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;
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 mutex_lock(&extent_root->fs_info->chunk_mutex);
1534 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1535 if (ret == -ENOSPC) {
1536 printk("space info full %Lu\n", flags);
1537 space_info->full = 1;
1542 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1543 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1547 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1552 static int update_block_group(struct btrfs_trans_handle *trans,
1553 struct btrfs_root *root,
1554 u64 bytenr, u64 num_bytes, int alloc,
1557 struct btrfs_block_group_cache *cache;
1558 struct btrfs_fs_info *info = root->fs_info;
1559 u64 total = num_bytes;
1563 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1565 cache = btrfs_lookup_block_group(info, bytenr);
1569 byte_in_group = bytenr - cache->key.objectid;
1570 WARN_ON(byte_in_group > cache->key.offset);
1572 spin_lock(&cache->lock);
1574 old_val = btrfs_block_group_used(&cache->item);
1575 num_bytes = min(total, cache->key.offset - byte_in_group);
1577 old_val += num_bytes;
1578 cache->space_info->bytes_used += num_bytes;
1579 btrfs_set_block_group_used(&cache->item, old_val);
1580 spin_unlock(&cache->lock);
1582 old_val -= num_bytes;
1583 cache->space_info->bytes_used -= num_bytes;
1584 btrfs_set_block_group_used(&cache->item, old_val);
1585 spin_unlock(&cache->lock);
1588 ret = btrfs_add_free_space(cache, bytenr,
1595 bytenr += num_bytes;
1600 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1602 struct btrfs_block_group_cache *cache;
1604 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1608 return cache->key.objectid;
1611 int btrfs_update_pinned_extents(struct btrfs_root *root,
1612 u64 bytenr, u64 num, int pin)
1615 struct btrfs_block_group_cache *cache;
1616 struct btrfs_fs_info *fs_info = root->fs_info;
1618 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1620 set_extent_dirty(&fs_info->pinned_extents,
1621 bytenr, bytenr + num - 1, GFP_NOFS);
1623 clear_extent_dirty(&fs_info->pinned_extents,
1624 bytenr, bytenr + num - 1, GFP_NOFS);
1627 cache = btrfs_lookup_block_group(fs_info, bytenr);
1629 len = min(num, cache->key.offset -
1630 (bytenr - cache->key.objectid));
1632 spin_lock(&cache->lock);
1633 cache->pinned += len;
1634 cache->space_info->bytes_pinned += len;
1635 spin_unlock(&cache->lock);
1636 fs_info->total_pinned += len;
1638 spin_lock(&cache->lock);
1639 cache->pinned -= len;
1640 cache->space_info->bytes_pinned -= len;
1641 spin_unlock(&cache->lock);
1642 fs_info->total_pinned -= len;
1650 static int update_reserved_extents(struct btrfs_root *root,
1651 u64 bytenr, u64 num, int reserve)
1654 struct btrfs_block_group_cache *cache;
1655 struct btrfs_fs_info *fs_info = root->fs_info;
1657 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1659 cache = btrfs_lookup_block_group(fs_info, bytenr);
1661 len = min(num, cache->key.offset -
1662 (bytenr - cache->key.objectid));
1664 spin_lock(&cache->lock);
1665 cache->reserved += len;
1666 cache->space_info->bytes_reserved += len;
1667 spin_unlock(&cache->lock);
1669 spin_lock(&cache->lock);
1670 cache->reserved -= len;
1671 cache->space_info->bytes_reserved -= len;
1672 spin_unlock(&cache->lock);
1680 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1685 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1689 ret = find_first_extent_bit(pinned_extents, last,
1690 &start, &end, EXTENT_DIRTY);
1693 set_extent_dirty(copy, start, end, GFP_NOFS);
1699 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1700 struct btrfs_root *root,
1701 struct extent_io_tree *unpin)
1706 struct btrfs_block_group_cache *cache;
1708 mutex_lock(&root->fs_info->alloc_mutex);
1710 ret = find_first_extent_bit(unpin, 0, &start, &end,
1714 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1715 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1716 cache = btrfs_lookup_block_group(root->fs_info, start);
1718 btrfs_add_free_space(cache, start, end - start + 1);
1719 if (need_resched()) {
1720 mutex_unlock(&root->fs_info->alloc_mutex);
1722 mutex_lock(&root->fs_info->alloc_mutex);
1725 mutex_unlock(&root->fs_info->alloc_mutex);
1729 static int finish_current_insert(struct btrfs_trans_handle *trans,
1730 struct btrfs_root *extent_root)
