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 = NULL;
3570 if (root->ref_cows || root == root->fs_info->csum_root)
3571 block_rsv = trans->block_rsv;
3574 block_rsv = root->block_rsv;
3577 block_rsv = &root->fs_info->empty_block_rsv;
3582 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3586 spin_lock(&block_rsv->lock);
3587 if (block_rsv->reserved >= num_bytes) {
3588 block_rsv->reserved -= num_bytes;
3589 if (block_rsv->reserved < block_rsv->size)
3590 block_rsv->full = 0;
3593 spin_unlock(&block_rsv->lock);
3597 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3598 u64 num_bytes, int update_size)
3600 spin_lock(&block_rsv->lock);
3601 block_rsv->reserved += num_bytes;
3603 block_rsv->size += num_bytes;
3604 else if (block_rsv->reserved >= block_rsv->size)
3605 block_rsv->full = 1;
3606 spin_unlock(&block_rsv->lock);
3609 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3610 struct btrfs_block_rsv *dest, u64 num_bytes)
3612 struct btrfs_space_info *space_info = block_rsv->space_info;
3614 spin_lock(&block_rsv->lock);
3615 if (num_bytes == (u64)-1)
3616 num_bytes = block_rsv->size;
3617 block_rsv->size -= num_bytes;
3618 if (block_rsv->reserved >= block_rsv->size) {
3619 num_bytes = block_rsv->reserved - block_rsv->size;
3620 block_rsv->reserved = block_rsv->size;
3621 block_rsv->full = 1;
3625 spin_unlock(&block_rsv->lock);
3627 if (num_bytes > 0) {
3629 spin_lock(&dest->lock);
3633 bytes_to_add = dest->size - dest->reserved;
3634 bytes_to_add = min(num_bytes, bytes_to_add);
3635 dest->reserved += bytes_to_add;
3636 if (dest->reserved >= dest->size)
3638 num_bytes -= bytes_to_add;
3640 spin_unlock(&dest->lock);
3643 spin_lock(&space_info->lock);
3644 space_info->bytes_may_use -= num_bytes;
3645 space_info->reservation_progress++;
3646 spin_unlock(&space_info->lock);
3651 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3652 struct btrfs_block_rsv *dst, u64 num_bytes)
3656 ret = block_rsv_use_bytes(src, num_bytes);
3660 block_rsv_add_bytes(dst, num_bytes, 1);
3664 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3666 memset(rsv, 0, sizeof(*rsv));
3667 spin_lock_init(&rsv->lock);
3670 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3672 struct btrfs_block_rsv *block_rsv;
3673 struct btrfs_fs_info *fs_info = root->fs_info;
3675 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3679 btrfs_init_block_rsv(block_rsv);
3680 block_rsv->space_info = __find_space_info(fs_info,
3681 BTRFS_BLOCK_GROUP_METADATA);
3685 void btrfs_free_block_rsv(struct btrfs_root *root,
3686 struct btrfs_block_rsv *rsv)
3688 btrfs_block_rsv_release(root, rsv, (u64)-1);
3692 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3693 struct btrfs_root *root,
3694 struct btrfs_block_rsv *block_rsv,
3702 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3704 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3711 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3712 struct btrfs_root *root,
3713 struct btrfs_block_rsv *block_rsv,
3714 u64 min_reserved, int min_factor, int flush)
3722 spin_lock(&block_rsv->lock);
3724 num_bytes = div_factor(block_rsv->size, min_factor);
3725 if (min_reserved > num_bytes)
3726 num_bytes = min_reserved;
3728 if (block_rsv->reserved >= num_bytes)
3731 num_bytes -= block_rsv->reserved;
3732 spin_unlock(&block_rsv->lock);
3737 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, flush);
3739 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3746 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3747 struct btrfs_block_rsv *dst_rsv,
3750 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3753 void btrfs_block_rsv_release(struct btrfs_root *root,
3754 struct btrfs_block_rsv *block_rsv,
3757 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3758 if (global_rsv->full || global_rsv == block_rsv ||
3759 block_rsv->space_info != global_rsv->space_info)
3761 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3765 * helper to calculate size of global block reservation.
3766 * the desired value is sum of space used by extent tree,
3767 * checksum tree and root tree
3769 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3771 struct btrfs_space_info *sinfo;
3775 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3777 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3778 spin_lock(&sinfo->lock);
3779 data_used = sinfo->bytes_used;
3780 spin_unlock(&sinfo->lock);
3782 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3783 spin_lock(&sinfo->lock);
3784 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3786 meta_used = sinfo->bytes_used;
3787 spin_unlock(&sinfo->lock);
3789 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3791 num_bytes += div64_u64(data_used + meta_used, 50);
3793 if (num_bytes * 3 > meta_used)
3794 num_bytes = div64_u64(meta_used, 3);
3796 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3799 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3801 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3802 struct btrfs_space_info *sinfo = block_rsv->space_info;
3805 num_bytes = calc_global_metadata_size(fs_info);
3807 spin_lock(&block_rsv->lock);
3808 spin_lock(&sinfo->lock);
3810 block_rsv->size = num_bytes;
3812 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3813 sinfo->bytes_reserved + sinfo->bytes_readonly +
3814 sinfo->bytes_may_use;
3816 if (sinfo->total_bytes > num_bytes) {
3817 num_bytes = sinfo->total_bytes - num_bytes;
3818 block_rsv->reserved += num_bytes;
3819 sinfo->bytes_may_use += num_bytes;
3822 if (block_rsv->reserved >= block_rsv->size) {
3823 num_bytes = block_rsv->reserved - block_rsv->size;
3824 sinfo->bytes_may_use -= num_bytes;
3825 sinfo->reservation_progress++;
3826 block_rsv->reserved = block_rsv->size;
3827 block_rsv->full = 1;
3830 spin_unlock(&sinfo->lock);
3831 spin_unlock(&block_rsv->lock);
3834 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3836 struct btrfs_space_info *space_info;
3838 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3839 fs_info->chunk_block_rsv.space_info = space_info;
3841 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3842 fs_info->global_block_rsv.space_info = space_info;
3843 fs_info->delalloc_block_rsv.space_info = space_info;
3844 fs_info->trans_block_rsv.space_info = space_info;
3845 fs_info->empty_block_rsv.space_info = space_info;
3847 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3848 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3849 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3850 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3851 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3853 update_global_block_rsv(fs_info);
3856 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3858 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3859 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3860 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3861 WARN_ON(fs_info->trans_block_rsv.size > 0);
3862 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3863 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3864 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3867 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root)
3870 struct btrfs_block_rsv *block_rsv;
3872 if (!trans->bytes_reserved)
3875 block_rsv = &root->fs_info->trans_block_rsv;
3876 btrfs_block_rsv_release(root, block_rsv, trans->bytes_reserved);
3877 trans->bytes_reserved = 0;
3880 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3881 struct inode *inode)
3883 struct btrfs_root *root = BTRFS_I(inode)->root;
3884 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3885 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3888 * We need to hold space in order to delete our orphan item once we've
3889 * added it, so this takes the reservation so we can release it later
3890 * when we are truly done with the orphan item.
3892 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3893 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3896 void btrfs_orphan_release_metadata(struct inode *inode)
3898 struct btrfs_root *root = BTRFS_I(inode)->root;
3899 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3900 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3903 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3904 struct btrfs_pending_snapshot *pending)
3906 struct btrfs_root *root = pending->root;
3907 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3908 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3910 * two for root back/forward refs, two for directory entries
3911 * and one for root of the snapshot.
3913 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3914 dst_rsv->space_info = src_rsv->space_info;
3915 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3919 * drop_outstanding_extent - drop an outstanding extent
3920 * @inode: the inode we're dropping the extent for
3922 * This is called when we are freeing up an outstanding extent, either called
3923 * after an error or after an extent is written. This will return the number of
3924 * reserved extents that need to be freed. This must be called with
3925 * BTRFS_I(inode)->lock held.
3927 static unsigned drop_outstanding_extent(struct inode *inode)
3929 unsigned dropped_extents = 0;
3931 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3932 BTRFS_I(inode)->outstanding_extents--;
3935 * If we have more or the same amount of outsanding extents than we have
3936 * reserved then we need to leave the reserved extents count alone.
3938 if (BTRFS_I(inode)->outstanding_extents >=
3939 BTRFS_I(inode)->reserved_extents)
3942 dropped_extents = BTRFS_I(inode)->reserved_extents -
3943 BTRFS_I(inode)->outstanding_extents;
3944 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3945 return dropped_extents;
3949 * calc_csum_metadata_size - return the amount of metada space that must be
3950 * reserved/free'd for the given bytes.
3951 * @inode: the inode we're manipulating
3952 * @num_bytes: the number of bytes in question
3953 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3955 * This adjusts the number of csum_bytes in the inode and then returns the
3956 * correct amount of metadata that must either be reserved or freed. We
3957 * calculate how many checksums we can fit into one leaf and then divide the
3958 * number of bytes that will need to be checksumed by this value to figure out
3959 * how many checksums will be required. If we are adding bytes then the number
3960 * may go up and we will return the number of additional bytes that must be
3961 * reserved. If it is going down we will return the number of bytes that must
3964 * This must be called with BTRFS_I(inode)->lock held.
3966 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
3969 struct btrfs_root *root = BTRFS_I(inode)->root;
3971 int num_csums_per_leaf;
3975 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
3976 BTRFS_I(inode)->csum_bytes == 0)
3979 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
3981 BTRFS_I(inode)->csum_bytes += num_bytes;
3983 BTRFS_I(inode)->csum_bytes -= num_bytes;
3984 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
3985 num_csums_per_leaf = (int)div64_u64(csum_size,
3986 sizeof(struct btrfs_csum_item) +
3987 sizeof(struct btrfs_disk_key));
3988 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
3989 num_csums = num_csums + num_csums_per_leaf - 1;
3990 num_csums = num_csums / num_csums_per_leaf;
3992 old_csums = old_csums + num_csums_per_leaf - 1;
3993 old_csums = old_csums / num_csums_per_leaf;
3995 /* No change, no need to reserve more */
3996 if (old_csums == num_csums)
4000 return btrfs_calc_trans_metadata_size(root,
4001 num_csums - old_csums);
4003 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4006 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4008 struct btrfs_root *root = BTRFS_I(inode)->root;
4009 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4011 unsigned nr_extents = 0;
4015 if (btrfs_is_free_space_inode(root, inode))
4018 if (flush && btrfs_transaction_in_commit(root->fs_info))
4019 schedule_timeout(1);
4021 num_bytes = ALIGN(num_bytes, root->sectorsize);
4023 spin_lock(&BTRFS_I(inode)->lock);
4024 BTRFS_I(inode)->outstanding_extents++;
4026 if (BTRFS_I(inode)->outstanding_extents >
4027 BTRFS_I(inode)->reserved_extents) {
4028 nr_extents = BTRFS_I(inode)->outstanding_extents -
4029 BTRFS_I(inode)->reserved_extents;
4030 BTRFS_I(inode)->reserved_extents += nr_extents;
4032 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4034 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4035 spin_unlock(&BTRFS_I(inode)->lock);
4037 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, flush);
4042 spin_lock(&BTRFS_I(inode)->lock);
4043 dropped = drop_outstanding_extent(inode);
4044 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4045 spin_unlock(&BTRFS_I(inode)->lock);
4046 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4049 * Somebody could have come in and twiddled with the
4050 * reservation, so if we have to free more than we would have
4051 * reserved from this reservation go ahead and release those
4054 to_free -= to_reserve;
4056 btrfs_block_rsv_release(root, block_rsv, to_free);
4060 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4066 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4067 * @inode: the inode to release the reservation for
4068 * @num_bytes: the number of bytes we're releasing
4070 * This will release the metadata reservation for an inode. This can be called
4071 * once we complete IO for a given set of bytes to release their metadata
4074 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4076 struct btrfs_root *root = BTRFS_I(inode)->root;
4080 num_bytes = ALIGN(num_bytes, root->sectorsize);
4081 spin_lock(&BTRFS_I(inode)->lock);
4082 dropped = drop_outstanding_extent(inode);
4084 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4085 spin_unlock(&BTRFS_I(inode)->lock);
4087 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4089 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4094 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4095 * @inode: inode we're writing to
4096 * @num_bytes: the number of bytes we want to allocate
4098 * This will do the following things
4100 * o reserve space in the data space info for num_bytes
4101 * o reserve space in the metadata space info based on number of outstanding
4102 * extents and how much csums will be needed
4103 * o add to the inodes ->delalloc_bytes
4104 * o add it to the fs_info's delalloc inodes list.
