blocksize,
5893 wc->reada_slot = slot;
5897 * hepler to process tree block while walking down the tree.
5899 * when wc->stage == UPDATE_BACKREF, this function updates
5900 * back refs for pointers in the block.
5902 * NOTE: return value 1 means we should stop walking down.
5904 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5905 struct btrfs_root *root,
5906 struct btrfs_path *path,
5907 struct walk_control *wc, int lookup_info)
5909 int level = wc->level;
5910 struct extent_buffer *eb = path->nodes[level];
5911 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5914 if (wc->stage == UPDATE_BACKREF &&
5915 btrfs_header_owner(eb) != root->root_key.objectid)
5919 * when reference count of tree block is 1, it won't increase
5920 * again. once full backref flag is set, we never clear it.
5923 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5924 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5925 BUG_ON(!path->locks[level]);
5926 ret = btrfs_lookup_extent_info(trans, root,
5931 BUG_ON(wc->refs[level] == 0);
5934 if (wc->stage == DROP_REFERENCE) {
5935 if (wc->refs[level] > 1)
5938 if (path->locks[level] && !wc->keep_locks) {
5939 btrfs_tree_unlock_rw(eb, path->locks[level]);
5940 path->locks[level] = 0;
5945 /* wc->stage == UPDATE_BACKREF */
5946 if (!(wc->flags[level] & flag)) {
5947 BUG_ON(!path->locks[level]);
5948 ret = btrfs_inc_ref(trans, root, eb, 1);
5950 ret = btrfs_dec_ref(trans, root, eb, 0);
5952 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5955 wc->flags[level] |= flag;
5959 * the block is shared by multiple trees, so it's not good to
5960 * keep the tree lock
5962 if (path->locks[level] && level > 0) {
5963 btrfs_tree_unlock_rw(eb, path->locks[level]);
5964 path->locks[level] = 0;
5970 * hepler to process tree block pointer.
5972 * when wc->stage == DROP_REFERENCE, this function checks
5973 * reference count of the block pointed to. if the block
5974 * is shared and we need update back refs for the subtree
5975 * rooted at the block, this function changes wc->stage to
5976 * UPDATE_BACKREF. if the block is shared and there is no
5977 * need to update back, this function drops the reference
5980 * NOTE: return value 1 means we should stop walking down.
5982 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5983 struct btrfs_root *root,
5984 struct btrfs_path *path,
5985 struct walk_control *wc, int *lookup_info)
5991 struct btrfs_key key;
5992 struct extent_buffer *next;
5993 int level = wc->level;
5997 generation = btrfs_node_ptr_generation(path->nodes[level],
5998 path->slots[level]);
6000 * if the lower level block was created before the snapshot
6001 * was created, we know there is no need to update back refs
6004 if (wc->stage == UPDATE_BACKREF &&
6005 generation <= root->root_key.offset) {
6010 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6011 blocksize = btrfs_level_size(root, level - 1);
6013 next = btrfs_find_tree_block(root, bytenr, blocksize);
6015 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6020 btrfs_tree_lock(next);
6021 btrfs_set_lock_blocking(next);
6023 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6024 &wc->refs[level - 1],
6025 &wc->flags[level - 1]);
6027 BUG_ON(wc->refs[level - 1] == 0);
6030 if (wc->stage == DROP_REFERENCE) {
6031 if (wc->refs[level - 1] > 1) {
6033 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6036 if (!wc->update_ref ||
6037 generation <= root->root_key.offset)
6040 btrfs_node_key_to_cpu(path->nodes[level], &key,
6041 path->slots[level]);
6042 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6046 wc->stage = UPDATE_BACKREF;
6047 wc->shared_level = level - 1;
6051 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6055 if (!btrfs_buffer_uptodate(next, generation)) {
6056 btrfs_tree_unlock(next);
6057 free_extent_buffer(next);
6063 if (reada && level == 1)
6064 reada_walk_down(trans, root, wc, path);
6065 next = read_tree_block(root, bytenr, blocksize, generation);
6068 btrfs_tree_lock(next);
6069 btrfs_set_lock_blocking(next);
6073 BUG_ON(level != btrfs_header_level(next));
6074 path->nodes[level] = next;
6075 path->slots[level] = 0;
6076 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6082 wc->refs[level - 1] = 0;
6083 wc->flags[level - 1] = 0;
6084 if (wc->stage == DROP_REFERENCE) {
6085 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6086 parent = path->nodes[level]->start;
6088 BUG_ON(root->root_key.objectid !=
6089 btrfs_header_owner(path->nodes[level]));
6093 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6094 root->root_key.objectid, level - 1, 0);
6097 btrfs_tree_unlock(next);
6098 free_extent_buffer(next);
6104 * hepler to process tree block while walking up the tree.