1735 struct btrfs_fs_info *info = extent_root->fs_info;
1736 struct btrfs_path *path;
1737 struct btrfs_extent_ref *ref;
1738 struct pending_extent_op *extent_op;
1739 struct btrfs_key key;
1740 struct btrfs_extent_item extent_item;
1744 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
1745 btrfs_set_stack_extent_refs(&extent_item, 1);
1746 path = btrfs_alloc_path();
1749 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1750 &end, EXTENT_LOCKED);
1754 ret = get_state_private(&info->extent_ins, start, &priv);
1756 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1758 if (extent_op->type == PENDING_EXTENT_INSERT) {
1759 key.objectid = start;
1760 key.offset = end + 1 - start;
1761 key.type = BTRFS_EXTENT_ITEM_KEY;
1762 err = btrfs_insert_item(trans, extent_root, &key,
1763 &extent_item, sizeof(extent_item));
1766 clear_extent_bits(&info->extent_ins, start, end,
1767 EXTENT_LOCKED, GFP_NOFS);
1769 err = insert_extent_backref(trans, extent_root, path,
1770 start, extent_op->parent,
1771 extent_root->root_key.objectid,
1772 extent_op->generation,
1773 extent_op->level, 0);
1775 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1776 err = lookup_extent_backref(trans, extent_root, path,
1777 start, extent_op->orig_parent,
1778 extent_root->root_key.objectid,
1779 extent_op->orig_generation, 0);
1782 clear_extent_bits(&info->extent_ins, start, end,
1783 EXTENT_LOCKED, GFP_NOFS);
1785 key.objectid = start;
1786 key.offset = extent_op->parent;
1787 key.type = BTRFS_EXTENT_REF_KEY;
1788 err = btrfs_set_item_key_safe(trans, extent_root, path,
1791 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1792 struct btrfs_extent_ref);
1793 btrfs_set_ref_generation(path->nodes[0], ref,
1794 extent_op->generation);
1795 btrfs_mark_buffer_dirty(path->nodes[0]);
1796 btrfs_release_path(extent_root, path);
1802 if (need_resched()) {
1803 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1805 mutex_lock(&extent_root->fs_info->alloc_mutex);
1808 btrfs_free_path(path);
1812 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1813 struct btrfs_root *root,
1814 u64 bytenr, u64 num_bytes, int is_data)
1817 struct extent_buffer *buf;
1819 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1823 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1827 /* we can reuse a block if it hasn't been written
1828 * and it is from this transaction. We can't
1829 * reuse anything from the tree log root because
1830 * it has tiny sub-transactions.
1832 if (btrfs_buffer_uptodate(buf, 0) &&
1833 btrfs_try_tree_lock(buf)) {
1834 u64 header_owner = btrfs_header_owner(buf);
1835 u64 header_transid = btrfs_header_generation(buf);
1836 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1837 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1838 header_transid == trans->transid &&
1839 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1840 clean_tree_block(NULL, root, buf);
1841 btrfs_tree_unlock(buf);
1842 free_extent_buffer(buf);
1845 btrfs_tree_unlock(buf);
1847 free_extent_buffer(buf);
1849 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1856 * remove an extent from the root, returns 0 on success
1858 static int __free_extent(struct btrfs_trans_handle *trans,
1859 struct btrfs_root *root,
1860 u64 bytenr, u64 num_bytes, u64 parent,
1861 u64 root_objectid, u64 ref_generation,
1862 u64 owner_objectid, u64 owner_offset,
1863 int pin, int mark_free)
1865 struct btrfs_path *path;
1866 struct btrfs_key key;
1867 struct btrfs_fs_info *info = root->fs_info;
1868 struct btrfs_root *extent_root = info->extent_root;
1869 struct extent_buffer *leaf;
1871 int extent_slot = 0;
1872 int found_extent = 0;
1874 struct btrfs_extent_item *ei;
1877 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1878 key.objectid = bytenr;
1879 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1880 key.offset = num_bytes;
1881 path = btrfs_alloc_path();
1886 ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent,
1887 root_objectid, ref_generation, 1);
1889 struct btrfs_key found_key;
1890 extent_slot = path->slots[0];
1891 while(extent_slot > 0) {
1893 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1895 if (found_key.objectid != bytenr)
1897 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1898 found_key.offset == num_bytes) {
1902 if (path->slots[0] - extent_slot > 5)
1905 if (!found_extent) {