4106 * This will return 0 for success and -ENOSPC if there is no space left.
4108 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4112 ret = btrfs_check_data_free_space(inode, num_bytes);
4116 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4118 btrfs_free_reserved_data_space(inode, num_bytes);
4126 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4127 * @inode: inode we're releasing space for
4128 * @num_bytes: the number of bytes we want to free up
4130 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4131 * called in the case that we don't need the metadata AND data reservations
4132 * anymore. So if there is an error or we insert an inline extent.
4134 * This function will release the metadata space that was not used and will
4135 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4136 * list if there are no delalloc bytes left.
4138 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4140 btrfs_delalloc_release_metadata(inode, num_bytes);
4141 btrfs_free_reserved_data_space(inode, num_bytes);
4144 static int update_block_group(struct btrfs_trans_handle *trans,
4145 struct btrfs_root *root,
4146 u64 bytenr, u64 num_bytes, int alloc)
4148 struct btrfs_block_group_cache *cache = NULL;
4149 struct btrfs_fs_info *info = root->fs_info;
4150 u64 total = num_bytes;
4155 /* block accounting for super block */
4156 spin_lock(&info->delalloc_lock);
4157 old_val = btrfs_super_bytes_used(&info->super_copy);
4159 old_val += num_bytes;
4161 old_val -= num_bytes;
4162 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4163 spin_unlock(&info->delalloc_lock);
4166 cache = btrfs_lookup_block_group(info, bytenr);
4169 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4170 BTRFS_BLOCK_GROUP_RAID1 |
4171 BTRFS_BLOCK_GROUP_RAID10))
4176 * If this block group has free space cache written out, we
4177 * need to make sure to load it if we are removing space. This
4178 * is because we need the unpinning stage to actually add the
4179 * space back to the block group, otherwise we will leak space.
4181 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4182 cache_block_group(cache, trans, NULL, 1);
4184 byte_in_group = bytenr - cache->key.objectid;
4185 WARN_ON(byte_in_group > cache->key.offset);
4187 spin_lock(&cache->space_info->lock);
4188 spin_lock(&cache->lock);
4190 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4191 cache->disk_cache_state < BTRFS_DC_CLEAR)
4192 cache->disk_cache_state = BTRFS_DC_CLEAR;
4195 old_val = btrfs_block_group_used(&cache->item);
4196 num_bytes = min(total, cache->key.offset - byte_in_group);
4198 old_val += num_bytes;
4199 btrfs_set_block_group_used(&cache->item, old_val);
4200 cache->reserved -= num_bytes;
4201 cache->space_info->bytes_reserved -= num_bytes;
4202 cache->space_info->bytes_used += num_bytes;
4203 cache->space_info->disk_used += num_bytes * factor;
4204 spin_unlock(&cache->lock);
4205 spin_unlock(&cache->space_info->lock);
4207 old_val -= num_bytes;
4208 btrfs_set_block_group_used(&cache->item, old_val);
4209 cache->pinned += num_bytes;
4210 cache->space_info->bytes_pinned += num_bytes;
4211 cache->space_info->bytes_used -= num_bytes;
4212 cache->space_info->disk_used -= num_bytes * factor;
4213 spin_unlock(&cache->lock);
4214 spin_unlock(&cache->space_info->lock);
4216 set_extent_dirty(info->pinned_extents,
4217 bytenr, bytenr + num_bytes - 1,
4218 GFP_NOFS | __GFP_NOFAIL);
4220 btrfs_put_block_group(cache);
4222 bytenr += num_bytes;
4227 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4229 struct btrfs_block_group_cache *cache;
4232 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4236 bytenr = cache->key.objectid;
4237 btrfs_put_block_group(cache);
4242 static int pin_down_extent(struct btrfs_root *root,
4243 struct btrfs_block_group_cache *cache,
4244 u64 bytenr, u64 num_bytes, int reserved)
4246 spin_lock(&cache->space_info->lock);
4247 spin_lock(&cache->lock);
4248 cache->pinned += num_bytes;
4249 cache->space_info->bytes_pinned += num_bytes;
4251 cache->reserved -= num_bytes;
4252 cache->space_info->bytes_reserved -= num_bytes;
4254 spin_unlock(&cache->lock);
4255 spin_unlock(&cache->space_info->lock);
4257 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4258 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4263 * this function must be called within transaction
4265 int btrfs_pin_extent(struct btrfs_root *root,
4266 u64 bytenr, u64 num_bytes, int reserved)
4268 struct btrfs_block_group_cache *cache;
4270 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4273 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4275 btrfs_put_block_group(cache);
4280 * btrfs_update_reserved_bytes - update the block_group and space info counters
4281 * @cache: The cache we are manipulating
4282 * @num_bytes: The number of bytes in question
4283 * @reserve: One of the reservation enums
4285 * This is called by the allocator when it reserves space, or by somebody who is
4286 * freeing space that was never actually used on disk. For example if you
4287 * reserve some space for a new leaf in transaction A and before transaction A
4288 * commits you free that leaf, you call this with reserve set to 0 in order to
4289 * clear the reservation.
4291 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4292 * ENOSPC accounting. For data we handle the reservation through clearing the
4293 * delalloc bits in the io_tree. We have to do this since we could end up
4294 * allocating less disk space for the amount of data we have reserved in the
4295 * case of compression.
4297 * If this is a reservation and the block group has become read only we cannot
4298 * make the reservation and return -EAGAIN, otherwise this function always
4301 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4302 u64 num_bytes, int reserve)
4304 struct btrfs_space_info *space_info = cache->space_info;
4306 spin_lock(&space_info->lock);
4307 spin_lock(&cache->lock);
4308 if (reserve != RESERVE_FREE) {
4312 cache->reserved += num_bytes;
4313 space_info->bytes_reserved += num_bytes;
4314 if (reserve == RESERVE_ALLOC) {
4315 BUG_ON(space_info->bytes_may_use < num_bytes);
4316 space_info->bytes_may_use -= num_bytes;
4321 space_info->bytes_readonly += num_bytes;
4322 cache->reserved -= num_bytes;
4323 space_info->bytes_reserved -= num_bytes;
4324 space_info->reservation_progress++;
4326 spin_unlock(&cache->lock);
4327 spin_unlock(&space_info->lock);
4331 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4332 struct btrfs_root *root)
4334 struct btrfs_fs_info *fs_info = root->fs_info;
4335 struct btrfs_caching_control *next;
4336 struct btrfs_caching_control *caching_ctl;
4337 struct btrfs_block_group_cache *cache;
4339 down_write(&fs_info->extent_commit_sem);
4341 list_for_each_entry_safe(caching_ctl, next,
4342 &fs_info->caching_block_groups, list) {
4343 cache = caching_ctl->block_group;
4344 if (block_group_cache_done(cache)) {
4345 cache->last_byte_to_unpin = (u64)-1;
4346 list_del_init(&caching_ctl->list);
4347 put_caching_control(caching_ctl);
4349 cache->last_byte_to_unpin = caching_ctl->progress;
4353 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4354 fs_info->pinned_extents = &fs_info->freed_extents[1];
4356 fs_info->pinned_extents = &fs_info->freed_extents[0];
4358 up_write(&fs_info->extent_commit_sem);
4360 update_global_block_rsv(fs_info);
4364 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4366 struct btrfs_fs_info *fs_info = root->fs_info;
4367 struct btrfs_block_group_cache *cache = NULL;
4370 while (start <= end) {
4372 start >= cache->key.objectid + cache->key.offset) {
4374 btrfs_put_block_group(cache);
4375 cache = btrfs_lookup_block_group(fs_info, start);
4379 len = cache->key.objectid + cache->key.offset - start;
4380 len = min(len, end + 1 - start);
4382 if (start < cache->last_byte_to_unpin) {
4383 len = min(len, cache->last_byte_to_unpin - start);
4384 btrfs_add_free_space(cache, start, len);
4389 spin_lock(&cache->space_info->lock);
4390 spin_lock(&cache->lock);
4391 cache->pinned -= len;
4392 cache->space_info->bytes_pinned -= len;
4394 cache->space_info->bytes_readonly += len;
4395 spin_unlock(&cache->lock);
4396 spin_unlock(&cache->space_info->lock);
4400 btrfs_put_block_group(cache);
4404 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4405 struct btrfs_root *root)
4407 struct btrfs_fs_info *fs_info = root->fs_info;
4408 struct extent_io_tree *unpin;
4413 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4414 unpin = &fs_info->freed_extents[1];
4416 unpin = &fs_info->freed_extents[0];
4419 ret = find_first_extent_bit(unpin, 0, &start, &end,
4424 if (btrfs_test_opt(root, DISCARD))
4425 ret = btrfs_discard_extent(root, start,
4426 end + 1 - start, NULL);
4428 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4429 unpin_extent_range(root, start, end);
4436 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4437 struct btrfs_root *root,
4438 u64 bytenr, u64 num_bytes, u64 parent,
4439 u64 root_objectid, u64 owner_objectid,
4440 u64 owner_offset, int refs_to_drop,
4441 struct btrfs_delayed_extent_op *extent_op)
4443 struct btrfs_key key;
4444 struct btrfs_path *path;
4445 struct btrfs_fs_info *info = root->fs_info;
4446 struct btrfs_root *extent_root = info->extent_root;
4447 struct extent_buffer *leaf;
4448 struct btrfs_extent_item *ei;
4449 struct btrfs_extent_inline_ref *iref;
4452 int extent_slot = 0;
4453 int found_extent = 0;
4458 path = btrfs_alloc_path();
4463 path->leave_spinning = 1;
4465 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4466 BUG_ON(!is_data && refs_to_drop != 1);
4468 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4469 bytenr, num_bytes, parent,
4470 root_objectid, owner_objectid,
4473 extent_slot = path->slots[0];
4474 while (extent_slot >= 0) {
4475 btrfs_item_key_to_cpu(path->nodes[0], &key,
4477 if (key.objectid != bytenr)
4479 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4480 key.offset == num_bytes) {
4484 if (path->slots[0] - extent_slot > 5)
4488 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4489 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4490 if (found_extent && item_size < sizeof(*ei))
4493 if (!found_extent) {
4495 ret = remove_extent_backref(trans, extent_root, path,
4499 btrfs_release_path(path);
4500 path->leave_spinning = 1;
4502 key.objectid = bytenr;
4503 key.type = BTRFS_EXTENT_ITEM_KEY;
4504 key.offset = num_bytes;
4506 ret = btrfs_search_slot(trans, extent_root,
4509 printk(KERN_ERR "umm, got %d back from search"
4510 ", was looking for %llu\n", ret,
4511 (unsigned long long)bytenr);
4513 btrfs_print_leaf(extent_root,
4517 extent_slot = path->slots[0];
4520 btrfs_print_leaf(extent_root, path->nodes[0]);
