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_reserved_extents(struct btrfs_root *root,
36 u64 bytenr, u64 num, int reserve);
37 static int update_block_group(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 u64 bytenr, u64 num_bytes, int alloc,
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);
61 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
62 struct btrfs_root *extent_root, u64 alloc_bytes,
63 u64 flags, int force);
66 block_group_cache_done(struct btrfs_block_group_cache *cache)
69 return cache->cached == BTRFS_CACHE_FINISHED;
72 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
74 return (cache->flags & bits) == bits;
78 * this adds the block group to the fs_info rb tree for the block group
81 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
82 struct btrfs_block_group_cache *block_group)
85 struct rb_node *parent = NULL;
86 struct btrfs_block_group_cache *cache;
88 spin_lock(&info->block_group_cache_lock);
89 p = &info->block_group_cache_tree.rb_node;
93 cache = rb_entry(parent, struct btrfs_block_group_cache,
95 if (block_group->key.objectid < cache->key.objectid) {
97 } else if (block_group->key.objectid > cache->key.objectid) {
100 spin_unlock(&info->block_group_cache_lock);
105 rb_link_node(&block_group->cache_node, parent, p);
106 rb_insert_color(&block_group->cache_node,
107 &info->block_group_cache_tree);
108 spin_unlock(&info->block_group_cache_lock);
114 * This will return the block group at or after bytenr if contains is 0, else
115 * it will return the block group that contains the bytenr
117 static struct btrfs_block_group_cache *
118 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
121 struct btrfs_block_group_cache *cache, *ret = NULL;
125 spin_lock(&info->block_group_cache_lock);
126 n = info->block_group_cache_tree.rb_node;
129 cache = rb_entry(n, struct btrfs_block_group_cache,
131 end = cache->key.objectid + cache->key.offset - 1;
132 start = cache->key.objectid;
134 if (bytenr < start) {
135 if (!contains && (!ret || start < ret->key.objectid))
138 } else if (bytenr > start) {
139 if (contains && bytenr <= end) {
150 atomic_inc(&ret->count);
151 spin_unlock(&info->block_group_cache_lock);
156 void btrfs_free_super_mirror_extents(struct btrfs_fs_info *info)
158 u64 start, end, last = 0;
162 ret = find_first_extent_bit(&info->pinned_extents, last,
163 &start, &end, EXTENT_LOCKED);
167 unlock_extent(&info->pinned_extents, start, end, GFP_NOFS);
172 static int remove_sb_from_cache(struct btrfs_root *root,
173 struct btrfs_block_group_cache *cache)
175 struct btrfs_fs_info *fs_info = root->fs_info;
181 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
182 bytenr = btrfs_sb_offset(i);
183 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
184 cache->key.objectid, bytenr,
185 0, &logical, &nr, &stripe_len);
188 try_lock_extent(&fs_info->pinned_extents,
190 logical[nr] + stripe_len - 1, GFP_NOFS);
199 * this is only called by cache_block_group, since we could have freed extents
200 * we need to check the pinned_extents for any extents that can't be used yet
201 * since their free space will be released as soon as the transaction commits.
203 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
204 struct btrfs_fs_info *info, u64 start, u64 end)
206 u64 extent_start, extent_end, size, total_added = 0;
209 while (start < end) {
210 ret = find_first_extent_bit(&info->pinned_extents, start,
211 &extent_start, &extent_end,
212 EXTENT_DIRTY|EXTENT_LOCKED|
217 if (extent_start == start) {
218 start = extent_end + 1;
219 } else if (extent_start > start && extent_start < end) {
220 size = extent_start - start;
222 ret = btrfs_add_free_space(block_group, start,
225 start = extent_end + 1;
234 ret = btrfs_add_free_space(block_group, start, size);
241 DEFINE_MUTEX(discard_mutex);
244 * if async kthreads are running when we cross transactions, we mark any pinned
245 * extents with EXTENT_DELALLOC and then let the caching kthreads clean up those
246 * extents when they are done. Also we run this from btrfs_finish_extent_commit
247 * in case there were some pinned extents that were missed because we had
248 * already cached that block group.
250 static void btrfs_discard_pinned_extents(struct btrfs_fs_info *fs_info,
251 struct btrfs_block_group_cache *cache)
253 u64 start, end, last;
259 last = cache->key.objectid;
261 mutex_lock(&discard_mutex);
263 ret = find_first_extent_bit(&fs_info->pinned_extents, last,
264 &start, &end, EXTENT_DELALLOC);
268 if (cache && start >= cache->key.objectid + cache->key.offset)
273 cache = btrfs_lookup_block_group(fs_info, start);
276 start = max(start, cache->key.objectid);
277 end = min(end, cache->key.objectid + cache->key.offset - 1);
279 if (block_group_cache_done(cache))
280 btrfs_add_free_space(cache, start,
284 start = max(start, cache->key.objectid);
285 end = min(end, cache->key.objectid + cache->key.offset - 1);
286 btrfs_add_free_space(cache, start, end - start + 1);
289 clear_extent_bits(&fs_info->pinned_extents, start, end,
290 EXTENT_DELALLOC, GFP_NOFS);
293 if (need_resched()) {
294 mutex_unlock(&discard_mutex);
296 mutex_lock(&discard_mutex);
299 mutex_unlock(&discard_mutex);
302 static int caching_kthread(void *data)
304 struct btrfs_block_group_cache *block_group = data;
305 struct btrfs_fs_info *fs_info = block_group->fs_info;
307 struct btrfs_path *path;
309 struct btrfs_key key;
310 struct extent_buffer *leaf;
316 path = btrfs_alloc_path();
320 atomic_inc(&fs_info->async_caching_threads);
321 atomic_inc(&block_group->space_info->caching_threads);
322 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
324 /* need to make sure the commit_root doesn't disappear */
325 down_read(&fs_info->extent_root->commit_root_sem);
328 * We don't want to deadlock with somebody trying to allocate a new
329 * extent for the extent root while also trying to search the extent
330 * root to add free space. So we skip locking and search the commit
331 * root, since its read-only
333 path->skip_locking = 1;
334 path->search_commit_root = 1;
339 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
340 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
346 if (block_group->fs_info->closing)
349 leaf = path->nodes[0];
350 slot = path->slots[0];
351 if (slot >= btrfs_header_nritems(leaf)) {
352 ret = btrfs_next_leaf(fs_info->extent_root, path);
358 if (need_resched()) {
359 btrfs_release_path(fs_info->extent_root, path);
360 up_read(&fs_info->extent_root->commit_root_sem);
367 btrfs_item_key_to_cpu(leaf, &key, slot);
368 if (key.objectid < block_group->key.objectid)
371 if (key.objectid >= block_group->key.objectid +
372 block_group->key.offset)
375 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
376 total_found += add_new_free_space(block_group,
379 last = key.objectid + key.offset;
382 if (total_found > (1024 * 1024 * 2)) {
384 wake_up(&block_group->caching_q);
391 total_found += add_new_free_space(block_group, fs_info, last,
392 block_group->key.objectid +
393 block_group->key.offset);
395 spin_lock(&block_group->lock);
396 block_group->cached = BTRFS_CACHE_FINISHED;
397 spin_unlock(&block_group->lock);
400 btrfs_free_path(path);
401 up_read(&fs_info->extent_root->commit_root_sem);
402 atomic_dec(&fs_info->async_caching_threads);
403 atomic_dec(&block_group->space_info->caching_threads);
404 wake_up(&block_group->caching_q);
407 btrfs_discard_pinned_extents(fs_info, block_group);
412 static int cache_block_group(struct btrfs_block_group_cache *cache)
414 struct task_struct *tsk;
417 spin_lock(&cache->lock);
418 if (cache->cached != BTRFS_CACHE_NO) {
419 spin_unlock(&cache->lock);
422 cache->cached = BTRFS_CACHE_STARTED;
423 spin_unlock(&cache->lock);
425 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
426 cache->key.objectid);
429 printk(KERN_ERR "error running thread %d\n", ret);
437 * return the block group that starts at or after bytenr
439 static struct btrfs_block_group_cache *
440 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
442 struct btrfs_block_group_cache *cache;
444 cache = block_group_cache_tree_search(info, bytenr, 0);
450 * return the block group that contains the given bytenr
452 struct btrfs_block_group_cache *btrfs_lookup_block_group(
453 struct btrfs_fs_info *info,
456 struct btrfs_block_group_cache *cache;
458 cache = block_group_cache_tree_search(info, bytenr, 1);
463 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
465 if (atomic_dec_and_test(&cache->count))
469 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
472 struct list_head *head = &info->space_info;
473 struct btrfs_space_info *found;
476 list_for_each_entry_rcu(found, head, list) {
477 if (found->flags == flags) {
487 * after adding space to the filesystem, we need to clear the full flags
488 * on all the space infos.
490 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
492 struct list_head *head = &info->space_info;
493 struct btrfs_space_info *found;
496 list_for_each_entry_rcu(found, head, list)
501 static u64 div_factor(u64 num, int factor)
510 u64 btrfs_find_block_group(struct btrfs_root *root,
511 u64 search_start, u64 search_hint, int owner)
513 struct btrfs_block_group_cache *cache;
515 u64 last = max(search_hint, search_start);
522 cache = btrfs_lookup_first_block_group(root->fs_info, last);
526 spin_lock(&cache->lock);
527 last = cache->key.objectid + cache->key.offset;
528 used = btrfs_block_group_used(&cache->item);
530 if ((full_search || !cache->ro) &&
531 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
532 if (used + cache->pinned + cache->reserved <
533 div_factor(cache->key.offset, factor)) {
534 group_start = cache->key.objectid;
535 spin_unlock(&cache->lock);
536 btrfs_put_block_group(cache);
540 spin_unlock(&cache->lock);
541 btrfs_put_block_group(cache);
549 if (!full_search && factor < 10) {
559 /* simple helper to search for an existing extent at a given offset */
560 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
563 struct btrfs_key key;
564 struct btrfs_path *path;
566 path = btrfs_alloc_path();
568 key.objectid = start;
570 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
571 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
573 btrfs_free_path(path);
578 * Back reference rules. Back refs have three main goals:
580 * 1) differentiate between all holders of references to an extent so that
581 * when a reference is dropped we can make sure it was a valid reference
582 * before freeing the extent.
584 * 2) Provide enough information to quickly find the holders of an extent
585 * if we notice a given block is corrupted or bad.
587 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
588 * maintenance. This is actually the same as #2, but with a slightly
589 * different use case.
591 * There are two kinds of back refs. The implicit back refs is optimized
592 * for pointers in non-shared tree blocks. For a given pointer in a block,
593 * back refs of this kind provide information about the block's owner tree
594 * and the pointer's key. These information allow us to find the block by
595 * b-tree searching. The full back refs is for pointers in tree blocks not
596 * referenced by their owner trees. The location of tree block is recorded
597 * in the back refs. Actually the full back refs is generic, and can be
598 * used in all cases the implicit back refs is used. The major shortcoming
599 * of the full back refs is its overhead. Every time a tree block gets
600 * COWed, we have to update back refs entry for all pointers in it.
602 * For a newly allocated tree block, we use implicit back refs for
603 * pointers in it. This means most tree related operations only involve
604 * implicit back refs. For a tree block created in old transaction, the
605 * only way to drop a reference to it is COW it. So we can detect the
606 * event that tree block loses its owner tree's reference and do the
607 * back refs conversion.
609 * When a tree block is COW'd through a tree, there are four cases:
611 * The reference count of the block is one and the tree is the block's
612 * owner tree. Nothing to do in this case.
614 * The reference count of the block is one and the tree is not the
615 * block's owner tree. In this case, full back refs is used for pointers
616 * in the block. Remove these full back refs, add implicit back refs for
617 * every pointers in the new block.
619 * The reference count of the block is greater than one and the tree is
620 * the block's owner tree. In this case, implicit back refs is used for
621 * pointers in the block. Add full back refs for every pointers in the
622 * block, increase lower level extents' reference counts. The original
623 * implicit back refs are entailed to the new block.
625 * The reference count of the block is greater than one and the tree is
626 * not the block's owner tree. Add implicit back refs for every pointer in
627 * the new block, increase lower level extents' reference count.
629 * Back Reference Key composing:
631 * The key objectid corresponds to the first byte in the extent,
632 * The key type is used to differentiate between types of back refs.
633 * There are different meanings of the key offset for different types
636 * File extents can be referenced by:
638 * - multiple snapshots, subvolumes, or different generations in one subvol
639 * - different files inside a single subvolume
640 * - different offsets inside a file (bookend extents in file.c)
642 * The extent ref structure for the implicit back refs has fields for:
644 * - Objectid of the subvolume root
645 * - objectid of the file holding the reference
646 * - original offset in the file
647 * - how many bookend extents
649 * The key offset for the implicit back refs is hash of the first
652 * The extent ref structure for the full back refs has field for:
654 * - number of pointers in the tree leaf
656 * The key offset for the implicit back refs is the first byte of
659 * When a file extent is allocated, The implicit back refs is used.
660 * the fields are filled in:
662 * (root_key.objectid, inode objectid, offset in file, 1)
664 * When a file extent is removed file truncation, we find the
665 * corresponding implicit back refs and check the following fields:
667 * (btrfs_header_owner(leaf), inode objectid, offset in file)
669 * Btree extents can be referenced by:
671 * - Different subvolumes
673 * Both the implicit back refs and the full back refs for tree blocks
674 * only consist of key. The key offset for the implicit back refs is
675 * objectid of block's owner tree. The key offset for the full back refs
676 * is the first byte of parent block.
678 * When implicit back refs is used, information about the lowest key and
679 * level of the tree block are required. These information are stored in
680 * tree block info structure.
683 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
684 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
685 struct btrfs_root *root,
686 struct btrfs_path *path,
687 u64 owner, u32 extra_size)
689 struct btrfs_extent_item *item;
690 struct btrfs_extent_item_v0 *ei0;
691 struct btrfs_extent_ref_v0 *ref0;
692 struct btrfs_tree_block_info *bi;
693 struct extent_buffer *leaf;
694 struct btrfs_key key;
695 struct btrfs_key found_key;
696 u32 new_size = sizeof(*item);
700 leaf = path->nodes[0];
701 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
703 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
704 ei0 = btrfs_item_ptr(leaf, path->slots[0],
705 struct btrfs_extent_item_v0);
706 refs = btrfs_extent_refs_v0(leaf, ei0);
708 if (owner == (u64)-1) {
710 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
711 ret = btrfs_next_leaf(root, path);
715 leaf = path->nodes[0];
717 btrfs_item_key_to_cpu(leaf, &found_key,
719 BUG_ON(key.objectid != found_key.objectid);
720 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
724 ref0 = btrfs_item_ptr(leaf, path->slots[0],
725 struct btrfs_extent_ref_v0);
726 owner = btrfs_ref_objectid_v0(leaf, ref0);
730 btrfs_release_path(root, path);
732 if (owner < BTRFS_FIRST_FREE_OBJECTID)
733 new_size += sizeof(*bi);
735 new_size -= sizeof(*ei0);
736 ret = btrfs_search_slot(trans, root, &key, path,
737 new_size + extra_size, 1);
742 ret = btrfs_extend_item(trans, root, path, new_size);
745 leaf = path->nodes[0];
746 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
747 btrfs_set_extent_refs(leaf, item, refs);
748 /* FIXME: get real generation */
749 btrfs_set_extent_generation(leaf, item, 0);
750 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
751 btrfs_set_extent_flags(leaf, item,
752 BTRFS_EXTENT_FLAG_TREE_BLOCK |
753 BTRFS_BLOCK_FLAG_FULL_BACKREF);
754 bi = (struct btrfs_tree_block_info *)(item + 1);
755 /* FIXME: get first key of the block */
756 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
757 btrfs_set_tree_block_level(leaf, bi, (int)owner);
759 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
761 btrfs_mark_buffer_dirty(leaf);
766 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
768 u32 high_crc = ~(u32)0;
769 u32 low_crc = ~(u32)0;
772 lenum = cpu_to_le64(root_objectid);
773 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
774 lenum = cpu_to_le64(owner);
775 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
776 lenum = cpu_to_le64(offset);
777 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
779 return ((u64)high_crc << 31) ^ (u64)low_crc;
782 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
783 struct btrfs_extent_data_ref *ref)
785 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
786 btrfs_extent_data_ref_objectid(leaf, ref),
787 btrfs_extent_data_ref_offset(leaf, ref));
790 static int match_extent_data_ref(struct extent_buffer *leaf,
791 struct btrfs_extent_data_ref *ref,
792 u64 root_objectid, u64 owner, u64 offset)
794 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
795 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
796 btrfs_extent_data_ref_offset(leaf, ref) != offset)
801 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
802 struct btrfs_root *root,
803 struct btrfs_path *path,
804 u64 bytenr, u64 parent,
806 u64 owner, u64 offset)
808 struct btrfs_key key;
809 struct btrfs_extent_data_ref *ref;
810 struct extent_buffer *leaf;
816 key.objectid = bytenr;
818 key.type = BTRFS_SHARED_DATA_REF_KEY;
821 key.type = BTRFS_EXTENT_DATA_REF_KEY;
822 key.offset = hash_extent_data_ref(root_objectid,
827 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
836 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
837 key.type = BTRFS_EXTENT_REF_V0_KEY;
838 btrfs_release_path(root, path);
839 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
850 leaf = path->nodes[0];
851 nritems = btrfs_header_nritems(leaf);
853 if (path->slots[0] >= nritems) {
854 ret = btrfs_next_leaf(root, path);
860 leaf = path->nodes[0];
861 nritems = btrfs_header_nritems(leaf);
865 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
866 if (key.objectid != bytenr ||
867 key.type != BTRFS_EXTENT_DATA_REF_KEY)
870 ref = btrfs_item_ptr(leaf, path->slots[0],
871 struct btrfs_extent_data_ref);
873 if (match_extent_data_ref(leaf, ref, root_objectid,
876 btrfs_release_path(root, path);
888 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
889 struct btrfs_root *root,
890 struct btrfs_path *path,
891 u64 bytenr, u64 parent,
892 u64 root_objectid, u64 owner,
893 u64 offset, int refs_to_add)
895 struct btrfs_key key;
896 struct extent_buffer *leaf;
901 key.objectid = bytenr;
903 key.type = BTRFS_SHARED_DATA_REF_KEY;
905 size = sizeof(struct btrfs_shared_data_ref);
907 key.type = BTRFS_EXTENT_DATA_REF_KEY;
908 key.offset = hash_extent_data_ref(root_objectid,
910 size = sizeof(struct btrfs_extent_data_ref);
913 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
914 if (ret && ret != -EEXIST)
917 leaf = path->nodes[0];
919 struct btrfs_shared_data_ref *ref;
920 ref = btrfs_item_ptr(leaf, path->slots[0],
921 struct btrfs_shared_data_ref);
923 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
925 num_refs = btrfs_shared_data_ref_count(leaf, ref);
926 num_refs += refs_to_add;
927 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
930 struct btrfs_extent_data_ref *ref;
931 while (ret == -EEXIST) {
932 ref = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_extent_data_ref);
934 if (match_extent_data_ref(leaf, ref, root_objectid,
937 btrfs_release_path(root, path);
939 ret = btrfs_insert_empty_item(trans, root, path, &key,
941 if (ret && ret != -EEXIST)
944 leaf = path->nodes[0];
946 ref = btrfs_item_ptr(leaf, path->slots[0],
947 struct btrfs_extent_data_ref);
949 btrfs_set_extent_data_ref_root(leaf, ref,
951 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
952 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
953 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
955 num_refs = btrfs_extent_data_ref_count(leaf, ref);
956 num_refs += refs_to_add;
957 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
960 btrfs_mark_buffer_dirty(leaf);
963 btrfs_release_path(root, path);
967 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
968 struct btrfs_root *root,
969 struct btrfs_path *path,
972 struct btrfs_key key;
973 struct btrfs_extent_data_ref *ref1 = NULL;
974 struct btrfs_shared_data_ref *ref2 = NULL;
975 struct extent_buffer *leaf;
979 leaf = path->nodes[0];
980 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
982 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
983 ref1 = btrfs_item_ptr(leaf, path->slots[0],
984 struct btrfs_extent_data_ref);
985 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
986 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
987 ref2 = btrfs_item_ptr(leaf, path->slots[0],
988 struct btrfs_shared_data_ref);
989 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
990 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
991 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
992 struct btrfs_extent_ref_v0 *ref0;
993 ref0 = btrfs_item_ptr(leaf, path->slots[0],
994 struct btrfs_extent_ref_v0);
995 num_refs = btrfs_ref_count_v0(leaf, ref0);
1001 BUG_ON(num_refs < refs_to_drop);
1002 num_refs -= refs_to_drop;
1004 if (num_refs == 0) {
1005 ret = btrfs_del_item(trans, root, path);
1007 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1008 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1009 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1010 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1011 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1013 struct btrfs_extent_ref_v0 *ref0;
1014 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1015 struct btrfs_extent_ref_v0);
1016 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1019 btrfs_mark_buffer_dirty(leaf);
1024 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1025 struct btrfs_path *path,
1026 struct btrfs_extent_inline_ref *iref)
1028 struct btrfs_key key;
1029 struct extent_buffer *leaf;
1030 struct btrfs_extent_data_ref *ref1;
1031 struct btrfs_shared_data_ref *ref2;
1034 leaf = path->nodes[0];
1035 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1037 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1038 BTRFS_EXTENT_DATA_REF_KEY) {
1039 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1040 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1042 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1043 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1045 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1046 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1047 struct btrfs_extent_data_ref);
1048 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1049 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1050 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1051 struct btrfs_shared_data_ref);
1052 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1054 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1055 struct btrfs_extent_ref_v0 *ref0;
1056 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1057 struct btrfs_extent_ref_v0);
1058 num_refs = btrfs_ref_count_v0(leaf, ref0);
1066 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1067 struct btrfs_root *root,
1068 struct btrfs_path *path,
1069 u64 bytenr, u64 parent,
1072 struct btrfs_key key;
1075 key.objectid = bytenr;
1077 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1078 key.offset = parent;
1080 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1081 key.offset = root_objectid;
1084 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1087 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1088 if (ret == -ENOENT && parent) {
1089 btrfs_release_path(root, path);
1090 key.type = BTRFS_EXTENT_REF_V0_KEY;
1091 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1100 struct btrfs_root *root,
1101 struct btrfs_path *path,
1102 u64 bytenr, u64 parent,
1105 struct btrfs_key key;
1108 key.objectid = bytenr;
1110 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1111 key.offset = parent;
1113 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1114 key.offset = root_objectid;
1117 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1118 btrfs_release_path(root, path);
1122 static inline int extent_ref_type(u64 parent, u64 owner)
1125 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1127 type = BTRFS_SHARED_BLOCK_REF_KEY;
1129 type = BTRFS_TREE_BLOCK_REF_KEY;
1132 type = BTRFS_SHARED_DATA_REF_KEY;
1134 type = BTRFS_EXTENT_DATA_REF_KEY;
1139 static int find_next_key(struct btrfs_path *path, int level,
1140 struct btrfs_key *key)
1143 for (; level < BTRFS_MAX_LEVEL; level++) {
1144 if (!path->nodes[level])
1146 if (path->slots[level] + 1 >=
1147 btrfs_header_nritems(path->nodes[level]))
1150 btrfs_item_key_to_cpu(path->nodes[level], key,
1151 path->slots[level] + 1);
1153 btrfs_node_key_to_cpu(path->nodes[level], key,
1154 path->slots[level] + 1);
1161 * look for inline back ref. if back ref is found, *ref_ret is set
1162 * to the address of inline back ref, and 0 is returned.
