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 static int update_block_group(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 u64 bytenr, u64 num_bytes, int alloc);
39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve, int sinfo);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int find_next_key(struct btrfs_path *path, int level,
64 struct btrfs_key *key);
65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
66 int dump_block_groups);
69 block_group_cache_done(struct btrfs_block_group_cache *cache)
72 return cache->cached == BTRFS_CACHE_FINISHED;
75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
77 return (cache->flags & bits) == bits;
80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
82 atomic_inc(&cache->count);
85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
87 if (atomic_dec_and_test(&cache->count)) {
88 WARN_ON(cache->pinned > 0);
89 WARN_ON(cache->reserved > 0);
90 WARN_ON(cache->reserved_pinned > 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
100 struct btrfs_block_group_cache *block_group)
103 struct rb_node *parent = NULL;
104 struct btrfs_block_group_cache *cache;
106 spin_lock(&info->block_group_cache_lock);
107 p = &info->block_group_cache_tree.rb_node;
111 cache = rb_entry(parent, struct btrfs_block_group_cache,
113 if (block_group->key.objectid < cache->key.objectid) {
115 } else if (block_group->key.objectid > cache->key.objectid) {
118 spin_unlock(&info->block_group_cache_lock);
123 rb_link_node(&block_group->cache_node, parent, p);
124 rb_insert_color(&block_group->cache_node,
125 &info->block_group_cache_tree);
126 spin_unlock(&info->block_group_cache_lock);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache *
136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
139 struct btrfs_block_group_cache *cache, *ret = NULL;
143 spin_lock(&info->block_group_cache_lock);
144 n = info->block_group_cache_tree.rb_node;
147 cache = rb_entry(n, struct btrfs_block_group_cache,
149 end = cache->key.objectid + cache->key.offset - 1;
150 start = cache->key.objectid;
152 if (bytenr < start) {
153 if (!contains && (!ret || start < ret->key.objectid))
156 } else if (bytenr > start) {
157 if (contains && bytenr <= end) {
168 btrfs_get_block_group(ret);
169 spin_unlock(&info->block_group_cache_lock);
174 static int add_excluded_extent(struct btrfs_root *root,
175 u64 start, u64 num_bytes)
177 u64 end = start + num_bytes - 1;
178 set_extent_bits(&root->fs_info->freed_extents[0],
179 start, end, EXTENT_UPTODATE, GFP_NOFS);
180 set_extent_bits(&root->fs_info->freed_extents[1],
181 start, end, EXTENT_UPTODATE, GFP_NOFS);
185 static void free_excluded_extents(struct btrfs_root *root,
186 struct btrfs_block_group_cache *cache)
190 start = cache->key.objectid;
191 end = start + cache->key.offset - 1;
193 clear_extent_bits(&root->fs_info->freed_extents[0],
194 start, end, EXTENT_UPTODATE, GFP_NOFS);
195 clear_extent_bits(&root->fs_info->freed_extents[1],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
199 static int exclude_super_stripes(struct btrfs_root *root,
200 struct btrfs_block_group_cache *cache)
207 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
208 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
209 cache->bytes_super += stripe_len;
210 ret = add_excluded_extent(root, cache->key.objectid,
215 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
216 bytenr = btrfs_sb_offset(i);
217 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
218 cache->key.objectid, bytenr,
219 0, &logical, &nr, &stripe_len);
223 cache->bytes_super += stripe_len;
224 ret = add_excluded_extent(root, logical[nr],
234 static struct btrfs_caching_control *
235 get_caching_control(struct btrfs_block_group_cache *cache)
237 struct btrfs_caching_control *ctl;
239 spin_lock(&cache->lock);
240 if (cache->cached != BTRFS_CACHE_STARTED) {
241 spin_unlock(&cache->lock);
245 /* We're loading it the fast way, so we don't have a caching_ctl. */
246 if (!cache->caching_ctl) {
247 spin_unlock(&cache->lock);
251 ctl = cache->caching_ctl;
252 atomic_inc(&ctl->count);
253 spin_unlock(&cache->lock);
257 static void put_caching_control(struct btrfs_caching_control *ctl)
259 if (atomic_dec_and_test(&ctl->count))
264 * this is only called by cache_block_group, since we could have freed extents
265 * we need to check the pinned_extents for any extents that can't be used yet
266 * since their free space will be released as soon as the transaction commits.
268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
269 struct btrfs_fs_info *info, u64 start, u64 end)
271 u64 extent_start, extent_end, size, total_added = 0;
274 while (start < end) {
275 ret = find_first_extent_bit(info->pinned_extents, start,
276 &extent_start, &extent_end,
277 EXTENT_DIRTY | EXTENT_UPTODATE);
281 if (extent_start <= start) {
282 start = extent_end + 1;
283 } else if (extent_start > start && extent_start < end) {
284 size = extent_start - start;
286 ret = btrfs_add_free_space(block_group, start,
289 start = extent_end + 1;
298 ret = btrfs_add_free_space(block_group, start, size);
305 static int caching_kthread(void *data)
307 struct btrfs_block_group_cache *block_group = data;
308 struct btrfs_fs_info *fs_info = block_group->fs_info;
309 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
310 struct btrfs_root *extent_root = fs_info->extent_root;
311 struct btrfs_path *path;
312 struct extent_buffer *leaf;
313 struct btrfs_key key;
319 path = btrfs_alloc_path();
323 exclude_super_stripes(extent_root, block_group);
324 spin_lock(&block_group->space_info->lock);
325 block_group->space_info->bytes_readonly += block_group->bytes_super;
326 spin_unlock(&block_group->space_info->lock);
328 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
331 * We don't want to deadlock with somebody trying to allocate a new
332 * extent for the extent root while also trying to search the extent
333 * root to add free space. So we skip locking and search the commit
334 * root, since its read-only
336 path->skip_locking = 1;
337 path->search_commit_root = 1;
342 key.type = BTRFS_EXTENT_ITEM_KEY;
344 mutex_lock(&caching_ctl->mutex);
345 /* need to make sure the commit_root doesn't disappear */
346 down_read(&fs_info->extent_commit_sem);
348 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
352 leaf = path->nodes[0];
353 nritems = btrfs_header_nritems(leaf);
357 if (fs_info->closing > 1) {
362 if (path->slots[0] < nritems) {
363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
365 ret = find_next_key(path, 0, &key);
369 caching_ctl->progress = last;
370 btrfs_release_path(extent_root, path);
371 up_read(&fs_info->extent_commit_sem);
372 mutex_unlock(&caching_ctl->mutex);
373 if (btrfs_transaction_in_commit(fs_info))
380 if (key.objectid < block_group->key.objectid) {
385 if (key.objectid >= block_group->key.objectid +
386 block_group->key.offset)
389 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
390 total_found += add_new_free_space(block_group,
393 last = key.objectid + key.offset;
395 if (total_found > (1024 * 1024 * 2)) {
397 wake_up(&caching_ctl->wait);
404 total_found += add_new_free_space(block_group, fs_info, last,
405 block_group->key.objectid +
406 block_group->key.offset);
407 caching_ctl->progress = (u64)-1;
409 spin_lock(&block_group->lock);
410 block_group->caching_ctl = NULL;
411 block_group->cached = BTRFS_CACHE_FINISHED;
412 spin_unlock(&block_group->lock);
415 btrfs_free_path(path);
416 up_read(&fs_info->extent_commit_sem);
418 free_excluded_extents(extent_root, block_group);
420 mutex_unlock(&caching_ctl->mutex);
421 wake_up(&caching_ctl->wait);
423 put_caching_control(caching_ctl);
424 atomic_dec(&block_group->space_info->caching_threads);
425 btrfs_put_block_group(block_group);
430 static int cache_block_group(struct btrfs_block_group_cache *cache,
431 struct btrfs_trans_handle *trans,
434 struct btrfs_fs_info *fs_info = cache->fs_info;
435 struct btrfs_caching_control *caching_ctl;
436 struct task_struct *tsk;
440 if (cache->cached != BTRFS_CACHE_NO)
444 * We can't do the read from on-disk cache during a commit since we need
445 * to have the normal tree locking.
447 if (!trans->transaction->in_commit) {
448 spin_lock(&cache->lock);
449 if (cache->cached != BTRFS_CACHE_NO) {
450 spin_unlock(&cache->lock);
453 cache->cached = BTRFS_CACHE_STARTED;
454 spin_unlock(&cache->lock);
456 ret = load_free_space_cache(fs_info, cache);
458 spin_lock(&cache->lock);
460 cache->cached = BTRFS_CACHE_FINISHED;
461 cache->last_byte_to_unpin = (u64)-1;
463 cache->cached = BTRFS_CACHE_NO;
465 spin_unlock(&cache->lock);
473 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
474 BUG_ON(!caching_ctl);
476 INIT_LIST_HEAD(&caching_ctl->list);
477 mutex_init(&caching_ctl->mutex);
478 init_waitqueue_head(&caching_ctl->wait);
479 caching_ctl->block_group = cache;
480 caching_ctl->progress = cache->key.objectid;
481 /* one for caching kthread, one for caching block group list */
482 atomic_set(&caching_ctl->count, 2);
484 spin_lock(&cache->lock);
485 if (cache->cached != BTRFS_CACHE_NO) {
486 spin_unlock(&cache->lock);
490 cache->caching_ctl = caching_ctl;
491 cache->cached = BTRFS_CACHE_STARTED;
492 spin_unlock(&cache->lock);
494 down_write(&fs_info->extent_commit_sem);
495 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
496 up_write(&fs_info->extent_commit_sem);
498 atomic_inc(&cache->space_info->caching_threads);
499 btrfs_get_block_group(cache);
501 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
502 cache->key.objectid);
505 printk(KERN_ERR "error running thread %d\n", ret);
513 * return the block group that starts at or after bytenr
515 static struct btrfs_block_group_cache *
516 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
518 struct btrfs_block_group_cache *cache;
520 cache = block_group_cache_tree_search(info, bytenr, 0);
526 * return the block group that contains the given bytenr
528 struct btrfs_block_group_cache *btrfs_lookup_block_group(
529 struct btrfs_fs_info *info,
532 struct btrfs_block_group_cache *cache;
534 cache = block_group_cache_tree_search(info, bytenr, 1);
539 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
542 struct list_head *head = &info->space_info;
543 struct btrfs_space_info *found;
545 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
546 BTRFS_BLOCK_GROUP_METADATA;
549 list_for_each_entry_rcu(found, head, list) {
550 if (found->flags & flags) {
560 * after adding space to the filesystem, we need to clear the full flags
561 * on all the space infos.
563 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
565 struct list_head *head = &info->space_info;
566 struct btrfs_space_info *found;
569 list_for_each_entry_rcu(found, head, list)
574 static u64 div_factor(u64 num, int factor)
583 u64 btrfs_find_block_group(struct btrfs_root *root,
584 u64 search_start, u64 search_hint, int owner)
586 struct btrfs_block_group_cache *cache;
588 u64 last = max(search_hint, search_start);
595 cache = btrfs_lookup_first_block_group(root->fs_info, last);
599 spin_lock(&cache->lock);
600 last = cache->key.objectid + cache->key.offset;
601 used = btrfs_block_group_used(&cache->item);
603 if ((full_search || !cache->ro) &&
604 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
605 if (used + cache->pinned + cache->reserved <
606 div_factor(cache->key.offset, factor)) {
607 group_start = cache->key.objectid;
608 spin_unlock(&cache->lock);
609 btrfs_put_block_group(cache);
613 spin_unlock(&cache->lock);
614 btrfs_put_block_group(cache);
622 if (!full_search && factor < 10) {
632 /* simple helper to search for an existing extent at a given offset */
633 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
636 struct btrfs_key key;
637 struct btrfs_path *path;
639 path = btrfs_alloc_path();
641 key.objectid = start;
643 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
644 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
646 btrfs_free_path(path);
651 * helper function to lookup reference count and flags of extent.
653 * the head node for delayed ref is used to store the sum of all the
654 * reference count modifications queued up in the rbtree. the head
655 * node may also store the extent flags to set. This way you can check
656 * to see what the reference count and extent flags would be if all of
657 * the delayed refs are not processed.
659 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
660 struct btrfs_root *root, u64 bytenr,
661 u64 num_bytes, u64 *refs, u64 *flags)
663 struct btrfs_delayed_ref_head *head;
664 struct btrfs_delayed_ref_root *delayed_refs;
665 struct btrfs_path *path;
666 struct btrfs_extent_item *ei;
667 struct extent_buffer *leaf;
668 struct btrfs_key key;
674 path = btrfs_alloc_path();
678 key.objectid = bytenr;
679 key.type = BTRFS_EXTENT_ITEM_KEY;
680 key.offset = num_bytes;
682 path->skip_locking = 1;
683 path->search_commit_root = 1;
686 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
692 leaf = path->nodes[0];
693 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
694 if (item_size >= sizeof(*ei)) {
695 ei = btrfs_item_ptr(leaf, path->slots[0],
696 struct btrfs_extent_item);
697 num_refs = btrfs_extent_refs(leaf, ei);
698 extent_flags = btrfs_extent_flags(leaf, ei);
700 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
701 struct btrfs_extent_item_v0 *ei0;
702 BUG_ON(item_size != sizeof(*ei0));
703 ei0 = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_extent_item_v0);
705 num_refs = btrfs_extent_refs_v0(leaf, ei0);
706 /* FIXME: this isn't correct for data */
707 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
712 BUG_ON(num_refs == 0);
722 delayed_refs = &trans->transaction->delayed_refs;
723 spin_lock(&delayed_refs->lock);
724 head = btrfs_find_delayed_ref_head(trans, bytenr);
726 if (!mutex_trylock(&head->mutex)) {
727 atomic_inc(&head->node.refs);
728 spin_unlock(&delayed_refs->lock);
730 btrfs_release_path(root->fs_info->extent_root, path);
732 mutex_lock(&head->mutex);
733 mutex_unlock(&head->mutex);
734 btrfs_put_delayed_ref(&head->node);
737 if (head->extent_op && head->extent_op->update_flags)
738 extent_flags |= head->extent_op->flags_to_set;
740 BUG_ON(num_refs == 0);
742 num_refs += head->node.ref_mod;
743 mutex_unlock(&head->mutex);
745 spin_unlock(&delayed_refs->lock);
747 WARN_ON(num_refs == 0);
751 *flags = extent_flags;
753 btrfs_free_path(path);
758 * Back reference rules. Back refs have three main goals:
760 * 1) differentiate between all holders of references to an extent so that
761 * when a reference is dropped we can make sure it was a valid reference
762 * before freeing the extent.
764 * 2) Provide enough information to quickly find the holders of an extent
765 * if we notice a given block is corrupted or bad.
767 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
768 * maintenance. This is actually the same as #2, but with a slightly
769 * different use case.
771 * There are two kinds of back refs. The implicit back refs is optimized
772 * for pointers in non-shared tree blocks. For a given pointer in a block,
773 * back refs of this kind provide information about the block's owner tree
774 * and the pointer's key. These information allow us to find the block by
775 * b-tree searching. The full back refs is for pointers in tree blocks not
776 * referenced by their owner trees. The location of tree block is recorded
777 * in the back refs. Actually the full back refs is generic, and can be
778 * used in all cases the implicit back refs is used. The major shortcoming
779 * of the full back refs is its overhead. Every time a tree block gets
780 * COWed, we have to update back refs entry for all pointers in it.
782 * For a newly allocated tree block, we use implicit back refs for
783 * pointers in it. This means most tree related operations only involve
784 * implicit back refs. For a tree block created in old transaction, the
785 * only way to drop a reference to it is COW it. So we can detect the
786 * event that tree block loses its owner tree's reference and do the
787 * back refs conversion.
789 * When a tree block is COW'd through a tree, there are four cases:
791 * The reference count of the block is one and the tree is the block's
792 * owner tree. Nothing to do in this case.
794 * The reference count of the block is one and the tree is not the
795 * block's owner tree. In this case, full back refs is used for pointers
796 * in the block. Remove these full back refs, add implicit back refs for
797 * every pointers in the new block.
799 * The reference count of the block is greater than one and the tree is
800 * the block's owner tree. In this case, implicit back refs is used for
801 * pointers in the block. Add full back refs for every pointers in the
802 * block, increase lower level extents' reference counts. The original
803 * implicit back refs are entailed to the new block.
805 * The reference count of the block is greater than one and the tree is
806 * not the block's owner tree. Add implicit back refs for every pointer in
807 * the new block, increase lower level extents' reference count.
809 * Back Reference Key composing:
811 * The key objectid corresponds to the first byte in the extent,
812 * The key type is used to differentiate between types of back refs.
813 * There are different meanings of the key offset for different types
816 * File extents can be referenced by:
818 * - multiple snapshots, subvolumes, or different generations in one subvol
819 * - different files inside a single subvolume
820 * - different offsets inside a file (bookend extents in file.c)
822 * The extent ref structure for the implicit back refs has fields for:
824 * - Objectid of the subvolume root
825 * - objectid of the file holding the reference
826 * - original offset in the file
827 * - how many bookend extents
829 * The key offset for the implicit back refs is hash of the first
832 * The extent ref structure for the full back refs has field for:
834 * - number of pointers in the tree leaf
836 * The key offset for the implicit back refs is the first byte of
839 * When a file extent is allocated, The implicit back refs is used.
840 * the fields are filled in:
842 * (root_key.objectid, inode objectid, offset in file, 1)
844 * When a file extent is removed file truncation, we find the
845 * corresponding implicit back refs and check the following fields:
847 * (btrfs_header_owner(leaf), inode objectid, offset in file)
849 * Btree extents can be referenced by:
851 * - Different subvolumes
853 * Both the implicit back refs and the full back refs for tree blocks
854 * only consist of key. The key offset for the implicit back refs is
855 * objectid of block's owner tree. The key offset for the full back refs
856 * is the first byte of parent block.
858 * When implicit back refs is used, information about the lowest key and
859 * level of the tree block are required. These information are stored in
860 * tree block info structure.
863 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
864 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
865 struct btrfs_root *root,
866 struct btrfs_path *path,
867 u64 owner, u32 extra_size)
869 struct btrfs_extent_item *item;
870 struct btrfs_extent_item_v0 *ei0;
871 struct btrfs_extent_ref_v0 *ref0;
872 struct btrfs_tree_block_info *bi;
873 struct extent_buffer *leaf;
874 struct btrfs_key key;
875 struct btrfs_key found_key;
876 u32 new_size = sizeof(*item);
880 leaf = path->nodes[0];
881 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
883 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884 ei0 = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_extent_item_v0);
886 refs = btrfs_extent_refs_v0(leaf, ei0);
888 if (owner == (u64)-1) {
890 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
891 ret = btrfs_next_leaf(root, path);
895 leaf = path->nodes[0];
897 btrfs_item_key_to_cpu(leaf, &found_key,
899 BUG_ON(key.objectid != found_key.objectid);
900 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
904 ref0 = btrfs_item_ptr(leaf, path->slots[0],
905 struct btrfs_extent_ref_v0);
906 owner = btrfs_ref_objectid_v0(leaf, ref0);
910 btrfs_release_path(root, path);
912 if (owner < BTRFS_FIRST_FREE_OBJECTID)
913 new_size += sizeof(*bi);
915 new_size -= sizeof(*ei0);
916 ret = btrfs_search_slot(trans, root, &key, path,
917 new_size + extra_size, 1);
922 ret = btrfs_extend_item(trans, root, path, new_size);
925 leaf = path->nodes[0];
926 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
927 btrfs_set_extent_refs(leaf, item, refs);
928 /* FIXME: get real generation */
929 btrfs_set_extent_generation(leaf, item, 0);
930 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
931 btrfs_set_extent_flags(leaf, item,
932 BTRFS_EXTENT_FLAG_TREE_BLOCK |
933 BTRFS_BLOCK_FLAG_FULL_BACKREF);
934 bi = (struct btrfs_tree_block_info *)(item + 1);
935 /* FIXME: get first key of the block */
936 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
937 btrfs_set_tree_block_level(leaf, bi, (int)owner);
939 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
941 btrfs_mark_buffer_dirty(leaf);
946 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
948 u32 high_crc = ~(u32)0;
949 u32 low_crc = ~(u32)0;
952 lenum = cpu_to_le64(root_objectid);
953 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
954 lenum = cpu_to_le64(owner);
955 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
956 lenum = cpu_to_le64(offset);
957 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
959 return ((u64)high_crc << 31) ^ (u64)low_crc;
962 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
963 struct btrfs_extent_data_ref *ref)
965 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
966 btrfs_extent_data_ref_objectid(leaf, ref),
967 btrfs_extent_data_ref_offset(leaf, ref));
970 static int match_extent_data_ref(struct extent_buffer *leaf,
971 struct btrfs_extent_data_ref *ref,
972 u64 root_objectid, u64 owner, u64 offset)
974 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
975 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
976 btrfs_extent_data_ref_offset(leaf, ref) != offset)
981 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
982 struct btrfs_root *root,
983 struct btrfs_path *path,
984 u64 bytenr, u64 parent,
986 u64 owner, u64 offset)
988 struct btrfs_key key;
989 struct btrfs_extent_data_ref *ref;
990 struct extent_buffer *leaf;
996 key.objectid = bytenr;
998 key.type = BTRFS_SHARED_DATA_REF_KEY;
1001 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1002 key.offset = hash_extent_data_ref(root_objectid,
1007 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1016 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1017 key.type = BTRFS_EXTENT_REF_V0_KEY;
1018 btrfs_release_path(root, path);
1019 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1030 leaf = path->nodes[0];
1031 nritems = btrfs_header_nritems(leaf);
1033 if (path->slots[0] >= nritems) {
1034 ret = btrfs_next_leaf(root, path);
1040 leaf = path->nodes[0];
1041 nritems = btrfs_header_nritems(leaf);
1045 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1046 if (key.objectid != bytenr ||
1047 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1050 ref = btrfs_item_ptr(leaf, path->slots[0],
1051 struct btrfs_extent_data_ref);
1053 if (match_extent_data_ref(leaf, ref, root_objectid,
1056 btrfs_release_path(root, path);
1068 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1069 struct btrfs_root *root,
1070 struct btrfs_path *path,
1071 u64 bytenr, u64 parent,
1072 u64 root_objectid, u64 owner,
1073 u64 offset, int refs_to_add)
1075 struct btrfs_key key;
1076 struct extent_buffer *leaf;
1081 key.objectid = bytenr;
1083 key.type = BTRFS_SHARED_DATA_REF_KEY;
1084 key.offset = parent;
1085 size = sizeof(struct btrfs_shared_data_ref);
1087 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1088 key.offset = hash_extent_data_ref(root_objectid,
1090 size = sizeof(struct btrfs_extent_data_ref);
1093 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1094 if (ret && ret != -EEXIST)
1097 leaf = path->nodes[0];
1099 struct btrfs_shared_data_ref *ref;
1100 ref = btrfs_item_ptr(leaf, path->slots[0],
1101 struct btrfs_shared_data_ref);
1103 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1105 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1106 num_refs += refs_to_add;
1107 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1110 struct btrfs_extent_data_ref *ref;
1111 while (ret == -EEXIST) {
1112 ref = btrfs_item_ptr(leaf, path->slots[0],
1113 struct btrfs_extent_data_ref);
1114 if (match_extent_data_ref(leaf, ref, root_objectid,
1117 btrfs_release_path(root, path);
1119 ret = btrfs_insert_empty_item(trans, root, path, &key,
1121 if (ret && ret != -EEXIST)
1124 leaf = path->nodes[0];
1126 ref = btrfs_item_ptr(leaf, path->slots[0],
1127 struct btrfs_extent_data_ref);
1129 btrfs_set_extent_data_ref_root(leaf, ref,
1131 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1132 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1133 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1135 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1136 num_refs += refs_to_add;
1137 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1140 btrfs_mark_buffer_dirty(leaf);
1143 btrfs_release_path(root, path);
1147 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1148 struct btrfs_root *root,
1149 struct btrfs_path *path,
1152 struct btrfs_key key;
1153 struct btrfs_extent_data_ref *ref1 = NULL;
1154 struct btrfs_shared_data_ref *ref2 = NULL;
1155 struct extent_buffer *leaf;
1159 leaf = path->nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1162 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1163 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1164 struct btrfs_extent_data_ref);
1165 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1166 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1167 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1168 struct btrfs_shared_data_ref);
1169 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1170 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1171 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1172 struct btrfs_extent_ref_v0 *ref0;
1173 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_extent_ref_v0);
1175 num_refs = btrfs_ref_count_v0(leaf, ref0);
1181 BUG_ON(num_refs < refs_to_drop);
1182 num_refs -= refs_to_drop;
1184 if (num_refs == 0) {
1185 ret = btrfs_del_item(trans, root, path);
1187 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1188 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1189 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1190 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1191 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1193 struct btrfs_extent_ref_v0 *ref0;
1194 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1195 struct btrfs_extent_ref_v0);
1196 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1199 btrfs_mark_buffer_dirty(leaf);
1204 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1205 struct btrfs_path *path,
1206 struct btrfs_extent_inline_ref *iref)
1208 struct btrfs_key key;
1209 struct extent_buffer *leaf;
1210 struct btrfs_extent_data_ref *ref1;
1211 struct btrfs_shared_data_ref *ref2;
1214 leaf = path->nodes[0];
1215 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1217 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1218 BTRFS_EXTENT_DATA_REF_KEY) {
1219 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1220 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1222 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1223 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1225 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1226 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1229 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1230 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1231 struct btrfs_shared_data_ref);
1232 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1235 struct btrfs_extent_ref_v0 *ref0;
1236 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1237 struct btrfs_extent_ref_v0);
1238 num_refs = btrfs_ref_count_v0(leaf, ref0);
1246 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1247 struct btrfs_root *root,
1248 struct btrfs_path *path,
1249 u64 bytenr, u64 parent,
1252 struct btrfs_key key;
1255 key.objectid = bytenr;
1257 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1258 key.offset = parent;
1260 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1261 key.offset = root_objectid;
1264 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1267 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1268 if (ret == -ENOENT && parent) {
1269 btrfs_release_path(root, path);
1270 key.type = BTRFS_EXTENT_REF_V0_KEY;
1271 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1279 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1280 struct btrfs_root *root,
1281 struct btrfs_path *path,
1282 u64 bytenr, u64 parent,
1285 struct btrfs_key key;
1288 key.objectid = bytenr;
1290 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1291 key.offset = parent;
1293 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1294 key.offset = root_objectid;
1297 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1298 btrfs_release_path(root, path);
1302 static inline int extent_ref_type(u64 parent, u64 owner)
1305 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1307 type = BTRFS_SHARED_BLOCK_REF_KEY;
1309 type = BTRFS_TREE_BLOCK_REF_KEY;
1312 type = BTRFS_SHARED_DATA_REF_KEY;
1314 type = BTRFS_EXTENT_DATA_REF_KEY;
1319 static int find_next_key(struct btrfs_path *path, int level,
1320 struct btrfs_key *key)
1323 for (; level < BTRFS_MAX_LEVEL; level++) {
1324 if (!path->nodes[level])
1326 if (path->slots[level] + 1 >=
1327 btrfs_header_nritems(path->nodes[level]))
1330 btrfs_item_key_to_cpu(path->nodes[level], key,
1331 path->slots[level] + 1);
1333 btrfs_node_key_to_cpu(path->nodes[level], key,
1334 path->slots[level] + 1);
1341 * look for inline back ref. if back ref is found, *ref_ret is set
1342 * to the address of inline back ref, and 0 is returned.