6106 * when wc->stage == DROP_REFERENCE, this function drops
6107 * reference count on the block.
6109 * when wc->stage == UPDATE_BACKREF, this function changes
6110 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6111 * to UPDATE_BACKREF previously while processing the block.
6113 * NOTE: return value 1 means we should stop walking up.
6115 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6116 struct btrfs_root *root,
6117 struct btrfs_path *path,
6118 struct walk_control *wc)
6121 int level = wc->level;
6122 struct extent_buffer *eb = path->nodes[level];
6125 if (wc->stage == UPDATE_BACKREF) {
6126 BUG_ON(wc->shared_level < level);
6127 if (level < wc->shared_level)
6130 ret = find_next_key(path, level + 1, &wc->update_progress);
6134 wc->stage = DROP_REFERENCE;
6135 wc->shared_level = -1;
6136 path->slots[level] = 0;
6139 * check reference count again if the block isn't locked.
6140 * we should start walking down the tree again if reference
6143 if (!path->locks[level]) {
6145 btrfs_tree_lock(eb);
6146 btrfs_set_lock_blocking(eb);
6147 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6149 ret = btrfs_lookup_extent_info(trans, root,
6154 BUG_ON(wc->refs[level] == 0);
6155 if (wc->refs[level] == 1) {
6156 btrfs_tree_unlock_rw(eb, path->locks[level]);
6162 /* wc->stage == DROP_REFERENCE */
6163 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6165 if (wc->refs[level] == 1) {
6167 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6168 ret = btrfs_dec_ref(trans, root, eb, 1);
6170 ret = btrfs_dec_ref(trans, root, eb, 0);
6173 /* make block locked assertion in clean_tree_block happy */
6174 if (!path->locks[level] &&
6175 btrfs_header_generation(eb) == trans->transid) {
6176 btrfs_tree_lock(eb);
6177 btrfs_set_lock_blocking(eb);
6178 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6180 clean_tree_block(trans, root, eb);
6183 if (eb == root->node) {
6184 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6187 BUG_ON(root->root_key.objectid !=
6188 btrfs_header_owner(eb));
6190 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6191 parent = path->nodes[level + 1]->start;
6193 BUG_ON(root->root_key.objectid !=
6194 btrfs_header_owner(path->nodes[level + 1]));
6197 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6199 wc->refs[level] = 0;
6200 wc->flags[level] = 0;
6204 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6205 struct btrfs_root *root,
6206 struct btrfs_path *path,
6207 struct walk_control *wc)
6209 int level = wc->level;
6210 int lookup_info = 1;
6213 while (level >= 0) {
6214 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6221 if (path->slots[level] >=
6222 btrfs_header_nritems(path->nodes[level]))
6225 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6227 path->slots[level]++;
6236 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6237 struct btrfs_root *root,
6238 struct btrfs_path *path,
6239 struct walk_control *wc, int max_level)
6241 int level = wc->level;
6244 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6245 while (level < max_level && path->nodes[level]) {
6247 if (path->slots[level] + 1 <
6248 btrfs_header_nritems(path->nodes[level])) {
6249 path->slots[level]++;
6252 ret = walk_up_proc(trans, root, path, wc);
6256 if (path->locks[level]) {
6257 btrfs_tree_unlock_rw(path->nodes[level],
6258 path->locks[level]);
6259 path->locks[level] = 0;
6261 free_extent_buffer(path->nodes[level]);
6262 path->nodes[level] = NULL;
6270 * drop a subvolume tree.
6272 * this function traverses the tree freeing any blocks that only
6273 * referenced by the tree.
6275 * when a shared tree block is found. this function decreases its
6276 * reference count by one. if update_ref is true, this function
6277 * also make sure backrefs for the shared block and all lower level
6278 * blocks are properly updated.
