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>
26 #include "print-tree.h"
27 #include "transaction.h"
30 #include "ref-cache.h"
32 #define PENDING_EXTENT_INSERT 0
33 #define PENDING_EXTENT_DELETE 1
34 #define PENDING_BACKREF_UPDATE 2
36 struct pending_extent_op {
47 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
48 btrfs_root *extent_root);
49 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
50 btrfs_root *extent_root);
51 static struct btrfs_block_group_cache *
52 __btrfs_find_block_group(struct btrfs_root *root,
53 struct btrfs_block_group_cache *hint,
54 u64 search_start, int data, int owner);
56 void maybe_lock_mutex(struct btrfs_root *root)
58 if (root != root->fs_info->extent_root &&
59 root != root->fs_info->chunk_root &&
60 root != root->fs_info->dev_root) {
61 mutex_lock(&root->fs_info->alloc_mutex);
65 void maybe_unlock_mutex(struct btrfs_root *root)
67 if (root != root->fs_info->extent_root &&
68 root != root->fs_info->chunk_root &&
69 root != root->fs_info->dev_root) {
70 mutex_unlock(&root->fs_info->alloc_mutex);
74 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
76 return (cache->flags & bits) == bits;
80 * this adds the block group to the fs_info rb tree for the block group
83 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
84 struct btrfs_block_group_cache *block_group)
87 struct rb_node *parent = NULL;
88 struct btrfs_block_group_cache *cache;
90 spin_lock(&info->block_group_cache_lock);
91 p = &info->block_group_cache_tree.rb_node;
95 cache = rb_entry(parent, struct btrfs_block_group_cache,
97 if (block_group->key.objectid < cache->key.objectid) {
99 } else if (block_group->key.objectid > cache->key.objectid) {
102 spin_unlock(&info->block_group_cache_lock);
107 rb_link_node(&block_group->cache_node, parent, p);
108 rb_insert_color(&block_group->cache_node,
109 &info->block_group_cache_tree);
110 spin_unlock(&info->block_group_cache_lock);
116 * This will return the block group at or after bytenr if contains is 0, else
117 * it will return the block group that contains the bytenr
119 static struct btrfs_block_group_cache *
120 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
123 struct btrfs_block_group_cache *cache, *ret = NULL;
127 spin_lock(&info->block_group_cache_lock);
128 n = info->block_group_cache_tree.rb_node;
131 cache = rb_entry(n, struct btrfs_block_group_cache,
133 end = cache->key.objectid + cache->key.offset - 1;
134 start = cache->key.objectid;
136 if (bytenr < start) {
137 if (!contains && (!ret || start < ret->key.objectid))
140 } else if (bytenr > start) {
141 if (contains && bytenr <= end) {
151 spin_unlock(&info->block_group_cache_lock);
157 * this is only called by cache_block_group, since we could have freed extents
158 * we need to check the pinned_extents for any extents that can't be used yet
159 * since their free space will be released as soon as the transaction commits.
161 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
162 struct btrfs_fs_info *info, u64 start, u64 end)
164 u64 extent_start, extent_end, size;
167 while (start < end) {
168 ret = find_first_extent_bit(&info->pinned_extents, start,
169 &extent_start, &extent_end,
174 if (extent_start == start) {
175 start = extent_end + 1;
176 } else if (extent_start > start && extent_start < end) {
177 size = extent_start - start;
178 ret = btrfs_add_free_space(block_group, start, size);
180 start = extent_end + 1;
188 ret = btrfs_add_free_space(block_group, start, size);
195 static int cache_block_group(struct btrfs_root *root,
196 struct btrfs_block_group_cache *block_group)
198 struct btrfs_path *path;
200 struct btrfs_key key;
201 struct extent_buffer *leaf;
210 root = root->fs_info->extent_root;
212 if (block_group->cached)
215 path = btrfs_alloc_path();
221 * we get into deadlocks with paths held by callers of this function.
222 * since the alloc_mutex is protecting things right now, just
223 * skip the locking here
225 path->skip_locking = 1;
226 first_free = max_t(u64, block_group->key.objectid,
227 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
228 key.objectid = block_group->key.objectid;
230 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
231 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
234 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
238 leaf = path->nodes[0];
239 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
240 if (key.objectid + key.offset > first_free)
241 first_free = key.objectid + key.offset;
244 leaf = path->nodes[0];
245 slot = path->slots[0];
246 if (slot >= btrfs_header_nritems(leaf)) {
247 ret = btrfs_next_leaf(root, path);
255 btrfs_item_key_to_cpu(leaf, &key, slot);
256 if (key.objectid < block_group->key.objectid)
259 if (key.objectid >= block_group->key.objectid +
260 block_group->key.offset)
263 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
269 add_new_free_space(block_group, root->fs_info, last,
272 last = key.objectid + key.offset;
281 add_new_free_space(block_group, root->fs_info, last,
282 block_group->key.objectid +
283 block_group->key.offset);
285 block_group->cached = 1;
288 btrfs_free_path(path);
293 * return the block group that starts at or after bytenr
295 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
299 struct btrfs_block_group_cache *cache;
301 cache = block_group_cache_tree_search(info, bytenr, 0);
307 * return the block group that contains teh given bytenr
309 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
313 struct btrfs_block_group_cache *cache;
315 cache = block_group_cache_tree_search(info, bytenr, 1);
320 static int noinline find_free_space(struct btrfs_root *root,
321 struct btrfs_block_group_cache **cache_ret,
322 u64 *start_ret, u64 num, int data)
325 struct btrfs_block_group_cache *cache = *cache_ret;
326 struct btrfs_free_space *info = NULL;
328 u64 search_start = *start_ret;
330 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
334 last = max(search_start, cache->key.objectid);
337 ret = cache_block_group(root, cache);
341 if (cache->ro || !block_group_bits(cache, data))
344 info = btrfs_find_free_space(cache, last, num);
346 *start_ret = info->offset;
351 last = cache->key.objectid + cache->key.offset;
353 cache = btrfs_lookup_first_block_group(root->fs_info, last);
364 static u64 div_factor(u64 num, int factor)
373 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
376 struct list_head *head = &info->space_info;
377 struct list_head *cur;
378 struct btrfs_space_info *found;
379 list_for_each(cur, head) {
380 found = list_entry(cur, struct btrfs_space_info, list);
381 if (found->flags == flags)
387 static struct btrfs_block_group_cache *
388 __btrfs_find_block_group(struct btrfs_root *root,
389 struct btrfs_block_group_cache *hint,
390 u64 search_start, int data, int owner)
392 struct btrfs_block_group_cache *cache;
393 struct btrfs_block_group_cache *found_group = NULL;
394 struct btrfs_fs_info *info = root->fs_info;
402 if (data & BTRFS_BLOCK_GROUP_METADATA)
406 struct btrfs_block_group_cache *shint;
407 shint = btrfs_lookup_first_block_group(info, search_start);
408 if (shint && block_group_bits(shint, data) && !shint->ro) {
409 spin_lock(&shint->lock);
410 used = btrfs_block_group_used(&shint->item);
411 if (used + shint->pinned + shint->reserved <
412 div_factor(shint->key.offset, factor)) {
413 spin_unlock(&shint->lock);
416 spin_unlock(&shint->lock);
419 if (hint && !hint->ro && block_group_bits(hint, data)) {
420 spin_lock(&hint->lock);
421 used = btrfs_block_group_used(&hint->item);
422 if (used + hint->pinned + hint->reserved <
423 div_factor(hint->key.offset, factor)) {
424 spin_unlock(&hint->lock);
427 spin_unlock(&hint->lock);
428 last = hint->key.objectid + hint->key.offset;
431 last = max(hint->key.objectid, search_start);
437 cache = btrfs_lookup_first_block_group(root->fs_info, last);
441 spin_lock(&cache->lock);
442 last = cache->key.objectid + cache->key.offset;
443 used = btrfs_block_group_used(&cache->item);
445 if (!cache->ro && block_group_bits(cache, data)) {
446 free_check = div_factor(cache->key.offset, factor);
447 if (used + cache->pinned + cache->reserved <
450 spin_unlock(&cache->lock);
454 spin_unlock(&cache->lock);
462 if (!full_search && factor < 10) {
472 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
473 struct btrfs_block_group_cache
474 *hint, u64 search_start,
478 struct btrfs_block_group_cache *ret;
479 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
483 /* simple helper to search for an existing extent at a given offset */
484 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
487 struct btrfs_key key;
488 struct btrfs_path *path;
490 path = btrfs_alloc_path();
492 maybe_lock_mutex(root);
493 key.objectid = start;
495 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
496 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
498 maybe_unlock_mutex(root);
499 btrfs_free_path(path);
504 * Back reference rules. Back refs have three main goals:
506 * 1) differentiate between all holders of references to an extent so that
507 * when a reference is dropped we can make sure it was a valid reference
508 * before freeing the extent.
510 * 2) Provide enough information to quickly find the holders of an extent
511 * if we notice a given block is corrupted or bad.
513 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
514 * maintenance. This is actually the same as #2, but with a slightly
515 * different use case.
517 * File extents can be referenced by:
519 * - multiple snapshots, subvolumes, or different generations in one subvol
520 * - different files inside a single subvolume
521 * - different offsets inside a file (bookend extents in file.c)
523 * The extent ref structure has fields for:
525 * - Objectid of the subvolume root
526 * - Generation number of the tree holding the reference
527 * - objectid of the file holding the reference
528 * - offset in the file corresponding to the key holding the reference
529 * - number of references holding by parent node (alway 1 for tree blocks)
531 * Btree leaf may hold multiple references to a file extent. In most cases,
532 * these references are from same file and the corresponding offsets inside
533 * the file are close together. So inode objectid and offset in file are
534 * just hints, they provide hints about where in the btree the references
535 * can be found and when we can stop searching.
537 * When a file extent is allocated the fields are filled in:
538 * (root_key.objectid, trans->transid, inode objectid, offset in file, 1)
540 * When a leaf is cow'd new references are added for every file extent found
541 * in the leaf. It looks similar to the create case, but trans->transid will
542 * be different when the block is cow'd.
544 * (root_key.objectid, trans->transid, inode objectid, offset in file,
545 * number of references in the leaf)
547 * Because inode objectid and offset in file are just hints, they are not
548 * used when backrefs are deleted. When a file extent is removed either
549 * during snapshot deletion or file truncation, we find the corresponding
550 * back back reference and check the following fields.
552 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf))
554 * Btree extents can be referenced by:
556 * - Different subvolumes
557 * - Different generations of the same subvolume
559 * When a tree block is created, back references are inserted:
561 * (root->root_key.objectid, trans->transid, level, 0, 1)
563 * When a tree block is cow'd, new back references are added for all the
564 * blocks it points to. If the tree block isn't in reference counted root,
565 * the old back references are removed. These new back references are of
566 * the form (trans->transid will have increased since creation):
568 * (root->root_key.objectid, trans->transid, level, 0, 1)
570 * When a backref is in deleting, the following fields are checked:
572 * if backref was for a tree root:
573 * (btrfs_header_owner(itself), btrfs_header_generation(itself))
575 * (btrfs_header_owner(parent), btrfs_header_generation(parent))
577 * Back Reference Key composing:
579 * The key objectid corresponds to the first byte in the extent, the key
580 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
581 * byte of parent extent. If a extent is tree root, the key offset is set
582 * to the key objectid.