1344 * if back ref isn't found, *ref_ret is set to the address where it
1345 * should be inserted, and -ENOENT is returned.
1347 * if insert is true and there are too many inline back refs, the path
1348 * points to the extent item, and -EAGAIN is returned.
1350 * NOTE: inline back refs are ordered in the same way that back ref
1351 * items in the tree are ordered.
1353 static noinline_for_stack
1354 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root,
1356 struct btrfs_path *path,
1357 struct btrfs_extent_inline_ref **ref_ret,
1358 u64 bytenr, u64 num_bytes,
1359 u64 parent, u64 root_objectid,
1360 u64 owner, u64 offset, int insert)
1362 struct btrfs_key key;
1363 struct extent_buffer *leaf;
1364 struct btrfs_extent_item *ei;
1365 struct btrfs_extent_inline_ref *iref;
1376 key.objectid = bytenr;
1377 key.type = BTRFS_EXTENT_ITEM_KEY;
1378 key.offset = num_bytes;
1380 want = extent_ref_type(parent, owner);
1382 extra_size = btrfs_extent_inline_ref_size(want);
1383 path->keep_locks = 1;
1386 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1393 leaf = path->nodes[0];
1394 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1396 if (item_size < sizeof(*ei)) {
1401 ret = convert_extent_item_v0(trans, root, path, owner,
1407 leaf = path->nodes[0];
1408 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1411 BUG_ON(item_size < sizeof(*ei));
1413 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1414 flags = btrfs_extent_flags(leaf, ei);
1416 ptr = (unsigned long)(ei + 1);
1417 end = (unsigned long)ei + item_size;
1419 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1420 ptr += sizeof(struct btrfs_tree_block_info);
1423 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1432 iref = (struct btrfs_extent_inline_ref *)ptr;
1433 type = btrfs_extent_inline_ref_type(leaf, iref);
1437 ptr += btrfs_extent_inline_ref_size(type);
1441 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1442 struct btrfs_extent_data_ref *dref;
1443 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1444 if (match_extent_data_ref(leaf, dref, root_objectid,
1449 if (hash_extent_data_ref_item(leaf, dref) <
1450 hash_extent_data_ref(root_objectid, owner, offset))
1454 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1456 if (parent == ref_offset) {
1460 if (ref_offset < parent)
1463 if (root_objectid == ref_offset) {
1467 if (ref_offset < root_objectid)
1471 ptr += btrfs_extent_inline_ref_size(type);
1473 if (err == -ENOENT && insert) {
1474 if (item_size + extra_size >=
1475 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1480 * To add new inline back ref, we have to make sure
1481 * there is no corresponding back ref item.
1482 * For simplicity, we just do not add new inline back
1483 * ref if there is any kind of item for this block
1485 if (find_next_key(path, 0, &key) == 0 &&
1486 key.objectid == bytenr &&
1487 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1492 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1495 path->keep_locks = 0;
1496 btrfs_unlock_up_safe(path, 1);
1502 * helper to add new inline back ref
1504 static noinline_for_stack
1505 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root,
1507 struct btrfs_path *path,
1508 struct btrfs_extent_inline_ref *iref,
1509 u64 parent, u64 root_objectid,
1510 u64 owner, u64 offset, int refs_to_add,
1511 struct btrfs_delayed_extent_op *extent_op)
1513 struct extent_buffer *leaf;
1514 struct btrfs_extent_item *ei;
1517 unsigned long item_offset;
1523 leaf = path->nodes[0];
1524 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1525 item_offset = (unsigned long)iref - (unsigned long)ei;
1527 type = extent_ref_type(parent, owner);
1528 size = btrfs_extent_inline_ref_size(type);
1530 ret = btrfs_extend_item(trans, root, path, size);
1533 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1534 refs = btrfs_extent_refs(leaf, ei);
1535 refs += refs_to_add;
1536 btrfs_set_extent_refs(leaf, ei, refs);
1538 __run_delayed_extent_op(extent_op, leaf, ei);
1540 ptr = (unsigned long)ei + item_offset;
1541 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1542 if (ptr < end - size)
1543 memmove_extent_buffer(leaf, ptr + size, ptr,
1546 iref = (struct btrfs_extent_inline_ref *)ptr;
1547 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1548 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1549 struct btrfs_extent_data_ref *dref;
1550 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1551 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1552 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1553 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1554 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1555 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1556 struct btrfs_shared_data_ref *sref;
1557 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1558 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1559 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1560 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1561 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1563 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1565 btrfs_mark_buffer_dirty(leaf);
1569 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1570 struct btrfs_root *root,
1571 struct btrfs_path *path,
1572 struct btrfs_extent_inline_ref **ref_ret,
1573 u64 bytenr, u64 num_bytes, u64 parent,
1574 u64 root_objectid, u64 owner, u64 offset)
1578 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1579 bytenr, num_bytes, parent,
1580 root_objectid, owner, offset, 0);
1584 btrfs_release_path(root, path);
1587 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1588 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1591 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1592 root_objectid, owner, offset);
1598 * helper to update/remove inline back ref
1600 static noinline_for_stack
1601 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1602 struct btrfs_root *root,
1603 struct btrfs_path *path,
1604 struct btrfs_extent_inline_ref *iref,
1606 struct btrfs_delayed_extent_op *extent_op)
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1610 struct btrfs_extent_data_ref *dref = NULL;
1611 struct btrfs_shared_data_ref *sref = NULL;
1620 leaf = path->nodes[0];
1621 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622 refs = btrfs_extent_refs(leaf, ei);
1623 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1624 refs += refs_to_mod;
1625 btrfs_set_extent_refs(leaf, ei, refs);
1627 __run_delayed_extent_op(extent_op, leaf, ei);
1629 type = btrfs_extent_inline_ref_type(leaf, iref);
1631 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1632 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1633 refs = btrfs_extent_data_ref_count(leaf, dref);
1634 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1635 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1636 refs = btrfs_shared_data_ref_count(leaf, sref);
1639 BUG_ON(refs_to_mod != -1);
1642 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1643 refs += refs_to_mod;
1646 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1649 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1651 size = btrfs_extent_inline_ref_size(type);
1652 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1653 ptr = (unsigned long)iref;
1654 end = (unsigned long)ei + item_size;
1655 if (ptr + size < end)
1656 memmove_extent_buffer(leaf, ptr, ptr + size,
1659 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1662 btrfs_mark_buffer_dirty(leaf);
1666 static noinline_for_stack
1667 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1668 struct btrfs_root *root,
1669 struct btrfs_path *path,
1670 u64 bytenr, u64 num_bytes, u64 parent,
1671 u64 root_objectid, u64 owner,
1672 u64 offset, int refs_to_add,
1673 struct btrfs_delayed_extent_op *extent_op)
1675 struct btrfs_extent_inline_ref *iref;
1678 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1679 bytenr, num_bytes, parent,
1680 root_objectid, owner, offset, 1);
1682 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1683 ret = update_inline_extent_backref(trans, root, path, iref,
1684 refs_to_add, extent_op);
1685 } else if (ret == -ENOENT) {
1686 ret = setup_inline_extent_backref(trans, root, path, iref,
1687 parent, root_objectid,
1688 owner, offset, refs_to_add,
1694 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct btrfs_path *path,
1697 u64 bytenr, u64 parent, u64 root_objectid,
1698 u64 owner, u64 offset, int refs_to_add)
1701 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1702 BUG_ON(refs_to_add != 1);
1703 ret = insert_tree_block_ref(trans, root, path, bytenr,
1704 parent, root_objectid);
1706 ret = insert_extent_data_ref(trans, root, path, bytenr,
1707 parent, root_objectid,
1708 owner, offset, refs_to_add);
1713 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1714 struct btrfs_root *root,
1715 struct btrfs_path *path,
1716 struct btrfs_extent_inline_ref *iref,
1717 int refs_to_drop, int is_data)
1721 BUG_ON(!is_data && refs_to_drop != 1);
1723 ret = update_inline_extent_backref(trans, root, path, iref,
1724 -refs_to_drop, NULL);
1725 } else if (is_data) {
1726 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1728 ret = btrfs_del_item(trans, root, path);
1733 static void btrfs_issue_discard(struct block_device *bdev,
1736 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1737 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1740 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1744 u64 map_length = num_bytes;
1745 struct btrfs_multi_bio *multi = NULL;
1747 if (!btrfs_test_opt(root, DISCARD))
1750 /* Tell the block device(s) that the sectors can be discarded */
1751 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1752 bytenr, &map_length, &multi, 0);
1754 struct btrfs_bio_stripe *stripe = multi->stripes;
1757 if (map_length > num_bytes)
1758 map_length = num_bytes;
1760 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1761 btrfs_issue_discard(stripe->dev->bdev,
1771 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1772 struct btrfs_root *root,
1773 u64 bytenr, u64 num_bytes, u64 parent,
1774 u64 root_objectid, u64 owner, u64 offset)
1777 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1778 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1780 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1781 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1782 parent, root_objectid, (int)owner,
1783 BTRFS_ADD_DELAYED_REF, NULL);
1785 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1786 parent, root_objectid, owner, offset,
1787 BTRFS_ADD_DELAYED_REF, NULL);
1792 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 u64 bytenr, u64 num_bytes,
1795 u64 parent, u64 root_objectid,
1796 u64 owner, u64 offset, int refs_to_add,
1797 struct btrfs_delayed_extent_op *extent_op)
1799 struct btrfs_path *path;
1800 struct extent_buffer *leaf;
1801 struct btrfs_extent_item *item;
1806 path = btrfs_alloc_path();
1811 path->leave_spinning = 1;
1812 /* this will setup the path even if it fails to insert the back ref */
1813 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1814 path, bytenr, num_bytes, parent,
1815 root_objectid, owner, offset,
1816 refs_to_add, extent_op);
1820 if (ret != -EAGAIN) {
1825 leaf = path->nodes[0];
1826 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1827 refs = btrfs_extent_refs(leaf, item);
1828 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1830 __run_delayed_extent_op(extent_op, leaf, item);
1832 btrfs_mark_buffer_dirty(leaf);
1833 btrfs_release_path(root->fs_info->extent_root, path);
1836 path->leave_spinning = 1;
1838 /* now insert the actual backref */
1839 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1840 path, bytenr, parent, root_objectid,
1841 owner, offset, refs_to_add);
1844 btrfs_free_path(path);
1848 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1849 struct btrfs_root *root,
1850 struct btrfs_delayed_ref_node *node,
1851 struct btrfs_delayed_extent_op *extent_op,
1852 int insert_reserved)
1855 struct btrfs_delayed_data_ref *ref;
1856 struct btrfs_key ins;
1861 ins.objectid = node->bytenr;
1862 ins.offset = node->num_bytes;
1863 ins.type = BTRFS_EXTENT_ITEM_KEY;
1865 ref = btrfs_delayed_node_to_data_ref(node);
1866 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1867 parent = ref->parent;
1869 ref_root = ref->root;
1871 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1873 BUG_ON(extent_op->update_key);
1874 flags |= extent_op->flags_to_set;
1876 ret = alloc_reserved_file_extent(trans, root,
1877 parent, ref_root, flags,
1878 ref->objectid, ref->offset,
1879 &ins, node->ref_mod);
1880 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1881 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1882 node->num_bytes, parent,
1883 ref_root, ref->objectid,
1884 ref->offset, node->ref_mod,
1886 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1887 ret = __btrfs_free_extent(trans, root, node->bytenr,
1888 node->num_bytes, parent,
1889 ref_root, ref->objectid,
1890 ref->offset, node->ref_mod,
1898 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1899 struct extent_buffer *leaf,
1900 struct btrfs_extent_item *ei)
1902 u64 flags = btrfs_extent_flags(leaf, ei);
1903 if (extent_op->update_flags) {
1904 flags |= extent_op->flags_to_set;
1905 btrfs_set_extent_flags(leaf, ei, flags);
1908 if (extent_op->update_key) {
1909 struct btrfs_tree_block_info *bi;
1910 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1911 bi = (struct btrfs_tree_block_info *)(ei + 1);
1912 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1916 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1917 struct btrfs_root *root,
1918 struct btrfs_delayed_ref_node *node,
1919 struct btrfs_delayed_extent_op *extent_op)
1921 struct btrfs_key key;
1922 struct btrfs_path *path;
1923 struct btrfs_extent_item *ei;
1924 struct extent_buffer *leaf;
1929 path = btrfs_alloc_path();
1933 key.objectid = node->bytenr;
1934 key.type = BTRFS_EXTENT_ITEM_KEY;
1935 key.offset = node->num_bytes;
1938 path->leave_spinning = 1;
1939 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1950 leaf = path->nodes[0];
1951 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1952 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1953 if (item_size < sizeof(*ei)) {
1954 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1960 leaf = path->nodes[0];
1961 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1964 BUG_ON(item_size < sizeof(*ei));
1965 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1966 __run_delayed_extent_op(extent_op, leaf, ei);
1968 btrfs_mark_buffer_dirty(leaf);
1970 btrfs_free_path(path);
1974 static int run_delayed_tree_ref(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,
1978 int insert_reserved)
1981 struct btrfs_delayed_tree_ref *ref;
1982 struct btrfs_key ins;
1986 ins.objectid = node->bytenr;
1987 ins.offset = node->num_bytes;
1988 ins.type = BTRFS_EXTENT_ITEM_KEY;
1990 ref = btrfs_delayed_node_to_tree_ref(node);
1991 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1992 parent = ref->parent;
1994 ref_root = ref->root;
1996 BUG_ON(node->ref_mod != 1);
1997 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1998 BUG_ON(!extent_op || !extent_op->update_flags ||
1999 !extent_op->update_key);
2000 ret = alloc_reserved_tree_block(trans, root,
2002 extent_op->flags_to_set,
2005 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2006 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2007 node->num_bytes, parent, ref_root,
2008 ref->level, 0, 1, extent_op);
2009 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2010 ret = __btrfs_free_extent(trans, root, node->bytenr,
2011 node->num_bytes, parent, ref_root,
2012 ref->level, 0, 1, extent_op);
2019 /* helper function to actually process a single delayed ref entry */
2020 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2021 struct btrfs_root *root,
2022 struct btrfs_delayed_ref_node *node,
2023 struct btrfs_delayed_extent_op *extent_op,
2024 int insert_reserved)
2027 if (btrfs_delayed_ref_is_head(node)) {
2028 struct btrfs_delayed_ref_head *head;
2030 * we've hit the end of the chain and we were supposed
2031 * to insert this extent into the tree. But, it got
2032 * deleted before we ever needed to insert it, so all
2033 * we have to do is clean up the accounting
2036 head = btrfs_delayed_node_to_head(node);
2037 if (insert_reserved) {
2038 btrfs_pin_extent(root, node->bytenr,
2039 node->num_bytes, 1);
2040 if (head->is_data) {
2041 ret = btrfs_del_csums(trans, root,
2047 mutex_unlock(&head->mutex);
2051 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2052 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2053 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2055 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2056 node->type == BTRFS_SHARED_DATA_REF_KEY)
2057 ret = run_delayed_data_ref(trans, root, node, extent_op,
2064 static noinline struct btrfs_delayed_ref_node *
2065 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2067 struct rb_node *node;
2068 struct btrfs_delayed_ref_node *ref;
2069 int action = BTRFS_ADD_DELAYED_REF;
2072 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2073 * this prevents ref count from going down to zero when
2074 * there still are pending delayed ref.
2076 node = rb_prev(&head->node.rb_node);
2080 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2082 if (ref->bytenr != head->node.bytenr)
2084 if (ref->action == action)
2086 node = rb_prev(node);
2088 if (action == BTRFS_ADD_DELAYED_REF) {
2089 action = BTRFS_DROP_DELAYED_REF;
2095 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2096 struct btrfs_root *root,
2097 struct list_head *cluster)
2099 struct btrfs_delayed_ref_root *delayed_refs;
2100 struct btrfs_delayed_ref_node *ref;
2101 struct btrfs_delayed_ref_head *locked_ref = NULL;
2102 struct btrfs_delayed_extent_op *extent_op;
2105 int must_insert_reserved = 0;
2107 delayed_refs = &trans->transaction->delayed_refs;
2110 /* pick a new head ref from the cluster list */
2111 if (list_empty(cluster))
2114 locked_ref = list_entry(cluster->next,
2115 struct btrfs_delayed_ref_head, cluster);
2117 /* grab the lock that says we are going to process
2118 * all the refs for this head */
2119 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2122 * we may have dropped the spin lock to get the head
2123 * mutex lock, and that might have given someone else
2124 * time to free the head. If that's true, it has been
2125 * removed from our list and we can move on.
2127 if (ret == -EAGAIN) {
2135 * record the must insert reserved flag before we
2136 * drop the spin lock.
2138 must_insert_reserved = locked_ref->must_insert_reserved;
2139 locked_ref->must_insert_reserved = 0;
2141 extent_op = locked_ref->extent_op;
2142 locked_ref->extent_op = NULL;
2145 * locked_ref is the head node, so we have to go one
2146 * node back for any delayed ref updates
2148 ref = select_delayed_ref(locked_ref);
2150 /* All delayed refs have been processed, Go ahead
2151 * and send the head node to run_one_delayed_ref,
2152 * so that any accounting fixes can happen
2154 ref = &locked_ref->node;
2156 if (extent_op && must_insert_reserved) {
2162 spin_unlock(&delayed_refs->lock);
2164 ret = run_delayed_extent_op(trans, root,
2170 spin_lock(&delayed_refs->lock);
2174 list_del_init(&locked_ref->cluster);
2179 rb_erase(&ref->rb_node, &delayed_refs->root);
2180 delayed_refs->num_entries--;
2182 spin_unlock(&delayed_refs->lock);
2184 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2185 must_insert_reserved);
2188 btrfs_put_delayed_ref(ref);
2193 spin_lock(&delayed_refs->lock);
2199 * this starts processing the delayed reference count updates and
2200 * extent insertions we have queued up so far. count can be
2201 * 0, which means to process everything in the tree at the start
2202 * of the run (but not newly added entries), or it can be some target
2203 * number you'd like to process.
