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>
29 #include "print-tree.h"
30 #include "transaction.h"
33 #include "free-space-cache.h"
35 static int update_block_group(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 u64 bytenr, u64 num_bytes, int alloc,
39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve);
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 pin_down_bytes(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 struct btrfs_path *path,
66 u64 bytenr, u64 num_bytes,
67 int is_data, int reserved,
68 struct extent_buffer **must_clean);
69 static int find_next_key(struct btrfs_path *path, int level,
70 struct btrfs_key *key);
73 block_group_cache_done(struct btrfs_block_group_cache *cache)
76 return cache->cached == BTRFS_CACHE_FINISHED;
79 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
81 return (cache->flags & bits) == bits;
85 * this adds the block group to the fs_info rb tree for the block group
88 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
89 struct btrfs_block_group_cache *block_group)
92 struct rb_node *parent = NULL;
93 struct btrfs_block_group_cache *cache;
95 spin_lock(&info->block_group_cache_lock);
96 p = &info->block_group_cache_tree.rb_node;
100 cache = rb_entry(parent, struct btrfs_block_group_cache,
102 if (block_group->key.objectid < cache->key.objectid) {
104 } else if (block_group->key.objectid > cache->key.objectid) {
107 spin_unlock(&info->block_group_cache_lock);
112 rb_link_node(&block_group->cache_node, parent, p);
113 rb_insert_color(&block_group->cache_node,
114 &info->block_group_cache_tree);
115 spin_unlock(&info->block_group_cache_lock);
121 * This will return the block group at or after bytenr if contains is 0, else
122 * it will return the block group that contains the bytenr
124 static struct btrfs_block_group_cache *
125 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
128 struct btrfs_block_group_cache *cache, *ret = NULL;
132 spin_lock(&info->block_group_cache_lock);
133 n = info->block_group_cache_tree.rb_node;
136 cache = rb_entry(n, struct btrfs_block_group_cache,
138 end = cache->key.objectid + cache->key.offset - 1;
139 start = cache->key.objectid;
141 if (bytenr < start) {
142 if (!contains && (!ret || start < ret->key.objectid))
145 } else if (bytenr > start) {
146 if (contains && bytenr <= end) {
157 atomic_inc(&ret->count);
158 spin_unlock(&info->block_group_cache_lock);
163 static int add_excluded_extent(struct btrfs_root *root,
164 u64 start, u64 num_bytes)
166 u64 end = start + num_bytes - 1;
167 set_extent_bits(&root->fs_info->freed_extents[0],
168 start, end, EXTENT_UPTODATE, GFP_NOFS);
169 set_extent_bits(&root->fs_info->freed_extents[1],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
174 static void free_excluded_extents(struct btrfs_root *root,
175 struct btrfs_block_group_cache *cache)
179 start = cache->key.objectid;
180 end = start + cache->key.offset - 1;
182 clear_extent_bits(&root->fs_info->freed_extents[0],
183 start, end, EXTENT_UPTODATE, GFP_NOFS);
184 clear_extent_bits(&root->fs_info->freed_extents[1],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
188 static int exclude_super_stripes(struct btrfs_root *root,
189 struct btrfs_block_group_cache *cache)
196 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
197 bytenr = btrfs_sb_offset(i);
198 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
199 cache->key.objectid, bytenr,
200 0, &logical, &nr, &stripe_len);
204 ret = add_excluded_extent(root, logical[nr],
214 static struct btrfs_caching_control *
215 get_caching_control(struct btrfs_block_group_cache *cache)
217 struct btrfs_caching_control *ctl;
219 spin_lock(&cache->lock);
220 if (cache->cached != BTRFS_CACHE_STARTED) {
221 spin_unlock(&cache->lock);
225 ctl = cache->caching_ctl;
226 atomic_inc(&ctl->count);
227 spin_unlock(&cache->lock);
231 static void put_caching_control(struct btrfs_caching_control *ctl)
233 if (atomic_dec_and_test(&ctl->count))
238 * this is only called by cache_block_group, since we could have freed extents
239 * we need to check the pinned_extents for any extents that can't be used yet
240 * since their free space will be released as soon as the transaction commits.
242 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
243 struct btrfs_fs_info *info, u64 start, u64 end)
245 u64 extent_start, extent_end, size, total_added = 0;
248 while (start < end) {
249 ret = find_first_extent_bit(info->pinned_extents, start,
250 &extent_start, &extent_end,
251 EXTENT_DIRTY | EXTENT_UPTODATE);
255 if (extent_start == start) {
256 start = extent_end + 1;
257 } else if (extent_start > start && extent_start < end) {
258 size = extent_start - start;
260 ret = btrfs_add_free_space(block_group, start,
263 start = extent_end + 1;
272 ret = btrfs_add_free_space(block_group, start, size);
279 static int caching_kthread(void *data)
281 struct btrfs_block_group_cache *block_group = data;
282 struct btrfs_fs_info *fs_info = block_group->fs_info;
283 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
284 struct btrfs_root *extent_root = fs_info->extent_root;
285 struct btrfs_path *path;
286 struct extent_buffer *leaf;
287 struct btrfs_key key;
293 path = btrfs_alloc_path();
297 exclude_super_stripes(extent_root, block_group);
299 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
302 * We don't want to deadlock with somebody trying to allocate a new
303 * extent for the extent root while also trying to search the extent
304 * root to add free space. So we skip locking and search the commit
305 * root, since its read-only
307 path->skip_locking = 1;
308 path->search_commit_root = 1;
313 key.type = BTRFS_EXTENT_ITEM_KEY;
315 mutex_lock(&caching_ctl->mutex);
316 /* need to make sure the commit_root doesn't disappear */
317 down_read(&fs_info->extent_commit_sem);
319 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
323 leaf = path->nodes[0];
324 nritems = btrfs_header_nritems(leaf);
328 if (fs_info->closing > 1) {
333 if (path->slots[0] < nritems) {
334 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
336 ret = find_next_key(path, 0, &key);
340 caching_ctl->progress = last;
341 btrfs_release_path(extent_root, path);
342 up_read(&fs_info->extent_commit_sem);
343 mutex_unlock(&caching_ctl->mutex);
344 if (btrfs_transaction_in_commit(fs_info))
351 if (key.objectid < block_group->key.objectid) {
356 if (key.objectid >= block_group->key.objectid +
357 block_group->key.offset)
360 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
361 total_found += add_new_free_space(block_group,
364 last = key.objectid + key.offset;
366 if (total_found > (1024 * 1024 * 2)) {
368 wake_up(&caching_ctl->wait);
375 total_found += add_new_free_space(block_group, fs_info, last,
376 block_group->key.objectid +
377 block_group->key.offset);
378 caching_ctl->progress = (u64)-1;
380 spin_lock(&block_group->lock);
381 block_group->caching_ctl = NULL;
382 block_group->cached = BTRFS_CACHE_FINISHED;
383 spin_unlock(&block_group->lock);
386 btrfs_free_path(path);
387 up_read(&fs_info->extent_commit_sem);
389 free_excluded_extents(extent_root, block_group);
391 mutex_unlock(&caching_ctl->mutex);
392 wake_up(&caching_ctl->wait);
394 put_caching_control(caching_ctl);
395 atomic_dec(&block_group->space_info->caching_threads);
399 static int cache_block_group(struct btrfs_block_group_cache *cache)
401 struct btrfs_fs_info *fs_info = cache->fs_info;
402 struct btrfs_caching_control *caching_ctl;
403 struct task_struct *tsk;
407 if (cache->cached != BTRFS_CACHE_NO)
410 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
411 BUG_ON(!caching_ctl);
413 INIT_LIST_HEAD(&caching_ctl->list);
414 mutex_init(&caching_ctl->mutex);
415 init_waitqueue_head(&caching_ctl->wait);
416 caching_ctl->block_group = cache;
417 caching_ctl->progress = cache->key.objectid;
418 /* one for caching kthread, one for caching block group list */
419 atomic_set(&caching_ctl->count, 2);
421 spin_lock(&cache->lock);
422 if (cache->cached != BTRFS_CACHE_NO) {
423 spin_unlock(&cache->lock);
427 cache->caching_ctl = caching_ctl;
428 cache->cached = BTRFS_CACHE_STARTED;
429 spin_unlock(&cache->lock);
431 down_write(&fs_info->extent_commit_sem);
432 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
433 up_write(&fs_info->extent_commit_sem);
435 atomic_inc(&cache->space_info->caching_threads);
437 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
438 cache->key.objectid);
441 printk(KERN_ERR "error running thread %d\n", ret);
449 * return the block group that starts at or after bytenr
451 static struct btrfs_block_group_cache *
452 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
454 struct btrfs_block_group_cache *cache;
456 cache = block_group_cache_tree_search(info, bytenr, 0);
462 * return the block group that contains the given bytenr
464 struct btrfs_block_group_cache *btrfs_lookup_block_group(
465 struct btrfs_fs_info *info,
468 struct btrfs_block_group_cache *cache;
470 cache = block_group_cache_tree_search(info, bytenr, 1);
475 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
477 if (atomic_dec_and_test(&cache->count))
481 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
484 struct list_head *head = &info->space_info;
485 struct btrfs_space_info *found;
488 list_for_each_entry_rcu(found, head, list) {
489 if (found->flags == flags) {
499 * after adding space to the filesystem, we need to clear the full flags
500 * on all the space infos.
502 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
504 struct list_head *head = &info->space_info;
505 struct btrfs_space_info *found;
508 list_for_each_entry_rcu(found, head, list)
513 static u64 div_factor(u64 num, int factor)
522 u64 btrfs_find_block_group(struct btrfs_root *root,
523 u64 search_start, u64 search_hint, int owner)
525 struct btrfs_block_group_cache *cache;
527 u64 last = max(search_hint, search_start);
534 cache = btrfs_lookup_first_block_group(root->fs_info, last);
538 spin_lock(&cache->lock);
539 last = cache->key.objectid + cache->key.offset;
540 used = btrfs_block_group_used(&cache->item);
542 if ((full_search || !cache->ro) &&
543 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
544 if (used + cache->pinned + cache->reserved <
545 div_factor(cache->key.offset, factor)) {
546 group_start = cache->key.objectid;
547 spin_unlock(&cache->lock);
548 btrfs_put_block_group(cache);
552 spin_unlock(&cache->lock);
553 btrfs_put_block_group(cache);
561 if (!full_search && factor < 10) {
571 /* simple helper to search for an existing extent at a given offset */
572 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
575 struct btrfs_key key;
576 struct btrfs_path *path;
578 path = btrfs_alloc_path();
580 key.objectid = start;
582 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
583 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
585 btrfs_free_path(path);
590 * Back reference rules. Back refs have three main goals:
592 * 1) differentiate between all holders of references to an extent so that
593 * when a reference is dropped we can make sure it was a valid reference
594 * before freeing the extent.
596 * 2) Provide enough information to quickly find the holders of an extent
597 * if we notice a given block is corrupted or bad.
599 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
600 * maintenance. This is actually the same as #2, but with a slightly
601 * different use case.
603 * There are two kinds of back refs. The implicit back refs is optimized
604 * for pointers in non-shared tree blocks. For a given pointer in a block,
605 * back refs of this kind provide information about the block's owner tree
606 * and the pointer's key. These information allow us to find the block by
607 * b-tree searching. The full back refs is for pointers in tree blocks not
608 * referenced by their owner trees. The location of tree block is recorded
609 * in the back refs. Actually the full back refs is generic, and can be
610 * used in all cases the implicit back refs is used. The major shortcoming
611 * of the full back refs is its overhead. Every time a tree block gets
612 * COWed, we have to update back refs entry for all pointers in it.
614 * For a newly allocated tree block, we use implicit back refs for
615 * pointers in it. This means most tree related operations only involve
616 * implicit back refs. For a tree block created in old transaction, the
617 * only way to drop a reference to it is COW it. So we can detect the
618 * event that tree block loses its owner tree's reference and do the
619 * back refs conversion.
621 * When a tree block is COW'd through a tree, there are four cases:
623 * The reference count of the block is one and the tree is the block's
624 * owner tree. Nothing to do in this case.
626 * The reference count of the block is one and the tree is not the
627 * block's owner tree. In this case, full back refs is used for pointers
628 * in the block. Remove these full back refs, add implicit back refs for
629 * every pointers in the new block.
631 * The reference count of the block is greater than one and the tree is
632 * the block's owner tree. In this case, implicit back refs is used for
633 * pointers in the block. Add full back refs for every pointers in the
634 * block, increase lower level extents' reference counts. The original
635 * implicit back refs are entailed to the new block.
637 * The reference count of the block is greater than one and the tree is
638 * not the block's owner tree. Add implicit back refs for every pointer in
639 * the new block, increase lower level extents' reference count.
641 * Back Reference Key composing:
643 * The key objectid corresponds to the first byte in the extent,
644 * The key type is used to differentiate between types of back refs.
645 * There are different meanings of the key offset for different types
648 * File extents can be referenced by:
650 * - multiple snapshots, subvolumes, or different generations in one subvol
651 * - different files inside a single subvolume
652 * - different offsets inside a file (bookend extents in file.c)
654 * The extent ref structure for the implicit back refs has fields for:
656 * - Objectid of the subvolume root
657 * - objectid of the file holding the reference
658 * - original offset in the file
659 * - how many bookend extents
661 * The key offset for the implicit back refs is hash of the first
664 * The extent ref structure for the full back refs has field for:
666 * - number of pointers in the tree leaf
668 * The key offset for the implicit back refs is the first byte of
671 * When a file extent is allocated, The implicit back refs is used.
672 * the fields are filled in:
674 * (root_key.objectid, inode objectid, offset in file, 1)
676 * When a file extent is removed file truncation, we find the
677 * corresponding implicit back refs and check the following fields:
679 * (btrfs_header_owner(leaf), inode objectid, offset in file)
681 * Btree extents can be referenced by:
683 * - Different subvolumes
685 * Both the implicit back refs and the full back refs for tree blocks
686 * only consist of key. The key offset for the implicit back refs is
687 * objectid of block's owner tree. The key offset for the full back refs
688 * is the first byte of parent block.
690 * When implicit back refs is used, information about the lowest key and
691 * level of the tree block are required. These information are stored in
692 * tree block info structure.
695 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
696 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
697 struct btrfs_root *root,
698 struct btrfs_path *path,
699 u64 owner, u32 extra_size)
701 struct btrfs_extent_item *item;
702 struct btrfs_extent_item_v0 *ei0;
703 struct btrfs_extent_ref_v0 *ref0;
704 struct btrfs_tree_block_info *bi;
705 struct extent_buffer *leaf;
706 struct btrfs_key key;
707 struct btrfs_key found_key;
708 u32 new_size = sizeof(*item);
712 leaf = path->nodes[0];
713 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
715 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
716 ei0 = btrfs_item_ptr(leaf, path->slots[0],
717 struct btrfs_extent_item_v0);
718 refs = btrfs_extent_refs_v0(leaf, ei0);
720 if (owner == (u64)-1) {
722 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
723 ret = btrfs_next_leaf(root, path);
727 leaf = path->nodes[0];
729 btrfs_item_key_to_cpu(leaf, &found_key,
731 BUG_ON(key.objectid != found_key.objectid);
732 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
736 ref0 = btrfs_item_ptr(leaf, path->slots[0],
737 struct btrfs_extent_ref_v0);
738 owner = btrfs_ref_objectid_v0(leaf, ref0);
742 btrfs_release_path(root, path);
744 if (owner < BTRFS_FIRST_FREE_OBJECTID)
745 new_size += sizeof(*bi);
747 new_size -= sizeof(*ei0);
748 ret = btrfs_search_slot(trans, root, &key, path,
749 new_size + extra_size, 1);
754 ret = btrfs_extend_item(trans, root, path, new_size);
757 leaf = path->nodes[0];
758 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
759 btrfs_set_extent_refs(leaf, item, refs);
760 /* FIXME: get real generation */
761 btrfs_set_extent_generation(leaf, item, 0);
762 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
763 btrfs_set_extent_flags(leaf, item,
764 BTRFS_EXTENT_FLAG_TREE_BLOCK |
765 BTRFS_BLOCK_FLAG_FULL_BACKREF);
766 bi = (struct btrfs_tree_block_info *)(item + 1);
767 /* FIXME: get first key of the block */
768 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
769 btrfs_set_tree_block_level(leaf, bi, (int)owner);
771 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
773 btrfs_mark_buffer_dirty(leaf);
778 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
780 u32 high_crc = ~(u32)0;
781 u32 low_crc = ~(u32)0;
784 lenum = cpu_to_le64(root_objectid);
785 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
786 lenum = cpu_to_le64(owner);
787 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
788 lenum = cpu_to_le64(offset);
789 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
791 return ((u64)high_crc << 31) ^ (u64)low_crc;
794 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
795 struct btrfs_extent_data_ref *ref)
797 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
798 btrfs_extent_data_ref_objectid(leaf, ref),
799 btrfs_extent_data_ref_offset(leaf, ref));
802 static int match_extent_data_ref(struct extent_buffer *leaf,
803 struct btrfs_extent_data_ref *ref,
804 u64 root_objectid, u64 owner, u64 offset)
806 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
807 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
808 btrfs_extent_data_ref_offset(leaf, ref) != offset)
813 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
814 struct btrfs_root *root,
815 struct btrfs_path *path,
816 u64 bytenr, u64 parent,
818 u64 owner, u64 offset)
820 struct btrfs_key key;
821 struct btrfs_extent_data_ref *ref;
822 struct extent_buffer *leaf;
828 key.objectid = bytenr;
830 key.type = BTRFS_SHARED_DATA_REF_KEY;
833 key.type = BTRFS_EXTENT_DATA_REF_KEY;
834 key.offset = hash_extent_data_ref(root_objectid,
839 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
848 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
849 key.type = BTRFS_EXTENT_REF_V0_KEY;
850 btrfs_release_path(root, path);
851 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
862 leaf = path->nodes[0];
863 nritems = btrfs_header_nritems(leaf);
865 if (path->slots[0] >= nritems) {
866 ret = btrfs_next_leaf(root, path);
872 leaf = path->nodes[0];
873 nritems = btrfs_header_nritems(leaf);
877 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
878 if (key.objectid != bytenr ||
879 key.type != BTRFS_EXTENT_DATA_REF_KEY)
882 ref = btrfs_item_ptr(leaf, path->slots[0],
883 struct btrfs_extent_data_ref);
885 if (match_extent_data_ref(leaf, ref, root_objectid,
888 btrfs_release_path(root, path);
900 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
901 struct btrfs_root *root,
902 struct btrfs_path *path,
903 u64 bytenr, u64 parent,
904 u64 root_objectid, u64 owner,
905 u64 offset, int refs_to_add)
907 struct btrfs_key key;
908 struct extent_buffer *leaf;
913 key.objectid = bytenr;
915 key.type = BTRFS_SHARED_DATA_REF_KEY;
917 size = sizeof(struct btrfs_shared_data_ref);
919 key.type = BTRFS_EXTENT_DATA_REF_KEY;
920 key.offset = hash_extent_data_ref(root_objectid,
922 size = sizeof(struct btrfs_extent_data_ref);
925 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
926 if (ret && ret != -EEXIST)
929 leaf = path->nodes[0];
931 struct btrfs_shared_data_ref *ref;
932 ref = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_shared_data_ref);
935 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
937 num_refs = btrfs_shared_data_ref_count(leaf, ref);
938 num_refs += refs_to_add;
939 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
942 struct btrfs_extent_data_ref *ref;
943 while (ret == -EEXIST) {
944 ref = btrfs_item_ptr(leaf, path->slots[0],
945 struct btrfs_extent_data_ref);
946 if (match_extent_data_ref(leaf, ref, root_objectid,
949 btrfs_release_path(root, path);
951 ret = btrfs_insert_empty_item(trans, root, path, &key,
953 if (ret && ret != -EEXIST)
956 leaf = path->nodes[0];
958 ref = btrfs_item_ptr(leaf, path->slots[0],
959 struct btrfs_extent_data_ref);
961 btrfs_set_extent_data_ref_root(leaf, ref,
963 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
964 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
965 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
967 num_refs = btrfs_extent_data_ref_count(leaf, ref);
968 num_refs += refs_to_add;
969 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
972 btrfs_mark_buffer_dirty(leaf);
975 btrfs_release_path(root, path);
979 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
984 struct btrfs_key key;
985 struct btrfs_extent_data_ref *ref1 = NULL;
986 struct btrfs_shared_data_ref *ref2 = NULL;
987 struct extent_buffer *leaf;
991 leaf = path->nodes[0];
992 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
994 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
995 ref1 = btrfs_item_ptr(leaf, path->slots[0],
996 struct btrfs_extent_data_ref);
997 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
998 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
999 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_shared_data_ref);
1001 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1003 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1004 struct btrfs_extent_ref_v0 *ref0;
1005 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_extent_ref_v0);
1007 num_refs = btrfs_ref_count_v0(leaf, ref0);
1013 BUG_ON(num_refs < refs_to_drop);
1014 num_refs -= refs_to_drop;
1016 if (num_refs == 0) {
1017 ret = btrfs_del_item(trans, root, path);
1019 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1020 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1021 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1022 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1023 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1025 struct btrfs_extent_ref_v0 *ref0;
1026 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_extent_ref_v0);
1028 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1031 btrfs_mark_buffer_dirty(leaf);
1036 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1037 struct btrfs_path *path,
1038 struct btrfs_extent_inline_ref *iref)
1040 struct btrfs_key key;
1041 struct extent_buffer *leaf;
1042 struct btrfs_extent_data_ref *ref1;
1043 struct btrfs_shared_data_ref *ref2;
1046 leaf = path->nodes[0];
1047 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1049 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1050 BTRFS_EXTENT_DATA_REF_KEY) {
1051 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1052 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1054 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1055 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1057 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1058 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1059 struct btrfs_extent_data_ref);
1060 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1061 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1062 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_shared_data_ref);
1064 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1065 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1066 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1067 struct btrfs_extent_ref_v0 *ref0;
1068 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1069 struct btrfs_extent_ref_v0);
1070 num_refs = btrfs_ref_count_v0(leaf, ref0);
1078 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1079 struct btrfs_root *root,
1080 struct btrfs_path *path,
1081 u64 bytenr, u64 parent,
1084 struct btrfs_key key;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1093 key.offset = root_objectid;
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1100 if (ret == -ENOENT && parent) {
1101 btrfs_release_path(root, path);
1102 key.type = BTRFS_EXTENT_REF_V0_KEY;
1103 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1111 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1112 struct btrfs_root *root,
1113 struct btrfs_path *path,
1114 u64 bytenr, u64 parent,
1117 struct btrfs_key key;
1120 key.objectid = bytenr;
1122 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1123 key.offset = parent;
1125 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1126 key.offset = root_objectid;
1129 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1130 btrfs_release_path(root, path);
1134 static inline int extent_ref_type(u64 parent, u64 owner)
1137 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1139 type = BTRFS_SHARED_BLOCK_REF_KEY;
1141 type = BTRFS_TREE_BLOCK_REF_KEY;
1144 type = BTRFS_SHARED_DATA_REF_KEY;
1146 type = BTRFS_EXTENT_DATA_REF_KEY;
1151 static int find_next_key(struct btrfs_path *path, int level,
1152 struct btrfs_key *key)
1155 for (; level < BTRFS_MAX_LEVEL; level++) {
1156 if (!path->nodes[level])
1158 if (path->slots[level] + 1 >=
1159 btrfs_header_nritems(path->nodes[level]))
1162 btrfs_item_key_to_cpu(path->nodes[level], key,
1163 path->slots[level] + 1);
1165 btrfs_node_key_to_cpu(path->nodes[level], key,
1166 path->slots[level] + 1);
1173 * look for inline back ref. if back ref is found, *ref_ret is set
1174 * to the address of inline back ref, and 0 is returned.