1164 * if back ref isn't found, *ref_ret is set to the address where it
1165 * should be inserted, and -ENOENT is returned.
1167 * if insert is true and there are too many inline back refs, the path
1168 * points to the extent item, and -EAGAIN is returned.
1170 * NOTE: inline back refs are ordered in the same way that back ref
1171 * items in the tree are ordered.
1173 static noinline_for_stack
1174 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 struct btrfs_path *path,
1177 struct btrfs_extent_inline_ref **ref_ret,
1178 u64 bytenr, u64 num_bytes,
1179 u64 parent, u64 root_objectid,
1180 u64 owner, u64 offset, int insert)
1182 struct btrfs_key key;
1183 struct extent_buffer *leaf;
1184 struct btrfs_extent_item *ei;
1185 struct btrfs_extent_inline_ref *iref;
1196 key.objectid = bytenr;
1197 key.type = BTRFS_EXTENT_ITEM_KEY;
1198 key.offset = num_bytes;
1200 want = extent_ref_type(parent, owner);
1202 extra_size = btrfs_extent_inline_ref_size(want);
1203 path->keep_locks = 1;
1206 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1213 leaf = path->nodes[0];
1214 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1215 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1216 if (item_size < sizeof(*ei)) {
1221 ret = convert_extent_item_v0(trans, root, path, owner,
1227 leaf = path->nodes[0];
1228 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1231 BUG_ON(item_size < sizeof(*ei));
1233 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1234 flags = btrfs_extent_flags(leaf, ei);
1236 ptr = (unsigned long)(ei + 1);
1237 end = (unsigned long)ei + item_size;
1239 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1240 ptr += sizeof(struct btrfs_tree_block_info);
1243 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1252 iref = (struct btrfs_extent_inline_ref *)ptr;
1253 type = btrfs_extent_inline_ref_type(leaf, iref);
1257 ptr += btrfs_extent_inline_ref_size(type);
1261 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1262 struct btrfs_extent_data_ref *dref;
1263 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1264 if (match_extent_data_ref(leaf, dref, root_objectid,
1269 if (hash_extent_data_ref_item(leaf, dref) <
1270 hash_extent_data_ref(root_objectid, owner, offset))
1274 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1276 if (parent == ref_offset) {
1280 if (ref_offset < parent)
1283 if (root_objectid == ref_offset) {
1287 if (ref_offset < root_objectid)
1291 ptr += btrfs_extent_inline_ref_size(type);
1293 if (err == -ENOENT && insert) {
1294 if (item_size + extra_size >=
1295 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1300 * To add new inline back ref, we have to make sure
1301 * there is no corresponding back ref item.
1302 * For simplicity, we just do not add new inline back
1303 * ref if there is any kind of item for this block
1305 if (find_next_key(path, 0, &key) == 0 &&
1306 key.objectid == bytenr &&
1307 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1312 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1315 path->keep_locks = 0;
1316 btrfs_unlock_up_safe(path, 1);
1322 * helper to add new inline back ref
1324 static noinline_for_stack
1325 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 struct btrfs_extent_inline_ref *iref,
1329 u64 parent, u64 root_objectid,
1330 u64 owner, u64 offset, int refs_to_add,
1331 struct btrfs_delayed_extent_op *extent_op)
1333 struct extent_buffer *leaf;
1334 struct btrfs_extent_item *ei;
1337 unsigned long item_offset;
1343 leaf = path->nodes[0];
1344 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1345 item_offset = (unsigned long)iref - (unsigned long)ei;
1347 type = extent_ref_type(parent, owner);
1348 size = btrfs_extent_inline_ref_size(type);
1350 ret = btrfs_extend_item(trans, root, path, size);
1353 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1354 refs = btrfs_extent_refs(leaf, ei);
1355 refs += refs_to_add;
1356 btrfs_set_extent_refs(leaf, ei, refs);
1358 __run_delayed_extent_op(extent_op, leaf, ei);
1360 ptr = (unsigned long)ei + item_offset;
1361 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1362 if (ptr < end - size)
1363 memmove_extent_buffer(leaf, ptr + size, ptr,
1366 iref = (struct btrfs_extent_inline_ref *)ptr;
1367 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1368 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1369 struct btrfs_extent_data_ref *dref;
1370 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1371 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1372 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1373 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1374 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1375 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1376 struct btrfs_shared_data_ref *sref;
1377 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1378 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1379 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1380 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1381 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1383 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1385 btrfs_mark_buffer_dirty(leaf);
1389 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root,
1391 struct btrfs_path *path,
1392 struct btrfs_extent_inline_ref **ref_ret,
1393 u64 bytenr, u64 num_bytes, u64 parent,
1394 u64 root_objectid, u64 owner, u64 offset)
1398 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1399 bytenr, num_bytes, parent,
1400 root_objectid, owner, offset, 0);
1404 btrfs_release_path(root, path);
1407 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1408 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1411 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1412 root_objectid, owner, offset);
1418 * helper to update/remove inline back ref
1420 static noinline_for_stack
1421 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root,
1423 struct btrfs_path *path,
1424 struct btrfs_extent_inline_ref *iref,
1426 struct btrfs_delayed_extent_op *extent_op)
1428 struct extent_buffer *leaf;
1429 struct btrfs_extent_item *ei;
1430 struct btrfs_extent_data_ref *dref = NULL;
1431 struct btrfs_shared_data_ref *sref = NULL;
1440 leaf = path->nodes[0];
1441 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1442 refs = btrfs_extent_refs(leaf, ei);
1443 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1444 refs += refs_to_mod;
1445 btrfs_set_extent_refs(leaf, ei, refs);
1447 __run_delayed_extent_op(extent_op, leaf, ei);
1449 type = btrfs_extent_inline_ref_type(leaf, iref);
1451 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1452 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1453 refs = btrfs_extent_data_ref_count(leaf, dref);
1454 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1455 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1456 refs = btrfs_shared_data_ref_count(leaf, sref);
1459 BUG_ON(refs_to_mod != -1);
1462 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1463 refs += refs_to_mod;
1466 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1467 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1469 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1471 size = btrfs_extent_inline_ref_size(type);
1472 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1473 ptr = (unsigned long)iref;
1474 end = (unsigned long)ei + item_size;
1475 if (ptr + size < end)
1476 memmove_extent_buffer(leaf, ptr, ptr + size,
1479 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1482 btrfs_mark_buffer_dirty(leaf);
1486 static noinline_for_stack
1487 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1488 struct btrfs_root *root,
1489 struct btrfs_path *path,
1490 u64 bytenr, u64 num_bytes, u64 parent,
1491 u64 root_objectid, u64 owner,
1492 u64 offset, int refs_to_add,
1493 struct btrfs_delayed_extent_op *extent_op)
1495 struct btrfs_extent_inline_ref *iref;
1498 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1499 bytenr, num_bytes, parent,
1500 root_objectid, owner, offset, 1);
1502 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1503 ret = update_inline_extent_backref(trans, root, path, iref,
1504 refs_to_add, extent_op);
1505 } else if (ret == -ENOENT) {
1506 ret = setup_inline_extent_backref(trans, root, path, iref,
1507 parent, root_objectid,
1508 owner, offset, refs_to_add,
1514 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1515 struct btrfs_root *root,
1516 struct btrfs_path *path,
1517 u64 bytenr, u64 parent, u64 root_objectid,
1518 u64 owner, u64 offset, int refs_to_add)
1521 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1522 BUG_ON(refs_to_add != 1);
1523 ret = insert_tree_block_ref(trans, root, path, bytenr,
1524 parent, root_objectid);
1526 ret = insert_extent_data_ref(trans, root, path, bytenr,
1527 parent, root_objectid,
1528 owner, offset, refs_to_add);
1533 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1534 struct btrfs_root *root,
1535 struct btrfs_path *path,
1536 struct btrfs_extent_inline_ref *iref,
1537 int refs_to_drop, int is_data)
1541 BUG_ON(!is_data && refs_to_drop != 1);
1543 ret = update_inline_extent_backref(trans, root, path, iref,
1544 -refs_to_drop, NULL);
1545 } else if (is_data) {
1546 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1548 ret = btrfs_del_item(trans, root, path);
1553 #ifdef BIO_RW_DISCARD
1554 static void btrfs_issue_discard(struct block_device *bdev,
1557 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1561 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1564 #ifdef BIO_RW_DISCARD
1566 u64 map_length = num_bytes;
1567 struct btrfs_multi_bio *multi = NULL;
1569 /* Tell the block device(s) that the sectors can be discarded */
1570 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1571 bytenr, &map_length, &multi, 0);
1573 struct btrfs_bio_stripe *stripe = multi->stripes;
1576 if (map_length > num_bytes)
1577 map_length = num_bytes;
1579 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1580 btrfs_issue_discard(stripe->dev->bdev,
1593 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1594 struct btrfs_root *root,
1595 u64 bytenr, u64 num_bytes, u64 parent,
1596 u64 root_objectid, u64 owner, u64 offset)
1599 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1600 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1602 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1603 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1604 parent, root_objectid, (int)owner,
1605 BTRFS_ADD_DELAYED_REF, NULL);
1607 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1608 parent, root_objectid, owner, offset,
1609 BTRFS_ADD_DELAYED_REF, NULL);
1614 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1615 struct btrfs_root *root,
1616 u64 bytenr, u64 num_bytes,
1617 u64 parent, u64 root_objectid,
1618 u64 owner, u64 offset, int refs_to_add,
1619 struct btrfs_delayed_extent_op *extent_op)
1621 struct btrfs_path *path;
1622 struct extent_buffer *leaf;
1623 struct btrfs_extent_item *item;
1628 path = btrfs_alloc_path();
1633 path->leave_spinning = 1;
1634 /* this will setup the path even if it fails to insert the back ref */
1635 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1636 path, bytenr, num_bytes, parent,
1637 root_objectid, owner, offset,
1638 refs_to_add, extent_op);
1642 if (ret != -EAGAIN) {
1647 leaf = path->nodes[0];
1648 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1649 refs = btrfs_extent_refs(leaf, item);
1650 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1652 __run_delayed_extent_op(extent_op, leaf, item);
1654 btrfs_mark_buffer_dirty(leaf);
1655 btrfs_release_path(root->fs_info->extent_root, path);
1658 path->leave_spinning = 1;
1660 /* now insert the actual backref */
1661 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1662 path, bytenr, parent, root_objectid,
1663 owner, offset, refs_to_add);
1666 btrfs_free_path(path);
1670 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1671 struct btrfs_root *root,
1672 struct btrfs_delayed_ref_node *node,
1673 struct btrfs_delayed_extent_op *extent_op,
1674 int insert_reserved)
1677 struct btrfs_delayed_data_ref *ref;
1678 struct btrfs_key ins;
1683 ins.objectid = node->bytenr;
1684 ins.offset = node->num_bytes;
1685 ins.type = BTRFS_EXTENT_ITEM_KEY;
1687 ref = btrfs_delayed_node_to_data_ref(node);
1688 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1689 parent = ref->parent;
1691 ref_root = ref->root;
1693 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1695 BUG_ON(extent_op->update_key);
1696 flags |= extent_op->flags_to_set;
1698 ret = alloc_reserved_file_extent(trans, root,
1699 parent, ref_root, flags,
1700 ref->objectid, ref->offset,
1701 &ins, node->ref_mod);
1702 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1703 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1704 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1705 node->num_bytes, parent,
1706 ref_root, ref->objectid,
1707 ref->offset, node->ref_mod,
1709 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1710 ret = __btrfs_free_extent(trans, root, node->bytenr,
1711 node->num_bytes, parent,
1712 ref_root, ref->objectid,
1713 ref->offset, node->ref_mod,
1721 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1722 struct extent_buffer *leaf,
1723 struct btrfs_extent_item *ei)
1725 u64 flags = btrfs_extent_flags(leaf, ei);
1726 if (extent_op->update_flags) {
1727 flags |= extent_op->flags_to_set;
1728 btrfs_set_extent_flags(leaf, ei, flags);
1731 if (extent_op->update_key) {
1732 struct btrfs_tree_block_info *bi;
1733 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1734 bi = (struct btrfs_tree_block_info *)(ei + 1);
1735 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1739 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1740 struct btrfs_root *root,
1741 struct btrfs_delayed_ref_node *node,
1742 struct btrfs_delayed_extent_op *extent_op)
1744 struct btrfs_key key;
1745 struct btrfs_path *path;
1746 struct btrfs_extent_item *ei;
1747 struct extent_buffer *leaf;
1752 path = btrfs_alloc_path();
1756 key.objectid = node->bytenr;
1757 key.type = BTRFS_EXTENT_ITEM_KEY;
1758 key.offset = node->num_bytes;
1761 path->leave_spinning = 1;
1762 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1773 leaf = path->nodes[0];
1774 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1775 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1776 if (item_size < sizeof(*ei)) {
1777 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1783 leaf = path->nodes[0];
1784 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1787 BUG_ON(item_size < sizeof(*ei));
1788 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1789 __run_delayed_extent_op(extent_op, leaf, ei);
1791 btrfs_mark_buffer_dirty(leaf);
1793 btrfs_free_path(path);
1797 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1798 struct btrfs_root *root,
1799 struct btrfs_delayed_ref_node *node,
1800 struct btrfs_delayed_extent_op *extent_op,
1801 int insert_reserved)
1804 struct btrfs_delayed_tree_ref *ref;
1805 struct btrfs_key ins;
1809 ins.objectid = node->bytenr;
1810 ins.offset = node->num_bytes;
1811 ins.type = BTRFS_EXTENT_ITEM_KEY;
1813 ref = btrfs_delayed_node_to_tree_ref(node);
1814 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1815 parent = ref->parent;
1817 ref_root = ref->root;
1819 BUG_ON(node->ref_mod != 1);
1820 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1821 BUG_ON(!extent_op || !extent_op->update_flags ||
1822 !extent_op->update_key);
1823 ret = alloc_reserved_tree_block(trans, root,
1825 extent_op->flags_to_set,
1828 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1829 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1830 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1831 node->num_bytes, parent, ref_root,
1832 ref->level, 0, 1, extent_op);
1833 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1834 ret = __btrfs_free_extent(trans, root, node->bytenr,
1835 node->num_bytes, parent, ref_root,
1836 ref->level, 0, 1, extent_op);
1844 /* helper function to actually process a single delayed ref entry */
1845 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_delayed_ref_node *node,
1848 struct btrfs_delayed_extent_op *extent_op,
1849 int insert_reserved)
1852 if (btrfs_delayed_ref_is_head(node)) {
1853 struct btrfs_delayed_ref_head *head;
1855 * we've hit the end of the chain and we were supposed
1856 * to insert this extent into the tree. But, it got
1857 * deleted before we ever needed to insert it, so all
1858 * we have to do is clean up the accounting
1861 head = btrfs_delayed_node_to_head(node);
1862 if (insert_reserved) {
1863 if (head->is_data) {
1864 ret = btrfs_del_csums(trans, root,
1869 btrfs_update_pinned_extents(root, node->bytenr,
1870 node->num_bytes, 1, 0);
1871 update_reserved_extents(root, node->bytenr,
1872 node->num_bytes, 0);
1874 mutex_unlock(&head->mutex);
1878 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1879 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1880 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1882 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1883 node->type == BTRFS_SHARED_DATA_REF_KEY)
1884 ret = run_delayed_data_ref(trans, root, node, extent_op,
1891 static noinline struct btrfs_delayed_ref_node *
1892 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1894 struct rb_node *node;
1895 struct btrfs_delayed_ref_node *ref;
1896 int action = BTRFS_ADD_DELAYED_REF;
1899 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1900 * this prevents ref count from going down to zero when
1901 * there still are pending delayed ref.
1903 node = rb_prev(&head->node.rb_node);
1907 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1909 if (ref->bytenr != head->node.bytenr)
1911 if (ref->action == action)
1913 node = rb_prev(node);
1915 if (action == BTRFS_ADD_DELAYED_REF) {
1916 action = BTRFS_DROP_DELAYED_REF;
1922 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1923 struct btrfs_root *root,
1924 struct list_head *cluster)
1926 struct btrfs_delayed_ref_root *delayed_refs;
1927 struct btrfs_delayed_ref_node *ref;
1928 struct btrfs_delayed_ref_head *locked_ref = NULL;
1929 struct btrfs_delayed_extent_op *extent_op;
1932 int must_insert_reserved = 0;
1934 delayed_refs = &trans->transaction->delayed_refs;
1937 /* pick a new head ref from the cluster list */
1938 if (list_empty(cluster))
1941 locked_ref = list_entry(cluster->next,
1942 struct btrfs_delayed_ref_head, cluster);
1944 /* grab the lock that says we are going to process
1945 * all the refs for this head */
1946 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1949 * we may have dropped the spin lock to get the head
1950 * mutex lock, and that might have given someone else
1951 * time to free the head. If that's true, it has been
1952 * removed from our list and we can move on.
1954 if (ret == -EAGAIN) {
1962 * record the must insert reserved flag before we
1963 * drop the spin lock.
1965 must_insert_reserved = locked_ref->must_insert_reserved;
1966 locked_ref->must_insert_reserved = 0;
1968 extent_op = locked_ref->extent_op;
1969 locked_ref->extent_op = NULL;
1972 * locked_ref is the head node, so we have to go one
1973 * node back for any delayed ref updates
1975 ref = select_delayed_ref(locked_ref);
1977 /* All delayed refs have been processed, Go ahead
1978 * and send the head node to run_one_delayed_ref,
1979 * so that any accounting fixes can happen
1981 ref = &locked_ref->node;
1983 if (extent_op && must_insert_reserved) {
1989 spin_unlock(&delayed_refs->lock);
1991 ret = run_delayed_extent_op(trans, root,
1997 spin_lock(&delayed_refs->lock);
2001 list_del_init(&locked_ref->cluster);
2006 rb_erase(&ref->rb_node, &delayed_refs->root);
2007 delayed_refs->num_entries--;
2009 spin_unlock(&delayed_refs->lock);
2011 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2012 must_insert_reserved);
2015 btrfs_put_delayed_ref(ref);
2020 spin_lock(&delayed_refs->lock);
2026 * this starts processing the delayed reference count updates and
2027 * extent insertions we have queued up so far. count can be
2028 * 0, which means to process everything in the tree at the start
2029 * of the run (but not newly added entries), or it can be some target
2030 * number you'd like to process.