4522 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4523 "parent %llu root %llu owner %llu offset %llu\n",
4524 (unsigned long long)bytenr,
4525 (unsigned long long)parent,
4526 (unsigned long long)root_objectid,
4527 (unsigned long long)owner_objectid,
4528 (unsigned long long)owner_offset);
4531 leaf = path->nodes[0];
4532 item_size = btrfs_item_size_nr(leaf, extent_slot);
4533 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4534 if (item_size < sizeof(*ei)) {
4535 BUG_ON(found_extent || extent_slot != path->slots[0]);
4536 ret = convert_extent_item_v0(trans, extent_root, path,
4540 btrfs_release_path(path);
4541 path->leave_spinning = 1;
4543 key.objectid = bytenr;
4544 key.type = BTRFS_EXTENT_ITEM_KEY;
4545 key.offset = num_bytes;
4547 ret = btrfs_search_slot(trans, extent_root, &key, path,
4550 printk(KERN_ERR "umm, got %d back from search"
4551 ", was looking for %llu\n", ret,
4552 (unsigned long long)bytenr);
4553 btrfs_print_leaf(extent_root, path->nodes[0]);
4556 extent_slot = path->slots[0];
4557 leaf = path->nodes[0];
4558 item_size = btrfs_item_size_nr(leaf, extent_slot);
4561 BUG_ON(item_size < sizeof(*ei));
4562 ei = btrfs_item_ptr(leaf, extent_slot,
4563 struct btrfs_extent_item);
4564 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4565 struct btrfs_tree_block_info *bi;
4566 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4567 bi = (struct btrfs_tree_block_info *)(ei + 1);
4568 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4571 refs = btrfs_extent_refs(leaf, ei);
4572 BUG_ON(refs < refs_to_drop);
4573 refs -= refs_to_drop;
4577 __run_delayed_extent_op(extent_op, leaf, ei);
4579 * In the case of inline back ref, reference count will
4580 * be updated by remove_extent_backref
4583 BUG_ON(!found_extent);
4585 btrfs_set_extent_refs(leaf, ei, refs);
4586 btrfs_mark_buffer_dirty(leaf);
4589 ret = remove_extent_backref(trans, extent_root, path,
4596 BUG_ON(is_data && refs_to_drop !=
4597 extent_data_ref_count(root, path, iref));
4599 BUG_ON(path->slots[0] != extent_slot);
4601 BUG_ON(path->slots[0] != extent_slot + 1);
4602 path->slots[0] = extent_slot;
4607 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4610 btrfs_release_path(path);
4613 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4616 invalidate_mapping_pages(info->btree_inode->i_mapping,
4617 bytenr >> PAGE_CACHE_SHIFT,
4618 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4621 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4624 btrfs_free_path(path);
4629 * when we free an block, it is possible (and likely) that we free the last
4630 * delayed ref for that extent as well. This searches the delayed ref tree for
4631 * a given extent, and if there are no other delayed refs to be processed, it
4632 * removes it from the tree.
4634 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4635 struct btrfs_root *root, u64 bytenr)
4637 struct btrfs_delayed_ref_head *head;
4638 struct btrfs_delayed_ref_root *delayed_refs;
4639 struct btrfs_delayed_ref_node *ref;
4640 struct rb_node *node;
4643 delayed_refs = &trans->transaction->delayed_refs;
4644 spin_lock(&delayed_refs->lock);
4645 head = btrfs_find_delayed_ref_head(trans, bytenr);
4649 node = rb_prev(&head->node.rb_node);
4653 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4655 /* there are still entries for this ref, we can't drop it */
4656 if (ref->bytenr == bytenr)
4659 if (head->extent_op) {
4660 if (!head->must_insert_reserved)
4662 kfree(head->extent_op);
4663 head->extent_op = NULL;
4667 * waiting for the lock here would deadlock. If someone else has it
4668 * locked they are already in the process of dropping it anyway
4670 if (!mutex_trylock(&head->mutex))
4674 * at this point we have a head with no other entries. Go
4675 * ahead and process it.
4677 head->node.in_tree = 0;
4678 rb_erase(&head->node.rb_node, &delayed_refs->root);
4680 delayed_refs->num_entries--;
4683 * we don't take a ref on the node because we're removing it from the
4684 * tree, so we just steal the ref the tree was holding.
4686 delayed_refs->num_heads--;
4687 if (list_empty(&head->cluster))
4688 delayed_refs->num_heads_ready--;
4690 list_del_init(&head->cluster);
4691 spin_unlock(&delayed_refs->lock);
4693 BUG_ON(head->extent_op);
4694 if (head->must_insert_reserved)
4697 mutex_unlock(&head->mutex);
4698 btrfs_put_delayed_ref(&head->node);
4701 spin_unlock(&delayed_refs->lock);
4705 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4706 struct btrfs_root *root,
4707 struct extent_buffer *buf,
4708 u64 parent, int last_ref)
4710 struct btrfs_block_group_cache *cache = NULL;
4713 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4714 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4715 parent, root->root_key.objectid,
4716 btrfs_header_level(buf),
4717 BTRFS_DROP_DELAYED_REF, NULL);
4724 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4726 if (btrfs_header_generation(buf) == trans->transid) {
4727 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4728 ret = check_ref_cleanup(trans, root, buf->start);
4733 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4734 pin_down_extent(root, cache, buf->start, buf->len, 1);
4738 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4740 btrfs_add_free_space(cache, buf->start, buf->len);
4741 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4745 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4748 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4749 btrfs_put_block_group(cache);
4752 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4753 struct btrfs_root *root,
4754 u64 bytenr, u64 num_bytes, u64 parent,
4755 u64 root_objectid, u64 owner, u64 offset)
4760 * tree log blocks never actually go into the extent allocation
4761 * tree, just update pinning info and exit early.
4763 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4764 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4765 /* unlocks the pinned mutex */
4766 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4768 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4769 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4770 parent, root_objectid, (int)owner,
4771 BTRFS_DROP_DELAYED_REF, NULL);
4774 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4775 parent, root_objectid, owner,
4776 offset, BTRFS_DROP_DELAYED_REF, NULL);
4782 static u64 stripe_align(struct btrfs_root *root, u64 val)
4784 u64 mask = ((u64)root->stripesize - 1);
4785 u64 ret = (val + mask) & ~mask;
4790 * when we wait for progress in the block group caching, its because
4791 * our allocation attempt failed at least once. So, we must sleep
4792 * and let some progress happen before we try again.
4794 * This function will sleep at least once waiting for new free space to
4795 * show up, and then it will check the block group free space numbers
4796 * for our min num_bytes. Another option is to have it go ahead
4797 * and look in the rbtree for a free extent of a given size, but this
4801 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4804 struct btrfs_caching_control *caching_ctl;
4807 caching_ctl = get_caching_control(cache);
4811 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4812 (cache->free_space_ctl->free_space >= num_bytes));
4814 put_caching_control(caching_ctl);
4819 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4821 struct btrfs_caching_control *caching_ctl;
4824 caching_ctl = get_caching_control(cache);
4828 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4830 put_caching_control(caching_ctl);
4834 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4837 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4839 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4841 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4843 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4850 enum btrfs_loop_type {
4851 LOOP_FIND_IDEAL = 0,
4852 LOOP_CACHING_NOWAIT = 1,
4853 LOOP_CACHING_WAIT = 2,
4854 LOOP_ALLOC_CHUNK = 3,
4855 LOOP_NO_EMPTY_SIZE = 4,
4859 * walks the btree of allocated extents and find a hole of a given size.
4860 * The key ins is changed to record the hole:
4861 * ins->objectid == block start
4862 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4863 * ins->offset == number of blocks
4864 * Any available blocks before search_start are skipped.
4866 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4867 struct btrfs_root *orig_root,
4868 u64 num_bytes, u64 empty_size,
4869 u64 search_start, u64 search_end,
4870 u64 hint_byte, struct btrfs_key *ins,
4874 struct btrfs_root *root = orig_root->fs_info->extent_root;
4875 struct btrfs_free_cluster *last_ptr = NULL;
4876 struct btrfs_block_group_cache *block_group = NULL;
4877 int empty_cluster = 2 * 1024 * 1024;
4878 int allowed_chunk_alloc = 0;
4879 int done_chunk_alloc = 0;
4880 struct btrfs_space_info *space_info;
4881 int last_ptr_loop = 0;
4884 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
4885 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
4886 bool found_uncached_bg = false;
4887 bool failed_cluster_refill = false;
4888 bool failed_alloc = false;
4889 bool use_cluster = true;
4890 u64 ideal_cache_percent = 0;
4891 u64 ideal_cache_offset = 0;
4893 WARN_ON(num_bytes < root->sectorsize);
4894 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4898 space_info = __find_space_info(root->fs_info, data);
4900 printk(KERN_ERR "No space info for %llu\n", data);
4905 * If the space info is for both data and metadata it means we have a
4906 * small filesystem and we can't use the clustering stuff.
4908 if (btrfs_mixed_space_info(space_info))
4909 use_cluster = false;
4911 if (orig_root->ref_cows || empty_size)
4912 allowed_chunk_alloc = 1;
4914 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4915 last_ptr = &root->fs_info->meta_alloc_cluster;
4916 if (!btrfs_test_opt(root, SSD))
4917 empty_cluster = 64 * 1024;
4920 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4921 btrfs_test_opt(root, SSD)) {
4922 last_ptr = &root->fs_info->data_alloc_cluster;
4926 spin_lock(&last_ptr->lock);
4927 if (last_ptr->block_group)
4928 hint_byte = last_ptr->window_start;
4929 spin_unlock(&last_ptr->lock);
4932 search_start = max(search_start, first_logical_byte(root, 0));
4933 search_start = max(search_start, hint_byte);
4938 if (search_start == hint_byte) {
4940 block_group = btrfs_lookup_block_group(root->fs_info,
4943 * we don't want to use the block group if it doesn't match our
4944 * allocation bits, or if its not cached.
4946 * However if we are re-searching with an ideal block group
4947 * picked out then we don't care that the block group is cached.