1176 * if back ref isn't found, *ref_ret is set to the address where it
1177 * should be inserted, and -ENOENT is returned.
1179 * if insert is true and there are too many inline back refs, the path
1180 * points to the extent item, and -EAGAIN is returned.
1182 * NOTE: inline back refs are ordered in the same way that back ref
1183 * items in the tree are ordered.
1185 static noinline_for_stack
1186 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1187 struct btrfs_root *root,
1188 struct btrfs_path *path,
1189 struct btrfs_extent_inline_ref **ref_ret,
1190 u64 bytenr, u64 num_bytes,
1191 u64 parent, u64 root_objectid,
1192 u64 owner, u64 offset, int insert)
1194 struct btrfs_key key;
1195 struct extent_buffer *leaf;
1196 struct btrfs_extent_item *ei;
1197 struct btrfs_extent_inline_ref *iref;
1208 key.objectid = bytenr;
1209 key.type = BTRFS_EXTENT_ITEM_KEY;
1210 key.offset = num_bytes;
1212 want = extent_ref_type(parent, owner);
1214 extra_size = btrfs_extent_inline_ref_size(want);
1215 path->keep_locks = 1;
1218 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1225 leaf = path->nodes[0];
1226 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1227 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1228 if (item_size < sizeof(*ei)) {
1233 ret = convert_extent_item_v0(trans, root, path, owner,
1239 leaf = path->nodes[0];
1240 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1243 BUG_ON(item_size < sizeof(*ei));
1245 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1246 flags = btrfs_extent_flags(leaf, ei);
1248 ptr = (unsigned long)(ei + 1);
1249 end = (unsigned long)ei + item_size;
1251 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1252 ptr += sizeof(struct btrfs_tree_block_info);
1255 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1264 iref = (struct btrfs_extent_inline_ref *)ptr;
1265 type = btrfs_extent_inline_ref_type(leaf, iref);
1269 ptr += btrfs_extent_inline_ref_size(type);
1273 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1274 struct btrfs_extent_data_ref *dref;
1275 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1276 if (match_extent_data_ref(leaf, dref, root_objectid,
1281 if (hash_extent_data_ref_item(leaf, dref) <
1282 hash_extent_data_ref(root_objectid, owner, offset))
1286 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1288 if (parent == ref_offset) {
1292 if (ref_offset < parent)
1295 if (root_objectid == ref_offset) {
1299 if (ref_offset < root_objectid)
1303 ptr += btrfs_extent_inline_ref_size(type);
1305 if (err == -ENOENT && insert) {
1306 if (item_size + extra_size >=
1307 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1312 * To add new inline back ref, we have to make sure
1313 * there is no corresponding back ref item.
1314 * For simplicity, we just do not add new inline back
1315 * ref if there is any kind of item for this block
1317 if (find_next_key(path, 0, &key) == 0 &&
1318 key.objectid == bytenr &&
1319 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1324 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1327 path->keep_locks = 0;
1328 btrfs_unlock_up_safe(path, 1);
1334 * helper to add new inline back ref
1336 static noinline_for_stack
1337 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 struct btrfs_extent_inline_ref *iref,
1341 u64 parent, u64 root_objectid,
1342 u64 owner, u64 offset, int refs_to_add,
1343 struct btrfs_delayed_extent_op *extent_op)
1345 struct extent_buffer *leaf;
1346 struct btrfs_extent_item *ei;
1349 unsigned long item_offset;
1355 leaf = path->nodes[0];
1356 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1357 item_offset = (unsigned long)iref - (unsigned long)ei;
1359 type = extent_ref_type(parent, owner);
1360 size = btrfs_extent_inline_ref_size(type);
1362 ret = btrfs_extend_item(trans, root, path, size);
1365 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1366 refs = btrfs_extent_refs(leaf, ei);
1367 refs += refs_to_add;
1368 btrfs_set_extent_refs(leaf, ei, refs);
1370 __run_delayed_extent_op(extent_op, leaf, ei);
1372 ptr = (unsigned long)ei + item_offset;
1373 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1374 if (ptr < end - size)
1375 memmove_extent_buffer(leaf, ptr + size, ptr,
1378 iref = (struct btrfs_extent_inline_ref *)ptr;
1379 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1380 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1381 struct btrfs_extent_data_ref *dref;
1382 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1383 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1384 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1385 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1386 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1387 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1388 struct btrfs_shared_data_ref *sref;
1389 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1390 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1391 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1392 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1393 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1395 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1397 btrfs_mark_buffer_dirty(leaf);
1401 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 struct btrfs_extent_inline_ref **ref_ret,
1405 u64 bytenr, u64 num_bytes, u64 parent,
1406 u64 root_objectid, u64 owner, u64 offset)
1410 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1411 bytenr, num_bytes, parent,
1412 root_objectid, owner, offset, 0);
1416 btrfs_release_path(root, path);
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1420 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1423 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1424 root_objectid, owner, offset);
1430 * helper to update/remove inline back ref
1432 static noinline_for_stack
1433 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1434 struct btrfs_root *root,
1435 struct btrfs_path *path,
1436 struct btrfs_extent_inline_ref *iref,
1438 struct btrfs_delayed_extent_op *extent_op)
1440 struct extent_buffer *leaf;
1441 struct btrfs_extent_item *ei;
1442 struct btrfs_extent_data_ref *dref = NULL;
1443 struct btrfs_shared_data_ref *sref = NULL;
1452 leaf = path->nodes[0];
1453 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1454 refs = btrfs_extent_refs(leaf, ei);
1455 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1456 refs += refs_to_mod;
1457 btrfs_set_extent_refs(leaf, ei, refs);
1459 __run_delayed_extent_op(extent_op, leaf, ei);
1461 type = btrfs_extent_inline_ref_type(leaf, iref);
1463 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1464 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1465 refs = btrfs_extent_data_ref_count(leaf, dref);
1466 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1467 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1468 refs = btrfs_shared_data_ref_count(leaf, sref);
1471 BUG_ON(refs_to_mod != -1);
1474 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1475 refs += refs_to_mod;
1478 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1479 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1481 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1483 size = btrfs_extent_inline_ref_size(type);
1484 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 ptr = (unsigned long)iref;
1486 end = (unsigned long)ei + item_size;
1487 if (ptr + size < end)
1488 memmove_extent_buffer(leaf, ptr, ptr + size,
1491 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1494 btrfs_mark_buffer_dirty(leaf);
1498 static noinline_for_stack
1499 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1500 struct btrfs_root *root,
1501 struct btrfs_path *path,
1502 u64 bytenr, u64 num_bytes, u64 parent,
1503 u64 root_objectid, u64 owner,
1504 u64 offset, int refs_to_add,
1505 struct btrfs_delayed_extent_op *extent_op)
1507 struct btrfs_extent_inline_ref *iref;
1510 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1511 bytenr, num_bytes, parent,
1512 root_objectid, owner, offset, 1);
1514 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1515 ret = update_inline_extent_backref(trans, root, path, iref,
1516 refs_to_add, extent_op);
1517 } else if (ret == -ENOENT) {
1518 ret = setup_inline_extent_backref(trans, root, path, iref,
1519 parent, root_objectid,
1520 owner, offset, refs_to_add,
1526 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1527 struct btrfs_root *root,
1528 struct btrfs_path *path,
1529 u64 bytenr, u64 parent, u64 root_objectid,
1530 u64 owner, u64 offset, int refs_to_add)
1533 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1534 BUG_ON(refs_to_add != 1);
1535 ret = insert_tree_block_ref(trans, root, path, bytenr,
1536 parent, root_objectid);
1538 ret = insert_extent_data_ref(trans, root, path, bytenr,
1539 parent, root_objectid,
1540 owner, offset, refs_to_add);
1545 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1546 struct btrfs_root *root,
1547 struct btrfs_path *path,
1548 struct btrfs_extent_inline_ref *iref,
1549 int refs_to_drop, int is_data)
1553 BUG_ON(!is_data && refs_to_drop != 1);
1555 ret = update_inline_extent_backref(trans, root, path, iref,
1556 -refs_to_drop, NULL);
1557 } else if (is_data) {
1558 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1560 ret = btrfs_del_item(trans, root, path);
1565 #ifdef BIO_RW_DISCARD
1566 static void btrfs_issue_discard(struct block_device *bdev,
1569 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1573 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1576 #ifdef BIO_RW_DISCARD
1578 u64 map_length = num_bytes;
1579 struct btrfs_multi_bio *multi = NULL;
1581 /* Tell the block device(s) that the sectors can be discarded */
1582 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1583 bytenr, &map_length, &multi, 0);
1585 struct btrfs_bio_stripe *stripe = multi->stripes;
1588 if (map_length > num_bytes)
1589 map_length = num_bytes;
1591 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1592 btrfs_issue_discard(stripe->dev->bdev,
1605 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1606 struct btrfs_root *root,
1607 u64 bytenr, u64 num_bytes, u64 parent,
1608 u64 root_objectid, u64 owner, u64 offset)
1611 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1612 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1614 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1615 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1616 parent, root_objectid, (int)owner,
1617 BTRFS_ADD_DELAYED_REF, NULL);
1619 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1620 parent, root_objectid, owner, offset,
1621 BTRFS_ADD_DELAYED_REF, NULL);
1626 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root,
1628 u64 bytenr, u64 num_bytes,
1629 u64 parent, u64 root_objectid,
1630 u64 owner, u64 offset, int refs_to_add,
1631 struct btrfs_delayed_extent_op *extent_op)
1633 struct btrfs_path *path;
1634 struct extent_buffer *leaf;
1635 struct btrfs_extent_item *item;
1640 path = btrfs_alloc_path();
1645 path->leave_spinning = 1;
1646 /* this will setup the path even if it fails to insert the back ref */
1647 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1648 path, bytenr, num_bytes, parent,
1649 root_objectid, owner, offset,
1650 refs_to_add, extent_op);
1654 if (ret != -EAGAIN) {
1659 leaf = path->nodes[0];
1660 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1661 refs = btrfs_extent_refs(leaf, item);
1662 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1664 __run_delayed_extent_op(extent_op, leaf, item);
1666 btrfs_mark_buffer_dirty(leaf);
1667 btrfs_release_path(root->fs_info->extent_root, path);
1670 path->leave_spinning = 1;
1672 /* now insert the actual backref */
1673 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1674 path, bytenr, parent, root_objectid,
1675 owner, offset, refs_to_add);
1678 btrfs_free_path(path);
1682 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1683 struct btrfs_root *root,
1684 struct btrfs_delayed_ref_node *node,
1685 struct btrfs_delayed_extent_op *extent_op,
1686 int insert_reserved)
1689 struct btrfs_delayed_data_ref *ref;
1690 struct btrfs_key ins;
1695 ins.objectid = node->bytenr;
1696 ins.offset = node->num_bytes;
1697 ins.type = BTRFS_EXTENT_ITEM_KEY;
1699 ref = btrfs_delayed_node_to_data_ref(node);
1700 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1701 parent = ref->parent;
1703 ref_root = ref->root;
1705 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1707 BUG_ON(extent_op->update_key);
1708 flags |= extent_op->flags_to_set;
1710 ret = alloc_reserved_file_extent(trans, root,
1711 parent, ref_root, flags,
1712 ref->objectid, ref->offset,
1713 &ins, node->ref_mod);
1714 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1715 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1716 node->num_bytes, parent,
1717 ref_root, ref->objectid,
1718 ref->offset, node->ref_mod,
1720 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1721 ret = __btrfs_free_extent(trans, root, node->bytenr,
1722 node->num_bytes, parent,
1723 ref_root, ref->objectid,
1724 ref->offset, node->ref_mod,
1732 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1733 struct extent_buffer *leaf,
1734 struct btrfs_extent_item *ei)
1736 u64 flags = btrfs_extent_flags(leaf, ei);
1737 if (extent_op->update_flags) {
1738 flags |= extent_op->flags_to_set;
1739 btrfs_set_extent_flags(leaf, ei, flags);
1742 if (extent_op->update_key) {
1743 struct btrfs_tree_block_info *bi;
1744 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1745 bi = (struct btrfs_tree_block_info *)(ei + 1);
1746 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1750 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1751 struct btrfs_root *root,
1752 struct btrfs_delayed_ref_node *node,
1753 struct btrfs_delayed_extent_op *extent_op)
1755 struct btrfs_key key;
1756 struct btrfs_path *path;
1757 struct btrfs_extent_item *ei;
1758 struct extent_buffer *leaf;
1763 path = btrfs_alloc_path();
1767 key.objectid = node->bytenr;
1768 key.type = BTRFS_EXTENT_ITEM_KEY;
1769 key.offset = node->num_bytes;
1772 path->leave_spinning = 1;
1773 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1784 leaf = path->nodes[0];
1785 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1787 if (item_size < sizeof(*ei)) {
1788 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1794 leaf = path->nodes[0];
1795 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1798 BUG_ON(item_size < sizeof(*ei));
1799 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1800 __run_delayed_extent_op(extent_op, leaf, ei);
1802 btrfs_mark_buffer_dirty(leaf);
1804 btrfs_free_path(path);
1808 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1809 struct btrfs_root *root,
1810 struct btrfs_delayed_ref_node *node,
1811 struct btrfs_delayed_extent_op *extent_op,
1812 int insert_reserved)
1815 struct btrfs_delayed_tree_ref *ref;
1816 struct btrfs_key ins;
1820 ins.objectid = node->bytenr;
1821 ins.offset = node->num_bytes;
1822 ins.type = BTRFS_EXTENT_ITEM_KEY;
1824 ref = btrfs_delayed_node_to_tree_ref(node);
1825 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1826 parent = ref->parent;
1828 ref_root = ref->root;
1830 BUG_ON(node->ref_mod != 1);
1831 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1832 BUG_ON(!extent_op || !extent_op->update_flags ||
1833 !extent_op->update_key);
1834 ret = alloc_reserved_tree_block(trans, root,
1836 extent_op->flags_to_set,
1839 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1840 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1841 node->num_bytes, parent, ref_root,
1842 ref->level, 0, 1, extent_op);
1843 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1844 ret = __btrfs_free_extent(trans, root, node->bytenr,
1845 node->num_bytes, parent, ref_root,
1846 ref->level, 0, 1, extent_op);
1854 /* helper function to actually process a single delayed ref entry */
1855 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1856 struct btrfs_root *root,
1857 struct btrfs_delayed_ref_node *node,
1858 struct btrfs_delayed_extent_op *extent_op,
1859 int insert_reserved)
1862 if (btrfs_delayed_ref_is_head(node)) {
1863 struct btrfs_delayed_ref_head *head;
1865 * we've hit the end of the chain and we were supposed
1866 * to insert this extent into the tree. But, it got
1867 * deleted before we ever needed to insert it, so all
1868 * we have to do is clean up the accounting
1871 head = btrfs_delayed_node_to_head(node);
1872 if (insert_reserved) {
1874 struct extent_buffer *must_clean = NULL;
1876 ret = pin_down_bytes(trans, root, NULL,
1877 node->bytenr, node->num_bytes,
1878 head->is_data, 1, &must_clean);
1883 clean_tree_block(NULL, root, must_clean);
1884 btrfs_tree_unlock(must_clean);
1885 free_extent_buffer(must_clean);
1887 if (head->is_data) {
1888 ret = btrfs_del_csums(trans, root,
1894 ret = btrfs_free_reserved_extent(root,
1900 mutex_unlock(&head->mutex);
1904 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1905 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1906 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1908 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1909 node->type == BTRFS_SHARED_DATA_REF_KEY)
1910 ret = run_delayed_data_ref(trans, root, node, extent_op,
1917 static noinline struct btrfs_delayed_ref_node *
1918 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1920 struct rb_node *node;
1921 struct btrfs_delayed_ref_node *ref;
1922 int action = BTRFS_ADD_DELAYED_REF;
1925 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1926 * this prevents ref count from going down to zero when
1927 * there still are pending delayed ref.