6280 void btrfs_drop_snapshot(struct btrfs_root *root,
6281 struct btrfs_block_rsv *block_rsv, int update_ref)
6283 struct btrfs_path *path;
6284 struct btrfs_trans_handle *trans;
6285 struct btrfs_root *tree_root = root->fs_info->tree_root;
6286 struct btrfs_root_item *root_item = &root->root_item;
6287 struct walk_control *wc;
6288 struct btrfs_key key;
6293 path = btrfs_alloc_path();
6299 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6301 btrfs_free_path(path);
6306 trans = btrfs_start_transaction(tree_root, 0);
6307 BUG_ON(IS_ERR(trans));
6310 trans->block_rsv = block_rsv;
6312 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6313 level = btrfs_header_level(root->node);
6314 path->nodes[level] = btrfs_lock_root_node(root);
6315 btrfs_set_lock_blocking(path->nodes[level]);
6316 path->slots[level] = 0;
6317 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6318 memset(&wc->update_progress, 0,
6319 sizeof(wc->update_progress));
6321 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6322 memcpy(&wc->update_progress, &key,
6323 sizeof(wc->update_progress));
6325 level = root_item->drop_level;
6327 path->lowest_level = level;
6328 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6329 path->lowest_level = 0;
6337 * unlock our path, this is safe because only this
6338 * function is allowed to delete this snapshot
6340 btrfs_unlock_up_safe(path, 0);
6342 level = btrfs_header_level(root->node);
6344 btrfs_tree_lock(path->nodes[level]);
6345 btrfs_set_lock_blocking(path->nodes[level]);
6347 ret = btrfs_lookup_extent_info(trans, root,
6348 path->nodes[level]->start,
6349 path->nodes[level]->len,
6353 BUG_ON(wc->refs[level] == 0);
6355 if (level == root_item->drop_level)
6358 btrfs_tree_unlock(path->nodes[level]);
6359 WARN_ON(wc->refs[level] != 1);
6365 wc->shared_level = -1;
6366 wc->stage = DROP_REFERENCE;
6367 wc->update_ref = update_ref;
6369 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6372 ret = walk_down_tree(trans, root, path, wc);
6378 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6385 BUG_ON(wc->stage != DROP_REFERENCE);
6389 if (wc->stage == DROP_REFERENCE) {
6391 btrfs_node_key(path->nodes[level],
6392 &root_item->drop_progress,
6393 path->slots[level]);
6394 root_item->drop_level = level;
6397 BUG_ON(wc->level == 0);
6398 if (btrfs_should_end_transaction(trans, tree_root)) {
6399 ret = btrfs_update_root(trans, tree_root,
6404 btrfs_end_transaction_throttle(trans, tree_root);
6405 trans = btrfs_start_transaction(tree_root, 0);
6406 BUG_ON(IS_ERR(trans));
6408 trans->block_rsv = block_rsv;
6411 btrfs_release_path(path);
6414 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6417 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6418 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6422 /* if we fail to delete the orphan item this time
6423 * around, it'll get picked up the next time.
6425 * The most common failure here is just -ENOENT.
6427 btrfs_del_orphan_item(trans, tree_root,
6428 root->root_key.objectid);
6432 if (root->in_radix) {
6433 btrfs_free_fs_root(tree_root->fs_info, root);
6435 free_extent_buffer(root->node);
6436 free_extent_buffer(root->commit_root);
6440 btrfs_end_transaction_throttle(trans, tree_root);
6442 btrfs_free_path(path);
6445 btrfs_std_error(root->fs_info, err);
6450 * drop subtree rooted at tree block 'node'.
6452 * NOTE: this function will unlock and release tree block 'node'
6454 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6455 struct btrfs_root *root,
6456 struct extent_buffer *node,
6457 struct extent_buffer *parent)
6459 struct btrfs_path *path;
6460 struct walk_control *wc;
6466 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6468 path = btrfs_alloc_path();
6472 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6474 btrfs_free_path(path);
6478 btrfs_assert_tree_locked(parent);
6479 parent_level = btrfs_header_level(parent);
6480 extent_buffer_get(parent);
6481 path->nodes[parent_level] = parent;
6482 path->slots[parent_level] = btrfs_header_nritems(parent);
6484 btrfs_assert_tree_locked(node);
6485 level = btrfs_header_level(node);
6486 path->nodes[level] = node;
6487 path->slots[level] = 0;
6488 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6490 wc->refs[parent_level] = 1;
6491 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6493 wc->shared_level = -1;
6494 wc->stage = DROP_REFERENCE;
6497 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6500 wret = walk_down_tree(trans, root, path, wc);
6506 wret = walk_up_tree(trans, root, path, wc, parent_level);
6514 btrfs_free_path(path);
6518 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6521 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6522 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6525 * we add in the count of missing devices because we want
6526 * to make sure that any RAID levels on a degraded FS
6527 * continue to be honored.