2032 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2033 struct btrfs_root *root, unsigned long count)
2035 struct rb_node *node;
2036 struct btrfs_delayed_ref_root *delayed_refs;
2037 struct btrfs_delayed_ref_node *ref;
2038 struct list_head cluster;
2040 int run_all = count == (unsigned long)-1;
2043 if (root == root->fs_info->extent_root)
2044 root = root->fs_info->tree_root;
2046 delayed_refs = &trans->transaction->delayed_refs;
2047 INIT_LIST_HEAD(&cluster);
2049 spin_lock(&delayed_refs->lock);
2051 count = delayed_refs->num_entries * 2;
2055 if (!(run_all || run_most) &&
2056 delayed_refs->num_heads_ready < 64)
2060 * go find something we can process in the rbtree. We start at
2061 * the beginning of the tree, and then build a cluster
2062 * of refs to process starting at the first one we are able to
2065 ret = btrfs_find_ref_cluster(trans, &cluster,
2066 delayed_refs->run_delayed_start);
2070 ret = run_clustered_refs(trans, root, &cluster);
2073 count -= min_t(unsigned long, ret, count);
2080 node = rb_first(&delayed_refs->root);
2083 count = (unsigned long)-1;
2086 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2088 if (btrfs_delayed_ref_is_head(ref)) {
2089 struct btrfs_delayed_ref_head *head;
2091 head = btrfs_delayed_node_to_head(ref);
2092 atomic_inc(&ref->refs);
2094 spin_unlock(&delayed_refs->lock);
2095 mutex_lock(&head->mutex);
2096 mutex_unlock(&head->mutex);
2098 btrfs_put_delayed_ref(ref);
2102 node = rb_next(node);
2104 spin_unlock(&delayed_refs->lock);
2105 schedule_timeout(1);
2109 spin_unlock(&delayed_refs->lock);
2113 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
2115 u64 bytenr, u64 num_bytes, u64 flags,
2118 struct btrfs_delayed_extent_op *extent_op;
2121 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2125 extent_op->flags_to_set = flags;
2126 extent_op->update_flags = 1;
2127 extent_op->update_key = 0;
2128 extent_op->is_data = is_data ? 1 : 0;
2130 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2136 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2137 struct btrfs_root *root,
2138 struct btrfs_path *path,
2139 u64 objectid, u64 offset, u64 bytenr)
2141 struct btrfs_delayed_ref_head *head;
2142 struct btrfs_delayed_ref_node *ref;
2143 struct btrfs_delayed_data_ref *data_ref;
2144 struct btrfs_delayed_ref_root *delayed_refs;
2145 struct rb_node *node;
2149 delayed_refs = &trans->transaction->delayed_refs;
2150 spin_lock(&delayed_refs->lock);
2151 head = btrfs_find_delayed_ref_head(trans, bytenr);
2155 if (!mutex_trylock(&head->mutex)) {
2156 atomic_inc(&head->node.refs);
2157 spin_unlock(&delayed_refs->lock);
2159 btrfs_release_path(root->fs_info->extent_root, path);
2161 mutex_lock(&head->mutex);
2162 mutex_unlock(&head->mutex);
2163 btrfs_put_delayed_ref(&head->node);
2167 node = rb_prev(&head->node.rb_node);
2171 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2173 if (ref->bytenr != bytenr)
2177 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2180 data_ref = btrfs_delayed_node_to_data_ref(ref);
2182 node = rb_prev(node);
2184 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2185 if (ref->bytenr == bytenr)
2189 if (data_ref->root != root->root_key.objectid ||
2190 data_ref->objectid != objectid || data_ref->offset != offset)
2195 mutex_unlock(&head->mutex);
2197 spin_unlock(&delayed_refs->lock);
2201 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2202 struct btrfs_root *root,
2203 struct btrfs_path *path,
2204 u64 objectid, u64 offset, u64 bytenr)
2206 struct btrfs_root *extent_root = root->fs_info->extent_root;
2207 struct extent_buffer *leaf;
2208 struct btrfs_extent_data_ref *ref;
2209 struct btrfs_extent_inline_ref *iref;
2210 struct btrfs_extent_item *ei;
2211 struct btrfs_key key;
2215 key.objectid = bytenr;
2216 key.offset = (u64)-1;
2217 key.type = BTRFS_EXTENT_ITEM_KEY;
2219 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2225 if (path->slots[0] == 0)
2229 leaf = path->nodes[0];
2230 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2232 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2236 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2237 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2238 if (item_size < sizeof(*ei)) {
2239 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2243 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2245 if (item_size != sizeof(*ei) +
2246 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2249 if (btrfs_extent_generation(leaf, ei) <=
2250 btrfs_root_last_snapshot(&root->root_item))
2253 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2254 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2255 BTRFS_EXTENT_DATA_REF_KEY)
2258 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2259 if (btrfs_extent_refs(leaf, ei) !=
2260 btrfs_extent_data_ref_count(leaf, ref) ||
2261 btrfs_extent_data_ref_root(leaf, ref) !=
2262 root->root_key.objectid ||
2263 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2264 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2272 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2273 struct btrfs_root *root,
2274 u64 objectid, u64 offset, u64 bytenr)
2276 struct btrfs_path *path;
2280 path = btrfs_alloc_path();
2285 ret = check_committed_ref(trans, root, path, objectid,
2287 if (ret && ret != -ENOENT)
2290 ret2 = check_delayed_ref(trans, root, path, objectid,
2292 } while (ret2 == -EAGAIN);
2294 if (ret2 && ret2 != -ENOENT) {
2299 if (ret != -ENOENT || ret2 != -ENOENT)
2302 btrfs_free_path(path);
2307 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2308 struct extent_buffer *buf, u32 nr_extents)
2310 struct btrfs_key key;
2311 struct btrfs_file_extent_item *fi;
2319 if (!root->ref_cows)
2322 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2324 root_gen = root->root_key.offset;
2327 root_gen = trans->transid - 1;
2330 level = btrfs_header_level(buf);
2331 nritems = btrfs_header_nritems(buf);
2334 struct btrfs_leaf_ref *ref;
2335 struct btrfs_extent_info *info;
2337 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2343 ref->root_gen = root_gen;
2344 ref->bytenr = buf->start;
2345 ref->owner = btrfs_header_owner(buf);
2346 ref->generation = btrfs_header_generation(buf);
2347 ref->nritems = nr_extents;
2348 info = ref->extents;
2350 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2352 btrfs_item_key_to_cpu(buf, &key, i);
2353 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2355 fi = btrfs_item_ptr(buf, i,
2356 struct btrfs_file_extent_item);
2357 if (btrfs_file_extent_type(buf, fi) ==
2358 BTRFS_FILE_EXTENT_INLINE)
2360 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2361 if (disk_bytenr == 0)
2364 info->bytenr = disk_bytenr;
2366 btrfs_file_extent_disk_num_bytes(buf, fi);
2367 info->objectid = key.objectid;
2368 info->offset = key.offset;
2372 ret = btrfs_add_leaf_ref(root, ref, shared);
2373 if (ret == -EEXIST && shared) {
2374 struct btrfs_leaf_ref *old;
2375 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2377 btrfs_remove_leaf_ref(root, old);
2378 btrfs_free_leaf_ref(root, old);
2379 ret = btrfs_add_leaf_ref(root, ref, shared);
2382 btrfs_free_leaf_ref(root, ref);
2388 /* when a block goes through cow, we update the reference counts of
2389 * everything that block points to. The internal pointers of the block
2390 * can be in just about any order, and it is likely to have clusters of
2391 * things that are close together and clusters of things that are not.
2393 * To help reduce the seeks that come with updating all of these reference
2394 * counts, sort them by byte number before actual updates are done.
2396 * struct refsort is used to match byte number to slot in the btree block.
2397 * we sort based on the byte number and then use the slot to actually
2400 * struct refsort is smaller than strcut btrfs_item and smaller than
2401 * struct btrfs_key_ptr. Since we're currently limited to the page size
2402 * for a btree block, there's no way for a kmalloc of refsorts for a
2403 * single node to be bigger than a page.
2411 * for passing into sort()
2413 static int refsort_cmp(const void *a_void, const void *b_void)
2415 const struct refsort *a = a_void;
2416 const struct refsort *b = b_void;
2418 if (a->bytenr < b->bytenr)
2420 if (a->bytenr > b->bytenr)
2426 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2427 struct btrfs_root *root,
2428 struct extent_buffer *buf,
2429 int full_backref, int inc)
2436 struct btrfs_key key;
2437 struct btrfs_file_extent_item *fi;
2441 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2442 u64, u64, u64, u64, u64, u64);
2444 ref_root = btrfs_header_owner(buf);
2445 nritems = btrfs_header_nritems(buf);
2446 level = btrfs_header_level(buf);
2448 if (!root->ref_cows && level == 0)
2452 process_func = btrfs_inc_extent_ref;
2454 process_func = btrfs_free_extent;
2457 parent = buf->start;
2461 for (i = 0; i < nritems; i++) {
2463 btrfs_item_key_to_cpu(buf, &key, i);
2464 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2466 fi = btrfs_item_ptr(buf, i,
2467 struct btrfs_file_extent_item);
2468 if (btrfs_file_extent_type(buf, fi) ==
2469 BTRFS_FILE_EXTENT_INLINE)
2471 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2475 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2476 key.offset -= btrfs_file_extent_offset(buf, fi);
2477 ret = process_func(trans, root, bytenr, num_bytes,
2478 parent, ref_root, key.objectid,
2483 bytenr = btrfs_node_blockptr(buf, i);
2484 num_bytes = btrfs_level_size(root, level - 1);
2485 ret = process_func(trans, root, bytenr, num_bytes,
2486 parent, ref_root, level - 1, 0);
2497 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2498 struct extent_buffer *buf, int full_backref)
2500 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2503 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2504 struct extent_buffer *buf, int full_backref)
2506 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2509 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2510 struct btrfs_root *root,
2511 struct btrfs_path *path,
2512 struct btrfs_block_group_cache *cache)
2515 struct btrfs_root *extent_root = root->fs_info->extent_root;
2517 struct extent_buffer *leaf;
2519 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2524 leaf = path->nodes[0];
2525 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2526 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2527 btrfs_mark_buffer_dirty(leaf);
2528 btrfs_release_path(extent_root, path);
2536 static struct btrfs_block_group_cache *
2537 next_block_group(struct btrfs_root *root,
2538 struct btrfs_block_group_cache *cache)
2540 struct rb_node *node;
2541 spin_lock(&root->fs_info->block_group_cache_lock);
2542 node = rb_next(&cache->cache_node);
2543 btrfs_put_block_group(cache);
2545 cache = rb_entry(node, struct btrfs_block_group_cache,
2547 atomic_inc(&cache->count);
2550 spin_unlock(&root->fs_info->block_group_cache_lock);
2554 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2555 struct btrfs_root *root)
2557 struct btrfs_block_group_cache *cache;
2559 struct btrfs_path *path;
2562 path = btrfs_alloc_path();
2568 err = btrfs_run_delayed_refs(trans, root,
2573 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2577 cache = next_block_group(root, cache);
2587 last = cache->key.objectid + cache->key.offset;
2589 err = write_one_cache_group(trans, root, path, cache);
2591 btrfs_put_block_group(cache);
2594 btrfs_free_path(path);
2598 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2600 struct btrfs_block_group_cache *block_group;
2603 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2604 if (!block_group || block_group->ro)
2607 btrfs_put_block_group(block_group);
2611 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2612 u64 total_bytes, u64 bytes_used,
2613 struct btrfs_space_info **space_info)
2615 struct btrfs_space_info *found;
2617 found = __find_space_info(info, flags);
2619 spin_lock(&found->lock);
2620 found->total_bytes += total_bytes;
2621 found->bytes_used += bytes_used;
2623 spin_unlock(&found->lock);
2624 *space_info = found;
2627 found = kzalloc(sizeof(*found), GFP_NOFS);
2631 INIT_LIST_HEAD(&found->block_groups);
2632 init_rwsem(&found->groups_sem);
2633 spin_lock_init(&found->lock);
2634 found->flags = flags;
2635 found->total_bytes = total_bytes;
2636 found->bytes_used = bytes_used;
2637 found->bytes_pinned = 0;
2638 found->bytes_reserved = 0;
2639 found->bytes_readonly = 0;
2640 found->bytes_delalloc = 0;
2642 found->force_alloc = 0;
2643 *space_info = found;
2644 list_add_rcu(&found->list, &info->space_info);
2645 atomic_set(&found->caching_threads, 0);
2649 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2651 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2652 BTRFS_BLOCK_GROUP_RAID1 |
2653 BTRFS_BLOCK_GROUP_RAID10 |
2654 BTRFS_BLOCK_GROUP_DUP);
2656 if (flags & BTRFS_BLOCK_GROUP_DATA)
2657 fs_info->avail_data_alloc_bits |= extra_flags;
2658 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2659 fs_info->avail_metadata_alloc_bits |= extra_flags;
2660 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2661 fs_info->avail_system_alloc_bits |= extra_flags;
2665 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2667 spin_lock(&cache->space_info->lock);
2668 spin_lock(&cache->lock);
2670 cache->space_info->bytes_readonly += cache->key.offset -
2671 btrfs_block_group_used(&cache->item);
2674 spin_unlock(&cache->lock);
2675 spin_unlock(&cache->space_info->lock);
2678 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2680 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2682 if (num_devices == 1)
2683 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2684 if (num_devices < 4)
2685 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2687 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2688 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2689 BTRFS_BLOCK_GROUP_RAID10))) {
2690 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2693 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2694 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2695 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2698 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2699 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2700 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2701 (flags & BTRFS_BLOCK_GROUP_DUP)))
2702 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2706 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2708 struct btrfs_fs_info *info = root->fs_info;
2712 alloc_profile = info->avail_data_alloc_bits &
2713 info->data_alloc_profile;
2714 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2715 } else if (root == root->fs_info->chunk_root) {
2716 alloc_profile = info->avail_system_alloc_bits &
2717 info->system_alloc_profile;
2718 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2720 alloc_profile = info->avail_metadata_alloc_bits &
2721 info->metadata_alloc_profile;
2722 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2725 return btrfs_reduce_alloc_profile(root, data);
2728 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2732 alloc_target = btrfs_get_alloc_profile(root, 1);
2733 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2738 * for now this just makes sure we have at least 5% of our metadata space free
2741 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2743 struct btrfs_fs_info *info = root->fs_info;
2744 struct btrfs_space_info *meta_sinfo;
2745 u64 alloc_target, thresh;
2746 int committed = 0, ret;
2748 /* get the space info for where the metadata will live */
2749 alloc_target = btrfs_get_alloc_profile(root, 0);
2750 meta_sinfo = __find_space_info(info, alloc_target);
2753 spin_lock(&meta_sinfo->lock);
2754 if (!meta_sinfo->full)
2755 thresh = meta_sinfo->total_bytes * 80;
2757 thresh = meta_sinfo->total_bytes * 95;
2759 do_div(thresh, 100);
2761 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2762 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2763 struct btrfs_trans_handle *trans;
2764 if (!meta_sinfo->full) {
2765 meta_sinfo->force_alloc = 1;
2766 spin_unlock(&meta_sinfo->lock);
2768 trans = btrfs_start_transaction(root, 1);
2772 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2773 2 * 1024 * 1024, alloc_target, 0);
2774 btrfs_end_transaction(trans, root);
2777 spin_unlock(&meta_sinfo->lock);
2781 trans = btrfs_join_transaction(root, 1);
2784 ret = btrfs_commit_transaction(trans, root);
2791 spin_unlock(&meta_sinfo->lock);
2797 * This will check the space that the inode allocates from to make sure we have
2798 * enough space for bytes.
2800 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2803 struct btrfs_space_info *data_sinfo;
2804 int ret = 0, committed = 0;
2806 /* make sure bytes are sectorsize aligned */
2807 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2809 data_sinfo = BTRFS_I(inode)->space_info;
2811 /* make sure we have enough space to handle the data first */
2812 spin_lock(&data_sinfo->lock);
2813 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2814 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2815 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2816 data_sinfo->bytes_may_use < bytes) {
2817 struct btrfs_trans_handle *trans;
2820 * if we don't have enough free bytes in this space then we need
2821 * to alloc a new chunk.