1906 ret = remove_extent_backref(trans, extent_root, path);
1908 btrfs_release_path(extent_root, path);
1909 ret = btrfs_search_slot(trans, extent_root,
1912 extent_slot = path->slots[0];
1915 btrfs_print_leaf(extent_root, path->nodes[0]);
1917 printk("Unable to find ref byte nr %Lu root %Lu "
1918 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1919 root_objectid, ref_generation, owner_objectid,
1923 leaf = path->nodes[0];
1924 ei = btrfs_item_ptr(leaf, extent_slot,
1925 struct btrfs_extent_item);
1926 refs = btrfs_extent_refs(leaf, ei);
1929 btrfs_set_extent_refs(leaf, ei, refs);
1931 btrfs_mark_buffer_dirty(leaf);
1933 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1934 struct btrfs_extent_ref *ref;
1935 ref = btrfs_item_ptr(leaf, path->slots[0],
1936 struct btrfs_extent_ref);
1937 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1938 /* if the back ref and the extent are next to each other
1939 * they get deleted below in one shot
1941 path->slots[0] = extent_slot;
1943 } else if (found_extent) {
1944 /* otherwise delete the extent back ref */
1945 ret = remove_extent_backref(trans, extent_root, path);
1947 /* if refs are 0, we need to setup the path for deletion */
1949 btrfs_release_path(extent_root, path);
1950 ret = btrfs_search_slot(trans, extent_root, &key, path,
1959 #ifdef BIO_RW_DISCARD
1960 u64 map_length = num_bytes;
1961 struct btrfs_multi_bio *multi = NULL;
1965 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1966 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1972 /* block accounting for super block */
1973 spin_lock_irq(&info->delalloc_lock);
1974 super_used = btrfs_super_bytes_used(&info->super_copy);
1975 btrfs_set_super_bytes_used(&info->super_copy,
1976 super_used - num_bytes);
1977 spin_unlock_irq(&info->delalloc_lock);
1979 /* block accounting for root item */
1980 root_used = btrfs_root_used(&root->root_item);
1981 btrfs_set_root_used(&root->root_item,
1982 root_used - num_bytes);
1983 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1986 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1990 #ifdef BIO_RW_DISCARD
1991 /* Tell the block device(s) that the sectors can be discarded */
1992 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1993 bytenr, &map_length, &multi, 0);
1995 struct btrfs_bio_stripe *stripe = multi->stripes;
1998 if (map_length > num_bytes)
1999 map_length = num_bytes;
2001 for (i = 0; i < multi->num_stripes; i++, stripe++) {
2002 blkdev_issue_discard(stripe->dev->bdev,
2003 stripe->physical >> 9,
2010 btrfs_free_path(path);
2011 finish_current_insert(trans, extent_root);
2016 * find all the blocks marked as pending in the radix tree and remove
2017 * them from the extent map
2019 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2020 btrfs_root *extent_root)
2028 struct extent_io_tree *pending_del;
2029 struct extent_io_tree *extent_ins;
2030 struct pending_extent_op *extent_op;
2032 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
2033 extent_ins = &extent_root->fs_info->extent_ins;
2034 pending_del = &extent_root->fs_info->pending_del;
2037 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2042 ret = get_state_private(pending_del, start, &priv);
2044 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2046 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2049 ret = pin_down_bytes(trans, extent_root, start,
2050 end + 1 - start, 0);
2051 mark_free = ret > 0;
2052 if (!test_range_bit(extent_ins, start, end,
2053 EXTENT_LOCKED, 0)) {
2055 ret = __free_extent(trans, extent_root,
2056 start, end + 1 - start,
2057 extent_op->orig_parent,
2058 extent_root->root_key.objectid,
2059 extent_op->orig_generation,
2060 extent_op->level, 0, 0, mark_free);
2064 ret = get_state_private(extent_ins, start, &priv);
2066 extent_op = (struct pending_extent_op *)
2067 (unsigned long)priv;
2069 clear_extent_bits(extent_ins, start, end,
2070 EXTENT_LOCKED, GFP_NOFS);
2072 if (extent_op->type == PENDING_BACKREF_UPDATE)
2075 ret = update_block_group(trans, extent_root, start,
2076 end + 1 - start, 0, mark_free);
2083 if (need_resched()) {
2084 mutex_unlock(&extent_root->fs_info->alloc_mutex);
2086 mutex_lock(&extent_root->fs_info->alloc_mutex);
2093 * remove an extent from the root, returns 0 on success
2095 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2096 struct btrfs_root *root,