6529 num_devices = root->fs_info->fs_devices->rw_devices +
6530 root->fs_info->fs_devices->missing_devices;
6532 if (num_devices == 1) {
6533 stripped |= BTRFS_BLOCK_GROUP_DUP;
6534 stripped = flags & ~stripped;
6536 /* turn raid0 into single device chunks */
6537 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6540 /* turn mirroring into duplication */
6541 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6542 BTRFS_BLOCK_GROUP_RAID10))
6543 return stripped | BTRFS_BLOCK_GROUP_DUP;
6546 /* they already had raid on here, just return */
6547 if (flags & stripped)
6550 stripped |= BTRFS_BLOCK_GROUP_DUP;
6551 stripped = flags & ~stripped;
6553 /* switch duplicated blocks with raid1 */
6554 if (flags & BTRFS_BLOCK_GROUP_DUP)
6555 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6557 /* turn single device chunks into raid0 */
6558 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6563 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6565 struct btrfs_space_info *sinfo = cache->space_info;
6567 u64 min_allocable_bytes;
6572 * We need some metadata space and system metadata space for
6573 * allocating chunks in some corner cases until we force to set
6574 * it to be readonly.
6577 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6579 min_allocable_bytes = 1 * 1024 * 1024;
6581 min_allocable_bytes = 0;
6583 spin_lock(&sinfo->lock);
6584 spin_lock(&cache->lock);
6591 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6592 cache->bytes_super - btrfs_block_group_used(&cache->item);
6594 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6595 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6596 min_allocable_bytes <= sinfo->total_bytes) {
6597 sinfo->bytes_readonly += num_bytes;
6602 spin_unlock(&cache->lock);
6603 spin_unlock(&sinfo->lock);
6607 int btrfs_set_block_group_ro(struct btrfs_root *root,
6608 struct btrfs_block_group_cache *cache)
6611 struct btrfs_trans_handle *trans;
6617 trans = btrfs_join_transaction(root);
6618 BUG_ON(IS_ERR(trans));
6620 alloc_flags = update_block_group_flags(root, cache->flags);
6621 if (alloc_flags != cache->flags)
6622 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6625 ret = set_block_group_ro(cache, 0);
6628 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6629 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6633 ret = set_block_group_ro(cache, 0);
6635 btrfs_end_transaction(trans, root);
6639 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6640 struct btrfs_root *root, u64 type)
6642 u64 alloc_flags = get_alloc_profile(root, type);
6643 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6648 * helper to account the unused space of all the readonly block group in the
6649 * list. takes mirrors into account.
6651 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6653 struct btrfs_block_group_cache *block_group;
6657 list_for_each_entry(block_group, groups_list, list) {
6658 spin_lock(&block_group->lock);
6660 if (!block_group->ro) {
6661 spin_unlock(&block_group->lock);
6665 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6666 BTRFS_BLOCK_GROUP_RAID10 |
6667 BTRFS_BLOCK_GROUP_DUP))
6672 free_bytes += (block_group->key.offset -
6673 btrfs_block_group_used(&block_group->item)) *
6676 spin_unlock(&block_group->lock);
6683 * helper to account the unused space of all the readonly block group in the
6684 * space_info. takes mirrors into account.