2205 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2206 struct btrfs_root *root, unsigned long count)
2208 struct rb_node *node;
2209 struct btrfs_delayed_ref_root *delayed_refs;
2210 struct btrfs_delayed_ref_node *ref;
2211 struct list_head cluster;
2213 int run_all = count == (unsigned long)-1;
2216 if (root == root->fs_info->extent_root)
2217 root = root->fs_info->tree_root;
2219 delayed_refs = &trans->transaction->delayed_refs;
2220 INIT_LIST_HEAD(&cluster);
2222 spin_lock(&delayed_refs->lock);
2224 count = delayed_refs->num_entries * 2;
2228 if (!(run_all || run_most) &&
2229 delayed_refs->num_heads_ready < 64)
2233 * go find something we can process in the rbtree. We start at
2234 * the beginning of the tree, and then build a cluster
2235 * of refs to process starting at the first one we are able to
2238 ret = btrfs_find_ref_cluster(trans, &cluster,
2239 delayed_refs->run_delayed_start);
2243 ret = run_clustered_refs(trans, root, &cluster);
2246 count -= min_t(unsigned long, ret, count);
2253 node = rb_first(&delayed_refs->root);
2256 count = (unsigned long)-1;
2259 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2261 if (btrfs_delayed_ref_is_head(ref)) {
2262 struct btrfs_delayed_ref_head *head;
2264 head = btrfs_delayed_node_to_head(ref);
2265 atomic_inc(&ref->refs);
2267 spin_unlock(&delayed_refs->lock);
2268 mutex_lock(&head->mutex);
2269 mutex_unlock(&head->mutex);
2271 btrfs_put_delayed_ref(ref);
2275 node = rb_next(node);
2277 spin_unlock(&delayed_refs->lock);
2278 schedule_timeout(1);
2282 spin_unlock(&delayed_refs->lock);
2286 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2287 struct btrfs_root *root,
2288 u64 bytenr, u64 num_bytes, u64 flags,
2291 struct btrfs_delayed_extent_op *extent_op;
2294 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2298 extent_op->flags_to_set = flags;
2299 extent_op->update_flags = 1;
2300 extent_op->update_key = 0;
2301 extent_op->is_data = is_data ? 1 : 0;
2303 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2309 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2310 struct btrfs_root *root,
2311 struct btrfs_path *path,
2312 u64 objectid, u64 offset, u64 bytenr)
2314 struct btrfs_delayed_ref_head *head;
2315 struct btrfs_delayed_ref_node *ref;
2316 struct btrfs_delayed_data_ref *data_ref;
2317 struct btrfs_delayed_ref_root *delayed_refs;
2318 struct rb_node *node;
2322 delayed_refs = &trans->transaction->delayed_refs;
2323 spin_lock(&delayed_refs->lock);
2324 head = btrfs_find_delayed_ref_head(trans, bytenr);
2328 if (!mutex_trylock(&head->mutex)) {
2329 atomic_inc(&head->node.refs);
2330 spin_unlock(&delayed_refs->lock);
2332 btrfs_release_path(root->fs_info->extent_root, path);
2334 mutex_lock(&head->mutex);
2335 mutex_unlock(&head->mutex);
2336 btrfs_put_delayed_ref(&head->node);
2340 node = rb_prev(&head->node.rb_node);
2344 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2346 if (ref->bytenr != bytenr)
2350 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2353 data_ref = btrfs_delayed_node_to_data_ref(ref);
2355 node = rb_prev(node);
2357 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2358 if (ref->bytenr == bytenr)
2362 if (data_ref->root != root->root_key.objectid ||
2363 data_ref->objectid != objectid || data_ref->offset != offset)
2368 mutex_unlock(&head->mutex);
2370 spin_unlock(&delayed_refs->lock);
2374 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2375 struct btrfs_root *root,
2376 struct btrfs_path *path,
2377 u64 objectid, u64 offset, u64 bytenr)
2379 struct btrfs_root *extent_root = root->fs_info->extent_root;
2380 struct extent_buffer *leaf;
2381 struct btrfs_extent_data_ref *ref;
2382 struct btrfs_extent_inline_ref *iref;
2383 struct btrfs_extent_item *ei;
2384 struct btrfs_key key;
2388 key.objectid = bytenr;
2389 key.offset = (u64)-1;
2390 key.type = BTRFS_EXTENT_ITEM_KEY;
2392 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2398 if (path->slots[0] == 0)
2402 leaf = path->nodes[0];
2403 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2405 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2409 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2410 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2411 if (item_size < sizeof(*ei)) {
2412 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2416 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2418 if (item_size != sizeof(*ei) +
2419 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2422 if (btrfs_extent_generation(leaf, ei) <=
2423 btrfs_root_last_snapshot(&root->root_item))
2426 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2427 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2428 BTRFS_EXTENT_DATA_REF_KEY)
2431 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2432 if (btrfs_extent_refs(leaf, ei) !=
2433 btrfs_extent_data_ref_count(leaf, ref) ||
2434 btrfs_extent_data_ref_root(leaf, ref) !=
2435 root->root_key.objectid ||
2436 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2437 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2445 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2446 struct btrfs_root *root,
2447 u64 objectid, u64 offset, u64 bytenr)
2449 struct btrfs_path *path;
2453 path = btrfs_alloc_path();
2458 ret = check_committed_ref(trans, root, path, objectid,
2460 if (ret && ret != -ENOENT)
2463 ret2 = check_delayed_ref(trans, root, path, objectid,
2465 } while (ret2 == -EAGAIN);
2467 if (ret2 && ret2 != -ENOENT) {
2472 if (ret != -ENOENT || ret2 != -ENOENT)
2475 btrfs_free_path(path);
2476 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2482 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2483 struct extent_buffer *buf, u32 nr_extents)
2485 struct btrfs_key key;
2486 struct btrfs_file_extent_item *fi;
2494 if (!root->ref_cows)
2497 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2499 root_gen = root->root_key.offset;
2502 root_gen = trans->transid - 1;
2505 level = btrfs_header_level(buf);
2506 nritems = btrfs_header_nritems(buf);
2509 struct btrfs_leaf_ref *ref;
2510 struct btrfs_extent_info *info;
2512 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2518 ref->root_gen = root_gen;
2519 ref->bytenr = buf->start;
2520 ref->owner = btrfs_header_owner(buf);
2521 ref->generation = btrfs_header_generation(buf);
2522 ref->nritems = nr_extents;
2523 info = ref->extents;
2525 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2527 btrfs_item_key_to_cpu(buf, &key, i);
2528 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2530 fi = btrfs_item_ptr(buf, i,
2531 struct btrfs_file_extent_item);
2532 if (btrfs_file_extent_type(buf, fi) ==
2533 BTRFS_FILE_EXTENT_INLINE)
2535 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2536 if (disk_bytenr == 0)
2539 info->bytenr = disk_bytenr;
2541 btrfs_file_extent_disk_num_bytes(buf, fi);
2542 info->objectid = key.objectid;
2543 info->offset = key.offset;
2547 ret = btrfs_add_leaf_ref(root, ref, shared);
2548 if (ret == -EEXIST && shared) {
2549 struct btrfs_leaf_ref *old;
2550 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2552 btrfs_remove_leaf_ref(root, old);
2553 btrfs_free_leaf_ref(root, old);
2554 ret = btrfs_add_leaf_ref(root, ref, shared);
2557 btrfs_free_leaf_ref(root, ref);
2563 /* when a block goes through cow, we update the reference counts of
2564 * everything that block points to. The internal pointers of the block
2565 * can be in just about any order, and it is likely to have clusters of
2566 * things that are close together and clusters of things that are not.
2568 * To help reduce the seeks that come with updating all of these reference
2569 * counts, sort them by byte number before actual updates are done.
2571 * struct refsort is used to match byte number to slot in the btree block.
2572 * we sort based on the byte number and then use the slot to actually
2575 * struct refsort is smaller than strcut btrfs_item and smaller than
2576 * struct btrfs_key_ptr. Since we're currently limited to the page size
2577 * for a btree block, there's no way for a kmalloc of refsorts for a
2578 * single node to be bigger than a page.
2586 * for passing into sort()
2588 static int refsort_cmp(const void *a_void, const void *b_void)
2590 const struct refsort *a = a_void;
2591 const struct refsort *b = b_void;
2593 if (a->bytenr < b->bytenr)
2595 if (a->bytenr > b->bytenr)
2601 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2602 struct btrfs_root *root,
2603 struct extent_buffer *buf,
2604 int full_backref, int inc)
2611 struct btrfs_key key;
2612 struct btrfs_file_extent_item *fi;
2616 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2617 u64, u64, u64, u64, u64, u64);
2619 ref_root = btrfs_header_owner(buf);
2620 nritems = btrfs_header_nritems(buf);
2621 level = btrfs_header_level(buf);
2623 if (!root->ref_cows && level == 0)
2627 process_func = btrfs_inc_extent_ref;
2629 process_func = btrfs_free_extent;
2632 parent = buf->start;
2636 for (i = 0; i < nritems; i++) {
2638 btrfs_item_key_to_cpu(buf, &key, i);
2639 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2641 fi = btrfs_item_ptr(buf, i,
2642 struct btrfs_file_extent_item);
2643 if (btrfs_file_extent_type(buf, fi) ==
2644 BTRFS_FILE_EXTENT_INLINE)
2646 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2650 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2651 key.offset -= btrfs_file_extent_offset(buf, fi);
2652 ret = process_func(trans, root, bytenr, num_bytes,
2653 parent, ref_root, key.objectid,
2658 bytenr = btrfs_node_blockptr(buf, i);
2659 num_bytes = btrfs_level_size(root, level - 1);
2660 ret = process_func(trans, root, bytenr, num_bytes,
2661 parent, ref_root, level - 1, 0);
2672 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2673 struct extent_buffer *buf, int full_backref)
2675 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2678 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2679 struct extent_buffer *buf, int full_backref)
2681 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2684 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2685 struct btrfs_root *root,
2686 struct btrfs_path *path,
2687 struct btrfs_block_group_cache *cache)
2690 struct btrfs_root *extent_root = root->fs_info->extent_root;
2692 struct extent_buffer *leaf;
2694 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2699 leaf = path->nodes[0];
2700 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2701 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2702 btrfs_mark_buffer_dirty(leaf);
2703 btrfs_release_path(extent_root, path);
2711 static struct btrfs_block_group_cache *
2712 next_block_group(struct btrfs_root *root,
2713 struct btrfs_block_group_cache *cache)
2715 struct rb_node *node;
2716 spin_lock(&root->fs_info->block_group_cache_lock);
2717 node = rb_next(&cache->cache_node);
2718 btrfs_put_block_group(cache);
2720 cache = rb_entry(node, struct btrfs_block_group_cache,
2722 btrfs_get_block_group(cache);
2725 spin_unlock(&root->fs_info->block_group_cache_lock);
2729 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2730 struct btrfs_trans_handle *trans,
2731 struct btrfs_path *path)
2733 struct btrfs_root *root = block_group->fs_info->tree_root;
2734 struct inode *inode = NULL;
2741 * If this block group is smaller than 100 megs don't bother caching the
2744 if (block_group->key.offset < (100 * 1024 * 1024)) {
2745 spin_lock(&block_group->lock);
2746 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2747 spin_unlock(&block_group->lock);
2752 inode = lookup_free_space_inode(root, block_group, path);
2753 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2754 ret = PTR_ERR(inode);
2755 btrfs_release_path(root, path);
2759 if (IS_ERR(inode)) {
2763 if (block_group->ro)
2766 ret = create_free_space_inode(root, trans, block_group, path);
2773 * We want to set the generation to 0, that way if anything goes wrong
2774 * from here on out we know not to trust this cache when we load up next
2777 BTRFS_I(inode)->generation = 0;
2778 ret = btrfs_update_inode(trans, root, inode);
2781 if (i_size_read(inode) > 0) {
2782 ret = btrfs_truncate_free_space_cache(root, trans, path,
2788 spin_lock(&block_group->lock);
2789 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2790 spin_unlock(&block_group->lock);
2793 spin_unlock(&block_group->lock);
2795 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2800 * Just to make absolutely sure we have enough space, we're going to
2801 * preallocate 12 pages worth of space for each block group. In
2802 * practice we ought to use at most 8, but we need extra space so we can
2803 * add our header and have a terminator between the extents and the
2807 num_pages *= PAGE_CACHE_SIZE;
2809 ret = btrfs_check_data_free_space(inode, num_pages);
2813 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2814 num_pages, num_pages,
2816 btrfs_free_reserved_data_space(inode, num_pages);
2820 btrfs_release_path(root, path);
2822 spin_lock(&block_group->lock);
2824 block_group->disk_cache_state = BTRFS_DC_ERROR;
2826 block_group->disk_cache_state = BTRFS_DC_SETUP;
2827 spin_unlock(&block_group->lock);
2832 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2833 struct btrfs_root *root)
2835 struct btrfs_block_group_cache *cache;
2837 struct btrfs_path *path;
2840 path = btrfs_alloc_path();
2846 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2848 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2850 cache = next_block_group(root, cache);
2858 err = cache_save_setup(cache, trans, path);
2859 last = cache->key.objectid + cache->key.offset;
2860 btrfs_put_block_group(cache);
2865 err = btrfs_run_delayed_refs(trans, root,
2870 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2872 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2873 btrfs_put_block_group(cache);
2879 cache = next_block_group(root, cache);
2888 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2889 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2891 last = cache->key.objectid + cache->key.offset;
2893 err = write_one_cache_group(trans, root, path, cache);
2895 btrfs_put_block_group(cache);
2900 * I don't think this is needed since we're just marking our
2901 * preallocated extent as written, but just in case it can't
2905 err = btrfs_run_delayed_refs(trans, root,
2910 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2913 * Really this shouldn't happen, but it could if we
2914 * couldn't write the entire preallocated extent and
2915 * splitting the extent resulted in a new block.
2918 btrfs_put_block_group(cache);
2921 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2923 cache = next_block_group(root, cache);
2932 btrfs_write_out_cache(root, trans, cache, path);
2935 * If we didn't have an error then the cache state is still
2936 * NEED_WRITE, so we can set it to WRITTEN.
2938 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2939 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2940 last = cache->key.objectid + cache->key.offset;
2941 btrfs_put_block_group(cache);
2944 btrfs_free_path(path);
2948 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2950 struct btrfs_block_group_cache *block_group;
2953 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2954 if (!block_group || block_group->ro)
2957 btrfs_put_block_group(block_group);
2961 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2962 u64 total_bytes, u64 bytes_used,
2963 struct btrfs_space_info **space_info)
2965 struct btrfs_space_info *found;
2969 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2970 BTRFS_BLOCK_GROUP_RAID10))
2975 found = __find_space_info(info, flags);
2977 spin_lock(&found->lock);
2978 found->total_bytes += total_bytes;
2979 found->disk_total += total_bytes * factor;
2980 found->bytes_used += bytes_used;
2981 found->disk_used += bytes_used * factor;
2983 spin_unlock(&found->lock);
2984 *space_info = found;
2987 found = kzalloc(sizeof(*found), GFP_NOFS);
2991 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2992 INIT_LIST_HEAD(&found->block_groups[i]);
2993 init_rwsem(&found->groups_sem);
2994 spin_lock_init(&found->lock);
2995 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2996 BTRFS_BLOCK_GROUP_SYSTEM |
2997 BTRFS_BLOCK_GROUP_METADATA);
2998 found->total_bytes = total_bytes;
2999 found->disk_total = total_bytes * factor;
3000 found->bytes_used = bytes_used;
3001 found->disk_used = bytes_used * factor;
3002 found->bytes_pinned = 0;
3003 found->bytes_reserved = 0;
3004 found->bytes_readonly = 0;
3005 found->bytes_may_use = 0;
3007 found->force_alloc = 0;
3008 *space_info = found;
3009 list_add_rcu(&found->list, &info->space_info);
3010 atomic_set(&found->caching_threads, 0);
3014 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3016 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3017 BTRFS_BLOCK_GROUP_RAID1 |
3018 BTRFS_BLOCK_GROUP_RAID10 |
3019 BTRFS_BLOCK_GROUP_DUP);
3021 if (flags & BTRFS_BLOCK_GROUP_DATA)
3022 fs_info->avail_data_alloc_bits |= extra_flags;
3023 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3024 fs_info->avail_metadata_alloc_bits |= extra_flags;
3025 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3026 fs_info->avail_system_alloc_bits |= extra_flags;
3030 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3032 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3034 if (num_devices == 1)
3035 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3036 if (num_devices < 4)
3037 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3039 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3040 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3041 BTRFS_BLOCK_GROUP_RAID10))) {
3042 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3045 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3046 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3047 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3050 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3051 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3052 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3053 (flags & BTRFS_BLOCK_GROUP_DUP)))
3054 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3058 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3060 if (flags & BTRFS_BLOCK_GROUP_DATA)
3061 flags |= root->fs_info->avail_data_alloc_bits &
3062 root->fs_info->data_alloc_profile;
3063 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3064 flags |= root->fs_info->avail_system_alloc_bits &
3065 root->fs_info->system_alloc_profile;
3066 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3067 flags |= root->fs_info->avail_metadata_alloc_bits &
3068 root->fs_info->metadata_alloc_profile;
3069 return btrfs_reduce_alloc_profile(root, flags);
3072 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3077 flags = BTRFS_BLOCK_GROUP_DATA;
3078 else if (root == root->fs_info->chunk_root)
3079 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3081 flags = BTRFS_BLOCK_GROUP_METADATA;
3083 return get_alloc_profile(root, flags);
3086 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3088 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3089 BTRFS_BLOCK_GROUP_DATA);
3093 * This will check the space that the inode allocates from to make sure we have
3094 * enough space for bytes.
3096 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3098 struct btrfs_space_info *data_sinfo;
3099 struct btrfs_root *root = BTRFS_I(inode)->root;
3101 int ret = 0, committed = 0, alloc_chunk = 1;
3103 /* make sure bytes are sectorsize aligned */
3104 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3106 if (root == root->fs_info->tree_root) {
3111 data_sinfo = BTRFS_I(inode)->space_info;
3116 /* make sure we have enough space to handle the data first */
3117 spin_lock(&data_sinfo->lock);
3118 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3119 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3120 data_sinfo->bytes_may_use;
3122 if (used + bytes > data_sinfo->total_bytes) {
3123 struct btrfs_trans_handle *trans;
3126 * if we don't have enough free bytes in this space then we need
3127 * to alloc a new chunk.
3129 if (!data_sinfo->full && alloc_chunk) {
3132 data_sinfo->force_alloc = 1;
3133 spin_unlock(&data_sinfo->lock);
3135 alloc_target = btrfs_get_alloc_profile(root, 1);
3136 trans = btrfs_join_transaction(root, 1);
3138 return PTR_ERR(trans);
3140 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3141 bytes + 2 * 1024 * 1024,
3143 btrfs_end_transaction(trans, root);
3148 btrfs_set_inode_space_info(root, inode);
3149 data_sinfo = BTRFS_I(inode)->space_info;
3153 spin_unlock(&data_sinfo->lock);
3155 /* commit the current transaction and try again */
3156 if (!committed && !root->fs_info->open_ioctl_trans) {
3158 trans = btrfs_join_transaction(root, 1);
3160 return PTR_ERR(trans);
3161 ret = btrfs_commit_transaction(trans, root);
3167 #if 0 /* I hope we never need this code again, just in case */
3168 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3169 "%llu bytes_reserved, " "%llu bytes_pinned, "
3170 "%llu bytes_readonly, %llu may use %llu total\n",
3171 (unsigned long long)bytes,
3172 (unsigned long long)data_sinfo->bytes_used,
3173 (unsigned long long)data_sinfo->bytes_reserved,
3174 (unsigned long long)data_sinfo->bytes_pinned,
3175 (unsigned long long)data_sinfo->bytes_readonly,
3176 (unsigned long long)data_sinfo->bytes_may_use,
3177 (unsigned long long)data_sinfo->total_bytes);
3181 data_sinfo->bytes_may_use += bytes;
3182 BTRFS_I(inode)->reserved_bytes += bytes;
3183 spin_unlock(&data_sinfo->lock);
3189 * called when we are clearing an delalloc extent from the
3190 * inode's io_tree or there was an error for whatever reason
3191 * after calling btrfs_check_data_free_space
3193 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3195 struct btrfs_root *root = BTRFS_I(inode)->root;
3196 struct btrfs_space_info *data_sinfo;
3198 /* make sure bytes are sectorsize aligned */
3199 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3201 data_sinfo = BTRFS_I(inode)->space_info;
3202 spin_lock(&data_sinfo->lock);
3203 data_sinfo->bytes_may_use -= bytes;
3204 BTRFS_I(inode)->reserved_bytes -= bytes;
3205 spin_unlock(&data_sinfo->lock);
3208 static void force_metadata_allocation(struct btrfs_fs_info *info)
3210 struct list_head *head = &info->space_info;
3211 struct btrfs_space_info *found;
3214 list_for_each_entry_rcu(found, head, list) {
3215 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3216 found->force_alloc = 1;
3221 static int should_alloc_chunk(struct btrfs_space_info *sinfo, u64 alloc_bytes)
3223 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3225 if (sinfo->bytes_used + sinfo->bytes_reserved +
3226 alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3229 if (sinfo->bytes_used + sinfo->bytes_reserved +
3230 alloc_bytes < div_factor(num_bytes, 8))
3233 if (num_bytes > 256 * 1024 * 1024 &&
3234 sinfo->bytes_used < div_factor(num_bytes, 3))
3240 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *extent_root, u64 alloc_bytes,
3242 u64 flags, int force)
3244 struct btrfs_space_info *space_info;
3245 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3248 mutex_lock(&fs_info->chunk_mutex);
3250 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3252 space_info = __find_space_info(extent_root->fs_info, flags);
3254 ret = update_space_info(extent_root->fs_info, flags,
3258 BUG_ON(!space_info);
3260 spin_lock(&space_info->lock);
3261 if (space_info->force_alloc)
3263 if (space_info->full) {
3264 spin_unlock(&space_info->lock);
3268 if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3269 spin_unlock(&space_info->lock);
3272 spin_unlock(&space_info->lock);
3275 * If we have mixed data/metadata chunks we want to make sure we keep
3276 * allocating mixed chunks instead of individual chunks.
3278 if (btrfs_mixed_space_info(space_info))
3279 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3282 * if we're doing a data chunk, go ahead and make sure that
3283 * we keep a reasonable number of metadata chunks allocated in the
3286 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3287 fs_info->data_chunk_allocations++;
3288 if (!(fs_info->data_chunk_allocations %
3289 fs_info->metadata_ratio))
3290 force_metadata_allocation(fs_info);
3293 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3294 spin_lock(&space_info->lock);
3296 space_info->full = 1;
3299 space_info->force_alloc = 0;
3300 spin_unlock(&space_info->lock);
3302 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3307 * shrink metadata reservation for delalloc
3309 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3310 struct btrfs_root *root, u64 to_reclaim, int sync)
3312 struct btrfs_block_rsv *block_rsv;
3313 struct btrfs_space_info *space_info;
3321 block_rsv = &root->fs_info->delalloc_block_rsv;
3322 space_info = block_rsv->space_info;
3323 spin_lock(&space_info->lock);
3324 reserved = space_info->bytes_reserved;
3325 spin_unlock(&space_info->lock);
3330 max_reclaim = min(reserved, to_reclaim);
3333 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0, sync);
3338 __set_current_state(TASK_INTERRUPTIBLE);
3339 schedule_timeout(pause);
3341 if (pause > HZ / 10)
3348 spin_lock(&space_info->lock);
3349 if (reserved > space_info->bytes_reserved)
3350 reclaimed += reserved - space_info->bytes_reserved;
3351 reserved = space_info->bytes_reserved;
3352 spin_unlock(&space_info->lock);
3354 if (reserved == 0 || reclaimed >= max_reclaim)
3357 if (trans && trans->transaction->blocked)
3360 return reclaimed >= to_reclaim;
3364 * Retries tells us how many times we've called reserve_metadata_bytes. The
3365 * idea is if this is the first call (retries == 0) then we will add to our
3366 * reserved count if we can't make the allocation in order to hold our place
3367 * while we go and try and free up space. That way for retries > 1 we don't try
3368 * and add space, we just check to see if the amount of unused space is >= the
3369 * total space, meaning that our reservation is valid.
3371 * However if we don't intend to retry this reservation, pass -1 as retries so
3372 * that it short circuits this logic.