585 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path, u64 bytenr,
588 u64 parent, u64 ref_root,
589 u64 ref_generation, int del)
591 struct btrfs_key key;
592 struct btrfs_extent_ref *ref;
593 struct extent_buffer *leaf;
596 key.objectid = bytenr;
597 key.type = BTRFS_EXTENT_REF_KEY;
600 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
608 leaf = path->nodes[0];
609 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
610 if (btrfs_ref_root(leaf, ref) != ref_root ||
611 btrfs_ref_generation(leaf, ref) != ref_generation) {
621 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path,
624 u64 bytenr, u64 parent,
625 u64 ref_root, u64 ref_generation,
626 u64 owner_objectid, u64 owner_offset)
628 struct btrfs_key key;
629 struct extent_buffer *leaf;
630 struct btrfs_extent_ref *ref;
634 key.objectid = bytenr;
635 key.type = BTRFS_EXTENT_REF_KEY;
638 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
640 leaf = path->nodes[0];
641 ref = btrfs_item_ptr(leaf, path->slots[0],
642 struct btrfs_extent_ref);
643 btrfs_set_ref_root(leaf, ref, ref_root);
644 btrfs_set_ref_generation(leaf, ref, ref_generation);
645 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
646 btrfs_set_ref_offset(leaf, ref, owner_offset);
647 btrfs_set_ref_num_refs(leaf, ref, 1);
648 } else if (ret == -EEXIST) {
650 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
651 leaf = path->nodes[0];
652 ref = btrfs_item_ptr(leaf, path->slots[0],
653 struct btrfs_extent_ref);
654 if (btrfs_ref_root(leaf, ref) != ref_root ||
655 btrfs_ref_generation(leaf, ref) != ref_generation) {
661 num_refs = btrfs_ref_num_refs(leaf, ref);
662 BUG_ON(num_refs == 0);
663 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
665 existing_owner = btrfs_ref_objectid(leaf, ref);
666 if (existing_owner == owner_objectid &&
667 btrfs_ref_offset(leaf, ref) > owner_offset) {
668 btrfs_set_ref_offset(leaf, ref, owner_offset);
669 } else if (existing_owner != owner_objectid &&
670 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
671 btrfs_set_ref_objectid(leaf, ref,
672 BTRFS_MULTIPLE_OBJECTIDS);
673 btrfs_set_ref_offset(leaf, ref, 0);
679 btrfs_mark_buffer_dirty(path->nodes[0]);
681 btrfs_release_path(root, path);
685 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
686 struct btrfs_root *root,
687 struct btrfs_path *path)
689 struct extent_buffer *leaf;
690 struct btrfs_extent_ref *ref;
694 leaf = path->nodes[0];
695 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
696 num_refs = btrfs_ref_num_refs(leaf, ref);
697 BUG_ON(num_refs == 0);
700 ret = btrfs_del_item(trans, root, path);
702 btrfs_set_ref_num_refs(leaf, ref, num_refs);
703 btrfs_mark_buffer_dirty(leaf);
705 btrfs_release_path(root, path);
709 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
710 struct btrfs_root *root, u64 bytenr,
711 u64 orig_parent, u64 parent,
712 u64 orig_root, u64 ref_root,
713 u64 orig_generation, u64 ref_generation,
714 u64 owner_objectid, u64 owner_offset)
717 struct btrfs_root *extent_root = root->fs_info->extent_root;
718 struct btrfs_path *path;
720 if (root == root->fs_info->extent_root) {
721 struct pending_extent_op *extent_op;
724 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
725 num_bytes = btrfs_level_size(root, (int)owner_objectid);
726 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
727 bytenr + num_bytes - 1, EXTENT_LOCKED, 0)) {
729 ret = get_state_private(&root->fs_info->extent_ins,
732 extent_op = (struct pending_extent_op *)
734 BUG_ON(extent_op->parent != orig_parent);
735 BUG_ON(extent_op->generation != orig_generation);
736 extent_op->parent = parent;
737 extent_op->generation = ref_generation;
739 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
742 extent_op->type = PENDING_BACKREF_UPDATE;
743 extent_op->bytenr = bytenr;
744 extent_op->num_bytes = num_bytes;
745 extent_op->parent = parent;
746 extent_op->orig_parent = orig_parent;
747 extent_op->generation = ref_generation;
748 extent_op->orig_generation = orig_generation;
749 extent_op->level = (int)owner_objectid;
751 set_extent_bits(&root->fs_info->extent_ins,
752 bytenr, bytenr + num_bytes - 1,
753 EXTENT_LOCKED, GFP_NOFS);
754 set_state_private(&root->fs_info->extent_ins,
755 bytenr, (unsigned long)extent_op);
760 path = btrfs_alloc_path();
763 ret = lookup_extent_backref(trans, extent_root, path,
764 bytenr, orig_parent, orig_root,
768 ret = remove_extent_backref(trans, extent_root, path);
771 ret = insert_extent_backref(trans, extent_root, path, bytenr,
772 parent, ref_root, ref_generation,
773 owner_objectid, owner_offset);
775 finish_current_insert(trans, extent_root);
776 del_pending_extents(trans, extent_root);
778 btrfs_free_path(path);
782 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
783 struct btrfs_root *root, u64 bytenr,
784 u64 orig_parent, u64 parent,
785 u64 ref_root, u64 ref_generation,
786 u64 owner_objectid, u64 owner_offset)
789 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
790 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
792 maybe_lock_mutex(root);
793 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
794 parent, ref_root, ref_root,
795 ref_generation, ref_generation,
796 owner_objectid, owner_offset);
797 maybe_unlock_mutex(root);
801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
802 struct btrfs_root *root, u64 bytenr,
803 u64 orig_parent, u64 parent,
804 u64 orig_root, u64 ref_root,
805 u64 orig_generation, u64 ref_generation,
806 u64 owner_objectid, u64 owner_offset)
808 struct btrfs_path *path;
810 struct btrfs_key key;
811 struct extent_buffer *l;
812 struct btrfs_extent_item *item;
815 path = btrfs_alloc_path();
820 key.objectid = bytenr;
821 key.type = BTRFS_EXTENT_ITEM_KEY;
822 key.offset = (u64)-1;
824 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
828 BUG_ON(ret == 0 || path->slots[0] == 0);
833 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
834 BUG_ON(key.objectid != bytenr);
835 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
837 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
838 refs = btrfs_extent_refs(l, item);
839 btrfs_set_extent_refs(l, item, refs + 1);
840 btrfs_mark_buffer_dirty(path->nodes[0]);
842 btrfs_release_path(root->fs_info->extent_root, path);
845 ret = insert_extent_backref(trans, root->fs_info->extent_root,
846 path, bytenr, parent,
847 ref_root, ref_generation,
848 owner_objectid, owner_offset);
850 finish_current_insert(trans, root->fs_info->extent_root);
851 del_pending_extents(trans, root->fs_info->extent_root);
853 btrfs_free_path(path);
857 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
858 struct btrfs_root *root,
859 u64 bytenr, u64 num_bytes, u64 parent,
860 u64 ref_root, u64 ref_generation,
861 u64 owner_objectid, u64 owner_offset)
864 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
865 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
867 maybe_lock_mutex(root);
868 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
869 0, ref_root, 0, ref_generation,
870 owner_objectid, owner_offset);
871 maybe_unlock_mutex(root);
875 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
876 struct btrfs_root *root)
878 finish_current_insert(trans, root->fs_info->extent_root);
879 del_pending_extents(trans, root->fs_info->extent_root);
883 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
884 struct btrfs_root *root, u64 bytenr,
885 u64 num_bytes, u32 *refs)
887 struct btrfs_path *path;
889 struct btrfs_key key;
890 struct extent_buffer *l;
891 struct btrfs_extent_item *item;
893 WARN_ON(num_bytes < root->sectorsize);
894 path = btrfs_alloc_path();
896 key.objectid = bytenr;
897 key.offset = num_bytes;
898 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
899 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
904 btrfs_print_leaf(root, path->nodes[0]);
905 printk("failed to find block number %Lu\n", bytenr);
909 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
910 *refs = btrfs_extent_refs(l, item);
912 btrfs_free_path(path);
916 static int get_reference_status(struct btrfs_root *root, u64 bytenr,
917 u64 parent_gen, u64 ref_objectid,
918 u64 *min_generation, u32 *ref_count)
920 struct btrfs_root *extent_root = root->fs_info->extent_root;
921 struct btrfs_path *path;
922 struct extent_buffer *leaf;
923 struct btrfs_extent_ref *ref_item;
924 struct btrfs_key key;
925 struct btrfs_key found_key;
926 u64 root_objectid = root->root_key.objectid;
931 key.objectid = bytenr;
932 key.offset = (u64)-1;
933 key.type = BTRFS_EXTENT_ITEM_KEY;
935 path = btrfs_alloc_path();
936 mutex_lock(&root->fs_info->alloc_mutex);
937 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
941 if (ret < 0 || path->slots[0] == 0)
945 leaf = path->nodes[0];
946 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
948 if (found_key.objectid != bytenr ||
949 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
955 *min_generation = (u64)-1;
958 leaf = path->nodes[0];
959 nritems = btrfs_header_nritems(leaf);
960 if (path->slots[0] >= nritems) {
961 ret = btrfs_next_leaf(extent_root, path);
968 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
969 if (found_key.objectid != bytenr)
972 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
977 ref_item = btrfs_item_ptr(leaf, path->slots[0],
978 struct btrfs_extent_ref);
979 ref_generation = btrfs_ref_generation(leaf, ref_item);
981 * For (parent_gen > 0 && parent_gen > ref_generation):
983 * we reach here through the oldest root, therefore
984 * all other reference from same snapshot should have
985 * a larger generation.
987 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
988 (parent_gen > 0 && parent_gen > ref_generation) ||
989 (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
990 ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
996 if (*min_generation > ref_generation)
997 *min_generation = ref_generation;
1003 mutex_unlock(&root->fs_info->alloc_mutex);
1004 btrfs_free_path(path);
1008 int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
1009 struct btrfs_root *root,
1010 struct btrfs_key *key, u64 bytenr)
1012 struct btrfs_root *old_root;
1013 struct btrfs_path *path = NULL;
1014 struct extent_buffer *eb;
1015 struct btrfs_file_extent_item *item;
1023 BUG_ON(trans == NULL);
1024 BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
1025 ret = get_reference_status(root, bytenr, 0, key->objectid,
1026 &min_generation, &ref_count);
1033 old_root = root->dirty_root->root;
1034 ref_generation = old_root->root_key.offset;
1036 /* all references are created in running transaction */
1037 if (min_generation > ref_generation) {
1042 path = btrfs_alloc_path();
1048 path->skip_locking = 1;
1049 /* if no item found, the extent is referenced by other snapshot */
1050 ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
1054 eb = path->nodes[0];
1055 item = btrfs_item_ptr(eb, path->slots[0],
1056 struct btrfs_file_extent_item);
1057 if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
1058 btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
1063 for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
1065 eb = path->nodes[level];
1068 extent_start = eb->start;
1070 extent_start = bytenr;
1072 ret = get_reference_status(root, extent_start, ref_generation,
1073 0, &min_generation, &ref_count);
1077 if (ref_count != 1) {
1082 ref_generation = btrfs_header_generation(eb);
1087 btrfs_free_path(path);
1091 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1092 struct extent_buffer *buf, u32 nr_extents)
1094 struct btrfs_key key;
1095 struct btrfs_file_extent_item *fi;
1103 if (!root->ref_cows)
1106 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1108 root_gen = root->root_key.offset;
1111 root_gen = trans->transid - 1;
1114 level = btrfs_header_level(buf);
1115 nritems = btrfs_header_nritems(buf);
1118 struct btrfs_leaf_ref *ref;
1119 struct btrfs_extent_info *info;
1121 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1127 ref->root_gen = root_gen;
1128 ref->bytenr = buf->start;
1129 ref->owner = btrfs_header_owner(buf);
1130 ref->generation = btrfs_header_generation(buf);
1131 ref->nritems = nr_extents;
1132 info = ref->extents;
1134 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1136 btrfs_item_key_to_cpu(buf, &key, i);
1137 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1139 fi = btrfs_item_ptr(buf, i,
1140 struct btrfs_file_extent_item);
1141 if (btrfs_file_extent_type(buf, fi) ==
1142 BTRFS_FILE_EXTENT_INLINE)
1144 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1145 if (disk_bytenr == 0)
1148 info->bytenr = disk_bytenr;
1150 btrfs_file_extent_disk_num_bytes(buf, fi);
1151 info->objectid = key.objectid;
1152 info->offset = key.offset;
1156 ret = btrfs_add_leaf_ref(root, ref, shared);
1158 btrfs_free_leaf_ref(root, ref);
1164 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1165 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1172 u64 orig_generation;
1174 u32 nr_file_extents = 0;
1175 struct btrfs_key key;
1176 struct btrfs_file_extent_item *fi;
1181 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1182 u64, u64, u64, u64, u64, u64, u64, u64, u64);
1184 ref_root = btrfs_header_owner(buf);
1185 ref_generation = btrfs_header_generation(buf);
1186 orig_root = btrfs_header_owner(orig_buf);
1187 orig_generation = btrfs_header_generation(orig_buf);
1189 nritems = btrfs_header_nritems(buf);
1190 level = btrfs_header_level(buf);
1192 if (root->ref_cows) {
1193 process_func = __btrfs_inc_extent_ref;
1196 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1199 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1201 process_func = __btrfs_update_extent_ref;
1204 for (i = 0; i < nritems; i++) {
1207 btrfs_item_key_to_cpu(buf, &key, i);
1208 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1210 fi = btrfs_item_ptr(buf, i,
1211 struct btrfs_file_extent_item);
1212 if (btrfs_file_extent_type(buf, fi) ==
1213 BTRFS_FILE_EXTENT_INLINE)
1215 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1221 maybe_lock_mutex(root);
1222 ret = process_func(trans, root, bytenr,
1223 orig_buf->start, buf->start,
1224 orig_root, ref_root,
1225 orig_generation, ref_generation,
1226 key.objectid, key.offset);
1227 maybe_unlock_mutex(root);
1235 bytenr = btrfs_node_blockptr(buf, i);
1236 maybe_lock_mutex(root);
1237 ret = process_func(trans, root, bytenr,
1238 orig_buf->start, buf->start,
1239 orig_root, ref_root,
1240 orig_generation, ref_generation,
1242 maybe_unlock_mutex(root);
1253 *nr_extents = nr_file_extents;
1255 *nr_extents = nritems;
1263 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1264 struct btrfs_root *root, struct extent_buffer *orig_buf,
1265 struct extent_buffer *buf, int start_slot, int nr)
1272 u64 orig_generation;
1273 struct btrfs_key key;
1274 struct btrfs_file_extent_item *fi;
1280 BUG_ON(start_slot < 0);
1281 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1283 ref_root = btrfs_header_owner(buf);
1284 ref_generation = btrfs_header_generation(buf);
1285 orig_root = btrfs_header_owner(orig_buf);
1286 orig_generation = btrfs_header_generation(orig_buf);
1287 level = btrfs_header_level(buf);
1289 if (!root->ref_cows) {
1291 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1294 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1298 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1301 btrfs_item_key_to_cpu(buf, &key, slot);
1302 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1304 fi = btrfs_item_ptr(buf, slot,
1305 struct btrfs_file_extent_item);
1306 if (btrfs_file_extent_type(buf, fi) ==
1307 BTRFS_FILE_EXTENT_INLINE)
1309 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1312 maybe_lock_mutex(root);
1313 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1314 orig_buf->start, buf->start,
1315 orig_root, ref_root,
1316 orig_generation, ref_generation,
1317 key.objectid, key.offset);
1318 maybe_unlock_mutex(root);
1322 bytenr = btrfs_node_blockptr(buf, slot);
1323 maybe_lock_mutex(root);
1324 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1325 orig_buf->start, buf->start,
1326 orig_root, ref_root,
1327 orig_generation, ref_generation,
1329 maybe_unlock_mutex(root);
1340 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root,
1342 struct btrfs_path *path,
1343 struct btrfs_block_group_cache *cache)
1347 struct btrfs_root *extent_root = root->fs_info->extent_root;
1349 struct extent_buffer *leaf;
1351 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1356 leaf = path->nodes[0];
1357 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1358 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1359 btrfs_mark_buffer_dirty(leaf);
1360 btrfs_release_path(extent_root, path);
1362 finish_current_insert(trans, extent_root);
1363 pending_ret = del_pending_extents(trans, extent_root);
1372 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1373 struct btrfs_root *root)
1375 struct btrfs_block_group_cache *cache, *entry;
1379 struct btrfs_path *path;
1382 path = btrfs_alloc_path();
1386 mutex_lock(&root->fs_info->alloc_mutex);
1389 spin_lock(&root->fs_info->block_group_cache_lock);
1390 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1391 n; n = rb_next(n)) {
1392 entry = rb_entry(n, struct btrfs_block_group_cache,
1399 spin_unlock(&root->fs_info->block_group_cache_lock);
1405 last += cache->key.offset;
1407 err = write_one_cache_group(trans, root,
1410 * if we fail to write the cache group, we want
1411 * to keep it marked dirty in hopes that a later
1419 btrfs_free_path(path);
1420 mutex_unlock(&root->fs_info->alloc_mutex);
1424 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1425 u64 total_bytes, u64 bytes_used,
1426 struct btrfs_space_info **space_info)
1428 struct btrfs_space_info *found;
1430 found = __find_space_info(info, flags);
1432 found->total_bytes += total_bytes;
1433 found->bytes_used += bytes_used;
1435 *space_info = found;
1438 found = kmalloc(sizeof(*found), GFP_NOFS);