1929 node = rb_prev(&head->node.rb_node);
1933 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1935 if (ref->bytenr != head->node.bytenr)
1937 if (ref->action == action)
1939 node = rb_prev(node);
1941 if (action == BTRFS_ADD_DELAYED_REF) {
1942 action = BTRFS_DROP_DELAYED_REF;
1948 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct list_head *cluster)
1952 struct btrfs_delayed_ref_root *delayed_refs;
1953 struct btrfs_delayed_ref_node *ref;
1954 struct btrfs_delayed_ref_head *locked_ref = NULL;
1955 struct btrfs_delayed_extent_op *extent_op;
1958 int must_insert_reserved = 0;
1960 delayed_refs = &trans->transaction->delayed_refs;
1963 /* pick a new head ref from the cluster list */
1964 if (list_empty(cluster))
1967 locked_ref = list_entry(cluster->next,
1968 struct btrfs_delayed_ref_head, cluster);
1970 /* grab the lock that says we are going to process
1971 * all the refs for this head */
1972 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1975 * we may have dropped the spin lock to get the head
1976 * mutex lock, and that might have given someone else
1977 * time to free the head. If that's true, it has been
1978 * removed from our list and we can move on.
1980 if (ret == -EAGAIN) {
1988 * record the must insert reserved flag before we
1989 * drop the spin lock.
1991 must_insert_reserved = locked_ref->must_insert_reserved;
1992 locked_ref->must_insert_reserved = 0;
1994 extent_op = locked_ref->extent_op;
1995 locked_ref->extent_op = NULL;
1998 * locked_ref is the head node, so we have to go one
1999 * node back for any delayed ref updates
2001 ref = select_delayed_ref(locked_ref);
2003 /* All delayed refs have been processed, Go ahead
2004 * and send the head node to run_one_delayed_ref,
2005 * so that any accounting fixes can happen
2007 ref = &locked_ref->node;
2009 if (extent_op && must_insert_reserved) {
2015 spin_unlock(&delayed_refs->lock);
2017 ret = run_delayed_extent_op(trans, root,
2023 spin_lock(&delayed_refs->lock);
2027 list_del_init(&locked_ref->cluster);
2032 rb_erase(&ref->rb_node, &delayed_refs->root);
2033 delayed_refs->num_entries--;
2035 spin_unlock(&delayed_refs->lock);
2037 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2038 must_insert_reserved);
2041 btrfs_put_delayed_ref(ref);
2046 spin_lock(&delayed_refs->lock);
2052 * this starts processing the delayed reference count updates and
2053 * extent insertions we have queued up so far. count can be
2054 * 0, which means to process everything in the tree at the start
2055 * of the run (but not newly added entries), or it can be some target
2056 * number you'd like to process.
2058 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root, unsigned long count)
2061 struct rb_node *node;
2062 struct btrfs_delayed_ref_root *delayed_refs;
2063 struct btrfs_delayed_ref_node *ref;
2064 struct list_head cluster;
2066 int run_all = count == (unsigned long)-1;
2069 if (root == root->fs_info->extent_root)
2070 root = root->fs_info->tree_root;
2072 delayed_refs = &trans->transaction->delayed_refs;
2073 INIT_LIST_HEAD(&cluster);
2075 spin_lock(&delayed_refs->lock);
2077 count = delayed_refs->num_entries * 2;
2081 if (!(run_all || run_most) &&
2082 delayed_refs->num_heads_ready < 64)
2086 * go find something we can process in the rbtree. We start at
2087 * the beginning of the tree, and then build a cluster
2088 * of refs to process starting at the first one we are able to
2091 ret = btrfs_find_ref_cluster(trans, &cluster,
2092 delayed_refs->run_delayed_start);
2096 ret = run_clustered_refs(trans, root, &cluster);
2099 count -= min_t(unsigned long, ret, count);
2106 node = rb_first(&delayed_refs->root);
2109 count = (unsigned long)-1;
2112 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2114 if (btrfs_delayed_ref_is_head(ref)) {
2115 struct btrfs_delayed_ref_head *head;
2117 head = btrfs_delayed_node_to_head(ref);
2118 atomic_inc(&ref->refs);
2120 spin_unlock(&delayed_refs->lock);
2121 mutex_lock(&head->mutex);
2122 mutex_unlock(&head->mutex);
2124 btrfs_put_delayed_ref(ref);
2128 node = rb_next(node);
2130 spin_unlock(&delayed_refs->lock);
2131 schedule_timeout(1);
2135 spin_unlock(&delayed_refs->lock);
2139 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2140 struct btrfs_root *root,
2141 u64 bytenr, u64 num_bytes, u64 flags,
2144 struct btrfs_delayed_extent_op *extent_op;
2147 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2151 extent_op->flags_to_set = flags;
2152 extent_op->update_flags = 1;
2153 extent_op->update_key = 0;
2154 extent_op->is_data = is_data ? 1 : 0;
2156 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2162 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2163 struct btrfs_root *root,
2164 struct btrfs_path *path,
2165 u64 objectid, u64 offset, u64 bytenr)
2167 struct btrfs_delayed_ref_head *head;
2168 struct btrfs_delayed_ref_node *ref;
2169 struct btrfs_delayed_data_ref *data_ref;
2170 struct btrfs_delayed_ref_root *delayed_refs;
2171 struct rb_node *node;
2175 delayed_refs = &trans->transaction->delayed_refs;
2176 spin_lock(&delayed_refs->lock);
2177 head = btrfs_find_delayed_ref_head(trans, bytenr);
2181 if (!mutex_trylock(&head->mutex)) {
2182 atomic_inc(&head->node.refs);
2183 spin_unlock(&delayed_refs->lock);
2185 btrfs_release_path(root->fs_info->extent_root, path);
2187 mutex_lock(&head->mutex);
2188 mutex_unlock(&head->mutex);
2189 btrfs_put_delayed_ref(&head->node);
2193 node = rb_prev(&head->node.rb_node);
2197 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2199 if (ref->bytenr != bytenr)
2203 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2206 data_ref = btrfs_delayed_node_to_data_ref(ref);
2208 node = rb_prev(node);
2210 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2211 if (ref->bytenr == bytenr)
2215 if (data_ref->root != root->root_key.objectid ||
2216 data_ref->objectid != objectid || data_ref->offset != offset)
2221 mutex_unlock(&head->mutex);
2223 spin_unlock(&delayed_refs->lock);
2227 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2228 struct btrfs_root *root,
2229 struct btrfs_path *path,
2230 u64 objectid, u64 offset, u64 bytenr)
2232 struct btrfs_root *extent_root = root->fs_info->extent_root;
2233 struct extent_buffer *leaf;
2234 struct btrfs_extent_data_ref *ref;
2235 struct btrfs_extent_inline_ref *iref;
2236 struct btrfs_extent_item *ei;
2237 struct btrfs_key key;
2241 key.objectid = bytenr;
2242 key.offset = (u64)-1;
2243 key.type = BTRFS_EXTENT_ITEM_KEY;
2245 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2251 if (path->slots[0] == 0)
2255 leaf = path->nodes[0];
2256 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2258 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2262 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2263 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2264 if (item_size < sizeof(*ei)) {
2265 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2269 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2271 if (item_size != sizeof(*ei) +
2272 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2275 if (btrfs_extent_generation(leaf, ei) <=
2276 btrfs_root_last_snapshot(&root->root_item))
2279 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2280 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2281 BTRFS_EXTENT_DATA_REF_KEY)
2284 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2285 if (btrfs_extent_refs(leaf, ei) !=
2286 btrfs_extent_data_ref_count(leaf, ref) ||
2287 btrfs_extent_data_ref_root(leaf, ref) !=
2288 root->root_key.objectid ||
2289 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2290 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2298 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 u64 objectid, u64 offset, u64 bytenr)
2302 struct btrfs_path *path;
2306 path = btrfs_alloc_path();
2311 ret = check_committed_ref(trans, root, path, objectid,
2313 if (ret && ret != -ENOENT)
2316 ret2 = check_delayed_ref(trans, root, path, objectid,
2318 } while (ret2 == -EAGAIN);
2320 if (ret2 && ret2 != -ENOENT) {
2325 if (ret != -ENOENT || ret2 != -ENOENT)
2328 btrfs_free_path(path);
2333 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2334 struct extent_buffer *buf, u32 nr_extents)
2336 struct btrfs_key key;
2337 struct btrfs_file_extent_item *fi;
2345 if (!root->ref_cows)
2348 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2350 root_gen = root->root_key.offset;
2353 root_gen = trans->transid - 1;
2356 level = btrfs_header_level(buf);
2357 nritems = btrfs_header_nritems(buf);
2360 struct btrfs_leaf_ref *ref;
2361 struct btrfs_extent_info *info;
2363 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2369 ref->root_gen = root_gen;
2370 ref->bytenr = buf->start;
2371 ref->owner = btrfs_header_owner(buf);
2372 ref->generation = btrfs_header_generation(buf);
2373 ref->nritems = nr_extents;
2374 info = ref->extents;
2376 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2378 btrfs_item_key_to_cpu(buf, &key, i);
2379 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2381 fi = btrfs_item_ptr(buf, i,
2382 struct btrfs_file_extent_item);
2383 if (btrfs_file_extent_type(buf, fi) ==
2384 BTRFS_FILE_EXTENT_INLINE)
2386 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2387 if (disk_bytenr == 0)
2390 info->bytenr = disk_bytenr;
2392 btrfs_file_extent_disk_num_bytes(buf, fi);
2393 info->objectid = key.objectid;
2394 info->offset = key.offset;
2398 ret = btrfs_add_leaf_ref(root, ref, shared);
2399 if (ret == -EEXIST && shared) {
2400 struct btrfs_leaf_ref *old;
2401 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2403 btrfs_remove_leaf_ref(root, old);
2404 btrfs_free_leaf_ref(root, old);
2405 ret = btrfs_add_leaf_ref(root, ref, shared);
2408 btrfs_free_leaf_ref(root, ref);
2414 /* when a block goes through cow, we update the reference counts of
2415 * everything that block points to. The internal pointers of the block
2416 * can be in just about any order, and it is likely to have clusters of
2417 * things that are close together and clusters of things that are not.
2419 * To help reduce the seeks that come with updating all of these reference
2420 * counts, sort them by byte number before actual updates are done.
2422 * struct refsort is used to match byte number to slot in the btree block.
2423 * we sort based on the byte number and then use the slot to actually
2426 * struct refsort is smaller than strcut btrfs_item and smaller than
2427 * struct btrfs_key_ptr. Since we're currently limited to the page size
2428 * for a btree block, there's no way for a kmalloc of refsorts for a
2429 * single node to be bigger than a page.
2437 * for passing into sort()
2439 static int refsort_cmp(const void *a_void, const void *b_void)
2441 const struct refsort *a = a_void;
2442 const struct refsort *b = b_void;
2444 if (a->bytenr < b->bytenr)
2446 if (a->bytenr > b->bytenr)
2452 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2453 struct btrfs_root *root,
2454 struct extent_buffer *buf,
2455 int full_backref, int inc)
2462 struct btrfs_key key;
2463 struct btrfs_file_extent_item *fi;
2467 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2468 u64, u64, u64, u64, u64, u64);
2470 ref_root = btrfs_header_owner(buf);
2471 nritems = btrfs_header_nritems(buf);
2472 level = btrfs_header_level(buf);
2474 if (!root->ref_cows && level == 0)
2478 process_func = btrfs_inc_extent_ref;
2480 process_func = btrfs_free_extent;
2483 parent = buf->start;
2487 for (i = 0; i < nritems; i++) {
2489 btrfs_item_key_to_cpu(buf, &key, i);
2490 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2492 fi = btrfs_item_ptr(buf, i,
2493 struct btrfs_file_extent_item);
2494 if (btrfs_file_extent_type(buf, fi) ==
2495 BTRFS_FILE_EXTENT_INLINE)
2497 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2501 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2502 key.offset -= btrfs_file_extent_offset(buf, fi);
2503 ret = process_func(trans, root, bytenr, num_bytes,
2504 parent, ref_root, key.objectid,
2509 bytenr = btrfs_node_blockptr(buf, i);
2510 num_bytes = btrfs_level_size(root, level - 1);
2511 ret = process_func(trans, root, bytenr, num_bytes,
2512 parent, ref_root, level - 1, 0);
2523 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2524 struct extent_buffer *buf, int full_backref)
2526 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2529 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2530 struct extent_buffer *buf, int full_backref)
2532 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2535 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2536 struct btrfs_root *root,
2537 struct btrfs_path *path,
2538 struct btrfs_block_group_cache *cache)
2541 struct btrfs_root *extent_root = root->fs_info->extent_root;
2543 struct extent_buffer *leaf;
2545 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2550 leaf = path->nodes[0];
2551 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2552 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2553 btrfs_mark_buffer_dirty(leaf);
2554 btrfs_release_path(extent_root, path);
2562 static struct btrfs_block_group_cache *
2563 next_block_group(struct btrfs_root *root,
2564 struct btrfs_block_group_cache *cache)
2566 struct rb_node *node;
2567 spin_lock(&root->fs_info->block_group_cache_lock);
2568 node = rb_next(&cache->cache_node);
2569 btrfs_put_block_group(cache);
2571 cache = rb_entry(node, struct btrfs_block_group_cache,
2573 atomic_inc(&cache->count);
2576 spin_unlock(&root->fs_info->block_group_cache_lock);
2580 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2581 struct btrfs_root *root)
2583 struct btrfs_block_group_cache *cache;
2585 struct btrfs_path *path;
2588 path = btrfs_alloc_path();
2594 err = btrfs_run_delayed_refs(trans, root,
2599 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2603 cache = next_block_group(root, cache);
2613 last = cache->key.objectid + cache->key.offset;
2615 err = write_one_cache_group(trans, root, path, cache);
2617 btrfs_put_block_group(cache);
2620 btrfs_free_path(path);
2624 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2626 struct btrfs_block_group_cache *block_group;
2629 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2630 if (!block_group || block_group->ro)
2633 btrfs_put_block_group(block_group);
2637 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2638 u64 total_bytes, u64 bytes_used,
2639 struct btrfs_space_info **space_info)
2641 struct btrfs_space_info *found;
2643 found = __find_space_info(info, flags);
2645 spin_lock(&found->lock);
2646 found->total_bytes += total_bytes;
2647 found->bytes_used += bytes_used;
2649 spin_unlock(&found->lock);
2650 *space_info = found;
2653 found = kzalloc(sizeof(*found), GFP_NOFS);
2657 INIT_LIST_HEAD(&found->block_groups);
2658 init_rwsem(&found->groups_sem);
2659 spin_lock_init(&found->lock);
2660 found->flags = flags;
2661 found->total_bytes = total_bytes;
2662 found->bytes_used = bytes_used;
2663 found->bytes_pinned = 0;
2664 found->bytes_reserved = 0;
2665 found->bytes_readonly = 0;
2666 found->bytes_delalloc = 0;
2668 found->force_alloc = 0;
2669 *space_info = found;
2670 list_add_rcu(&found->list, &info->space_info);
2671 atomic_set(&found->caching_threads, 0);
2675 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2677 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2678 BTRFS_BLOCK_GROUP_RAID1 |
2679 BTRFS_BLOCK_GROUP_RAID10 |
2680 BTRFS_BLOCK_GROUP_DUP);
2682 if (flags & BTRFS_BLOCK_GROUP_DATA)
2683 fs_info->avail_data_alloc_bits |= extra_flags;
2684 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2685 fs_info->avail_metadata_alloc_bits |= extra_flags;
2686 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2687 fs_info->avail_system_alloc_bits |= extra_flags;
2691 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2693 spin_lock(&cache->space_info->lock);
2694 spin_lock(&cache->lock);
2696 cache->space_info->bytes_readonly += cache->key.offset -
2697 btrfs_block_group_used(&cache->item);
2700 spin_unlock(&cache->lock);
2701 spin_unlock(&cache->space_info->lock);
2704 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2706 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2708 if (num_devices == 1)
2709 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2710 if (num_devices < 4)
2711 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2713 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2714 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2715 BTRFS_BLOCK_GROUP_RAID10))) {
2716 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2719 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2720 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2721 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2724 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2725 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2726 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2727 (flags & BTRFS_BLOCK_GROUP_DUP)))
2728 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2732 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2734 struct btrfs_fs_info *info = root->fs_info;
2738 alloc_profile = info->avail_data_alloc_bits &
2739 info->data_alloc_profile;
2740 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2741 } else if (root == root->fs_info->chunk_root) {
2742 alloc_profile = info->avail_system_alloc_bits &
2743 info->system_alloc_profile;
2744 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2746 alloc_profile = info->avail_metadata_alloc_bits &
2747 info->metadata_alloc_profile;
2748 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2751 return btrfs_reduce_alloc_profile(root, data);
2754 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2758 alloc_target = btrfs_get_alloc_profile(root, 1);
2759 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2764 * for now this just makes sure we have at least 5% of our metadata space free
2767 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2769 struct btrfs_fs_info *info = root->fs_info;
2770 struct btrfs_space_info *meta_sinfo;
2771 u64 alloc_target, thresh;
2772 int committed = 0, ret;
2774 /* get the space info for where the metadata will live */
2775 alloc_target = btrfs_get_alloc_profile(root, 0);
2776 meta_sinfo = __find_space_info(info, alloc_target);
2779 spin_lock(&meta_sinfo->lock);
2780 if (!meta_sinfo->full)
2781 thresh = meta_sinfo->total_bytes * 80;
2783 thresh = meta_sinfo->total_bytes * 95;
2785 do_div(thresh, 100);
2787 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2788 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2789 struct btrfs_trans_handle *trans;
2790 if (!meta_sinfo->full) {
2791 meta_sinfo->force_alloc = 1;
2792 spin_unlock(&meta_sinfo->lock);
2794 trans = btrfs_start_transaction(root, 1);
2798 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2799 2 * 1024 * 1024, alloc_target, 0);
2800 btrfs_end_transaction(trans, root);
2803 spin_unlock(&meta_sinfo->lock);
2807 trans = btrfs_join_transaction(root, 1);
2810 ret = btrfs_commit_transaction(trans, root);
2817 spin_unlock(&meta_sinfo->lock);
2823 * This will check the space that the inode allocates from to make sure we have
2824 * enough space for bytes.
2826 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2829 struct btrfs_space_info *data_sinfo;
2830 int ret = 0, committed = 0;
2832 /* make sure bytes are sectorsize aligned */
2833 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2835 data_sinfo = BTRFS_I(inode)->space_info;
2837 /* make sure we have enough space to handle the data first */
2838 spin_lock(&data_sinfo->lock);
2839 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2840 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2841 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2842 data_sinfo->bytes_may_use < bytes) {
2843 struct btrfs_trans_handle *trans;
2846 * if we don't have enough free bytes in this space then we need
2847 * to alloc a new chunk.