4949 if (block_group && block_group_bits(block_group, data) &&
4950 (block_group->cached != BTRFS_CACHE_NO ||
4951 search_start == ideal_cache_offset)) {
4952 down_read(&space_info->groups_sem);
4953 if (list_empty(&block_group->list) ||
4956 * someone is removing this block group,
4957 * we can't jump into the have_block_group
4958 * target because our list pointers are not
4961 btrfs_put_block_group(block_group);
4962 up_read(&space_info->groups_sem);
4964 index = get_block_group_index(block_group);
4965 goto have_block_group;
4967 } else if (block_group) {
4968 btrfs_put_block_group(block_group);
4972 down_read(&space_info->groups_sem);
4973 list_for_each_entry(block_group, &space_info->block_groups[index],
4978 btrfs_get_block_group(block_group);
4979 search_start = block_group->key.objectid;
4982 * this can happen if we end up cycling through all the
4983 * raid types, but we want to make sure we only allocate
4984 * for the proper type.
4986 if (!block_group_bits(block_group, data)) {
4987 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4988 BTRFS_BLOCK_GROUP_RAID1 |
4989 BTRFS_BLOCK_GROUP_RAID10;
4992 * if they asked for extra copies and this block group
4993 * doesn't provide them, bail. This does allow us to
4994 * fill raid0 from raid1.
4996 if ((data & extra) && !(block_group->flags & extra))
5001 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5004 ret = cache_block_group(block_group, trans,
5006 if (block_group->cached == BTRFS_CACHE_FINISHED)
5007 goto have_block_group;
5009 free_percent = btrfs_block_group_used(&block_group->item);
5010 free_percent *= 100;
5011 free_percent = div64_u64(free_percent,
5012 block_group->key.offset);
5013 free_percent = 100 - free_percent;
5014 if (free_percent > ideal_cache_percent &&
5015 likely(!block_group->ro)) {
5016 ideal_cache_offset = block_group->key.objectid;
5017 ideal_cache_percent = free_percent;
5021 * The caching workers are limited to 2 threads, so we
5022 * can queue as much work as we care to.
5024 if (loop > LOOP_FIND_IDEAL) {
5025 ret = cache_block_group(block_group, trans,
5029 found_uncached_bg = true;
5032 * If loop is set for cached only, try the next block
5035 if (loop == LOOP_FIND_IDEAL)
5039 cached = block_group_cache_done(block_group);
5040 if (unlikely(!cached))
5041 found_uncached_bg = true;
5043 if (unlikely(block_group->ro))
5046 spin_lock(&block_group->free_space_ctl->tree_lock);
5048 block_group->free_space_ctl->free_space <
5049 num_bytes + empty_size) {
5050 spin_unlock(&block_group->free_space_ctl->tree_lock);
5053 spin_unlock(&block_group->free_space_ctl->tree_lock);
5056 * Ok we want to try and use the cluster allocator, so lets look
5057 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5058 * have tried the cluster allocator plenty of times at this
5059 * point and not have found anything, so we are likely way too
5060 * fragmented for the clustering stuff to find anything, so lets
5061 * just skip it and let the allocator find whatever block it can
5064 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5066 * the refill lock keeps out other
5067 * people trying to start a new cluster
5069 spin_lock(&last_ptr->refill_lock);
5070 if (last_ptr->block_group &&
5071 (last_ptr->block_group->ro ||
5072 !block_group_bits(last_ptr->block_group, data))) {
5074 goto refill_cluster;
5077 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5078 num_bytes, search_start);
5080 /* we have a block, we're done */
5081 spin_unlock(&last_ptr->refill_lock);
5085 spin_lock(&last_ptr->lock);
5087 * whoops, this cluster doesn't actually point to
5088 * this block group. Get a ref on the block
5089 * group is does point to and try again
5091 if (!last_ptr_loop && last_ptr->block_group &&
5092 last_ptr->block_group != block_group &&
5094 get_block_group_index(last_ptr->block_group)) {
5096 btrfs_put_block_group(block_group);
5097 block_group = last_ptr->block_group;
5098 btrfs_get_block_group(block_group);
5099 spin_unlock(&last_ptr->lock);
5100 spin_unlock(&last_ptr->refill_lock);
5103 search_start = block_group->key.objectid;
5105 * we know this block group is properly
5106 * in the list because
5107 * btrfs_remove_block_group, drops the
5108 * cluster before it removes the block
5109 * group from the list
5111 goto have_block_group;
5113 spin_unlock(&last_ptr->lock);
5116 * this cluster didn't work out, free it and
5119 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5123 /* allocate a cluster in this block group */
5124 ret = btrfs_find_space_cluster(trans, root,
5125 block_group, last_ptr,
5127 empty_cluster + empty_size);
5130 * now pull our allocation out of this
5133 offset = btrfs_alloc_from_cluster(block_group,
5134 last_ptr, num_bytes,
5137 /* we found one, proceed */
5138 spin_unlock(&last_ptr->refill_lock);
5141 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5142 && !failed_cluster_refill) {
5143 spin_unlock(&last_ptr->refill_lock);
5145 failed_cluster_refill = true;
5146 wait_block_group_cache_progress(block_group,
5147 num_bytes + empty_cluster + empty_size);
5148 goto have_block_group;
5152 * at this point we either didn't find a cluster
5153 * or we weren't able to allocate a block from our
5154 * cluster. Free the cluster we've been trying
5155 * to use, and go to the next block group
5157 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5158 spin_unlock(&last_ptr->refill_lock);
5162 offset = btrfs_find_space_for_alloc(block_group, search_start,
5163 num_bytes, empty_size);
5165 * If we didn't find a chunk, and we haven't failed on this
5166 * block group before, and this block group is in the middle of
5167 * caching and we are ok with waiting, then go ahead and wait
5168 * for progress to be made, and set failed_alloc to true.
5170 * If failed_alloc is true then we've already waited on this
5171 * block group once and should move on to the next block group.
5173 if (!offset && !failed_alloc && !cached &&
5174 loop > LOOP_CACHING_NOWAIT) {
5175 wait_block_group_cache_progress(block_group,
5176 num_bytes + empty_size);
5177 failed_alloc = true;
5178 goto have_block_group;
5179 } else if (!offset) {
5183 search_start = stripe_align(root, offset);
5184 /* move on to the next group */
5185 if (search_start + num_bytes >= search_end) {
5186 btrfs_add_free_space(block_group, offset, num_bytes);
5190 /* move on to the next group */
5191 if (search_start + num_bytes >
5192 block_group->key.objectid + block_group->key.offset) {
5193 btrfs_add_free_space(block_group, offset, num_bytes);
5197 ins->objectid = search_start;
5198 ins->offset = num_bytes;
5200 if (offset < search_start)
5201 btrfs_add_free_space(block_group, offset,
5202 search_start - offset);
5203 BUG_ON(offset > search_start);
5205 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5207 if (ret == -EAGAIN) {
5208 btrfs_add_free_space(block_group, offset, num_bytes);
5212 /* we are all good, lets return */
5213 ins->objectid = search_start;
5214 ins->offset = num_bytes;
5216 if (offset < search_start)
5217 btrfs_add_free_space(block_group, offset,
5218 search_start - offset);
5219 BUG_ON(offset > search_start);
5220 btrfs_put_block_group(block_group);
5223 failed_cluster_refill = false;
5224 failed_alloc = false;
5225 BUG_ON(index != get_block_group_index(block_group));
5226 btrfs_put_block_group(block_group);
5228 up_read(&space_info->groups_sem);
5230 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5233 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5234 * for them to make caching progress. Also
5235 * determine the best possible bg to cache
5236 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5237 * caching kthreads as we move along
5238 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5239 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5240 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5243 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5245 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5246 found_uncached_bg = false;
5248 if (!ideal_cache_percent)
5252 * 1 of the following 2 things have happened so far
5254 * 1) We found an ideal block group for caching that
5255 * is mostly full and will cache quickly, so we might
5256 * as well wait for it.
5258 * 2) We searched for cached only and we didn't find
5259 * anything, and we didn't start any caching kthreads
5260 * either, so chances are we will loop through and
5261 * start a couple caching kthreads, and then come back
5262 * around and just wait for them. This will be slower
5263 * because we will have 2 caching kthreads reading at
5264 * the same time when we could have just started one
5265 * and waited for it to get far enough to give us an
5266 * allocation, so go ahead and go to the wait caching
5269 loop = LOOP_CACHING_WAIT;
5270 search_start = ideal_cache_offset;
5271 ideal_cache_percent = 0;
5273 } else if (loop == LOOP_FIND_IDEAL) {
5275 * Didn't find a uncached bg, wait on anything we find
5278 loop = LOOP_CACHING_WAIT;
5284 if (loop == LOOP_ALLOC_CHUNK) {
5285 if (allowed_chunk_alloc) {
5286 ret = do_chunk_alloc(trans, root, num_bytes +
5287 2 * 1024 * 1024, data,
5288 CHUNK_ALLOC_LIMITED);
5289 allowed_chunk_alloc = 0;
5291 done_chunk_alloc = 1;
5292 } else if (!done_chunk_alloc &&
5293 space_info->force_alloc ==
5294 CHUNK_ALLOC_NO_FORCE) {
5295 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5299 * We didn't allocate a chunk, go ahead and drop the
5300 * empty size and loop again.