1442 list_add(&found->list, &info->space_info);
1443 INIT_LIST_HEAD(&found->block_groups);
1444 spin_lock_init(&found->lock);
1445 found->flags = flags;
1446 found->total_bytes = total_bytes;
1447 found->bytes_used = bytes_used;
1448 found->bytes_pinned = 0;
1449 found->bytes_reserved = 0;
1451 found->force_alloc = 0;
1452 *space_info = found;
1456 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1458 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1459 BTRFS_BLOCK_GROUP_RAID1 |
1460 BTRFS_BLOCK_GROUP_RAID10 |
1461 BTRFS_BLOCK_GROUP_DUP);
1463 if (flags & BTRFS_BLOCK_GROUP_DATA)
1464 fs_info->avail_data_alloc_bits |= extra_flags;
1465 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1466 fs_info->avail_metadata_alloc_bits |= extra_flags;
1467 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1468 fs_info->avail_system_alloc_bits |= extra_flags;
1472 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1474 u64 num_devices = root->fs_info->fs_devices->num_devices;
1476 if (num_devices == 1)
1477 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1478 if (num_devices < 4)
1479 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1481 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1482 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1483 BTRFS_BLOCK_GROUP_RAID10))) {
1484 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1487 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1488 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1489 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1492 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1493 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1494 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1495 (flags & BTRFS_BLOCK_GROUP_DUP)))
1496 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1500 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1501 struct btrfs_root *extent_root, u64 alloc_bytes,
1502 u64 flags, int force)
1504 struct btrfs_space_info *space_info;
1508 int ret = 0, waited = 0;
1510 flags = reduce_alloc_profile(extent_root, flags);
1512 space_info = __find_space_info(extent_root->fs_info, flags);
1514 ret = update_space_info(extent_root->fs_info, flags,
1518 BUG_ON(!space_info);
1520 if (space_info->force_alloc) {
1522 space_info->force_alloc = 0;
1524 if (space_info->full)
1527 thresh = div_factor(space_info->total_bytes, 6);
1529 (space_info->bytes_used + space_info->bytes_pinned +
1530 space_info->bytes_reserved + alloc_bytes) < thresh)
1533 while (!mutex_trylock(&extent_root->fs_info->chunk_mutex)) {
1536 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1538 mutex_lock(&extent_root->fs_info->alloc_mutex);
1542 if (waited && space_info->full)
1545 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1546 if (ret == -ENOSPC) {
1547 printk("space info full %Lu\n", flags);
1548 space_info->full = 1;
1553 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1554 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1558 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1563 static int update_block_group(struct btrfs_trans_handle *trans,
1564 struct btrfs_root *root,
1565 u64 bytenr, u64 num_bytes, int alloc,
1568 struct btrfs_block_group_cache *cache;
1569 struct btrfs_fs_info *info = root->fs_info;
1570 u64 total = num_bytes;
1574 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1576 cache = btrfs_lookup_block_group(info, bytenr);
1580 byte_in_group = bytenr - cache->key.objectid;
1581 WARN_ON(byte_in_group > cache->key.offset);
1583 spin_lock(&cache->lock);
1585 old_val = btrfs_block_group_used(&cache->item);
1586 num_bytes = min(total, cache->key.offset - byte_in_group);
1588 old_val += num_bytes;
1589 cache->space_info->bytes_used += num_bytes;
1590 btrfs_set_block_group_used(&cache->item, old_val);
1591 spin_unlock(&cache->lock);
1593 old_val -= num_bytes;
1594 cache->space_info->bytes_used -= num_bytes;
1595 btrfs_set_block_group_used(&cache->item, old_val);
1596 spin_unlock(&cache->lock);
1599 ret = btrfs_add_free_space(cache, bytenr,
1606 bytenr += num_bytes;
1611 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1613 struct btrfs_block_group_cache *cache;
1615 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1619 return cache->key.objectid;
1622 int btrfs_update_pinned_extents(struct btrfs_root *root,
1623 u64 bytenr, u64 num, int pin)
1626 struct btrfs_block_group_cache *cache;
1627 struct btrfs_fs_info *fs_info = root->fs_info;
1629 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1631 set_extent_dirty(&fs_info->pinned_extents,
1632 bytenr, bytenr + num - 1, GFP_NOFS);
1634 clear_extent_dirty(&fs_info->pinned_extents,
1635 bytenr, bytenr + num - 1, GFP_NOFS);
1638 cache = btrfs_lookup_block_group(fs_info, bytenr);
1640 len = min(num, cache->key.offset -
1641 (bytenr - cache->key.objectid));
1643 spin_lock(&cache->lock);
1644 cache->pinned += len;
1645 cache->space_info->bytes_pinned += len;
1646 spin_unlock(&cache->lock);
1647 fs_info->total_pinned += len;
1649 spin_lock(&cache->lock);
1650 cache->pinned -= len;
1651 cache->space_info->bytes_pinned -= len;
1652 spin_unlock(&cache->lock);
1653 fs_info->total_pinned -= len;
1661 static int update_reserved_extents(struct btrfs_root *root,
1662 u64 bytenr, u64 num, int reserve)
1665 struct btrfs_block_group_cache *cache;
1666 struct btrfs_fs_info *fs_info = root->fs_info;
1668 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1670 cache = btrfs_lookup_block_group(fs_info, bytenr);
1672 len = min(num, cache->key.offset -
1673 (bytenr - cache->key.objectid));
1675 spin_lock(&cache->lock);
1676 cache->reserved += len;
1677 cache->space_info->bytes_reserved += len;
1678 spin_unlock(&cache->lock);
1680 spin_lock(&cache->lock);
1681 cache->reserved -= len;
1682 cache->space_info->bytes_reserved -= len;
1683 spin_unlock(&cache->lock);
1691 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1696 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1700 ret = find_first_extent_bit(pinned_extents, last,
1701 &start, &end, EXTENT_DIRTY);
1704 set_extent_dirty(copy, start, end, GFP_NOFS);
1710 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1711 struct btrfs_root *root,
1712 struct extent_io_tree *unpin)
1717 struct btrfs_block_group_cache *cache;
1719 mutex_lock(&root->fs_info->alloc_mutex);
1721 ret = find_first_extent_bit(unpin, 0, &start, &end,
1725 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1726 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1727 cache = btrfs_lookup_block_group(root->fs_info, start);
1729 btrfs_add_free_space(cache, start, end - start + 1);
1730 if (need_resched()) {
1731 mutex_unlock(&root->fs_info->alloc_mutex);
1733 mutex_lock(&root->fs_info->alloc_mutex);
1736 mutex_unlock(&root->fs_info->alloc_mutex);
1740 static int finish_current_insert(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *extent_root)
1746 struct btrfs_fs_info *info = extent_root->fs_info;
1747 struct btrfs_path *path;
1748 struct btrfs_extent_ref *ref;
1749 struct pending_extent_op *extent_op;
1750 struct btrfs_key key;
1751 struct btrfs_extent_item extent_item;
1755 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1756 btrfs_set_stack_extent_refs(&extent_item, 1);
1757 path = btrfs_alloc_path();
1760 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1761 &end, EXTENT_LOCKED);
1765 ret = get_state_private(&info->extent_ins, start, &priv);
1767 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1769 if (extent_op->type == PENDING_EXTENT_INSERT) {
1770 key.objectid = start;
1771 key.offset = end + 1 - start;
1772 key.type = BTRFS_EXTENT_ITEM_KEY;
1773 err = btrfs_insert_item(trans, extent_root, &key,
1774 &extent_item, sizeof(extent_item));
1777 clear_extent_bits(&info->extent_ins, start, end,
1778 EXTENT_LOCKED, GFP_NOFS);
1780 err = insert_extent_backref(trans, extent_root, path,
1781 start, extent_op->parent,
1782 extent_root->root_key.objectid,
1783 extent_op->generation,
1784 extent_op->level, 0);
1786 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1787 err = lookup_extent_backref(trans, extent_root, path,
1788 start, extent_op->orig_parent,
1789 extent_root->root_key.objectid,
1790 extent_op->orig_generation, 0);
1793 clear_extent_bits(&info->extent_ins, start, end,
1794 EXTENT_LOCKED, GFP_NOFS);
1796 key.objectid = start;
1797 key.offset = extent_op->parent;
1798 key.type = BTRFS_EXTENT_REF_KEY;
1799 err = btrfs_set_item_key_safe(trans, extent_root, path,
1802 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1803 struct btrfs_extent_ref);
1804 btrfs_set_ref_generation(path->nodes[0], ref,
1805 extent_op->generation);
1806 btrfs_mark_buffer_dirty(path->nodes[0]);
1807 btrfs_release_path(extent_root, path);
1813 if (need_resched()) {
1814 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1816 mutex_lock(&extent_root->fs_info->alloc_mutex);
1819 btrfs_free_path(path);
1823 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root,
1825 u64 bytenr, u64 num_bytes, int is_data)
1828 struct extent_buffer *buf;
1830 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1834 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1838 /* we can reuse a block if it hasn't been written
1839 * and it is from this transaction. We can't
1840 * reuse anything from the tree log root because
1841 * it has tiny sub-transactions.
1843 if (btrfs_buffer_uptodate(buf, 0) &&
1844 btrfs_try_tree_lock(buf)) {
1845 u64 header_owner = btrfs_header_owner(buf);
1846 u64 header_transid = btrfs_header_generation(buf);
1847 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1848 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1849 header_transid == trans->transid &&
1850 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1851 clean_tree_block(NULL, root, buf);
1852 btrfs_tree_unlock(buf);
1853 free_extent_buffer(buf);
1856 btrfs_tree_unlock(buf);
1858 free_extent_buffer(buf);
1860 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1867 * remove an extent from the root, returns 0 on success
1869 static int __free_extent(struct btrfs_trans_handle *trans,
1870 struct btrfs_root *root,
1871 u64 bytenr, u64 num_bytes, u64 parent,
1872 u64 root_objectid, u64 ref_generation,
1873 u64 owner_objectid, u64 owner_offset,
1874 int pin, int mark_free)
1876 struct btrfs_path *path;
1877 struct btrfs_key key;
1878 struct btrfs_fs_info *info = root->fs_info;
1879 struct btrfs_root *extent_root = info->extent_root;
1880 struct extent_buffer *leaf;
1882 int extent_slot = 0;
1883 int found_extent = 0;
1885 struct btrfs_extent_item *ei;
1888 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1889 key.objectid = bytenr;
1890 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1891 key.offset = num_bytes;
1892 path = btrfs_alloc_path();
1897 ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent,
1898 root_objectid, ref_generation, 1);
1900 struct btrfs_key found_key;
1901 extent_slot = path->slots[0];
1902 while(extent_slot > 0) {
1904 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1906 if (found_key.objectid != bytenr)
1908 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1909 found_key.offset == num_bytes) {
1913 if (path->slots[0] - extent_slot > 5)
1916 if (!found_extent) {
1917 ret = remove_extent_backref(trans, extent_root, path);
1919 btrfs_release_path(extent_root, path);
1920 ret = btrfs_search_slot(trans, extent_root,
1923 extent_slot = path->slots[0];
1926 btrfs_print_leaf(extent_root, path->nodes[0]);
1928 printk("Unable to find ref byte nr %Lu root %Lu "
1929 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1930 root_objectid, ref_generation, owner_objectid,
1934 leaf = path->nodes[0];
1935 ei = btrfs_item_ptr(leaf, extent_slot,
1936 struct btrfs_extent_item);
1937 refs = btrfs_extent_refs(leaf, ei);
1940 btrfs_set_extent_refs(leaf, ei, refs);
1942 btrfs_mark_buffer_dirty(leaf);
1944 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1945 struct btrfs_extent_ref *ref;
1946 ref = btrfs_item_ptr(leaf, path->slots[0],
1947 struct btrfs_extent_ref);
1948 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1949 /* if the back ref and the extent are next to each other
1950 * they get deleted below in one shot
1952 path->slots[0] = extent_slot;
1954 } else if (found_extent) {
1955 /* otherwise delete the extent back ref */
1956 ret = remove_extent_backref(trans, extent_root, path);
1958 /* if refs are 0, we need to setup the path for deletion */
1960 btrfs_release_path(extent_root, path);
1961 ret = btrfs_search_slot(trans, extent_root, &key, path,
1970 #ifdef BIO_RW_DISCARD
1971 u64 map_length = num_bytes;
1972 struct btrfs_multi_bio *multi = NULL;
1976 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1977 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1983 /* block accounting for super block */
1984 spin_lock_irq(&info->delalloc_lock);
1985 super_used = btrfs_super_bytes_used(&info->super_copy);
1986 btrfs_set_super_bytes_used(&info->super_copy,
1987 super_used - num_bytes);
1988 spin_unlock_irq(&info->delalloc_lock);
1990 /* block accounting for root item */
1991 root_used = btrfs_root_used(&root->root_item);
1992 btrfs_set_root_used(&root->root_item,
1993 root_used - num_bytes);
1994 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1997 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2001 #ifdef BIO_RW_DISCARD
2002 /* Tell the block device(s) that the sectors can be discarded */
2003 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2004 bytenr, &map_length, &multi, 0);
2006 struct btrfs_bio_stripe *stripe = multi->stripes;
2009 if (map_length > num_bytes)
2010 map_length = num_bytes;
2012 for (i = 0; i < multi->num_stripes; i++, stripe++) {
2013 blkdev_issue_discard(stripe->dev->bdev,
2014 stripe->physical >> 9,
2021 btrfs_free_path(path);
2022 finish_current_insert(trans, extent_root);
2027 * find all the blocks marked as pending in the radix tree and remove
2028 * them from the extent map
2030 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2031 btrfs_root *extent_root)
2039 struct extent_io_tree *pending_del;
2040 struct extent_io_tree *extent_ins;
2041 struct pending_extent_op *extent_op;
2043 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
2044 extent_ins = &extent_root->fs_info->extent_ins;
2045 pending_del = &extent_root->fs_info->pending_del;
2048 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2053 ret = get_state_private(pending_del, start, &priv);
2055 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2057 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2060 ret = pin_down_bytes(trans, extent_root, start,
2061 end + 1 - start, 0);
2062 mark_free = ret > 0;
2063 if (!test_range_bit(extent_ins, start, end,
2064 EXTENT_LOCKED, 0)) {
2066 ret = __free_extent(trans, extent_root,
2067 start, end + 1 - start,
2068 extent_op->orig_parent,
2069 extent_root->root_key.objectid,
2070 extent_op->orig_generation,
2071 extent_op->level, 0, 0, mark_free);
2075 ret = get_state_private(extent_ins, start, &priv);
2077 extent_op = (struct pending_extent_op *)
2078 (unsigned long)priv;
2080 clear_extent_bits(extent_ins, start, end,
2081 EXTENT_LOCKED, GFP_NOFS);
2083 if (extent_op->type == PENDING_BACKREF_UPDATE)
2086 ret = update_block_group(trans, extent_root, start,
2087 end + 1 - start, 0, mark_free);
2094 if (need_resched()) {
2095 mutex_unlock(&extent_root->fs_info->alloc_mutex);
2097 mutex_lock(&extent_root->fs_info->alloc_mutex);
2104 * remove an extent from the root, returns 0 on success
2106 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2107 struct btrfs_root *root,
2108 u64 bytenr, u64 num_bytes, u64 parent,
2109 u64 root_objectid, u64 ref_generation,
2110 u64 owner_objectid, u64 owner_offset, int pin)
2112 struct btrfs_root *extent_root = root->fs_info->extent_root;
2116 WARN_ON(num_bytes < root->sectorsize);
2117 if (root == extent_root) {
2118 struct pending_extent_op *extent_op;
2120 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2123 extent_op->type = PENDING_EXTENT_DELETE;
2124 extent_op->bytenr = bytenr;
2125 extent_op->num_bytes = num_bytes;
2126 extent_op->parent = parent;
2127 extent_op->orig_parent = parent;
2128 extent_op->generation = ref_generation;
2129 extent_op->orig_generation = ref_generation;
2130 extent_op->level = (int)owner_objectid;
2132 set_extent_bits(&root->fs_info->pending_del,
2133 bytenr, bytenr + num_bytes - 1,
2134 EXTENT_LOCKED, GFP_NOFS);
2135 set_state_private(&root->fs_info->pending_del,
2136 bytenr, (unsigned long)extent_op);
2139 /* if metadata always pin */
2140 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2141 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2142 struct btrfs_block_group_cache *cache;
2144 /* btrfs_free_reserved_extent */
2145 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2147 btrfs_add_free_space(cache, bytenr, num_bytes);
2148 update_reserved_extents(root, bytenr, num_bytes, 0);
2154 /* if data pin when any transaction has committed this */
2155 if (ref_generation != trans->transid)
2158 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2159 root_objectid, ref_generation, owner_objectid,
2160 owner_offset, pin, pin == 0);
2162 finish_current_insert(trans, root->fs_info->extent_root);
2163 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2164 return ret ? ret : pending_ret;
2167 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root,
2169 u64 bytenr, u64 num_bytes, u64 parent,
2170 u64 root_objectid, u64 ref_generation,
2171 u64 owner_objectid, u64 owner_offset, int pin)
2175 maybe_lock_mutex(root);
2176 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2177 root_objectid, ref_generation,
2178 owner_objectid, owner_offset, pin);
2179 maybe_unlock_mutex(root);
2183 static u64 stripe_align(struct btrfs_root *root, u64 val)
2185 u64 mask = ((u64)root->stripesize - 1);
2186 u64 ret = (val + mask) & ~mask;
2191 * walks the btree of allocated extents and find a hole of a given size.