2097 u64 bytenr, u64 num_bytes, u64 parent,
2098 u64 root_objectid, u64 ref_generation,
2099 u64 owner_objectid, u64 owner_offset, int pin)
2101 struct btrfs_root *extent_root = root->fs_info->extent_root;
2105 WARN_ON(num_bytes < root->sectorsize);
2106 if (root == extent_root) {
2107 struct pending_extent_op *extent_op;
2109 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2112 extent_op->type = PENDING_EXTENT_DELETE;
2113 extent_op->bytenr = bytenr;
2114 extent_op->num_bytes = num_bytes;
2115 extent_op->parent = parent;
2116 extent_op->orig_parent = parent;
2117 extent_op->generation = ref_generation;
2118 extent_op->orig_generation = ref_generation;
2119 extent_op->level = (int)owner_objectid;
2121 set_extent_bits(&root->fs_info->pending_del,
2122 bytenr, bytenr + num_bytes - 1,
2123 EXTENT_LOCKED, GFP_NOFS);
2124 set_state_private(&root->fs_info->pending_del,
2125 bytenr, (unsigned long)extent_op);
2128 /* if metadata always pin */
2129 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2130 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2131 struct btrfs_block_group_cache *cache;
2133 /* btrfs_free_reserved_extent */
2134 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2136 btrfs_add_free_space(cache, bytenr, num_bytes);
2137 update_reserved_extents(root, bytenr, num_bytes, 0);
2143 /* if data pin when any transaction has committed this */
2144 if (ref_generation != trans->transid)
2147 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2148 root_objectid, ref_generation, owner_objectid,
2149 owner_offset, pin, pin == 0);
2151 finish_current_insert(trans, root->fs_info->extent_root);
2152 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2153 return ret ? ret : pending_ret;
2156 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2157 struct btrfs_root *root,
2158 u64 bytenr, u64 num_bytes, u64 parent,
2159 u64 root_objectid, u64 ref_generation,
2160 u64 owner_objectid, u64 owner_offset, int pin)
2164 maybe_lock_mutex(root);
2165 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2166 root_objectid, ref_generation,
2167 owner_objectid, owner_offset, pin);
2168 maybe_unlock_mutex(root);
2172 static u64 stripe_align(struct btrfs_root *root, u64 val)
2174 u64 mask = ((u64)root->stripesize - 1);
2175 u64 ret = (val + mask) & ~mask;
2180 * walks the btree of allocated extents and find a hole of a given size.
2181 * The key ins is changed to record the hole:
2182 * ins->objectid == block start
2183 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2184 * ins->offset == number of blocks
2185 * Any available blocks before search_start are skipped.
2187 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2188 struct btrfs_root *orig_root,
2189 u64 num_bytes, u64 empty_size,
2190 u64 search_start, u64 search_end,
2191 u64 hint_byte, struct btrfs_key *ins,
2192 u64 exclude_start, u64 exclude_nr,
2196 u64 orig_search_start;
2197 struct btrfs_root * root = orig_root->fs_info->extent_root;
2198 struct btrfs_fs_info *info = root->fs_info;
2199 u64 total_needed = num_bytes;
2200 u64 *last_ptr = NULL;
2201 struct btrfs_block_group_cache *block_group;
2202 int chunk_alloc_done = 0;
2203 int empty_cluster = 2 * 1024 * 1024;
2204 int allowed_chunk_alloc = 0;
2206 WARN_ON(num_bytes < root->sectorsize);
2207 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2209 if (orig_root->ref_cows || empty_size)
2210 allowed_chunk_alloc = 1;
2212 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2213 last_ptr = &root->fs_info->last_alloc;
2214 empty_cluster = 256 * 1024;
2217 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2218 last_ptr = &root->fs_info->last_data_alloc;
2220 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2221 last_ptr = &root->fs_info->last_log_alloc;
2222 if (!last_ptr == 0 && root->fs_info->last_alloc) {
2223 *last_ptr = root->fs_info->last_alloc + empty_cluster;
2229 hint_byte = *last_ptr;
2231 empty_size += empty_cluster;
2234 search_start = max(search_start, first_logical_byte(root, 0));
2235 orig_search_start = search_start;
2237 search_start = max(search_start, hint_byte);
2238 total_needed += empty_size;
2241 block_group = btrfs_lookup_first_block_group(info, search_start);
2244 * Ok this looks a little tricky, buts its really simple. First if we
2245 * didn't find a block group obviously we want to start over.