6686 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6691 spin_lock(&sinfo->lock);
6693 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6694 if (!list_empty(&sinfo->block_groups[i]))
6695 free_bytes += __btrfs_get_ro_block_group_free_space(
6696 &sinfo->block_groups[i]);
6698 spin_unlock(&sinfo->lock);
6703 int btrfs_set_block_group_rw(struct btrfs_root *root,
6704 struct btrfs_block_group_cache *cache)
6706 struct btrfs_space_info *sinfo = cache->space_info;
6711 spin_lock(&sinfo->lock);
6712 spin_lock(&cache->lock);
6713 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6714 cache->bytes_super - btrfs_block_group_used(&cache->item);
6715 sinfo->bytes_readonly -= num_bytes;
6717 spin_unlock(&cache->lock);
6718 spin_unlock(&sinfo->lock);
6723 * checks to see if its even possible to relocate this block group.
6725 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6726 * ok to go ahead and try.
6728 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6730 struct btrfs_block_group_cache *block_group;
6731 struct btrfs_space_info *space_info;
6732 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6733 struct btrfs_device *device;
6741 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6743 /* odd, couldn't find the block group, leave it alone */
6747 min_free = btrfs_block_group_used(&block_group->item);
6749 /* no bytes used, we're good */
6753 space_info = block_group->space_info;
6754 spin_lock(&space_info->lock);
6756 full = space_info->full;
6759 * if this is the last block group we have in this space, we can't
6760 * relocate it unless we're able to allocate a new chunk below.
6762 * Otherwise, we need to make sure we have room in the space to handle
6763 * all of the extents from this block group. If we can, we're good
6765 if ((space_info->total_bytes != block_group->key.offset) &&
6766 (space_info->bytes_used + space_info->bytes_reserved +
6767 space_info->bytes_pinned + space_info->bytes_readonly +
6768 min_free < space_info->total_bytes)) {
6769 spin_unlock(&space_info->lock);
6772 spin_unlock(&space_info->lock);
6775 * ok we don't have enough space, but maybe we have free space on our
6776 * devices to allocate new chunks for relocation, so loop through our
6777 * alloc devices and guess if we have enough space. However, if we
6778 * were marked as full, then we know there aren't enough chunks, and we
6793 index = get_block_group_index(block_group);
6798 } else if (index == 1) {
6800 } else if (index == 2) {
6803 } else if (index == 3) {
6804 dev_min = fs_devices->rw_devices;
6805 do_div(min_free, dev_min);
6808 mutex_lock(&root->fs_info->chunk_mutex);
6809 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6813 * check to make sure we can actually find a chunk with enough
6814 * space to fit our block group in.
6816 if (device->total_bytes > device->bytes_used + min_free) {
6817 ret = find_free_dev_extent(NULL, device, min_free,
6822 if (dev_nr >= dev_min)
6828 mutex_unlock(&root->fs_info->chunk_mutex);
6830 btrfs_put_block_group(block_group);
6834 static int find_first_block_group(struct btrfs_root *root,
6835 struct btrfs_path *path, struct btrfs_key *key)
6838 struct btrfs_key found_key;
6839 struct extent_buffer *leaf;
6842 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6847 slot = path->slots[0];
6848 leaf = path->nodes[0];
6849 if (slot >= btrfs_header_nritems(leaf)) {
6850 ret = btrfs_next_leaf(root, path);
6857 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6859 if (found_key.objectid >= key->objectid &&
6860 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6870 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6872 struct btrfs_block_group_cache *block_group;
6876 struct inode *inode;
6878 block_group = btrfs_lookup_first_block_group(info, last);
6879 while (block_group) {
6880 spin_lock(&block_group->lock);
6881 if (block_group->iref)
6883 spin_unlock(&block_group->lock);
6884 block_group = next_block_group(info->tree_root,
6894 inode = block_group->inode;
6895 block_group->iref = 0;
6896 block_group->inode = NULL;
6897 spin_unlock(&block_group->lock);
6899 last = block_group->key.objectid + block_group->key.offset;
6900 btrfs_put_block_group(block_group);
6904 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6906 struct btrfs_block_group_cache *block_group;
6907 struct btrfs_space_info *space_info;
6908 struct btrfs_caching_control *caching_ctl;
6911 down_write(&info->extent_commit_sem);
6912 while (!list_empty(&info->caching_block_groups)) {
6913 caching_ctl = list_entry(info->caching_block_groups.next,
6914 struct btrfs_caching_control, list);
6915 list_del(&caching_ctl->list);
6916 put_caching_control(caching_ctl);
6918 up_write(&info->extent_commit_sem);
6920 spin_lock(&info->block_group_cache_lock);
6921 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6922 block_group = rb_entry(n, struct btrfs_block_group_cache,
6924 rb_erase(&block_group->cache_node,
6925 &info->block_group_cache_tree);
6926 spin_unlock(&info->block_group_cache_lock);
6928 down_write(&block_group->space_info->groups_sem);
6929 list_del(&block_group->list);
6930 up_write(&block_group->space_info->groups_sem);
6932 if (block_group->cached == BTRFS_CACHE_STARTED)
6933 wait_block_group_cache_done(block_group);
6936 * We haven't cached this block group, which means we could
6937 * possibly have excluded extents on this block group.