3374 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3375 struct btrfs_root *root,
3376 struct btrfs_block_rsv *block_rsv,
3377 u64 orig_bytes, int flush)
3379 struct btrfs_space_info *space_info = block_rsv->space_info;
3381 u64 num_bytes = orig_bytes;
3384 bool reserved = false;
3385 bool committed = false;
3392 spin_lock(&space_info->lock);
3393 unused = space_info->bytes_used + space_info->bytes_reserved +
3394 space_info->bytes_pinned + space_info->bytes_readonly +
3395 space_info->bytes_may_use;
3398 * The idea here is that we've not already over-reserved the block group
3399 * then we can go ahead and save our reservation first and then start
3400 * flushing if we need to. Otherwise if we've already overcommitted
3401 * lets start flushing stuff first and then come back and try to make
3404 if (unused <= space_info->total_bytes) {
3405 unused -= space_info->total_bytes;
3406 if (unused >= num_bytes) {
3408 space_info->bytes_reserved += orig_bytes;
3412 * Ok set num_bytes to orig_bytes since we aren't
3413 * overocmmitted, this way we only try and reclaim what
3416 num_bytes = orig_bytes;
3420 * Ok we're over committed, set num_bytes to the overcommitted
3421 * amount plus the amount of bytes that we need for this
3424 num_bytes = unused - space_info->total_bytes +
3425 (orig_bytes * (retries + 1));
3429 * Couldn't make our reservation, save our place so while we're trying
3430 * to reclaim space we can actually use it instead of somebody else
3431 * stealing it from us.
3433 if (ret && !reserved) {
3434 space_info->bytes_reserved += orig_bytes;
3438 spin_unlock(&space_info->lock);
3447 * We do synchronous shrinking since we don't actually unreserve
3448 * metadata until after the IO is completed.
3450 ret = shrink_delalloc(trans, root, num_bytes, 1);
3457 * So if we were overcommitted it's possible that somebody else flushed
3458 * out enough space and we simply didn't have enough space to reclaim,
3459 * so go back around and try again.
3466 spin_lock(&space_info->lock);
3468 * Not enough space to be reclaimed, don't bother committing the
3471 if (space_info->bytes_pinned < orig_bytes)
3473 spin_unlock(&space_info->lock);
3478 if (trans || committed)
3482 trans = btrfs_join_transaction(root, 1);
3485 ret = btrfs_commit_transaction(trans, root);
3494 spin_lock(&space_info->lock);
3495 space_info->bytes_reserved -= orig_bytes;
3496 spin_unlock(&space_info->lock);
3502 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3503 struct btrfs_root *root)
3505 struct btrfs_block_rsv *block_rsv;
3507 block_rsv = trans->block_rsv;
3509 block_rsv = root->block_rsv;
3512 block_rsv = &root->fs_info->empty_block_rsv;
3517 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3521 spin_lock(&block_rsv->lock);
3522 if (block_rsv->reserved >= num_bytes) {
3523 block_rsv->reserved -= num_bytes;
3524 if (block_rsv->reserved < block_rsv->size)
3525 block_rsv->full = 0;
3528 spin_unlock(&block_rsv->lock);
3532 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3533 u64 num_bytes, int update_size)
3535 spin_lock(&block_rsv->lock);
3536 block_rsv->reserved += num_bytes;
3538 block_rsv->size += num_bytes;
3539 else if (block_rsv->reserved >= block_rsv->size)
3540 block_rsv->full = 1;
3541 spin_unlock(&block_rsv->lock);
3544 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3545 struct btrfs_block_rsv *dest, u64 num_bytes)
3547 struct btrfs_space_info *space_info = block_rsv->space_info;
3549 spin_lock(&block_rsv->lock);
3550 if (num_bytes == (u64)-1)
3551 num_bytes = block_rsv->size;
3552 block_rsv->size -= num_bytes;
3553 if (block_rsv->reserved >= block_rsv->size) {
3554 num_bytes = block_rsv->reserved - block_rsv->size;
3555 block_rsv->reserved = block_rsv->size;
3556 block_rsv->full = 1;
3560 spin_unlock(&block_rsv->lock);
3562 if (num_bytes > 0) {
3564 block_rsv_add_bytes(dest, num_bytes, 0);
3566 spin_lock(&space_info->lock);
3567 space_info->bytes_reserved -= num_bytes;
3568 spin_unlock(&space_info->lock);
3573 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3574 struct btrfs_block_rsv *dst, u64 num_bytes)
3578 ret = block_rsv_use_bytes(src, num_bytes);
3582 block_rsv_add_bytes(dst, num_bytes, 1);
3586 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3588 memset(rsv, 0, sizeof(*rsv));
3589 spin_lock_init(&rsv->lock);
3590 atomic_set(&rsv->usage, 1);
3592 INIT_LIST_HEAD(&rsv->list);
3595 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3597 struct btrfs_block_rsv *block_rsv;
3598 struct btrfs_fs_info *fs_info = root->fs_info;
3601 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3605 btrfs_init_block_rsv(block_rsv);
3607 alloc_target = btrfs_get_alloc_profile(root, 0);
3608 block_rsv->space_info = __find_space_info(fs_info,
3609 BTRFS_BLOCK_GROUP_METADATA);
3614 void btrfs_free_block_rsv(struct btrfs_root *root,
3615 struct btrfs_block_rsv *rsv)
3617 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3618 btrfs_block_rsv_release(root, rsv, (u64)-1);
3625 * make the block_rsv struct be able to capture freed space.
3626 * the captured space will re-add to the the block_rsv struct
3627 * after transaction commit
3629 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3630 struct btrfs_block_rsv *block_rsv)
3632 block_rsv->durable = 1;
3633 mutex_lock(&fs_info->durable_block_rsv_mutex);
3634 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3635 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3638 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3639 struct btrfs_root *root,
3640 struct btrfs_block_rsv *block_rsv,
3648 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3650 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3657 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3658 struct btrfs_root *root,
3659 struct btrfs_block_rsv *block_rsv,
3660 u64 min_reserved, int min_factor)
3663 int commit_trans = 0;
3669 spin_lock(&block_rsv->lock);
3671 num_bytes = div_factor(block_rsv->size, min_factor);
3672 if (min_reserved > num_bytes)
3673 num_bytes = min_reserved;
3675 if (block_rsv->reserved >= num_bytes) {
3678 num_bytes -= block_rsv->reserved;
3679 if (block_rsv->durable &&
3680 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3683 spin_unlock(&block_rsv->lock);
3687 if (block_rsv->refill_used) {
3688 ret = reserve_metadata_bytes(trans, root, block_rsv,
3691 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3700 trans = btrfs_join_transaction(root, 1);
3701 BUG_ON(IS_ERR(trans));
3702 ret = btrfs_commit_transaction(trans, root);
3707 printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3708 block_rsv->size, block_rsv->reserved,
3709 block_rsv->freed[0], block_rsv->freed[1]);
3714 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3715 struct btrfs_block_rsv *dst_rsv,
3718 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3721 void btrfs_block_rsv_release(struct btrfs_root *root,
3722 struct btrfs_block_rsv *block_rsv,
3725 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3726 if (global_rsv->full || global_rsv == block_rsv ||
3727 block_rsv->space_info != global_rsv->space_info)
3729 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3733 * helper to calculate size of global block reservation.
3734 * the desired value is sum of space used by extent tree,
3735 * checksum tree and root tree
3737 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3739 struct btrfs_space_info *sinfo;
3743 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3746 * per tree used space accounting can be inaccuracy, so we
3749 spin_lock(&fs_info->extent_root->accounting_lock);
3750 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3751 spin_unlock(&fs_info->extent_root->accounting_lock);
3753 spin_lock(&fs_info->csum_root->accounting_lock);
3754 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3755 spin_unlock(&fs_info->csum_root->accounting_lock);
3757 spin_lock(&fs_info->tree_root->accounting_lock);
3758 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3759 spin_unlock(&fs_info->tree_root->accounting_lock);
3761 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3762 spin_lock(&sinfo->lock);
3763 data_used = sinfo->bytes_used;
3764 spin_unlock(&sinfo->lock);
3766 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3767 spin_lock(&sinfo->lock);
3768 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3770 meta_used = sinfo->bytes_used;
3771 spin_unlock(&sinfo->lock);
3773 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3775 num_bytes += div64_u64(data_used + meta_used, 50);
3777 if (num_bytes * 3 > meta_used)
3778 num_bytes = div64_u64(meta_used, 3);
3780 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3783 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3785 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3786 struct btrfs_space_info *sinfo = block_rsv->space_info;
3789 num_bytes = calc_global_metadata_size(fs_info);
3791 spin_lock(&block_rsv->lock);
3792 spin_lock(&sinfo->lock);
3794 block_rsv->size = num_bytes;
3796 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3797 sinfo->bytes_reserved + sinfo->bytes_readonly +
3798 sinfo->bytes_may_use;
3800 if (sinfo->total_bytes > num_bytes) {
3801 num_bytes = sinfo->total_bytes - num_bytes;
3802 block_rsv->reserved += num_bytes;
3803 sinfo->bytes_reserved += num_bytes;
3806 if (block_rsv->reserved >= block_rsv->size) {
3807 num_bytes = block_rsv->reserved - block_rsv->size;
3808 sinfo->bytes_reserved -= num_bytes;
3809 block_rsv->reserved = block_rsv->size;
3810 block_rsv->full = 1;
3813 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3814 block_rsv->size, block_rsv->reserved);
3816 spin_unlock(&sinfo->lock);
3817 spin_unlock(&block_rsv->lock);
3820 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3822 struct btrfs_space_info *space_info;
3824 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3825 fs_info->chunk_block_rsv.space_info = space_info;
3826 fs_info->chunk_block_rsv.priority = 10;
3828 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3829 fs_info->global_block_rsv.space_info = space_info;
3830 fs_info->global_block_rsv.priority = 10;
3831 fs_info->global_block_rsv.refill_used = 1;
3832 fs_info->delalloc_block_rsv.space_info = space_info;
3833 fs_info->trans_block_rsv.space_info = space_info;
3834 fs_info->empty_block_rsv.space_info = space_info;
3835 fs_info->empty_block_rsv.priority = 10;
3837 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3838 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3839 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3840 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3841 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3843 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3845 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3847 update_global_block_rsv(fs_info);
3850 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3852 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3853 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3854 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3855 WARN_ON(fs_info->trans_block_rsv.size > 0);
3856 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3857 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3858 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3861 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3863 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3867 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root,
3874 if (num_items == 0 || root->fs_info->chunk_root == root)
3877 num_bytes = calc_trans_metadata_size(root, num_items);
3878 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3881 trans->bytes_reserved += num_bytes;
3882 trans->block_rsv = &root->fs_info->trans_block_rsv;
3887 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3888 struct btrfs_root *root)
3890 if (!trans->bytes_reserved)
3893 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3894 btrfs_block_rsv_release(root, trans->block_rsv,
3895 trans->bytes_reserved);
3896 trans->bytes_reserved = 0;
3899 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3900 struct inode *inode)
3902 struct btrfs_root *root = BTRFS_I(inode)->root;
3903 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3904 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3907 * one for deleting orphan item, one for updating inode and
3908 * two for calling btrfs_truncate_inode_items.
3910 * btrfs_truncate_inode_items is a delete operation, it frees
3911 * more space than it uses in most cases. So two units of
3912 * metadata space should be enough for calling it many times.
3913 * If all of the metadata space is used, we can commit
3914 * transaction and use space it freed.
3916 u64 num_bytes = calc_trans_metadata_size(root, 4);
3917 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3920 void btrfs_orphan_release_metadata(struct inode *inode)
3922 struct btrfs_root *root = BTRFS_I(inode)->root;
3923 u64 num_bytes = calc_trans_metadata_size(root, 4);
3924 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3927 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3928 struct btrfs_pending_snapshot *pending)
3930 struct btrfs_root *root = pending->root;
3931 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3932 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3934 * two for root back/forward refs, two for directory entries
3935 * and one for root of the snapshot.
3937 u64 num_bytes = calc_trans_metadata_size(root, 5);
3938 dst_rsv->space_info = src_rsv->space_info;
3939 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3942 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3944 return num_bytes >>= 3;
3947 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3949 struct btrfs_root *root = BTRFS_I(inode)->root;
3950 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3955 if (btrfs_transaction_in_commit(root->fs_info))
3956 schedule_timeout(1);
3958 num_bytes = ALIGN(num_bytes, root->sectorsize);
3960 spin_lock(&BTRFS_I(inode)->accounting_lock);
3961 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3962 if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3963 nr_extents -= BTRFS_I(inode)->reserved_extents;
3964 to_reserve = calc_trans_metadata_size(root, nr_extents);
3969 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3971 to_reserve += calc_csum_metadata_size(inode, num_bytes);
3972 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3976 spin_lock(&BTRFS_I(inode)->accounting_lock);
3977 BTRFS_I(inode)->reserved_extents += nr_extents;
3978 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3979 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3981 block_rsv_add_bytes(block_rsv, to_reserve, 1);
3983 if (block_rsv->size > 512 * 1024 * 1024)
3984 shrink_delalloc(NULL, root, to_reserve, 0);
3989 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3991 struct btrfs_root *root = BTRFS_I(inode)->root;
3995 num_bytes = ALIGN(num_bytes, root->sectorsize);
3996 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3998 spin_lock(&BTRFS_I(inode)->accounting_lock);
3999 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4000 if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4001 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4002 BTRFS_I(inode)->reserved_extents -= nr_extents;
4006 spin_unlock(&BTRFS_I(inode)->accounting_lock);
4008 to_free = calc_csum_metadata_size(inode, num_bytes);
4010 to_free += calc_trans_metadata_size(root, nr_extents);
4012 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4016 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4020 ret = btrfs_check_data_free_space(inode, num_bytes);
4024 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4026 btrfs_free_reserved_data_space(inode, num_bytes);
4033 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4035 btrfs_delalloc_release_metadata(inode, num_bytes);
4036 btrfs_free_reserved_data_space(inode, num_bytes);
4039 static int update_block_group(struct btrfs_trans_handle *trans,
4040 struct btrfs_root *root,
4041 u64 bytenr, u64 num_bytes, int alloc)
4043 struct btrfs_block_group_cache *cache = NULL;
4044 struct btrfs_fs_info *info = root->fs_info;
4045 u64 total = num_bytes;
4050 /* block accounting for super block */
4051 spin_lock(&info->delalloc_lock);
4052 old_val = btrfs_super_bytes_used(&info->super_copy);
4054 old_val += num_bytes;
4056 old_val -= num_bytes;
4057 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4058 spin_unlock(&info->delalloc_lock);
4061 cache = btrfs_lookup_block_group(info, bytenr);
4064 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4065 BTRFS_BLOCK_GROUP_RAID1 |
4066 BTRFS_BLOCK_GROUP_RAID10))
4071 * If this block group has free space cache written out, we
4072 * need to make sure to load it if we are removing space. This
4073 * is because we need the unpinning stage to actually add the
4074 * space back to the block group, otherwise we will leak space.
4076 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4077 cache_block_group(cache, trans, 1);
4079 byte_in_group = bytenr - cache->key.objectid;
4080 WARN_ON(byte_in_group > cache->key.offset);
4082 spin_lock(&cache->space_info->lock);
4083 spin_lock(&cache->lock);
4085 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4086 cache->disk_cache_state < BTRFS_DC_CLEAR)
4087 cache->disk_cache_state = BTRFS_DC_CLEAR;
4090 old_val = btrfs_block_group_used(&cache->item);
4091 num_bytes = min(total, cache->key.offset - byte_in_group);
4093 old_val += num_bytes;
4094 btrfs_set_block_group_used(&cache->item, old_val);
4095 cache->reserved -= num_bytes;
4096 cache->space_info->bytes_reserved -= num_bytes;
4097 cache->space_info->bytes_used += num_bytes;
4098 cache->space_info->disk_used += num_bytes * factor;
4099 spin_unlock(&cache->lock);
4100 spin_unlock(&cache->space_info->lock);
4102 old_val -= num_bytes;
4103 btrfs_set_block_group_used(&cache->item, old_val);
4104 cache->pinned += num_bytes;
4105 cache->space_info->bytes_pinned += num_bytes;
4106 cache->space_info->bytes_used -= num_bytes;
4107 cache->space_info->disk_used -= num_bytes * factor;
4108 spin_unlock(&cache->lock);
4109 spin_unlock(&cache->space_info->lock);
4111 set_extent_dirty(info->pinned_extents,
4112 bytenr, bytenr + num_bytes - 1,
4113 GFP_NOFS | __GFP_NOFAIL);
4115 btrfs_put_block_group(cache);
4117 bytenr += num_bytes;
4122 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4124 struct btrfs_block_group_cache *cache;
4127 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4131 bytenr = cache->key.objectid;
4132 btrfs_put_block_group(cache);
4137 static int pin_down_extent(struct btrfs_root *root,
4138 struct btrfs_block_group_cache *cache,
4139 u64 bytenr, u64 num_bytes, int reserved)
4141 spin_lock(&cache->space_info->lock);
4142 spin_lock(&cache->lock);
4143 cache->pinned += num_bytes;
4144 cache->space_info->bytes_pinned += num_bytes;
4146 cache->reserved -= num_bytes;
4147 cache->space_info->bytes_reserved -= num_bytes;
4149 spin_unlock(&cache->lock);
4150 spin_unlock(&cache->space_info->lock);
4152 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4153 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4158 * this function must be called within transaction
4160 int btrfs_pin_extent(struct btrfs_root *root,
4161 u64 bytenr, u64 num_bytes, int reserved)
4163 struct btrfs_block_group_cache *cache;
4165 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4168 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4170 btrfs_put_block_group(cache);
4175 * update size of reserved extents. this function may return -EAGAIN
4176 * if 'reserve' is true or 'sinfo' is false.
4178 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4179 u64 num_bytes, int reserve, int sinfo)
4183 struct btrfs_space_info *space_info = cache->space_info;
4184 spin_lock(&space_info->lock);
4185 spin_lock(&cache->lock);
4190 cache->reserved += num_bytes;
4191 space_info->bytes_reserved += num_bytes;
4195 space_info->bytes_readonly += num_bytes;
4196 cache->reserved -= num_bytes;
4197 space_info->bytes_reserved -= num_bytes;
4199 spin_unlock(&cache->lock);
4200 spin_unlock(&space_info->lock);
4202 spin_lock(&cache->lock);
4207 cache->reserved += num_bytes;
4209 cache->reserved -= num_bytes;
4211 spin_unlock(&cache->lock);
4216 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4217 struct btrfs_root *root)
4219 struct btrfs_fs_info *fs_info = root->fs_info;
4220 struct btrfs_caching_control *next;
4221 struct btrfs_caching_control *caching_ctl;
4222 struct btrfs_block_group_cache *cache;
4224 down_write(&fs_info->extent_commit_sem);
4226 list_for_each_entry_safe(caching_ctl, next,
4227 &fs_info->caching_block_groups, list) {
4228 cache = caching_ctl->block_group;
4229 if (block_group_cache_done(cache)) {
4230 cache->last_byte_to_unpin = (u64)-1;
4231 list_del_init(&caching_ctl->list);
4232 put_caching_control(caching_ctl);
4234 cache->last_byte_to_unpin = caching_ctl->progress;
4238 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4239 fs_info->pinned_extents = &fs_info->freed_extents[1];
4241 fs_info->pinned_extents = &fs_info->freed_extents[0];
4243 up_write(&fs_info->extent_commit_sem);
4245 update_global_block_rsv(fs_info);
4249 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4251 struct btrfs_fs_info *fs_info = root->fs_info;
4252 struct btrfs_block_group_cache *cache = NULL;
4255 while (start <= end) {
4257 start >= cache->key.objectid + cache->key.offset) {
4259 btrfs_put_block_group(cache);
4260 cache = btrfs_lookup_block_group(fs_info, start);
4264 len = cache->key.objectid + cache->key.offset - start;
4265 len = min(len, end + 1 - start);
4267 if (start < cache->last_byte_to_unpin) {
4268 len = min(len, cache->last_byte_to_unpin - start);
4269 btrfs_add_free_space(cache, start, len);
4274 spin_lock(&cache->space_info->lock);
4275 spin_lock(&cache->lock);
4276 cache->pinned -= len;
4277 cache->space_info->bytes_pinned -= len;
4279 cache->space_info->bytes_readonly += len;
4280 } else if (cache->reserved_pinned > 0) {
4281 len = min(len, cache->reserved_pinned);
4282 cache->reserved_pinned -= len;
4283 cache->space_info->bytes_reserved += len;
4285 spin_unlock(&cache->lock);
4286 spin_unlock(&cache->space_info->lock);
4290 btrfs_put_block_group(cache);
4294 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4295 struct btrfs_root *root)
4297 struct btrfs_fs_info *fs_info = root->fs_info;
4298 struct extent_io_tree *unpin;
4299 struct btrfs_block_rsv *block_rsv;
4300 struct btrfs_block_rsv *next_rsv;
4306 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4307 unpin = &fs_info->freed_extents[1];
4309 unpin = &fs_info->freed_extents[0];
4312 ret = find_first_extent_bit(unpin, 0, &start, &end,
4317 ret = btrfs_discard_extent(root, start, end + 1 - start);
4319 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4320 unpin_extent_range(root, start, end);
4324 mutex_lock(&fs_info->durable_block_rsv_mutex);
4325 list_for_each_entry_safe(block_rsv, next_rsv,
4326 &fs_info->durable_block_rsv_list, list) {
4328 idx = trans->transid & 0x1;
4329 if (block_rsv->freed[idx] > 0) {
4330 block_rsv_add_bytes(block_rsv,
4331 block_rsv->freed[idx], 0);
4332 block_rsv->freed[idx] = 0;
4334 if (atomic_read(&block_rsv->usage) == 0) {
4335 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4337 if (block_rsv->freed[0] == 0 &&
4338 block_rsv->freed[1] == 0) {
4339 list_del_init(&block_rsv->list);
4343 btrfs_block_rsv_release(root, block_rsv, 0);
4346 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4351 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4352 struct btrfs_root *root,
4353 u64 bytenr, u64 num_bytes, u64 parent,
4354 u64 root_objectid, u64 owner_objectid,
4355 u64 owner_offset, int refs_to_drop,
4356 struct btrfs_delayed_extent_op *extent_op)
4358 struct btrfs_key key;
4359 struct btrfs_path *path;
4360 struct btrfs_fs_info *info = root->fs_info;
4361 struct btrfs_root *extent_root = info->extent_root;
4362 struct extent_buffer *leaf;
4363 struct btrfs_extent_item *ei;
4364 struct btrfs_extent_inline_ref *iref;
4367 int extent_slot = 0;
4368 int found_extent = 0;
4373 path = btrfs_alloc_path();
4378 path->leave_spinning = 1;
4380 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4381 BUG_ON(!is_data && refs_to_drop != 1);
4383 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4384 bytenr, num_bytes, parent,
4385 root_objectid, owner_objectid,
4388 extent_slot = path->slots[0];
4389 while (extent_slot >= 0) {
4390 btrfs_item_key_to_cpu(path->nodes[0], &key,
4392 if (key.objectid != bytenr)
4394 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4395 key.offset == num_bytes) {
4399 if (path->slots[0] - extent_slot > 5)
4403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4404 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4405 if (found_extent && item_size < sizeof(*ei))
4408 if (!found_extent) {
4410 ret = remove_extent_backref(trans, extent_root, path,
4414 btrfs_release_path(extent_root, path);
4415 path->leave_spinning = 1;
4417 key.objectid = bytenr;
4418 key.type = BTRFS_EXTENT_ITEM_KEY;
4419 key.offset = num_bytes;
4421 ret = btrfs_search_slot(trans, extent_root,
4424 printk(KERN_ERR "umm, got %d back from search"
4425 ", was looking for %llu\n", ret,
4426 (unsigned long long)bytenr);
4427 btrfs_print_leaf(extent_root, path->nodes[0]);
4430 extent_slot = path->slots[0];
4433 btrfs_print_leaf(extent_root, path->nodes[0]);
4435 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4436 "parent %llu root %llu owner %llu offset %llu\n",
4437 (unsigned long long)bytenr,
4438 (unsigned long long)parent,
4439 (unsigned long long)root_objectid,
4440 (unsigned long long)owner_objectid,
4441 (unsigned long long)owner_offset);
4444 leaf = path->nodes[0];
4445 item_size = btrfs_item_size_nr(leaf, extent_slot);
4446 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4447 if (item_size < sizeof(*ei)) {
4448 BUG_ON(found_extent || extent_slot != path->slots[0]);
4449 ret = convert_extent_item_v0(trans, extent_root, path,
4453 btrfs_release_path(extent_root, path);
4454 path->leave_spinning = 1;
4456 key.objectid = bytenr;
4457 key.type = BTRFS_EXTENT_ITEM_KEY;
4458 key.offset = num_bytes;
4460 ret = btrfs_search_slot(trans, extent_root, &key, path,
4463 printk(KERN_ERR "umm, got %d back from search"
4464 ", was looking for %llu\n", ret,
4465 (unsigned long long)bytenr);
4466 btrfs_print_leaf(extent_root, path->nodes[0]);
4469 extent_slot = path->slots[0];
4470 leaf = path->nodes[0];
4471 item_size = btrfs_item_size_nr(leaf, extent_slot);
4474 BUG_ON(item_size < sizeof(*ei));
4475 ei = btrfs_item_ptr(leaf, extent_slot,
4476 struct btrfs_extent_item);
4477 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4478 struct btrfs_tree_block_info *bi;
4479 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4480 bi = (struct btrfs_tree_block_info *)(ei + 1);
4481 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4484 refs = btrfs_extent_refs(leaf, ei);
4485 BUG_ON(refs < refs_to_drop);
4486 refs -= refs_to_drop;
4490 __run_delayed_extent_op(extent_op, leaf, ei);
4492 * In the case of inline back ref, reference count will
4493 * be updated by remove_extent_backref
4496 BUG_ON(!found_extent);
4498 btrfs_set_extent_refs(leaf, ei, refs);
4499 btrfs_mark_buffer_dirty(leaf);
4502 ret = remove_extent_backref(trans, extent_root, path,
4509 BUG_ON(is_data && refs_to_drop !=
4510 extent_data_ref_count(root, path, iref));
4512 BUG_ON(path->slots[0] != extent_slot);
4514 BUG_ON(path->slots[0] != extent_slot + 1);
4515 path->slots[0] = extent_slot;
4520 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4523 btrfs_release_path(extent_root, path);
4526 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4529 invalidate_mapping_pages(info->btree_inode->i_mapping,
4530 bytenr >> PAGE_CACHE_SHIFT,
4531 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4534 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4537 btrfs_free_path(path);
4542 * when we free an block, it is possible (and likely) that we free the last
4543 * delayed ref for that extent as well. This searches the delayed ref tree for
4544 * a given extent, and if there are no other delayed refs to be processed, it
4545 * removes it from the tree.