5302 if (!done_chunk_alloc)
5303 loop = LOOP_NO_EMPTY_SIZE;
5306 if (loop == LOOP_NO_EMPTY_SIZE) {
5312 } else if (!ins->objectid) {
5314 } else if (ins->objectid) {
5321 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5322 int dump_block_groups)
5324 struct btrfs_block_group_cache *cache;
5327 spin_lock(&info->lock);
5328 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5329 (unsigned long long)info->flags,
5330 (unsigned long long)(info->total_bytes - info->bytes_used -
5331 info->bytes_pinned - info->bytes_reserved -
5332 info->bytes_readonly),
5333 (info->full) ? "" : "not ");
5334 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5335 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5336 (unsigned long long)info->total_bytes,
5337 (unsigned long long)info->bytes_used,
5338 (unsigned long long)info->bytes_pinned,
5339 (unsigned long long)info->bytes_reserved,
5340 (unsigned long long)info->bytes_may_use,
5341 (unsigned long long)info->bytes_readonly);
5342 spin_unlock(&info->lock);
5344 if (!dump_block_groups)
5347 down_read(&info->groups_sem);
5349 list_for_each_entry(cache, &info->block_groups[index], list) {
5350 spin_lock(&cache->lock);
5351 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5352 "%llu pinned %llu reserved\n",
5353 (unsigned long long)cache->key.objectid,
5354 (unsigned long long)cache->key.offset,
5355 (unsigned long long)btrfs_block_group_used(&cache->item),
5356 (unsigned long long)cache->pinned,
5357 (unsigned long long)cache->reserved);
5358 btrfs_dump_free_space(cache, bytes);
5359 spin_unlock(&cache->lock);
5361 if (++index < BTRFS_NR_RAID_TYPES)
5363 up_read(&info->groups_sem);
5366 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5367 struct btrfs_root *root,
5368 u64 num_bytes, u64 min_alloc_size,
5369 u64 empty_size, u64 hint_byte,
5370 u64 search_end, struct btrfs_key *ins,
5374 u64 search_start = 0;
5376 data = btrfs_get_alloc_profile(root, data);
5379 * the only place that sets empty_size is btrfs_realloc_node, which
5380 * is not called recursively on allocations
5382 if (empty_size || root->ref_cows)
5383 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5384 num_bytes + 2 * 1024 * 1024, data,
5385 CHUNK_ALLOC_NO_FORCE);
5387 WARN_ON(num_bytes < root->sectorsize);
5388 ret = find_free_extent(trans, root, num_bytes, empty_size,
5389 search_start, search_end, hint_byte,
5392 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5393 num_bytes = num_bytes >> 1;
5394 num_bytes = num_bytes & ~(root->sectorsize - 1);
5395 num_bytes = max(num_bytes, min_alloc_size);
5396 do_chunk_alloc(trans, root->fs_info->extent_root,
5397 num_bytes, data, CHUNK_ALLOC_FORCE);
5400 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5401 struct btrfs_space_info *sinfo;
5403 sinfo = __find_space_info(root->fs_info, data);
5404 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5405 "wanted %llu\n", (unsigned long long)data,
5406 (unsigned long long)num_bytes);
5407 dump_space_info(sinfo, num_bytes, 1);
5410 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5415 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5417 struct btrfs_block_group_cache *cache;
5420 cache = btrfs_lookup_block_group(root->fs_info, start);
5422 printk(KERN_ERR "Unable to find block group for %llu\n",
5423 (unsigned long long)start);
5427 if (btrfs_test_opt(root, DISCARD))
5428 ret = btrfs_discard_extent(root, start, len, NULL);
5430 btrfs_add_free_space(cache, start, len);
5431 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5432 btrfs_put_block_group(cache);
5434 trace_btrfs_reserved_extent_free(root, start, len);
5439 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5440 struct btrfs_root *root,
5441 u64 parent, u64 root_objectid,
5442 u64 flags, u64 owner, u64 offset,
5443 struct btrfs_key *ins, int ref_mod)
5446 struct btrfs_fs_info *fs_info = root->fs_info;
5447 struct btrfs_extent_item *extent_item;
5448 struct btrfs_extent_inline_ref *iref;
5449 struct btrfs_path *path;
5450 struct extent_buffer *leaf;
5455 type = BTRFS_SHARED_DATA_REF_KEY;
5457 type = BTRFS_EXTENT_DATA_REF_KEY;
5459 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5461 path = btrfs_alloc_path();
5465 path->leave_spinning = 1;
5466 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5470 leaf = path->nodes[0];
5471 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5472 struct btrfs_extent_item);
5473 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5474 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5475 btrfs_set_extent_flags(leaf, extent_item,
5476 flags | BTRFS_EXTENT_FLAG_DATA);
5478 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5479 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5481 struct btrfs_shared_data_ref *ref;
5482 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5483 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5484 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5486 struct btrfs_extent_data_ref *ref;
5487 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5488 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5489 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5490 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5491 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5494 btrfs_mark_buffer_dirty(path->nodes[0]);
5495 btrfs_free_path(path);
5497 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5499 printk(KERN_ERR "btrfs update block group failed for %llu "
5500 "%llu\n", (unsigned long long)ins->objectid,
5501 (unsigned long long)ins->offset);
5507 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5508 struct btrfs_root *root,
5509 u64 parent, u64 root_objectid,
5510 u64 flags, struct btrfs_disk_key *key,
5511 int level, struct btrfs_key *ins)
5514 struct btrfs_fs_info *fs_info = root->fs_info;
5515 struct btrfs_extent_item *extent_item;
5516 struct btrfs_tree_block_info *block_info;
5517 struct btrfs_extent_inline_ref *iref;
5518 struct btrfs_path *path;
5519 struct extent_buffer *leaf;
5520 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5522 path = btrfs_alloc_path();
5526 path->leave_spinning = 1;
5527 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5531 leaf = path->nodes[0];
5532 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5533 struct btrfs_extent_item);
5534 btrfs_set_extent_refs(leaf, extent_item, 1);
5535 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5536 btrfs_set_extent_flags(leaf, extent_item,
5537 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5538 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5540 btrfs_set_tree_block_key(leaf, block_info, key);
5541 btrfs_set_tree_block_level(leaf, block_info, level);
5543 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5545 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5546 btrfs_set_extent_inline_ref_type(leaf, iref,
5547 BTRFS_SHARED_BLOCK_REF_KEY);
5548 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5550 btrfs_set_extent_inline_ref_type(leaf, iref,
5551 BTRFS_TREE_BLOCK_REF_KEY);
5552 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5555 btrfs_mark_buffer_dirty(leaf);
5556 btrfs_free_path(path);
5558 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5560 printk(KERN_ERR "btrfs update block group failed for %llu "
5561 "%llu\n", (unsigned long long)ins->objectid,
5562 (unsigned long long)ins->offset);
5568 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5569 struct btrfs_root *root,
5570 u64 root_objectid, u64 owner,
5571 u64 offset, struct btrfs_key *ins)
5575 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5577 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5578 0, root_objectid, owner, offset,
5579 BTRFS_ADD_DELAYED_EXTENT, NULL);
5584 * this is used by the tree logging recovery code. It records that
5585 * an extent has been allocated and makes sure to clear the free
5586 * space cache bits as well
5588 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5589 struct btrfs_root *root,
5590 u64 root_objectid, u64 owner, u64 offset,
5591 struct btrfs_key *ins)
5594 struct btrfs_block_group_cache *block_group;
5595 struct btrfs_caching_control *caching_ctl;
5596 u64 start = ins->objectid;
5597 u64 num_bytes = ins->offset;
5599 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5600 cache_block_group(block_group, trans, NULL, 0);
5601 caching_ctl = get_caching_control(block_group);
5604 BUG_ON(!block_group_cache_done(block_group));
5605 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5608 mutex_lock(&caching_ctl->mutex);
5610 if (start >= caching_ctl->progress) {
5611 ret = add_excluded_extent(root, start, num_bytes);
5613 } else if (start + num_bytes <= caching_ctl->progress) {
5614 ret = btrfs_remove_free_space(block_group,
5618 num_bytes = caching_ctl->progress - start;
5619 ret = btrfs_remove_free_space(block_group,
5623 start = caching_ctl->progress;
5624 num_bytes = ins->objectid + ins->offset -
5625 caching_ctl->progress;
5626 ret = add_excluded_extent(root, start, num_bytes);
5630 mutex_unlock(&caching_ctl->mutex);
5631 put_caching_control(caching_ctl);
5634 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5635 RESERVE_ALLOC_NO_ACCOUNT);
5637 btrfs_put_block_group(block_group);
5638 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5639 0, owner, offset, ins, 1);
5643 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5644 struct btrfs_root *root,
5645 u64 bytenr, u32 blocksize,
5648 struct extent_buffer *buf;
5650 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5652 return ERR_PTR(-ENOMEM);
5653 btrfs_set_header_generation(buf, trans->transid);
5654 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5655 btrfs_tree_lock(buf);
5656 clean_tree_block(trans, root, buf);
5658 btrfs_set_lock_blocking(buf);
5659 btrfs_set_buffer_uptodate(buf);
5661 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5663 * we allow two log transactions at a time, use different
5664 * EXENT bit to differentiate dirty pages.
5666 if (root->log_transid % 2 == 0)
5667 set_extent_dirty(&root->dirty_log_pages, buf->start,
5668 buf->start + buf->len - 1, GFP_NOFS);
5670 set_extent_new(&root->dirty_log_pages, buf->start,
5671 buf->start + buf->len - 1, GFP_NOFS);
5673 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5674 buf->start + buf->len - 1, GFP_NOFS);
5676 trans->blocks_used++;
5677 /* this returns a buffer locked for blocking */
5681 static struct btrfs_block_rsv *
5682 use_block_rsv(struct btrfs_trans_handle *trans,
5683 struct btrfs_root *root, u32 blocksize)
5685 struct btrfs_block_rsv *block_rsv;
5686 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5689 block_rsv = get_block_rsv(trans, root);
5691 if (block_rsv->size == 0) {
5692 ret = reserve_metadata_bytes(trans, root, block_rsv,
5695 * If we couldn't reserve metadata bytes try and use some from
5696 * the global reserve.
5698 if (ret && block_rsv != global_rsv) {
5699 ret = block_rsv_use_bytes(global_rsv, blocksize);
5702 return ERR_PTR(ret);
5704 return ERR_PTR(ret);
5709 ret = block_rsv_use_bytes(block_rsv, blocksize);
5714 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5718 } else if (ret && block_rsv != global_rsv) {
5719 ret = block_rsv_use_bytes(global_rsv, blocksize);
5725 return ERR_PTR(-ENOSPC);
5728 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5730 block_rsv_add_bytes(block_rsv, blocksize, 0);
5731 block_rsv_release_bytes(block_rsv, NULL, 0);
5735 * finds a free extent and does all the dirty work required for allocation
5736 * returns the key for the extent through ins, and a tree buffer for
5737 * the first block of the extent through buf.
5739 * returns the tree buffer or NULL.
5741 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5742 struct btrfs_root *root, u32 blocksize,
5743 u64 parent, u64 root_objectid,
5744 struct btrfs_disk_key *key, int level,
5745 u64 hint, u64 empty_size)
5747 struct btrfs_key ins;
5748 struct btrfs_block_rsv *block_rsv;
5749 struct extent_buffer *buf;
5754 block_rsv = use_block_rsv(trans, root, blocksize);
5755 if (IS_ERR(block_rsv))
5756 return ERR_CAST(block_rsv);
5758 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5759 empty_size, hint, (u64)-1, &ins, 0);
5761 unuse_block_rsv(block_rsv, blocksize);
5762 return ERR_PTR(ret);
5765 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5767 BUG_ON(IS_ERR(buf));
5769 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5771 parent = ins.objectid;
5772 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5776 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5777 struct btrfs_delayed_extent_op *extent_op;
5778 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5781 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5783 memset(&extent_op->key, 0, sizeof(extent_op->key));
5784 extent_op->flags_to_set = flags;
5785 extent_op->update_key = 1;
5786 extent_op->update_flags = 1;
5787 extent_op->is_data = 0;
5789 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5790 ins.offset, parent, root_objectid,
5791 level, BTRFS_ADD_DELAYED_EXTENT,
5798 struct walk_control {
5799 u64 refs[BTRFS_MAX_LEVEL];
5800 u64 flags[BTRFS_MAX_LEVEL];
5801 struct btrfs_key update_progress;
5811 #define DROP_REFERENCE 1
5812 #define UPDATE_BACKREF 2
5814 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5815 struct btrfs_root *root,
5816 struct walk_control *wc,
5817 struct btrfs_path *path)
5825 struct btrfs_key key;
5826 struct extent_buffer *eb;
5831 if (path->slots[wc->level] < wc->reada_slot) {
5832 wc->reada_count = wc->reada_count * 2 / 3;
5833 wc->reada_count = max(wc->reada_count, 2);
5835 wc->reada_count = wc->reada_count * 3 / 2;
5836 wc->reada_count = min_t(int, wc->reada_count,
5837 BTRFS_NODEPTRS_PER_BLOCK(root));
5840 eb = path->nodes[wc->level];
5841 nritems = btrfs_header_nritems(eb);
5842 blocksize = btrfs_level_size(root, wc->level - 1);
5844 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5845 if (nread >= wc->reada_count)
5849 bytenr = btrfs_node_blockptr(eb, slot);
5850 generation = btrfs_node_ptr_generation(eb, slot);
5852 if (slot == path->slots[wc->level])
5855 if (wc->stage == UPDATE_BACKREF &&
5856 generation <= root->root_key.offset)
5859 /* We don't lock the tree block, it's OK to be racy here */
5860 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5865 if (wc->stage == DROP_REFERENCE) {
5869 if (wc->level == 1 &&
5870 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5872 if (!wc->update_ref ||
5873 generation <= root->root_key.offset)
5875 btrfs_node_key_to_cpu(eb, &key, slot);
5876 ret = btrfs_comp_cpu_keys(&key,
5877 &wc->update_progress);
5881 if (wc->level == 1 &&
5882 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5886 ret = readahead_tree_block(root, bytenr, blocksize,
5892 wc->reada_slot = slot;
5896 * hepler to process tree block while walking down the tree.