2849 if (!data_sinfo->full) {
2852 data_sinfo->force_alloc = 1;
2853 spin_unlock(&data_sinfo->lock);
2855 alloc_target = btrfs_get_alloc_profile(root, 1);
2856 trans = btrfs_start_transaction(root, 1);
2860 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2861 bytes + 2 * 1024 * 1024,
2863 btrfs_end_transaction(trans, root);
2868 spin_unlock(&data_sinfo->lock);
2870 /* commit the current transaction and try again */
2873 trans = btrfs_join_transaction(root, 1);
2876 ret = btrfs_commit_transaction(trans, root);
2882 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2883 ", %llu bytes_used, %llu bytes_reserved, "
2884 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2885 "%llu total\n", (unsigned long long)bytes,
2886 (unsigned long long)data_sinfo->bytes_delalloc,
2887 (unsigned long long)data_sinfo->bytes_used,
2888 (unsigned long long)data_sinfo->bytes_reserved,
2889 (unsigned long long)data_sinfo->bytes_pinned,
2890 (unsigned long long)data_sinfo->bytes_readonly,
2891 (unsigned long long)data_sinfo->bytes_may_use,
2892 (unsigned long long)data_sinfo->total_bytes);
2895 data_sinfo->bytes_may_use += bytes;
2896 BTRFS_I(inode)->reserved_bytes += bytes;
2897 spin_unlock(&data_sinfo->lock);
2899 return btrfs_check_metadata_free_space(root);
2903 * if there was an error for whatever reason after calling
2904 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2906 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2907 struct inode *inode, u64 bytes)
2909 struct btrfs_space_info *data_sinfo;
2911 /* make sure bytes are sectorsize aligned */
2912 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2914 data_sinfo = BTRFS_I(inode)->space_info;
2915 spin_lock(&data_sinfo->lock);
2916 data_sinfo->bytes_may_use -= bytes;
2917 BTRFS_I(inode)->reserved_bytes -= bytes;
2918 spin_unlock(&data_sinfo->lock);
2921 /* called when we are adding a delalloc extent to the inode's io_tree */
2922 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2925 struct btrfs_space_info *data_sinfo;
2927 /* get the space info for where this inode will be storing its data */
2928 data_sinfo = BTRFS_I(inode)->space_info;
2930 /* make sure we have enough space to handle the data first */
2931 spin_lock(&data_sinfo->lock);
2932 data_sinfo->bytes_delalloc += bytes;
2935 * we are adding a delalloc extent without calling
2936 * btrfs_check_data_free_space first. This happens on a weird
2937 * writepage condition, but shouldn't hurt our accounting
2939 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2940 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2941 BTRFS_I(inode)->reserved_bytes = 0;
2943 data_sinfo->bytes_may_use -= bytes;
2944 BTRFS_I(inode)->reserved_bytes -= bytes;
2947 spin_unlock(&data_sinfo->lock);
2950 /* called when we are clearing an delalloc extent from the inode's io_tree */
2951 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2954 struct btrfs_space_info *info;
2956 info = BTRFS_I(inode)->space_info;
2958 spin_lock(&info->lock);
2959 info->bytes_delalloc -= bytes;
2960 spin_unlock(&info->lock);
2963 static void force_metadata_allocation(struct btrfs_fs_info *info)
2965 struct list_head *head = &info->space_info;
2966 struct btrfs_space_info *found;
2969 list_for_each_entry_rcu(found, head, list) {
2970 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2971 found->force_alloc = 1;
2976 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2977 struct btrfs_root *extent_root, u64 alloc_bytes,
2978 u64 flags, int force)
2980 struct btrfs_space_info *space_info;
2981 struct btrfs_fs_info *fs_info = extent_root->fs_info;
2985 mutex_lock(&fs_info->chunk_mutex);
2987 flags = btrfs_reduce_alloc_profile(extent_root, flags);
2989 space_info = __find_space_info(extent_root->fs_info, flags);
2991 ret = update_space_info(extent_root->fs_info, flags,
2995 BUG_ON(!space_info);
2997 spin_lock(&space_info->lock);
2998 if (space_info->force_alloc) {
3000 space_info->force_alloc = 0;
3002 if (space_info->full) {
3003 spin_unlock(&space_info->lock);
3007 thresh = space_info->total_bytes - space_info->bytes_readonly;
3008 thresh = div_factor(thresh, 6);
3010 (space_info->bytes_used + space_info->bytes_pinned +
3011 space_info->bytes_reserved + alloc_bytes) < thresh) {
3012 spin_unlock(&space_info->lock);
3015 spin_unlock(&space_info->lock);
3018 * if we're doing a data chunk, go ahead and make sure that
3019 * we keep a reasonable number of metadata chunks allocated in the
3022 if (flags & BTRFS_BLOCK_GROUP_DATA) {
3023 fs_info->data_chunk_allocations++;
3024 if (!(fs_info->data_chunk_allocations %
3025 fs_info->metadata_ratio))
3026 force_metadata_allocation(fs_info);
3029 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3031 space_info->full = 1;
3033 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3037 static int update_block_group(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 bytenr, u64 num_bytes, int alloc,
3042 struct btrfs_block_group_cache *cache;
3043 struct btrfs_fs_info *info = root->fs_info;
3044 u64 total = num_bytes;
3048 /* block accounting for super block */
3049 spin_lock(&info->delalloc_lock);
3050 old_val = btrfs_super_bytes_used(&info->super_copy);
3052 old_val += num_bytes;
3054 old_val -= num_bytes;
3055 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3057 /* block accounting for root item */
3058 old_val = btrfs_root_used(&root->root_item);
3060 old_val += num_bytes;
3062 old_val -= num_bytes;
3063 btrfs_set_root_used(&root->root_item, old_val);
3064 spin_unlock(&info->delalloc_lock);
3067 cache = btrfs_lookup_block_group(info, bytenr);
3070 byte_in_group = bytenr - cache->key.objectid;
3071 WARN_ON(byte_in_group > cache->key.offset);
3073 spin_lock(&cache->space_info->lock);
3074 spin_lock(&cache->lock);
3076 old_val = btrfs_block_group_used(&cache->item);
3077 num_bytes = min(total, cache->key.offset - byte_in_group);
3079 old_val += num_bytes;
3080 btrfs_set_block_group_used(&cache->item, old_val);
3081 cache->reserved -= num_bytes;
3082 cache->space_info->bytes_used += num_bytes;
3083 cache->space_info->bytes_reserved -= num_bytes;
3085 cache->space_info->bytes_readonly -= num_bytes;
3086 spin_unlock(&cache->lock);
3087 spin_unlock(&cache->space_info->lock);
3089 old_val -= num_bytes;
3090 cache->space_info->bytes_used -= num_bytes;
3092 cache->space_info->bytes_readonly += num_bytes;
3093 btrfs_set_block_group_used(&cache->item, old_val);
3094 spin_unlock(&cache->lock);
3095 spin_unlock(&cache->space_info->lock);
3099 ret = btrfs_discard_extent(root, bytenr,
3103 ret = btrfs_add_free_space(cache, bytenr,
3108 btrfs_put_block_group(cache);
3110 bytenr += num_bytes;
3115 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3117 struct btrfs_block_group_cache *cache;
3120 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3124 bytenr = cache->key.objectid;
3125 btrfs_put_block_group(cache);
3131 * this function must be called within transaction
3133 int btrfs_pin_extent(struct btrfs_root *root,
3134 u64 bytenr, u64 num_bytes, int reserved)
3136 struct btrfs_fs_info *fs_info = root->fs_info;
3137 struct btrfs_block_group_cache *cache;
3139 cache = btrfs_lookup_block_group(fs_info, bytenr);
3142 spin_lock(&cache->space_info->lock);
3143 spin_lock(&cache->lock);
3144 cache->pinned += num_bytes;
3145 cache->space_info->bytes_pinned += num_bytes;
3147 cache->reserved -= num_bytes;
3148 cache->space_info->bytes_reserved -= num_bytes;
3150 spin_unlock(&cache->lock);
3151 spin_unlock(&cache->space_info->lock);
3153 btrfs_put_block_group(cache);
3155 set_extent_dirty(fs_info->pinned_extents,
3156 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3160 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3161 u64 num_bytes, int reserve)
3163 spin_lock(&cache->space_info->lock);
3164 spin_lock(&cache->lock);
3166 cache->reserved += num_bytes;
3167 cache->space_info->bytes_reserved += num_bytes;
3169 cache->reserved -= num_bytes;
3170 cache->space_info->bytes_reserved -= num_bytes;
3172 spin_unlock(&cache->lock);
3173 spin_unlock(&cache->space_info->lock);
3177 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3178 struct btrfs_root *root)
3180 struct btrfs_fs_info *fs_info = root->fs_info;
3181 struct btrfs_caching_control *next;
3182 struct btrfs_caching_control *caching_ctl;
3183 struct btrfs_block_group_cache *cache;
3185 down_write(&fs_info->extent_commit_sem);
3187 list_for_each_entry_safe(caching_ctl, next,
3188 &fs_info->caching_block_groups, list) {
3189 cache = caching_ctl->block_group;
3190 if (block_group_cache_done(cache)) {
3191 cache->last_byte_to_unpin = (u64)-1;
3192 list_del_init(&caching_ctl->list);
3193 put_caching_control(caching_ctl);
3195 cache->last_byte_to_unpin = caching_ctl->progress;
3199 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3200 fs_info->pinned_extents = &fs_info->freed_extents[1];
3202 fs_info->pinned_extents = &fs_info->freed_extents[0];
3204 up_write(&fs_info->extent_commit_sem);
3208 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3210 struct btrfs_fs_info *fs_info = root->fs_info;
3211 struct btrfs_block_group_cache *cache = NULL;
3214 while (start <= end) {
3216 start >= cache->key.objectid + cache->key.offset) {
3218 btrfs_put_block_group(cache);
3219 cache = btrfs_lookup_block_group(fs_info, start);
3223 len = cache->key.objectid + cache->key.offset - start;
3224 len = min(len, end + 1 - start);
3226 if (start < cache->last_byte_to_unpin) {
3227 len = min(len, cache->last_byte_to_unpin - start);
3228 btrfs_add_free_space(cache, start, len);
3231 spin_lock(&cache->space_info->lock);
3232 spin_lock(&cache->lock);
3233 cache->pinned -= len;
3234 cache->space_info->bytes_pinned -= len;
3235 spin_unlock(&cache->lock);
3236 spin_unlock(&cache->space_info->lock);
3242 btrfs_put_block_group(cache);
3246 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3247 struct btrfs_root *root)
3249 struct btrfs_fs_info *fs_info = root->fs_info;
3250 struct extent_io_tree *unpin;
3255 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3256 unpin = &fs_info->freed_extents[1];
3258 unpin = &fs_info->freed_extents[0];
3261 ret = find_first_extent_bit(unpin, 0, &start, &end,
3266 ret = btrfs_discard_extent(root, start, end + 1 - start);
3268 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3269 unpin_extent_range(root, start, end);
3276 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3277 struct btrfs_root *root,
3278 struct btrfs_path *path,
3279 u64 bytenr, u64 num_bytes,
3280 int is_data, int reserved,
3281 struct extent_buffer **must_clean)
3284 struct extent_buffer *buf;
3289 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3293 /* we can reuse a block if it hasn't been written
3294 * and it is from this transaction. We can't
3295 * reuse anything from the tree log root because
3296 * it has tiny sub-transactions.
3298 if (btrfs_buffer_uptodate(buf, 0) &&
3299 btrfs_try_tree_lock(buf)) {
3300 u64 header_owner = btrfs_header_owner(buf);
3301 u64 header_transid = btrfs_header_generation(buf);
3302 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3303 header_transid == trans->transid &&
3304 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3308 btrfs_tree_unlock(buf);
3310 free_extent_buffer(buf);
3313 btrfs_set_path_blocking(path);
3314 /* unlocks the pinned mutex */
3315 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3321 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3322 struct btrfs_root *root,
3323 u64 bytenr, u64 num_bytes, u64 parent,
3324 u64 root_objectid, u64 owner_objectid,
3325 u64 owner_offset, int refs_to_drop,
3326 struct btrfs_delayed_extent_op *extent_op)
3328 struct btrfs_key key;
3329 struct btrfs_path *path;
3330 struct btrfs_fs_info *info = root->fs_info;
3331 struct btrfs_root *extent_root = info->extent_root;
3332 struct extent_buffer *leaf;
3333 struct btrfs_extent_item *ei;
3334 struct btrfs_extent_inline_ref *iref;
3337 int extent_slot = 0;
3338 int found_extent = 0;
3343 path = btrfs_alloc_path();
3348 path->leave_spinning = 1;
3350 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3351 BUG_ON(!is_data && refs_to_drop != 1);
3353 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3354 bytenr, num_bytes, parent,
3355 root_objectid, owner_objectid,
3358 extent_slot = path->slots[0];
3359 while (extent_slot >= 0) {
3360 btrfs_item_key_to_cpu(path->nodes[0], &key,
3362 if (key.objectid != bytenr)
3364 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3365 key.offset == num_bytes) {
3369 if (path->slots[0] - extent_slot > 5)
3373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3374 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3375 if (found_extent && item_size < sizeof(*ei))
3378 if (!found_extent) {
3380 ret = remove_extent_backref(trans, extent_root, path,
3384 btrfs_release_path(extent_root, path);
3385 path->leave_spinning = 1;
3387 key.objectid = bytenr;
3388 key.type = BTRFS_EXTENT_ITEM_KEY;
3389 key.offset = num_bytes;
3391 ret = btrfs_search_slot(trans, extent_root,
3394 printk(KERN_ERR "umm, got %d back from search"
3395 ", was looking for %llu\n", ret,
3396 (unsigned long long)bytenr);
3397 btrfs_print_leaf(extent_root, path->nodes[0]);
3400 extent_slot = path->slots[0];
3403 btrfs_print_leaf(extent_root, path->nodes[0]);
3405 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3406 "parent %llu root %llu owner %llu offset %llu\n",
3407 (unsigned long long)bytenr,
3408 (unsigned long long)parent,
3409 (unsigned long long)root_objectid,
3410 (unsigned long long)owner_objectid,
3411 (unsigned long long)owner_offset);
3414 leaf = path->nodes[0];
3415 item_size = btrfs_item_size_nr(leaf, extent_slot);
3416 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3417 if (item_size < sizeof(*ei)) {
3418 BUG_ON(found_extent || extent_slot != path->slots[0]);
3419 ret = convert_extent_item_v0(trans, extent_root, path,
3423 btrfs_release_path(extent_root, path);
3424 path->leave_spinning = 1;
3426 key.objectid = bytenr;
3427 key.type = BTRFS_EXTENT_ITEM_KEY;
3428 key.offset = num_bytes;
3430 ret = btrfs_search_slot(trans, extent_root, &key, path,
3433 printk(KERN_ERR "umm, got %d back from search"
3434 ", was looking for %llu\n", ret,
3435 (unsigned long long)bytenr);
3436 btrfs_print_leaf(extent_root, path->nodes[0]);
3439 extent_slot = path->slots[0];
3440 leaf = path->nodes[0];
3441 item_size = btrfs_item_size_nr(leaf, extent_slot);
3444 BUG_ON(item_size < sizeof(*ei));
3445 ei = btrfs_item_ptr(leaf, extent_slot,
3446 struct btrfs_extent_item);
3447 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3448 struct btrfs_tree_block_info *bi;
3449 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3450 bi = (struct btrfs_tree_block_info *)(ei + 1);
3451 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3454 refs = btrfs_extent_refs(leaf, ei);
3455 BUG_ON(refs < refs_to_drop);
3456 refs -= refs_to_drop;
3460 __run_delayed_extent_op(extent_op, leaf, ei);
3462 * In the case of inline back ref, reference count will
3463 * be updated by remove_extent_backref
3466 BUG_ON(!found_extent);
3468 btrfs_set_extent_refs(leaf, ei, refs);
3469 btrfs_mark_buffer_dirty(leaf);
3472 ret = remove_extent_backref(trans, extent_root, path,
3479 struct extent_buffer *must_clean = NULL;
3482 BUG_ON(is_data && refs_to_drop !=
3483 extent_data_ref_count(root, path, iref));
3485 BUG_ON(path->slots[0] != extent_slot);
3487 BUG_ON(path->slots[0] != extent_slot + 1);
3488 path->slots[0] = extent_slot;
3493 ret = pin_down_bytes(trans, root, path, bytenr,
3494 num_bytes, is_data, 0, &must_clean);
3499 * it is going to be very rare for someone to be waiting
3500 * on the block we're freeing. del_items might need to
3501 * schedule, so rather than get fancy, just force it
3505 btrfs_set_lock_blocking(must_clean);
3507 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3510 btrfs_release_path(extent_root, path);
3513 clean_tree_block(NULL, root, must_clean);
3514 btrfs_tree_unlock(must_clean);
3515 free_extent_buffer(must_clean);
3519 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3522 invalidate_mapping_pages(info->btree_inode->i_mapping,
3523 bytenr >> PAGE_CACHE_SHIFT,
3524 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3527 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3531 btrfs_free_path(path);
3536 * when we free an extent, it is possible (and likely) that we free the last
3537 * delayed ref for that extent as well. This searches the delayed ref tree for
3538 * a given extent, and if there are no other delayed refs to be processed, it
3539 * removes it from the tree.
3541 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3542 struct btrfs_root *root, u64 bytenr)
3544 struct btrfs_delayed_ref_head *head;
3545 struct btrfs_delayed_ref_root *delayed_refs;
3546 struct btrfs_delayed_ref_node *ref;
3547 struct rb_node *node;
3550 delayed_refs = &trans->transaction->delayed_refs;
3551 spin_lock(&delayed_refs->lock);
3552 head = btrfs_find_delayed_ref_head(trans, bytenr);
3556 node = rb_prev(&head->node.rb_node);
3560 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3562 /* there are still entries for this ref, we can't drop it */
3563 if (ref->bytenr == bytenr)
3566 if (head->extent_op) {
3567 if (!head->must_insert_reserved)
3569 kfree(head->extent_op);
3570 head->extent_op = NULL;
3574 * waiting for the lock here would deadlock. If someone else has it
3575 * locked they are already in the process of dropping it anyway
3577 if (!mutex_trylock(&head->mutex))
3581 * at this point we have a head with no other entries. Go
3582 * ahead and process it.
3584 head->node.in_tree = 0;
3585 rb_erase(&head->node.rb_node, &delayed_refs->root);
3587 delayed_refs->num_entries--;
3590 * we don't take a ref on the node because we're removing it from the
3591 * tree, so we just steal the ref the tree was holding.
3593 delayed_refs->num_heads--;
3594 if (list_empty(&head->cluster))
3595 delayed_refs->num_heads_ready--;
3597 list_del_init(&head->cluster);
3598 spin_unlock(&delayed_refs->lock);
3600 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3601 &head->node, head->extent_op,
3602 head->must_insert_reserved);
3604 btrfs_put_delayed_ref(&head->node);
3607 spin_unlock(&delayed_refs->lock);
3611 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3612 struct btrfs_root *root,
3613 u64 bytenr, u64 num_bytes, u64 parent,
3614 u64 root_objectid, u64 owner, u64 offset)
3619 * tree log blocks never actually go into the extent allocation
3620 * tree, just update pinning info and exit early.
3622 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3623 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3624 /* unlocks the pinned mutex */
3625 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3627 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3628 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3629 parent, root_objectid, (int)owner,
3630 BTRFS_DROP_DELAYED_REF, NULL);
3632 ret = check_ref_cleanup(trans, root, bytenr);
3635 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3636 parent, root_objectid, owner,
3637 offset, BTRFS_DROP_DELAYED_REF, NULL);
3643 static u64 stripe_align(struct btrfs_root *root, u64 val)
3645 u64 mask = ((u64)root->stripesize - 1);
3646 u64 ret = (val + mask) & ~mask;
3651 * when we wait for progress in the block group caching, its because
3652 * our allocation attempt failed at least once. So, we must sleep
3653 * and let some progress happen before we try again.
3655 * This function will sleep at least once waiting for new free space to
3656 * show up, and then it will check the block group free space numbers
3657 * for our min num_bytes. Another option is to have it go ahead
3658 * and look in the rbtree for a free extent of a given size, but this
3662 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3665 struct btrfs_caching_control *caching_ctl;
3668 caching_ctl = get_caching_control(cache);
3672 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3673 (cache->free_space >= num_bytes));
3675 put_caching_control(caching_ctl);
3680 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3682 struct btrfs_caching_control *caching_ctl;
3685 caching_ctl = get_caching_control(cache);
3689 wait_event(caching_ctl->wait, block_group_cache_done(cache));
3691 put_caching_control(caching_ctl);
3695 enum btrfs_loop_type {
3696 LOOP_CACHED_ONLY = 0,
3697 LOOP_CACHING_NOWAIT = 1,
3698 LOOP_CACHING_WAIT = 2,
3699 LOOP_ALLOC_CHUNK = 3,
3700 LOOP_NO_EMPTY_SIZE = 4,
3704 * walks the btree of allocated extents and find a hole of a given size.