4547 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4548 struct btrfs_root *root, u64 bytenr)
4550 struct btrfs_delayed_ref_head *head;
4551 struct btrfs_delayed_ref_root *delayed_refs;
4552 struct btrfs_delayed_ref_node *ref;
4553 struct rb_node *node;
4556 delayed_refs = &trans->transaction->delayed_refs;
4557 spin_lock(&delayed_refs->lock);
4558 head = btrfs_find_delayed_ref_head(trans, bytenr);
4562 node = rb_prev(&head->node.rb_node);
4566 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4568 /* there are still entries for this ref, we can't drop it */
4569 if (ref->bytenr == bytenr)
4572 if (head->extent_op) {
4573 if (!head->must_insert_reserved)
4575 kfree(head->extent_op);
4576 head->extent_op = NULL;
4580 * waiting for the lock here would deadlock. If someone else has it
4581 * locked they are already in the process of dropping it anyway
4583 if (!mutex_trylock(&head->mutex))
4587 * at this point we have a head with no other entries. Go
4588 * ahead and process it.
4590 head->node.in_tree = 0;
4591 rb_erase(&head->node.rb_node, &delayed_refs->root);
4593 delayed_refs->num_entries--;
4596 * we don't take a ref on the node because we're removing it from the
4597 * tree, so we just steal the ref the tree was holding.
4599 delayed_refs->num_heads--;
4600 if (list_empty(&head->cluster))
4601 delayed_refs->num_heads_ready--;
4603 list_del_init(&head->cluster);
4604 spin_unlock(&delayed_refs->lock);
4606 BUG_ON(head->extent_op);
4607 if (head->must_insert_reserved)
4610 mutex_unlock(&head->mutex);
4611 btrfs_put_delayed_ref(&head->node);
4614 spin_unlock(&delayed_refs->lock);
4618 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4619 struct btrfs_root *root,
4620 struct extent_buffer *buf,
4621 u64 parent, int last_ref)
4623 struct btrfs_block_rsv *block_rsv;
4624 struct btrfs_block_group_cache *cache = NULL;
4627 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4628 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4629 parent, root->root_key.objectid,
4630 btrfs_header_level(buf),
4631 BTRFS_DROP_DELAYED_REF, NULL);
4638 block_rsv = get_block_rsv(trans, root);
4639 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4640 if (block_rsv->space_info != cache->space_info)
4643 if (btrfs_header_generation(buf) == trans->transid) {
4644 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4645 ret = check_ref_cleanup(trans, root, buf->start);
4650 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4651 pin_down_extent(root, cache, buf->start, buf->len, 1);
4655 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4657 btrfs_add_free_space(cache, buf->start, buf->len);
4658 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4659 if (ret == -EAGAIN) {
4660 /* block group became read-only */
4661 update_reserved_bytes(cache, buf->len, 0, 1);
4666 spin_lock(&block_rsv->lock);
4667 if (block_rsv->reserved < block_rsv->size) {
4668 block_rsv->reserved += buf->len;
4671 spin_unlock(&block_rsv->lock);
4674 spin_lock(&cache->space_info->lock);
4675 cache->space_info->bytes_reserved -= buf->len;
4676 spin_unlock(&cache->space_info->lock);
4681 if (block_rsv->durable && !cache->ro) {
4683 spin_lock(&cache->lock);
4685 cache->reserved_pinned += buf->len;
4688 spin_unlock(&cache->lock);
4691 spin_lock(&block_rsv->lock);
4692 block_rsv->freed[trans->transid & 0x1] += buf->len;
4693 spin_unlock(&block_rsv->lock);
4697 btrfs_put_block_group(cache);
4700 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4701 struct btrfs_root *root,
4702 u64 bytenr, u64 num_bytes, u64 parent,
4703 u64 root_objectid, u64 owner, u64 offset)
4708 * tree log blocks never actually go into the extent allocation
4709 * tree, just update pinning info and exit early.
4711 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4712 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4713 /* unlocks the pinned mutex */
4714 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4716 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4717 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4718 parent, root_objectid, (int)owner,
4719 BTRFS_DROP_DELAYED_REF, NULL);
4722 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4723 parent, root_objectid, owner,
4724 offset, BTRFS_DROP_DELAYED_REF, NULL);
4730 static u64 stripe_align(struct btrfs_root *root, u64 val)
4732 u64 mask = ((u64)root->stripesize - 1);
4733 u64 ret = (val + mask) & ~mask;
4738 * when we wait for progress in the block group caching, its because
4739 * our allocation attempt failed at least once. So, we must sleep
4740 * and let some progress happen before we try again.
4742 * This function will sleep at least once waiting for new free space to
4743 * show up, and then it will check the block group free space numbers
4744 * for our min num_bytes. Another option is to have it go ahead
4745 * and look in the rbtree for a free extent of a given size, but this
4749 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4752 struct btrfs_caching_control *caching_ctl;
4755 caching_ctl = get_caching_control(cache);
4759 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4760 (cache->free_space >= num_bytes));
4762 put_caching_control(caching_ctl);
4767 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4769 struct btrfs_caching_control *caching_ctl;
4772 caching_ctl = get_caching_control(cache);
4776 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4778 put_caching_control(caching_ctl);
4782 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4785 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4787 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4789 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4791 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4798 enum btrfs_loop_type {
4799 LOOP_FIND_IDEAL = 0,
4800 LOOP_CACHING_NOWAIT = 1,
4801 LOOP_CACHING_WAIT = 2,
4802 LOOP_ALLOC_CHUNK = 3,
4803 LOOP_NO_EMPTY_SIZE = 4,
4807 * walks the btree of allocated extents and find a hole of a given size.
4808 * The key ins is changed to record the hole:
4809 * ins->objectid == block start
4810 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4811 * ins->offset == number of blocks
4812 * Any available blocks before search_start are skipped.
4814 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4815 struct btrfs_root *orig_root,
4816 u64 num_bytes, u64 empty_size,
4817 u64 search_start, u64 search_end,
4818 u64 hint_byte, struct btrfs_key *ins,
4822 struct btrfs_root *root = orig_root->fs_info->extent_root;
4823 struct btrfs_free_cluster *last_ptr = NULL;
4824 struct btrfs_block_group_cache *block_group = NULL;
4825 int empty_cluster = 2 * 1024 * 1024;
4826 int allowed_chunk_alloc = 0;
4827 int done_chunk_alloc = 0;
4828 struct btrfs_space_info *space_info;
4829 int last_ptr_loop = 0;
4832 bool found_uncached_bg = false;
4833 bool failed_cluster_refill = false;
4834 bool failed_alloc = false;
4835 bool use_cluster = true;
4836 u64 ideal_cache_percent = 0;
4837 u64 ideal_cache_offset = 0;
4839 WARN_ON(num_bytes < root->sectorsize);
4840 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4844 space_info = __find_space_info(root->fs_info, data);
4846 printk(KERN_ERR "No space info for %d\n", data);
4851 * If the space info is for both data and metadata it means we have a
4852 * small filesystem and we can't use the clustering stuff.
4854 if (btrfs_mixed_space_info(space_info))
4855 use_cluster = false;
4857 if (orig_root->ref_cows || empty_size)
4858 allowed_chunk_alloc = 1;
4860 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4861 last_ptr = &root->fs_info->meta_alloc_cluster;
4862 if (!btrfs_test_opt(root, SSD))
4863 empty_cluster = 64 * 1024;
4866 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4867 btrfs_test_opt(root, SSD)) {
4868 last_ptr = &root->fs_info->data_alloc_cluster;
4872 spin_lock(&last_ptr->lock);
4873 if (last_ptr->block_group)
4874 hint_byte = last_ptr->window_start;
4875 spin_unlock(&last_ptr->lock);
4878 search_start = max(search_start, first_logical_byte(root, 0));
4879 search_start = max(search_start, hint_byte);
4884 if (search_start == hint_byte) {
4886 block_group = btrfs_lookup_block_group(root->fs_info,
4889 * we don't want to use the block group if it doesn't match our
4890 * allocation bits, or if its not cached.
4892 * However if we are re-searching with an ideal block group
4893 * picked out then we don't care that the block group is cached.
4895 if (block_group && block_group_bits(block_group, data) &&
4896 (block_group->cached != BTRFS_CACHE_NO ||
4897 search_start == ideal_cache_offset)) {
4898 down_read(&space_info->groups_sem);
4899 if (list_empty(&block_group->list) ||
4902 * someone is removing this block group,
4903 * we can't jump into the have_block_group
4904 * target because our list pointers are not
4907 btrfs_put_block_group(block_group);
4908 up_read(&space_info->groups_sem);
4910 index = get_block_group_index(block_group);
4911 goto have_block_group;
4913 } else if (block_group) {
4914 btrfs_put_block_group(block_group);
4918 down_read(&space_info->groups_sem);
4919 list_for_each_entry(block_group, &space_info->block_groups[index],
4924 btrfs_get_block_group(block_group);
4925 search_start = block_group->key.objectid;
4928 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4931 ret = cache_block_group(block_group, trans, 1);
4932 if (block_group->cached == BTRFS_CACHE_FINISHED)
4933 goto have_block_group;
4935 free_percent = btrfs_block_group_used(&block_group->item);
4936 free_percent *= 100;
4937 free_percent = div64_u64(free_percent,
4938 block_group->key.offset);
4939 free_percent = 100 - free_percent;
4940 if (free_percent > ideal_cache_percent &&
4941 likely(!block_group->ro)) {
4942 ideal_cache_offset = block_group->key.objectid;
4943 ideal_cache_percent = free_percent;
4947 * We only want to start kthread caching if we are at
4948 * the point where we will wait for caching to make
4949 * progress, or if our ideal search is over and we've
4950 * found somebody to start caching.
4952 if (loop > LOOP_CACHING_NOWAIT ||
4953 (loop > LOOP_FIND_IDEAL &&
4954 atomic_read(&space_info->caching_threads) < 2)) {
4955 ret = cache_block_group(block_group, trans, 0);
4958 found_uncached_bg = true;
4961 * If loop is set for cached only, try the next block
4964 if (loop == LOOP_FIND_IDEAL)
4968 cached = block_group_cache_done(block_group);
4969 if (unlikely(!cached))
4970 found_uncached_bg = true;
4972 if (unlikely(block_group->ro))
4976 * Ok we want to try and use the cluster allocator, so lets look
4977 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4978 * have tried the cluster allocator plenty of times at this
4979 * point and not have found anything, so we are likely way too
4980 * fragmented for the clustering stuff to find anything, so lets
4981 * just skip it and let the allocator find whatever block it can
4984 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4986 * the refill lock keeps out other
4987 * people trying to start a new cluster
4989 spin_lock(&last_ptr->refill_lock);
4990 if (last_ptr->block_group &&
4991 (last_ptr->block_group->ro ||
4992 !block_group_bits(last_ptr->block_group, data))) {
4994 goto refill_cluster;
4997 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4998 num_bytes, search_start);
5000 /* we have a block, we're done */
5001 spin_unlock(&last_ptr->refill_lock);
5005 spin_lock(&last_ptr->lock);
5007 * whoops, this cluster doesn't actually point to
5008 * this block group. Get a ref on the block
5009 * group is does point to and try again
5011 if (!last_ptr_loop && last_ptr->block_group &&
5012 last_ptr->block_group != block_group) {
5014 btrfs_put_block_group(block_group);
5015 block_group = last_ptr->block_group;
5016 btrfs_get_block_group(block_group);
5017 spin_unlock(&last_ptr->lock);
5018 spin_unlock(&last_ptr->refill_lock);
5021 search_start = block_group->key.objectid;
5023 * we know this block group is properly
5024 * in the list because
5025 * btrfs_remove_block_group, drops the
5026 * cluster before it removes the block
5027 * group from the list
5029 goto have_block_group;
5031 spin_unlock(&last_ptr->lock);
5034 * this cluster didn't work out, free it and
5037 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5041 /* allocate a cluster in this block group */
5042 ret = btrfs_find_space_cluster(trans, root,
5043 block_group, last_ptr,
5045 empty_cluster + empty_size);
5048 * now pull our allocation out of this
5051 offset = btrfs_alloc_from_cluster(block_group,
5052 last_ptr, num_bytes,
5055 /* we found one, proceed */
5056 spin_unlock(&last_ptr->refill_lock);
5059 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5060 && !failed_cluster_refill) {
5061 spin_unlock(&last_ptr->refill_lock);
5063 failed_cluster_refill = true;
5064 wait_block_group_cache_progress(block_group,
5065 num_bytes + empty_cluster + empty_size);
5066 goto have_block_group;
5070 * at this point we either didn't find a cluster
5071 * or we weren't able to allocate a block from our
5072 * cluster. Free the cluster we've been trying
5073 * to use, and go to the next block group
5075 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5076 spin_unlock(&last_ptr->refill_lock);
5080 offset = btrfs_find_space_for_alloc(block_group, search_start,
5081 num_bytes, empty_size);
5083 * If we didn't find a chunk, and we haven't failed on this
5084 * block group before, and this block group is in the middle of
5085 * caching and we are ok with waiting, then go ahead and wait
5086 * for progress to be made, and set failed_alloc to true.
5088 * If failed_alloc is true then we've already waited on this
5089 * block group once and should move on to the next block group.
5091 if (!offset && !failed_alloc && !cached &&
5092 loop > LOOP_CACHING_NOWAIT) {
5093 wait_block_group_cache_progress(block_group,
5094 num_bytes + empty_size);
5095 failed_alloc = true;
5096 goto have_block_group;
5097 } else if (!offset) {
5101 search_start = stripe_align(root, offset);
5102 /* move on to the next group */
5103 if (search_start + num_bytes >= search_end) {
5104 btrfs_add_free_space(block_group, offset, num_bytes);
5108 /* move on to the next group */
5109 if (search_start + num_bytes >
5110 block_group->key.objectid + block_group->key.offset) {
5111 btrfs_add_free_space(block_group, offset, num_bytes);
5115 ins->objectid = search_start;
5116 ins->offset = num_bytes;
5118 if (offset < search_start)
5119 btrfs_add_free_space(block_group, offset,
5120 search_start - offset);
5121 BUG_ON(offset > search_start);
5123 ret = update_reserved_bytes(block_group, num_bytes, 1,
5124 (data & BTRFS_BLOCK_GROUP_DATA));
5125 if (ret == -EAGAIN) {
5126 btrfs_add_free_space(block_group, offset, num_bytes);
5130 /* we are all good, lets return */
5131 ins->objectid = search_start;
5132 ins->offset = num_bytes;
5134 if (offset < search_start)
5135 btrfs_add_free_space(block_group, offset,
5136 search_start - offset);
5137 BUG_ON(offset > search_start);
5140 failed_cluster_refill = false;
5141 failed_alloc = false;
5142 BUG_ON(index != get_block_group_index(block_group));
5143 btrfs_put_block_group(block_group);
5145 up_read(&space_info->groups_sem);
5147 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5150 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5151 * for them to make caching progress. Also
5152 * determine the best possible bg to cache
5153 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5154 * caching kthreads as we move along
5155 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5156 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5157 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5160 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5161 (found_uncached_bg || empty_size || empty_cluster ||
5162 allowed_chunk_alloc)) {
5164 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5165 found_uncached_bg = false;
5167 if (!ideal_cache_percent &&
5168 atomic_read(&space_info->caching_threads))
5172 * 1 of the following 2 things have happened so far
5174 * 1) We found an ideal block group for caching that
5175 * is mostly full and will cache quickly, so we might
5176 * as well wait for it.
5178 * 2) We searched for cached only and we didn't find
5179 * anything, and we didn't start any caching kthreads
5180 * either, so chances are we will loop through and
5181 * start a couple caching kthreads, and then come back
5182 * around and just wait for them. This will be slower
5183 * because we will have 2 caching kthreads reading at
5184 * the same time when we could have just started one
5185 * and waited for it to get far enough to give us an
5186 * allocation, so go ahead and go to the wait caching
5189 loop = LOOP_CACHING_WAIT;
5190 search_start = ideal_cache_offset;
5191 ideal_cache_percent = 0;
5193 } else if (loop == LOOP_FIND_IDEAL) {
5195 * Didn't find a uncached bg, wait on anything we find
5198 loop = LOOP_CACHING_WAIT;
5202 if (loop < LOOP_CACHING_WAIT) {
5207 if (loop == LOOP_ALLOC_CHUNK) {
5212 if (allowed_chunk_alloc) {
5213 ret = do_chunk_alloc(trans, root, num_bytes +
5214 2 * 1024 * 1024, data, 1);
5215 allowed_chunk_alloc = 0;
5216 done_chunk_alloc = 1;
5217 } else if (!done_chunk_alloc) {
5218 space_info->force_alloc = 1;
5221 if (loop < LOOP_NO_EMPTY_SIZE) {
5226 } else if (!ins->objectid) {
5230 /* we found what we needed */
5231 if (ins->objectid) {
5232 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5233 trans->block_group = block_group->key.objectid;
5235 btrfs_put_block_group(block_group);
5242 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5243 int dump_block_groups)
5245 struct btrfs_block_group_cache *cache;
5248 spin_lock(&info->lock);
5249 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5250 (unsigned long long)(info->total_bytes - info->bytes_used -
5251 info->bytes_pinned - info->bytes_reserved -
5252 info->bytes_readonly),
5253 (info->full) ? "" : "not ");
5254 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5255 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5256 (unsigned long long)info->total_bytes,
5257 (unsigned long long)info->bytes_used,
5258 (unsigned long long)info->bytes_pinned,
5259 (unsigned long long)info->bytes_reserved,
5260 (unsigned long long)info->bytes_may_use,
5261 (unsigned long long)info->bytes_readonly);
5262 spin_unlock(&info->lock);
5264 if (!dump_block_groups)
5267 down_read(&info->groups_sem);
5269 list_for_each_entry(cache, &info->block_groups[index], list) {
5270 spin_lock(&cache->lock);
5271 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5272 "%llu pinned %llu reserved\n",
5273 (unsigned long long)cache->key.objectid,
5274 (unsigned long long)cache->key.offset,
5275 (unsigned long long)btrfs_block_group_used(&cache->item),
5276 (unsigned long long)cache->pinned,
5277 (unsigned long long)cache->reserved);
5278 btrfs_dump_free_space(cache, bytes);
5279 spin_unlock(&cache->lock);
5281 if (++index < BTRFS_NR_RAID_TYPES)
5283 up_read(&info->groups_sem);
5286 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5287 struct btrfs_root *root,
5288 u64 num_bytes, u64 min_alloc_size,
5289 u64 empty_size, u64 hint_byte,
5290 u64 search_end, struct btrfs_key *ins,
5294 u64 search_start = 0;
5296 data = btrfs_get_alloc_profile(root, data);
5299 * the only place that sets empty_size is btrfs_realloc_node, which
5300 * is not called recursively on allocations
5302 if (empty_size || root->ref_cows)
5303 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5304 num_bytes + 2 * 1024 * 1024, data, 0);
5306 WARN_ON(num_bytes < root->sectorsize);
5307 ret = find_free_extent(trans, root, num_bytes, empty_size,
5308 search_start, search_end, hint_byte,
5311 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5312 num_bytes = num_bytes >> 1;
5313 num_bytes = num_bytes & ~(root->sectorsize - 1);
5314 num_bytes = max(num_bytes, min_alloc_size);
5315 do_chunk_alloc(trans, root->fs_info->extent_root,
5316 num_bytes, data, 1);
5319 if (ret == -ENOSPC) {
5320 struct btrfs_space_info *sinfo;
5322 sinfo = __find_space_info(root->fs_info, data);
5323 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5324 "wanted %llu\n", (unsigned long long)data,
5325 (unsigned long long)num_bytes);
5326 dump_space_info(sinfo, num_bytes, 1);
5332 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5334 struct btrfs_block_group_cache *cache;
5337 cache = btrfs_lookup_block_group(root->fs_info, start);
5339 printk(KERN_ERR "Unable to find block group for %llu\n",
5340 (unsigned long long)start);
5344 ret = btrfs_discard_extent(root, start, len);
5346 btrfs_add_free_space(cache, start, len);
5347 update_reserved_bytes(cache, len, 0, 1);
5348 btrfs_put_block_group(cache);
5353 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5354 struct btrfs_root *root,
5355 u64 parent, u64 root_objectid,
5356 u64 flags, u64 owner, u64 offset,
5357 struct btrfs_key *ins, int ref_mod)
5360 struct btrfs_fs_info *fs_info = root->fs_info;
5361 struct btrfs_extent_item *extent_item;
5362 struct btrfs_extent_inline_ref *iref;
5363 struct btrfs_path *path;
5364 struct extent_buffer *leaf;
5369 type = BTRFS_SHARED_DATA_REF_KEY;
5371 type = BTRFS_EXTENT_DATA_REF_KEY;
5373 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5375 path = btrfs_alloc_path();
5378 path->leave_spinning = 1;
5379 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5383 leaf = path->nodes[0];
5384 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5385 struct btrfs_extent_item);
5386 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5387 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5388 btrfs_set_extent_flags(leaf, extent_item,
5389 flags | BTRFS_EXTENT_FLAG_DATA);
5391 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5392 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5394 struct btrfs_shared_data_ref *ref;
5395 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5396 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5397 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5399 struct btrfs_extent_data_ref *ref;
5400 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5401 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5402 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5403 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5404 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5407 btrfs_mark_buffer_dirty(path->nodes[0]);
5408 btrfs_free_path(path);
5410 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5412 printk(KERN_ERR "btrfs update block group failed for %llu "
5413 "%llu\n", (unsigned long long)ins->objectid,
5414 (unsigned long long)ins->offset);
5420 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5421 struct btrfs_root *root,
5422 u64 parent, u64 root_objectid,
5423 u64 flags, struct btrfs_disk_key *key,
5424 int level, struct btrfs_key *ins)
5427 struct btrfs_fs_info *fs_info = root->fs_info;
5428 struct btrfs_extent_item *extent_item;
5429 struct btrfs_tree_block_info *block_info;
5430 struct btrfs_extent_inline_ref *iref;
5431 struct btrfs_path *path;
5432 struct extent_buffer *leaf;
5433 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5435 path = btrfs_alloc_path();
5438 path->leave_spinning = 1;
5439 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5443 leaf = path->nodes[0];
5444 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5445 struct btrfs_extent_item);
5446 btrfs_set_extent_refs(leaf, extent_item, 1);
5447 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5448 btrfs_set_extent_flags(leaf, extent_item,
5449 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5450 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5452 btrfs_set_tree_block_key(leaf, block_info, key);
5453 btrfs_set_tree_block_level(leaf, block_info, level);
5455 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5457 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5458 btrfs_set_extent_inline_ref_type(leaf, iref,
5459 BTRFS_SHARED_BLOCK_REF_KEY);
5460 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5462 btrfs_set_extent_inline_ref_type(leaf, iref,
5463 BTRFS_TREE_BLOCK_REF_KEY);
5464 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5467 btrfs_mark_buffer_dirty(leaf);
5468 btrfs_free_path(path);
5470 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5472 printk(KERN_ERR "btrfs update block group failed for %llu "
5473 "%llu\n", (unsigned long long)ins->objectid,
5474 (unsigned long long)ins->offset);
5480 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5481 struct btrfs_root *root,
5482 u64 root_objectid, u64 owner,
5483 u64 offset, struct btrfs_key *ins)
5487 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5489 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5490 0, root_objectid, owner, offset,
5491 BTRFS_ADD_DELAYED_EXTENT, NULL);
5496 * this is used by the tree logging recovery code. It records that
5497 * an extent has been allocated and makes sure to clear the free
5498 * space cache bits as well
5500 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5501 struct btrfs_root *root,
5502 u64 root_objectid, u64 owner, u64 offset,
5503 struct btrfs_key *ins)
5506 struct btrfs_block_group_cache *block_group;
5507 struct btrfs_caching_control *caching_ctl;
5508 u64 start = ins->objectid;
5509 u64 num_bytes = ins->offset;
5511 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5512 cache_block_group(block_group, trans, 0);
5513 caching_ctl = get_caching_control(block_group);
5516 BUG_ON(!block_group_cache_done(block_group));
5517 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5520 mutex_lock(&caching_ctl->mutex);
5522 if (start >= caching_ctl->progress) {
5523 ret = add_excluded_extent(root, start, num_bytes);
5525 } else if (start + num_bytes <= caching_ctl->progress) {
5526 ret = btrfs_remove_free_space(block_group,
5530 num_bytes = caching_ctl->progress - start;
5531 ret = btrfs_remove_free_space(block_group,
5535 start = caching_ctl->progress;
5536 num_bytes = ins->objectid + ins->offset -
5537 caching_ctl->progress;
5538 ret = add_excluded_extent(root, start, num_bytes);
5542 mutex_unlock(&caching_ctl->mutex);
5543 put_caching_control(caching_ctl);
5546 ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5548 btrfs_put_block_group(block_group);
5549 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5550 0, owner, offset, ins, 1);
5554 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5555 struct btrfs_root *root,
5556 u64 bytenr, u32 blocksize,
5559 struct extent_buffer *buf;
5561 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5563 return ERR_PTR(-ENOMEM);
5564 btrfs_set_header_generation(buf, trans->transid);
5565 btrfs_set_buffer_lockdep_class(buf, level);
5566 btrfs_tree_lock(buf);
5567 clean_tree_block(trans, root, buf);
5569 btrfs_set_lock_blocking(buf);
5570 btrfs_set_buffer_uptodate(buf);
5572 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5574 * we allow two log transactions at a time, use different
5575 * EXENT bit to differentiate dirty pages.