5898 * when wc->stage == UPDATE_BACKREF, this function updates
5899 * back refs for pointers in the block.
5901 * NOTE: return value 1 means we should stop walking down.
5903 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5904 struct btrfs_root *root,
5905 struct btrfs_path *path,
5906 struct walk_control *wc, int lookup_info)
5908 int level = wc->level;
5909 struct extent_buffer *eb = path->nodes[level];
5910 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5913 if (wc->stage == UPDATE_BACKREF &&
5914 btrfs_header_owner(eb) != root->root_key.objectid)
5918 * when reference count of tree block is 1, it won't increase
5919 * again. once full backref flag is set, we never clear it.
5922 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5923 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5924 BUG_ON(!path->locks[level]);
5925 ret = btrfs_lookup_extent_info(trans, root,
5930 BUG_ON(wc->refs[level] == 0);
5933 if (wc->stage == DROP_REFERENCE) {
5934 if (wc->refs[level] > 1)
5937 if (path->locks[level] && !wc->keep_locks) {
5938 btrfs_tree_unlock_rw(eb, path->locks[level]);
5939 path->locks[level] = 0;
5944 /* wc->stage == UPDATE_BACKREF */
5945 if (!(wc->flags[level] & flag)) {
5946 BUG_ON(!path->locks[level]);
5947 ret = btrfs_inc_ref(trans, root, eb, 1);
5949 ret = btrfs_dec_ref(trans, root, eb, 0);
5951 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5954 wc->flags[level] |= flag;
5958 * the block is shared by multiple trees, so it's not good to
5959 * keep the tree lock
5961 if (path->locks[level] && level > 0) {
5962 btrfs_tree_unlock_rw(eb, path->locks[level]);
5963 path->locks[level] = 0;
5969 * hepler to process tree block pointer.
5971 * when wc->stage == DROP_REFERENCE, this function checks
5972 * reference count of the block pointed to. if the block
5973 * is shared and we need update back refs for the subtree
5974 * rooted at the block, this function changes wc->stage to
5975 * UPDATE_BACKREF. if the block is shared and there is no
5976 * need to update back, this function drops the reference
5979 * NOTE: return value 1 means we should stop walking down.
5981 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5982 struct btrfs_root *root,
5983 struct btrfs_path *path,
5984 struct walk_control *wc, int *lookup_info)
5990 struct btrfs_key key;
5991 struct extent_buffer *next;
5992 int level = wc->level;
5996 generation = btrfs_node_ptr_generation(path->nodes[level],
5997 path->slots[level]);
5999 * if the lower level block was created before the snapshot
6000 * was created, we know there is no need to update back refs
6003 if (wc->stage == UPDATE_BACKREF &&
6004 generation <= root->root_key.offset) {
6009 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6010 blocksize = btrfs_level_size(root, level - 1);
6012 next = btrfs_find_tree_block(root, bytenr, blocksize);
6014 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6019 btrfs_tree_lock(next);
6020 btrfs_set_lock_blocking(next);
6022 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6023 &wc->refs[level - 1],
6024 &wc->flags[level - 1]);
6026 BUG_ON(wc->refs[level - 1] == 0);
6029 if (wc->stage == DROP_REFERENCE) {
6030 if (wc->refs[level - 1] > 1) {
6032 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6035 if (!wc->update_ref ||
6036 generation <= root->root_key.offset)
6039 btrfs_node_key_to_cpu(path->nodes[level], &key,
6040 path->slots[level]);
6041 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6045 wc->stage = UPDATE_BACKREF;
6046 wc->shared_level = level - 1;
6050 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6054 if (!btrfs_buffer_uptodate(next, generation)) {
6055 btrfs_tree_unlock(next);
6056 free_extent_buffer(next);
6062 if (reada && level == 1)
6063 reada_walk_down(trans, root, wc, path);
6064 next = read_tree_block(root, bytenr, blocksize, generation);
6067 btrfs_tree_lock(next);
6068 btrfs_set_lock_blocking(next);
6072 BUG_ON(level != btrfs_header_level(next));
6073 path->nodes[level] = next;
6074 path->slots[level] = 0;
6075 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6081 wc->refs[level - 1] = 0;
6082 wc->flags[level - 1] = 0;
6083 if (wc->stage == DROP_REFERENCE) {
6084 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6085 parent = path->nodes[level]->start;
6087 BUG_ON(root->root_key.objectid !=
6088 btrfs_header_owner(path->nodes[level]));
6092 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6093 root->root_key.objectid, level - 1, 0);
6096 btrfs_tree_unlock(next);
6097 free_extent_buffer(next);
6103 * hepler to process tree block while walking up the tree.
6105 * when wc->stage == DROP_REFERENCE, this function drops
6106 * reference count on the block.
6108 * when wc->stage == UPDATE_BACKREF, this function changes
6109 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6110 * to UPDATE_BACKREF previously while processing the block.
6112 * NOTE: return value 1 means we should stop walking up.
6114 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6115 struct btrfs_root *root,
6116 struct btrfs_path *path,
6117 struct walk_control *wc)
6120 int level = wc->level;
6121 struct extent_buffer *eb = path->nodes[level];
6124 if (wc->stage == UPDATE_BACKREF) {
6125 BUG_ON(wc->shared_level < level);
6126 if (level < wc->shared_level)
6129 ret = find_next_key(path, level + 1, &wc->update_progress);
6133 wc->stage = DROP_REFERENCE;
6134 wc->shared_level = -1;
6135 path->slots[level] = 0;
6138 * check reference count again if the block isn't locked.
6139 * we should start walking down the tree again if reference
6142 if (!path->locks[level]) {
6144 btrfs_tree_lock(eb);
6145 btrfs_set_lock_blocking(eb);
6146 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6148 ret = btrfs_lookup_extent_info(trans, root,
6153 BUG_ON(wc->refs[level] == 0);
6154 if (wc->refs[level] == 1) {
6155 btrfs_tree_unlock_rw(eb, path->locks[level]);
6161 /* wc->stage == DROP_REFERENCE */
6162 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6164 if (wc->refs[level] == 1) {
6166 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6167 ret = btrfs_dec_ref(trans, root, eb, 1);
6169 ret = btrfs_dec_ref(trans, root, eb, 0);
6172 /* make block locked assertion in clean_tree_block happy */
6173 if (!path->locks[level] &&
6174 btrfs_header_generation(eb) == trans->transid) {
6175 btrfs_tree_lock(eb);
6176 btrfs_set_lock_blocking(eb);
6177 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6179 clean_tree_block(trans, root, eb);
6182 if (eb == root->node) {
6183 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6186 BUG_ON(root->root_key.objectid !=
6187 btrfs_header_owner(eb));
6189 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6190 parent = path->nodes[level + 1]->start;
6192 BUG_ON(root->root_key.objectid !=
6193 btrfs_header_owner(path->nodes[level + 1]));
6196 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6198 wc->refs[level] = 0;
6199 wc->flags[level] = 0;
6203 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6204 struct btrfs_root *root,
6205 struct btrfs_path *path,
6206 struct walk_control *wc)
6208 int level = wc->level;
6209 int lookup_info = 1;
6212 while (level >= 0) {
6213 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6220 if (path->slots[level] >=
6221 btrfs_header_nritems(path->nodes[level]))
6224 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6226 path->slots[level]++;
6235 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6236 struct btrfs_root *root,
6237 struct btrfs_path *path,
6238 struct walk_control *wc, int max_level)
6240 int level = wc->level;
6243 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6244 while (level < max_level && path->nodes[level]) {
6246 if (path->slots[level] + 1 <
6247 btrfs_header_nritems(path->nodes[level])) {
6248 path->slots[level]++;
6251 ret = walk_up_proc(trans, root, path, wc);
6255 if (path->locks[level]) {
6256 btrfs_tree_unlock_rw(path->nodes[level],
6257 path->locks[level]);
6258 path->locks[level] = 0;
6260 free_extent_buffer(path->nodes[level]);
6261 path->nodes[level] = NULL;
6269 * drop a subvolume tree.
6271 * this function traverses the tree freeing any blocks that only
6272 * referenced by the tree.
6274 * when a shared tree block is found. this function decreases its
6275 * reference count by one. if update_ref is true, this function
6276 * also make sure backrefs for the shared block and all lower level
6277 * blocks are properly updated.
6279 void btrfs_drop_snapshot(struct btrfs_root *root,
6280 struct btrfs_block_rsv *block_rsv, int update_ref)
6282 struct btrfs_path *path;
6283 struct btrfs_trans_handle *trans;
6284 struct btrfs_root *tree_root = root->fs_info->tree_root;
6285 struct btrfs_root_item *root_item = &root->root_item;
6286 struct walk_control *wc;
6287 struct btrfs_key key;
6292 path = btrfs_alloc_path();
6298 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6300 btrfs_free_path(path);
6305 trans = btrfs_start_transaction(tree_root, 0);
6306 BUG_ON(IS_ERR(trans));
6309 trans->block_rsv = block_rsv;
6311 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6312 level = btrfs_header_level(root->node);
6313 path->nodes[level] = btrfs_lock_root_node(root);
6314 btrfs_set_lock_blocking(path->nodes[level]);
6315 path->slots[level] = 0;
6316 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6317 memset(&wc->update_progress, 0,
6318 sizeof(wc->update_progress));
6320 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6321 memcpy(&wc->update_progress, &key,
6322 sizeof(wc->update_progress));
6324 level = root_item->drop_level;
6326 path->lowest_level = level;
6327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6328 path->lowest_level = 0;
6336 * unlock our path, this is safe because only this
6337 * function is allowed to delete this snapshot
6339 btrfs_unlock_up_safe(path, 0);
6341 level = btrfs_header_level(root->node);
6343 btrfs_tree_lock(path->nodes[level]);
6344 btrfs_set_lock_blocking(path->nodes[level]);
6346 ret = btrfs_lookup_extent_info(trans, root,
6347 path->nodes[level]->start,
6348 path->nodes[level]->len,
6352 BUG_ON(wc->refs[level] == 0);
6354 if (level == root_item->drop_level)
6357 btrfs_tree_unlock(path->nodes[level]);
6358 WARN_ON(wc->refs[level] != 1);
6364 wc->shared_level = -1;
6365 wc->stage = DROP_REFERENCE;
6366 wc->update_ref = update_ref;
6368 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6371 ret = walk_down_tree(trans, root, path, wc);
6377 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6384 BUG_ON(wc->stage != DROP_REFERENCE);
6388 if (wc->stage == DROP_REFERENCE) {
6390 btrfs_node_key(path->nodes[level],
6391 &root_item->drop_progress,
6392 path->slots[level]);
6393 root_item->drop_level = level;
6396 BUG_ON(wc->level == 0);
6397 if (btrfs_should_end_transaction(trans, tree_root)) {
6398 ret = btrfs_update_root(trans, tree_root,
6403 btrfs_end_transaction_throttle(trans, tree_root);
6404 trans = btrfs_start_transaction(tree_root, 0);
6405 BUG_ON(IS_ERR(trans));
6407 trans->block_rsv = block_rsv;
6410 btrfs_release_path(path);
6413 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6416 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6417 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6421 /* if we fail to delete the orphan item this time
6422 * around, it'll get picked up the next time.