6939 if (block_group->cached == BTRFS_CACHE_NO)
6940 free_excluded_extents(info->extent_root, block_group);
6942 btrfs_remove_free_space_cache(block_group);
6943 btrfs_put_block_group(block_group);
6945 spin_lock(&info->block_group_cache_lock);
6947 spin_unlock(&info->block_group_cache_lock);
6949 /* now that all the block groups are freed, go through and
6950 * free all the space_info structs. This is only called during
6951 * the final stages of unmount, and so we know nobody is
6952 * using them. We call synchronize_rcu() once before we start,
6953 * just to be on the safe side.
6957 release_global_block_rsv(info);
6959 while(!list_empty(&info->space_info)) {
6960 space_info = list_entry(info->space_info.next,
6961 struct btrfs_space_info,
6963 if (space_info->bytes_pinned > 0 ||
6964 space_info->bytes_reserved > 0 ||
6965 space_info->bytes_may_use > 0) {
6967 dump_space_info(space_info, 0, 0);
6969 list_del(&space_info->list);
6975 static void __link_block_group(struct btrfs_space_info *space_info,
6976 struct btrfs_block_group_cache *cache)
6978 int index = get_block_group_index(cache);
6980 down_write(&space_info->groups_sem);
6981 list_add_tail(&cache->list, &space_info->block_groups[index]);
6982 up_write(&space_info->groups_sem);
6985 int btrfs_read_block_groups(struct btrfs_root *root)
6987 struct btrfs_path *path;
6989 struct btrfs_block_group_cache *cache;
6990 struct btrfs_fs_info *info = root->fs_info;
6991 struct btrfs_space_info *space_info;
6992 struct btrfs_key key;
6993 struct btrfs_key found_key;
6994 struct extent_buffer *leaf;
6998 root = info->extent_root;
7001 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7002 path = btrfs_alloc_path();
7007 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7008 if (cache_gen != 0 &&
7009 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7011 if (btrfs_test_opt(root, CLEAR_CACHE))
7013 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
7014 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
7017 ret = find_first_block_group(root, path, &key);
7022 leaf = path->nodes[0];
7023 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7024 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7029 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7031 if (!cache->free_space_ctl) {
7037 atomic_set(&cache->count, 1);
7038 spin_lock_init(&cache->lock);
7039 cache->fs_info = info;
7040 INIT_LIST_HEAD(&cache->list);
7041 INIT_LIST_HEAD(&cache->cluster_list);
7044 cache->disk_cache_state = BTRFS_DC_CLEAR;
7046 read_extent_buffer(leaf, &cache->item,
7047 btrfs_item_ptr_offset(leaf, path->slots[0]),
7048 sizeof(cache->item));
7049 memcpy(&cache->key, &found_key, sizeof(found_key));
7051 key.objectid = found_key.objectid + found_key.offset;
7052 btrfs_release_path(path);
7053 cache->flags = btrfs_block_group_flags(&cache->item);
7054 cache->sectorsize = root->sectorsize;
7056 btrfs_init_free_space_ctl(cache);
7059 * We need to exclude the super stripes now so that the space
7060 * info has super bytes accounted for, otherwise we'll think
7061 * we have more space than we actually do.
7063 exclude_super_stripes(root, cache);
7066 * check for two cases, either we are full, and therefore
7067 * don't need to bother with the caching work since we won't
7068 * find any space, or we are empty, and we can just add all
7069 * the space in and be done with it. This saves us _alot_ of
7070 * time, particularly in the full case.