5577 if (root->log_transid % 2 == 0)
5578 set_extent_dirty(&root->dirty_log_pages, buf->start,
5579 buf->start + buf->len - 1, GFP_NOFS);
5581 set_extent_new(&root->dirty_log_pages, buf->start,
5582 buf->start + buf->len - 1, GFP_NOFS);
5584 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5585 buf->start + buf->len - 1, GFP_NOFS);
5587 trans->blocks_used++;
5588 /* this returns a buffer locked for blocking */
5592 static struct btrfs_block_rsv *
5593 use_block_rsv(struct btrfs_trans_handle *trans,
5594 struct btrfs_root *root, u32 blocksize)
5596 struct btrfs_block_rsv *block_rsv;
5599 block_rsv = get_block_rsv(trans, root);
5601 if (block_rsv->size == 0) {
5602 ret = reserve_metadata_bytes(trans, root, block_rsv,
5605 return ERR_PTR(ret);
5609 ret = block_rsv_use_bytes(block_rsv, blocksize);
5613 return ERR_PTR(-ENOSPC);
5616 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5618 block_rsv_add_bytes(block_rsv, blocksize, 0);
5619 block_rsv_release_bytes(block_rsv, NULL, 0);
5623 * finds a free extent and does all the dirty work required for allocation
5624 * returns the key for the extent through ins, and a tree buffer for
5625 * the first block of the extent through buf.
5627 * returns the tree buffer or NULL.
5629 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5630 struct btrfs_root *root, u32 blocksize,
5631 u64 parent, u64 root_objectid,
5632 struct btrfs_disk_key *key, int level,
5633 u64 hint, u64 empty_size)
5635 struct btrfs_key ins;
5636 struct btrfs_block_rsv *block_rsv;
5637 struct extent_buffer *buf;
5642 block_rsv = use_block_rsv(trans, root, blocksize);
5643 if (IS_ERR(block_rsv))
5644 return ERR_CAST(block_rsv);
5646 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5647 empty_size, hint, (u64)-1, &ins, 0);
5649 unuse_block_rsv(block_rsv, blocksize);
5650 return ERR_PTR(ret);
5653 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5655 BUG_ON(IS_ERR(buf));
5657 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5659 parent = ins.objectid;
5660 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5664 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5665 struct btrfs_delayed_extent_op *extent_op;
5666 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5669 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5671 memset(&extent_op->key, 0, sizeof(extent_op->key));
5672 extent_op->flags_to_set = flags;
5673 extent_op->update_key = 1;
5674 extent_op->update_flags = 1;
5675 extent_op->is_data = 0;
5677 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5678 ins.offset, parent, root_objectid,
5679 level, BTRFS_ADD_DELAYED_EXTENT,
5686 struct walk_control {
5687 u64 refs[BTRFS_MAX_LEVEL];
5688 u64 flags[BTRFS_MAX_LEVEL];
5689 struct btrfs_key update_progress;
5699 #define DROP_REFERENCE 1
5700 #define UPDATE_BACKREF 2
5702 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5703 struct btrfs_root *root,
5704 struct walk_control *wc,
5705 struct btrfs_path *path)
5714 struct btrfs_key key;
5715 struct extent_buffer *eb;
5720 if (path->slots[wc->level] < wc->reada_slot) {
5721 wc->reada_count = wc->reada_count * 2 / 3;
5722 wc->reada_count = max(wc->reada_count, 2);
5724 wc->reada_count = wc->reada_count * 3 / 2;
5725 wc->reada_count = min_t(int, wc->reada_count,
5726 BTRFS_NODEPTRS_PER_BLOCK(root));
5729 eb = path->nodes[wc->level];
5730 nritems = btrfs_header_nritems(eb);
5731 blocksize = btrfs_level_size(root, wc->level - 1);
5733 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5734 if (nread >= wc->reada_count)
5738 bytenr = btrfs_node_blockptr(eb, slot);
5739 generation = btrfs_node_ptr_generation(eb, slot);
5741 if (slot == path->slots[wc->level])
5744 if (wc->stage == UPDATE_BACKREF &&
5745 generation <= root->root_key.offset)
5748 /* We don't lock the tree block, it's OK to be racy here */
5749 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5754 if (wc->stage == DROP_REFERENCE) {
5758 if (wc->level == 1 &&
5759 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5761 if (!wc->update_ref ||
5762 generation <= root->root_key.offset)
5764 btrfs_node_key_to_cpu(eb, &key, slot);
5765 ret = btrfs_comp_cpu_keys(&key,
5766 &wc->update_progress);
5770 if (wc->level == 1 &&
5771 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5775 ret = readahead_tree_block(root, bytenr, blocksize,
5779 last = bytenr + blocksize;
5782 wc->reada_slot = slot;
5786 * hepler to process tree block while walking down the tree.
5788 * when wc->stage == UPDATE_BACKREF, this function updates
5789 * back refs for pointers in the block.
5791 * NOTE: return value 1 means we should stop walking down.
5793 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5794 struct btrfs_root *root,
5795 struct btrfs_path *path,
5796 struct walk_control *wc, int lookup_info)
5798 int level = wc->level;
5799 struct extent_buffer *eb = path->nodes[level];
5800 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5803 if (wc->stage == UPDATE_BACKREF &&
5804 btrfs_header_owner(eb) != root->root_key.objectid)
5808 * when reference count of tree block is 1, it won't increase
5809 * again. once full backref flag is set, we never clear it.
5812 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5813 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5814 BUG_ON(!path->locks[level]);
5815 ret = btrfs_lookup_extent_info(trans, root,
5820 BUG_ON(wc->refs[level] == 0);
5823 if (wc->stage == DROP_REFERENCE) {
5824 if (wc->refs[level] > 1)
5827 if (path->locks[level] && !wc->keep_locks) {
5828 btrfs_tree_unlock(eb);
5829 path->locks[level] = 0;
5834 /* wc->stage == UPDATE_BACKREF */
5835 if (!(wc->flags[level] & flag)) {
5836 BUG_ON(!path->locks[level]);
5837 ret = btrfs_inc_ref(trans, root, eb, 1);
5839 ret = btrfs_dec_ref(trans, root, eb, 0);
5841 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5844 wc->flags[level] |= flag;
5848 * the block is shared by multiple trees, so it's not good to
5849 * keep the tree lock
5851 if (path->locks[level] && level > 0) {
5852 btrfs_tree_unlock(eb);
5853 path->locks[level] = 0;
5859 * hepler to process tree block pointer.
5861 * when wc->stage == DROP_REFERENCE, this function checks
5862 * reference count of the block pointed to. if the block
5863 * is shared and we need update back refs for the subtree
5864 * rooted at the block, this function changes wc->stage to
5865 * UPDATE_BACKREF. if the block is shared and there is no
5866 * need to update back, this function drops the reference
5869 * NOTE: return value 1 means we should stop walking down.
5871 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5872 struct btrfs_root *root,
5873 struct btrfs_path *path,
5874 struct walk_control *wc, int *lookup_info)
5880 struct btrfs_key key;
5881 struct extent_buffer *next;
5882 int level = wc->level;
5886 generation = btrfs_node_ptr_generation(path->nodes[level],
5887 path->slots[level]);
5889 * if the lower level block was created before the snapshot
5890 * was created, we know there is no need to update back refs
5893 if (wc->stage == UPDATE_BACKREF &&
5894 generation <= root->root_key.offset) {
5899 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5900 blocksize = btrfs_level_size(root, level - 1);
5902 next = btrfs_find_tree_block(root, bytenr, blocksize);
5904 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5909 btrfs_tree_lock(next);
5910 btrfs_set_lock_blocking(next);
5912 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5913 &wc->refs[level - 1],
5914 &wc->flags[level - 1]);
5916 BUG_ON(wc->refs[level - 1] == 0);
5919 if (wc->stage == DROP_REFERENCE) {
5920 if (wc->refs[level - 1] > 1) {
5922 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5925 if (!wc->update_ref ||
5926 generation <= root->root_key.offset)
5929 btrfs_node_key_to_cpu(path->nodes[level], &key,
5930 path->slots[level]);
5931 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5935 wc->stage = UPDATE_BACKREF;
5936 wc->shared_level = level - 1;
5940 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5944 if (!btrfs_buffer_uptodate(next, generation)) {
5945 btrfs_tree_unlock(next);
5946 free_extent_buffer(next);
5952 if (reada && level == 1)
5953 reada_walk_down(trans, root, wc, path);
5954 next = read_tree_block(root, bytenr, blocksize, generation);
5955 btrfs_tree_lock(next);
5956 btrfs_set_lock_blocking(next);
5960 BUG_ON(level != btrfs_header_level(next));
5961 path->nodes[level] = next;
5962 path->slots[level] = 0;
5963 path->locks[level] = 1;
5969 wc->refs[level - 1] = 0;
5970 wc->flags[level - 1] = 0;
5971 if (wc->stage == DROP_REFERENCE) {
5972 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5973 parent = path->nodes[level]->start;
5975 BUG_ON(root->root_key.objectid !=
5976 btrfs_header_owner(path->nodes[level]));
5980 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5981 root->root_key.objectid, level - 1, 0);
5984 btrfs_tree_unlock(next);
5985 free_extent_buffer(next);
5991 * hepler to process tree block while walking up the tree.
5993 * when wc->stage == DROP_REFERENCE, this function drops
5994 * reference count on the block.
5996 * when wc->stage == UPDATE_BACKREF, this function changes
5997 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5998 * to UPDATE_BACKREF previously while processing the block.
6000 * NOTE: return value 1 means we should stop walking up.
6002 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6003 struct btrfs_root *root,
6004 struct btrfs_path *path,
6005 struct walk_control *wc)
6008 int level = wc->level;
6009 struct extent_buffer *eb = path->nodes[level];
6012 if (wc->stage == UPDATE_BACKREF) {
6013 BUG_ON(wc->shared_level < level);
6014 if (level < wc->shared_level)
6017 ret = find_next_key(path, level + 1, &wc->update_progress);
6021 wc->stage = DROP_REFERENCE;
6022 wc->shared_level = -1;
6023 path->slots[level] = 0;
6026 * check reference count again if the block isn't locked.
6027 * we should start walking down the tree again if reference
6030 if (!path->locks[level]) {
6032 btrfs_tree_lock(eb);
6033 btrfs_set_lock_blocking(eb);
6034 path->locks[level] = 1;
6036 ret = btrfs_lookup_extent_info(trans, root,
6041 BUG_ON(wc->refs[level] == 0);
6042 if (wc->refs[level] == 1) {
6043 btrfs_tree_unlock(eb);
6044 path->locks[level] = 0;
6050 /* wc->stage == DROP_REFERENCE */
6051 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6053 if (wc->refs[level] == 1) {
6055 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6056 ret = btrfs_dec_ref(trans, root, eb, 1);
6058 ret = btrfs_dec_ref(trans, root, eb, 0);
6061 /* make block locked assertion in clean_tree_block happy */
6062 if (!path->locks[level] &&
6063 btrfs_header_generation(eb) == trans->transid) {
6064 btrfs_tree_lock(eb);
6065 btrfs_set_lock_blocking(eb);
6066 path->locks[level] = 1;
6068 clean_tree_block(trans, root, eb);
6071 if (eb == root->node) {
6072 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6075 BUG_ON(root->root_key.objectid !=
6076 btrfs_header_owner(eb));
6078 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6079 parent = path->nodes[level + 1]->start;
6081 BUG_ON(root->root_key.objectid !=
6082 btrfs_header_owner(path->nodes[level + 1]));
6085 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6087 wc->refs[level] = 0;
6088 wc->flags[level] = 0;
6092 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6093 struct btrfs_root *root,
6094 struct btrfs_path *path,
6095 struct walk_control *wc)
6097 int level = wc->level;
6098 int lookup_info = 1;
6101 while (level >= 0) {
6102 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6109 if (path->slots[level] >=
6110 btrfs_header_nritems(path->nodes[level]))
6113 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6115 path->slots[level]++;
6124 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6125 struct btrfs_root *root,
6126 struct btrfs_path *path,
6127 struct walk_control *wc, int max_level)
6129 int level = wc->level;
6132 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6133 while (level < max_level && path->nodes[level]) {
6135 if (path->slots[level] + 1 <
6136 btrfs_header_nritems(path->nodes[level])) {
6137 path->slots[level]++;
6140 ret = walk_up_proc(trans, root, path, wc);
6144 if (path->locks[level]) {
6145 btrfs_tree_unlock(path->nodes[level]);
6146 path->locks[level] = 0;
6148 free_extent_buffer(path->nodes[level]);
6149 path->nodes[level] = NULL;
6157 * drop a subvolume tree.
6159 * this function traverses the tree freeing any blocks that only
6160 * referenced by the tree.
6162 * when a shared tree block is found. this function decreases its
6163 * reference count by one. if update_ref is true, this function
6164 * also make sure backrefs for the shared block and all lower level
6165 * blocks are properly updated.
6167 int btrfs_drop_snapshot(struct btrfs_root *root,
6168 struct btrfs_block_rsv *block_rsv, int update_ref)
6170 struct btrfs_path *path;
6171 struct btrfs_trans_handle *trans;
6172 struct btrfs_root *tree_root = root->fs_info->tree_root;
6173 struct btrfs_root_item *root_item = &root->root_item;
6174 struct walk_control *wc;
6175 struct btrfs_key key;
6180 path = btrfs_alloc_path();
6183 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6186 trans = btrfs_start_transaction(tree_root, 0);
6188 trans->block_rsv = block_rsv;
6190 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6191 level = btrfs_header_level(root->node);
6192 path->nodes[level] = btrfs_lock_root_node(root);
6193 btrfs_set_lock_blocking(path->nodes[level]);
6194 path->slots[level] = 0;
6195 path->locks[level] = 1;
6196 memset(&wc->update_progress, 0,
6197 sizeof(wc->update_progress));
6199 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6200 memcpy(&wc->update_progress, &key,
6201 sizeof(wc->update_progress));
6203 level = root_item->drop_level;
6205 path->lowest_level = level;
6206 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6207 path->lowest_level = 0;
6215 * unlock our path, this is safe because only this
6216 * function is allowed to delete this snapshot
6218 btrfs_unlock_up_safe(path, 0);
6220 level = btrfs_header_level(root->node);
6222 btrfs_tree_lock(path->nodes[level]);
6223 btrfs_set_lock_blocking(path->nodes[level]);
6225 ret = btrfs_lookup_extent_info(trans, root,
6226 path->nodes[level]->start,
6227 path->nodes[level]->len,
6231 BUG_ON(wc->refs[level] == 0);
6233 if (level == root_item->drop_level)
6236 btrfs_tree_unlock(path->nodes[level]);
6237 WARN_ON(wc->refs[level] != 1);
6243 wc->shared_level = -1;
6244 wc->stage = DROP_REFERENCE;
6245 wc->update_ref = update_ref;
6247 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6250 ret = walk_down_tree(trans, root, path, wc);
6256 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6263 BUG_ON(wc->stage != DROP_REFERENCE);
6267 if (wc->stage == DROP_REFERENCE) {
6269 btrfs_node_key(path->nodes[level],
6270 &root_item->drop_progress,
6271 path->slots[level]);
6272 root_item->drop_level = level;
6275 BUG_ON(wc->level == 0);
6276 if (btrfs_should_end_transaction(trans, tree_root)) {
6277 ret = btrfs_update_root(trans, tree_root,
6282 btrfs_end_transaction_throttle(trans, tree_root);
6283 trans = btrfs_start_transaction(tree_root, 0);
6285 trans->block_rsv = block_rsv;
6288 btrfs_release_path(root, path);
6291 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6294 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6295 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6299 ret = btrfs_del_orphan_item(trans, tree_root,
6300 root->root_key.objectid);
6305 if (root->in_radix) {
6306 btrfs_free_fs_root(tree_root->fs_info, root);
6308 free_extent_buffer(root->node);
6309 free_extent_buffer(root->commit_root);
6313 btrfs_end_transaction_throttle(trans, tree_root);
6315 btrfs_free_path(path);
6320 * drop subtree rooted at tree block 'node'.