6424 * The most common failure here is just -ENOENT.
6426 btrfs_del_orphan_item(trans, tree_root,
6427 root->root_key.objectid);
6431 if (root->in_radix) {
6432 btrfs_free_fs_root(tree_root->fs_info, root);
6434 free_extent_buffer(root->node);
6435 free_extent_buffer(root->commit_root);
6439 btrfs_end_transaction_throttle(trans, tree_root);
6441 btrfs_free_path(path);
6444 btrfs_std_error(root->fs_info, err);
6449 * drop subtree rooted at tree block 'node'.
6451 * NOTE: this function will unlock and release tree block 'node'
6453 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6454 struct btrfs_root *root,
6455 struct extent_buffer *node,
6456 struct extent_buffer *parent)
6458 struct btrfs_path *path;
6459 struct walk_control *wc;
6465 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6467 path = btrfs_alloc_path();
6471 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6473 btrfs_free_path(path);
6477 btrfs_assert_tree_locked(parent);
6478 parent_level = btrfs_header_level(parent);
6479 extent_buffer_get(parent);
6480 path->nodes[parent_level] = parent;
6481 path->slots[parent_level] = btrfs_header_nritems(parent);
6483 btrfs_assert_tree_locked(node);
6484 level = btrfs_header_level(node);
6485 path->nodes[level] = node;
6486 path->slots[level] = 0;
6487 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6489 wc->refs[parent_level] = 1;
6490 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6492 wc->shared_level = -1;
6493 wc->stage = DROP_REFERENCE;
6496 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6499 wret = walk_down_tree(trans, root, path, wc);
6505 wret = walk_up_tree(trans, root, path, wc, parent_level);
6513 btrfs_free_path(path);
6517 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6520 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6521 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6524 * we add in the count of missing devices because we want
6525 * to make sure that any RAID levels on a degraded FS
6526 * continue to be honored.
6528 num_devices = root->fs_info->fs_devices->rw_devices +
6529 root->fs_info->fs_devices->missing_devices;
6531 if (num_devices == 1) {
6532 stripped |= BTRFS_BLOCK_GROUP_DUP;
6533 stripped = flags & ~stripped;
6535 /* turn raid0 into single device chunks */
6536 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6539 /* turn mirroring into duplication */
6540 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6541 BTRFS_BLOCK_GROUP_RAID10))
6542 return stripped | BTRFS_BLOCK_GROUP_DUP;
6545 /* they already had raid on here, just return */
6546 if (flags & stripped)
6549 stripped |= BTRFS_BLOCK_GROUP_DUP;
6550 stripped = flags & ~stripped;
6552 /* switch duplicated blocks with raid1 */
6553 if (flags & BTRFS_BLOCK_GROUP_DUP)
6554 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6556 /* turn single device chunks into raid0 */
6557 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6562 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6564 struct btrfs_space_info *sinfo = cache->space_info;
6566 u64 min_allocable_bytes;
6571 * We need some metadata space and system metadata space for
6572 * allocating chunks in some corner cases until we force to set
6573 * it to be readonly.
6576 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6578 min_allocable_bytes = 1 * 1024 * 1024;
6580 min_allocable_bytes = 0;
6582 spin_lock(&sinfo->lock);
6583 spin_lock(&cache->lock);
6590 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6591 cache->bytes_super - btrfs_block_group_used(&cache->item);
6593 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6594 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6595 min_allocable_bytes <= sinfo->total_bytes) {
6596 sinfo->bytes_readonly += num_bytes;
6601 spin_unlock(&cache->lock);
6602 spin_unlock(&sinfo->lock);
6606 int btrfs_set_block_group_ro(struct btrfs_root *root,
6607 struct btrfs_block_group_cache *cache)
6610 struct btrfs_trans_handle *trans;
6616 trans = btrfs_join_transaction(root);
6617 BUG_ON(IS_ERR(trans));
6619 alloc_flags = update_block_group_flags(root, cache->flags);
6620 if (alloc_flags != cache->flags)
6621 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6624 ret = set_block_group_ro(cache, 0);
6627 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6628 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6632 ret = set_block_group_ro(cache, 0);
6634 btrfs_end_transaction(trans, root);
6638 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6639 struct btrfs_root *root, u64 type)
6641 u64 alloc_flags = get_alloc_profile(root, type);
6642 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6647 * helper to account the unused space of all the readonly block group in the
6648 * list. takes mirrors into account.
6650 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6652 struct btrfs_block_group_cache *block_group;
6656 list_for_each_entry(block_group, groups_list, list) {
6657 spin_lock(&block_group->lock);
6659 if (!block_group->ro) {
6660 spin_unlock(&block_group->lock);
6664 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6665 BTRFS_BLOCK_GROUP_RAID10 |
6666 BTRFS_BLOCK_GROUP_DUP))
6671 free_bytes += (block_group->key.offset -
6672 btrfs_block_group_used(&block_group->item)) *
6675 spin_unlock(&block_group->lock);
6682 * helper to account the unused space of all the readonly block group in the
6683 * space_info. takes mirrors into account.
6685 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6690 spin_lock(&sinfo->lock);
6692 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6693 if (!list_empty(&sinfo->block_groups[i]))
6694 free_bytes += __btrfs_get_ro_block_group_free_space(
6695 &sinfo->block_groups[i]);
6697 spin_unlock(&sinfo->lock);
6702 int btrfs_set_block_group_rw(struct btrfs_root *root,
6703 struct btrfs_block_group_cache *cache)
6705 struct btrfs_space_info *sinfo = cache->space_info;
6710 spin_lock(&sinfo->lock);
6711 spin_lock(&cache->lock);
6712 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6713 cache->bytes_super - btrfs_block_group_used(&cache->item);
6714 sinfo->bytes_readonly -= num_bytes;
6716 spin_unlock(&cache->lock);
6717 spin_unlock(&sinfo->lock);
6722 * checks to see if its even possible to relocate this block group.
6724 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6725 * ok to go ahead and try.
6727 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6729 struct btrfs_block_group_cache *block_group;
6730 struct btrfs_space_info *space_info;
6731 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6732 struct btrfs_device *device;
6740 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6742 /* odd, couldn't find the block group, leave it alone */
6746 min_free = btrfs_block_group_used(&block_group->item);
6748 /* no bytes used, we're good */
6752 space_info = block_group->space_info;
6753 spin_lock(&space_info->lock);
6755 full = space_info->full;
6758 * if this is the last block group we have in this space, we can't
6759 * relocate it unless we're able to allocate a new chunk below.
6761 * Otherwise, we need to make sure we have room in the space to handle
6762 * all of the extents from this block group. If we can, we're good
6764 if ((space_info->total_bytes != block_group->key.offset) &&
6765 (space_info->bytes_used + space_info->bytes_reserved +
6766 space_info->bytes_pinned + space_info->bytes_readonly +
6767 min_free < space_info->total_bytes)) {
6768 spin_unlock(&space_info->lock);
6771 spin_unlock(&space_info->lock);
6774 * ok we don't have enough space, but maybe we have free space on our
6775 * devices to allocate new chunks for relocation, so loop through our
6776 * alloc devices and guess if we have enough space. However, if we
6777 * were marked as full, then we know there aren't enough chunks, and we
6792 index = get_block_group_index(block_group);
6797 } else if (index == 1) {
6799 } else if (index == 2) {
6802 } else if (index == 3) {
6803 dev_min = fs_devices->rw_devices;
6804 do_div(min_free, dev_min);
6807 mutex_lock(&root->fs_info->chunk_mutex);
6808 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6812 * check to make sure we can actually find a chunk with enough
6813 * space to fit our block group in.
6815 if (device->total_bytes > device->bytes_used + min_free) {
6816 ret = find_free_dev_extent(NULL, device, min_free,
6821 if (dev_nr >= dev_min)
6827 mutex_unlock(&root->fs_info->chunk_mutex);
6829 btrfs_put_block_group(block_group);
6833 static int find_first_block_group(struct btrfs_root *root,
6834 struct btrfs_path *path, struct btrfs_key *key)
6837 struct btrfs_key found_key;
6838 struct extent_buffer *leaf;
6841 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6846 slot = path->slots[0];
6847 leaf = path->nodes[0];
6848 if (slot >= btrfs_header_nritems(leaf)) {
6849 ret = btrfs_next_leaf(root, path);
6856 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6858 if (found_key.objectid >= key->objectid &&
6859 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6869 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6871 struct btrfs_block_group_cache *block_group;
6875 struct inode *inode;
6877 block_group = btrfs_lookup_first_block_group(info, last);
6878 while (block_group) {
6879 spin_lock(&block_group->lock);
6880 if (block_group->iref)
6882 spin_unlock(&block_group->lock);
6883 block_group = next_block_group(info->tree_root,
6893 inode = block_group->inode;
6894 block_group->iref = 0;
6895 block_group->inode = NULL;
6896 spin_unlock(&block_group->lock);
6898 last = block_group->key.objectid + block_group->key.offset;
6899 btrfs_put_block_group(block_group);
6903 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6905 struct btrfs_block_group_cache *block_group;
6906 struct btrfs_space_info *space_info;
6907 struct btrfs_caching_control *caching_ctl;
6910 down_write(&info->extent_commit_sem);
6911 while (!list_empty(&info->caching_block_groups)) {
6912 caching_ctl = list_entry(info->caching_block_groups.next,
6913 struct btrfs_caching_control, list);
6914 list_del(&caching_ctl->list);
6915 put_caching_control(caching_ctl);
6917 up_write(&info->extent_commit_sem);
6919 spin_lock(&info->block_group_cache_lock);
6920 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6921 block_group = rb_entry(n, struct btrfs_block_group_cache,
6923 rb_erase(&block_group->cache_node,
6924 &info->block_group_cache_tree);
6925 spin_unlock(&info->block_group_cache_lock);
6927 down_write(&block_group->space_info->groups_sem);
6928 list_del(&block_group->list);
6929 up_write(&block_group->space_info->groups_sem);
6931 if (block_group->cached == BTRFS_CACHE_STARTED)
6932 wait_block_group_cache_done(block_group);
6935 * We haven't cached this block group, which means we could
6936 * possibly have excluded extents on this block group.
6938 if (block_group->cached == BTRFS_CACHE_NO)
6939 free_excluded_extents(info->extent_root, block_group);
6941 btrfs_remove_free_space_cache(block_group);
6942 btrfs_put_block_group(block_group);
6944 spin_lock(&info->block_group_cache_lock);
6946 spin_unlock(&info->block_group_cache_lock);
6948 /* now that all the block groups are freed, go through and
6949 * free all the space_info structs. This is only called during
6950 * the final stages of unmount, and so we know nobody is
6951 * using them. We call synchronize_rcu() once before we start,
6952 * just to be on the safe side.