3705 * The key ins is changed to record the hole:
3706 * ins->objectid == block start
3707 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3708 * ins->offset == number of blocks
3709 * Any available blocks before search_start are skipped.
3711 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3712 struct btrfs_root *orig_root,
3713 u64 num_bytes, u64 empty_size,
3714 u64 search_start, u64 search_end,
3715 u64 hint_byte, struct btrfs_key *ins,
3716 u64 exclude_start, u64 exclude_nr,
3720 struct btrfs_root *root = orig_root->fs_info->extent_root;
3721 struct btrfs_free_cluster *last_ptr = NULL;
3722 struct btrfs_block_group_cache *block_group = NULL;
3723 int empty_cluster = 2 * 1024 * 1024;
3724 int allowed_chunk_alloc = 0;
3725 struct btrfs_space_info *space_info;
3726 int last_ptr_loop = 0;
3728 bool found_uncached_bg = false;
3729 bool failed_cluster_refill = false;
3731 WARN_ON(num_bytes < root->sectorsize);
3732 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3736 space_info = __find_space_info(root->fs_info, data);
3738 if (orig_root->ref_cows || empty_size)
3739 allowed_chunk_alloc = 1;
3741 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3742 last_ptr = &root->fs_info->meta_alloc_cluster;
3743 if (!btrfs_test_opt(root, SSD))
3744 empty_cluster = 64 * 1024;
3747 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3748 last_ptr = &root->fs_info->data_alloc_cluster;
3752 spin_lock(&last_ptr->lock);
3753 if (last_ptr->block_group)
3754 hint_byte = last_ptr->window_start;
3755 spin_unlock(&last_ptr->lock);
3758 search_start = max(search_start, first_logical_byte(root, 0));
3759 search_start = max(search_start, hint_byte);
3764 if (search_start == hint_byte) {
3765 block_group = btrfs_lookup_block_group(root->fs_info,
3768 * we don't want to use the block group if it doesn't match our
3769 * allocation bits, or if its not cached.
3771 if (block_group && block_group_bits(block_group, data) &&
3772 block_group_cache_done(block_group)) {
3773 down_read(&space_info->groups_sem);
3774 if (list_empty(&block_group->list) ||
3777 * someone is removing this block group,
3778 * we can't jump into the have_block_group
3779 * target because our list pointers are not
3782 btrfs_put_block_group(block_group);
3783 up_read(&space_info->groups_sem);
3785 goto have_block_group;
3786 } else if (block_group) {
3787 btrfs_put_block_group(block_group);
3792 down_read(&space_info->groups_sem);
3793 list_for_each_entry(block_group, &space_info->block_groups, list) {
3797 atomic_inc(&block_group->count);
3798 search_start = block_group->key.objectid;
3801 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
3803 * we want to start caching kthreads, but not too many
3804 * right off the bat so we don't overwhelm the system,
3805 * so only start them if there are less than 2 and we're
3806 * in the initial allocation phase.
3808 if (loop > LOOP_CACHING_NOWAIT ||
3809 atomic_read(&space_info->caching_threads) < 2) {
3810 ret = cache_block_group(block_group);
3815 cached = block_group_cache_done(block_group);
3816 if (unlikely(!cached)) {
3817 found_uncached_bg = true;
3819 /* if we only want cached bgs, loop */
3820 if (loop == LOOP_CACHED_ONLY)
3824 if (unlikely(block_group->ro))
3828 * Ok we want to try and use the cluster allocator, so lets look
3829 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
3830 * have tried the cluster allocator plenty of times at this
3831 * point and not have found anything, so we are likely way too
3832 * fragmented for the clustering stuff to find anything, so lets
3833 * just skip it and let the allocator find whatever block it can
3836 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
3838 * the refill lock keeps out other
3839 * people trying to start a new cluster
3841 spin_lock(&last_ptr->refill_lock);
3842 if (last_ptr->block_group &&
3843 (last_ptr->block_group->ro ||
3844 !block_group_bits(last_ptr->block_group, data))) {
3846 goto refill_cluster;
3849 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3850 num_bytes, search_start);
3852 /* we have a block, we're done */
3853 spin_unlock(&last_ptr->refill_lock);
3857 spin_lock(&last_ptr->lock);
3859 * whoops, this cluster doesn't actually point to
3860 * this block group. Get a ref on the block
3861 * group is does point to and try again
3863 if (!last_ptr_loop && last_ptr->block_group &&
3864 last_ptr->block_group != block_group) {
3866 btrfs_put_block_group(block_group);
3867 block_group = last_ptr->block_group;
3868 atomic_inc(&block_group->count);
3869 spin_unlock(&last_ptr->lock);
3870 spin_unlock(&last_ptr->refill_lock);
3873 search_start = block_group->key.objectid;
3875 * we know this block group is properly
3876 * in the list because
3877 * btrfs_remove_block_group, drops the
3878 * cluster before it removes the block
3879 * group from the list
3881 goto have_block_group;
3883 spin_unlock(&last_ptr->lock);
3886 * this cluster didn't work out, free it and
3889 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3893 /* allocate a cluster in this block group */
3894 ret = btrfs_find_space_cluster(trans, root,
3895 block_group, last_ptr,
3897 empty_cluster + empty_size);
3900 * now pull our allocation out of this
3903 offset = btrfs_alloc_from_cluster(block_group,
3904 last_ptr, num_bytes,
3907 /* we found one, proceed */
3908 spin_unlock(&last_ptr->refill_lock);
3911 } else if (!cached && loop > LOOP_CACHING_NOWAIT
3912 && !failed_cluster_refill) {
3913 spin_unlock(&last_ptr->refill_lock);
3915 failed_cluster_refill = true;
3916 wait_block_group_cache_progress(block_group,
3917 num_bytes + empty_cluster + empty_size);
3918 goto have_block_group;
3922 * at this point we either didn't find a cluster
3923 * or we weren't able to allocate a block from our
3924 * cluster. Free the cluster we've been trying
3925 * to use, and go to the next block group
3927 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3928 spin_unlock(&last_ptr->refill_lock);
3932 offset = btrfs_find_space_for_alloc(block_group, search_start,
3933 num_bytes, empty_size);
3934 if (!offset && (cached || (!cached &&
3935 loop == LOOP_CACHING_NOWAIT))) {
3937 } else if (!offset && (!cached &&
3938 loop > LOOP_CACHING_NOWAIT)) {
3939 wait_block_group_cache_progress(block_group,
3940 num_bytes + empty_size);
3941 goto have_block_group;
3944 search_start = stripe_align(root, offset);
3945 /* move on to the next group */
3946 if (search_start + num_bytes >= search_end) {
3947 btrfs_add_free_space(block_group, offset, num_bytes);
3951 /* move on to the next group */
3952 if (search_start + num_bytes >
3953 block_group->key.objectid + block_group->key.offset) {
3954 btrfs_add_free_space(block_group, offset, num_bytes);
3958 if (exclude_nr > 0 &&
3959 (search_start + num_bytes > exclude_start &&
3960 search_start < exclude_start + exclude_nr)) {
3961 search_start = exclude_start + exclude_nr;
3963 btrfs_add_free_space(block_group, offset, num_bytes);
3965 * if search_start is still in this block group
3966 * then we just re-search this block group
3968 if (search_start >= block_group->key.objectid &&
3969 search_start < (block_group->key.objectid +
3970 block_group->key.offset))
3971 goto have_block_group;
3975 ins->objectid = search_start;
3976 ins->offset = num_bytes;
3978 if (offset < search_start)
3979 btrfs_add_free_space(block_group, offset,
3980 search_start - offset);
3981 BUG_ON(offset > search_start);
3983 update_reserved_extents(block_group, num_bytes, 1);
3985 /* we are all good, lets return */
3988 failed_cluster_refill = false;
3989 btrfs_put_block_group(block_group);
3991 up_read(&space_info->groups_sem);
3993 /* LOOP_CACHED_ONLY, only search fully cached block groups
3994 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3995 * dont wait foR them to finish caching
3996 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3997 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3998 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4001 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4002 (found_uncached_bg || empty_size || empty_cluster ||
4003 allowed_chunk_alloc)) {
4004 if (found_uncached_bg) {
4005 found_uncached_bg = false;
4006 if (loop < LOOP_CACHING_WAIT) {
4012 if (loop == LOOP_ALLOC_CHUNK) {
4017 if (allowed_chunk_alloc) {
4018 ret = do_chunk_alloc(trans, root, num_bytes +
4019 2 * 1024 * 1024, data, 1);
4020 allowed_chunk_alloc = 0;
4022 space_info->force_alloc = 1;
4025 if (loop < LOOP_NO_EMPTY_SIZE) {
4030 } else if (!ins->objectid) {
4034 /* we found what we needed */
4035 if (ins->objectid) {
4036 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4037 trans->block_group = block_group->key.objectid;
4039 btrfs_put_block_group(block_group);
4046 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
4048 struct btrfs_block_group_cache *cache;
4050 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4051 (unsigned long long)(info->total_bytes - info->bytes_used -
4052 info->bytes_pinned - info->bytes_reserved),
4053 (info->full) ? "" : "not ");
4054 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4055 " may_use=%llu, used=%llu\n",
4056 (unsigned long long)info->total_bytes,
4057 (unsigned long long)info->bytes_pinned,
4058 (unsigned long long)info->bytes_delalloc,
4059 (unsigned long long)info->bytes_may_use,
4060 (unsigned long long)info->bytes_used);
4062 down_read(&info->groups_sem);
4063 list_for_each_entry(cache, &info->block_groups, list) {
4064 spin_lock(&cache->lock);
4065 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4066 "%llu pinned %llu reserved\n",
4067 (unsigned long long)cache->key.objectid,
4068 (unsigned long long)cache->key.offset,
4069 (unsigned long long)btrfs_block_group_used(&cache->item),
4070 (unsigned long long)cache->pinned,
4071 (unsigned long long)cache->reserved);
4072 btrfs_dump_free_space(cache, bytes);
4073 spin_unlock(&cache->lock);
4075 up_read(&info->groups_sem);
4078 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4079 struct btrfs_root *root,
4080 u64 num_bytes, u64 min_alloc_size,
4081 u64 empty_size, u64 hint_byte,
4082 u64 search_end, struct btrfs_key *ins,
4086 u64 search_start = 0;
4087 struct btrfs_fs_info *info = root->fs_info;
4089 data = btrfs_get_alloc_profile(root, data);
4092 * the only place that sets empty_size is btrfs_realloc_node, which
4093 * is not called recursively on allocations
4095 if (empty_size || root->ref_cows) {
4096 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4097 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4099 BTRFS_BLOCK_GROUP_METADATA |
4100 (info->metadata_alloc_profile &
4101 info->avail_metadata_alloc_bits), 0);
4103 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4104 num_bytes + 2 * 1024 * 1024, data, 0);
4107 WARN_ON(num_bytes < root->sectorsize);
4108 ret = find_free_extent(trans, root, num_bytes, empty_size,
4109 search_start, search_end, hint_byte, ins,
4110 trans->alloc_exclude_start,
4111 trans->alloc_exclude_nr, data);
4113 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4114 num_bytes = num_bytes >> 1;
4115 num_bytes = num_bytes & ~(root->sectorsize - 1);
4116 num_bytes = max(num_bytes, min_alloc_size);
4117 do_chunk_alloc(trans, root->fs_info->extent_root,
4118 num_bytes, data, 1);
4121 if (ret == -ENOSPC) {
4122 struct btrfs_space_info *sinfo;
4124 sinfo = __find_space_info(root->fs_info, data);
4125 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4126 "wanted %llu\n", (unsigned long long)data,
4127 (unsigned long long)num_bytes);
4128 dump_space_info(sinfo, num_bytes);
4134 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4136 struct btrfs_block_group_cache *cache;
4139 cache = btrfs_lookup_block_group(root->fs_info, start);
4141 printk(KERN_ERR "Unable to find block group for %llu\n",
4142 (unsigned long long)start);
4146 ret = btrfs_discard_extent(root, start, len);
4148 btrfs_add_free_space(cache, start, len);
4149 update_reserved_extents(cache, len, 0);
4150 btrfs_put_block_group(cache);
4155 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4156 struct btrfs_root *root,
4157 u64 parent, u64 root_objectid,
4158 u64 flags, u64 owner, u64 offset,
4159 struct btrfs_key *ins, int ref_mod)
4162 struct btrfs_fs_info *fs_info = root->fs_info;
4163 struct btrfs_extent_item *extent_item;
4164 struct btrfs_extent_inline_ref *iref;
4165 struct btrfs_path *path;
4166 struct extent_buffer *leaf;
4171 type = BTRFS_SHARED_DATA_REF_KEY;
4173 type = BTRFS_EXTENT_DATA_REF_KEY;
4175 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4177 path = btrfs_alloc_path();
4180 path->leave_spinning = 1;
4181 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4185 leaf = path->nodes[0];
4186 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4187 struct btrfs_extent_item);
4188 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4189 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4190 btrfs_set_extent_flags(leaf, extent_item,
4191 flags | BTRFS_EXTENT_FLAG_DATA);
4193 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4194 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4196 struct btrfs_shared_data_ref *ref;
4197 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4198 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4199 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4201 struct btrfs_extent_data_ref *ref;
4202 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4203 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4204 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4205 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4206 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4209 btrfs_mark_buffer_dirty(path->nodes[0]);
4210 btrfs_free_path(path);
4212 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4215 printk(KERN_ERR "btrfs update block group failed for %llu "
4216 "%llu\n", (unsigned long long)ins->objectid,
4217 (unsigned long long)ins->offset);
4223 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4224 struct btrfs_root *root,
4225 u64 parent, u64 root_objectid,
4226 u64 flags, struct btrfs_disk_key *key,
4227 int level, struct btrfs_key *ins)
4230 struct btrfs_fs_info *fs_info = root->fs_info;
4231 struct btrfs_extent_item *extent_item;
4232 struct btrfs_tree_block_info *block_info;
4233 struct btrfs_extent_inline_ref *iref;
4234 struct btrfs_path *path;
4235 struct extent_buffer *leaf;
4236 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4238 path = btrfs_alloc_path();
4241 path->leave_spinning = 1;
4242 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4246 leaf = path->nodes[0];
4247 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4248 struct btrfs_extent_item);
4249 btrfs_set_extent_refs(leaf, extent_item, 1);
4250 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4251 btrfs_set_extent_flags(leaf, extent_item,
4252 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4253 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4255 btrfs_set_tree_block_key(leaf, block_info, key);
4256 btrfs_set_tree_block_level(leaf, block_info, level);
4258 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4260 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4261 btrfs_set_extent_inline_ref_type(leaf, iref,
4262 BTRFS_SHARED_BLOCK_REF_KEY);
4263 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4265 btrfs_set_extent_inline_ref_type(leaf, iref,
4266 BTRFS_TREE_BLOCK_REF_KEY);
4267 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4270 btrfs_mark_buffer_dirty(leaf);
4271 btrfs_free_path(path);
4273 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4276 printk(KERN_ERR "btrfs update block group failed for %llu "
4277 "%llu\n", (unsigned long long)ins->objectid,
4278 (unsigned long long)ins->offset);
4284 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4285 struct btrfs_root *root,
4286 u64 root_objectid, u64 owner,
4287 u64 offset, struct btrfs_key *ins)
4291 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4293 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4294 0, root_objectid, owner, offset,
4295 BTRFS_ADD_DELAYED_EXTENT, NULL);
4300 * this is used by the tree logging recovery code. It records that
4301 * an extent has been allocated and makes sure to clear the free
4302 * space cache bits as well
4304 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4305 struct btrfs_root *root,
4306 u64 root_objectid, u64 owner, u64 offset,
4307 struct btrfs_key *ins)
4310 struct btrfs_block_group_cache *block_group;
4311 struct btrfs_caching_control *caching_ctl;
4312 u64 start = ins->objectid;
4313 u64 num_bytes = ins->offset;
4315 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4316 cache_block_group(block_group);
4317 caching_ctl = get_caching_control(block_group);
4320 BUG_ON(!block_group_cache_done(block_group));
4321 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4324 mutex_lock(&caching_ctl->mutex);
4326 if (start >= caching_ctl->progress) {
4327 ret = add_excluded_extent(root, start, num_bytes);
4329 } else if (start + num_bytes <= caching_ctl->progress) {
4330 ret = btrfs_remove_free_space(block_group,
4334 num_bytes = caching_ctl->progress - start;
4335 ret = btrfs_remove_free_space(block_group,
4339 start = caching_ctl->progress;
4340 num_bytes = ins->objectid + ins->offset -
4341 caching_ctl->progress;
4342 ret = add_excluded_extent(root, start, num_bytes);
4346 mutex_unlock(&caching_ctl->mutex);
4347 put_caching_control(caching_ctl);
4350 update_reserved_extents(block_group, ins->offset, 1);
4351 btrfs_put_block_group(block_group);
4352 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4353 0, owner, offset, ins, 1);
4358 * finds a free extent and does all the dirty work required for allocation
4359 * returns the key for the extent through ins, and a tree buffer for
4360 * the first block of the extent through buf.
4362 * returns 0 if everything worked, non-zero otherwise.
4364 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4365 struct btrfs_root *root,
4366 u64 num_bytes, u64 parent, u64 root_objectid,
4367 struct btrfs_disk_key *key, int level,
4368 u64 empty_size, u64 hint_byte, u64 search_end,
4369 struct btrfs_key *ins)
4374 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4375 empty_size, hint_byte, search_end,
4380 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4382 parent = ins->objectid;
4383 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4387 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4388 struct btrfs_delayed_extent_op *extent_op;
4389 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4392 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4394 memset(&extent_op->key, 0, sizeof(extent_op->key));
4395 extent_op->flags_to_set = flags;
4396 extent_op->update_key = 1;
4397 extent_op->update_flags = 1;
4398 extent_op->is_data = 0;
4400 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4401 ins->offset, parent, root_objectid,
4402 level, BTRFS_ADD_DELAYED_EXTENT,
4409 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4410 struct btrfs_root *root,
4411 u64 bytenr, u32 blocksize,
4414 struct extent_buffer *buf;
4416 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4418 return ERR_PTR(-ENOMEM);
4419 btrfs_set_header_generation(buf, trans->transid);
4420 btrfs_set_buffer_lockdep_class(buf, level);
4421 btrfs_tree_lock(buf);
4422 clean_tree_block(trans, root, buf);
4424 btrfs_set_lock_blocking(buf);
4425 btrfs_set_buffer_uptodate(buf);
4427 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4428 set_extent_dirty(&root->dirty_log_pages, buf->start,
4429 buf->start + buf->len - 1, GFP_NOFS);
4431 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4432 buf->start + buf->len - 1, GFP_NOFS);
4434 trans->blocks_used++;
4435 /* this returns a buffer locked for blocking */
4440 * helper function to allocate a block for a given tree
4441 * returns the tree buffer or NULL.