2192 * The key ins is changed to record the hole:
2193 * ins->objectid == block start
2194 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2195 * ins->offset == number of blocks
2196 * Any available blocks before search_start are skipped.
2198 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *orig_root,
2200 u64 num_bytes, u64 empty_size,
2201 u64 search_start, u64 search_end,
2202 u64 hint_byte, struct btrfs_key *ins,
2203 u64 exclude_start, u64 exclude_nr,
2207 u64 orig_search_start;
2208 struct btrfs_root * root = orig_root->fs_info->extent_root;
2209 struct btrfs_fs_info *info = root->fs_info;
2210 u64 total_needed = num_bytes;
2211 u64 *last_ptr = NULL;
2212 struct btrfs_block_group_cache *block_group;
2213 int chunk_alloc_done = 0;
2214 int empty_cluster = 2 * 1024 * 1024;
2215 int allowed_chunk_alloc = 0;
2217 WARN_ON(num_bytes < root->sectorsize);
2218 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2220 if (orig_root->ref_cows || empty_size)
2221 allowed_chunk_alloc = 1;
2223 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2224 last_ptr = &root->fs_info->last_alloc;
2225 empty_cluster = 256 * 1024;
2228 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2229 last_ptr = &root->fs_info->last_data_alloc;
2233 hint_byte = *last_ptr;
2235 empty_size += empty_cluster;
2238 search_start = max(search_start, first_logical_byte(root, 0));
2239 orig_search_start = search_start;
2241 search_start = max(search_start, hint_byte);
2242 total_needed += empty_size;
2245 block_group = btrfs_lookup_block_group(info, search_start);
2247 block_group = btrfs_lookup_first_block_group(info,
2251 * Ok this looks a little tricky, buts its really simple. First if we
2252 * didn't find a block group obviously we want to start over.
2253 * Secondly, if the block group we found does not match the type we
2254 * need, and we have a last_ptr and its not 0, chances are the last
2255 * allocation we made was at the end of the block group, so lets go
2256 * ahead and skip the looking through the rest of the block groups and
2257 * start at the beginning. This helps with metadata allocations,
2258 * since you are likely to have a bunch of data block groups to search
2259 * through first before you realize that you need to start over, so go
2260 * ahead and start over and save the time.
2262 if (!block_group || (!block_group_bits(block_group, data) &&
2263 last_ptr && *last_ptr)) {
2264 if (search_start != orig_search_start) {
2265 if (last_ptr && *last_ptr) {
2266 total_needed += empty_cluster;
2269 search_start = orig_search_start;
2271 } else if (!chunk_alloc_done && allowed_chunk_alloc) {
2272 ret = do_chunk_alloc(trans, root,
2273 num_bytes + 2 * 1024 * 1024,
2278 chunk_alloc_done = 1;
2279 search_start = orig_search_start;
2288 * this is going to seach through all of the existing block groups it
2289 * can find, so if we don't find something we need to see if we can
2290 * allocate what we need.
2292 ret = find_free_space(root, &block_group, &search_start,
2293 total_needed, data);
2294 if (ret == -ENOSPC) {
2296 * instead of allocating, start at the original search start
2297 * and see if there is something to be found, if not then we
2300 if (search_start != orig_search_start) {
2301 if (last_ptr && *last_ptr) {
2303 total_needed += empty_cluster;
2305 search_start = orig_search_start;
2310 * we've already allocated, we're pretty screwed
2312 if (chunk_alloc_done) {
2314 } else if (!allowed_chunk_alloc && block_group &&
2315 block_group_bits(block_group, data)) {
2316 block_group->space_info->force_alloc = 1;
2318 } else if (!allowed_chunk_alloc) {
2322 ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
2328 chunk_alloc_done = 1;
2330 search_start = block_group->key.objectid +
2331 block_group->key.offset;
2333 search_start = orig_search_start;
2340 search_start = stripe_align(root, search_start);
2341 ins->objectid = search_start;
2342 ins->offset = num_bytes;
2344 if (ins->objectid + num_bytes >= search_end) {
2345 search_start = orig_search_start;
2346 if (chunk_alloc_done) {
2353 if (ins->objectid + num_bytes >
2354 block_group->key.objectid + block_group->key.offset) {
2355 if (search_start == orig_search_start && chunk_alloc_done) {
2359 search_start = block_group->key.objectid +
2360 block_group->key.offset;
2364 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2365 ins->objectid < exclude_start + exclude_nr)) {
2366 search_start = exclude_start + exclude_nr;
2370 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2371 trans->block_group = block_group;
2373 ins->offset = num_bytes;
2375 *last_ptr = ins->objectid + ins->offset;
2377 btrfs_super_total_bytes(&root->fs_info->super_copy))
2386 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2388 struct btrfs_block_group_cache *cache;
2389 struct list_head *l;
2391 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2392 info->total_bytes - info->bytes_used - info->bytes_pinned -
2393 info->bytes_reserved, (info->full) ? "" : "not ");
2395 spin_lock(&info->lock);
2396 list_for_each(l, &info->block_groups) {
2397 cache = list_entry(l, struct btrfs_block_group_cache, list);
2398 spin_lock(&cache->lock);
2399 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2400 "%Lu pinned %Lu reserved\n",
2401 cache->key.objectid, cache->key.offset,
2402 btrfs_block_group_used(&cache->item),
2403 cache->pinned, cache->reserved);
2404 btrfs_dump_free_space(cache, bytes);
2405 spin_unlock(&cache->lock);
2407 spin_unlock(&info->lock);
2410 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2411 struct btrfs_root *root,
2412 u64 num_bytes, u64 min_alloc_size,
2413 u64 empty_size, u64 hint_byte,
2414 u64 search_end, struct btrfs_key *ins,
2418 u64 search_start = 0;
2420 struct btrfs_fs_info *info = root->fs_info;
2421 struct btrfs_block_group_cache *cache;
2424 alloc_profile = info->avail_data_alloc_bits &
2425 info->data_alloc_profile;
2426 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2427 } else if (root == root->fs_info->chunk_root) {
2428 alloc_profile = info->avail_system_alloc_bits &
2429 info->system_alloc_profile;
2430 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2432 alloc_profile = info->avail_metadata_alloc_bits &
2433 info->metadata_alloc_profile;
2434 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2437 data = reduce_alloc_profile(root, data);
2439 * the only place that sets empty_size is btrfs_realloc_node, which
2440 * is not called recursively on allocations
2442 if (empty_size || root->ref_cows) {
2443 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2444 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2446 BTRFS_BLOCK_GROUP_METADATA |
2447 (info->metadata_alloc_profile &
2448 info->avail_metadata_alloc_bits), 0);
2450 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2451 num_bytes + 2 * 1024 * 1024, data, 0);
2454 WARN_ON(num_bytes < root->sectorsize);
2455 ret = find_free_extent(trans, root, num_bytes, empty_size,
2456 search_start, search_end, hint_byte, ins,
2457 trans->alloc_exclude_start,
2458 trans->alloc_exclude_nr, data);
2460 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2461 num_bytes = num_bytes >> 1;
2462 num_bytes = num_bytes & ~(root->sectorsize - 1);
2463 num_bytes = max(num_bytes, min_alloc_size);
2464 do_chunk_alloc(trans, root->fs_info->extent_root,
2465 num_bytes, data, 1);
2469 struct btrfs_space_info *sinfo;
2471 sinfo = __find_space_info(root->fs_info, data);
2472 printk("allocation failed flags %Lu, wanted %Lu\n",
2474 dump_space_info(sinfo, num_bytes);
2477 cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2479 printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
2483 ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
2488 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2490 struct btrfs_block_group_cache *cache;
2492 maybe_lock_mutex(root);
2493 cache = btrfs_lookup_block_group(root->fs_info, start);
2495 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2496 maybe_unlock_mutex(root);
2499 btrfs_add_free_space(cache, start, len);
2500 update_reserved_extents(root, start, len, 0);
2501 maybe_unlock_mutex(root);
2505 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2506 struct btrfs_root *root,
2507 u64 num_bytes, u64 min_alloc_size,
2508 u64 empty_size, u64 hint_byte,
2509 u64 search_end, struct btrfs_key *ins,
2513 maybe_lock_mutex(root);
2514 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2515 empty_size, hint_byte, search_end, ins,
2517 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2518 maybe_unlock_mutex(root);
2522 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2523 struct btrfs_root *root, u64 parent,
2524 u64 root_objectid, u64 ref_generation,
2525 u64 owner, u64 owner_offset,
2526 struct btrfs_key *ins)
2532 u64 num_bytes = ins->offset;
2534 struct btrfs_fs_info *info = root->fs_info;
2535 struct btrfs_root *extent_root = info->extent_root;
2536 struct btrfs_extent_item *extent_item;
2537 struct btrfs_extent_ref *ref;
2538 struct btrfs_path *path;
2539 struct btrfs_key keys[2];
2542 parent = ins->objectid;
2544 /* block accounting for super block */
2545 spin_lock_irq(&info->delalloc_lock);
2546 super_used = btrfs_super_bytes_used(&info->super_copy);
2547 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2548 spin_unlock_irq(&info->delalloc_lock);
2550 /* block accounting for root item */
2551 root_used = btrfs_root_used(&root->root_item);
2552 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2554 if (root == extent_root) {
2555 struct pending_extent_op *extent_op;
2557 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2560 extent_op->type = PENDING_EXTENT_INSERT;
2561 extent_op->bytenr = ins->objectid;
2562 extent_op->num_bytes = ins->offset;
2563 extent_op->parent = parent;
2564 extent_op->orig_parent = 0;
2565 extent_op->generation = ref_generation;
2566 extent_op->orig_generation = 0;
2567 extent_op->level = (int)owner;
2569 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2570 ins->objectid + ins->offset - 1,
2571 EXTENT_LOCKED, GFP_NOFS);
2572 set_state_private(&root->fs_info->extent_ins,
2573 ins->objectid, (unsigned long)extent_op);
2577 memcpy(&keys[0], ins, sizeof(*ins));
2578 keys[1].objectid = ins->objectid;
2579 keys[1].type = BTRFS_EXTENT_REF_KEY;
2580 keys[1].offset = parent;
2581 sizes[0] = sizeof(*extent_item);
2582 sizes[1] = sizeof(*ref);
2584 path = btrfs_alloc_path();
2587 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2591 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2592 struct btrfs_extent_item);
2593 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2594 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2595 struct btrfs_extent_ref);
2597 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2598 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2599 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2600 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2601 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2603 btrfs_mark_buffer_dirty(path->nodes[0]);
2605 trans->alloc_exclude_start = 0;
2606 trans->alloc_exclude_nr = 0;
2607 btrfs_free_path(path);
2608 finish_current_insert(trans, extent_root);
2609 pending_ret = del_pending_extents(trans, extent_root);
2619 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2621 printk("update block group failed for %Lu %Lu\n",
2622 ins->objectid, ins->offset);
2629 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root, u64 parent,
2631 u64 root_objectid, u64 ref_generation,
2632 u64 owner, u64 owner_offset,
2633 struct btrfs_key *ins)
2637 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2639 maybe_lock_mutex(root);
2640 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2641 root_objectid, ref_generation,
2642 owner, owner_offset, ins);
2643 update_reserved_extents(root, ins->objectid, ins->offset, 0);
2644 maybe_unlock_mutex(root);
2649 * this is used by the tree logging recovery code. It records that
2650 * an extent has been allocated and makes sure to clear the free
2651 * space cache bits as well
2653 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2654 struct btrfs_root *root, u64 parent,
2655 u64 root_objectid, u64 ref_generation,
2656 u64 owner, u64 owner_offset,
2657 struct btrfs_key *ins)
2660 struct btrfs_block_group_cache *block_group;
2662 maybe_lock_mutex(root);
2663 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2664 cache_block_group(root, block_group);
2666 ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
2668 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2669 root_objectid, ref_generation,
2670 owner, owner_offset, ins);
2671 maybe_unlock_mutex(root);
2676 * finds a free extent and does all the dirty work required for allocation
2677 * returns the key for the extent through ins, and a tree buffer for
2678 * the first block of the extent through buf.