6956 release_global_block_rsv(info);
6958 while(!list_empty(&info->space_info)) {
6959 space_info = list_entry(info->space_info.next,
6960 struct btrfs_space_info,
6962 if (space_info->bytes_pinned > 0 ||
6963 space_info->bytes_reserved > 0 ||
6964 space_info->bytes_may_use > 0) {
6966 dump_space_info(space_info, 0, 0);
6968 list_del(&space_info->list);
6974 static void __link_block_group(struct btrfs_space_info *space_info,
6975 struct btrfs_block_group_cache *cache)
6977 int index = get_block_group_index(cache);
6979 down_write(&space_info->groups_sem);
6980 list_add_tail(&cache->list, &space_info->block_groups[index]);
6981 up_write(&space_info->groups_sem);
6984 int btrfs_read_block_groups(struct btrfs_root *root)
6986 struct btrfs_path *path;
6988 struct btrfs_block_group_cache *cache;
6989 struct btrfs_fs_info *info = root->fs_info;
6990 struct btrfs_space_info *space_info;
6991 struct btrfs_key key;
6992 struct btrfs_key found_key;
6993 struct extent_buffer *leaf;
6997 root = info->extent_root;
7000 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7001 path = btrfs_alloc_path();
7006 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7007 if (cache_gen != 0 &&
7008 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7010 if (btrfs_test_opt(root, CLEAR_CACHE))
7012 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
7013 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
7016 ret = find_first_block_group(root, path, &key);
7021 leaf = path->nodes[0];
7022 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7023 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7028 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7030 if (!cache->free_space_ctl) {
7036 atomic_set(&cache->count, 1);
7037 spin_lock_init(&cache->lock);
7038 cache->fs_info = info;
7039 INIT_LIST_HEAD(&cache->list);
7040 INIT_LIST_HEAD(&cache->cluster_list);
7043 cache->disk_cache_state = BTRFS_DC_CLEAR;
7045 read_extent_buffer(leaf, &cache->item,
7046 btrfs_item_ptr_offset(leaf, path->slots[0]),
7047 sizeof(cache->item));
7048 memcpy(&cache->key, &found_key, sizeof(found_key));
7050 key.objectid = found_key.objectid + found_key.offset;
7051 btrfs_release_path(path);
7052 cache->flags = btrfs_block_group_flags(&cache->item);
7053 cache->sectorsize = root->sectorsize;
7055 btrfs_init_free_space_ctl(cache);
7058 * We need to exclude the super stripes now so that the space
7059 * info has super bytes accounted for, otherwise we'll think
7060 * we have more space than we actually do.
7062 exclude_super_stripes(root, cache);
7065 * check for two cases, either we are full, and therefore
7066 * don't need to bother with the caching work since we won't
7067 * find any space, or we are empty, and we can just add all
7068 * the space in and be done with it. This saves us _alot_ of
7069 * time, particularly in the full case.
7071 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7072 cache->last_byte_to_unpin = (u64)-1;
7073 cache->cached = BTRFS_CACHE_FINISHED;
7074 free_excluded_extents(root, cache);
7075 } else if (btrfs_block_group_used(&cache->item) == 0) {
7076 cache->last_byte_to_unpin = (u64)-1;
7077 cache->cached = BTRFS_CACHE_FINISHED;
7078 add_new_free_space(cache, root->fs_info,
7080 found_key.objectid +
7082 free_excluded_extents(root, cache);
7085 ret = update_space_info(info, cache->flags, found_key.offset,
7086 btrfs_block_group_used(&cache->item),
7089 cache->space_info = space_info;
7090 spin_lock(&cache->space_info->lock);
7091 cache->space_info->bytes_readonly += cache->bytes_super;
7092 spin_unlock(&cache->space_info->lock);
7094 __link_block_group(space_info, cache);
7096 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7099 set_avail_alloc_bits(root->fs_info, cache->flags);
7100 if (btrfs_chunk_readonly(root, cache->key.objectid))
7101 set_block_group_ro(cache, 1);
7104 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7105 if (!(get_alloc_profile(root, space_info->flags) &
7106 (BTRFS_BLOCK_GROUP_RAID10 |
7107 BTRFS_BLOCK_GROUP_RAID1 |
7108 BTRFS_BLOCK_GROUP_DUP)))
7111 * avoid allocating from un-mirrored block group if there are
7112 * mirrored block groups.
7114 list_for_each_entry(cache, &space_info->block_groups[3], list)
7115 set_block_group_ro(cache, 1);
7116 list_for_each_entry(cache, &space_info->block_groups[4], list)
7117 set_block_group_ro(cache, 1);
7120 init_global_block_rsv(info);
7123 btrfs_free_path(path);
7127 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7128 struct btrfs_root *root, u64 bytes_used,
7129 u64 type, u64 chunk_objectid, u64 chunk_offset,
7133 struct btrfs_root *extent_root;
7134 struct btrfs_block_group_cache *cache;
7136 extent_root = root->fs_info->extent_root;
7138 root->fs_info->last_trans_log_full_commit = trans->transid;
7140 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7143 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7145 if (!cache->free_space_ctl) {
7150 cache->key.objectid = chunk_offset;
7151 cache->key.offset = size;
7152 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7153 cache->sectorsize = root->sectorsize;
7154 cache->fs_info = root->fs_info;
7156 atomic_set(&cache->count, 1);
7157 spin_lock_init(&cache->lock);
7158 INIT_LIST_HEAD(&cache->list);
7159 INIT_LIST_HEAD(&cache->cluster_list);
7161 btrfs_init_free_space_ctl(cache);
7163 btrfs_set_block_group_used(&cache->item, bytes_used);
7164 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7165 cache->flags = type;
7166 btrfs_set_block_group_flags(&cache->item, type);
7168 cache->last_byte_to_unpin = (u64)-1;
7169 cache->cached = BTRFS_CACHE_FINISHED;
7170 exclude_super_stripes(root, cache);
7172 add_new_free_space(cache, root->fs_info, chunk_offset,
7173 chunk_offset + size);
7175 free_excluded_extents(root, cache);
7177 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7178 &cache->space_info);
7181 spin_lock(&cache->space_info->lock);
7182 cache->space_info->bytes_readonly += cache->bytes_super;
7183 spin_unlock(&cache->space_info->lock);
7185 __link_block_group(cache->space_info, cache);
7187 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7190 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7191 sizeof(cache->item));
7194 set_avail_alloc_bits(extent_root->fs_info, type);
7199 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7200 struct btrfs_root *root, u64 group_start)
7202 struct btrfs_path *path;
7203 struct btrfs_block_group_cache *block_group;
7204 struct btrfs_free_cluster *cluster;
7205 struct btrfs_root *tree_root = root->fs_info->tree_root;
7206 struct btrfs_key key;
7207 struct inode *inode;
7211 root = root->fs_info->extent_root;
7213 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7214 BUG_ON(!block_group);
7215 BUG_ON(!block_group->ro);
7218 * Free the reserved super bytes from this block group before
7221 free_excluded_extents(root, block_group);
7223 memcpy(&key, &block_group->key, sizeof(key));
7224 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7225 BTRFS_BLOCK_GROUP_RAID1 |
7226 BTRFS_BLOCK_GROUP_RAID10))
7231 /* make sure this block group isn't part of an allocation cluster */
7232 cluster = &root->fs_info->data_alloc_cluster;
7233 spin_lock(&cluster->refill_lock);
7234 btrfs_return_cluster_to_free_space(block_group, cluster);
7235 spin_unlock(&cluster->refill_lock);
7238 * make sure this block group isn't part of a metadata
7239 * allocation cluster
7241 cluster = &root->fs_info->meta_alloc_cluster;
7242 spin_lock(&cluster->refill_lock);
7243 btrfs_return_cluster_to_free_space(block_group, cluster);
7244 spin_unlock(&cluster->refill_lock);
7246 path = btrfs_alloc_path();
7252 inode = lookup_free_space_inode(root, block_group, path);
7253 if (!IS_ERR(inode)) {
7254 ret = btrfs_orphan_add(trans, inode);
7257 /* One for the block groups ref */
7258 spin_lock(&block_group->lock);
7259 if (block_group->iref) {
7260 block_group->iref = 0;
7261 block_group->inode = NULL;
7262 spin_unlock(&block_group->lock);
7265 spin_unlock(&block_group->lock);
7267 /* One for our lookup ref */
7271 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7272 key.offset = block_group->key.objectid;
7275 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7279 btrfs_release_path(path);
7281 ret = btrfs_del_item(trans, tree_root, path);
7284 btrfs_release_path(path);
7287 spin_lock(&root->fs_info->block_group_cache_lock);
7288 rb_erase(&block_group->cache_node,
7289 &root->fs_info->block_group_cache_tree);
7290 spin_unlock(&root->fs_info->block_group_cache_lock);
7292 down_write(&block_group->space_info->groups_sem);
7294 * we must use list_del_init so people can check to see if they
7295 * are still on the list after taking the semaphore
7297 list_del_init(&block_group->list);
7298 up_write(&block_group->space_info->groups_sem);
7300 if (block_group->cached == BTRFS_CACHE_STARTED)
7301 wait_block_group_cache_done(block_group);
7303 btrfs_remove_free_space_cache(block_group);
7305 spin_lock(&block_group->space_info->lock);
7306 block_group->space_info->total_bytes -= block_group->key.offset;
7307 block_group->space_info->bytes_readonly -= block_group->key.offset;
7308 block_group->space_info->disk_total -= block_group->key.offset * factor;
7309 spin_unlock(&block_group->space_info->lock);
7311 memcpy(&key, &block_group->key, sizeof(key));
7313 btrfs_clear_space_info_full(root->fs_info);
7315 btrfs_put_block_group(block_group);
7316 btrfs_put_block_group(block_group);
7318 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7324 ret = btrfs_del_item(trans, root, path);
7326 btrfs_free_path(path);
7330 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7332 struct btrfs_space_info *space_info;
7333 struct btrfs_super_block *disk_super;
7339 disk_super = &fs_info->super_copy;
7340 if (!btrfs_super_root(disk_super))
7343 features = btrfs_super_incompat_flags(disk_super);
7344 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7347 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7348 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7353 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7354 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7356 flags = BTRFS_BLOCK_GROUP_METADATA;
7357 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7361 flags = BTRFS_BLOCK_GROUP_DATA;
7362 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7368 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7370 return unpin_extent_range(root, start, end);
7373 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7374 u64 num_bytes, u64 *actual_bytes)
7376 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7379 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7381 struct btrfs_fs_info *fs_info = root->fs_info;
7382 struct btrfs_block_group_cache *cache = NULL;
7389 cache = btrfs_lookup_block_group(fs_info, range->start);
7392 if (cache->key.objectid >= (range->start + range->len)) {
7393 btrfs_put_block_group(cache);
7397 start = max(range->start, cache->key.objectid);
7398 end = min(range->start + range->len,
7399 cache->key.objectid + cache->key.offset);
7401 if (end - start >= range->minlen) {
7402 if (!block_group_cache_done(cache)) {
7403 ret = cache_block_group(cache, NULL, root, 0);
7405 wait_block_group_cache_done(cache);
7407 ret = btrfs_trim_block_group(cache,
7413 trimmed += group_trimmed;
7415 btrfs_put_block_group(cache);
7420 cache = next_block_group(fs_info->tree_root, cache);
7423 range->len = trimmed;