2246 * Secondly, if the block group we found does not match the type we
2247 * need, and we have a last_ptr and its not 0, chances are the last
2248 * allocation we made was at the end of the block group, so lets go
2249 * ahead and skip the looking through the rest of the block groups and
2250 * start at the beginning. This helps with metadata allocations,
2251 * since you are likely to have a bunch of data block groups to search
2252 * through first before you realize that you need to start over, so go
2253 * ahead and start over and save the time.
2255 if (!block_group || (!block_group_bits(block_group, data) &&
2256 last_ptr && *last_ptr)) {
2257 if (search_start != orig_search_start) {
2258 if (last_ptr && *last_ptr)
2260 search_start = orig_search_start;
2262 } else if (!chunk_alloc_done && allowed_chunk_alloc) {
2263 ret = do_chunk_alloc(trans, root,
2264 num_bytes + 2 * 1024 * 1024,
2269 chunk_alloc_done = 1;
2270 search_start = orig_search_start;
2279 * this is going to seach through all of the existing block groups it
2280 * can find, so if we don't find something we need to see if we can
2281 * allocate what we need.
2283 ret = find_free_space(root, &block_group, &search_start,
2284 total_needed, data);
2285 if (ret == -ENOSPC) {
2287 * instead of allocating, start at the original search start
2288 * and see if there is something to be found, if not then we
2291 if (search_start != orig_search_start) {
2292 if (last_ptr && *last_ptr) {
2294 total_needed += empty_cluster;
2296 search_start = orig_search_start;
2301 * we've already allocated, we're pretty screwed
2303 if (chunk_alloc_done) {
2305 } else if (!allowed_chunk_alloc && block_group &&
2306 block_group_bits(block_group, data)) {
2307 block_group->space_info->force_alloc = 1;
2309 } else if (!allowed_chunk_alloc) {
2313 ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
2319 chunk_alloc_done = 1;
2321 search_start = block_group->key.objectid +
2322 block_group->key.offset;
2324 search_start = orig_search_start;
2331 search_start = stripe_align(root, search_start);
2332 ins->objectid = search_start;
2333 ins->offset = num_bytes;
2335 if (ins->objectid + num_bytes >= search_end) {
2336 search_start = orig_search_start;
2337 if (chunk_alloc_done) {
2344 if (ins->objectid + num_bytes >
2345 block_group->key.objectid + block_group->key.offset) {
2346 if (search_start == orig_search_start && chunk_alloc_done) {
2350 search_start = block_group->key.objectid +
2351 block_group->key.offset;
2355 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2356 ins->objectid < exclude_start + exclude_nr)) {
2357 search_start = exclude_start + exclude_nr;
2361 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2362 trans->block_group = block_group;
2364 ins->offset = num_bytes;
2366 *last_ptr = ins->objectid + ins->offset;
2368 btrfs_super_total_bytes(&root->fs_info->super_copy))
2377 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2379 struct btrfs_block_group_cache *cache;
2380 struct list_head *l;
2382 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2383 info->total_bytes - info->bytes_used - info->bytes_pinned -
2384 info->bytes_reserved, (info->full) ? "" : "not ");
2386 spin_lock(&info->lock);
2387 list_for_each(l, &info->block_groups) {
2388 cache = list_entry(l, struct btrfs_block_group_cache, list);
2389 spin_lock(&cache->lock);
2390 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2391 "%Lu pinned %Lu reserved\n",
2392 cache->key.objectid, cache->key.offset,
2393 btrfs_block_group_used(&cache->item),
2394 cache->pinned, cache->reserved);
2395 btrfs_dump_free_space(cache, bytes);
2396 spin_unlock(&cache->lock);
2398 spin_unlock(&info->lock);
2401 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 u64 num_bytes, u64 min_alloc_size,
2404 u64 empty_size, u64 hint_byte,
2405 u64 search_end, struct btrfs_key *ins,
2409 u64 search_start = 0;
2411 struct btrfs_fs_info *info = root->fs_info;