2823 if (!data_sinfo->full) {
2826 data_sinfo->force_alloc = 1;
2827 spin_unlock(&data_sinfo->lock);
2829 alloc_target = btrfs_get_alloc_profile(root, 1);
2830 trans = btrfs_start_transaction(root, 1);
2834 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2835 bytes + 2 * 1024 * 1024,
2837 btrfs_end_transaction(trans, root);
2842 spin_unlock(&data_sinfo->lock);
2844 /* commit the current transaction and try again */
2847 trans = btrfs_join_transaction(root, 1);
2850 ret = btrfs_commit_transaction(trans, root);
2856 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2857 ", %llu bytes_used, %llu bytes_reserved, "
2858 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2859 "%llu total\n", (unsigned long long)bytes,
2860 (unsigned long long)data_sinfo->bytes_delalloc,
2861 (unsigned long long)data_sinfo->bytes_used,
2862 (unsigned long long)data_sinfo->bytes_reserved,
2863 (unsigned long long)data_sinfo->bytes_pinned,
2864 (unsigned long long)data_sinfo->bytes_readonly,
2865 (unsigned long long)data_sinfo->bytes_may_use,
2866 (unsigned long long)data_sinfo->total_bytes);
2869 data_sinfo->bytes_may_use += bytes;
2870 BTRFS_I(inode)->reserved_bytes += bytes;
2871 spin_unlock(&data_sinfo->lock);
2873 return btrfs_check_metadata_free_space(root);
2877 * if there was an error for whatever reason after calling
2878 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2880 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2881 struct inode *inode, u64 bytes)
2883 struct btrfs_space_info *data_sinfo;
2885 /* make sure bytes are sectorsize aligned */
2886 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2888 data_sinfo = BTRFS_I(inode)->space_info;
2889 spin_lock(&data_sinfo->lock);
2890 data_sinfo->bytes_may_use -= bytes;
2891 BTRFS_I(inode)->reserved_bytes -= bytes;
2892 spin_unlock(&data_sinfo->lock);
2895 /* called when we are adding a delalloc extent to the inode's io_tree */
2896 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2899 struct btrfs_space_info *data_sinfo;
2901 /* get the space info for where this inode will be storing its data */
2902 data_sinfo = BTRFS_I(inode)->space_info;
2904 /* make sure we have enough space to handle the data first */
2905 spin_lock(&data_sinfo->lock);
2906 data_sinfo->bytes_delalloc += bytes;
2909 * we are adding a delalloc extent without calling
2910 * btrfs_check_data_free_space first. This happens on a weird
2911 * writepage condition, but shouldn't hurt our accounting
2913 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2914 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2915 BTRFS_I(inode)->reserved_bytes = 0;
2917 data_sinfo->bytes_may_use -= bytes;
2918 BTRFS_I(inode)->reserved_bytes -= bytes;
2921 spin_unlock(&data_sinfo->lock);
2924 /* called when we are clearing an delalloc extent from the inode's io_tree */
2925 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2928 struct btrfs_space_info *info;
2930 info = BTRFS_I(inode)->space_info;
2932 spin_lock(&info->lock);
2933 info->bytes_delalloc -= bytes;
2934 spin_unlock(&info->lock);
2937 static void force_metadata_allocation(struct btrfs_fs_info *info)
2939 struct list_head *head = &info->space_info;
2940 struct btrfs_space_info *found;
2943 list_for_each_entry_rcu(found, head, list) {
2944 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2945 found->force_alloc = 1;
2950 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2951 struct btrfs_root *extent_root, u64 alloc_bytes,
2952 u64 flags, int force)
2954 struct btrfs_space_info *space_info;
2955 struct btrfs_fs_info *fs_info = extent_root->fs_info;
2959 mutex_lock(&fs_info->chunk_mutex);
2961 flags = btrfs_reduce_alloc_profile(extent_root, flags);
2963 space_info = __find_space_info(extent_root->fs_info, flags);
2965 ret = update_space_info(extent_root->fs_info, flags,
2969 BUG_ON(!space_info);
2971 spin_lock(&space_info->lock);
2972 if (space_info->force_alloc) {
2974 space_info->force_alloc = 0;
2976 if (space_info->full) {
2977 spin_unlock(&space_info->lock);
2981 thresh = space_info->total_bytes - space_info->bytes_readonly;
2982 thresh = div_factor(thresh, 6);
2984 (space_info->bytes_used + space_info->bytes_pinned +
2985 space_info->bytes_reserved + alloc_bytes) < thresh) {
2986 spin_unlock(&space_info->lock);
2989 spin_unlock(&space_info->lock);
2992 * if we're doing a data chunk, go ahead and make sure that
2993 * we keep a reasonable number of metadata chunks allocated in the
2996 if (flags & BTRFS_BLOCK_GROUP_DATA) {
2997 fs_info->data_chunk_allocations++;
2998 if (!(fs_info->data_chunk_allocations %
2999 fs_info->metadata_ratio))
3000 force_metadata_allocation(fs_info);
3003 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3005 space_info->full = 1;
3007 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3011 static int update_block_group(struct btrfs_trans_handle *trans,
3012 struct btrfs_root *root,
3013 u64 bytenr, u64 num_bytes, int alloc,
3016 struct btrfs_block_group_cache *cache;
3017 struct btrfs_fs_info *info = root->fs_info;
3018 u64 total = num_bytes;
3022 /* block accounting for super block */
3023 spin_lock(&info->delalloc_lock);
3024 old_val = btrfs_super_bytes_used(&info->super_copy);
3026 old_val += num_bytes;
3028 old_val -= num_bytes;
3029 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3031 /* block accounting for root item */
3032 old_val = btrfs_root_used(&root->root_item);
3034 old_val += num_bytes;
3036 old_val -= num_bytes;
3037 btrfs_set_root_used(&root->root_item, old_val);
3038 spin_unlock(&info->delalloc_lock);
3041 cache = btrfs_lookup_block_group(info, bytenr);
3044 byte_in_group = bytenr - cache->key.objectid;
3045 WARN_ON(byte_in_group > cache->key.offset);
3047 spin_lock(&cache->space_info->lock);
3048 spin_lock(&cache->lock);
3050 old_val = btrfs_block_group_used(&cache->item);
3051 num_bytes = min(total, cache->key.offset - byte_in_group);
3053 old_val += num_bytes;
3054 cache->space_info->bytes_used += num_bytes;
3056 cache->space_info->bytes_readonly -= num_bytes;
3057 btrfs_set_block_group_used(&cache->item, old_val);
3058 spin_unlock(&cache->lock);
3059 spin_unlock(&cache->space_info->lock);
3061 old_val -= num_bytes;
3062 cache->space_info->bytes_used -= num_bytes;
3064 cache->space_info->bytes_readonly += num_bytes;
3065 btrfs_set_block_group_used(&cache->item, old_val);
3066 spin_unlock(&cache->lock);
3067 spin_unlock(&cache->space_info->lock);
3071 ret = btrfs_discard_extent(root, bytenr,
3075 ret = btrfs_add_free_space(cache, bytenr,
3080 btrfs_put_block_group(cache);
3082 bytenr += num_bytes;
3087 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3089 struct btrfs_block_group_cache *cache;
3092 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3096 bytenr = cache->key.objectid;
3097 btrfs_put_block_group(cache);
3102 int btrfs_update_pinned_extents(struct btrfs_root *root,
3103 u64 bytenr, u64 num, int pin, int mark_free)
3106 struct btrfs_block_group_cache *cache;
3107 struct btrfs_fs_info *fs_info = root->fs_info;
3110 set_extent_dirty(&fs_info->pinned_extents,
3111 bytenr, bytenr + num - 1, GFP_NOFS);
3113 clear_extent_dirty(&fs_info->pinned_extents,
3114 bytenr, bytenr + num - 1, GFP_NOFS);
3118 cache = btrfs_lookup_block_group(fs_info, bytenr);
3120 len = min(num, cache->key.offset -
3121 (bytenr - cache->key.objectid));
3123 spin_lock(&cache->space_info->lock);
3124 spin_lock(&cache->lock);
3125 cache->pinned += len;
3126 cache->space_info->bytes_pinned += len;
3127 spin_unlock(&cache->lock);
3128 spin_unlock(&cache->space_info->lock);
3129 fs_info->total_pinned += len;
3131 spin_lock(&cache->space_info->lock);
3132 spin_lock(&cache->lock);
3133 cache->pinned -= len;
3134 cache->space_info->bytes_pinned -= len;
3135 spin_unlock(&cache->lock);
3136 spin_unlock(&cache->space_info->lock);
3137 fs_info->total_pinned -= len;
3138 if (block_group_cache_done(cache) && mark_free)
3139 btrfs_add_free_space(cache, bytenr, len);
3141 btrfs_put_block_group(cache);
3148 static int update_reserved_extents(struct btrfs_root *root,
3149 u64 bytenr, u64 num, int reserve)
3152 struct btrfs_block_group_cache *cache;
3153 struct btrfs_fs_info *fs_info = root->fs_info;
3156 cache = btrfs_lookup_block_group(fs_info, bytenr);
3158 len = min(num, cache->key.offset -
3159 (bytenr - cache->key.objectid));
3161 spin_lock(&cache->space_info->lock);
3162 spin_lock(&cache->lock);
3164 cache->reserved += len;
3165 cache->space_info->bytes_reserved += len;
3167 cache->reserved -= len;
3168 cache->space_info->bytes_reserved -= len;
3170 spin_unlock(&cache->lock);
3171 spin_unlock(&cache->space_info->lock);
3172 btrfs_put_block_group(cache);
3179 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
3184 bool caching_kthreads = false;
3185 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3188 if (atomic_read(&root->fs_info->async_caching_threads))
3189 caching_kthreads = true;
3192 ret = find_first_extent_bit(pinned_extents, last,
3193 &start, &end, EXTENT_DIRTY);
3198 * we need to make sure that the pinned extents don't go away
3199 * while we are caching block groups
3201 if (unlikely(caching_kthreads))
3202 set_extent_delalloc(pinned_extents, start, end,
3205 set_extent_dirty(copy, start, end, GFP_NOFS);
3211 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3212 struct btrfs_root *root,
3213 struct extent_io_tree *unpin)
3220 ret = find_first_extent_bit(&root->fs_info->pinned_extents, 0,
3221 &start, &end, EXTENT_DELALLOC);
3226 ret = find_first_extent_bit(unpin, 0, &start, &end,
3231 ret = btrfs_discard_extent(root, start, end + 1 - start);
3233 /* unlocks the pinned mutex */
3234 btrfs_update_pinned_extents(root, start, end + 1 - start, 0,
3236 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3241 if (unlikely(!mark_free))
3242 btrfs_discard_pinned_extents(root->fs_info, NULL);
3247 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3248 struct btrfs_root *root,
3249 struct btrfs_path *path,
3250 u64 bytenr, u64 num_bytes, int is_data,
3251 struct extent_buffer **must_clean)
3254 struct extent_buffer *buf;
3259 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3263 /* we can reuse a block if it hasn't been written
3264 * and it is from this transaction. We can't
3265 * reuse anything from the tree log root because
3266 * it has tiny sub-transactions.
3268 if (btrfs_buffer_uptodate(buf, 0) &&
3269 btrfs_try_tree_lock(buf)) {
3270 u64 header_owner = btrfs_header_owner(buf);
3271 u64 header_transid = btrfs_header_generation(buf);
3272 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3273 header_transid == trans->transid &&
3274 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3278 btrfs_tree_unlock(buf);
3280 free_extent_buffer(buf);
3282 btrfs_set_path_blocking(path);
3283 /* unlocks the pinned mutex */
3284 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1, 0);
3291 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3292 struct btrfs_root *root,
3293 u64 bytenr, u64 num_bytes, u64 parent,
3294 u64 root_objectid, u64 owner_objectid,
3295 u64 owner_offset, int refs_to_drop,
3296 struct btrfs_delayed_extent_op *extent_op)
3298 struct btrfs_key key;
3299 struct btrfs_path *path;
3300 struct btrfs_fs_info *info = root->fs_info;
3301 struct btrfs_root *extent_root = info->extent_root;
3302 struct extent_buffer *leaf;
3303 struct btrfs_extent_item *ei;
3304 struct btrfs_extent_inline_ref *iref;
3307 int extent_slot = 0;
3308 int found_extent = 0;
3313 path = btrfs_alloc_path();
3318 path->leave_spinning = 1;
3320 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3321 BUG_ON(!is_data && refs_to_drop != 1);
3323 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3324 bytenr, num_bytes, parent,
3325 root_objectid, owner_objectid,
3328 extent_slot = path->slots[0];
3329 while (extent_slot >= 0) {
3330 btrfs_item_key_to_cpu(path->nodes[0], &key,
3332 if (key.objectid != bytenr)
3334 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3335 key.offset == num_bytes) {
3339 if (path->slots[0] - extent_slot > 5)
3343 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3344 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3345 if (found_extent && item_size < sizeof(*ei))
3348 if (!found_extent) {
3350 ret = remove_extent_backref(trans, extent_root, path,
3354 btrfs_release_path(extent_root, path);
3355 path->leave_spinning = 1;
3357 key.objectid = bytenr;
3358 key.type = BTRFS_EXTENT_ITEM_KEY;
3359 key.offset = num_bytes;
3361 ret = btrfs_search_slot(trans, extent_root,
3364 printk(KERN_ERR "umm, got %d back from search"
3365 ", was looking for %llu\n", ret,
3366 (unsigned long long)bytenr);
3367 btrfs_print_leaf(extent_root, path->nodes[0]);
3370 extent_slot = path->slots[0];
3373 btrfs_print_leaf(extent_root, path->nodes[0]);
3375 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3376 "parent %llu root %llu owner %llu offset %llu\n",
3377 (unsigned long long)bytenr,
3378 (unsigned long long)parent,
3379 (unsigned long long)root_objectid,
3380 (unsigned long long)owner_objectid,
3381 (unsigned long long)owner_offset);
3384 leaf = path->nodes[0];
3385 item_size = btrfs_item_size_nr(leaf, extent_slot);
3386 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3387 if (item_size < sizeof(*ei)) {
3388 BUG_ON(found_extent || extent_slot != path->slots[0]);
3389 ret = convert_extent_item_v0(trans, extent_root, path,
3393 btrfs_release_path(extent_root, path);
3394 path->leave_spinning = 1;
3396 key.objectid = bytenr;
3397 key.type = BTRFS_EXTENT_ITEM_KEY;
3398 key.offset = num_bytes;
3400 ret = btrfs_search_slot(trans, extent_root, &key, path,
3403 printk(KERN_ERR "umm, got %d back from search"
3404 ", was looking for %llu\n", ret,
3405 (unsigned long long)bytenr);
3406 btrfs_print_leaf(extent_root, path->nodes[0]);
3409 extent_slot = path->slots[0];
3410 leaf = path->nodes[0];
3411 item_size = btrfs_item_size_nr(leaf, extent_slot);
3414 BUG_ON(item_size < sizeof(*ei));
3415 ei = btrfs_item_ptr(leaf, extent_slot,
3416 struct btrfs_extent_item);
3417 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3418 struct btrfs_tree_block_info *bi;
3419 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3420 bi = (struct btrfs_tree_block_info *)(ei + 1);
3421 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3424 refs = btrfs_extent_refs(leaf, ei);
3425 BUG_ON(refs < refs_to_drop);
3426 refs -= refs_to_drop;
3430 __run_delayed_extent_op(extent_op, leaf, ei);
3432 * In the case of inline back ref, reference count will
3433 * be updated by remove_extent_backref
3436 BUG_ON(!found_extent);
3438 btrfs_set_extent_refs(leaf, ei, refs);
3439 btrfs_mark_buffer_dirty(leaf);
3442 ret = remove_extent_backref(trans, extent_root, path,
3449 struct extent_buffer *must_clean = NULL;
3452 BUG_ON(is_data && refs_to_drop !=
3453 extent_data_ref_count(root, path, iref));
3455 BUG_ON(path->slots[0] != extent_slot);
3457 BUG_ON(path->slots[0] != extent_slot + 1);
3458 path->slots[0] = extent_slot;
3463 ret = pin_down_bytes(trans, root, path, bytenr,
3464 num_bytes, is_data, &must_clean);
3469 * it is going to be very rare for someone to be waiting
3470 * on the block we're freeing. del_items might need to
3471 * schedule, so rather than get fancy, just force it
3475 btrfs_set_lock_blocking(must_clean);
3477 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3480 btrfs_release_path(extent_root, path);
3483 clean_tree_block(NULL, root, must_clean);
3484 btrfs_tree_unlock(must_clean);
3485 free_extent_buffer(must_clean);
3489 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3492 invalidate_mapping_pages(info->btree_inode->i_mapping,
3493 bytenr >> PAGE_CACHE_SHIFT,
3494 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3497 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3501 btrfs_free_path(path);
3506 * when we free an extent, it is possible (and likely) that we free the last
3507 * delayed ref for that extent as well. This searches the delayed ref tree for
3508 * a given extent, and if there are no other delayed refs to be processed, it
3509 * removes it from the tree.
3511 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root, u64 bytenr)
3514 struct btrfs_delayed_ref_head *head;
3515 struct btrfs_delayed_ref_root *delayed_refs;
3516 struct btrfs_delayed_ref_node *ref;
3517 struct rb_node *node;
3520 delayed_refs = &trans->transaction->delayed_refs;
3521 spin_lock(&delayed_refs->lock);
3522 head = btrfs_find_delayed_ref_head(trans, bytenr);
3526 node = rb_prev(&head->node.rb_node);
3530 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3532 /* there are still entries for this ref, we can't drop it */
3533 if (ref->bytenr == bytenr)
3536 if (head->extent_op) {
3537 if (!head->must_insert_reserved)
3539 kfree(head->extent_op);
3540 head->extent_op = NULL;
3544 * waiting for the lock here would deadlock. If someone else has it
3545 * locked they are already in the process of dropping it anyway
3547 if (!mutex_trylock(&head->mutex))
3551 * at this point we have a head with no other entries. Go
3552 * ahead and process it.
3554 head->node.in_tree = 0;
3555 rb_erase(&head->node.rb_node, &delayed_refs->root);
3557 delayed_refs->num_entries--;
3560 * we don't take a ref on the node because we're removing it from the
3561 * tree, so we just steal the ref the tree was holding.
3563 delayed_refs->num_heads--;
3564 if (list_empty(&head->cluster))
3565 delayed_refs->num_heads_ready--;
3567 list_del_init(&head->cluster);
3568 spin_unlock(&delayed_refs->lock);
3570 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3571 &head->node, head->extent_op,
3572 head->must_insert_reserved);
3574 btrfs_put_delayed_ref(&head->node);
3577 spin_unlock(&delayed_refs->lock);
3581 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3582 struct btrfs_root *root,
3583 u64 bytenr, u64 num_bytes, u64 parent,
3584 u64 root_objectid, u64 owner, u64 offset)
3589 * tree log blocks never actually go into the extent allocation
3590 * tree, just update pinning info and exit early.
3592 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3593 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3594 /* unlocks the pinned mutex */
3595 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1, 0);
3596 update_reserved_extents(root, bytenr, num_bytes, 0);
3598 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3599 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3600 parent, root_objectid, (int)owner,
3601 BTRFS_DROP_DELAYED_REF, NULL);
3603 ret = check_ref_cleanup(trans, root, bytenr);
3606 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3607 parent, root_objectid, owner,
3608 offset, BTRFS_DROP_DELAYED_REF, NULL);
3614 static u64 stripe_align(struct btrfs_root *root, u64 val)
3616 u64 mask = ((u64)root->stripesize - 1);
3617 u64 ret = (val + mask) & ~mask;
3622 * when we wait for progress in the block group caching, its because
3623 * our allocation attempt failed at least once. So, we must sleep
3624 * and let some progress happen before we try again.
3626 * This function will sleep at least once waiting for new free space to
3627 * show up, and then it will check the block group free space numbers
3628 * for our min num_bytes. Another option is to have it go ahead
3629 * and look in the rbtree for a free extent of a given size, but this
3633 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3638 prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE);
3640 if (block_group_cache_done(cache)) {
3641 finish_wait(&cache->caching_q, &wait);
3645 finish_wait(&cache->caching_q, &wait);
3647 wait_event(cache->caching_q, block_group_cache_done(cache) ||
3648 (cache->free_space >= num_bytes));
3652 enum btrfs_loop_type {
3653 LOOP_CACHED_ONLY = 0,
3654 LOOP_CACHING_NOWAIT = 1,
3655 LOOP_CACHING_WAIT = 2,
3656 LOOP_ALLOC_CHUNK = 3,
3657 LOOP_NO_EMPTY_SIZE = 4,
3661 * walks the btree of allocated extents and find a hole of a given size.
3662 * The key ins is changed to record the hole:
3663 * ins->objectid == block start
3664 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3665 * ins->offset == number of blocks
3666 * Any available blocks before search_start are skipped.
3668 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3669 struct btrfs_root *orig_root,
3670 u64 num_bytes, u64 empty_size,
3671 u64 search_start, u64 search_end,
3672 u64 hint_byte, struct btrfs_key *ins,
3673 u64 exclude_start, u64 exclude_nr,
3677 struct btrfs_root *root = orig_root->fs_info->extent_root;
3678 struct btrfs_free_cluster *last_ptr = NULL;
3679 struct btrfs_block_group_cache *block_group = NULL;
3680 int empty_cluster = 2 * 1024 * 1024;
3681 int allowed_chunk_alloc = 0;
3682 struct btrfs_space_info *space_info;
3683 int last_ptr_loop = 0;
3685 bool found_uncached_bg = false;
3687 WARN_ON(num_bytes < root->sectorsize);
3688 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3692 space_info = __find_space_info(root->fs_info, data);
3694 if (orig_root->ref_cows || empty_size)
3695 allowed_chunk_alloc = 1;
3697 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3698 last_ptr = &root->fs_info->meta_alloc_cluster;
3699 if (!btrfs_test_opt(root, SSD))
3700 empty_cluster = 64 * 1024;
3703 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3704 last_ptr = &root->fs_info->data_alloc_cluster;
3708 spin_lock(&last_ptr->lock);
3709 if (last_ptr->block_group)
3710 hint_byte = last_ptr->window_start;
3711 spin_unlock(&last_ptr->lock);
3714 search_start = max(search_start, first_logical_byte(root, 0));
3715 search_start = max(search_start, hint_byte);
3720 if (search_start == hint_byte) {
3721 block_group = btrfs_lookup_block_group(root->fs_info,
3724 * we don't want to use the block group if it doesn't match our
3725 * allocation bits, or if its not cached.
3727 if (block_group && block_group_bits(block_group, data) &&
3728 block_group_cache_done(block_group)) {
3729 down_read(&space_info->groups_sem);
3730 if (list_empty(&block_group->list) ||
3733 * someone is removing this block group,
3734 * we can't jump into the have_block_group
3735 * target because our list pointers are not
3738 btrfs_put_block_group(block_group);
3739 up_read(&space_info->groups_sem);
3741 goto have_block_group;
3742 } else if (block_group) {
3743 btrfs_put_block_group(block_group);
3748 down_read(&space_info->groups_sem);
3749 list_for_each_entry(block_group, &space_info->block_groups, list) {
3753 atomic_inc(&block_group->count);
3754 search_start = block_group->key.objectid;
3757 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
3759 * we want to start caching kthreads, but not too many
3760 * right off the bat so we don't overwhelm the system,
3761 * so only start them if there are less than 2 and we're
3762 * in the initial allocation phase.