7072 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7073 cache->last_byte_to_unpin = (u64)-1;
7074 cache->cached = BTRFS_CACHE_FINISHED;
7075 free_excluded_extents(root, cache);
7076 } else if (btrfs_block_group_used(&cache->item) == 0) {
7077 cache->last_byte_to_unpin = (u64)-1;
7078 cache->cached = BTRFS_CACHE_FINISHED;
7079 add_new_free_space(cache, root->fs_info,
7081 found_key.objectid +
7083 free_excluded_extents(root, cache);
7086 ret = update_space_info(info, cache->flags, found_key.offset,
7087 btrfs_block_group_used(&cache->item),
7090 cache->space_info = space_info;
7091 spin_lock(&cache->space_info->lock);
7092 cache->space_info->bytes_readonly += cache->bytes_super;
7093 spin_unlock(&cache->space_info->lock);
7095 __link_block_group(space_info, cache);
7097 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7100 set_avail_alloc_bits(root->fs_info, cache->flags);
7101 if (btrfs_chunk_readonly(root, cache->key.objectid))
7102 set_block_group_ro(cache, 1);
7105 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7106 if (!(get_alloc_profile(root, space_info->flags) &
7107 (BTRFS_BLOCK_GROUP_RAID10 |
7108 BTRFS_BLOCK_GROUP_RAID1 |
7109 BTRFS_BLOCK_GROUP_DUP)))
7112 * avoid allocating from un-mirrored block group if there are
7113 * mirrored block groups.
7115 list_for_each_entry(cache, &space_info->block_groups[3], list)
7116 set_block_group_ro(cache, 1);
7117 list_for_each_entry(cache, &space_info->block_groups[4], list)
7118 set_block_group_ro(cache, 1);
7121 init_global_block_rsv(info);
7124 btrfs_free_path(path);
7128 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7129 struct btrfs_root *root, u64 bytes_used,
7130 u64 type, u64 chunk_objectid, u64 chunk_offset,
7134 struct btrfs_root *extent_root;
7135 struct btrfs_block_group_cache *cache;
7137 extent_root = root->fs_info->extent_root;
7139 root->fs_info->last_trans_log_full_commit = trans->transid;
7141 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7144 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7146 if (!cache->free_space_ctl) {
7151 cache->key.objectid = chunk_offset;
7152 cache->key.offset = size;
7153 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7154 cache->sectorsize = root->sectorsize;
7155 cache->fs_info = root->fs_info;
7157 atomic_set(&cache->count, 1);
7158 spin_lock_init(&cache->lock);
7159 INIT_LIST_HEAD(&cache->list);
7160 INIT_LIST_HEAD(&cache->cluster_list);
7162 btrfs_init_free_space_ctl(cache);
7164 btrfs_set_block_group_used(&cache->item, bytes_used);
7165 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7166 cache->flags = type;
7167 btrfs_set_block_group_flags(&cache->item, type);
7169 cache->last_byte_to_unpin = (u64)-1;
7170 cache->cached = BTRFS_CACHE_FINISHED;
7171 exclude_super_stripes(root, cache);
7173 add_new_free_space(cache, root->fs_info, chunk_offset,
7174 chunk_offset + size);
7176 free_excluded_extents(root, cache);
7178 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7179 &cache->space_info);
7182 spin_lock(&cache->space_info->lock);
7183 cache->space_info->bytes_readonly += cache->bytes_super;
7184 spin_unlock(&cache->space_info->lock);
7186 __link_block_group(cache->space_info, cache);
7188 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7191 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7192 sizeof(cache->item));
7195 set_avail_alloc_bits(extent_root->fs_info, type);
7200 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7201 struct btrfs_root *root, u64 group_start)
7203 struct btrfs_path *path;
7204 struct btrfs_block_group_cache *block_group;
7205 struct btrfs_free_cluster *cluster;
7206 struct btrfs_root *tree_root = root->fs_info->tree_root;
7207 struct btrfs_key key;
7208 struct inode *inode;
7212 root = root->fs_info->extent_root;
7214 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7215 BUG_ON(!block_group);
7216 BUG_ON(!block_group->ro);
7219 * Free the reserved super bytes from this block group before
7222 free_excluded_extents(root, block_group);
7224 memcpy(&key, &block_group->key, sizeof(key));
7225 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7226 BTRFS_BLOCK_GROUP_RAID1 |
7227 BTRFS_BLOCK_GROUP_RAID10))
7232 /* make sure this block group isn't part of an allocation cluster */
7233 cluster = &root->fs_info->data_alloc_cluster;