2680 * returns 0 if everything worked, non-zero otherwise.
2682 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root,
2684 u64 num_bytes, u64 parent, u64 min_alloc_size,
2685 u64 root_objectid, u64 ref_generation,
2686 u64 owner_objectid, u64 owner_offset,
2687 u64 empty_size, u64 hint_byte,
2688 u64 search_end, struct btrfs_key *ins, u64 data)
2692 maybe_lock_mutex(root);
2694 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2695 min_alloc_size, empty_size, hint_byte,
2696 search_end, ins, data);
2698 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2699 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2700 root_objectid, ref_generation,
2701 owner_objectid, owner_offset, ins);
2705 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2707 maybe_unlock_mutex(root);
2711 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2712 struct btrfs_root *root,
2713 u64 bytenr, u32 blocksize)
2715 struct extent_buffer *buf;
2717 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2719 return ERR_PTR(-ENOMEM);
2720 btrfs_set_header_generation(buf, trans->transid);
2721 btrfs_tree_lock(buf);
2722 clean_tree_block(trans, root, buf);
2723 btrfs_set_buffer_uptodate(buf);
2724 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2725 set_extent_dirty(&root->dirty_log_pages, buf->start,
2726 buf->start + buf->len - 1, GFP_NOFS);
2728 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2729 buf->start + buf->len - 1, GFP_NOFS);
2731 trans->blocks_used++;
2736 * helper function to allocate a block for a given tree
2737 * returns the tree buffer or NULL.
2739 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2740 struct btrfs_root *root,
2741 u32 blocksize, u64 parent,
2748 struct btrfs_key ins;
2750 struct extent_buffer *buf;
2752 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2753 root_objectid, ref_generation, level, 0,
2754 empty_size, hint, (u64)-1, &ins, 0);
2757 return ERR_PTR(ret);
2760 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2764 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root, struct extent_buffer *leaf)
2768 u64 leaf_generation;
2769 struct btrfs_key key;
2770 struct btrfs_file_extent_item *fi;
2775 BUG_ON(!btrfs_is_leaf(leaf));
2776 nritems = btrfs_header_nritems(leaf);
2777 leaf_owner = btrfs_header_owner(leaf);
2778 leaf_generation = btrfs_header_generation(leaf);
2780 for (i = 0; i < nritems; i++) {
2784 btrfs_item_key_to_cpu(leaf, &key, i);
2785 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2787 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2788 if (btrfs_file_extent_type(leaf, fi) ==
2789 BTRFS_FILE_EXTENT_INLINE)
2792 * FIXME make sure to insert a trans record that
2793 * repeats the snapshot del on crash
2795 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2796 if (disk_bytenr == 0)
2799 mutex_lock(&root->fs_info->alloc_mutex);
2800 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2801 btrfs_file_extent_disk_num_bytes(leaf, fi),
2802 leaf->start, leaf_owner, leaf_generation,
2803 key.objectid, key.offset, 0);
2804 mutex_unlock(&root->fs_info->alloc_mutex);
2807 atomic_inc(&root->fs_info->throttle_gen);
2808 wake_up(&root->fs_info->transaction_throttle);
2814 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2815 struct btrfs_root *root,
2816 struct btrfs_leaf_ref *ref)
2820 struct btrfs_extent_info *info = ref->extents;
2822 for (i = 0; i < ref->nritems; i++) {
2823 mutex_lock(&root->fs_info->alloc_mutex);
2824 ret = __btrfs_free_extent(trans, root, info->bytenr,
2825 info->num_bytes, ref->bytenr,
2826 ref->owner, ref->generation,
2827 info->objectid, info->offset, 0);
2828 mutex_unlock(&root->fs_info->alloc_mutex);
2830 atomic_inc(&root->fs_info->throttle_gen);
2831 wake_up(&root->fs_info->transaction_throttle);
2841 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2846 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2849 #if 0 // some debugging code in case we see problems here
2850 /* if the refs count is one, it won't get increased again. But
2851 * if the ref count is > 1, someone may be decreasing it at
2852 * the same time we are.
2855 struct extent_buffer *eb = NULL;
2856 eb = btrfs_find_create_tree_block(root, start, len);
2858 btrfs_tree_lock(eb);
2860 mutex_lock(&root->fs_info->alloc_mutex);
2861 ret = lookup_extent_ref(NULL, root, start, len, refs);
2863 mutex_unlock(&root->fs_info->alloc_mutex);
2866 btrfs_tree_unlock(eb);
2867 free_extent_buffer(eb);
2870 printk("block %llu went down to one during drop_snap\n",
2871 (unsigned long long)start);
2882 * helper function for drop_snapshot, this walks down the tree dropping ref
2883 * counts as it goes.
2885 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2886 struct btrfs_root *root,
2887 struct btrfs_path *path, int *level)
2893 struct extent_buffer *next;
2894 struct extent_buffer *cur;
2895 struct extent_buffer *parent;
2896 struct btrfs_leaf_ref *ref;
2901 WARN_ON(*level < 0);
2902 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2903 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2904 path->nodes[*level]->len, &refs);
2910 * walk down to the last node level and free all the leaves
2912 while(*level >= 0) {
2913 WARN_ON(*level < 0);
2914 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2915 cur = path->nodes[*level];
2917 if (btrfs_header_level(cur) != *level)
2920 if (path->slots[*level] >=
2921 btrfs_header_nritems(cur))
2924 ret = btrfs_drop_leaf_ref(trans, root, cur);
2928 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2929 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2930 blocksize = btrfs_level_size(root, *level - 1);
2932 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2935 parent = path->nodes[*level];
2936 root_owner = btrfs_header_owner(parent);
2937 root_gen = btrfs_header_generation(parent);
2938 path->slots[*level]++;
2940 mutex_lock(&root->fs_info->alloc_mutex);
2941 ret = __btrfs_free_extent(trans, root, bytenr,
2942 blocksize, parent->start,
2943 root_owner, root_gen, 0, 0, 1);
2945 mutex_unlock(&root->fs_info->alloc_mutex);
2947 atomic_inc(&root->fs_info->throttle_gen);
2948 wake_up(&root->fs_info->transaction_throttle);
2954 * at this point, we have a single ref, and since the
2955 * only place referencing this extent is a dead root
2956 * the reference count should never go higher.
2957 * So, we don't need to check it again
2960 ref = btrfs_lookup_leaf_ref(root, bytenr);
2961 if (ref && ref->generation != ptr_gen) {
2962 btrfs_free_leaf_ref(root, ref);
2966 ret = cache_drop_leaf_ref(trans, root, ref);
2968 btrfs_remove_leaf_ref(root, ref);
2969 btrfs_free_leaf_ref(root, ref);
2973 if (printk_ratelimit())
2974 printk("leaf ref miss for bytenr %llu\n",
2975 (unsigned long long)bytenr);
2977 next = btrfs_find_tree_block(root, bytenr, blocksize);
2978 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2979 free_extent_buffer(next);
2981 next = read_tree_block(root, bytenr, blocksize,
2986 * this is a debugging check and can go away
2987 * the ref should never go all the way down to 1
2990 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2996 WARN_ON(*level <= 0);
2997 if (path->nodes[*level-1])
2998 free_extent_buffer(path->nodes[*level-1]);
2999 path->nodes[*level-1] = next;
3000 *level = btrfs_header_level(next);
3001 path->slots[*level] = 0;
3005 WARN_ON(*level < 0);
3006 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3008 if (path->nodes[*level] == root->node) {
3009 parent = path->nodes[*level];
3010 bytenr = path->nodes[*level]->start;
3012 parent = path->nodes[*level + 1];
3013 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3016 blocksize = btrfs_level_size(root, *level);
3017 root_owner = btrfs_header_owner(parent);
3018 root_gen = btrfs_header_generation(parent);
3020 mutex_lock(&root->fs_info->alloc_mutex);
3021 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3022 parent->start, root_owner, root_gen,
3024 mutex_unlock(&root->fs_info->alloc_mutex);
3025 free_extent_buffer(path->nodes[*level]);
3026 path->nodes[*level] = NULL;
3035 * helper for dropping snapshots. This walks back up the tree in the path
3036 * to find the first node higher up where we haven't yet gone through
3039 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 struct btrfs_path *path, int *level)
3045 struct btrfs_root_item *root_item = &root->root_item;
3050 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3051 slot = path->slots[i];
3052 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3053 struct extent_buffer *node;
3054 struct btrfs_disk_key disk_key;
3055 node = path->nodes[i];
3058 WARN_ON(*level == 0);
3059 btrfs_node_key(node, &disk_key, path->slots[i]);
3060 memcpy(&root_item->drop_progress,
3061 &disk_key, sizeof(disk_key));
3062 root_item->drop_level = i;
3065 struct extent_buffer *parent;
3066 if (path->nodes[*level] == root->node)
3067 parent = path->nodes[*level];
3069 parent = path->nodes[*level + 1];
3071 root_owner = btrfs_header_owner(parent);
3072 root_gen = btrfs_header_generation(parent);
3073 ret = btrfs_free_extent(trans, root,
3074 path->nodes[*level]->start,
3075 path->nodes[*level]->len,
3077 root_owner, root_gen, 0, 0, 1);
3079 free_extent_buffer(path->nodes[*level]);
3080 path->nodes[*level] = NULL;
3088 * drop the reference count on the tree rooted at 'snap'. This traverses
3089 * the tree freeing any blocks that have a ref count of zero after being
3092 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3098 struct btrfs_path *path;
3101 struct btrfs_root_item *root_item = &root->root_item;
3103 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3104 path = btrfs_alloc_path();
3107 level = btrfs_header_level(root->node);
3109 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3110 path->nodes[level] = root->node;
3111 extent_buffer_get(root->node);
3112 path->slots[level] = 0;
3114 struct btrfs_key key;
3115 struct btrfs_disk_key found_key;
3116 struct extent_buffer *node;
3118 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3119 level = root_item->drop_level;
3120 path->lowest_level = level;
3121 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3126 node = path->nodes[level];
3127 btrfs_node_key(node, &found_key, path->slots[level]);
3128 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3129 sizeof(found_key)));
3131 * unlock our path, this is safe because only this
3132 * function is allowed to delete this snapshot
3134 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3135 if (path->nodes[i] && path->locks[i]) {
3137 btrfs_tree_unlock(path->nodes[i]);
3142 wret = walk_down_tree(trans, root, path, &level);
3148 wret = walk_up_tree(trans, root, path, &level);
3153 if (trans->transaction->in_commit) {
3157 atomic_inc(&root->fs_info->throttle_gen);
3158 wake_up(&root->fs_info->transaction_throttle);
3160 for (i = 0; i <= orig_level; i++) {
3161 if (path->nodes[i]) {
3162 free_extent_buffer(path->nodes[i]);
3163 path->nodes[i] = NULL;
3167 btrfs_free_path(path);
3171 static unsigned long calc_ra(unsigned long start, unsigned long last,
3174 return min(last, start + nr - 1);
3177 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3182 unsigned long first_index;
3183 unsigned long last_index;
3186 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3187 struct file_ra_state *ra;
3188 struct btrfs_ordered_extent *ordered;
3189 unsigned int total_read = 0;
3190 unsigned int total_dirty = 0;
3193 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3195 mutex_lock(&inode->i_mutex);
3196 first_index = start >> PAGE_CACHE_SHIFT;
3197 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3199 /* make sure the dirty trick played by the caller work */
3200 ret = invalidate_inode_pages2_range(inode->i_mapping,
3201 first_index, last_index);
3205 file_ra_state_init(ra, inode->i_mapping);
3207 for (i = first_index ; i <= last_index; i++) {
3208 if (total_read % ra->ra_pages == 0) {
3209 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3210 calc_ra(i, last_index, ra->ra_pages));
3214 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3216 page = grab_cache_page(inode->i_mapping, i);
3221 if (!PageUptodate(page)) {
3222 btrfs_readpage(NULL, page);
3224 if (!PageUptodate(page)) {
3226 page_cache_release(page);
3231 wait_on_page_writeback(page);
3233 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3234 page_end = page_start + PAGE_CACHE_SIZE - 1;
3235 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3237 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3239 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3241 page_cache_release(page);
3242 btrfs_start_ordered_extent(inode, ordered, 1);
3243 btrfs_put_ordered_extent(ordered);
3246 set_page_extent_mapped(page);
3248 btrfs_set_extent_delalloc(inode, page_start, page_end);
3249 if (i == first_index)
3250 set_extent_bits(io_tree, page_start, page_end,
3251 EXTENT_BOUNDARY, GFP_NOFS);
3253 set_page_dirty(page);
3256 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3258 page_cache_release(page);
3263 mutex_unlock(&inode->i_mutex);
3264 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3268 static int noinline relocate_data_extent(struct inode *reloc_inode,
3269 struct btrfs_key *extent_key,
3272 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3273 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3274 struct extent_map *em;
3276 em = alloc_extent_map(GFP_NOFS);
3277 BUG_ON(!em || IS_ERR(em));
3279 em->start = extent_key->objectid - offset;
3280 em->len = extent_key->offset;
3281 em->block_start = extent_key->objectid;
3282 em->bdev = root->fs_info->fs_devices->latest_bdev;
3283 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3285 /* setup extent map to cheat btrfs_readpage */
3286 mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex);
3289 spin_lock(&em_tree->lock);
3290 ret = add_extent_mapping(em_tree, em);
3291 spin_unlock(&em_tree->lock);
3292 if (ret != -EEXIST) {
3293 free_extent_map(em);
3296 btrfs_drop_extent_cache(reloc_inode, em->start,
3297 em->start + em->len - 1, 0);
3299 mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex);
3301 return relocate_inode_pages(reloc_inode, extent_key->objectid - offset,
3302 extent_key->offset);
3305 struct btrfs_ref_path {
3307 u64 nodes[BTRFS_MAX_LEVEL];
3309 u64 root_generation;
3317 struct disk_extent {
3324 static int is_cowonly_root(u64 root_objectid)
3326 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3327 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3328 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3329 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3330 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3335 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3336 struct btrfs_root *extent_root,
3337 struct btrfs_ref_path *ref_path,
3340 struct extent_buffer *leaf;
3341 struct btrfs_path *path;
3342 struct btrfs_extent_ref *ref;
3343 struct btrfs_key key;
3344 struct btrfs_key found_key;
3350 path = btrfs_alloc_path();
3354 mutex_lock(&extent_root->fs_info->alloc_mutex);
3357 ref_path->lowest_level = -1;
3358 ref_path->current_level = -1;
3362 level = ref_path->current_level - 1;
3363 while (level >= -1) {
3365 if (level < ref_path->lowest_level)
3369 bytenr = ref_path->nodes[level];
3371 bytenr = ref_path->extent_start;
3373 BUG_ON(bytenr == 0);
3375 parent = ref_path->nodes[level + 1];
3376 ref_path->nodes[level + 1] = 0;
3377 ref_path->current_level = level;
3378 BUG_ON(parent == 0);
3380 key.objectid = bytenr;
3381 key.offset = parent + 1;
3382 key.type = BTRFS_EXTENT_REF_KEY;
3384 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3389 leaf = path->nodes[0];
3390 nritems = btrfs_header_nritems(leaf);
3391 if (path->slots[0] >= nritems) {
3392 ret = btrfs_next_leaf(extent_root, path);
3397 leaf = path->nodes[0];
3400 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3401 if (found_key.objectid == bytenr &&
3402 found_key.type == BTRFS_EXTENT_REF_KEY)
3406 btrfs_release_path(extent_root, path);
3407 if (need_resched()) {
3408 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3410 mutex_lock(&extent_root->fs_info->alloc_mutex);
3413 /* reached lowest level */
3417 level = ref_path->current_level;
3418 while (level < BTRFS_MAX_LEVEL - 1) {
3421 bytenr = ref_path->nodes[level];
3423 bytenr = ref_path->extent_start;
3425 BUG_ON(bytenr == 0);
3427 key.objectid = bytenr;
3429 key.type = BTRFS_EXTENT_REF_KEY;
3431 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3435 leaf = path->nodes[0];
3436 nritems = btrfs_header_nritems(leaf);
3437 if (path->slots[0] >= nritems) {
3438 ret = btrfs_next_leaf(extent_root, path);
3442 /* the extent was freed by someone */
3443 if (ref_path->lowest_level == level)
3445 btrfs_release_path(extent_root, path);
3448 leaf = path->nodes[0];
3451 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3452 if (found_key.objectid != bytenr ||
3453 found_key.type != BTRFS_EXTENT_REF_KEY) {
3454 /* the extent was freed by someone */
3455 if (ref_path->lowest_level == level) {
3459 btrfs_release_path(extent_root, path);
3463 ref = btrfs_item_ptr(leaf, path->slots[0],
3464 struct btrfs_extent_ref);
3465 ref_objectid = btrfs_ref_objectid(leaf, ref);
3466 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3468 level = (int)ref_objectid;
3469 BUG_ON(level >= BTRFS_MAX_LEVEL);
3470 ref_path->lowest_level = level;
3471 ref_path->current_level = level;
3472 ref_path->nodes[level] = bytenr;
3474 WARN_ON(ref_objectid != level);
3477 WARN_ON(level != -1);
3481 if (ref_path->lowest_level == level) {
3482 ref_path->owner_objectid = ref_objectid;
3483 ref_path->owner_offset = btrfs_ref_offset(leaf, ref);
3484 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3488 * the block is tree root or the block isn't in reference
3491 if (found_key.objectid == found_key.offset ||
3492 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3493 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3494 ref_path->root_generation =
3495 btrfs_ref_generation(leaf, ref);
3497 /* special reference from the tree log */
3498 ref_path->nodes[0] = found_key.offset;
3499 ref_path->current_level = 0;
3506 BUG_ON(ref_path->nodes[level] != 0);
3507 ref_path->nodes[level] = found_key.offset;
3508 ref_path->current_level = level;
3511 * the reference was created in the running transaction,
3512 * no need to continue walking up.