4443 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4444 struct btrfs_root *root, u32 blocksize,
4445 u64 parent, u64 root_objectid,
4446 struct btrfs_disk_key *key, int level,
4447 u64 hint, u64 empty_size)
4449 struct btrfs_key ins;
4451 struct extent_buffer *buf;
4453 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4454 key, level, empty_size, hint, (u64)-1, &ins);
4457 return ERR_PTR(ret);
4460 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4465 struct walk_control {
4466 u64 refs[BTRFS_MAX_LEVEL];
4467 u64 flags[BTRFS_MAX_LEVEL];
4468 struct btrfs_key update_progress;
4478 #define DROP_REFERENCE 1
4479 #define UPDATE_BACKREF 2
4481 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4482 struct btrfs_root *root,
4483 struct walk_control *wc,
4484 struct btrfs_path *path)
4492 struct btrfs_key key;
4493 struct extent_buffer *eb;
4498 if (path->slots[wc->level] < wc->reada_slot) {
4499 wc->reada_count = wc->reada_count * 2 / 3;
4500 wc->reada_count = max(wc->reada_count, 2);
4502 wc->reada_count = wc->reada_count * 3 / 2;
4503 wc->reada_count = min_t(int, wc->reada_count,
4504 BTRFS_NODEPTRS_PER_BLOCK(root));
4507 eb = path->nodes[wc->level];
4508 nritems = btrfs_header_nritems(eb);
4509 blocksize = btrfs_level_size(root, wc->level - 1);
4511 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4512 if (nread >= wc->reada_count)
4516 bytenr = btrfs_node_blockptr(eb, slot);
4517 generation = btrfs_node_ptr_generation(eb, slot);
4519 if (slot == path->slots[wc->level])
4522 if (wc->stage == UPDATE_BACKREF &&
4523 generation <= root->root_key.offset)
4526 if (wc->stage == DROP_REFERENCE) {
4527 ret = btrfs_lookup_extent_info(trans, root,
4535 if (!wc->update_ref ||
4536 generation <= root->root_key.offset)
4538 btrfs_node_key_to_cpu(eb, &key, slot);
4539 ret = btrfs_comp_cpu_keys(&key,
4540 &wc->update_progress);
4545 ret = readahead_tree_block(root, bytenr, blocksize,
4549 last = bytenr + blocksize;
4552 wc->reada_slot = slot;
4556 * hepler to process tree block while walking down the tree.
4558 * when wc->stage == UPDATE_BACKREF, this function updates
4559 * back refs for pointers in the block.
4561 * NOTE: return value 1 means we should stop walking down.
4563 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4564 struct btrfs_root *root,
4565 struct btrfs_path *path,
4566 struct walk_control *wc)
4568 int level = wc->level;
4569 struct extent_buffer *eb = path->nodes[level];
4570 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4573 if (wc->stage == UPDATE_BACKREF &&
4574 btrfs_header_owner(eb) != root->root_key.objectid)
4578 * when reference count of tree block is 1, it won't increase
4579 * again. once full backref flag is set, we never clear it.
4581 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4582 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4583 BUG_ON(!path->locks[level]);
4584 ret = btrfs_lookup_extent_info(trans, root,
4589 BUG_ON(wc->refs[level] == 0);
4592 if (wc->stage == DROP_REFERENCE) {
4593 if (wc->refs[level] > 1)
4596 if (path->locks[level] && !wc->keep_locks) {
4597 btrfs_tree_unlock(eb);
4598 path->locks[level] = 0;
4603 /* wc->stage == UPDATE_BACKREF */
4604 if (!(wc->flags[level] & flag)) {
4605 BUG_ON(!path->locks[level]);
4606 ret = btrfs_inc_ref(trans, root, eb, 1);
4608 ret = btrfs_dec_ref(trans, root, eb, 0);
4610 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4613 wc->flags[level] |= flag;
4617 * the block is shared by multiple trees, so it's not good to
4618 * keep the tree lock
4620 if (path->locks[level] && level > 0) {
4621 btrfs_tree_unlock(eb);
4622 path->locks[level] = 0;
4628 * hepler to process tree block pointer.
4630 * when wc->stage == DROP_REFERENCE, this function checks
4631 * reference count of the block pointed to. if the block
4632 * is shared and we need update back refs for the subtree
4633 * rooted at the block, this function changes wc->stage to
4634 * UPDATE_BACKREF. if the block is shared and there is no
4635 * need to update back, this function drops the reference
4638 * NOTE: return value 1 means we should stop walking down.
4640 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4641 struct btrfs_root *root,
4642 struct btrfs_path *path,
4643 struct walk_control *wc)
4649 struct btrfs_key key;
4650 struct extent_buffer *next;
4651 int level = wc->level;
4655 generation = btrfs_node_ptr_generation(path->nodes[level],
4656 path->slots[level]);
4658 * if the lower level block was created before the snapshot
4659 * was created, we know there is no need to update back refs
4662 if (wc->stage == UPDATE_BACKREF &&
4663 generation <= root->root_key.offset)
4666 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4667 blocksize = btrfs_level_size(root, level - 1);
4669 next = btrfs_find_tree_block(root, bytenr, blocksize);
4671 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
4674 btrfs_tree_lock(next);
4675 btrfs_set_lock_blocking(next);
4677 if (wc->stage == DROP_REFERENCE) {
4678 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4679 &wc->refs[level - 1],
4680 &wc->flags[level - 1]);
4682 BUG_ON(wc->refs[level - 1] == 0);
4684 if (wc->refs[level - 1] > 1) {
4685 if (!wc->update_ref ||
4686 generation <= root->root_key.offset)
4689 btrfs_node_key_to_cpu(path->nodes[level], &key,
4690 path->slots[level]);
4691 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4695 wc->stage = UPDATE_BACKREF;
4696 wc->shared_level = level - 1;
4700 if (!btrfs_buffer_uptodate(next, generation)) {
4701 btrfs_tree_unlock(next);
4702 free_extent_buffer(next);
4707 if (reada && level == 1)
4708 reada_walk_down(trans, root, wc, path);
4709 next = read_tree_block(root, bytenr, blocksize, generation);
4710 btrfs_tree_lock(next);
4711 btrfs_set_lock_blocking(next);
4715 BUG_ON(level != btrfs_header_level(next));
4716 path->nodes[level] = next;
4717 path->slots[level] = 0;
4718 path->locks[level] = 1;
4724 wc->refs[level - 1] = 0;
4725 wc->flags[level - 1] = 0;
4727 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4728 parent = path->nodes[level]->start;
4730 BUG_ON(root->root_key.objectid !=
4731 btrfs_header_owner(path->nodes[level]));
4735 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
4736 root->root_key.objectid, level - 1, 0);
4739 btrfs_tree_unlock(next);
4740 free_extent_buffer(next);
4745 * hepler to process tree block while walking up the tree.
4747 * when wc->stage == DROP_REFERENCE, this function drops
4748 * reference count on the block.
4750 * when wc->stage == UPDATE_BACKREF, this function changes
4751 * wc->stage back to DROP_REFERENCE if we changed wc->stage
4752 * to UPDATE_BACKREF previously while processing the block.
4754 * NOTE: return value 1 means we should stop walking up.
4756 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4757 struct btrfs_root *root,
4758 struct btrfs_path *path,
4759 struct walk_control *wc)
4762 int level = wc->level;
4763 struct extent_buffer *eb = path->nodes[level];
4766 if (wc->stage == UPDATE_BACKREF) {
4767 BUG_ON(wc->shared_level < level);
4768 if (level < wc->shared_level)
4771 ret = find_next_key(path, level + 1, &wc->update_progress);
4775 wc->stage = DROP_REFERENCE;
4776 wc->shared_level = -1;
4777 path->slots[level] = 0;
4780 * check reference count again if the block isn't locked.
4781 * we should start walking down the tree again if reference
4784 if (!path->locks[level]) {
4786 btrfs_tree_lock(eb);
4787 btrfs_set_lock_blocking(eb);
4788 path->locks[level] = 1;
4790 ret = btrfs_lookup_extent_info(trans, root,
4795 BUG_ON(wc->refs[level] == 0);
4796 if (wc->refs[level] == 1) {
4797 btrfs_tree_unlock(eb);
4798 path->locks[level] = 0;
4804 /* wc->stage == DROP_REFERENCE */
4805 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4807 if (wc->refs[level] == 1) {
4809 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4810 ret = btrfs_dec_ref(trans, root, eb, 1);
4812 ret = btrfs_dec_ref(trans, root, eb, 0);
4815 /* make block locked assertion in clean_tree_block happy */
4816 if (!path->locks[level] &&
4817 btrfs_header_generation(eb) == trans->transid) {
4818 btrfs_tree_lock(eb);
4819 btrfs_set_lock_blocking(eb);
4820 path->locks[level] = 1;
4822 clean_tree_block(trans, root, eb);
4825 if (eb == root->node) {
4826 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4829 BUG_ON(root->root_key.objectid !=
4830 btrfs_header_owner(eb));
4832 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4833 parent = path->nodes[level + 1]->start;
4835 BUG_ON(root->root_key.objectid !=
4836 btrfs_header_owner(path->nodes[level + 1]));
4839 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
4840 root->root_key.objectid, level, 0);
4843 wc->refs[level] = 0;
4844 wc->flags[level] = 0;
4848 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4849 struct btrfs_root *root,
4850 struct btrfs_path *path,
4851 struct walk_control *wc)
4853 int level = wc->level;
4856 while (level >= 0) {
4857 if (path->slots[level] >=
4858 btrfs_header_nritems(path->nodes[level]))
4861 ret = walk_down_proc(trans, root, path, wc);
4868 ret = do_walk_down(trans, root, path, wc);
4870 path->slots[level]++;
4878 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4879 struct btrfs_root *root,
4880 struct btrfs_path *path,
4881 struct walk_control *wc, int max_level)
4883 int level = wc->level;
4886 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
4887 while (level < max_level && path->nodes[level]) {
4889 if (path->slots[level] + 1 <
4890 btrfs_header_nritems(path->nodes[level])) {
4891 path->slots[level]++;
4894 ret = walk_up_proc(trans, root, path, wc);
4898 if (path->locks[level]) {
4899 btrfs_tree_unlock(path->nodes[level]);
4900 path->locks[level] = 0;
4902 free_extent_buffer(path->nodes[level]);
4903 path->nodes[level] = NULL;
4911 * drop a subvolume tree.
4913 * this function traverses the tree freeing any blocks that only
4914 * referenced by the tree.
4916 * when a shared tree block is found. this function decreases its
4917 * reference count by one. if update_ref is true, this function
4918 * also make sure backrefs for the shared block and all lower level
4919 * blocks are properly updated.
4921 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
4923 struct btrfs_path *path;
4924 struct btrfs_trans_handle *trans;
4925 struct btrfs_root *tree_root = root->fs_info->tree_root;
4926 struct btrfs_root_item *root_item = &root->root_item;
4927 struct walk_control *wc;
4928 struct btrfs_key key;
4933 path = btrfs_alloc_path();
4936 wc = kzalloc(sizeof(*wc), GFP_NOFS);
4939 trans = btrfs_start_transaction(tree_root, 1);
4941 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4942 level = btrfs_header_level(root->node);
4943 path->nodes[level] = btrfs_lock_root_node(root);
4944 btrfs_set_lock_blocking(path->nodes[level]);
4945 path->slots[level] = 0;
4946 path->locks[level] = 1;
4947 memset(&wc->update_progress, 0,
4948 sizeof(wc->update_progress));
4950 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4951 memcpy(&wc->update_progress, &key,
4952 sizeof(wc->update_progress));
4954 level = root_item->drop_level;
4956 path->lowest_level = level;
4957 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4958 path->lowest_level = 0;
4966 * unlock our path, this is safe because only this
4967 * function is allowed to delete this snapshot
4969 btrfs_unlock_up_safe(path, 0);
4971 level = btrfs_header_level(root->node);
4973 btrfs_tree_lock(path->nodes[level]);
4974 btrfs_set_lock_blocking(path->nodes[level]);
4976 ret = btrfs_lookup_extent_info(trans, root,
4977 path->nodes[level]->start,
4978 path->nodes[level]->len,
4982 BUG_ON(wc->refs[level] == 0);
4984 if (level == root_item->drop_level)
4987 btrfs_tree_unlock(path->nodes[level]);
4988 WARN_ON(wc->refs[level] != 1);
4994 wc->shared_level = -1;
4995 wc->stage = DROP_REFERENCE;
4996 wc->update_ref = update_ref;
4998 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5001 ret = walk_down_tree(trans, root, path, wc);
5007 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5014 BUG_ON(wc->stage != DROP_REFERENCE);
5018 if (wc->stage == DROP_REFERENCE) {
5020 btrfs_node_key(path->nodes[level],
5021 &root_item->drop_progress,
5022 path->slots[level]);
5023 root_item->drop_level = level;
5026 BUG_ON(wc->level == 0);
5027 if (trans->transaction->in_commit ||
5028 trans->transaction->delayed_refs.flushing) {
5029 ret = btrfs_update_root(trans, tree_root,
5034 btrfs_end_transaction(trans, tree_root);
5035 trans = btrfs_start_transaction(tree_root, 1);
5037 unsigned long update;
5038 update = trans->delayed_ref_updates;
5039 trans->delayed_ref_updates = 0;
5041 btrfs_run_delayed_refs(trans, tree_root,
5045 btrfs_release_path(root, path);
5048 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5051 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5052 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5056 ret = btrfs_del_orphan_item(trans, tree_root,
5057 root->root_key.objectid);
5062 if (root->in_radix) {
5063 btrfs_free_fs_root(tree_root->fs_info, root);
5065 free_extent_buffer(root->node);
5066 free_extent_buffer(root->commit_root);
5070 btrfs_end_transaction(trans, tree_root);
5072 btrfs_free_path(path);
5077 * drop subtree rooted at tree block 'node'.
5079 * NOTE: this function will unlock and release tree block 'node'
5081 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5082 struct btrfs_root *root,
5083 struct extent_buffer *node,
5084 struct extent_buffer *parent)
5086 struct btrfs_path *path;
5087 struct walk_control *wc;
5093 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5095 path = btrfs_alloc_path();
5098 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5101 btrfs_assert_tree_locked(parent);
5102 parent_level = btrfs_header_level(parent);
5103 extent_buffer_get(parent);
5104 path->nodes[parent_level] = parent;
5105 path->slots[parent_level] = btrfs_header_nritems(parent);
5107 btrfs_assert_tree_locked(node);
5108 level = btrfs_header_level(node);
5109 path->nodes[level] = node;
5110 path->slots[level] = 0;
5111 path->locks[level] = 1;
5113 wc->refs[parent_level] = 1;
5114 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5116 wc->shared_level = -1;
5117 wc->stage = DROP_REFERENCE;
5120 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5123 wret = walk_down_tree(trans, root, path, wc);
5129 wret = walk_up_tree(trans, root, path, wc, parent_level);
5137 btrfs_free_path(path);
5142 static unsigned long calc_ra(unsigned long start, unsigned long last,
5145 return min(last, start + nr - 1);
5148 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5153 unsigned long first_index;
5154 unsigned long last_index;
5157 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5158 struct file_ra_state *ra;
5159 struct btrfs_ordered_extent *ordered;
5160 unsigned int total_read = 0;
5161 unsigned int total_dirty = 0;
5164 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5166 mutex_lock(&inode->i_mutex);
5167 first_index = start >> PAGE_CACHE_SHIFT;
5168 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5170 /* make sure the dirty trick played by the caller work */
5171 ret = invalidate_inode_pages2_range(inode->i_mapping,
5172 first_index, last_index);
5176 file_ra_state_init(ra, inode->i_mapping);
5178 for (i = first_index ; i <= last_index; i++) {
5179 if (total_read % ra->ra_pages == 0) {
5180 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5181 calc_ra(i, last_index, ra->ra_pages));
5185 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5187 page = grab_cache_page(inode->i_mapping, i);
5192 if (!PageUptodate(page)) {
5193 btrfs_readpage(NULL, page);
5195 if (!PageUptodate(page)) {
5197 page_cache_release(page);
5202 wait_on_page_writeback(page);
5204 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5205 page_end = page_start + PAGE_CACHE_SIZE - 1;
5206 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5208 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5210 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5212 page_cache_release(page);
5213 btrfs_start_ordered_extent(inode, ordered, 1);
5214 btrfs_put_ordered_extent(ordered);
5217 set_page_extent_mapped(page);
5219 if (i == first_index)
5220 set_extent_bits(io_tree, page_start, page_end,
5221 EXTENT_BOUNDARY, GFP_NOFS);
5222 btrfs_set_extent_delalloc(inode, page_start, page_end);
5224 set_page_dirty(page);
5227 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5229 page_cache_release(page);
5234 mutex_unlock(&inode->i_mutex);
5235 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5239 static noinline int relocate_data_extent(struct inode *reloc_inode,
5240 struct btrfs_key *extent_key,
5243 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5244 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5245 struct extent_map *em;
5246 u64 start = extent_key->objectid - offset;
5247 u64 end = start + extent_key->offset - 1;
5249 em = alloc_extent_map(GFP_NOFS);
5250 BUG_ON(!em || IS_ERR(em));
5253 em->len = extent_key->offset;
5254 em->block_len = extent_key->offset;
5255 em->block_start = extent_key->objectid;
5256 em->bdev = root->fs_info->fs_devices->latest_bdev;
5257 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5259 /* setup extent map to cheat btrfs_readpage */
5260 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5263 write_lock(&em_tree->lock);
5264 ret = add_extent_mapping(em_tree, em);
5265 write_unlock(&em_tree->lock);
5266 if (ret != -EEXIST) {
5267 free_extent_map(em);
5270 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5272 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5274 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5277 struct btrfs_ref_path {
5279 u64 nodes[BTRFS_MAX_LEVEL];
5281 u64 root_generation;
5288 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5289 u64 new_nodes[BTRFS_MAX_LEVEL];
5292 struct disk_extent {
5303 static int is_cowonly_root(u64 root_objectid)
5305 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5306 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5307 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5308 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5309 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5310 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5315 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5316 struct btrfs_root *extent_root,
5317 struct btrfs_ref_path *ref_path,
5320 struct extent_buffer *leaf;
5321 struct btrfs_path *path;
5322 struct btrfs_extent_ref *ref;
5323 struct btrfs_key key;
5324 struct btrfs_key found_key;
5330 path = btrfs_alloc_path();
5335 ref_path->lowest_level = -1;
5336 ref_path->current_level = -1;
5337 ref_path->shared_level = -1;
5341 level = ref_path->current_level - 1;
5342 while (level >= -1) {
5344 if (level < ref_path->lowest_level)
5348 bytenr = ref_path->nodes[level];
5350 bytenr = ref_path->extent_start;
5351 BUG_ON(bytenr == 0);
5353 parent = ref_path->nodes[level + 1];
5354 ref_path->nodes[level + 1] = 0;
5355 ref_path->current_level = level;
5356 BUG_ON(parent == 0);
5358 key.objectid = bytenr;
5359 key.offset = parent + 1;
5360 key.type = BTRFS_EXTENT_REF_KEY;
5362 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5367 leaf = path->nodes[0];
5368 nritems = btrfs_header_nritems(leaf);
5369 if (path->slots[0] >= nritems) {
5370 ret = btrfs_next_leaf(extent_root, path);
5375 leaf = path->nodes[0];
5378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5379 if (found_key.objectid == bytenr &&
5380 found_key.type == BTRFS_EXTENT_REF_KEY) {
5381 if (level < ref_path->shared_level)
5382 ref_path->shared_level = level;
5387 btrfs_release_path(extent_root, path);
5390 /* reached lowest level */
5394 level = ref_path->current_level;
5395 while (level < BTRFS_MAX_LEVEL - 1) {
5399 bytenr = ref_path->nodes[level];
5401 bytenr = ref_path->extent_start;
5403 BUG_ON(bytenr == 0);
5405 key.objectid = bytenr;
5407 key.type = BTRFS_EXTENT_REF_KEY;
5409 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5413 leaf = path->nodes[0];
5414 nritems = btrfs_header_nritems(leaf);
5415 if (path->slots[0] >= nritems) {
5416 ret = btrfs_next_leaf(extent_root, path);
5420 /* the extent was freed by someone */
5421 if (ref_path->lowest_level == level)
5423 btrfs_release_path(extent_root, path);
5426 leaf = path->nodes[0];
5429 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5430 if (found_key.objectid != bytenr ||
5431 found_key.type != BTRFS_EXTENT_REF_KEY) {
5432 /* the extent was freed by someone */
5433 if (ref_path->lowest_level == level) {
5437 btrfs_release_path(extent_root, path);
5441 ref = btrfs_item_ptr(leaf, path->slots[0],
5442 struct btrfs_extent_ref);
5443 ref_objectid = btrfs_ref_objectid(leaf, ref);
5444 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5446 level = (int)ref_objectid;
5447 BUG_ON(level >= BTRFS_MAX_LEVEL);
5448 ref_path->lowest_level = level;
5449 ref_path->current_level = level;
5450 ref_path->nodes[level] = bytenr;
5452 WARN_ON(ref_objectid != level);
5455 WARN_ON(level != -1);
5459 if (ref_path->lowest_level == level) {
5460 ref_path->owner_objectid = ref_objectid;
5461 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5465 * the block is tree root or the block isn't in reference
5468 if (found_key.objectid == found_key.offset ||
5469 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5470 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5471 ref_path->root_generation =
5472 btrfs_ref_generation(leaf, ref);
5474 /* special reference from the tree log */
5475 ref_path->nodes[0] = found_key.offset;
5476 ref_path->current_level = 0;
5483 BUG_ON(ref_path->nodes[level] != 0);
5484 ref_path->nodes[level] = found_key.offset;
5485 ref_path->current_level = level;
5488 * the reference was created in the running transaction,
5489 * no need to continue walking up.