6322 * NOTE: this function will unlock and release tree block 'node'
6324 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6325 struct btrfs_root *root,
6326 struct extent_buffer *node,
6327 struct extent_buffer *parent)
6329 struct btrfs_path *path;
6330 struct walk_control *wc;
6336 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6338 path = btrfs_alloc_path();
6341 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6344 btrfs_assert_tree_locked(parent);
6345 parent_level = btrfs_header_level(parent);
6346 extent_buffer_get(parent);
6347 path->nodes[parent_level] = parent;
6348 path->slots[parent_level] = btrfs_header_nritems(parent);
6350 btrfs_assert_tree_locked(node);
6351 level = btrfs_header_level(node);
6352 path->nodes[level] = node;
6353 path->slots[level] = 0;
6354 path->locks[level] = 1;
6356 wc->refs[parent_level] = 1;
6357 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6359 wc->shared_level = -1;
6360 wc->stage = DROP_REFERENCE;
6363 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6366 wret = walk_down_tree(trans, root, path, wc);
6372 wret = walk_up_tree(trans, root, path, wc, parent_level);
6380 btrfs_free_path(path);
6385 static unsigned long calc_ra(unsigned long start, unsigned long last,
6388 return min(last, start + nr - 1);
6391 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6396 unsigned long first_index;
6397 unsigned long last_index;
6400 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6401 struct file_ra_state *ra;
6402 struct btrfs_ordered_extent *ordered;
6403 unsigned int total_read = 0;
6404 unsigned int total_dirty = 0;
6407 ra = kzalloc(sizeof(*ra), GFP_NOFS);
6409 mutex_lock(&inode->i_mutex);
6410 first_index = start >> PAGE_CACHE_SHIFT;
6411 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6413 /* make sure the dirty trick played by the caller work */
6414 ret = invalidate_inode_pages2_range(inode->i_mapping,
6415 first_index, last_index);
6419 file_ra_state_init(ra, inode->i_mapping);
6421 for (i = first_index ; i <= last_index; i++) {
6422 if (total_read % ra->ra_pages == 0) {
6423 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6424 calc_ra(i, last_index, ra->ra_pages));
6428 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6430 page = grab_cache_page(inode->i_mapping, i);
6435 if (!PageUptodate(page)) {
6436 btrfs_readpage(NULL, page);
6438 if (!PageUptodate(page)) {
6440 page_cache_release(page);
6445 wait_on_page_writeback(page);
6447 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6448 page_end = page_start + PAGE_CACHE_SIZE - 1;
6449 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6451 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6453 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6455 page_cache_release(page);
6456 btrfs_start_ordered_extent(inode, ordered, 1);
6457 btrfs_put_ordered_extent(ordered);
6460 set_page_extent_mapped(page);
6462 if (i == first_index)
6463 set_extent_bits(io_tree, page_start, page_end,
6464 EXTENT_BOUNDARY, GFP_NOFS);
6465 btrfs_set_extent_delalloc(inode, page_start, page_end);
6467 set_page_dirty(page);
6470 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6472 page_cache_release(page);
6477 mutex_unlock(&inode->i_mutex);
6478 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6482 static noinline int relocate_data_extent(struct inode *reloc_inode,
6483 struct btrfs_key *extent_key,
6486 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6487 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6488 struct extent_map *em;
6489 u64 start = extent_key->objectid - offset;
6490 u64 end = start + extent_key->offset - 1;
6492 em = alloc_extent_map(GFP_NOFS);
6493 BUG_ON(!em || IS_ERR(em));
6496 em->len = extent_key->offset;
6497 em->block_len = extent_key->offset;
6498 em->block_start = extent_key->objectid;
6499 em->bdev = root->fs_info->fs_devices->latest_bdev;
6500 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6502 /* setup extent map to cheat btrfs_readpage */
6503 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6506 write_lock(&em_tree->lock);
6507 ret = add_extent_mapping(em_tree, em);
6508 write_unlock(&em_tree->lock);
6509 if (ret != -EEXIST) {
6510 free_extent_map(em);
6513 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6515 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6517 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6520 struct btrfs_ref_path {
6522 u64 nodes[BTRFS_MAX_LEVEL];
6524 u64 root_generation;
6531 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6532 u64 new_nodes[BTRFS_MAX_LEVEL];
6535 struct disk_extent {
6546 static int is_cowonly_root(u64 root_objectid)
6548 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6549 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6550 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6551 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6552 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6553 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6558 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6559 struct btrfs_root *extent_root,
6560 struct btrfs_ref_path *ref_path,
6563 struct extent_buffer *leaf;
6564 struct btrfs_path *path;
6565 struct btrfs_extent_ref *ref;
6566 struct btrfs_key key;
6567 struct btrfs_key found_key;
6573 path = btrfs_alloc_path();
6578 ref_path->lowest_level = -1;
6579 ref_path->current_level = -1;
6580 ref_path->shared_level = -1;
6584 level = ref_path->current_level - 1;
6585 while (level >= -1) {
6587 if (level < ref_path->lowest_level)
6591 bytenr = ref_path->nodes[level];
6593 bytenr = ref_path->extent_start;
6594 BUG_ON(bytenr == 0);
6596 parent = ref_path->nodes[level + 1];
6597 ref_path->nodes[level + 1] = 0;
6598 ref_path->current_level = level;
6599 BUG_ON(parent == 0);
6601 key.objectid = bytenr;
6602 key.offset = parent + 1;
6603 key.type = BTRFS_EXTENT_REF_KEY;
6605 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6610 leaf = path->nodes[0];
6611 nritems = btrfs_header_nritems(leaf);
6612 if (path->slots[0] >= nritems) {
6613 ret = btrfs_next_leaf(extent_root, path);
6618 leaf = path->nodes[0];
6621 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6622 if (found_key.objectid == bytenr &&
6623 found_key.type == BTRFS_EXTENT_REF_KEY) {
6624 if (level < ref_path->shared_level)
6625 ref_path->shared_level = level;
6630 btrfs_release_path(extent_root, path);
6633 /* reached lowest level */
6637 level = ref_path->current_level;
6638 while (level < BTRFS_MAX_LEVEL - 1) {
6642 bytenr = ref_path->nodes[level];
6644 bytenr = ref_path->extent_start;
6646 BUG_ON(bytenr == 0);
6648 key.objectid = bytenr;
6650 key.type = BTRFS_EXTENT_REF_KEY;
6652 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6656 leaf = path->nodes[0];
6657 nritems = btrfs_header_nritems(leaf);
6658 if (path->slots[0] >= nritems) {
6659 ret = btrfs_next_leaf(extent_root, path);
6663 /* the extent was freed by someone */
6664 if (ref_path->lowest_level == level)
6666 btrfs_release_path(extent_root, path);
6669 leaf = path->nodes[0];
6672 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6673 if (found_key.objectid != bytenr ||
6674 found_key.type != BTRFS_EXTENT_REF_KEY) {
6675 /* the extent was freed by someone */
6676 if (ref_path->lowest_level == level) {
6680 btrfs_release_path(extent_root, path);
6684 ref = btrfs_item_ptr(leaf, path->slots[0],
6685 struct btrfs_extent_ref);
6686 ref_objectid = btrfs_ref_objectid(leaf, ref);
6687 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6689 level = (int)ref_objectid;
6690 BUG_ON(level >= BTRFS_MAX_LEVEL);
6691 ref_path->lowest_level = level;
6692 ref_path->current_level = level;
6693 ref_path->nodes[level] = bytenr;
6695 WARN_ON(ref_objectid != level);
6698 WARN_ON(level != -1);
6702 if (ref_path->lowest_level == level) {
6703 ref_path->owner_objectid = ref_objectid;
6704 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6708 * the block is tree root or the block isn't in reference
6711 if (found_key.objectid == found_key.offset ||
6712 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6713 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6714 ref_path->root_generation =
6715 btrfs_ref_generation(leaf, ref);
6717 /* special reference from the tree log */
6718 ref_path->nodes[0] = found_key.offset;
6719 ref_path->current_level = 0;
6726 BUG_ON(ref_path->nodes[level] != 0);
6727 ref_path->nodes[level] = found_key.offset;
6728 ref_path->current_level = level;
6731 * the reference was created in the running transaction,
6732 * no need to continue walking up.
6734 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6735 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6736 ref_path->root_generation =
6737 btrfs_ref_generation(leaf, ref);
6742 btrfs_release_path(extent_root, path);
6745 /* reached max tree level, but no tree root found. */
6748 btrfs_free_path(path);
6752 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6753 struct btrfs_root *extent_root,
6754 struct btrfs_ref_path *ref_path,
6757 memset(ref_path, 0, sizeof(*ref_path));
6758 ref_path->extent_start = extent_start;
6760 return __next_ref_path(trans, extent_root, ref_path, 1);
6763 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6764 struct btrfs_root *extent_root,
6765 struct btrfs_ref_path *ref_path)
6767 return __next_ref_path(trans, extent_root, ref_path, 0);
6770 static noinline int get_new_locations(struct inode *reloc_inode,
6771 struct btrfs_key *extent_key,
6772 u64 offset, int no_fragment,
6773 struct disk_extent **extents,
6776 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6777 struct btrfs_path *path;
6778 struct btrfs_file_extent_item *fi;
6779 struct extent_buffer *leaf;
6780 struct disk_extent *exts = *extents;
6781 struct btrfs_key found_key;
6786 int max = *nr_extents;
6789 WARN_ON(!no_fragment && *extents);
6792 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6797 path = btrfs_alloc_path();
6800 cur_pos = extent_key->objectid - offset;
6801 last_byte = extent_key->objectid + extent_key->offset;
6802 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6812 leaf = path->nodes[0];
6813 nritems = btrfs_header_nritems(leaf);
6814 if (path->slots[0] >= nritems) {
6815 ret = btrfs_next_leaf(root, path);
6820 leaf = path->nodes[0];
6823 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6824 if (found_key.offset != cur_pos ||
6825 found_key.type != BTRFS_EXTENT_DATA_KEY ||
6826 found_key.objectid != reloc_inode->i_ino)
6829 fi = btrfs_item_ptr(leaf, path->slots[0],
6830 struct btrfs_file_extent_item);
6831 if (btrfs_file_extent_type(leaf, fi) !=
6832 BTRFS_FILE_EXTENT_REG ||
6833 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6837 struct disk_extent *old = exts;
6839 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6840 memcpy(exts, old, sizeof(*exts) * nr);
6841 if (old != *extents)
6845 exts[nr].disk_bytenr =
6846 btrfs_file_extent_disk_bytenr(leaf, fi);
6847 exts[nr].disk_num_bytes =
6848 btrfs_file_extent_disk_num_bytes(leaf, fi);
6849 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6850 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6851 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6852 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6853 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6854 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6856 BUG_ON(exts[nr].offset > 0);
6857 BUG_ON(exts[nr].compression || exts[nr].encryption);
6858 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6860 cur_pos += exts[nr].num_bytes;
6863 if (cur_pos + offset >= last_byte)
6873 BUG_ON(cur_pos + offset > last_byte);
6874 if (cur_pos + offset < last_byte) {
6880 btrfs_free_path(path);
6882 if (exts != *extents)
6891 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6892 struct btrfs_root *root,
6893 struct btrfs_path *path,
6894 struct btrfs_key *extent_key,
6895 struct btrfs_key *leaf_key,
6896 struct btrfs_ref_path *ref_path,
6897 struct disk_extent *new_extents,
6900 struct extent_buffer *leaf;
6901 struct btrfs_file_extent_item *fi;
6902 struct inode *inode = NULL;
6903 struct btrfs_key key;
6908 u64 search_end = (u64)-1;
6911 int extent_locked = 0;
6915 memcpy(&key, leaf_key, sizeof(key));
6916 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6917 if (key.objectid < ref_path->owner_objectid ||
6918 (key.objectid == ref_path->owner_objectid &&
6919 key.type < BTRFS_EXTENT_DATA_KEY)) {
6920 key.objectid = ref_path->owner_objectid;
6921 key.type = BTRFS_EXTENT_DATA_KEY;
6927 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6931 leaf = path->nodes[0];
6932 nritems = btrfs_header_nritems(leaf);
6934 if (extent_locked && ret > 0) {
6936 * the file extent item was modified by someone
6937 * before the extent got locked.
6939 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6940 lock_end, GFP_NOFS);
6944 if (path->slots[0] >= nritems) {
6945 if (++nr_scaned > 2)
6948 BUG_ON(extent_locked);
6949 ret = btrfs_next_leaf(root, path);
6954 leaf = path->nodes[0];
6955 nritems = btrfs_header_nritems(leaf);
6958 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6960 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6961 if ((key.objectid > ref_path->owner_objectid) ||
6962 (key.objectid == ref_path->owner_objectid &&
6963 key.type > BTRFS_EXTENT_DATA_KEY) ||
6964 key.offset >= search_end)
6968 if (inode && key.objectid != inode->i_ino) {
6969 BUG_ON(extent_locked);
6970 btrfs_release_path(root, path);
6971 mutex_unlock(&inode->i_mutex);
6977 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6982 fi = btrfs_item_ptr(leaf, path->slots[0],
6983 struct btrfs_file_extent_item);
6984 extent_type = btrfs_file_extent_type(leaf, fi);
6985 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6986 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6987 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6988 extent_key->objectid)) {
6994 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6995 ext_offset = btrfs_file_extent_offset(leaf, fi);
6997 if (search_end == (u64)-1) {
6998 search_end = key.offset - ext_offset +
6999 btrfs_file_extent_ram_bytes(leaf, fi);
7002 if (!extent_locked) {
7003 lock_start = key.offset;
7004 lock_end = lock_start + num_bytes - 1;
7006 if (lock_start > key.offset ||
7007 lock_end + 1 < key.offset + num_bytes) {
7008 unlock_extent(&BTRFS_I(inode)->io_tree,
7009 lock_start, lock_end, GFP_NOFS);
7015 btrfs_release_path(root, path);
7017 inode = btrfs_iget_locked(root->fs_info->sb,
7018 key.objectid, root);
7019 if (inode->i_state & I_NEW) {
7020 BTRFS_I(inode)->root = root;
7021 BTRFS_I(inode)->location.objectid =
7023 BTRFS_I(inode)->location.type =
7024 BTRFS_INODE_ITEM_KEY;
7025 BTRFS_I(inode)->location.offset = 0;
7026 btrfs_read_locked_inode(inode);
7027 unlock_new_inode(inode);
7030 * some code call btrfs_commit_transaction while
7031 * holding the i_mutex, so we can't use mutex_lock
7034 if (is_bad_inode(inode) ||
7035 !mutex_trylock(&inode->i_mutex)) {
7038 key.offset = (u64)-1;
7043 if (!extent_locked) {
7044 struct btrfs_ordered_extent *ordered;
7046 btrfs_release_path(root, path);
7048 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7049 lock_end, GFP_NOFS);
7050 ordered = btrfs_lookup_first_ordered_extent(inode,
7053 ordered->file_offset <= lock_end &&
7054 ordered->file_offset + ordered->len > lock_start) {
7055 unlock_extent(&BTRFS_I(inode)->io_tree,
7056 lock_start, lock_end, GFP_NOFS);
7057 btrfs_start_ordered_extent(inode, ordered, 1);
7058 btrfs_put_ordered_extent(ordered);
7059 key.offset += num_bytes;
7063 btrfs_put_ordered_extent(ordered);
7069 if (nr_extents == 1) {
7070 /* update extent pointer in place */
7071 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7072 new_extents[0].disk_bytenr);
7073 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7074 new_extents[0].disk_num_bytes);
7075 btrfs_mark_buffer_dirty(leaf);
7077 btrfs_drop_extent_cache(inode, key.offset,
7078 key.offset + num_bytes - 1, 0);
7080 ret = btrfs_inc_extent_ref(trans, root,
7081 new_extents[0].disk_bytenr,
7082 new_extents[0].disk_num_bytes,
7084 root->root_key.objectid,
7089 ret = btrfs_free_extent(trans, root,
7090 extent_key->objectid,
7093 btrfs_header_owner(leaf),
7094 btrfs_header_generation(leaf),
7098 btrfs_release_path(root, path);
7099 key.offset += num_bytes;
7107 * drop old extent pointer at first, then insert the
7108 * new pointers one bye one
7110 btrfs_release_path(root, path);
7111 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7112 key.offset + num_bytes,
7113 key.offset, &alloc_hint);
7116 for (i = 0; i < nr_extents; i++) {
7117 if (ext_offset >= new_extents[i].num_bytes) {
7118 ext_offset -= new_extents[i].num_bytes;
7121 extent_len = min(new_extents[i].num_bytes -
7122 ext_offset, num_bytes);
7124 ret = btrfs_insert_empty_item(trans, root,
7129 leaf = path->nodes[0];
7130 fi = btrfs_item_ptr(leaf, path->slots[0],
7131 struct btrfs_file_extent_item);
7132 btrfs_set_file_extent_generation(leaf, fi,
7134 btrfs_set_file_extent_type(leaf, fi,
7135 BTRFS_FILE_EXTENT_REG);
7136 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7137 new_extents[i].disk_bytenr);
7138 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7139 new_extents[i].disk_num_bytes);
7140 btrfs_set_file_extent_ram_bytes(leaf, fi,
7141 new_extents[i].ram_bytes);
7143 btrfs_set_file_extent_compression(leaf, fi,
7144 new_extents[i].compression);
7145 btrfs_set_file_extent_encryption(leaf, fi,
7146 new_extents[i].encryption);
7147 btrfs_set_file_extent_other_encoding(leaf, fi,
7148 new_extents[i].other_encoding);
7150 btrfs_set_file_extent_num_bytes(leaf, fi,
7152 ext_offset += new_extents[i].offset;
7153 btrfs_set_file_extent_offset(leaf, fi,
7155 btrfs_mark_buffer_dirty(leaf);
7157 btrfs_drop_extent_cache(inode, key.offset,
7158 key.offset + extent_len - 1, 0);
7160 ret = btrfs_inc_extent_ref(trans, root,
7161 new_extents[i].disk_bytenr,
7162 new_extents[i].disk_num_bytes,
7164 root->root_key.objectid,
7165 trans->transid, key.objectid);
7167 btrfs_release_path(root, path);
7169 inode_add_bytes(inode, extent_len);
7172 num_bytes -= extent_len;
7173 key.offset += extent_len;
7178 BUG_ON(i >= nr_extents);
7182 if (extent_locked) {
7183 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7184 lock_end, GFP_NOFS);
7188 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7189 key.offset >= search_end)
7196 btrfs_release_path(root, path);
7198 mutex_unlock(&inode->i_mutex);
7199 if (extent_locked) {
7200 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7201 lock_end, GFP_NOFS);
7208 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7209 struct btrfs_root *root,
7210 struct extent_buffer *buf, u64 orig_start)
7215 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7216 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7218 level = btrfs_header_level(buf);
7220 struct btrfs_leaf_ref *ref;
7221 struct btrfs_leaf_ref *orig_ref;
7223 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7227 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7229 btrfs_free_leaf_ref(root, orig_ref);
7233 ref->nritems = orig_ref->nritems;
7234 memcpy(ref->extents, orig_ref->extents,
7235 sizeof(ref->extents[0]) * ref->nritems);
7237 btrfs_free_leaf_ref(root, orig_ref);
7239 ref->root_gen = trans->transid;
7240 ref->bytenr = buf->start;
7241 ref->owner = btrfs_header_owner(buf);
7242 ref->generation = btrfs_header_generation(buf);
7244 ret = btrfs_add_leaf_ref(root, ref, 0);
7246 btrfs_free_leaf_ref(root, ref);
7251 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7252 struct extent_buffer *leaf,
7253 struct btrfs_block_group_cache *group,
7254 struct btrfs_root *target_root)
7256 struct btrfs_key key;
7257 struct inode *inode = NULL;
7258 struct btrfs_file_extent_item *fi;
7259 struct extent_state *cached_state = NULL;
7261 u64 skip_objectid = 0;
7265 nritems = btrfs_header_nritems(leaf);
7266 for (i = 0; i < nritems; i++) {
7267 btrfs_item_key_to_cpu(leaf, &key, i);
7268 if (key.objectid == skip_objectid ||
7269 key.type != BTRFS_EXTENT_DATA_KEY)
7271 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7272 if (btrfs_file_extent_type(leaf, fi) ==
7273 BTRFS_FILE_EXTENT_INLINE)
7275 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7277 if (!inode || inode->i_ino != key.objectid) {
7279 inode = btrfs_ilookup(target_root->fs_info->sb,
7280 key.objectid, target_root, 1);
7283 skip_objectid = key.objectid;
7286 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7288 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7289 key.offset + num_bytes - 1, 0, &cached_state,
7291 btrfs_drop_extent_cache(inode, key.offset,
7292 key.offset + num_bytes - 1, 1);
7293 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7294 key.offset + num_bytes - 1, &cached_state,
7302 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7303 struct btrfs_root *root,
7304 struct extent_buffer *leaf,
7305 struct btrfs_block_group_cache *group,
7306 struct inode *reloc_inode)
7308 struct btrfs_key key;
7309 struct btrfs_key extent_key;
7310 struct btrfs_file_extent_item *fi;
7311 struct btrfs_leaf_ref *ref;
7312 struct disk_extent *new_extent;
7321 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7322 BUG_ON(!new_extent);
7324 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7328 nritems = btrfs_header_nritems(leaf);
7329 for (i = 0; i < nritems; i++) {
7330 btrfs_item_key_to_cpu(leaf, &key, i);
7331 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7333 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7334 if (btrfs_file_extent_type(leaf, fi) ==
7335 BTRFS_FILE_EXTENT_INLINE)
7337 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7338 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7343 if (bytenr >= group->key.objectid + group->key.offset ||
7344 bytenr + num_bytes <= group->key.objectid)
7347 extent_key.objectid = bytenr;
7348 extent_key.offset = num_bytes;
7349 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7351 ret = get_new_locations(reloc_inode, &extent_key,
7352 group->key.objectid, 1,
7353 &new_extent, &nr_extent);
7358 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7359 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7360 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7361 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7363 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7364 new_extent->disk_bytenr);
7365 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7366 new_extent->disk_num_bytes);
7367 btrfs_mark_buffer_dirty(leaf);
7369 ret = btrfs_inc_extent_ref(trans, root,
7370 new_extent->disk_bytenr,
7371 new_extent->disk_num_bytes,
7373 root->root_key.objectid,
7374 trans->transid, key.objectid);
7377 ret = btrfs_free_extent(trans, root,
7378 bytenr, num_bytes, leaf->start,
7379 btrfs_header_owner(leaf),
7380 btrfs_header_generation(leaf),
7386 BUG_ON(ext_index + 1 != ref->nritems);
7387 btrfs_free_leaf_ref(root, ref);
7391 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7392 struct btrfs_root *root)
7394 struct btrfs_root *reloc_root;
7397 if (root->reloc_root) {
7398 reloc_root = root->reloc_root;
7399 root->reloc_root = NULL;
7400 list_add(&reloc_root->dead_list,
7401 &root->fs_info->dead_reloc_roots);
7403 btrfs_set_root_bytenr(&reloc_root->root_item,
7404 reloc_root->node->start);
7405 btrfs_set_root_level(&root->root_item,
7406 btrfs_header_level(reloc_root->node));
7407 memset(&reloc_root->root_item.drop_progress, 0,
7408 sizeof(struct btrfs_disk_key));
7409 reloc_root->root_item.drop_level = 0;
7411 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7412 &reloc_root->root_key,
7413 &reloc_root->root_item);
7419 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7421 struct btrfs_trans_handle *trans;
7422 struct btrfs_root *reloc_root;
7423 struct btrfs_root *prev_root = NULL;
7424 struct list_head dead_roots;
7428 INIT_LIST_HEAD(&dead_roots);
7429 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7431 while (!list_empty(&dead_roots)) {
7432 reloc_root = list_entry(dead_roots.prev,
7433 struct btrfs_root, dead_list);
7434 list_del_init(&reloc_root->dead_list);
7436 BUG_ON(reloc_root->commit_root != NULL);
7438 trans = btrfs_join_transaction(root, 1);
7441 mutex_lock(&root->fs_info->drop_mutex);
7442 ret = btrfs_drop_snapshot(trans, reloc_root);
7445 mutex_unlock(&root->fs_info->drop_mutex);
7447 nr = trans->blocks_used;
7448 ret = btrfs_end_transaction(trans, root);
7450 btrfs_btree_balance_dirty(root, nr);
7453 free_extent_buffer(reloc_root->node);
7455 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7456 &reloc_root->root_key);
7458 mutex_unlock(&root->fs_info->drop_mutex);
7460 nr = trans->blocks_used;
7461 ret = btrfs_end_transaction(trans, root);
7463 btrfs_btree_balance_dirty(root, nr);
7466 prev_root = reloc_root;
7469 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7475 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7477 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7481 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7483 struct btrfs_root *reloc_root;
7484 struct btrfs_trans_handle *trans;
7485 struct btrfs_key location;
7489 mutex_lock(&root->fs_info->tree_reloc_mutex);
7490 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7492 found = !list_empty(&root->fs_info->dead_reloc_roots);
7493 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7496 trans = btrfs_start_transaction(root, 1);
7498 ret = btrfs_commit_transaction(trans, root);
7502 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7503 location.offset = (u64)-1;
7504 location.type = BTRFS_ROOT_ITEM_KEY;
7506 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7507 BUG_ON(!reloc_root);
7508 btrfs_orphan_cleanup(reloc_root);
7512 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7513 struct btrfs_root *root)
7515 struct btrfs_root *reloc_root;
7516 struct extent_buffer *eb;
7517 struct btrfs_root_item *root_item;
7518 struct btrfs_key root_key;
7521 BUG_ON(!root->ref_cows);
7522 if (root->reloc_root)
7525 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7528 ret = btrfs_copy_root(trans, root, root->commit_root,
7529 &eb, BTRFS_TREE_RELOC_OBJECTID);
7532 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7533 root_key.offset = root->root_key.objectid;
7534 root_key.type = BTRFS_ROOT_ITEM_KEY;
7536 memcpy(root_item, &root->root_item, sizeof(root_item));
7537 btrfs_set_root_refs(root_item, 0);
7538 btrfs_set_root_bytenr(root_item, eb->start);
7539 btrfs_set_root_level(root_item, btrfs_header_level(eb));
7540 btrfs_set_root_generation(root_item, trans->transid);
7542 btrfs_tree_unlock(eb);
7543 free_extent_buffer(eb);
7545 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7546 &root_key, root_item);
7550 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7552 BUG_ON(!reloc_root);
7553 reloc_root->last_trans = trans->transid;
7554 reloc_root->commit_root = NULL;
7555 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7557 root->reloc_root = reloc_root;
7562 * Core function of space balance.