3764 if (loop > LOOP_CACHING_NOWAIT ||
3765 atomic_read(&space_info->caching_threads) < 2) {
3766 ret = cache_block_group(block_group);
3771 cached = block_group_cache_done(block_group);
3772 if (unlikely(!cached)) {
3773 found_uncached_bg = true;
3775 /* if we only want cached bgs, loop */
3776 if (loop == LOOP_CACHED_ONLY)
3780 if (unlikely(block_group->ro))
3785 * the refill lock keeps out other
3786 * people trying to start a new cluster
3788 spin_lock(&last_ptr->refill_lock);
3789 if (last_ptr->block_group &&
3790 (last_ptr->block_group->ro ||
3791 !block_group_bits(last_ptr->block_group, data))) {
3793 goto refill_cluster;
3796 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3797 num_bytes, search_start);
3799 /* we have a block, we're done */
3800 spin_unlock(&last_ptr->refill_lock);
3804 spin_lock(&last_ptr->lock);
3806 * whoops, this cluster doesn't actually point to
3807 * this block group. Get a ref on the block
3808 * group is does point to and try again
3810 if (!last_ptr_loop && last_ptr->block_group &&
3811 last_ptr->block_group != block_group) {
3813 btrfs_put_block_group(block_group);
3814 block_group = last_ptr->block_group;
3815 atomic_inc(&block_group->count);
3816 spin_unlock(&last_ptr->lock);
3817 spin_unlock(&last_ptr->refill_lock);
3820 search_start = block_group->key.objectid;
3822 * we know this block group is properly
3823 * in the list because
3824 * btrfs_remove_block_group, drops the
3825 * cluster before it removes the block
3826 * group from the list
3828 goto have_block_group;
3830 spin_unlock(&last_ptr->lock);
3833 * this cluster didn't work out, free it and
3836 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3840 /* allocate a cluster in this block group */
3841 ret = btrfs_find_space_cluster(trans, root,
3842 block_group, last_ptr,
3844 empty_cluster + empty_size);
3847 * now pull our allocation out of this
3850 offset = btrfs_alloc_from_cluster(block_group,
3851 last_ptr, num_bytes,
3854 /* we found one, proceed */
3855 spin_unlock(&last_ptr->refill_lock);
3858 } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
3859 spin_unlock(&last_ptr->refill_lock);
3861 wait_block_group_cache_progress(block_group,
3862 num_bytes + empty_cluster + empty_size);
3863 goto have_block_group;
3867 * at this point we either didn't find a cluster
3868 * or we weren't able to allocate a block from our
3869 * cluster. Free the cluster we've been trying
3870 * to use, and go to the next block group
3872 if (loop < LOOP_NO_EMPTY_SIZE) {
3873 btrfs_return_cluster_to_free_space(NULL,
3875 spin_unlock(&last_ptr->refill_lock);
3878 spin_unlock(&last_ptr->refill_lock);
3881 offset = btrfs_find_space_for_alloc(block_group, search_start,
3882 num_bytes, empty_size);
3883 if (!offset && (cached || (!cached &&
3884 loop == LOOP_CACHING_NOWAIT))) {
3886 } else if (!offset && (!cached &&
3887 loop > LOOP_CACHING_NOWAIT)) {
3888 wait_block_group_cache_progress(block_group,
3889 num_bytes + empty_size);
3890 goto have_block_group;
3893 search_start = stripe_align(root, offset);
3894 /* move on to the next group */
3895 if (search_start + num_bytes >= search_end) {
3896 btrfs_add_free_space(block_group, offset, num_bytes);
3900 /* move on to the next group */
3901 if (search_start + num_bytes >
3902 block_group->key.objectid + block_group->key.offset) {
3903 btrfs_add_free_space(block_group, offset, num_bytes);
3907 if (exclude_nr > 0 &&
3908 (search_start + num_bytes > exclude_start &&
3909 search_start < exclude_start + exclude_nr)) {
3910 search_start = exclude_start + exclude_nr;
3912 btrfs_add_free_space(block_group, offset, num_bytes);
3914 * if search_start is still in this block group
3915 * then we just re-search this block group
3917 if (search_start >= block_group->key.objectid &&
3918 search_start < (block_group->key.objectid +
3919 block_group->key.offset))
3920 goto have_block_group;
3924 ins->objectid = search_start;
3925 ins->offset = num_bytes;
3927 if (offset < search_start)
3928 btrfs_add_free_space(block_group, offset,
3929 search_start - offset);
3930 BUG_ON(offset > search_start);
3932 /* we are all good, lets return */
3935 btrfs_put_block_group(block_group);
3937 up_read(&space_info->groups_sem);
3939 /* LOOP_CACHED_ONLY, only search fully cached block groups
3940 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3941 * dont wait foR them to finish caching
3942 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3943 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3944 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3947 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
3948 (found_uncached_bg || empty_size || empty_cluster ||
3949 allowed_chunk_alloc)) {
3950 if (found_uncached_bg) {
3951 found_uncached_bg = false;
3952 if (loop < LOOP_CACHING_WAIT) {
3958 if (loop == LOOP_ALLOC_CHUNK) {
3963 if (allowed_chunk_alloc) {
3964 ret = do_chunk_alloc(trans, root, num_bytes +
3965 2 * 1024 * 1024, data, 1);
3966 allowed_chunk_alloc = 0;
3968 space_info->force_alloc = 1;
3971 if (loop < LOOP_NO_EMPTY_SIZE) {
3976 } else if (!ins->objectid) {
3980 /* we found what we needed */
3981 if (ins->objectid) {
3982 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3983 trans->block_group = block_group->key.objectid;
3985 btrfs_put_block_group(block_group);
3992 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3994 struct btrfs_block_group_cache *cache;
3996 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3997 (unsigned long long)(info->total_bytes - info->bytes_used -
3998 info->bytes_pinned - info->bytes_reserved),
3999 (info->full) ? "" : "not ");
4000 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4001 " may_use=%llu, used=%llu\n",
4002 (unsigned long long)info->total_bytes,
4003 (unsigned long long)info->bytes_pinned,
4004 (unsigned long long)info->bytes_delalloc,
4005 (unsigned long long)info->bytes_may_use,
4006 (unsigned long long)info->bytes_used);
4008 down_read(&info->groups_sem);
4009 list_for_each_entry(cache, &info->block_groups, list) {
4010 spin_lock(&cache->lock);
4011 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4012 "%llu pinned %llu reserved\n",
4013 (unsigned long long)cache->key.objectid,
4014 (unsigned long long)cache->key.offset,
4015 (unsigned long long)btrfs_block_group_used(&cache->item),
4016 (unsigned long long)cache->pinned,
4017 (unsigned long long)cache->reserved);
4018 btrfs_dump_free_space(cache, bytes);
4019 spin_unlock(&cache->lock);
4021 up_read(&info->groups_sem);
4024 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4025 struct btrfs_root *root,
4026 u64 num_bytes, u64 min_alloc_size,
4027 u64 empty_size, u64 hint_byte,
4028 u64 search_end, struct btrfs_key *ins,
4032 u64 search_start = 0;
4033 struct btrfs_fs_info *info = root->fs_info;
4035 data = btrfs_get_alloc_profile(root, data);
4038 * the only place that sets empty_size is btrfs_realloc_node, which
4039 * is not called recursively on allocations
4041 if (empty_size || root->ref_cows) {
4042 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4043 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4045 BTRFS_BLOCK_GROUP_METADATA |
4046 (info->metadata_alloc_profile &
4047 info->avail_metadata_alloc_bits), 0);
4049 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4050 num_bytes + 2 * 1024 * 1024, data, 0);
4053 WARN_ON(num_bytes < root->sectorsize);
4054 ret = find_free_extent(trans, root, num_bytes, empty_size,
4055 search_start, search_end, hint_byte, ins,
4056 trans->alloc_exclude_start,
4057 trans->alloc_exclude_nr, data);
4059 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4060 num_bytes = num_bytes >> 1;
4061 num_bytes = num_bytes & ~(root->sectorsize - 1);
4062 num_bytes = max(num_bytes, min_alloc_size);
4063 do_chunk_alloc(trans, root->fs_info->extent_root,
4064 num_bytes, data, 1);
4067 if (ret == -ENOSPC) {
4068 struct btrfs_space_info *sinfo;
4070 sinfo = __find_space_info(root->fs_info, data);
4071 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4072 "wanted %llu\n", (unsigned long long)data,
4073 (unsigned long long)num_bytes);
4074 dump_space_info(sinfo, num_bytes);
4080 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4082 struct btrfs_block_group_cache *cache;
4085 cache = btrfs_lookup_block_group(root->fs_info, start);
4087 printk(KERN_ERR "Unable to find block group for %llu\n",
4088 (unsigned long long)start);
4092 ret = btrfs_discard_extent(root, start, len);
4094 btrfs_add_free_space(cache, start, len);
4095 btrfs_put_block_group(cache);
4096 update_reserved_extents(root, start, len, 0);
4101 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4102 struct btrfs_root *root,
4103 u64 num_bytes, u64 min_alloc_size,
4104 u64 empty_size, u64 hint_byte,
4105 u64 search_end, struct btrfs_key *ins,
4109 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
4110 empty_size, hint_byte, search_end, ins,
4113 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4118 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4119 struct btrfs_root *root,
4120 u64 parent, u64 root_objectid,
4121 u64 flags, u64 owner, u64 offset,
4122 struct btrfs_key *ins, int ref_mod)
4125 struct btrfs_fs_info *fs_info = root->fs_info;
4126 struct btrfs_extent_item *extent_item;
4127 struct btrfs_extent_inline_ref *iref;
4128 struct btrfs_path *path;
4129 struct extent_buffer *leaf;
4134 type = BTRFS_SHARED_DATA_REF_KEY;
4136 type = BTRFS_EXTENT_DATA_REF_KEY;
4138 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4140 path = btrfs_alloc_path();
4143 path->leave_spinning = 1;
4144 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4148 leaf = path->nodes[0];
4149 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4150 struct btrfs_extent_item);
4151 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4152 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4153 btrfs_set_extent_flags(leaf, extent_item,
4154 flags | BTRFS_EXTENT_FLAG_DATA);
4156 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4157 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4159 struct btrfs_shared_data_ref *ref;
4160 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4161 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4162 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4164 struct btrfs_extent_data_ref *ref;
4165 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4166 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4167 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4168 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4169 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4172 btrfs_mark_buffer_dirty(path->nodes[0]);
4173 btrfs_free_path(path);
4175 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4178 printk(KERN_ERR "btrfs update block group failed for %llu "
4179 "%llu\n", (unsigned long long)ins->objectid,
4180 (unsigned long long)ins->offset);
4186 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4187 struct btrfs_root *root,
4188 u64 parent, u64 root_objectid,
4189 u64 flags, struct btrfs_disk_key *key,
4190 int level, struct btrfs_key *ins)
4193 struct btrfs_fs_info *fs_info = root->fs_info;
4194 struct btrfs_extent_item *extent_item;
4195 struct btrfs_tree_block_info *block_info;
4196 struct btrfs_extent_inline_ref *iref;
4197 struct btrfs_path *path;
4198 struct extent_buffer *leaf;
4199 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4201 path = btrfs_alloc_path();
4204 path->leave_spinning = 1;
4205 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4209 leaf = path->nodes[0];
4210 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4211 struct btrfs_extent_item);
4212 btrfs_set_extent_refs(leaf, extent_item, 1);
4213 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4214 btrfs_set_extent_flags(leaf, extent_item,
4215 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4216 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4218 btrfs_set_tree_block_key(leaf, block_info, key);
4219 btrfs_set_tree_block_level(leaf, block_info, level);
4221 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4223 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4224 btrfs_set_extent_inline_ref_type(leaf, iref,
4225 BTRFS_SHARED_BLOCK_REF_KEY);
4226 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4228 btrfs_set_extent_inline_ref_type(leaf, iref,
4229 BTRFS_TREE_BLOCK_REF_KEY);
4230 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4233 btrfs_mark_buffer_dirty(leaf);
4234 btrfs_free_path(path);
4236 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4239 printk(KERN_ERR "btrfs update block group failed for %llu "
4240 "%llu\n", (unsigned long long)ins->objectid,
4241 (unsigned long long)ins->offset);
4247 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4248 struct btrfs_root *root,
4249 u64 root_objectid, u64 owner,
4250 u64 offset, struct btrfs_key *ins)
4254 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4256 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4257 0, root_objectid, owner, offset,
4258 BTRFS_ADD_DELAYED_EXTENT, NULL);
4263 * this is used by the tree logging recovery code. It records that
4264 * an extent has been allocated and makes sure to clear the free
4265 * space cache bits as well
4267 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4268 struct btrfs_root *root,
4269 u64 root_objectid, u64 owner, u64 offset,
4270 struct btrfs_key *ins)
4273 struct btrfs_block_group_cache *block_group;
4275 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4276 cache_block_group(block_group);
4277 wait_event(block_group->caching_q,
4278 block_group_cache_done(block_group));
4280 ret = btrfs_remove_free_space(block_group, ins->objectid,
4283 btrfs_put_block_group(block_group);
4284 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4285 0, owner, offset, ins, 1);
4290 * finds a free extent and does all the dirty work required for allocation
4291 * returns the key for the extent through ins, and a tree buffer for
4292 * the first block of the extent through buf.
4294 * returns 0 if everything worked, non-zero otherwise.
4296 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4297 struct btrfs_root *root,
4298 u64 num_bytes, u64 parent, u64 root_objectid,
4299 struct btrfs_disk_key *key, int level,
4300 u64 empty_size, u64 hint_byte, u64 search_end,
4301 struct btrfs_key *ins)
4306 ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4307 empty_size, hint_byte, search_end,
4312 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4314 parent = ins->objectid;
4315 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4319 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4320 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4321 struct btrfs_delayed_extent_op *extent_op;
4322 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4325 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4327 memset(&extent_op->key, 0, sizeof(extent_op->key));
4328 extent_op->flags_to_set = flags;
4329 extent_op->update_key = 1;
4330 extent_op->update_flags = 1;
4331 extent_op->is_data = 0;
4333 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4334 ins->offset, parent, root_objectid,
4335 level, BTRFS_ADD_DELAYED_EXTENT,
4342 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4343 struct btrfs_root *root,
4344 u64 bytenr, u32 blocksize,
4347 struct extent_buffer *buf;
4349 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4351 return ERR_PTR(-ENOMEM);
4352 btrfs_set_header_generation(buf, trans->transid);
4353 btrfs_set_buffer_lockdep_class(buf, level);
4354 btrfs_tree_lock(buf);
4355 clean_tree_block(trans, root, buf);
4357 btrfs_set_lock_blocking(buf);
4358 btrfs_set_buffer_uptodate(buf);
4360 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4361 set_extent_dirty(&root->dirty_log_pages, buf->start,
4362 buf->start + buf->len - 1, GFP_NOFS);
4364 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4365 buf->start + buf->len - 1, GFP_NOFS);
4367 trans->blocks_used++;
4368 /* this returns a buffer locked for blocking */
4373 * helper function to allocate a block for a given tree
4374 * returns the tree buffer or NULL.
4376 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4377 struct btrfs_root *root, u32 blocksize,
4378 u64 parent, u64 root_objectid,
4379 struct btrfs_disk_key *key, int level,
4380 u64 hint, u64 empty_size)
4382 struct btrfs_key ins;
4384 struct extent_buffer *buf;
4386 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4387 key, level, empty_size, hint, (u64)-1, &ins);
4390 return ERR_PTR(ret);
4393 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4399 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4400 struct btrfs_root *root, struct extent_buffer *leaf)
4404 struct btrfs_key key;
4405 struct btrfs_file_extent_item *fi;
4410 BUG_ON(!btrfs_is_leaf(leaf));
4411 nritems = btrfs_header_nritems(leaf);
4413 for (i = 0; i < nritems; i++) {
4415 btrfs_item_key_to_cpu(leaf, &key, i);
4417 /* only extents have references, skip everything else */
4418 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4421 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4423 /* inline extents live in the btree, they don't have refs */
4424 if (btrfs_file_extent_type(leaf, fi) ==
4425 BTRFS_FILE_EXTENT_INLINE)
4428 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4430 /* holes don't have refs */
4431 if (disk_bytenr == 0)
4434 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4435 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4436 leaf->start, 0, key.objectid, 0);
4442 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4443 struct btrfs_root *root,
4444 struct btrfs_leaf_ref *ref)
4448 struct btrfs_extent_info *info;
4449 struct refsort *sorted;
4451 if (ref->nritems == 0)
4454 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4455 for (i = 0; i < ref->nritems; i++) {
4456 sorted[i].bytenr = ref->extents[i].bytenr;
4459 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4462 * the items in the ref were sorted when the ref was inserted
4463 * into the ref cache, so this is already in order
4465 for (i = 0; i < ref->nritems; i++) {
4466 info = ref->extents + sorted[i].slot;
4467 ret = btrfs_free_extent(trans, root, info->bytenr,
4468 info->num_bytes, ref->bytenr,
4469 ref->owner, ref->generation,
4472 atomic_inc(&root->fs_info->throttle_gen);
4473 wake_up(&root->fs_info->transaction_throttle);
4485 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *root, u64 start,
4491 ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4494 #if 0 /* some debugging code in case we see problems here */
4495 /* if the refs count is one, it won't get increased again. But
4496 * if the ref count is > 1, someone may be decreasing it at
4497 * the same time we are.
4500 struct extent_buffer *eb = NULL;
4501 eb = btrfs_find_create_tree_block(root, start, len);
4503 btrfs_tree_lock(eb);
4505 mutex_lock(&root->fs_info->alloc_mutex);
4506 ret = lookup_extent_ref(NULL, root, start, len, refs);
4508 mutex_unlock(&root->fs_info->alloc_mutex);
4511 btrfs_tree_unlock(eb);
4512 free_extent_buffer(eb);
4515 printk(KERN_ERR "btrfs block %llu went down to one "
4516 "during drop_snap\n", (unsigned long long)start);
4528 * this is used while deleting old snapshots, and it drops the refs
4529 * on a whole subtree starting from a level 1 node.
4531 * The idea is to sort all the leaf pointers, and then drop the
4532 * ref on all the leaves in order. Most of the time the leaves
4533 * will have ref cache entries, so no leaf IOs will be required to
4534 * find the extents they have references on.
4536 * For each leaf, any references it has are also dropped in order
4538 * This ends up dropping the references in something close to optimal
4539 * order for reading and modifying the extent allocation tree.
4541 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4542 struct btrfs_root *root,
4543 struct btrfs_path *path)
4548 struct extent_buffer *eb = path->nodes[1];
4549 struct extent_buffer *leaf;
4550 struct btrfs_leaf_ref *ref;
4551 struct refsort *sorted = NULL;
4552 int nritems = btrfs_header_nritems(eb);
4556 int slot = path->slots[1];
4557 u32 blocksize = btrfs_level_size(root, 0);
4563 root_owner = btrfs_header_owner(eb);
4564 root_gen = btrfs_header_generation(eb);
4565 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4568 * step one, sort all the leaf pointers so we don't scribble
4569 * randomly into the extent allocation tree
4571 for (i = slot; i < nritems; i++) {
4572 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4573 sorted[refi].slot = i;
4578 * nritems won't be zero, but if we're picking up drop_snapshot
4579 * after a crash, slot might be > 0, so double check things
4585 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4588 * the first loop frees everything the leaves point to
4590 for (i = 0; i < refi; i++) {
4593 bytenr = sorted[i].bytenr;
4596 * check the reference count on this leaf. If it is > 1
4597 * we just decrement it below and don't update any
4598 * of the refs the leaf points to.
4600 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4606 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4609 * the leaf only had one reference, which means the
4610 * only thing pointing to this leaf is the snapshot
4611 * we're deleting. It isn't possible for the reference
4612 * count to increase again later
4614 * The reference cache is checked for the leaf,
4615 * and if found we'll be able to drop any refs held by
4616 * the leaf without needing to read it in.
4618 ref = btrfs_lookup_leaf_ref(root, bytenr);
4619 if (ref && ref->generation != ptr_gen) {
4620 btrfs_free_leaf_ref(root, ref);
4624 ret = cache_drop_leaf_ref(trans, root, ref);
4626 btrfs_remove_leaf_ref(root, ref);
4627 btrfs_free_leaf_ref(root, ref);
4630 * the leaf wasn't in the reference cache, so
4631 * we have to read it.
4633 leaf = read_tree_block(root, bytenr, blocksize,
4635 ret = btrfs_drop_leaf_ref(trans, root, leaf);
4637 free_extent_buffer(leaf);
4639 atomic_inc(&root->fs_info->throttle_gen);
4640 wake_up(&root->fs_info->transaction_throttle);
4645 * run through the loop again to free the refs on the leaves.
4646 * This is faster than doing it in the loop above because
4647 * the leaves are likely to be clustered together. We end up
4648 * working in nice chunks on the extent allocation tree.
4650 for (i = 0; i < refi; i++) {
4651 bytenr = sorted[i].bytenr;
4652 ret = btrfs_free_extent(trans, root, bytenr,
4653 blocksize, eb->start,
4654 root_owner, root_gen, 0, 1);
4657 atomic_inc(&root->fs_info->throttle_gen);
4658 wake_up(&root->fs_info->transaction_throttle);
4665 * update the path to show we've processed the entire level 1
4666 * node. This will get saved into the root's drop_snapshot_progress
4667 * field so these drops are not repeated again if this transaction
4670 path->slots[1] = nritems;
4675 * helper function for drop_snapshot, this walks down the tree dropping ref
4676 * counts as it goes.