3514 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3515 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3516 ref_path->root_generation =
3517 btrfs_ref_generation(leaf, ref);
3522 btrfs_release_path(extent_root, path);
3523 if (need_resched()) {
3524 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3526 mutex_lock(&extent_root->fs_info->alloc_mutex);
3529 /* reached max tree level, but no tree root found. */
3532 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3533 btrfs_free_path(path);
3537 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3538 struct btrfs_root *extent_root,
3539 struct btrfs_ref_path *ref_path,
3542 memset(ref_path, 0, sizeof(*ref_path));
3543 ref_path->extent_start = extent_start;
3545 return __next_ref_path(trans, extent_root, ref_path, 1);
3548 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3549 struct btrfs_root *extent_root,
3550 struct btrfs_ref_path *ref_path)
3552 return __next_ref_path(trans, extent_root, ref_path, 0);
3555 static int noinline get_new_locations(struct inode *reloc_inode,
3556 struct btrfs_key *extent_key,
3557 u64 offset, int no_fragment,
3558 struct disk_extent **extents,
3561 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3562 struct btrfs_path *path;
3563 struct btrfs_file_extent_item *fi;
3564 struct extent_buffer *leaf;
3565 struct disk_extent *exts = *extents;
3566 struct btrfs_key found_key;
3571 int max = *nr_extents;
3574 WARN_ON(!no_fragment && *extents);
3577 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3582 path = btrfs_alloc_path();
3585 cur_pos = extent_key->objectid - offset;
3586 last_byte = extent_key->objectid + extent_key->offset;
3587 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3597 leaf = path->nodes[0];
3598 nritems = btrfs_header_nritems(leaf);
3599 if (path->slots[0] >= nritems) {
3600 ret = btrfs_next_leaf(root, path);
3605 leaf = path->nodes[0];
3608 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3609 if (found_key.offset != cur_pos ||
3610 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3611 found_key.objectid != reloc_inode->i_ino)
3614 fi = btrfs_item_ptr(leaf, path->slots[0],
3615 struct btrfs_file_extent_item);
3616 if (btrfs_file_extent_type(leaf, fi) !=
3617 BTRFS_FILE_EXTENT_REG ||
3618 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3622 struct disk_extent *old = exts;
3624 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3625 memcpy(exts, old, sizeof(*exts) * nr);
3626 if (old != *extents)
3630 exts[nr].disk_bytenr =
3631 btrfs_file_extent_disk_bytenr(leaf, fi);
3632 exts[nr].disk_num_bytes =
3633 btrfs_file_extent_disk_num_bytes(leaf, fi);
3634 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3635 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3636 WARN_ON(exts[nr].offset > 0);
3637 WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3639 cur_pos += exts[nr].num_bytes;
3642 if (cur_pos + offset >= last_byte)
3652 WARN_ON(cur_pos + offset > last_byte);
3653 if (cur_pos + offset < last_byte) {
3659 btrfs_free_path(path);
3661 if (exts != *extents)
3670 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3671 struct btrfs_root *root,
3672 struct btrfs_path *path,
3673 struct btrfs_key *extent_key,
3674 struct btrfs_key *leaf_key,
3675 struct btrfs_ref_path *ref_path,
3676 struct disk_extent *new_extents,
3679 struct extent_buffer *leaf;
3680 struct btrfs_file_extent_item *fi;
3681 struct inode *inode = NULL;
3682 struct btrfs_key key;
3689 int extent_locked = 0;
3692 first_pos = ref_path->owner_offset;
3693 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3694 key.objectid = ref_path->owner_objectid;
3695 key.offset = ref_path->owner_offset;
3696 key.type = BTRFS_EXTENT_DATA_KEY;
3698 memcpy(&key, leaf_key, sizeof(key));
3702 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3706 leaf = path->nodes[0];
3707 nritems = btrfs_header_nritems(leaf);
3709 if (extent_locked && ret > 0) {
3711 * the file extent item was modified by someone
3712 * before the extent got locked.
3714 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3715 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3716 lock_end, GFP_NOFS);
3720 if (path->slots[0] >= nritems) {
3721 if (ref_path->owner_objectid ==
3722 BTRFS_MULTIPLE_OBJECTIDS)
3725 BUG_ON(extent_locked);
3726 ret = btrfs_next_leaf(root, path);
3731 leaf = path->nodes[0];
3732 nritems = btrfs_header_nritems(leaf);
3735 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3737 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3738 if ((key.objectid > ref_path->owner_objectid) ||
3739 (key.objectid == ref_path->owner_objectid &&
3740 key.type > BTRFS_EXTENT_DATA_KEY) ||
3741 (key.offset >= first_pos + extent_key->offset))
3745 if (inode && key.objectid != inode->i_ino) {
3746 BUG_ON(extent_locked);
3747 btrfs_release_path(root, path);
3748 mutex_unlock(&inode->i_mutex);
3754 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3759 fi = btrfs_item_ptr(leaf, path->slots[0],
3760 struct btrfs_file_extent_item);
3761 if ((btrfs_file_extent_type(leaf, fi) !=
3762 BTRFS_FILE_EXTENT_REG) ||
3763 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3764 extent_key->objectid)) {
3770 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3771 ext_offset = btrfs_file_extent_offset(leaf, fi);
3773 if (first_pos > key.offset - ext_offset)
3774 first_pos = key.offset - ext_offset;
3776 if (!extent_locked) {
3777 lock_start = key.offset;
3778 lock_end = lock_start + num_bytes - 1;
3780 BUG_ON(lock_start != key.offset);
3781 BUG_ON(lock_end - lock_start + 1 < num_bytes);
3785 btrfs_release_path(root, path);
3787 inode = btrfs_iget_locked(root->fs_info->sb,
3788 key.objectid, root);
3789 if (inode->i_state & I_NEW) {
3790 BTRFS_I(inode)->root = root;
3791 BTRFS_I(inode)->location.objectid =
3793 BTRFS_I(inode)->location.type =
3794 BTRFS_INODE_ITEM_KEY;
3795 BTRFS_I(inode)->location.offset = 0;
3796 btrfs_read_locked_inode(inode);
3797 unlock_new_inode(inode);
3800 * some code call btrfs_commit_transaction while
3801 * holding the i_mutex, so we can't use mutex_lock
3804 if (is_bad_inode(inode) ||
3805 !mutex_trylock(&inode->i_mutex)) {
3808 key.offset = (u64)-1;
3813 if (!extent_locked) {
3814 struct btrfs_ordered_extent *ordered;
3816 btrfs_release_path(root, path);
3818 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3819 lock_end, GFP_NOFS);
3820 ordered = btrfs_lookup_first_ordered_extent(inode,
3823 ordered->file_offset <= lock_end &&
3824 ordered->file_offset + ordered->len > lock_start) {
3825 unlock_extent(&BTRFS_I(inode)->io_tree,
3826 lock_start, lock_end, GFP_NOFS);
3827 btrfs_start_ordered_extent(inode, ordered, 1);
3828 btrfs_put_ordered_extent(ordered);
3829 key.offset += num_bytes;
3833 btrfs_put_ordered_extent(ordered);
3835 mutex_lock(&BTRFS_I(inode)->extent_mutex);
3840 if (nr_extents == 1) {
3841 /* update extent pointer in place */
3842 btrfs_set_file_extent_generation(leaf, fi,
3844 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3845 new_extents[0].disk_bytenr);
3846 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3847 new_extents[0].disk_num_bytes);
3848 ext_offset += new_extents[0].offset;
3849 btrfs_set_file_extent_offset(leaf, fi, ext_offset);
3850 btrfs_mark_buffer_dirty(leaf);
3852 btrfs_drop_extent_cache(inode, key.offset,
3853 key.offset + num_bytes - 1, 0);
3855 ret = btrfs_inc_extent_ref(trans, root,
3856 new_extents[0].disk_bytenr,
3857 new_extents[0].disk_num_bytes,
3859 root->root_key.objectid,
3861 key.objectid, key.offset);
3864 ret = btrfs_free_extent(trans, root,
3865 extent_key->objectid,
3868 btrfs_header_owner(leaf),
3869 btrfs_header_generation(leaf),
3870 key.objectid, key.offset, 0);
3873 btrfs_release_path(root, path);
3874 key.offset += num_bytes;
3880 * drop old extent pointer at first, then insert the
3881 * new pointers one bye one
3883 btrfs_release_path(root, path);
3884 ret = btrfs_drop_extents(trans, root, inode, key.offset,
3885 key.offset + num_bytes,
3886 key.offset, &alloc_hint);
3889 for (i = 0; i < nr_extents; i++) {
3890 if (ext_offset >= new_extents[i].num_bytes) {
3891 ext_offset -= new_extents[i].num_bytes;
3894 extent_len = min(new_extents[i].num_bytes -
3895 ext_offset, num_bytes);
3897 ret = btrfs_insert_empty_item(trans, root,
3902 leaf = path->nodes[0];
3903 fi = btrfs_item_ptr(leaf, path->slots[0],
3904 struct btrfs_file_extent_item);
3905 btrfs_set_file_extent_generation(leaf, fi,
3907 btrfs_set_file_extent_type(leaf, fi,
3908 BTRFS_FILE_EXTENT_REG);
3909 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3910 new_extents[i].disk_bytenr);
3911 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3912 new_extents[i].disk_num_bytes);
3913 btrfs_set_file_extent_num_bytes(leaf, fi,
3915 ext_offset += new_extents[i].offset;
3916 btrfs_set_file_extent_offset(leaf, fi,
3918 btrfs_mark_buffer_dirty(leaf);
3920 btrfs_drop_extent_cache(inode, key.offset,
3921 key.offset + extent_len - 1, 0);
3923 ret = btrfs_inc_extent_ref(trans, root,
3924 new_extents[i].disk_bytenr,
3925 new_extents[i].disk_num_bytes,
3927 root->root_key.objectid,
3929 key.objectid, key.offset);
3931 btrfs_release_path(root, path);
3933 inode_add_bytes(inode, extent_len);
3936 num_bytes -= extent_len;
3937 key.offset += extent_len;
3942 BUG_ON(i >= nr_extents);
3945 if (extent_locked) {
3946 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3947 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3948 lock_end, GFP_NOFS);
3952 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
3953 key.offset >= first_pos + extent_key->offset)
3960 btrfs_release_path(root, path);
3962 mutex_unlock(&inode->i_mutex);
3963 if (extent_locked) {
3964 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3965 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3966 lock_end, GFP_NOFS);
3973 int btrfs_add_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3974 u64 num_bytes, u64 new_bytenr)
3976 set_extent_bits(&root->fs_info->reloc_mapping_tree,
3977 orig_bytenr, orig_bytenr + num_bytes - 1,
3978 EXTENT_LOCKED, GFP_NOFS);
3979 set_state_private(&root->fs_info->reloc_mapping_tree,
3980 orig_bytenr, new_bytenr);
3984 int btrfs_get_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3985 u64 num_bytes, u64 *new_bytenr)
3988 u64 cur_bytenr = orig_bytenr;
3989 u64 prev_bytenr = orig_bytenr;
3993 ret = get_state_private(&root->fs_info->reloc_mapping_tree,
3994 cur_bytenr, &bytenr);
3997 prev_bytenr = cur_bytenr;
3998 cur_bytenr = bytenr;
4001 if (orig_bytenr == cur_bytenr)
4004 if (prev_bytenr != orig_bytenr) {
4005 set_state_private(&root->fs_info->reloc_mapping_tree,
4006 orig_bytenr, cur_bytenr);
4008 *new_bytenr = cur_bytenr;
4012 void btrfs_free_reloc_mappings(struct btrfs_root *root)
4014 clear_extent_bits(&root->fs_info->reloc_mapping_tree,
4015 0, (u64)-1, -1, GFP_NOFS);
4018 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4019 struct btrfs_root *root,
4020 struct extent_buffer *buf, u64 orig_start)
4025 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4026 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4028 level = btrfs_header_level(buf);
4030 struct btrfs_leaf_ref *ref;
4031 struct btrfs_leaf_ref *orig_ref;
4033 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4037 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4039 btrfs_free_leaf_ref(root, orig_ref);
4043 ref->nritems = orig_ref->nritems;
4044 memcpy(ref->extents, orig_ref->extents,
4045 sizeof(ref->extents[0]) * ref->nritems);
4047 btrfs_free_leaf_ref(root, orig_ref);
4049 ref->root_gen = trans->transid;
4050 ref->bytenr = buf->start;
4051 ref->owner = btrfs_header_owner(buf);
4052 ref->generation = btrfs_header_generation(buf);
4053 ret = btrfs_add_leaf_ref(root, ref, 0);
4055 btrfs_free_leaf_ref(root, ref);
4060 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4061 struct extent_buffer *leaf,
4062 struct btrfs_block_group_cache *group,
4063 struct btrfs_root *target_root)
4065 struct btrfs_key key;
4066 struct inode *inode = NULL;
4067 struct btrfs_file_extent_item *fi;
4069 u64 skip_objectid = 0;
4073 nritems = btrfs_header_nritems(leaf);
4074 for (i = 0; i < nritems; i++) {
4075 btrfs_item_key_to_cpu(leaf, &key, i);
4076 if (key.objectid == skip_objectid ||
4077 key.type != BTRFS_EXTENT_DATA_KEY)
4079 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4080 if (btrfs_file_extent_type(leaf, fi) ==
4081 BTRFS_FILE_EXTENT_INLINE)
4083 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4085 if (!inode || inode->i_ino != key.objectid) {
4087 inode = btrfs_ilookup(target_root->fs_info->sb,
4088 key.objectid, target_root, 1);
4091 skip_objectid = key.objectid;
4094 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4096 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4097 key.offset + num_bytes - 1, GFP_NOFS);
4098 mutex_lock(&BTRFS_I(inode)->extent_mutex);