5491 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5492 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5493 ref_path->root_generation =
5494 btrfs_ref_generation(leaf, ref);
5499 btrfs_release_path(extent_root, path);
5502 /* reached max tree level, but no tree root found. */
5505 btrfs_free_path(path);
5509 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5510 struct btrfs_root *extent_root,
5511 struct btrfs_ref_path *ref_path,
5514 memset(ref_path, 0, sizeof(*ref_path));
5515 ref_path->extent_start = extent_start;
5517 return __next_ref_path(trans, extent_root, ref_path, 1);
5520 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5521 struct btrfs_root *extent_root,
5522 struct btrfs_ref_path *ref_path)
5524 return __next_ref_path(trans, extent_root, ref_path, 0);
5527 static noinline int get_new_locations(struct inode *reloc_inode,
5528 struct btrfs_key *extent_key,
5529 u64 offset, int no_fragment,
5530 struct disk_extent **extents,
5533 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5534 struct btrfs_path *path;
5535 struct btrfs_file_extent_item *fi;
5536 struct extent_buffer *leaf;
5537 struct disk_extent *exts = *extents;
5538 struct btrfs_key found_key;
5543 int max = *nr_extents;
5546 WARN_ON(!no_fragment && *extents);
5549 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5554 path = btrfs_alloc_path();
5557 cur_pos = extent_key->objectid - offset;
5558 last_byte = extent_key->objectid + extent_key->offset;
5559 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5569 leaf = path->nodes[0];
5570 nritems = btrfs_header_nritems(leaf);
5571 if (path->slots[0] >= nritems) {
5572 ret = btrfs_next_leaf(root, path);
5577 leaf = path->nodes[0];
5580 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5581 if (found_key.offset != cur_pos ||
5582 found_key.type != BTRFS_EXTENT_DATA_KEY ||
5583 found_key.objectid != reloc_inode->i_ino)
5586 fi = btrfs_item_ptr(leaf, path->slots[0],
5587 struct btrfs_file_extent_item);
5588 if (btrfs_file_extent_type(leaf, fi) !=
5589 BTRFS_FILE_EXTENT_REG ||
5590 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5594 struct disk_extent *old = exts;
5596 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5597 memcpy(exts, old, sizeof(*exts) * nr);
5598 if (old != *extents)
5602 exts[nr].disk_bytenr =
5603 btrfs_file_extent_disk_bytenr(leaf, fi);
5604 exts[nr].disk_num_bytes =
5605 btrfs_file_extent_disk_num_bytes(leaf, fi);
5606 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5607 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5608 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5609 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5610 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5611 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5613 BUG_ON(exts[nr].offset > 0);
5614 BUG_ON(exts[nr].compression || exts[nr].encryption);
5615 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5617 cur_pos += exts[nr].num_bytes;
5620 if (cur_pos + offset >= last_byte)
5630 BUG_ON(cur_pos + offset > last_byte);
5631 if (cur_pos + offset < last_byte) {
5637 btrfs_free_path(path);
5639 if (exts != *extents)
5648 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5649 struct btrfs_root *root,
5650 struct btrfs_path *path,
5651 struct btrfs_key *extent_key,
5652 struct btrfs_key *leaf_key,
5653 struct btrfs_ref_path *ref_path,
5654 struct disk_extent *new_extents,
5657 struct extent_buffer *leaf;
5658 struct btrfs_file_extent_item *fi;
5659 struct inode *inode = NULL;
5660 struct btrfs_key key;
5665 u64 search_end = (u64)-1;
5668 int extent_locked = 0;
5672 memcpy(&key, leaf_key, sizeof(key));
5673 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5674 if (key.objectid < ref_path->owner_objectid ||
5675 (key.objectid == ref_path->owner_objectid &&
5676 key.type < BTRFS_EXTENT_DATA_KEY)) {
5677 key.objectid = ref_path->owner_objectid;
5678 key.type = BTRFS_EXTENT_DATA_KEY;
5684 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5688 leaf = path->nodes[0];
5689 nritems = btrfs_header_nritems(leaf);
5691 if (extent_locked && ret > 0) {
5693 * the file extent item was modified by someone
5694 * before the extent got locked.
5696 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5697 lock_end, GFP_NOFS);
5701 if (path->slots[0] >= nritems) {
5702 if (++nr_scaned > 2)
5705 BUG_ON(extent_locked);
5706 ret = btrfs_next_leaf(root, path);
5711 leaf = path->nodes[0];
5712 nritems = btrfs_header_nritems(leaf);
5715 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5717 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5718 if ((key.objectid > ref_path->owner_objectid) ||
5719 (key.objectid == ref_path->owner_objectid &&
5720 key.type > BTRFS_EXTENT_DATA_KEY) ||
5721 key.offset >= search_end)
5725 if (inode && key.objectid != inode->i_ino) {
5726 BUG_ON(extent_locked);
5727 btrfs_release_path(root, path);
5728 mutex_unlock(&inode->i_mutex);
5734 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5739 fi = btrfs_item_ptr(leaf, path->slots[0],
5740 struct btrfs_file_extent_item);
5741 extent_type = btrfs_file_extent_type(leaf, fi);
5742 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5743 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5744 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5745 extent_key->objectid)) {
5751 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5752 ext_offset = btrfs_file_extent_offset(leaf, fi);
5754 if (search_end == (u64)-1) {
5755 search_end = key.offset - ext_offset +
5756 btrfs_file_extent_ram_bytes(leaf, fi);
5759 if (!extent_locked) {
5760 lock_start = key.offset;
5761 lock_end = lock_start + num_bytes - 1;
5763 if (lock_start > key.offset ||
5764 lock_end + 1 < key.offset + num_bytes) {
5765 unlock_extent(&BTRFS_I(inode)->io_tree,
5766 lock_start, lock_end, GFP_NOFS);
5772 btrfs_release_path(root, path);
5774 inode = btrfs_iget_locked(root->fs_info->sb,
5775 key.objectid, root);
5776 if (inode->i_state & I_NEW) {
5777 BTRFS_I(inode)->root = root;
5778 BTRFS_I(inode)->location.objectid =
5780 BTRFS_I(inode)->location.type =
5781 BTRFS_INODE_ITEM_KEY;
5782 BTRFS_I(inode)->location.offset = 0;
5783 btrfs_read_locked_inode(inode);
5784 unlock_new_inode(inode);
5787 * some code call btrfs_commit_transaction while
5788 * holding the i_mutex, so we can't use mutex_lock
5791 if (is_bad_inode(inode) ||
5792 !mutex_trylock(&inode->i_mutex)) {
5795 key.offset = (u64)-1;
5800 if (!extent_locked) {
5801 struct btrfs_ordered_extent *ordered;
5803 btrfs_release_path(root, path);
5805 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5806 lock_end, GFP_NOFS);
5807 ordered = btrfs_lookup_first_ordered_extent(inode,
5810 ordered->file_offset <= lock_end &&
5811 ordered->file_offset + ordered->len > lock_start) {
5812 unlock_extent(&BTRFS_I(inode)->io_tree,
5813 lock_start, lock_end, GFP_NOFS);
5814 btrfs_start_ordered_extent(inode, ordered, 1);
5815 btrfs_put_ordered_extent(ordered);
5816 key.offset += num_bytes;
5820 btrfs_put_ordered_extent(ordered);
5826 if (nr_extents == 1) {
5827 /* update extent pointer in place */
5828 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5829 new_extents[0].disk_bytenr);
5830 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5831 new_extents[0].disk_num_bytes);
5832 btrfs_mark_buffer_dirty(leaf);
5834 btrfs_drop_extent_cache(inode, key.offset,
5835 key.offset + num_bytes - 1, 0);
5837 ret = btrfs_inc_extent_ref(trans, root,
5838 new_extents[0].disk_bytenr,
5839 new_extents[0].disk_num_bytes,
5841 root->root_key.objectid,
5846 ret = btrfs_free_extent(trans, root,
5847 extent_key->objectid,
5850 btrfs_header_owner(leaf),
5851 btrfs_header_generation(leaf),
5855 btrfs_release_path(root, path);
5856 key.offset += num_bytes;
5864 * drop old extent pointer at first, then insert the
5865 * new pointers one bye one
5867 btrfs_release_path(root, path);
5868 ret = btrfs_drop_extents(trans, root, inode, key.offset,
5869 key.offset + num_bytes,
5870 key.offset, &alloc_hint);
5873 for (i = 0; i < nr_extents; i++) {
5874 if (ext_offset >= new_extents[i].num_bytes) {
5875 ext_offset -= new_extents[i].num_bytes;
5878 extent_len = min(new_extents[i].num_bytes -
5879 ext_offset, num_bytes);
5881 ret = btrfs_insert_empty_item(trans, root,
5886 leaf = path->nodes[0];
5887 fi = btrfs_item_ptr(leaf, path->slots[0],
5888 struct btrfs_file_extent_item);
5889 btrfs_set_file_extent_generation(leaf, fi,
5891 btrfs_set_file_extent_type(leaf, fi,
5892 BTRFS_FILE_EXTENT_REG);
5893 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5894 new_extents[i].disk_bytenr);
5895 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5896 new_extents[i].disk_num_bytes);
5897 btrfs_set_file_extent_ram_bytes(leaf, fi,
5898 new_extents[i].ram_bytes);
5900 btrfs_set_file_extent_compression(leaf, fi,
5901 new_extents[i].compression);
5902 btrfs_set_file_extent_encryption(leaf, fi,
5903 new_extents[i].encryption);
5904 btrfs_set_file_extent_other_encoding(leaf, fi,
5905 new_extents[i].other_encoding);
5907 btrfs_set_file_extent_num_bytes(leaf, fi,
5909 ext_offset += new_extents[i].offset;
5910 btrfs_set_file_extent_offset(leaf, fi,
5912 btrfs_mark_buffer_dirty(leaf);
5914 btrfs_drop_extent_cache(inode, key.offset,
5915 key.offset + extent_len - 1, 0);
5917 ret = btrfs_inc_extent_ref(trans, root,
5918 new_extents[i].disk_bytenr,
5919 new_extents[i].disk_num_bytes,
5921 root->root_key.objectid,
5922 trans->transid, key.objectid);
5924 btrfs_release_path(root, path);
5926 inode_add_bytes(inode, extent_len);
5929 num_bytes -= extent_len;
5930 key.offset += extent_len;
5935 BUG_ON(i >= nr_extents);
5939 if (extent_locked) {
5940 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5941 lock_end, GFP_NOFS);
5945 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5946 key.offset >= search_end)
5953 btrfs_release_path(root, path);
5955 mutex_unlock(&inode->i_mutex);
5956 if (extent_locked) {
5957 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5958 lock_end, GFP_NOFS);
5965 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5966 struct btrfs_root *root,
5967 struct extent_buffer *buf, u64 orig_start)
5972 BUG_ON(btrfs_header_generation(buf) != trans->transid);
5973 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5975 level = btrfs_header_level(buf);
5977 struct btrfs_leaf_ref *ref;
5978 struct btrfs_leaf_ref *orig_ref;
5980 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5984 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5986 btrfs_free_leaf_ref(root, orig_ref);
5990 ref->nritems = orig_ref->nritems;
5991 memcpy(ref->extents, orig_ref->extents,
5992 sizeof(ref->extents[0]) * ref->nritems);
5994 btrfs_free_leaf_ref(root, orig_ref);
5996 ref->root_gen = trans->transid;
5997 ref->bytenr = buf->start;
5998 ref->owner = btrfs_header_owner(buf);
5999 ref->generation = btrfs_header_generation(buf);
6001 ret = btrfs_add_leaf_ref(root, ref, 0);
6003 btrfs_free_leaf_ref(root, ref);
6008 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6009 struct extent_buffer *leaf,
6010 struct btrfs_block_group_cache *group,
6011 struct btrfs_root *target_root)
6013 struct btrfs_key key;
6014 struct inode *inode = NULL;
6015 struct btrfs_file_extent_item *fi;
6017 u64 skip_objectid = 0;
6021 nritems = btrfs_header_nritems(leaf);
6022 for (i = 0; i < nritems; i++) {
6023 btrfs_item_key_to_cpu(leaf, &key, i);
6024 if (key.objectid == skip_objectid ||
6025 key.type != BTRFS_EXTENT_DATA_KEY)
6027 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6028 if (btrfs_file_extent_type(leaf, fi) ==
6029 BTRFS_FILE_EXTENT_INLINE)
6031 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6033 if (!inode || inode->i_ino != key.objectid) {
6035 inode = btrfs_ilookup(target_root->fs_info->sb,
6036 key.objectid, target_root, 1);
6039 skip_objectid = key.objectid;
6042 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6044 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6045 key.offset + num_bytes - 1, GFP_NOFS);
6046 btrfs_drop_extent_cache(inode, key.offset,
6047 key.offset + num_bytes - 1, 1);
6048 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6049 key.offset + num_bytes - 1, GFP_NOFS);
6056 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6057 struct btrfs_root *root,
6058 struct extent_buffer *leaf,
6059 struct btrfs_block_group_cache *group,
6060 struct inode *reloc_inode)
6062 struct btrfs_key key;
6063 struct btrfs_key extent_key;
6064 struct btrfs_file_extent_item *fi;
6065 struct btrfs_leaf_ref *ref;
6066 struct disk_extent *new_extent;
6075 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6076 BUG_ON(!new_extent);
6078 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6082 nritems = btrfs_header_nritems(leaf);
6083 for (i = 0; i < nritems; i++) {
6084 btrfs_item_key_to_cpu(leaf, &key, i);
6085 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6087 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6088 if (btrfs_file_extent_type(leaf, fi) ==
6089 BTRFS_FILE_EXTENT_INLINE)
6091 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6092 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6097 if (bytenr >= group->key.objectid + group->key.offset ||
6098 bytenr + num_bytes <= group->key.objectid)
6101 extent_key.objectid = bytenr;
6102 extent_key.offset = num_bytes;
6103 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6105 ret = get_new_locations(reloc_inode, &extent_key,
6106 group->key.objectid, 1,
6107 &new_extent, &nr_extent);
6112 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6113 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6114 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6115 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6117 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6118 new_extent->disk_bytenr);
6119 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6120 new_extent->disk_num_bytes);
6121 btrfs_mark_buffer_dirty(leaf);
6123 ret = btrfs_inc_extent_ref(trans, root,
6124 new_extent->disk_bytenr,
6125 new_extent->disk_num_bytes,
6127 root->root_key.objectid,
6128 trans->transid, key.objectid);
6131 ret = btrfs_free_extent(trans, root,
6132 bytenr, num_bytes, leaf->start,
6133 btrfs_header_owner(leaf),
6134 btrfs_header_generation(leaf),
6140 BUG_ON(ext_index + 1 != ref->nritems);
6141 btrfs_free_leaf_ref(root, ref);
6145 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6146 struct btrfs_root *root)
6148 struct btrfs_root *reloc_root;
6151 if (root->reloc_root) {
6152 reloc_root = root->reloc_root;
6153 root->reloc_root = NULL;
6154 list_add(&reloc_root->dead_list,
6155 &root->fs_info->dead_reloc_roots);
6157 btrfs_set_root_bytenr(&reloc_root->root_item,
6158 reloc_root->node->start);
6159 btrfs_set_root_level(&root->root_item,
6160 btrfs_header_level(reloc_root->node));
6161 memset(&reloc_root->root_item.drop_progress, 0,
6162 sizeof(struct btrfs_disk_key));
6163 reloc_root->root_item.drop_level = 0;
6165 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6166 &reloc_root->root_key,
6167 &reloc_root->root_item);
6173 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6175 struct btrfs_trans_handle *trans;
6176 struct btrfs_root *reloc_root;
6177 struct btrfs_root *prev_root = NULL;
6178 struct list_head dead_roots;
6182 INIT_LIST_HEAD(&dead_roots);
6183 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6185 while (!list_empty(&dead_roots)) {
6186 reloc_root = list_entry(dead_roots.prev,
6187 struct btrfs_root, dead_list);
6188 list_del_init(&reloc_root->dead_list);
6190 BUG_ON(reloc_root->commit_root != NULL);
6192 trans = btrfs_join_transaction(root, 1);
6195 mutex_lock(&root->fs_info->drop_mutex);
6196 ret = btrfs_drop_snapshot(trans, reloc_root);
6199 mutex_unlock(&root->fs_info->drop_mutex);
6201 nr = trans->blocks_used;
6202 ret = btrfs_end_transaction(trans, root);
6204 btrfs_btree_balance_dirty(root, nr);
6207 free_extent_buffer(reloc_root->node);
6209 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6210 &reloc_root->root_key);
6212 mutex_unlock(&root->fs_info->drop_mutex);
6214 nr = trans->blocks_used;
6215 ret = btrfs_end_transaction(trans, root);
6217 btrfs_btree_balance_dirty(root, nr);
6220 prev_root = reloc_root;
6223 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6229 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6231 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6235 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6237 struct btrfs_root *reloc_root;
6238 struct btrfs_trans_handle *trans;
6239 struct btrfs_key location;
6243 mutex_lock(&root->fs_info->tree_reloc_mutex);
6244 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6246 found = !list_empty(&root->fs_info->dead_reloc_roots);
6247 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6250 trans = btrfs_start_transaction(root, 1);
6252 ret = btrfs_commit_transaction(trans, root);
6256 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6257 location.offset = (u64)-1;
6258 location.type = BTRFS_ROOT_ITEM_KEY;
6260 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6261 BUG_ON(!reloc_root);
6262 btrfs_orphan_cleanup(reloc_root);
6266 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6267 struct btrfs_root *root)
6269 struct btrfs_root *reloc_root;
6270 struct extent_buffer *eb;
6271 struct btrfs_root_item *root_item;
6272 struct btrfs_key root_key;
6275 BUG_ON(!root->ref_cows);
6276 if (root->reloc_root)
6279 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6282 ret = btrfs_copy_root(trans, root, root->commit_root,
6283 &eb, BTRFS_TREE_RELOC_OBJECTID);
6286 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6287 root_key.offset = root->root_key.objectid;
6288 root_key.type = BTRFS_ROOT_ITEM_KEY;
6290 memcpy(root_item, &root->root_item, sizeof(root_item));
6291 btrfs_set_root_refs(root_item, 0);
6292 btrfs_set_root_bytenr(root_item, eb->start);
6293 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6294 btrfs_set_root_generation(root_item, trans->transid);
6296 btrfs_tree_unlock(eb);
6297 free_extent_buffer(eb);
6299 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6300 &root_key, root_item);
6304 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6306 BUG_ON(!reloc_root);
6307 reloc_root->last_trans = trans->transid;
6308 reloc_root->commit_root = NULL;
6309 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6311 root->reloc_root = reloc_root;
6316 * Core function of space balance.