4678 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4679 struct btrfs_root *root,
4680 struct btrfs_path *path, int *level)
4686 struct extent_buffer *next;
4687 struct extent_buffer *cur;
4688 struct extent_buffer *parent;
4693 WARN_ON(*level < 0);
4694 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4695 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4696 path->nodes[*level]->len, &refs);
4702 * walk down to the last node level and free all the leaves
4704 while (*level >= 0) {
4705 WARN_ON(*level < 0);
4706 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4707 cur = path->nodes[*level];
4709 if (btrfs_header_level(cur) != *level)
4712 if (path->slots[*level] >=
4713 btrfs_header_nritems(cur))
4716 /* the new code goes down to level 1 and does all the
4717 * leaves pointed to that node in bulk. So, this check
4718 * for level 0 will always be false.
4720 * But, the disk format allows the drop_snapshot_progress
4721 * field in the root to leave things in a state where
4722 * a leaf will need cleaning up here. If someone crashes
4723 * with the old code and then boots with the new code,
4724 * we might find a leaf here.
4727 ret = btrfs_drop_leaf_ref(trans, root, cur);
4733 * once we get to level one, process the whole node
4734 * at once, including everything below it.
4737 ret = drop_level_one_refs(trans, root, path);
4742 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4743 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4744 blocksize = btrfs_level_size(root, *level - 1);
4746 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4751 * if there is more than one reference, we don't need
4752 * to read that node to drop any references it has. We
4753 * just drop the ref we hold on that node and move on to the
4754 * next slot in this level.
4757 parent = path->nodes[*level];
4758 root_owner = btrfs_header_owner(parent);
4759 root_gen = btrfs_header_generation(parent);
4760 path->slots[*level]++;
4762 ret = btrfs_free_extent(trans, root, bytenr,
4763 blocksize, parent->start,
4764 root_owner, root_gen,
4768 atomic_inc(&root->fs_info->throttle_gen);
4769 wake_up(&root->fs_info->transaction_throttle);
4776 * we need to keep freeing things in the next level down.
4777 * read the block and loop around to process it
4779 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4780 WARN_ON(*level <= 0);
4781 if (path->nodes[*level-1])
4782 free_extent_buffer(path->nodes[*level-1]);
4783 path->nodes[*level-1] = next;
4784 *level = btrfs_header_level(next);
4785 path->slots[*level] = 0;
4789 WARN_ON(*level < 0);
4790 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4792 if (path->nodes[*level] == root->node) {
4793 parent = path->nodes[*level];
4794 bytenr = path->nodes[*level]->start;
4796 parent = path->nodes[*level + 1];
4797 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4800 blocksize = btrfs_level_size(root, *level);
4801 root_owner = btrfs_header_owner(parent);
4802 root_gen = btrfs_header_generation(parent);
4805 * cleanup and free the reference on the last node
4808 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4809 parent->start, root_owner, root_gen,
4811 free_extent_buffer(path->nodes[*level]);
4812 path->nodes[*level] = NULL;
4822 struct walk_control {
4823 u64 refs[BTRFS_MAX_LEVEL];
4824 u64 flags[BTRFS_MAX_LEVEL];
4825 struct btrfs_key update_progress;
4833 #define DROP_REFERENCE 1
4834 #define UPDATE_BACKREF 2
4837 * hepler to process tree block while walking down the tree.
4839 * when wc->stage == DROP_REFERENCE, this function checks
4840 * reference count of the block. if the block is shared and
4841 * we need update back refs for the subtree rooted at the
4842 * block, this function changes wc->stage to UPDATE_BACKREF
4844 * when wc->stage == UPDATE_BACKREF, this function updates
4845 * back refs for pointers in the block.
4847 * NOTE: return value 1 means we should stop walking down.
4849 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4850 struct btrfs_root *root,
4851 struct btrfs_path *path,
4852 struct walk_control *wc)
4854 int level = wc->level;
4855 struct extent_buffer *eb = path->nodes[level];
4856 struct btrfs_key key;
4857 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4860 if (wc->stage == UPDATE_BACKREF &&
4861 btrfs_header_owner(eb) != root->root_key.objectid)
4865 * when reference count of tree block is 1, it won't increase
4866 * again. once full backref flag is set, we never clear it.
4868 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4869 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4870 BUG_ON(!path->locks[level]);
4871 ret = btrfs_lookup_extent_info(trans, root,
4876 BUG_ON(wc->refs[level] == 0);
4879 if (wc->stage == DROP_REFERENCE &&
4880 wc->update_ref && wc->refs[level] > 1) {
4881 BUG_ON(eb == root->node);
4882 BUG_ON(path->slots[level] > 0);
4884 btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4886 btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4887 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4888 btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4889 wc->stage = UPDATE_BACKREF;
4890 wc->shared_level = level;
4894 if (wc->stage == DROP_REFERENCE) {
4895 if (wc->refs[level] > 1)
4898 if (path->locks[level] && !wc->keep_locks) {
4899 btrfs_tree_unlock(eb);
4900 path->locks[level] = 0;
4905 /* wc->stage == UPDATE_BACKREF */
4906 if (!(wc->flags[level] & flag)) {
4907 BUG_ON(!path->locks[level]);
4908 ret = btrfs_inc_ref(trans, root, eb, 1);
4910 ret = btrfs_dec_ref(trans, root, eb, 0);
4912 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4915 wc->flags[level] |= flag;
4919 * the block is shared by multiple trees, so it's not good to
4920 * keep the tree lock
4922 if (path->locks[level] && level > 0) {
4923 btrfs_tree_unlock(eb);
4924 path->locks[level] = 0;
4930 * hepler to process tree block while walking up the tree.
4932 * when wc->stage == DROP_REFERENCE, this function drops
4933 * reference count on the block.
4935 * when wc->stage == UPDATE_BACKREF, this function changes
4936 * wc->stage back to DROP_REFERENCE if we changed wc->stage
4937 * to UPDATE_BACKREF previously while processing the block.
4939 * NOTE: return value 1 means we should stop walking up.
4941 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4942 struct btrfs_root *root,
4943 struct btrfs_path *path,
4944 struct walk_control *wc)
4947 int level = wc->level;
4948 struct extent_buffer *eb = path->nodes[level];
4951 if (wc->stage == UPDATE_BACKREF) {
4952 BUG_ON(wc->shared_level < level);
4953 if (level < wc->shared_level)
4956 BUG_ON(wc->refs[level] <= 1);
4957 ret = find_next_key(path, level + 1, &wc->update_progress);
4961 wc->stage = DROP_REFERENCE;
4962 wc->shared_level = -1;
4963 path->slots[level] = 0;
4966 * check reference count again if the block isn't locked.
4967 * we should start walking down the tree again if reference
4970 if (!path->locks[level]) {
4972 btrfs_tree_lock(eb);
4973 btrfs_set_lock_blocking(eb);
4974 path->locks[level] = 1;
4976 ret = btrfs_lookup_extent_info(trans, root,
4981 BUG_ON(wc->refs[level] == 0);
4982 if (wc->refs[level] == 1) {
4983 btrfs_tree_unlock(eb);
4984 path->locks[level] = 0;
4992 /* wc->stage == DROP_REFERENCE */
4993 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4995 if (wc->refs[level] == 1) {
4997 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4998 ret = btrfs_dec_ref(trans, root, eb, 1);
5000 ret = btrfs_dec_ref(trans, root, eb, 0);
5003 /* make block locked assertion in clean_tree_block happy */
5004 if (!path->locks[level] &&
5005 btrfs_header_generation(eb) == trans->transid) {
5006 btrfs_tree_lock(eb);
5007 btrfs_set_lock_blocking(eb);
5008 path->locks[level] = 1;
5010 clean_tree_block(trans, root, eb);
5013 if (eb == root->node) {
5014 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5017 BUG_ON(root->root_key.objectid !=
5018 btrfs_header_owner(eb));
5020 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5021 parent = path->nodes[level + 1]->start;
5023 BUG_ON(root->root_key.objectid !=
5024 btrfs_header_owner(path->nodes[level + 1]));
5027 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5028 root->root_key.objectid, level, 0);
5031 wc->refs[level] = 0;
5032 wc->flags[level] = 0;
5036 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5037 struct btrfs_root *root,
5038 struct btrfs_path *path,
5039 struct walk_control *wc)
5041 struct extent_buffer *next;
5042 struct extent_buffer *cur;
5046 int level = wc->level;
5049 while (level >= 0) {
5050 cur = path->nodes[level];
5051 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
5053 ret = walk_down_proc(trans, root, path, wc);
5060 bytenr = btrfs_node_blockptr(cur, path->slots[level]);
5061 blocksize = btrfs_level_size(root, level - 1);
5062 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
5064 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
5065 btrfs_tree_lock(next);
5066 btrfs_set_lock_blocking(next);
5069 BUG_ON(level != btrfs_header_level(next));
5070 path->nodes[level] = next;
5071 path->slots[level] = 0;
5072 path->locks[level] = 1;
5078 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5079 struct btrfs_root *root,
5080 struct btrfs_path *path,
5081 struct walk_control *wc, int max_level)
5083 int level = wc->level;
5086 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5087 while (level < max_level && path->nodes[level]) {
5089 if (path->slots[level] + 1 <
5090 btrfs_header_nritems(path->nodes[level])) {
5091 path->slots[level]++;
5094 ret = walk_up_proc(trans, root, path, wc);
5098 if (path->locks[level]) {
5099 btrfs_tree_unlock(path->nodes[level]);
5100 path->locks[level] = 0;
5102 free_extent_buffer(path->nodes[level]);
5103 path->nodes[level] = NULL;
5111 * drop a subvolume tree.
5113 * this function traverses the tree freeing any blocks that only
5114 * referenced by the tree.
5116 * when a shared tree block is found. this function decreases its
5117 * reference count by one. if update_ref is true, this function
5118 * also make sure backrefs for the shared block and all lower level
5119 * blocks are properly updated.
5121 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5123 struct btrfs_path *path;
5124 struct btrfs_trans_handle *trans;
5125 struct btrfs_root *tree_root = root->fs_info->tree_root;
5126 struct btrfs_root_item *root_item = &root->root_item;
5127 struct walk_control *wc;
5128 struct btrfs_key key;
5133 path = btrfs_alloc_path();
5136 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5139 trans = btrfs_start_transaction(tree_root, 1);
5141 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5142 level = btrfs_header_level(root->node);
5143 path->nodes[level] = btrfs_lock_root_node(root);
5144 btrfs_set_lock_blocking(path->nodes[level]);
5145 path->slots[level] = 0;
5146 path->locks[level] = 1;
5147 memset(&wc->update_progress, 0,
5148 sizeof(wc->update_progress));
5150 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5151 memcpy(&wc->update_progress, &key,
5152 sizeof(wc->update_progress));
5154 level = root_item->drop_level;
5156 path->lowest_level = level;
5157 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5158 path->lowest_level = 0;
5163 btrfs_node_key_to_cpu(path->nodes[level], &key,
5164 path->slots[level]);
5165 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
5168 * unlock our path, this is safe because only this
5169 * function is allowed to delete this snapshot
5171 btrfs_unlock_up_safe(path, 0);
5173 level = btrfs_header_level(root->node);
5175 btrfs_tree_lock(path->nodes[level]);
5176 btrfs_set_lock_blocking(path->nodes[level]);
5178 ret = btrfs_lookup_extent_info(trans, root,
5179 path->nodes[level]->start,
5180 path->nodes[level]->len,
5184 BUG_ON(wc->refs[level] == 0);
5186 if (level == root_item->drop_level)
5189 btrfs_tree_unlock(path->nodes[level]);
5190 WARN_ON(wc->refs[level] != 1);
5196 wc->shared_level = -1;
5197 wc->stage = DROP_REFERENCE;
5198 wc->update_ref = update_ref;
5202 ret = walk_down_tree(trans, root, path, wc);
5208 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5215 BUG_ON(wc->stage != DROP_REFERENCE);
5219 if (wc->stage == DROP_REFERENCE) {
5221 btrfs_node_key(path->nodes[level],
5222 &root_item->drop_progress,
5223 path->slots[level]);
5224 root_item->drop_level = level;
5227 BUG_ON(wc->level == 0);
5228 if (trans->transaction->in_commit ||
5229 trans->transaction->delayed_refs.flushing) {
5230 ret = btrfs_update_root(trans, tree_root,
5235 btrfs_end_transaction(trans, tree_root);
5236 trans = btrfs_start_transaction(tree_root, 1);
5238 unsigned long update;
5239 update = trans->delayed_ref_updates;
5240 trans->delayed_ref_updates = 0;
5242 btrfs_run_delayed_refs(trans, tree_root,
5246 btrfs_release_path(root, path);
5249 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5252 free_extent_buffer(root->node);
5253 free_extent_buffer(root->commit_root);
5256 btrfs_end_transaction(trans, tree_root);
5258 btrfs_free_path(path);
5263 * drop subtree rooted at tree block 'node'.
5265 * NOTE: this function will unlock and release tree block 'node'
5267 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5268 struct btrfs_root *root,
5269 struct extent_buffer *node,
5270 struct extent_buffer *parent)
5272 struct btrfs_path *path;
5273 struct walk_control *wc;
5279 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5281 path = btrfs_alloc_path();
5284 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5287 btrfs_assert_tree_locked(parent);
5288 parent_level = btrfs_header_level(parent);
5289 extent_buffer_get(parent);
5290 path->nodes[parent_level] = parent;
5291 path->slots[parent_level] = btrfs_header_nritems(parent);
5293 btrfs_assert_tree_locked(node);
5294 level = btrfs_header_level(node);
5295 path->nodes[level] = node;
5296 path->slots[level] = 0;
5297 path->locks[level] = 1;
5299 wc->refs[parent_level] = 1;
5300 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5302 wc->shared_level = -1;
5303 wc->stage = DROP_REFERENCE;
5308 wret = walk_down_tree(trans, root, path, wc);
5314 wret = walk_up_tree(trans, root, path, wc, parent_level);
5322 btrfs_free_path(path);
5327 static unsigned long calc_ra(unsigned long start, unsigned long last,
5330 return min(last, start + nr - 1);
5333 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5338 unsigned long first_index;
5339 unsigned long last_index;
5342 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5343 struct file_ra_state *ra;
5344 struct btrfs_ordered_extent *ordered;
5345 unsigned int total_read = 0;
5346 unsigned int total_dirty = 0;
5349 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5351 mutex_lock(&inode->i_mutex);
5352 first_index = start >> PAGE_CACHE_SHIFT;
5353 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5355 /* make sure the dirty trick played by the caller work */
5356 ret = invalidate_inode_pages2_range(inode->i_mapping,
5357 first_index, last_index);
5361 file_ra_state_init(ra, inode->i_mapping);
5363 for (i = first_index ; i <= last_index; i++) {
5364 if (total_read % ra->ra_pages == 0) {
5365 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5366 calc_ra(i, last_index, ra->ra_pages));
5370 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5372 page = grab_cache_page(inode->i_mapping, i);
5377 if (!PageUptodate(page)) {
5378 btrfs_readpage(NULL, page);
5380 if (!PageUptodate(page)) {
5382 page_cache_release(page);
5387 wait_on_page_writeback(page);
5389 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5390 page_end = page_start + PAGE_CACHE_SIZE - 1;
5391 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5393 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5395 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5397 page_cache_release(page);
5398 btrfs_start_ordered_extent(inode, ordered, 1);
5399 btrfs_put_ordered_extent(ordered);
5402 set_page_extent_mapped(page);
5404 if (i == first_index)
5405 set_extent_bits(io_tree, page_start, page_end,
5406 EXTENT_BOUNDARY, GFP_NOFS);
5407 btrfs_set_extent_delalloc(inode, page_start, page_end);
5409 set_page_dirty(page);
5412 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5414 page_cache_release(page);
5419 mutex_unlock(&inode->i_mutex);
5420 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5424 static noinline int relocate_data_extent(struct inode *reloc_inode,
5425 struct btrfs_key *extent_key,
5428 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5429 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5430 struct extent_map *em;
5431 u64 start = extent_key->objectid - offset;
5432 u64 end = start + extent_key->offset - 1;
5434 em = alloc_extent_map(GFP_NOFS);
5435 BUG_ON(!em || IS_ERR(em));
5438 em->len = extent_key->offset;
5439 em->block_len = extent_key->offset;
5440 em->block_start = extent_key->objectid;
5441 em->bdev = root->fs_info->fs_devices->latest_bdev;
5442 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5444 /* setup extent map to cheat btrfs_readpage */
5445 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5448 spin_lock(&em_tree->lock);
5449 ret = add_extent_mapping(em_tree, em);
5450 spin_unlock(&em_tree->lock);
5451 if (ret != -EEXIST) {
5452 free_extent_map(em);
5455 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5457 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5459 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5462 struct btrfs_ref_path {
5464 u64 nodes[BTRFS_MAX_LEVEL];
5466 u64 root_generation;
5473 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5474 u64 new_nodes[BTRFS_MAX_LEVEL];
5477 struct disk_extent {
5488 static int is_cowonly_root(u64 root_objectid)
5490 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5491 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5492 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5493 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5494 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5495 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5500 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5501 struct btrfs_root *extent_root,
5502 struct btrfs_ref_path *ref_path,
5505 struct extent_buffer *leaf;
5506 struct btrfs_path *path;
5507 struct btrfs_extent_ref *ref;
5508 struct btrfs_key key;
5509 struct btrfs_key found_key;
5515 path = btrfs_alloc_path();
5520 ref_path->lowest_level = -1;
5521 ref_path->current_level = -1;
5522 ref_path->shared_level = -1;
5526 level = ref_path->current_level - 1;
5527 while (level >= -1) {
5529 if (level < ref_path->lowest_level)
5533 bytenr = ref_path->nodes[level];
5535 bytenr = ref_path->extent_start;
5536 BUG_ON(bytenr == 0);
5538 parent = ref_path->nodes[level + 1];
5539 ref_path->nodes[level + 1] = 0;
5540 ref_path->current_level = level;
5541 BUG_ON(parent == 0);
5543 key.objectid = bytenr;
5544 key.offset = parent + 1;
5545 key.type = BTRFS_EXTENT_REF_KEY;
5547 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5552 leaf = path->nodes[0];
5553 nritems = btrfs_header_nritems(leaf);
5554 if (path->slots[0] >= nritems) {
5555 ret = btrfs_next_leaf(extent_root, path);
5560 leaf = path->nodes[0];
5563 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5564 if (found_key.objectid == bytenr &&
5565 found_key.type == BTRFS_EXTENT_REF_KEY) {
5566 if (level < ref_path->shared_level)
5567 ref_path->shared_level = level;
5572 btrfs_release_path(extent_root, path);
5575 /* reached lowest level */
5579 level = ref_path->current_level;
5580 while (level < BTRFS_MAX_LEVEL - 1) {
5584 bytenr = ref_path->nodes[level];
5586 bytenr = ref_path->extent_start;
5588 BUG_ON(bytenr == 0);
5590 key.objectid = bytenr;
5592 key.type = BTRFS_EXTENT_REF_KEY;
5594 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5598 leaf = path->nodes[0];
5599 nritems = btrfs_header_nritems(leaf);
5600 if (path->slots[0] >= nritems) {
5601 ret = btrfs_next_leaf(extent_root, path);
5605 /* the extent was freed by someone */
5606 if (ref_path->lowest_level == level)
5608 btrfs_release_path(extent_root, path);
5611 leaf = path->nodes[0];
5614 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5615 if (found_key.objectid != bytenr ||
5616 found_key.type != BTRFS_EXTENT_REF_KEY) {
5617 /* the extent was freed by someone */
5618 if (ref_path->lowest_level == level) {
5622 btrfs_release_path(extent_root, path);
5626 ref = btrfs_item_ptr(leaf, path->slots[0],
5627 struct btrfs_extent_ref);
5628 ref_objectid = btrfs_ref_objectid(leaf, ref);
5629 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5631 level = (int)ref_objectid;
5632 BUG_ON(level >= BTRFS_MAX_LEVEL);
5633 ref_path->lowest_level = level;
5634 ref_path->current_level = level;
5635 ref_path->nodes[level] = bytenr;
5637 WARN_ON(ref_objectid != level);
5640 WARN_ON(level != -1);
5644 if (ref_path->lowest_level == level) {
5645 ref_path->owner_objectid = ref_objectid;
5646 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5650 * the block is tree root or the block isn't in reference
5653 if (found_key.objectid == found_key.offset ||
5654 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5655 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5656 ref_path->root_generation =
5657 btrfs_ref_generation(leaf, ref);
5659 /* special reference from the tree log */
5660 ref_path->nodes[0] = found_key.offset;
5661 ref_path->current_level = 0;
5668 BUG_ON(ref_path->nodes[level] != 0);
5669 ref_path->nodes[level] = found_key.offset;
5670 ref_path->current_level = level;
5673 * the reference was created in the running transaction,
5674 * no need to continue walking up.