4099 btrfs_drop_extent_cache(inode, key.offset,
4100 key.offset + num_bytes - 1, 1);
4101 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
4102 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4103 key.offset + num_bytes - 1, GFP_NOFS);
4110 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4111 struct btrfs_root *root,
4112 struct extent_buffer *leaf,
4113 struct btrfs_block_group_cache *group,
4114 struct inode *reloc_inode)
4116 struct btrfs_key key;
4117 struct btrfs_key extent_key;
4118 struct btrfs_file_extent_item *fi;
4119 struct btrfs_leaf_ref *ref;
4120 struct disk_extent *new_extent;
4129 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4130 BUG_ON(!new_extent);
4132 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4136 nritems = btrfs_header_nritems(leaf);
4137 for (i = 0; i < nritems; i++) {
4138 btrfs_item_key_to_cpu(leaf, &key, i);
4139 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4141 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4142 if (btrfs_file_extent_type(leaf, fi) ==
4143 BTRFS_FILE_EXTENT_INLINE)
4145 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4146 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4151 if (bytenr >= group->key.objectid + group->key.offset ||
4152 bytenr + num_bytes <= group->key.objectid)
4155 extent_key.objectid = bytenr;
4156 extent_key.offset = num_bytes;
4157 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4159 ret = get_new_locations(reloc_inode, &extent_key,
4160 group->key.objectid, 1,
4161 &new_extent, &nr_extent);
4166 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4167 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4168 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4169 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4171 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4172 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4173 new_extent->disk_bytenr);
4174 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4175 new_extent->disk_num_bytes);
4176 new_extent->offset += btrfs_file_extent_offset(leaf, fi);
4177 btrfs_set_file_extent_offset(leaf, fi, new_extent->offset);
4178 btrfs_mark_buffer_dirty(leaf);
4180 ret = btrfs_inc_extent_ref(trans, root,
4181 new_extent->disk_bytenr,
4182 new_extent->disk_num_bytes,
4184 root->root_key.objectid,
4186 key.objectid, key.offset);
4188 ret = btrfs_free_extent(trans, root,
4189 bytenr, num_bytes, leaf->start,
4190 btrfs_header_owner(leaf),
4191 btrfs_header_generation(leaf),
4192 key.objectid, key.offset, 0);
4197 BUG_ON(ext_index + 1 != ref->nritems);
4198 btrfs_free_leaf_ref(root, ref);
4202 int btrfs_free_reloc_root(struct btrfs_root *root)
4204 struct btrfs_root *reloc_root;
4206 if (root->reloc_root) {
4207 reloc_root = root->reloc_root;
4208 root->reloc_root = NULL;
4209 list_add(&reloc_root->dead_list,
4210 &root->fs_info->dead_reloc_roots);
4215 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4217 struct btrfs_trans_handle *trans;
4218 struct btrfs_root *reloc_root;
4219 struct btrfs_root *prev_root = NULL;
4220 struct list_head dead_roots;
4224 INIT_LIST_HEAD(&dead_roots);
4225 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4227 while (!list_empty(&dead_roots)) {
4228 reloc_root = list_entry(dead_roots.prev,
4229 struct btrfs_root, dead_list);
4230 list_del_init(&reloc_root->dead_list);
4232 BUG_ON(reloc_root->commit_root != NULL);
4234 trans = btrfs_join_transaction(root, 1);
4237 mutex_lock(&root->fs_info->drop_mutex);
4238 ret = btrfs_drop_snapshot(trans, reloc_root);
4241 mutex_unlock(&root->fs_info->drop_mutex);
4243 nr = trans->blocks_used;
4244 ret = btrfs_end_transaction(trans, root);
4246 btrfs_btree_balance_dirty(root, nr);
4249 free_extent_buffer(reloc_root->node);
4251 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4252 &reloc_root->root_key);
4254 mutex_unlock(&root->fs_info->drop_mutex);
4256 nr = trans->blocks_used;
4257 ret = btrfs_end_transaction(trans, root);
4259 btrfs_btree_balance_dirty(root, nr);
4262 prev_root = reloc_root;
4265 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4271 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4273 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4277 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4279 struct btrfs_root *reloc_root;
4280 struct btrfs_trans_handle *trans;
4281 struct btrfs_key location;
4285 mutex_lock(&root->fs_info->tree_reloc_mutex);
4286 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4288 found = !list_empty(&root->fs_info->dead_reloc_roots);
4289 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4292 trans = btrfs_start_transaction(root, 1);
4294 ret = btrfs_commit_transaction(trans, root);
4298 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4299 location.offset = (u64)-1;
4300 location.type = BTRFS_ROOT_ITEM_KEY;
4302 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4303 BUG_ON(!reloc_root);
4304 btrfs_orphan_cleanup(reloc_root);
4308 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4309 struct btrfs_root *root)
4311 struct btrfs_root *reloc_root;
4312 struct extent_buffer *eb;
4313 struct btrfs_root_item *root_item;
4314 struct btrfs_key root_key;
4317 BUG_ON(!root->ref_cows);
4318 if (root->reloc_root)
4321 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4324 ret = btrfs_copy_root(trans, root, root->commit_root,
4325 &eb, BTRFS_TREE_RELOC_OBJECTID);
4328 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4329 root_key.offset = root->root_key.objectid;
4330 root_key.type = BTRFS_ROOT_ITEM_KEY;
4332 memcpy(root_item, &root->root_item, sizeof(root_item));
4333 btrfs_set_root_refs(root_item, 0);
4334 btrfs_set_root_bytenr(root_item, eb->start);
4335 btrfs_set_root_level(root_item, btrfs_header_level(eb));
4336 memset(&root_item->drop_progress, 0, sizeof(root_item->drop_progress));
4337 root_item->drop_level = 0;
4339 btrfs_tree_unlock(eb);
4340 free_extent_buffer(eb);
4342 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4343 &root_key, root_item);
4347 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4349 BUG_ON(!reloc_root);
4350 reloc_root->last_trans = trans->transid;
4351 reloc_root->commit_root = NULL;
4352 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4354 root->reloc_root = reloc_root;
4359 * Core function of space balance.
4361 * The idea is using reloc trees to relocate tree blocks in reference
4362 * counted roots. There is one reloc tree for each subvol, all reloc
4363 * trees share same key objectid. Reloc trees are snapshots of the
4364 * latest committed roots (subvol root->commit_root). To relocate a tree
4365 * block referenced by a subvol, the code COW the block through the reloc
4366 * tree, then update pointer in the subvol to point to the new block.
4367 * Since all reloc trees share same key objectid, we can easily do special
4368 * handing to share tree blocks between reloc trees. Once a tree block has
4369 * been COWed in one reloc tree, we can use the result when the same block
4370 * is COWed again through other reloc trees.
4372 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4373 struct btrfs_root *root,
4374 struct btrfs_path *path,
4375 struct btrfs_key *first_key,
4376 struct btrfs_ref_path *ref_path,
4377 struct btrfs_block_group_cache *group,
4378 struct inode *reloc_inode)
4380 struct btrfs_root *reloc_root;
4381 struct extent_buffer *eb = NULL;
4382 struct btrfs_key *keys;
4386 int lowest_level = 0;
4390 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4391 lowest_level = ref_path->owner_objectid;
4393 if (is_cowonly_root(ref_path->root_objectid)) {
4394 path->lowest_level = lowest_level;
4395 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4397 path->lowest_level = 0;
4398 btrfs_release_path(root, path);
4402 keys = kzalloc(sizeof(*keys) * BTRFS_MAX_LEVEL, GFP_NOFS);
4404 nodes = kzalloc(sizeof(*nodes) * BTRFS_MAX_LEVEL, GFP_NOFS);
4407 mutex_lock(&root->fs_info->tree_reloc_mutex);
4408 ret = init_reloc_tree(trans, root);
4410 reloc_root = root->reloc_root;
4412 path->lowest_level = lowest_level;
4413 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 0);
4416 * get relocation mapping for tree blocks in the path
4418 lowest_merge = BTRFS_MAX_LEVEL;
4419 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4421 eb = path->nodes[level];
4422 if (!eb || eb == reloc_root->node)
4424 ret = btrfs_get_reloc_mapping(reloc_root, eb->start, eb->len,
4429 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4431 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4432 nodes[level] = new_bytenr;
4433 lowest_merge = level;
4437 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4438 eb = path->nodes[0];
4439 if (btrfs_header_generation(eb) < trans->transid)
4443 btrfs_release_path(reloc_root, path);
4445 * merge tree blocks that already relocated in other reloc trees
4447 if (lowest_merge != BTRFS_MAX_LEVEL) {
4448 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4453 * cow any tree blocks that still haven't been relocated
4455 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 1);
4458 * if we are relocating data block group, update extent pointers
4459 * in the newly created tree leaf.
4461 eb = path->nodes[0];
4462 if (update_refs && nodes[0] != eb->start) {
4463 ret = replace_extents_in_leaf(trans, reloc_root, eb, group,
4468 memset(keys, 0, sizeof(*keys) * BTRFS_MAX_LEVEL);
4469 memset(nodes, 0, sizeof(*nodes) * BTRFS_MAX_LEVEL);
4470 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4471 eb = path->nodes[level];
4472 if (!eb || eb == reloc_root->node)
4474 BUG_ON(btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID);
4475 nodes[level] = eb->start;
4477 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4479 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4482 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4483 eb = path->nodes[0];
4484 extent_buffer_get(eb);
4486 btrfs_release_path(reloc_root, path);
4488 * replace tree blocks in the fs tree with tree blocks in
4491 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4494 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4495 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4497 free_extent_buffer(eb);
4499 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4501 path->lowest_level = 0;
4507 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4508 struct btrfs_root *root,
4509 struct btrfs_path *path,
4510 struct btrfs_key *first_key,
4511 struct btrfs_ref_path *ref_path)
4516 if (root == root->fs_info->extent_root ||
4517 root == root->fs_info->chunk_root ||
4518 root == root->fs_info->dev_root) {
4520 mutex_lock(&root->fs_info->alloc_mutex);
4523 ret = relocate_one_path(trans, root, path, first_key,
4524 ref_path, NULL, NULL);
4527 if (root == root->fs_info->extent_root)
4528 btrfs_extent_post_op(trans, root);
4530 mutex_unlock(&root->fs_info->alloc_mutex);
4535 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4536 struct btrfs_root *extent_root,
4537 struct btrfs_path *path,
4538 struct btrfs_key *extent_key)
4542 mutex_lock(&extent_root->fs_info->alloc_mutex);
4543 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4546 ret = btrfs_del_item(trans, extent_root, path);
4548 btrfs_release_path(extent_root, path);
4549 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4553 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4554 struct btrfs_ref_path *ref_path)
4556 struct btrfs_key root_key;
4558 root_key.objectid = ref_path->root_objectid;
4559 root_key.type = BTRFS_ROOT_ITEM_KEY;
4560 if (is_cowonly_root(ref_path->root_objectid))
4561 root_key.offset = 0;
4563 root_key.offset = (u64)-1;
4565 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4568 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4569 struct btrfs_path *path,
4570 struct btrfs_key *extent_key,
4571 struct btrfs_block_group_cache *group,
4572 struct inode *reloc_inode, int pass)
4574 struct btrfs_trans_handle *trans;
4575 struct btrfs_root *found_root;
4576 struct btrfs_ref_path *ref_path = NULL;
4577 struct disk_extent *new_extents = NULL;
4582 struct btrfs_key first_key;
4585 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4587 trans = btrfs_start_transaction(extent_root, 1);
4590 if (extent_key->objectid == 0) {
4591 ret = del_extent_zero(trans, extent_root, path, extent_key);
4595 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4601 for (loops = 0; ; loops++) {
4603 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4604 extent_key->objectid);
4606 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4613 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4614 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4617 found_root = read_ref_root(extent_root->fs_info, ref_path);
4618 BUG_ON(!found_root);
4620 * for reference counted tree, only process reference paths
4621 * rooted at the latest committed root.