7564 * The idea is using reloc trees to relocate tree blocks in reference
7565 * counted roots. There is one reloc tree for each subvol, and all
7566 * reloc trees share same root key objectid. Reloc trees are snapshots
7567 * of the latest committed roots of subvols (root->commit_root).
7569 * To relocate a tree block referenced by a subvol, there are two steps.
7570 * COW the block through subvol's reloc tree, then update block pointer
7571 * in the subvol to point to the new block. Since all reloc trees share
7572 * same root key objectid, doing special handing for tree blocks owned
7573 * by them is easy. Once a tree block has been COWed in one reloc tree,
7574 * we can use the resulting new block directly when the same block is
7575 * required to COW again through other reloc trees. By this way, relocated
7576 * tree blocks are shared between reloc trees, so they are also shared
7579 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7580 struct btrfs_root *root,
7581 struct btrfs_path *path,
7582 struct btrfs_key *first_key,
7583 struct btrfs_ref_path *ref_path,
7584 struct btrfs_block_group_cache *group,
7585 struct inode *reloc_inode)
7587 struct btrfs_root *reloc_root;
7588 struct extent_buffer *eb = NULL;
7589 struct btrfs_key *keys;
7593 int lowest_level = 0;
7596 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7597 lowest_level = ref_path->owner_objectid;
7599 if (!root->ref_cows) {
7600 path->lowest_level = lowest_level;
7601 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7603 path->lowest_level = 0;
7604 btrfs_release_path(root, path);
7608 mutex_lock(&root->fs_info->tree_reloc_mutex);
7609 ret = init_reloc_tree(trans, root);
7611 reloc_root = root->reloc_root;
7613 shared_level = ref_path->shared_level;
7614 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7616 keys = ref_path->node_keys;
7617 nodes = ref_path->new_nodes;
7618 memset(&keys[shared_level + 1], 0,
7619 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7620 memset(&nodes[shared_level + 1], 0,
7621 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7623 if (nodes[lowest_level] == 0) {
7624 path->lowest_level = lowest_level;
7625 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7628 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7629 eb = path->nodes[level];
7630 if (!eb || eb == reloc_root->node)
7632 nodes[level] = eb->start;
7634 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7636 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7639 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7640 eb = path->nodes[0];
7641 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7642 group, reloc_inode);
7645 btrfs_release_path(reloc_root, path);
7647 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7653 * replace tree blocks in the fs tree with tree blocks in
7656 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7659 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7660 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7663 extent_buffer_get(path->nodes[0]);
7664 eb = path->nodes[0];
7665 btrfs_release_path(reloc_root, path);
7666 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7668 free_extent_buffer(eb);
7671 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7672 path->lowest_level = 0;
7676 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7677 struct btrfs_root *root,
7678 struct btrfs_path *path,
7679 struct btrfs_key *first_key,
7680 struct btrfs_ref_path *ref_path)
7684 ret = relocate_one_path(trans, root, path, first_key,
7685 ref_path, NULL, NULL);
7691 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7692 struct btrfs_root *extent_root,
7693 struct btrfs_path *path,
7694 struct btrfs_key *extent_key)
7698 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7701 ret = btrfs_del_item(trans, extent_root, path);
7703 btrfs_release_path(extent_root, path);
7707 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7708 struct btrfs_ref_path *ref_path)
7710 struct btrfs_key root_key;
7712 root_key.objectid = ref_path->root_objectid;
7713 root_key.type = BTRFS_ROOT_ITEM_KEY;
7714 if (is_cowonly_root(ref_path->root_objectid))
7715 root_key.offset = 0;
7717 root_key.offset = (u64)-1;
7719 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7722 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7723 struct btrfs_path *path,
7724 struct btrfs_key *extent_key,
7725 struct btrfs_block_group_cache *group,
7726 struct inode *reloc_inode, int pass)
7728 struct btrfs_trans_handle *trans;
7729 struct btrfs_root *found_root;
7730 struct btrfs_ref_path *ref_path = NULL;
7731 struct disk_extent *new_extents = NULL;
7736 struct btrfs_key first_key;
7740 trans = btrfs_start_transaction(extent_root, 1);
7743 if (extent_key->objectid == 0) {
7744 ret = del_extent_zero(trans, extent_root, path, extent_key);
7748 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7754 for (loops = 0; ; loops++) {
7756 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7757 extent_key->objectid);
7759 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7766 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7767 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7770 found_root = read_ref_root(extent_root->fs_info, ref_path);
7771 BUG_ON(!found_root);
7773 * for reference counted tree, only process reference paths
7774 * rooted at the latest committed root.
7776 if (found_root->ref_cows &&
7777 ref_path->root_generation != found_root->root_key.offset)
7780 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7783 * copy data extents to new locations
7785 u64 group_start = group->key.objectid;
7786 ret = relocate_data_extent(reloc_inode,
7795 level = ref_path->owner_objectid;
7798 if (prev_block != ref_path->nodes[level]) {
7799 struct extent_buffer *eb;
7800 u64 block_start = ref_path->nodes[level];
7801 u64 block_size = btrfs_level_size(found_root, level);
7803 eb = read_tree_block(found_root, block_start,
7805 btrfs_tree_lock(eb);
7806 BUG_ON(level != btrfs_header_level(eb));
7809 btrfs_item_key_to_cpu(eb, &first_key, 0);
7811 btrfs_node_key_to_cpu(eb, &first_key, 0);
7813 btrfs_tree_unlock(eb);
7814 free_extent_buffer(eb);
7815 prev_block = block_start;
7818 mutex_lock(&extent_root->fs_info->trans_mutex);
7819 btrfs_record_root_in_trans(found_root);
7820 mutex_unlock(&extent_root->fs_info->trans_mutex);
7821 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7823 * try to update data extent references while
7824 * keeping metadata shared between snapshots.
7827 ret = relocate_one_path(trans, found_root,
7828 path, &first_key, ref_path,
7829 group, reloc_inode);
7835 * use fallback method to process the remaining
7839 u64 group_start = group->key.objectid;
7840 new_extents = kmalloc(sizeof(*new_extents),
7843 ret = get_new_locations(reloc_inode,
7851 ret = replace_one_extent(trans, found_root,
7853 &first_key, ref_path,
7854 new_extents, nr_extents);
7856 ret = relocate_tree_block(trans, found_root, path,
7857 &first_key, ref_path);
7864 btrfs_end_transaction(trans, extent_root);
7871 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7874 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7875 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7877 num_devices = root->fs_info->fs_devices->rw_devices;
7878 if (num_devices == 1) {
7879 stripped |= BTRFS_BLOCK_GROUP_DUP;
7880 stripped = flags & ~stripped;
7882 /* turn raid0 into single device chunks */
7883 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7886 /* turn mirroring into duplication */
7887 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7888 BTRFS_BLOCK_GROUP_RAID10))
7889 return stripped | BTRFS_BLOCK_GROUP_DUP;
7892 /* they already had raid on here, just return */
7893 if (flags & stripped)
7896 stripped |= BTRFS_BLOCK_GROUP_DUP;
7897 stripped = flags & ~stripped;
7899 /* switch duplicated blocks with raid1 */
7900 if (flags & BTRFS_BLOCK_GROUP_DUP)
7901 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7903 /* turn single device chunks into raid0 */
7904 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7909 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7911 struct btrfs_space_info *sinfo = cache->space_info;
7918 spin_lock(&sinfo->lock);
7919 spin_lock(&cache->lock);
7920 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7921 cache->bytes_super - btrfs_block_group_used(&cache->item);
7923 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7924 sinfo->bytes_may_use + sinfo->bytes_readonly +
7925 cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7926 sinfo->bytes_readonly += num_bytes;
7927 sinfo->bytes_reserved += cache->reserved_pinned;
7928 cache->reserved_pinned = 0;
7932 spin_unlock(&cache->lock);
7933 spin_unlock(&sinfo->lock);
7937 int btrfs_set_block_group_ro(struct btrfs_root *root,
7938 struct btrfs_block_group_cache *cache)
7941 struct btrfs_trans_handle *trans;
7947 trans = btrfs_join_transaction(root, 1);
7948 BUG_ON(IS_ERR(trans));
7950 alloc_flags = update_block_group_flags(root, cache->flags);
7951 if (alloc_flags != cache->flags)
7952 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7954 ret = set_block_group_ro(cache);
7957 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7958 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7961 ret = set_block_group_ro(cache);
7963 btrfs_end_transaction(trans, root);
7967 int btrfs_set_block_group_rw(struct btrfs_root *root,
7968 struct btrfs_block_group_cache *cache)
7970 struct btrfs_space_info *sinfo = cache->space_info;
7975 spin_lock(&sinfo->lock);
7976 spin_lock(&cache->lock);
7977 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7978 cache->bytes_super - btrfs_block_group_used(&cache->item);
7979 sinfo->bytes_readonly -= num_bytes;
7981 spin_unlock(&cache->lock);
7982 spin_unlock(&sinfo->lock);
7987 * checks to see if its even possible to relocate this block group.
7989 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7990 * ok to go ahead and try.
7992 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7994 struct btrfs_block_group_cache *block_group;
7995 struct btrfs_space_info *space_info;
7996 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7997 struct btrfs_device *device;
8001 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8003 /* odd, couldn't find the block group, leave it alone */
8007 /* no bytes used, we're good */
8008 if (!btrfs_block_group_used(&block_group->item))
8011 space_info = block_group->space_info;
8012 spin_lock(&space_info->lock);
8014 full = space_info->full;
8017 * if this is the last block group we have in this space, we can't
8018 * relocate it unless we're able to allocate a new chunk below.
8020 * Otherwise, we need to make sure we have room in the space to handle
8021 * all of the extents from this block group. If we can, we're good
8023 if ((space_info->total_bytes != block_group->key.offset) &&
8024 (space_info->bytes_used + space_info->bytes_reserved +
8025 space_info->bytes_pinned + space_info->bytes_readonly +
8026 btrfs_block_group_used(&block_group->item) <
8027 space_info->total_bytes)) {
8028 spin_unlock(&space_info->lock);
8031 spin_unlock(&space_info->lock);
8034 * ok we don't have enough space, but maybe we have free space on our
8035 * devices to allocate new chunks for relocation, so loop through our
8036 * alloc devices and guess if we have enough space. However, if we
8037 * were marked as full, then we know there aren't enough chunks, and we
8044 mutex_lock(&root->fs_info->chunk_mutex);
8045 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8046 u64 min_free = btrfs_block_group_used(&block_group->item);
8047 u64 dev_offset, max_avail;
8050 * check to make sure we can actually find a chunk with enough
8051 * space to fit our block group in.
8053 if (device->total_bytes > device->bytes_used + min_free) {
8054 ret = find_free_dev_extent(NULL, device, min_free,
8055 &dev_offset, &max_avail);
8061 mutex_unlock(&root->fs_info->chunk_mutex);
8063 btrfs_put_block_group(block_group);
8067 static int find_first_block_group(struct btrfs_root *root,
8068 struct btrfs_path *path, struct btrfs_key *key)
8071 struct btrfs_key found_key;
8072 struct extent_buffer *leaf;
8075 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8080 slot = path->slots[0];
8081 leaf = path->nodes[0];
8082 if (slot >= btrfs_header_nritems(leaf)) {
8083 ret = btrfs_next_leaf(root, path);
8090 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8092 if (found_key.objectid >= key->objectid &&
8093 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8103 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8105 struct btrfs_block_group_cache *block_group;
8109 struct inode *inode;
8111 block_group = btrfs_lookup_first_block_group(info, last);
8112 while (block_group) {
8113 spin_lock(&block_group->lock);
8114 if (block_group->iref)
8116 spin_unlock(&block_group->lock);
8117 block_group = next_block_group(info->tree_root,
8127 inode = block_group->inode;
8128 block_group->iref = 0;
8129 block_group->inode = NULL;
8130 spin_unlock(&block_group->lock);
8132 last = block_group->key.objectid + block_group->key.offset;
8133 btrfs_put_block_group(block_group);
8137 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8139 struct btrfs_block_group_cache *block_group;
8140 struct btrfs_space_info *space_info;
8141 struct btrfs_caching_control *caching_ctl;
8144 down_write(&info->extent_commit_sem);
8145 while (!list_empty(&info->caching_block_groups)) {
8146 caching_ctl = list_entry(info->caching_block_groups.next,
8147 struct btrfs_caching_control, list);
8148 list_del(&caching_ctl->list);
8149 put_caching_control(caching_ctl);
8151 up_write(&info->extent_commit_sem);
8153 spin_lock(&info->block_group_cache_lock);
8154 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8155 block_group = rb_entry(n, struct btrfs_block_group_cache,
8157 rb_erase(&block_group->cache_node,
8158 &info->block_group_cache_tree);
8159 spin_unlock(&info->block_group_cache_lock);
8161 down_write(&block_group->space_info->groups_sem);
8162 list_del(&block_group->list);
8163 up_write(&block_group->space_info->groups_sem);
8165 if (block_group->cached == BTRFS_CACHE_STARTED)
8166 wait_block_group_cache_done(block_group);
8168 btrfs_remove_free_space_cache(block_group);
8169 btrfs_put_block_group(block_group);
8171 spin_lock(&info->block_group_cache_lock);
8173 spin_unlock(&info->block_group_cache_lock);
8175 /* now that all the block groups are freed, go through and
8176 * free all the space_info structs. This is only called during
8177 * the final stages of unmount, and so we know nobody is
8178 * using them. We call synchronize_rcu() once before we start,
8179 * just to be on the safe side.
8183 release_global_block_rsv(info);
8185 while(!list_empty(&info->space_info)) {
8186 space_info = list_entry(info->space_info.next,
8187 struct btrfs_space_info,
8189 if (space_info->bytes_pinned > 0 ||
8190 space_info->bytes_reserved > 0) {
8192 dump_space_info(space_info, 0, 0);
8194 list_del(&space_info->list);
8200 static void __link_block_group(struct btrfs_space_info *space_info,
8201 struct btrfs_block_group_cache *cache)
8203 int index = get_block_group_index(cache);
8205 down_write(&space_info->groups_sem);
8206 list_add_tail(&cache->list, &space_info->block_groups[index]);
8207 up_write(&space_info->groups_sem);
8210 int btrfs_read_block_groups(struct btrfs_root *root)
8212 struct btrfs_path *path;
8214 struct btrfs_block_group_cache *cache;
8215 struct btrfs_fs_info *info = root->fs_info;
8216 struct btrfs_space_info *space_info;
8217 struct btrfs_key key;
8218 struct btrfs_key found_key;
8219 struct extent_buffer *leaf;
8223 root = info->extent_root;
8226 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8227 path = btrfs_alloc_path();
8231 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8232 if (cache_gen != 0 &&
8233 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8235 if (btrfs_test_opt(root, CLEAR_CACHE))
8237 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8238 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8241 ret = find_first_block_group(root, path, &key);
8247 leaf = path->nodes[0];
8248 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8249 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8255 atomic_set(&cache->count, 1);
8256 spin_lock_init(&cache->lock);
8257 spin_lock_init(&cache->tree_lock);
8258 cache->fs_info = info;
8259 INIT_LIST_HEAD(&cache->list);
8260 INIT_LIST_HEAD(&cache->cluster_list);
8263 cache->disk_cache_state = BTRFS_DC_CLEAR;
8266 * we only want to have 32k of ram per block group for keeping
8267 * track of free space, and if we pass 1/2 of that we want to
8268 * start converting things over to using bitmaps
8270 cache->extents_thresh = ((1024 * 32) / 2) /
8271 sizeof(struct btrfs_free_space);
8273 read_extent_buffer(leaf, &cache->item,
8274 btrfs_item_ptr_offset(leaf, path->slots[0]),
8275 sizeof(cache->item));
8276 memcpy(&cache->key, &found_key, sizeof(found_key));
8278 key.objectid = found_key.objectid + found_key.offset;
8279 btrfs_release_path(root, path);
8280 cache->flags = btrfs_block_group_flags(&cache->item);
8281 cache->sectorsize = root->sectorsize;
8284 * check for two cases, either we are full, and therefore
8285 * don't need to bother with the caching work since we won't
8286 * find any space, or we are empty, and we can just add all
8287 * the space in and be done with it. This saves us _alot_ of
8288 * time, particularly in the full case.
8290 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8291 exclude_super_stripes(root, cache);
8292 cache->last_byte_to_unpin = (u64)-1;
8293 cache->cached = BTRFS_CACHE_FINISHED;
8294 free_excluded_extents(root, cache);
8295 } else if (btrfs_block_group_used(&cache->item) == 0) {
8296 exclude_super_stripes(root, cache);
8297 cache->last_byte_to_unpin = (u64)-1;
8298 cache->cached = BTRFS_CACHE_FINISHED;
8299 add_new_free_space(cache, root->fs_info,
8301 found_key.objectid +
8303 free_excluded_extents(root, cache);
8306 ret = update_space_info(info, cache->flags, found_key.offset,
8307 btrfs_block_group_used(&cache->item),
8310 cache->space_info = space_info;
8311 spin_lock(&cache->space_info->lock);
8312 cache->space_info->bytes_readonly += cache->bytes_super;
8313 spin_unlock(&cache->space_info->lock);
8315 __link_block_group(space_info, cache);
8317 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8320 set_avail_alloc_bits(root->fs_info, cache->flags);
8321 if (btrfs_chunk_readonly(root, cache->key.objectid))
8322 set_block_group_ro(cache);
8325 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8326 if (!(get_alloc_profile(root, space_info->flags) &
8327 (BTRFS_BLOCK_GROUP_RAID10 |
8328 BTRFS_BLOCK_GROUP_RAID1 |
8329 BTRFS_BLOCK_GROUP_DUP)))
8332 * avoid allocating from un-mirrored block group if there are
8333 * mirrored block groups.
8335 list_for_each_entry(cache, &space_info->block_groups[3], list)
8336 set_block_group_ro(cache);
8337 list_for_each_entry(cache, &space_info->block_groups[4], list)
8338 set_block_group_ro(cache);
8341 init_global_block_rsv(info);
8344 btrfs_free_path(path);
8348 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8349 struct btrfs_root *root, u64 bytes_used,
8350 u64 type, u64 chunk_objectid, u64 chunk_offset,
8354 struct btrfs_root *extent_root;
8355 struct btrfs_block_group_cache *cache;
8357 extent_root = root->fs_info->extent_root;
8359 root->fs_info->last_trans_log_full_commit = trans->transid;
8361 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8365 cache->key.objectid = chunk_offset;
8366 cache->key.offset = size;
8367 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8368 cache->sectorsize = root->sectorsize;
8369 cache->fs_info = root->fs_info;
8372 * we only want to have 32k of ram per block group for keeping track
8373 * of free space, and if we pass 1/2 of that we want to start
8374 * converting things over to using bitmaps
8376 cache->extents_thresh = ((1024 * 32) / 2) /
8377 sizeof(struct btrfs_free_space);
8378 atomic_set(&cache->count, 1);
8379 spin_lock_init(&cache->lock);
8380 spin_lock_init(&cache->tree_lock);
8381 INIT_LIST_HEAD(&cache->list);
8382 INIT_LIST_HEAD(&cache->cluster_list);
8384 btrfs_set_block_group_used(&cache->item, bytes_used);
8385 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8386 cache->flags = type;
8387 btrfs_set_block_group_flags(&cache->item, type);
8389 cache->last_byte_to_unpin = (u64)-1;
8390 cache->cached = BTRFS_CACHE_FINISHED;
8391 exclude_super_stripes(root, cache);
8393 add_new_free_space(cache, root->fs_info, chunk_offset,
8394 chunk_offset + size);
8396 free_excluded_extents(root, cache);
8398 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8399 &cache->space_info);
8402 spin_lock(&cache->space_info->lock);
8403 cache->space_info->bytes_readonly += cache->bytes_super;
8404 spin_unlock(&cache->space_info->lock);
8406 __link_block_group(cache->space_info, cache);
8408 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8411 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8412 sizeof(cache->item));
8415 set_avail_alloc_bits(extent_root->fs_info, type);
8420 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8421 struct btrfs_root *root, u64 group_start)
8423 struct btrfs_path *path;
8424 struct btrfs_block_group_cache *block_group;
8425 struct btrfs_free_cluster *cluster;
8426 struct btrfs_root *tree_root = root->fs_info->tree_root;
8427 struct btrfs_key key;
8428 struct inode *inode;
8432 root = root->fs_info->extent_root;
8434 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8435 BUG_ON(!block_group);
8436 BUG_ON(!block_group->ro);
8438 memcpy(&key, &block_group->key, sizeof(key));
8439 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8440 BTRFS_BLOCK_GROUP_RAID1 |
8441 BTRFS_BLOCK_GROUP_RAID10))
8446 /* make sure this block group isn't part of an allocation cluster */
8447 cluster = &root->fs_info->data_alloc_cluster;
8448 spin_lock(&cluster->refill_lock);
8449 btrfs_return_cluster_to_free_space(block_group, cluster);
8450 spin_unlock(&cluster->refill_lock);
8453 * make sure this block group isn't part of a metadata
8454 * allocation cluster
8456 cluster = &root->fs_info->meta_alloc_cluster;
8457 spin_lock(&cluster->refill_lock);
8458 btrfs_return_cluster_to_free_space(block_group, cluster);
8459 spin_unlock(&cluster->refill_lock);
8461 path = btrfs_alloc_path();
8464 inode = lookup_free_space_inode(root, block_group, path);
8465 if (!IS_ERR(inode)) {
8466 btrfs_orphan_add(trans, inode);
8468 /* One for the block groups ref */
8469 spin_lock(&block_group->lock);
8470 if (block_group->iref) {
8471 block_group->iref = 0;
8472 block_group->inode = NULL;
8473 spin_unlock(&block_group->lock);
8476 spin_unlock(&block_group->lock);
8478 /* One for our lookup ref */
8482 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8483 key.offset = block_group->key.objectid;
8486 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8490 btrfs_release_path(tree_root, path);
8492 ret = btrfs_del_item(trans, tree_root, path);
8495 btrfs_release_path(tree_root, path);
8498 spin_lock(&root->fs_info->block_group_cache_lock);
8499 rb_erase(&block_group->cache_node,
8500 &root->fs_info->block_group_cache_tree);
8501 spin_unlock(&root->fs_info->block_group_cache_lock);
8503 down_write(&block_group->space_info->groups_sem);
8505 * we must use list_del_init so people can check to see if they
8506 * are still on the list after taking the semaphore
8508 list_del_init(&block_group->list);
8509 up_write(&block_group->space_info->groups_sem);
8511 if (block_group->cached == BTRFS_CACHE_STARTED)
8512 wait_block_group_cache_done(block_group);
8514 btrfs_remove_free_space_cache(block_group);
8516 spin_lock(&block_group->space_info->lock);
8517 block_group->space_info->total_bytes -= block_group->key.offset;
8518 block_group->space_info->bytes_readonly -= block_group->key.offset;
8519 block_group->space_info->disk_total -= block_group->key.offset * factor;
8520 spin_unlock(&block_group->space_info->lock);
8522 memcpy(&key, &block_group->key, sizeof(key));
8524 btrfs_clear_space_info_full(root->fs_info);
8526 btrfs_put_block_group(block_group);
8527 btrfs_put_block_group(block_group);
8529 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8535 ret = btrfs_del_item(trans, root, path);
8537 btrfs_free_path(path);