6318 * The idea is using reloc trees to relocate tree blocks in reference
6319 * counted roots. There is one reloc tree for each subvol, and all
6320 * reloc trees share same root key objectid. Reloc trees are snapshots
6321 * of the latest committed roots of subvols (root->commit_root).
6323 * To relocate a tree block referenced by a subvol, there are two steps.
6324 * COW the block through subvol's reloc tree, then update block pointer
6325 * in the subvol to point to the new block. Since all reloc trees share
6326 * same root key objectid, doing special handing for tree blocks owned
6327 * by them is easy. Once a tree block has been COWed in one reloc tree,
6328 * we can use the resulting new block directly when the same block is
6329 * required to COW again through other reloc trees. By this way, relocated
6330 * tree blocks are shared between reloc trees, so they are also shared
6333 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6334 struct btrfs_root *root,
6335 struct btrfs_path *path,
6336 struct btrfs_key *first_key,
6337 struct btrfs_ref_path *ref_path,
6338 struct btrfs_block_group_cache *group,
6339 struct inode *reloc_inode)
6341 struct btrfs_root *reloc_root;
6342 struct extent_buffer *eb = NULL;
6343 struct btrfs_key *keys;
6347 int lowest_level = 0;
6350 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6351 lowest_level = ref_path->owner_objectid;
6353 if (!root->ref_cows) {
6354 path->lowest_level = lowest_level;
6355 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6357 path->lowest_level = 0;
6358 btrfs_release_path(root, path);
6362 mutex_lock(&root->fs_info->tree_reloc_mutex);
6363 ret = init_reloc_tree(trans, root);
6365 reloc_root = root->reloc_root;
6367 shared_level = ref_path->shared_level;
6368 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6370 keys = ref_path->node_keys;
6371 nodes = ref_path->new_nodes;
6372 memset(&keys[shared_level + 1], 0,
6373 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6374 memset(&nodes[shared_level + 1], 0,
6375 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6377 if (nodes[lowest_level] == 0) {
6378 path->lowest_level = lowest_level;
6379 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6382 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6383 eb = path->nodes[level];
6384 if (!eb || eb == reloc_root->node)
6386 nodes[level] = eb->start;
6388 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6390 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6393 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6394 eb = path->nodes[0];
6395 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6396 group, reloc_inode);
6399 btrfs_release_path(reloc_root, path);
6401 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6407 * replace tree blocks in the fs tree with tree blocks in
6410 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6413 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6414 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6417 extent_buffer_get(path->nodes[0]);
6418 eb = path->nodes[0];
6419 btrfs_release_path(reloc_root, path);
6420 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6422 free_extent_buffer(eb);
6425 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6426 path->lowest_level = 0;
6430 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6431 struct btrfs_root *root,
6432 struct btrfs_path *path,
6433 struct btrfs_key *first_key,
6434 struct btrfs_ref_path *ref_path)
6438 ret = relocate_one_path(trans, root, path, first_key,
6439 ref_path, NULL, NULL);
6445 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6446 struct btrfs_root *extent_root,
6447 struct btrfs_path *path,
6448 struct btrfs_key *extent_key)
6452 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6455 ret = btrfs_del_item(trans, extent_root, path);
6457 btrfs_release_path(extent_root, path);
6461 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6462 struct btrfs_ref_path *ref_path)
6464 struct btrfs_key root_key;
6466 root_key.objectid = ref_path->root_objectid;
6467 root_key.type = BTRFS_ROOT_ITEM_KEY;
6468 if (is_cowonly_root(ref_path->root_objectid))
6469 root_key.offset = 0;
6471 root_key.offset = (u64)-1;
6473 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6476 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6477 struct btrfs_path *path,
6478 struct btrfs_key *extent_key,
6479 struct btrfs_block_group_cache *group,
6480 struct inode *reloc_inode, int pass)
6482 struct btrfs_trans_handle *trans;
6483 struct btrfs_root *found_root;
6484 struct btrfs_ref_path *ref_path = NULL;
6485 struct disk_extent *new_extents = NULL;
6490 struct btrfs_key first_key;
6494 trans = btrfs_start_transaction(extent_root, 1);
6497 if (extent_key->objectid == 0) {
6498 ret = del_extent_zero(trans, extent_root, path, extent_key);
6502 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6508 for (loops = 0; ; loops++) {
6510 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6511 extent_key->objectid);
6513 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6520 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6521 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6524 found_root = read_ref_root(extent_root->fs_info, ref_path);
6525 BUG_ON(!found_root);
6527 * for reference counted tree, only process reference paths
6528 * rooted at the latest committed root.
6530 if (found_root->ref_cows &&
6531 ref_path->root_generation != found_root->root_key.offset)
6534 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6537 * copy data extents to new locations
6539 u64 group_start = group->key.objectid;
6540 ret = relocate_data_extent(reloc_inode,
6549 level = ref_path->owner_objectid;
6552 if (prev_block != ref_path->nodes[level]) {
6553 struct extent_buffer *eb;
6554 u64 block_start = ref_path->nodes[level];
6555 u64 block_size = btrfs_level_size(found_root, level);
6557 eb = read_tree_block(found_root, block_start,
6559 btrfs_tree_lock(eb);
6560 BUG_ON(level != btrfs_header_level(eb));
6563 btrfs_item_key_to_cpu(eb, &first_key, 0);
6565 btrfs_node_key_to_cpu(eb, &first_key, 0);
6567 btrfs_tree_unlock(eb);
6568 free_extent_buffer(eb);
6569 prev_block = block_start;
6572 mutex_lock(&extent_root->fs_info->trans_mutex);
6573 btrfs_record_root_in_trans(found_root);
6574 mutex_unlock(&extent_root->fs_info->trans_mutex);
6575 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6577 * try to update data extent references while
6578 * keeping metadata shared between snapshots.
6581 ret = relocate_one_path(trans, found_root,
6582 path, &first_key, ref_path,
6583 group, reloc_inode);
6589 * use fallback method to process the remaining
6593 u64 group_start = group->key.objectid;
6594 new_extents = kmalloc(sizeof(*new_extents),
6597 ret = get_new_locations(reloc_inode,
6605 ret = replace_one_extent(trans, found_root,
6607 &first_key, ref_path,
6608 new_extents, nr_extents);
6610 ret = relocate_tree_block(trans, found_root, path,
6611 &first_key, ref_path);
6618 btrfs_end_transaction(trans, extent_root);
6625 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6628 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6629 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6631 num_devices = root->fs_info->fs_devices->rw_devices;
6632 if (num_devices == 1) {
6633 stripped |= BTRFS_BLOCK_GROUP_DUP;
6634 stripped = flags & ~stripped;
6636 /* turn raid0 into single device chunks */
6637 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6640 /* turn mirroring into duplication */
6641 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6642 BTRFS_BLOCK_GROUP_RAID10))
6643 return stripped | BTRFS_BLOCK_GROUP_DUP;
6646 /* they already had raid on here, just return */
6647 if (flags & stripped)
6650 stripped |= BTRFS_BLOCK_GROUP_DUP;
6651 stripped = flags & ~stripped;
6653 /* switch duplicated blocks with raid1 */
6654 if (flags & BTRFS_BLOCK_GROUP_DUP)
6655 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6657 /* turn single device chunks into raid0 */
6658 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6663 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6664 struct btrfs_block_group_cache *shrink_block_group,
6667 struct btrfs_trans_handle *trans;
6668 u64 new_alloc_flags;
6671 spin_lock(&shrink_block_group->lock);
6672 if (btrfs_block_group_used(&shrink_block_group->item) +
6673 shrink_block_group->reserved > 0) {
6674 spin_unlock(&shrink_block_group->lock);
6676 trans = btrfs_start_transaction(root, 1);
6677 spin_lock(&shrink_block_group->lock);
6679 new_alloc_flags = update_block_group_flags(root,
6680 shrink_block_group->flags);
6681 if (new_alloc_flags != shrink_block_group->flags) {
6683 btrfs_block_group_used(&shrink_block_group->item);
6685 calc = shrink_block_group->key.offset;
6687 spin_unlock(&shrink_block_group->lock);
6689 do_chunk_alloc(trans, root->fs_info->extent_root,
6690 calc + 2 * 1024 * 1024, new_alloc_flags, force);
6692 btrfs_end_transaction(trans, root);
6694 spin_unlock(&shrink_block_group->lock);
6699 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6700 struct btrfs_block_group_cache *group)
6703 __alloc_chunk_for_shrink(root, group, 1);
6704 set_block_group_readonly(group);
6709 * checks to see if its even possible to relocate this block group.
6711 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6712 * ok to go ahead and try.
6714 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6716 struct btrfs_block_group_cache *block_group;
6717 struct btrfs_space_info *space_info;
6718 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6719 struct btrfs_device *device;
6723 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6725 /* odd, couldn't find the block group, leave it alone */
6729 /* no bytes used, we're good */
6730 if (!btrfs_block_group_used(&block_group->item))
6733 space_info = block_group->space_info;
6734 spin_lock(&space_info->lock);
6736 full = space_info->full;
6739 * if this is the last block group we have in this space, we can't
6742 if (space_info->total_bytes == block_group->key.offset) {
6744 spin_unlock(&space_info->lock);
6749 * need to make sure we have room in the space to handle all of the
6750 * extents from this block group. If we can, we're good
6752 if (space_info->bytes_used + space_info->bytes_reserved +
6753 space_info->bytes_pinned + space_info->bytes_readonly +
6754 btrfs_block_group_used(&block_group->item) <
6755 space_info->total_bytes) {
6756 spin_unlock(&space_info->lock);
6759 spin_unlock(&space_info->lock);
6762 * ok we don't have enough space, but maybe we have free space on our
6763 * devices to allocate new chunks for relocation, so loop through our
6764 * alloc devices and guess if we have enough space. However, if we
6765 * were marked as full, then we know there aren't enough chunks, and we
6772 mutex_lock(&root->fs_info->chunk_mutex);
6773 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6774 u64 min_free = btrfs_block_group_used(&block_group->item);
6775 u64 dev_offset, max_avail;
6778 * check to make sure we can actually find a chunk with enough
6779 * space to fit our block group in.
6781 if (device->total_bytes > device->bytes_used + min_free) {
6782 ret = find_free_dev_extent(NULL, device, min_free,
6783 &dev_offset, &max_avail);
6789 mutex_unlock(&root->fs_info->chunk_mutex);
6791 btrfs_put_block_group(block_group);
6795 static int find_first_block_group(struct btrfs_root *root,
6796 struct btrfs_path *path, struct btrfs_key *key)
6799 struct btrfs_key found_key;
6800 struct extent_buffer *leaf;
6803 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6808 slot = path->slots[0];
6809 leaf = path->nodes[0];
6810 if (slot >= btrfs_header_nritems(leaf)) {
6811 ret = btrfs_next_leaf(root, path);
6818 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6820 if (found_key.objectid >= key->objectid &&
6821 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6832 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6834 struct btrfs_block_group_cache *block_group;
6835 struct btrfs_space_info *space_info;
6836 struct btrfs_caching_control *caching_ctl;
6839 down_write(&info->extent_commit_sem);
6840 while (!list_empty(&info->caching_block_groups)) {
6841 caching_ctl = list_entry(info->caching_block_groups.next,
6842 struct btrfs_caching_control, list);
6843 list_del(&caching_ctl->list);
6844 put_caching_control(caching_ctl);
6846 up_write(&info->extent_commit_sem);
6848 spin_lock(&info->block_group_cache_lock);
6849 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6850 block_group = rb_entry(n, struct btrfs_block_group_cache,
6852 rb_erase(&block_group->cache_node,
6853 &info->block_group_cache_tree);
6854 spin_unlock(&info->block_group_cache_lock);
6856 down_write(&block_group->space_info->groups_sem);
6857 list_del(&block_group->list);
6858 up_write(&block_group->space_info->groups_sem);
6860 if (block_group->cached == BTRFS_CACHE_STARTED)
6861 wait_block_group_cache_done(block_group);
6863 btrfs_remove_free_space_cache(block_group);
6865 WARN_ON(atomic_read(&block_group->count) != 1);
6868 spin_lock(&info->block_group_cache_lock);
6870 spin_unlock(&info->block_group_cache_lock);
6872 /* now that all the block groups are freed, go through and
6873 * free all the space_info structs. This is only called during
6874 * the final stages of unmount, and so we know nobody is
6875 * using them. We call synchronize_rcu() once before we start,
6876 * just to be on the safe side.
6880 while(!list_empty(&info->space_info)) {
6881 space_info = list_entry(info->space_info.next,
6882 struct btrfs_space_info,
6885 list_del(&space_info->list);
6891 int btrfs_read_block_groups(struct btrfs_root *root)
6893 struct btrfs_path *path;
6895 struct btrfs_block_group_cache *cache;
6896 struct btrfs_fs_info *info = root->fs_info;
6897 struct btrfs_space_info *space_info;
6898 struct btrfs_key key;
6899 struct btrfs_key found_key;
6900 struct extent_buffer *leaf;
6902 root = info->extent_root;
6905 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6906 path = btrfs_alloc_path();
6911 ret = find_first_block_group(root, path, &key);
6919 leaf = path->nodes[0];
6920 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6921 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6927 atomic_set(&cache->count, 1);
6928 spin_lock_init(&cache->lock);
6929 spin_lock_init(&cache->tree_lock);
6930 cache->fs_info = info;
6931 INIT_LIST_HEAD(&cache->list);
6932 INIT_LIST_HEAD(&cache->cluster_list);
6935 * we only want to have 32k of ram per block group for keeping
6936 * track of free space, and if we pass 1/2 of that we want to
6937 * start converting things over to using bitmaps
6939 cache->extents_thresh = ((1024 * 32) / 2) /
6940 sizeof(struct btrfs_free_space);
6942 read_extent_buffer(leaf, &cache->item,
6943 btrfs_item_ptr_offset(leaf, path->slots[0]),
6944 sizeof(cache->item));
6945 memcpy(&cache->key, &found_key, sizeof(found_key));
6947 key.objectid = found_key.objectid + found_key.offset;
6948 btrfs_release_path(root, path);
6949 cache->flags = btrfs_block_group_flags(&cache->item);
6950 cache->sectorsize = root->sectorsize;
6953 * check for two cases, either we are full, and therefore
6954 * don't need to bother with the caching work since we won't
6955 * find any space, or we are empty, and we can just add all
6956 * the space in and be done with it. This saves us _alot_ of
6957 * time, particularly in the full case.
6959 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
6960 cache->last_byte_to_unpin = (u64)-1;
6961 cache->cached = BTRFS_CACHE_FINISHED;
6962 } else if (btrfs_block_group_used(&cache->item) == 0) {
6963 exclude_super_stripes(root, cache);
6964 cache->last_byte_to_unpin = (u64)-1;
6965 cache->cached = BTRFS_CACHE_FINISHED;
6966 add_new_free_space(cache, root->fs_info,
6968 found_key.objectid +
6970 free_excluded_extents(root, cache);
6973 ret = update_space_info(info, cache->flags, found_key.offset,
6974 btrfs_block_group_used(&cache->item),
6977 cache->space_info = space_info;
6978 down_write(&space_info->groups_sem);
6979 list_add_tail(&cache->list, &space_info->block_groups);
6980 up_write(&space_info->groups_sem);
6982 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6985 set_avail_alloc_bits(root->fs_info, cache->flags);
6986 if (btrfs_chunk_readonly(root, cache->key.objectid))
6987 set_block_group_readonly(cache);
6991 btrfs_free_path(path);
6995 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6996 struct btrfs_root *root, u64 bytes_used,
6997 u64 type, u64 chunk_objectid, u64 chunk_offset,
7001 struct btrfs_root *extent_root;
7002 struct btrfs_block_group_cache *cache;
7004 extent_root = root->fs_info->extent_root;
7006 root->fs_info->last_trans_log_full_commit = trans->transid;
7008 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7012 cache->key.objectid = chunk_offset;
7013 cache->key.offset = size;
7014 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7015 cache->sectorsize = root->sectorsize;
7018 * we only want to have 32k of ram per block group for keeping track
7019 * of free space, and if we pass 1/2 of that we want to start
7020 * converting things over to using bitmaps
7022 cache->extents_thresh = ((1024 * 32) / 2) /
7023 sizeof(struct btrfs_free_space);
7024 atomic_set(&cache->count, 1);
7025 spin_lock_init(&cache->lock);
7026 spin_lock_init(&cache->tree_lock);
7027 INIT_LIST_HEAD(&cache->list);
7028 INIT_LIST_HEAD(&cache->cluster_list);
7030 btrfs_set_block_group_used(&cache->item, bytes_used);
7031 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7032 cache->flags = type;
7033 btrfs_set_block_group_flags(&cache->item, type);
7035 cache->last_byte_to_unpin = (u64)-1;
7036 cache->cached = BTRFS_CACHE_FINISHED;
7037 exclude_super_stripes(root, cache);
7039 add_new_free_space(cache, root->fs_info, chunk_offset,
7040 chunk_offset + size);
7042 free_excluded_extents(root, cache);
7044 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7045 &cache->space_info);
7047 down_write(&cache->space_info->groups_sem);
7048 list_add_tail(&cache->list, &cache->space_info->block_groups);
7049 up_write(&cache->space_info->groups_sem);
7051 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7054 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7055 sizeof(cache->item));
7058 set_avail_alloc_bits(extent_root->fs_info, type);
7063 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7064 struct btrfs_root *root, u64 group_start)
7066 struct btrfs_path *path;
7067 struct btrfs_block_group_cache *block_group;
7068 struct btrfs_free_cluster *cluster;
7069 struct btrfs_key key;
7072 root = root->fs_info->extent_root;
7074 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7075 BUG_ON(!block_group);
7076 BUG_ON(!block_group->ro);
7078 memcpy(&key, &block_group->key, sizeof(key));
7080 /* make sure this block group isn't part of an allocation cluster */
7081 cluster = &root->fs_info->data_alloc_cluster;
7082 spin_lock(&cluster->refill_lock);
7083 btrfs_return_cluster_to_free_space(block_group, cluster);
7084 spin_unlock(&cluster->refill_lock);
7087 * make sure this block group isn't part of a metadata
7088 * allocation cluster
7090 cluster = &root->fs_info->meta_alloc_cluster;
7091 spin_lock(&cluster->refill_lock);
7092 btrfs_return_cluster_to_free_space(block_group, cluster);
7093 spin_unlock(&cluster->refill_lock);
7095 path = btrfs_alloc_path();
7098 spin_lock(&root->fs_info->block_group_cache_lock);
7099 rb_erase(&block_group->cache_node,
7100 &root->fs_info->block_group_cache_tree);
7101 spin_unlock(&root->fs_info->block_group_cache_lock);
7103 down_write(&block_group->space_info->groups_sem);
7105 * we must use list_del_init so people can check to see if they
7106 * are still on the list after taking the semaphore
7108 list_del_init(&block_group->list);
7109 up_write(&block_group->space_info->groups_sem);
7111 if (block_group->cached == BTRFS_CACHE_STARTED)
7112 wait_block_group_cache_done(block_group);
7114 btrfs_remove_free_space_cache(block_group);
7116 spin_lock(&block_group->space_info->lock);
7117 block_group->space_info->total_bytes -= block_group->key.offset;
7118 block_group->space_info->bytes_readonly -= block_group->key.offset;
7119 spin_unlock(&block_group->space_info->lock);
7121 btrfs_clear_space_info_full(root->fs_info);
7123 btrfs_put_block_group(block_group);
7124 btrfs_put_block_group(block_group);
7126 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7132 ret = btrfs_del_item(trans, root, path);
7134 btrfs_free_path(path);