5676 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5677 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5678 ref_path->root_generation =
5679 btrfs_ref_generation(leaf, ref);
5684 btrfs_release_path(extent_root, path);
5687 /* reached max tree level, but no tree root found. */
5690 btrfs_free_path(path);
5694 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5695 struct btrfs_root *extent_root,
5696 struct btrfs_ref_path *ref_path,
5699 memset(ref_path, 0, sizeof(*ref_path));
5700 ref_path->extent_start = extent_start;
5702 return __next_ref_path(trans, extent_root, ref_path, 1);
5705 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5706 struct btrfs_root *extent_root,
5707 struct btrfs_ref_path *ref_path)
5709 return __next_ref_path(trans, extent_root, ref_path, 0);
5712 static noinline int get_new_locations(struct inode *reloc_inode,
5713 struct btrfs_key *extent_key,
5714 u64 offset, int no_fragment,
5715 struct disk_extent **extents,
5718 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5719 struct btrfs_path *path;
5720 struct btrfs_file_extent_item *fi;
5721 struct extent_buffer *leaf;
5722 struct disk_extent *exts = *extents;
5723 struct btrfs_key found_key;
5728 int max = *nr_extents;
5731 WARN_ON(!no_fragment && *extents);
5734 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5739 path = btrfs_alloc_path();
5742 cur_pos = extent_key->objectid - offset;
5743 last_byte = extent_key->objectid + extent_key->offset;
5744 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5754 leaf = path->nodes[0];
5755 nritems = btrfs_header_nritems(leaf);
5756 if (path->slots[0] >= nritems) {
5757 ret = btrfs_next_leaf(root, path);
5762 leaf = path->nodes[0];
5765 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5766 if (found_key.offset != cur_pos ||
5767 found_key.type != BTRFS_EXTENT_DATA_KEY ||
5768 found_key.objectid != reloc_inode->i_ino)
5771 fi = btrfs_item_ptr(leaf, path->slots[0],
5772 struct btrfs_file_extent_item);
5773 if (btrfs_file_extent_type(leaf, fi) !=
5774 BTRFS_FILE_EXTENT_REG ||
5775 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5779 struct disk_extent *old = exts;
5781 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5782 memcpy(exts, old, sizeof(*exts) * nr);
5783 if (old != *extents)
5787 exts[nr].disk_bytenr =
5788 btrfs_file_extent_disk_bytenr(leaf, fi);
5789 exts[nr].disk_num_bytes =
5790 btrfs_file_extent_disk_num_bytes(leaf, fi);
5791 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5792 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5793 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5794 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5795 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5796 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5798 BUG_ON(exts[nr].offset > 0);
5799 BUG_ON(exts[nr].compression || exts[nr].encryption);
5800 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5802 cur_pos += exts[nr].num_bytes;
5805 if (cur_pos + offset >= last_byte)
5815 BUG_ON(cur_pos + offset > last_byte);
5816 if (cur_pos + offset < last_byte) {
5822 btrfs_free_path(path);
5824 if (exts != *extents)
5833 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5834 struct btrfs_root *root,
5835 struct btrfs_path *path,
5836 struct btrfs_key *extent_key,
5837 struct btrfs_key *leaf_key,
5838 struct btrfs_ref_path *ref_path,
5839 struct disk_extent *new_extents,
5842 struct extent_buffer *leaf;
5843 struct btrfs_file_extent_item *fi;
5844 struct inode *inode = NULL;
5845 struct btrfs_key key;
5850 u64 search_end = (u64)-1;
5853 int extent_locked = 0;
5857 memcpy(&key, leaf_key, sizeof(key));
5858 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5859 if (key.objectid < ref_path->owner_objectid ||
5860 (key.objectid == ref_path->owner_objectid &&
5861 key.type < BTRFS_EXTENT_DATA_KEY)) {
5862 key.objectid = ref_path->owner_objectid;
5863 key.type = BTRFS_EXTENT_DATA_KEY;
5869 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5873 leaf = path->nodes[0];
5874 nritems = btrfs_header_nritems(leaf);
5876 if (extent_locked && ret > 0) {
5878 * the file extent item was modified by someone
5879 * before the extent got locked.
5881 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5882 lock_end, GFP_NOFS);
5886 if (path->slots[0] >= nritems) {
5887 if (++nr_scaned > 2)
5890 BUG_ON(extent_locked);
5891 ret = btrfs_next_leaf(root, path);
5896 leaf = path->nodes[0];
5897 nritems = btrfs_header_nritems(leaf);
5900 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5902 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5903 if ((key.objectid > ref_path->owner_objectid) ||
5904 (key.objectid == ref_path->owner_objectid &&
5905 key.type > BTRFS_EXTENT_DATA_KEY) ||
5906 key.offset >= search_end)
5910 if (inode && key.objectid != inode->i_ino) {
5911 BUG_ON(extent_locked);
5912 btrfs_release_path(root, path);
5913 mutex_unlock(&inode->i_mutex);
5919 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5924 fi = btrfs_item_ptr(leaf, path->slots[0],
5925 struct btrfs_file_extent_item);
5926 extent_type = btrfs_file_extent_type(leaf, fi);
5927 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5928 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5929 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5930 extent_key->objectid)) {
5936 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5937 ext_offset = btrfs_file_extent_offset(leaf, fi);
5939 if (search_end == (u64)-1) {
5940 search_end = key.offset - ext_offset +
5941 btrfs_file_extent_ram_bytes(leaf, fi);
5944 if (!extent_locked) {
5945 lock_start = key.offset;
5946 lock_end = lock_start + num_bytes - 1;
5948 if (lock_start > key.offset ||
5949 lock_end + 1 < key.offset + num_bytes) {
5950 unlock_extent(&BTRFS_I(inode)->io_tree,
5951 lock_start, lock_end, GFP_NOFS);
5957 btrfs_release_path(root, path);
5959 inode = btrfs_iget_locked(root->fs_info->sb,
5960 key.objectid, root);
5961 if (inode->i_state & I_NEW) {
5962 BTRFS_I(inode)->root = root;
5963 BTRFS_I(inode)->location.objectid =
5965 BTRFS_I(inode)->location.type =
5966 BTRFS_INODE_ITEM_KEY;
5967 BTRFS_I(inode)->location.offset = 0;
5968 btrfs_read_locked_inode(inode);
5969 unlock_new_inode(inode);
5972 * some code call btrfs_commit_transaction while
5973 * holding the i_mutex, so we can't use mutex_lock
5976 if (is_bad_inode(inode) ||
5977 !mutex_trylock(&inode->i_mutex)) {
5980 key.offset = (u64)-1;
5985 if (!extent_locked) {
5986 struct btrfs_ordered_extent *ordered;
5988 btrfs_release_path(root, path);
5990 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5991 lock_end, GFP_NOFS);
5992 ordered = btrfs_lookup_first_ordered_extent(inode,
5995 ordered->file_offset <= lock_end &&
5996 ordered->file_offset + ordered->len > lock_start) {
5997 unlock_extent(&BTRFS_I(inode)->io_tree,
5998 lock_start, lock_end, GFP_NOFS);
5999 btrfs_start_ordered_extent(inode, ordered, 1);
6000 btrfs_put_ordered_extent(ordered);
6001 key.offset += num_bytes;
6005 btrfs_put_ordered_extent(ordered);
6011 if (nr_extents == 1) {
6012 /* update extent pointer in place */
6013 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6014 new_extents[0].disk_bytenr);
6015 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6016 new_extents[0].disk_num_bytes);
6017 btrfs_mark_buffer_dirty(leaf);
6019 btrfs_drop_extent_cache(inode, key.offset,
6020 key.offset + num_bytes - 1, 0);
6022 ret = btrfs_inc_extent_ref(trans, root,
6023 new_extents[0].disk_bytenr,
6024 new_extents[0].disk_num_bytes,
6026 root->root_key.objectid,
6031 ret = btrfs_free_extent(trans, root,
6032 extent_key->objectid,
6035 btrfs_header_owner(leaf),
6036 btrfs_header_generation(leaf),
6040 btrfs_release_path(root, path);
6041 key.offset += num_bytes;
6049 * drop old extent pointer at first, then insert the
6050 * new pointers one bye one
6052 btrfs_release_path(root, path);
6053 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6054 key.offset + num_bytes,
6055 key.offset, &alloc_hint);
6058 for (i = 0; i < nr_extents; i++) {
6059 if (ext_offset >= new_extents[i].num_bytes) {
6060 ext_offset -= new_extents[i].num_bytes;
6063 extent_len = min(new_extents[i].num_bytes -
6064 ext_offset, num_bytes);
6066 ret = btrfs_insert_empty_item(trans, root,
6071 leaf = path->nodes[0];
6072 fi = btrfs_item_ptr(leaf, path->slots[0],
6073 struct btrfs_file_extent_item);
6074 btrfs_set_file_extent_generation(leaf, fi,
6076 btrfs_set_file_extent_type(leaf, fi,
6077 BTRFS_FILE_EXTENT_REG);
6078 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6079 new_extents[i].disk_bytenr);
6080 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6081 new_extents[i].disk_num_bytes);
6082 btrfs_set_file_extent_ram_bytes(leaf, fi,
6083 new_extents[i].ram_bytes);
6085 btrfs_set_file_extent_compression(leaf, fi,
6086 new_extents[i].compression);
6087 btrfs_set_file_extent_encryption(leaf, fi,
6088 new_extents[i].encryption);
6089 btrfs_set_file_extent_other_encoding(leaf, fi,
6090 new_extents[i].other_encoding);
6092 btrfs_set_file_extent_num_bytes(leaf, fi,
6094 ext_offset += new_extents[i].offset;
6095 btrfs_set_file_extent_offset(leaf, fi,
6097 btrfs_mark_buffer_dirty(leaf);
6099 btrfs_drop_extent_cache(inode, key.offset,
6100 key.offset + extent_len - 1, 0);
6102 ret = btrfs_inc_extent_ref(trans, root,
6103 new_extents[i].disk_bytenr,
6104 new_extents[i].disk_num_bytes,
6106 root->root_key.objectid,
6107 trans->transid, key.objectid);
6109 btrfs_release_path(root, path);
6111 inode_add_bytes(inode, extent_len);
6114 num_bytes -= extent_len;
6115 key.offset += extent_len;
6120 BUG_ON(i >= nr_extents);
6124 if (extent_locked) {
6125 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6126 lock_end, GFP_NOFS);
6130 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6131 key.offset >= search_end)
6138 btrfs_release_path(root, path);
6140 mutex_unlock(&inode->i_mutex);
6141 if (extent_locked) {
6142 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6143 lock_end, GFP_NOFS);
6150 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6151 struct btrfs_root *root,
6152 struct extent_buffer *buf, u64 orig_start)
6157 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6158 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6160 level = btrfs_header_level(buf);
6162 struct btrfs_leaf_ref *ref;
6163 struct btrfs_leaf_ref *orig_ref;
6165 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6169 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6171 btrfs_free_leaf_ref(root, orig_ref);
6175 ref->nritems = orig_ref->nritems;
6176 memcpy(ref->extents, orig_ref->extents,
6177 sizeof(ref->extents[0]) * ref->nritems);
6179 btrfs_free_leaf_ref(root, orig_ref);
6181 ref->root_gen = trans->transid;
6182 ref->bytenr = buf->start;
6183 ref->owner = btrfs_header_owner(buf);
6184 ref->generation = btrfs_header_generation(buf);
6186 ret = btrfs_add_leaf_ref(root, ref, 0);
6188 btrfs_free_leaf_ref(root, ref);
6193 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6194 struct extent_buffer *leaf,
6195 struct btrfs_block_group_cache *group,
6196 struct btrfs_root *target_root)
6198 struct btrfs_key key;
6199 struct inode *inode = NULL;
6200 struct btrfs_file_extent_item *fi;
6202 u64 skip_objectid = 0;
6206 nritems = btrfs_header_nritems(leaf);
6207 for (i = 0; i < nritems; i++) {
6208 btrfs_item_key_to_cpu(leaf, &key, i);
6209 if (key.objectid == skip_objectid ||
6210 key.type != BTRFS_EXTENT_DATA_KEY)
6212 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6213 if (btrfs_file_extent_type(leaf, fi) ==
6214 BTRFS_FILE_EXTENT_INLINE)
6216 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6218 if (!inode || inode->i_ino != key.objectid) {
6220 inode = btrfs_ilookup(target_root->fs_info->sb,
6221 key.objectid, target_root, 1);
6224 skip_objectid = key.objectid;
6227 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6229 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6230 key.offset + num_bytes - 1, GFP_NOFS);
6231 btrfs_drop_extent_cache(inode, key.offset,
6232 key.offset + num_bytes - 1, 1);
6233 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6234 key.offset + num_bytes - 1, GFP_NOFS);
6241 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6242 struct btrfs_root *root,
6243 struct extent_buffer *leaf,
6244 struct btrfs_block_group_cache *group,
6245 struct inode *reloc_inode)
6247 struct btrfs_key key;
6248 struct btrfs_key extent_key;
6249 struct btrfs_file_extent_item *fi;
6250 struct btrfs_leaf_ref *ref;
6251 struct disk_extent *new_extent;
6260 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6261 BUG_ON(!new_extent);
6263 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6267 nritems = btrfs_header_nritems(leaf);
6268 for (i = 0; i < nritems; i++) {
6269 btrfs_item_key_to_cpu(leaf, &key, i);
6270 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6272 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6273 if (btrfs_file_extent_type(leaf, fi) ==
6274 BTRFS_FILE_EXTENT_INLINE)
6276 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6277 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6282 if (bytenr >= group->key.objectid + group->key.offset ||
6283 bytenr + num_bytes <= group->key.objectid)
6286 extent_key.objectid = bytenr;
6287 extent_key.offset = num_bytes;
6288 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6290 ret = get_new_locations(reloc_inode, &extent_key,
6291 group->key.objectid, 1,
6292 &new_extent, &nr_extent);
6297 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6298 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6299 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6300 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6302 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6303 new_extent->disk_bytenr);
6304 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6305 new_extent->disk_num_bytes);
6306 btrfs_mark_buffer_dirty(leaf);
6308 ret = btrfs_inc_extent_ref(trans, root,
6309 new_extent->disk_bytenr,
6310 new_extent->disk_num_bytes,
6312 root->root_key.objectid,
6313 trans->transid, key.objectid);
6316 ret = btrfs_free_extent(trans, root,
6317 bytenr, num_bytes, leaf->start,
6318 btrfs_header_owner(leaf),
6319 btrfs_header_generation(leaf),
6325 BUG_ON(ext_index + 1 != ref->nritems);
6326 btrfs_free_leaf_ref(root, ref);
6330 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6331 struct btrfs_root *root)
6333 struct btrfs_root *reloc_root;
6336 if (root->reloc_root) {
6337 reloc_root = root->reloc_root;
6338 root->reloc_root = NULL;
6339 list_add(&reloc_root->dead_list,
6340 &root->fs_info->dead_reloc_roots);
6342 btrfs_set_root_bytenr(&reloc_root->root_item,
6343 reloc_root->node->start);
6344 btrfs_set_root_level(&root->root_item,
6345 btrfs_header_level(reloc_root->node));
6346 memset(&reloc_root->root_item.drop_progress, 0,
6347 sizeof(struct btrfs_disk_key));
6348 reloc_root->root_item.drop_level = 0;
6350 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6351 &reloc_root->root_key,
6352 &reloc_root->root_item);
6358 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6360 struct btrfs_trans_handle *trans;
6361 struct btrfs_root *reloc_root;
6362 struct btrfs_root *prev_root = NULL;
6363 struct list_head dead_roots;
6367 INIT_LIST_HEAD(&dead_roots);
6368 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6370 while (!list_empty(&dead_roots)) {
6371 reloc_root = list_entry(dead_roots.prev,
6372 struct btrfs_root, dead_list);
6373 list_del_init(&reloc_root->dead_list);
6375 BUG_ON(reloc_root->commit_root != NULL);
6377 trans = btrfs_join_transaction(root, 1);
6380 mutex_lock(&root->fs_info->drop_mutex);
6381 ret = btrfs_drop_snapshot(trans, reloc_root);
6384 mutex_unlock(&root->fs_info->drop_mutex);
6386 nr = trans->blocks_used;
6387 ret = btrfs_end_transaction(trans, root);
6389 btrfs_btree_balance_dirty(root, nr);
6392 free_extent_buffer(reloc_root->node);
6394 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6395 &reloc_root->root_key);
6397 mutex_unlock(&root->fs_info->drop_mutex);
6399 nr = trans->blocks_used;
6400 ret = btrfs_end_transaction(trans, root);
6402 btrfs_btree_balance_dirty(root, nr);
6405 prev_root = reloc_root;
6408 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6414 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6416 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6420 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6422 struct btrfs_root *reloc_root;
6423 struct btrfs_trans_handle *trans;
6424 struct btrfs_key location;
6428 mutex_lock(&root->fs_info->tree_reloc_mutex);
6429 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6431 found = !list_empty(&root->fs_info->dead_reloc_roots);
6432 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6435 trans = btrfs_start_transaction(root, 1);
6437 ret = btrfs_commit_transaction(trans, root);
6441 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6442 location.offset = (u64)-1;
6443 location.type = BTRFS_ROOT_ITEM_KEY;
6445 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6446 BUG_ON(!reloc_root);
6447 btrfs_orphan_cleanup(reloc_root);
6451 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6452 struct btrfs_root *root)
6454 struct btrfs_root *reloc_root;
6455 struct extent_buffer *eb;
6456 struct btrfs_root_item *root_item;
6457 struct btrfs_key root_key;
6460 BUG_ON(!root->ref_cows);
6461 if (root->reloc_root)
6464 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6467 ret = btrfs_copy_root(trans, root, root->commit_root,
6468 &eb, BTRFS_TREE_RELOC_OBJECTID);
6471 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6472 root_key.offset = root->root_key.objectid;
6473 root_key.type = BTRFS_ROOT_ITEM_KEY;
6475 memcpy(root_item, &root->root_item, sizeof(root_item));
6476 btrfs_set_root_refs(root_item, 0);
6477 btrfs_set_root_bytenr(root_item, eb->start);
6478 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6479 btrfs_set_root_generation(root_item, trans->transid);
6481 btrfs_tree_unlock(eb);
6482 free_extent_buffer(eb);
6484 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6485 &root_key, root_item);
6489 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6491 BUG_ON(!reloc_root);
6492 reloc_root->last_trans = trans->transid;
6493 reloc_root->commit_root = NULL;
6494 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6496 root->reloc_root = reloc_root;
6501 * Core function of space balance.
6503 * The idea is using reloc trees to relocate tree blocks in reference
6504 * counted roots. There is one reloc tree for each subvol, and all
6505 * reloc trees share same root key objectid. Reloc trees are snapshots
6506 * of the latest committed roots of subvols (root->commit_root).
6508 * To relocate a tree block referenced by a subvol, there are two steps.