7234 spin_lock(&cluster->refill_lock);
7235 btrfs_return_cluster_to_free_space(block_group, cluster);
7236 spin_unlock(&cluster->refill_lock);
7239 * make sure this block group isn't part of a metadata
7240 * allocation cluster
7242 cluster = &root->fs_info->meta_alloc_cluster;
7243 spin_lock(&cluster->refill_lock);
7244 btrfs_return_cluster_to_free_space(block_group, cluster);
7245 spin_unlock(&cluster->refill_lock);
7247 path = btrfs_alloc_path();
7253 inode = lookup_free_space_inode(root, block_group, path);
7254 if (!IS_ERR(inode)) {
7255 ret = btrfs_orphan_add(trans, inode);
7258 /* One for the block groups ref */
7259 spin_lock(&block_group->lock);
7260 if (block_group->iref) {
7261 block_group->iref = 0;
7262 block_group->inode = NULL;
7263 spin_unlock(&block_group->lock);
7266 spin_unlock(&block_group->lock);
7268 /* One for our lookup ref */
7272 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7273 key.offset = block_group->key.objectid;
7276 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7280 btrfs_release_path(path);
7282 ret = btrfs_del_item(trans, tree_root, path);
7285 btrfs_release_path(path);
7288 spin_lock(&root->fs_info->block_group_cache_lock);
7289 rb_erase(&block_group->cache_node,
7290 &root->fs_info->block_group_cache_tree);
7291 spin_unlock(&root->fs_info->block_group_cache_lock);
7293 down_write(&block_group->space_info->groups_sem);
7295 * we must use list_del_init so people can check to see if they
7296 * are still on the list after taking the semaphore
7298 list_del_init(&block_group->list);
7299 up_write(&block_group->space_info->groups_sem);
7301 if (block_group->cached == BTRFS_CACHE_STARTED)
7302 wait_block_group_cache_done(block_group);
7304 btrfs_remove_free_space_cache(block_group);
7306 spin_lock(&block_group->space_info->lock);
7307 block_group->space_info->total_bytes -= block_group->key.offset;
7308 block_group->space_info->bytes_readonly -= block_group->key.offset;
7309 block_group->space_info->disk_total -= block_group->key.offset * factor;
7310 spin_unlock(&block_group->space_info->lock);
7312 memcpy(&key, &block_group->key, sizeof(key));
7314 btrfs_clear_space_info_full(root->fs_info);
7316 btrfs_put_block_group(block_group);
7317 btrfs_put_block_group(block_group);
7319 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7325 ret = btrfs_del_item(trans, root, path);
7327 btrfs_free_path(path);
7331 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7333 struct btrfs_space_info *space_info;
7334 struct btrfs_super_block *disk_super;
7340 disk_super = &fs_info->super_copy;
7341 if (!btrfs_super_root(disk_super))
7344 features = btrfs_super_incompat_flags(disk_super);
7345 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7348 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7349 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7354 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7355 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7357 flags = BTRFS_BLOCK_GROUP_METADATA;
7358 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7362 flags = BTRFS_BLOCK_GROUP_DATA;
7363 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7369 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7371 return unpin_extent_range(root, start, end);
7374 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7375 u64 num_bytes, u64 *actual_bytes)
7377 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7380 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7382 struct btrfs_fs_info *fs_info = root->fs_info;
7383 struct btrfs_block_group_cache *cache = NULL;
7390 cache = btrfs_lookup_block_group(fs_info, range->start);
7393 if (cache->key.objectid >= (range->start + range->len)) {
7394 btrfs_put_block_group(cache);
7398 start = max(range->start, cache->key.objectid);
7399 end = min(range->start + range->len,
7400 cache->key.objectid + cache->key.offset);
7402 if (end - start >= range->minlen) {
7403 if (!block_group_cache_done(cache)) {
7404 ret = cache_block_group(cache, NULL, root, 0);
7406 wait_block_group_cache_done(cache);
7408 ret = btrfs_trim_block_group(cache,
7414 trimmed += group_trimmed;
7416 btrfs_put_block_group(cache);
7421 cache = next_block_group(fs_info->tree_root, cache);
7424 range->len = trimmed;