4623 if (found_root->ref_cows &&
4624 ref_path->root_generation != found_root->root_key.offset)
4627 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4630 * copy data extents to new locations
4632 u64 group_start = group->key.objectid;
4633 ret = relocate_data_extent(reloc_inode,
4642 level = ref_path->owner_objectid;
4645 if (prev_block != ref_path->nodes[level]) {
4646 struct extent_buffer *eb;
4647 u64 block_start = ref_path->nodes[level];
4648 u64 block_size = btrfs_level_size(found_root, level);
4650 eb = read_tree_block(found_root, block_start,
4652 btrfs_tree_lock(eb);
4653 BUG_ON(level != btrfs_header_level(eb));
4656 btrfs_item_key_to_cpu(eb, &first_key, 0);
4658 btrfs_node_key_to_cpu(eb, &first_key, 0);
4660 btrfs_tree_unlock(eb);
4661 free_extent_buffer(eb);
4662 prev_block = block_start;
4665 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4668 * use fallback method to process the remaining
4672 u64 group_start = group->key.objectid;
4673 ret = get_new_locations(reloc_inode,
4681 btrfs_record_root_in_trans(found_root);
4682 ret = replace_one_extent(trans, found_root,
4684 &first_key, ref_path,
4685 new_extents, nr_extents);
4691 btrfs_record_root_in_trans(found_root);
4692 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4693 ret = relocate_tree_block(trans, found_root, path,
4694 &first_key, ref_path);
4697 * try to update data extent references while
4698 * keeping metadata shared between snapshots.
4700 ret = relocate_one_path(trans, found_root, path,
4701 &first_key, ref_path,
4702 group, reloc_inode);
4709 btrfs_end_transaction(trans, extent_root);
4712 mutex_lock(&extent_root->fs_info->alloc_mutex);
4716 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4719 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4720 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4722 num_devices = root->fs_info->fs_devices->num_devices;
4723 if (num_devices == 1) {
4724 stripped |= BTRFS_BLOCK_GROUP_DUP;
4725 stripped = flags & ~stripped;
4727 /* turn raid0 into single device chunks */
4728 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4731 /* turn mirroring into duplication */
4732 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4733 BTRFS_BLOCK_GROUP_RAID10))
4734 return stripped | BTRFS_BLOCK_GROUP_DUP;
4737 /* they already had raid on here, just return */
4738 if (flags & stripped)
4741 stripped |= BTRFS_BLOCK_GROUP_DUP;
4742 stripped = flags & ~stripped;
4744 /* switch duplicated blocks with raid1 */
4745 if (flags & BTRFS_BLOCK_GROUP_DUP)
4746 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4748 /* turn single device chunks into raid0 */
4749 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4754 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4755 struct btrfs_block_group_cache *shrink_block_group,
4758 struct btrfs_trans_handle *trans;
4759 u64 new_alloc_flags;
4762 spin_lock(&shrink_block_group->lock);
4763 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4764 spin_unlock(&shrink_block_group->lock);
4765 mutex_unlock(&root->fs_info->alloc_mutex);
4767 trans = btrfs_start_transaction(root, 1);
4768 mutex_lock(&root->fs_info->alloc_mutex);
4769 spin_lock(&shrink_block_group->lock);
4771 new_alloc_flags = update_block_group_flags(root,
4772 shrink_block_group->flags);
4773 if (new_alloc_flags != shrink_block_group->flags) {
4775 btrfs_block_group_used(&shrink_block_group->item);
4777 calc = shrink_block_group->key.offset;
4779 spin_unlock(&shrink_block_group->lock);
4781 do_chunk_alloc(trans, root->fs_info->extent_root,
4782 calc + 2 * 1024 * 1024, new_alloc_flags, force);
4784 mutex_unlock(&root->fs_info->alloc_mutex);
4785 btrfs_end_transaction(trans, root);
4786 mutex_lock(&root->fs_info->alloc_mutex);
4788 spin_unlock(&shrink_block_group->lock);
4792 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4793 struct btrfs_root *root,
4794 u64 objectid, u64 size)
4796 struct btrfs_path *path;
4797 struct btrfs_inode_item *item;
4798 struct extent_buffer *leaf;
4801 path = btrfs_alloc_path();
4805 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4809 leaf = path->nodes[0];
4810 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4811 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4812 btrfs_set_inode_generation(leaf, item, 1);
4813 btrfs_set_inode_size(leaf, item, size);
4814 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4815 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM);
4816 btrfs_mark_buffer_dirty(leaf);
4817 btrfs_release_path(root, path);
4819 btrfs_free_path(path);
4823 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4824 struct btrfs_block_group_cache *group)
4826 struct inode *inode = NULL;
4827 struct btrfs_trans_handle *trans;
4828 struct btrfs_root *root;
4829 struct btrfs_key root_key;
4830 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4833 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4834 root_key.type = BTRFS_ROOT_ITEM_KEY;
4835 root_key.offset = (u64)-1;
4836 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4838 return ERR_CAST(root);
4840 trans = btrfs_start_transaction(root, 1);
4843 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4847 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4850 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4851 group->key.offset, 0);
4854 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4855 if (inode->i_state & I_NEW) {
4856 BTRFS_I(inode)->root = root;
4857 BTRFS_I(inode)->location.objectid = objectid;
4858 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4859 BTRFS_I(inode)->location.offset = 0;
4860 btrfs_read_locked_inode(inode);
4861 unlock_new_inode(inode);
4862 BUG_ON(is_bad_inode(inode));
4867 err = btrfs_orphan_add(trans, inode);
4869 btrfs_end_transaction(trans, root);
4873 inode = ERR_PTR(err);
4878 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4880 struct btrfs_trans_handle *trans;
4881 struct btrfs_path *path;
4882 struct btrfs_fs_info *info = root->fs_info;
4883 struct extent_buffer *leaf;
4884 struct inode *reloc_inode;
4885 struct btrfs_block_group_cache *block_group;
4886 struct btrfs_key key;
4894 root = root->fs_info->extent_root;
4896 block_group = btrfs_lookup_block_group(info, group_start);
4897 BUG_ON(!block_group);
4899 printk("btrfs relocating block group %llu flags %llu\n",
4900 (unsigned long long)block_group->key.objectid,
4901 (unsigned long long)block_group->flags);
4903 path = btrfs_alloc_path();
4906 reloc_inode = create_reloc_inode(info, block_group);
4907 BUG_ON(IS_ERR(reloc_inode));
4909 mutex_lock(&root->fs_info->alloc_mutex);
4911 __alloc_chunk_for_shrink(root, block_group, 1);
4912 block_group->ro = 1;
4913 block_group->space_info->total_bytes -= block_group->key.offset;
4915 mutex_unlock(&root->fs_info->alloc_mutex);
4917 btrfs_start_delalloc_inodes(info->tree_root);
4918 btrfs_wait_ordered_extents(info->tree_root, 0);
4922 key.objectid = block_group->key.objectid;
4925 cur_byte = key.objectid;
4927 trans = btrfs_start_transaction(info->tree_root, 1);
4928 btrfs_commit_transaction(trans, info->tree_root);
4930 mutex_lock(&root->fs_info->cleaner_mutex);
4931 btrfs_clean_old_snapshots(info->tree_root);
4932 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4933 mutex_unlock(&root->fs_info->cleaner_mutex);
4935 mutex_lock(&root->fs_info->alloc_mutex);
4938 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4942 leaf = path->nodes[0];
4943 nritems = btrfs_header_nritems(leaf);
4944 if (path->slots[0] >= nritems) {
4945 ret = btrfs_next_leaf(root, path);
4952 leaf = path->nodes[0];
4953 nritems = btrfs_header_nritems(leaf);
4956 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4958 if (key.objectid >= block_group->key.objectid +
4959 block_group->key.offset)
4962 if (progress && need_resched()) {
4963 btrfs_release_path(root, path);
4964 mutex_unlock(&root->fs_info->alloc_mutex);
4966 mutex_lock(&root->fs_info->alloc_mutex);
4972 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4973 key.objectid + key.offset <= cur_byte) {
4979 cur_byte = key.objectid + key.offset;
4980 btrfs_release_path(root, path);
4982 __alloc_chunk_for_shrink(root, block_group, 0);
4983 ret = relocate_one_extent(root, path, &key, block_group,
4987 key.objectid = cur_byte;
4992 btrfs_release_path(root, path);
4993 mutex_unlock(&root->fs_info->alloc_mutex);
4996 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4997 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4998 WARN_ON(reloc_inode->i_mapping->nrpages);
5001 if (total_found > 0) {
5002 printk("btrfs found %llu extents in pass %d\n",
5003 (unsigned long long)total_found, pass);
5008 /* delete reloc_inode */
5011 /* unpin extents in this range */
5012 trans = btrfs_start_transaction(info->tree_root, 1);
5013 btrfs_commit_transaction(trans, info->tree_root);
5015 mutex_lock(&root->fs_info->alloc_mutex);
5017 spin_lock(&block_group->lock);
5018 WARN_ON(block_group->pinned > 0);
5019 WARN_ON(block_group->reserved > 0);
5020 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5021 spin_unlock(&block_group->lock);
5024 mutex_unlock(&root->fs_info->alloc_mutex);
5025 btrfs_free_path(path);
5029 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5030 struct btrfs_key *key)
5033 struct btrfs_key found_key;
5034 struct extent_buffer *leaf;
5037 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5042 slot = path->slots[0];
5043 leaf = path->nodes[0];
5044 if (slot >= btrfs_header_nritems(leaf)) {
5045 ret = btrfs_next_leaf(root, path);
5052 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5054 if (found_key.objectid >= key->objectid &&
5055 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5066 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5068 struct btrfs_block_group_cache *block_group;
5071 mutex_lock(&info->alloc_mutex);
5072 spin_lock(&info->block_group_cache_lock);
5073 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5074 block_group = rb_entry(n, struct btrfs_block_group_cache,
5077 spin_unlock(&info->block_group_cache_lock);
5078 btrfs_remove_free_space_cache(block_group);
5079 spin_lock(&info->block_group_cache_lock);
5081 rb_erase(&block_group->cache_node,
5082 &info->block_group_cache_tree);
5083 spin_lock(&block_group->space_info->lock);
5084 list_del(&block_group->list);
5085 spin_unlock(&block_group->space_info->lock);
5088 spin_unlock(&info->block_group_cache_lock);
5089 mutex_unlock(&info->alloc_mutex);
5093 int btrfs_read_block_groups(struct btrfs_root *root)
5095 struct btrfs_path *path;
5097 struct btrfs_block_group_cache *cache;
5098 struct btrfs_fs_info *info = root->fs_info;
5099 struct btrfs_space_info *space_info;
5100 struct btrfs_key key;
5101 struct btrfs_key found_key;
5102 struct extent_buffer *leaf;
5104 root = info->extent_root;
5107 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5108 path = btrfs_alloc_path();
5112 mutex_lock(&root->fs_info->alloc_mutex);
5114 ret = find_first_block_group(root, path, &key);
5122 leaf = path->nodes[0];
5123 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5124 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5130 spin_lock_init(&cache->lock);
5131 INIT_LIST_HEAD(&cache->list);
5132 read_extent_buffer(leaf, &cache->item,
5133 btrfs_item_ptr_offset(leaf, path->slots[0]),
5134 sizeof(cache->item));
5135 memcpy(&cache->key, &found_key, sizeof(found_key));
5137 key.objectid = found_key.objectid + found_key.offset;
5138 btrfs_release_path(root, path);
5139 cache->flags = btrfs_block_group_flags(&cache->item);
5141 ret = update_space_info(info, cache->flags, found_key.offset,
5142 btrfs_block_group_used(&cache->item),
5145 cache->space_info = space_info;
5146 spin_lock(&space_info->lock);
5147 list_add(&cache->list, &space_info->block_groups);
5148 spin_unlock(&space_info->lock);
5150 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5153 set_avail_alloc_bits(root->fs_info, cache->flags);
5157 btrfs_free_path(path);
5158 mutex_unlock(&root->fs_info->alloc_mutex);
5162 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5163 struct btrfs_root *root, u64 bytes_used,
5164 u64 type, u64 chunk_objectid, u64 chunk_offset,
5168 struct btrfs_root *extent_root;
5169 struct btrfs_block_group_cache *cache;
5171 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5172 extent_root = root->fs_info->extent_root;
5174 root->fs_info->last_trans_new_blockgroup = trans->transid;
5176 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5180 cache->key.objectid = chunk_offset;
5181 cache->key.offset = size;
5182 spin_lock_init(&cache->lock);
5183 INIT_LIST_HEAD(&cache->list);
5184 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5186 btrfs_set_block_group_used(&cache->item, bytes_used);
5187 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5188 cache->flags = type;
5189 btrfs_set_block_group_flags(&cache->item, type);
5191 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5192 &cache->space_info);
5194 spin_lock(&cache->space_info->lock);
5195 list_add(&cache->list, &cache->space_info->block_groups);
5196 spin_unlock(&cache->space_info->lock);
5198 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5201 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5202 sizeof(cache->item));
5205 finish_current_insert(trans, extent_root);
5206 ret = del_pending_extents(trans, extent_root);
5208 set_avail_alloc_bits(extent_root->fs_info, type);
5213 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5214 struct btrfs_root *root, u64 group_start)
5216 struct btrfs_path *path;
5217 struct btrfs_block_group_cache *block_group;
5218 struct btrfs_key key;
5221 BUG_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5222 root = root->fs_info->extent_root;
5224 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5225 BUG_ON(!block_group);
5227 memcpy(&key, &block_group->key, sizeof(key));
5229 path = btrfs_alloc_path();
5232 btrfs_remove_free_space_cache(block_group);
5233 rb_erase(&block_group->cache_node,
5234 &root->fs_info->block_group_cache_tree);
5235 spin_lock(&block_group->space_info->lock);
5236 list_del(&block_group->list);
5237 spin_unlock(&block_group->space_info->lock);
5240 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5241 kfree(shrink_block_group);
5244 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5250 ret = btrfs_del_item(trans, root, path);
5252 btrfs_free_path(path);