6509 * COW the block through subvol's reloc tree, then update block pointer
6510 * in the subvol to point to the new block. Since all reloc trees share
6511 * same root key objectid, doing special handing for tree blocks owned
6512 * by them is easy. Once a tree block has been COWed in one reloc tree,
6513 * we can use the resulting new block directly when the same block is
6514 * required to COW again through other reloc trees. By this way, relocated
6515 * tree blocks are shared between reloc trees, so they are also shared
6518 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6519 struct btrfs_root *root,
6520 struct btrfs_path *path,
6521 struct btrfs_key *first_key,
6522 struct btrfs_ref_path *ref_path,
6523 struct btrfs_block_group_cache *group,
6524 struct inode *reloc_inode)
6526 struct btrfs_root *reloc_root;
6527 struct extent_buffer *eb = NULL;
6528 struct btrfs_key *keys;
6532 int lowest_level = 0;
6535 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6536 lowest_level = ref_path->owner_objectid;
6538 if (!root->ref_cows) {
6539 path->lowest_level = lowest_level;
6540 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6542 path->lowest_level = 0;
6543 btrfs_release_path(root, path);
6547 mutex_lock(&root->fs_info->tree_reloc_mutex);
6548 ret = init_reloc_tree(trans, root);
6550 reloc_root = root->reloc_root;
6552 shared_level = ref_path->shared_level;
6553 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6555 keys = ref_path->node_keys;
6556 nodes = ref_path->new_nodes;
6557 memset(&keys[shared_level + 1], 0,
6558 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6559 memset(&nodes[shared_level + 1], 0,
6560 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6562 if (nodes[lowest_level] == 0) {
6563 path->lowest_level = lowest_level;
6564 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6567 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6568 eb = path->nodes[level];
6569 if (!eb || eb == reloc_root->node)
6571 nodes[level] = eb->start;
6573 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6575 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6578 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6579 eb = path->nodes[0];
6580 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6581 group, reloc_inode);
6584 btrfs_release_path(reloc_root, path);
6586 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6592 * replace tree blocks in the fs tree with tree blocks in
6595 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6598 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6599 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6602 extent_buffer_get(path->nodes[0]);
6603 eb = path->nodes[0];
6604 btrfs_release_path(reloc_root, path);
6605 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6607 free_extent_buffer(eb);
6610 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6611 path->lowest_level = 0;
6615 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6616 struct btrfs_root *root,
6617 struct btrfs_path *path,
6618 struct btrfs_key *first_key,
6619 struct btrfs_ref_path *ref_path)
6623 ret = relocate_one_path(trans, root, path, first_key,
6624 ref_path, NULL, NULL);
6630 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6631 struct btrfs_root *extent_root,
6632 struct btrfs_path *path,
6633 struct btrfs_key *extent_key)
6637 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6640 ret = btrfs_del_item(trans, extent_root, path);
6642 btrfs_release_path(extent_root, path);
6646 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6647 struct btrfs_ref_path *ref_path)
6649 struct btrfs_key root_key;
6651 root_key.objectid = ref_path->root_objectid;
6652 root_key.type = BTRFS_ROOT_ITEM_KEY;
6653 if (is_cowonly_root(ref_path->root_objectid))
6654 root_key.offset = 0;
6656 root_key.offset = (u64)-1;
6658 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6661 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6662 struct btrfs_path *path,
6663 struct btrfs_key *extent_key,
6664 struct btrfs_block_group_cache *group,
6665 struct inode *reloc_inode, int pass)
6667 struct btrfs_trans_handle *trans;
6668 struct btrfs_root *found_root;
6669 struct btrfs_ref_path *ref_path = NULL;
6670 struct disk_extent *new_extents = NULL;
6675 struct btrfs_key first_key;
6679 trans = btrfs_start_transaction(extent_root, 1);
6682 if (extent_key->objectid == 0) {
6683 ret = del_extent_zero(trans, extent_root, path, extent_key);
6687 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6693 for (loops = 0; ; loops++) {
6695 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6696 extent_key->objectid);
6698 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6705 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6706 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6709 found_root = read_ref_root(extent_root->fs_info, ref_path);
6710 BUG_ON(!found_root);
6712 * for reference counted tree, only process reference paths
6713 * rooted at the latest committed root.
6715 if (found_root->ref_cows &&
6716 ref_path->root_generation != found_root->root_key.offset)
6719 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6722 * copy data extents to new locations
6724 u64 group_start = group->key.objectid;
6725 ret = relocate_data_extent(reloc_inode,
6734 level = ref_path->owner_objectid;
6737 if (prev_block != ref_path->nodes[level]) {
6738 struct extent_buffer *eb;
6739 u64 block_start = ref_path->nodes[level];
6740 u64 block_size = btrfs_level_size(found_root, level);
6742 eb = read_tree_block(found_root, block_start,
6744 btrfs_tree_lock(eb);
6745 BUG_ON(level != btrfs_header_level(eb));
6748 btrfs_item_key_to_cpu(eb, &first_key, 0);
6750 btrfs_node_key_to_cpu(eb, &first_key, 0);
6752 btrfs_tree_unlock(eb);
6753 free_extent_buffer(eb);
6754 prev_block = block_start;
6757 mutex_lock(&extent_root->fs_info->trans_mutex);
6758 btrfs_record_root_in_trans(found_root);
6759 mutex_unlock(&extent_root->fs_info->trans_mutex);
6760 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6762 * try to update data extent references while
6763 * keeping metadata shared between snapshots.
6766 ret = relocate_one_path(trans, found_root,
6767 path, &first_key, ref_path,
6768 group, reloc_inode);
6774 * use fallback method to process the remaining
6778 u64 group_start = group->key.objectid;
6779 new_extents = kmalloc(sizeof(*new_extents),
6782 ret = get_new_locations(reloc_inode,
6790 ret = replace_one_extent(trans, found_root,
6792 &first_key, ref_path,
6793 new_extents, nr_extents);
6795 ret = relocate_tree_block(trans, found_root, path,
6796 &first_key, ref_path);
6803 btrfs_end_transaction(trans, extent_root);
6810 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6813 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6814 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6816 num_devices = root->fs_info->fs_devices->rw_devices;
6817 if (num_devices == 1) {
6818 stripped |= BTRFS_BLOCK_GROUP_DUP;
6819 stripped = flags & ~stripped;
6821 /* turn raid0 into single device chunks */
6822 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6825 /* turn mirroring into duplication */
6826 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6827 BTRFS_BLOCK_GROUP_RAID10))
6828 return stripped | BTRFS_BLOCK_GROUP_DUP;
6831 /* they already had raid on here, just return */
6832 if (flags & stripped)
6835 stripped |= BTRFS_BLOCK_GROUP_DUP;
6836 stripped = flags & ~stripped;
6838 /* switch duplicated blocks with raid1 */
6839 if (flags & BTRFS_BLOCK_GROUP_DUP)
6840 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6842 /* turn single device chunks into raid0 */
6843 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6848 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6849 struct btrfs_block_group_cache *shrink_block_group,
6852 struct btrfs_trans_handle *trans;
6853 u64 new_alloc_flags;
6856 spin_lock(&shrink_block_group->lock);
6857 if (btrfs_block_group_used(&shrink_block_group->item) +
6858 shrink_block_group->reserved > 0) {
6859 spin_unlock(&shrink_block_group->lock);
6861 trans = btrfs_start_transaction(root, 1);
6862 spin_lock(&shrink_block_group->lock);
6864 new_alloc_flags = update_block_group_flags(root,
6865 shrink_block_group->flags);
6866 if (new_alloc_flags != shrink_block_group->flags) {
6868 btrfs_block_group_used(&shrink_block_group->item);
6870 calc = shrink_block_group->key.offset;
6872 spin_unlock(&shrink_block_group->lock);
6874 do_chunk_alloc(trans, root->fs_info->extent_root,
6875 calc + 2 * 1024 * 1024, new_alloc_flags, force);
6877 btrfs_end_transaction(trans, root);
6879 spin_unlock(&shrink_block_group->lock);
6884 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6885 struct btrfs_block_group_cache *group)
6888 __alloc_chunk_for_shrink(root, group, 1);
6889 set_block_group_readonly(group);
6894 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6895 struct btrfs_root *root,
6896 u64 objectid, u64 size)
6898 struct btrfs_path *path;
6899 struct btrfs_inode_item *item;
6900 struct extent_buffer *leaf;
6903 path = btrfs_alloc_path();
6907 path->leave_spinning = 1;
6908 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6912 leaf = path->nodes[0];
6913 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6914 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6915 btrfs_set_inode_generation(leaf, item, 1);
6916 btrfs_set_inode_size(leaf, item, size);
6917 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6918 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6919 btrfs_mark_buffer_dirty(leaf);
6920 btrfs_release_path(root, path);
6922 btrfs_free_path(path);
6926 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6927 struct btrfs_block_group_cache *group)
6929 struct inode *inode = NULL;
6930 struct btrfs_trans_handle *trans;
6931 struct btrfs_root *root;
6932 struct btrfs_key root_key;
6933 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6936 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6937 root_key.type = BTRFS_ROOT_ITEM_KEY;
6938 root_key.offset = (u64)-1;
6939 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6941 return ERR_CAST(root);
6943 trans = btrfs_start_transaction(root, 1);
6946 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6950 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6953 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6954 group->key.offset, 0, group->key.offset,
6958 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6959 if (inode->i_state & I_NEW) {
6960 BTRFS_I(inode)->root = root;
6961 BTRFS_I(inode)->location.objectid = objectid;
6962 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6963 BTRFS_I(inode)->location.offset = 0;
6964 btrfs_read_locked_inode(inode);
6965 unlock_new_inode(inode);
6966 BUG_ON(is_bad_inode(inode));
6970 BTRFS_I(inode)->index_cnt = group->key.objectid;
6972 err = btrfs_orphan_add(trans, inode);
6974 btrfs_end_transaction(trans, root);
6978 inode = ERR_PTR(err);
6983 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6986 struct btrfs_ordered_sum *sums;
6987 struct btrfs_sector_sum *sector_sum;
6988 struct btrfs_ordered_extent *ordered;
6989 struct btrfs_root *root = BTRFS_I(inode)->root;
6990 struct list_head list;
6995 INIT_LIST_HEAD(&list);
6997 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6998 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
7000 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
7001 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
7002 disk_bytenr + len - 1, &list);
7004 while (!list_empty(&list)) {
7005 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
7006 list_del_init(&sums->list);
7008 sector_sum = sums->sums;
7009 sums->bytenr = ordered->start;
7012 while (offset < sums->len) {
7013 sector_sum->bytenr += ordered->start - disk_bytenr;
7015 offset += root->sectorsize;
7018 btrfs_add_ordered_sum(inode, ordered, sums);
7020 btrfs_put_ordered_extent(ordered);
7024 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
7026 struct btrfs_trans_handle *trans;
7027 struct btrfs_path *path;
7028 struct btrfs_fs_info *info = root->fs_info;
7029 struct extent_buffer *leaf;
7030 struct inode *reloc_inode;
7031 struct btrfs_block_group_cache *block_group;
7032 struct btrfs_key key;
7041 root = root->fs_info->extent_root;
7043 block_group = btrfs_lookup_block_group(info, group_start);
7044 BUG_ON(!block_group);
7046 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
7047 (unsigned long long)block_group->key.objectid,
7048 (unsigned long long)block_group->flags);
7050 path = btrfs_alloc_path();
7053 reloc_inode = create_reloc_inode(info, block_group);
7054 BUG_ON(IS_ERR(reloc_inode));
7056 __alloc_chunk_for_shrink(root, block_group, 1);
7057 set_block_group_readonly(block_group);
7059 btrfs_start_delalloc_inodes(info->tree_root);
7060 btrfs_wait_ordered_extents(info->tree_root, 0);
7065 key.objectid = block_group->key.objectid;
7068 cur_byte = key.objectid;
7070 trans = btrfs_start_transaction(info->tree_root, 1);
7071 btrfs_commit_transaction(trans, info->tree_root);
7073 mutex_lock(&root->fs_info->cleaner_mutex);
7074 btrfs_clean_old_snapshots(info->tree_root);
7075 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
7076 mutex_unlock(&root->fs_info->cleaner_mutex);
7078 trans = btrfs_start_transaction(info->tree_root, 1);
7079 btrfs_commit_transaction(trans, info->tree_root);
7082 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7086 leaf = path->nodes[0];
7087 nritems = btrfs_header_nritems(leaf);
7088 if (path->slots[0] >= nritems) {
7089 ret = btrfs_next_leaf(root, path);
7096 leaf = path->nodes[0];
7097 nritems = btrfs_header_nritems(leaf);
7100 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7102 if (key.objectid >= block_group->key.objectid +
7103 block_group->key.offset)
7106 if (progress && need_resched()) {
7107 btrfs_release_path(root, path);
7114 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7115 key.objectid + key.offset <= cur_byte) {
7121 cur_byte = key.objectid + key.offset;
7122 btrfs_release_path(root, path);
7124 __alloc_chunk_for_shrink(root, block_group, 0);
7125 ret = relocate_one_extent(root, path, &key, block_group,
7131 key.objectid = cur_byte;
7136 btrfs_release_path(root, path);
7139 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7140 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7143 if (total_found > 0) {
7144 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7145 (unsigned long long)total_found, pass);
7147 if (total_found == skipped && pass > 2) {
7149 reloc_inode = create_reloc_inode(info, block_group);
7155 /* delete reloc_inode */
7158 /* unpin extents in this range */
7159 trans = btrfs_start_transaction(info->tree_root, 1);
7160 btrfs_commit_transaction(trans, info->tree_root);
7162 spin_lock(&block_group->lock);
7163 WARN_ON(block_group->pinned > 0);
7164 WARN_ON(block_group->reserved > 0);
7165 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7166 spin_unlock(&block_group->lock);
7167 btrfs_put_block_group(block_group);
7170 btrfs_free_path(path);
7175 static int find_first_block_group(struct btrfs_root *root,
7176 struct btrfs_path *path, struct btrfs_key *key)
7179 struct btrfs_key found_key;
7180 struct extent_buffer *leaf;
7183 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7188 slot = path->slots[0];
7189 leaf = path->nodes[0];
7190 if (slot >= btrfs_header_nritems(leaf)) {
7191 ret = btrfs_next_leaf(root, path);
7198 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7200 if (found_key.objectid >= key->objectid &&
7201 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7212 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7214 struct btrfs_block_group_cache *block_group;
7215 struct btrfs_space_info *space_info;
7218 spin_lock(&info->block_group_cache_lock);
7219 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7220 block_group = rb_entry(n, struct btrfs_block_group_cache,
7222 rb_erase(&block_group->cache_node,
7223 &info->block_group_cache_tree);
7224 spin_unlock(&info->block_group_cache_lock);
7226 down_write(&block_group->space_info->groups_sem);
7227 list_del(&block_group->list);
7228 up_write(&block_group->space_info->groups_sem);
7230 if (block_group->cached == BTRFS_CACHE_STARTED)
7231 wait_event(block_group->caching_q,
7232 block_group_cache_done(block_group));
7234 btrfs_remove_free_space_cache(block_group);
7236 WARN_ON(atomic_read(&block_group->count) != 1);
7239 spin_lock(&info->block_group_cache_lock);
7241 spin_unlock(&info->block_group_cache_lock);
7243 /* now that all the block groups are freed, go through and
7244 * free all the space_info structs. This is only called during
7245 * the final stages of unmount, and so we know nobody is
7246 * using them. We call synchronize_rcu() once before we start,
7247 * just to be on the safe side.
7251 while(!list_empty(&info->space_info)) {
7252 space_info = list_entry(info->space_info.next,
7253 struct btrfs_space_info,
7256 list_del(&space_info->list);
7262 int btrfs_read_block_groups(struct btrfs_root *root)
7264 struct btrfs_path *path;
7266 struct btrfs_block_group_cache *cache;
7267 struct btrfs_fs_info *info = root->fs_info;
7268 struct btrfs_space_info *space_info;
7269 struct btrfs_key key;
7270 struct btrfs_key found_key;
7271 struct extent_buffer *leaf;
7273 root = info->extent_root;
7276 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7277 path = btrfs_alloc_path();
7282 ret = find_first_block_group(root, path, &key);
7290 leaf = path->nodes[0];
7291 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7292 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7298 atomic_set(&cache->count, 1);
7299 spin_lock_init(&cache->lock);
7300 spin_lock_init(&cache->tree_lock);
7301 cache->fs_info = info;
7302 init_waitqueue_head(&cache->caching_q);
7303 INIT_LIST_HEAD(&cache->list);
7304 INIT_LIST_HEAD(&cache->cluster_list);
7307 * we only want to have 32k of ram per block group for keeping
7308 * track of free space, and if we pass 1/2 of that we want to
7309 * start converting things over to using bitmaps
7311 cache->extents_thresh = ((1024 * 32) / 2) /
7312 sizeof(struct btrfs_free_space);
7314 read_extent_buffer(leaf, &cache->item,
7315 btrfs_item_ptr_offset(leaf, path->slots[0]),
7316 sizeof(cache->item));
7317 memcpy(&cache->key, &found_key, sizeof(found_key));
7319 key.objectid = found_key.objectid + found_key.offset;
7320 btrfs_release_path(root, path);
7321 cache->flags = btrfs_block_group_flags(&cache->item);
7322 cache->sectorsize = root->sectorsize;
7324 remove_sb_from_cache(root, cache);
7327 * check for two cases, either we are full, and therefore
7328 * don't need to bother with the caching work since we won't
7329 * find any space, or we are empty, and we can just add all
7330 * the space in and be done with it. This saves us _alot_ of
7331 * time, particularly in the full case.
7333 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7334 cache->cached = BTRFS_CACHE_FINISHED;
7335 } else if (btrfs_block_group_used(&cache->item) == 0) {
7336 cache->cached = BTRFS_CACHE_FINISHED;
7337 add_new_free_space(cache, root->fs_info,
7339 found_key.objectid +
7343 ret = update_space_info(info, cache->flags, found_key.offset,
7344 btrfs_block_group_used(&cache->item),
7347 cache->space_info = space_info;
7348 down_write(&space_info->groups_sem);
7349 list_add_tail(&cache->list, &space_info->block_groups);
7350 up_write(&space_info->groups_sem);
7352 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7355 set_avail_alloc_bits(root->fs_info, cache->flags);
7356 if (btrfs_chunk_readonly(root, cache->key.objectid))
7357 set_block_group_readonly(cache);
7361 btrfs_free_path(path);
7365 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7366 struct btrfs_root *root, u64 bytes_used,
7367 u64 type, u64 chunk_objectid, u64 chunk_offset,
7371 struct btrfs_root *extent_root;
7372 struct btrfs_block_group_cache *cache;
7374 extent_root = root->fs_info->extent_root;
7376 root->fs_info->last_trans_log_full_commit = trans->transid;
7378 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7382 cache->key.objectid = chunk_offset;
7383 cache->key.offset = size;
7384 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7385 cache->sectorsize = root->sectorsize;
7388 * we only want to have 32k of ram per block group for keeping track
7389 * of free space, and if we pass 1/2 of that we want to start
7390 * converting things over to using bitmaps
7392 cache->extents_thresh = ((1024 * 32) / 2) /
7393 sizeof(struct btrfs_free_space);
7394 atomic_set(&cache->count, 1);
7395 spin_lock_init(&cache->lock);
7396 spin_lock_init(&cache->tree_lock);
7397 init_waitqueue_head(&cache->caching_q);
7398 INIT_LIST_HEAD(&cache->list);
7399 INIT_LIST_HEAD(&cache->cluster_list);
7401 btrfs_set_block_group_used(&cache->item, bytes_used);
7402 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7403 cache->flags = type;
7404 btrfs_set_block_group_flags(&cache->item, type);
7406 cache->cached = BTRFS_CACHE_FINISHED;
7407 remove_sb_from_cache(root, cache);
7409 add_new_free_space(cache, root->fs_info, chunk_offset,
7410 chunk_offset + size);
7412 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7413 &cache->space_info);
7415 down_write(&cache->space_info->groups_sem);
7416 list_add_tail(&cache->list, &cache->space_info->block_groups);
7417 up_write(&cache->space_info->groups_sem);
7419 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7422 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7423 sizeof(cache->item));
7426 set_avail_alloc_bits(extent_root->fs_info, type);
7431 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7432 struct btrfs_root *root, u64 group_start)
7434 struct btrfs_path *path;
7435 struct btrfs_block_group_cache *block_group;
7436 struct btrfs_free_cluster *cluster;
7437 struct btrfs_key key;
7440 root = root->fs_info->extent_root;
7442 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7443 BUG_ON(!block_group);
7444 BUG_ON(!block_group->ro);
7446 memcpy(&key, &block_group->key, sizeof(key));
7448 /* make sure this block group isn't part of an allocation cluster */
7449 cluster = &root->fs_info->data_alloc_cluster;
7450 spin_lock(&cluster->refill_lock);
7451 btrfs_return_cluster_to_free_space(block_group, cluster);
7452 spin_unlock(&cluster->refill_lock);
7455 * make sure this block group isn't part of a metadata
7456 * allocation cluster
7458 cluster = &root->fs_info->meta_alloc_cluster;
7459 spin_lock(&cluster->refill_lock);
7460 btrfs_return_cluster_to_free_space(block_group, cluster);
7461 spin_unlock(&cluster->refill_lock);
7463 path = btrfs_alloc_path();
7466 spin_lock(&root->fs_info->block_group_cache_lock);
7467 rb_erase(&block_group->cache_node,
7468 &root->fs_info->block_group_cache_tree);
7469 spin_unlock(&root->fs_info->block_group_cache_lock);
7471 down_write(&block_group->space_info->groups_sem);
7473 * we must use list_del_init so people can check to see if they
7474 * are still on the list after taking the semaphore
7476 list_del_init(&block_group->list);
7477 up_write(&block_group->space_info->groups_sem);
7479 if (block_group->cached == BTRFS_CACHE_STARTED)
7480 wait_event(block_group->caching_q,
7481 block_group_cache_done(block_group));
7483 btrfs_remove_free_space_cache(block_group);
7485 spin_lock(&block_group->space_info->lock);
7486 block_group->space_info->total_bytes -= block_group->key.offset;
7487 block_group->space_info->bytes_readonly -= block_group->key.offset;
7488 spin_unlock(&block_group->space_info->lock);
7490 btrfs_clear_space_info_full(root->fs_info);
7492 btrfs_put_block_group(block_group);
7493 btrfs_put_block_group(block_group);
7495 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7501 ret = btrfs_del_item(trans, root, path);
7503 btrfs_free_path(path);