2412 struct btrfs_block_group_cache *cache;
2415 alloc_profile = info->avail_data_alloc_bits &
2416 info->data_alloc_profile;
2417 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2418 } else if (root == root->fs_info->chunk_root) {
2419 alloc_profile = info->avail_system_alloc_bits &
2420 info->system_alloc_profile;
2421 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2423 alloc_profile = info->avail_metadata_alloc_bits &
2424 info->metadata_alloc_profile;
2425 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2428 data = reduce_alloc_profile(root, data);
2430 * the only place that sets empty_size is btrfs_realloc_node, which
2431 * is not called recursively on allocations
2433 if (empty_size || root->ref_cows) {
2434 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2435 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2437 BTRFS_BLOCK_GROUP_METADATA |
2438 (info->metadata_alloc_profile &
2439 info->avail_metadata_alloc_bits), 0);
2441 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2442 num_bytes + 2 * 1024 * 1024, data, 0);
2445 WARN_ON(num_bytes < root->sectorsize);
2446 ret = find_free_extent(trans, root, num_bytes, empty_size,
2447 search_start, search_end, hint_byte, ins,
2448 trans->alloc_exclude_start,
2449 trans->alloc_exclude_nr, data);
2451 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2452 num_bytes = num_bytes >> 1;
2453 num_bytes = num_bytes & ~(root->sectorsize - 1);
2454 num_bytes = max(num_bytes, min_alloc_size);
2455 do_chunk_alloc(trans, root->fs_info->extent_root,
2456 num_bytes, data, 1);
2460 struct btrfs_space_info *sinfo;
2462 sinfo = __find_space_info(root->fs_info, data);
2463 printk("allocation failed flags %Lu, wanted %Lu\n",
2465 dump_space_info(sinfo, num_bytes);
2468 cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2470 printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
2474 ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
2479 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2481 struct btrfs_block_group_cache *cache;
2483 maybe_lock_mutex(root);
2484 cache = btrfs_lookup_block_group(root->fs_info, start);
2486 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2487 maybe_unlock_mutex(root);
2490 btrfs_add_free_space(cache, start, len);
2491 update_reserved_extents(root, start, len, 0);
2492 maybe_unlock_mutex(root);
2496 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2497 struct btrfs_root *root,
2498 u64 num_bytes, u64 min_alloc_size,
2499 u64 empty_size, u64 hint_byte,
2500 u64 search_end, struct btrfs_key *ins,
2504 maybe_lock_mutex(root);
2505 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2506 empty_size, hint_byte, search_end, ins,
2508 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2509 maybe_unlock_mutex(root);
2513 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2514 struct btrfs_root *root, u64 parent,
2515 u64 root_objectid, u64 ref_generation,
2516 u64 owner, u64 owner_offset,
2517 struct btrfs_key *ins)
2523 u64 num_bytes = ins->offset;
2525 struct btrfs_fs_info *info = root->fs_info;
2526 struct btrfs_root *extent_root = info->extent_root;
2527 struct btrfs_extent_item *extent_item;
2528 struct btrfs_extent_ref *ref;
2529 struct btrfs_path *path;
2530 struct btrfs_key keys[2];
2533 parent = ins->objectid;
2535 /* block accounting for super block */
2536 spin_lock_irq(&info->delalloc_lock);
2537 super_used = btrfs_super_bytes_used(&info->super_copy);
2538 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2539 spin_unlock_irq(&info->delalloc_lock);
2541 /* block accounting for root item */
2542 root_used = btrfs_root_used(&root->root_item);
2543 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2545 if (root == extent_root) {
2546 struct pending_extent_op *extent_op;
2548 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2551 extent_op->type = PENDING_EXTENT_INSERT;
2552 extent_op->bytenr = ins->objectid;
2553 extent_op->num_bytes = ins->offset;
2554 extent_op->parent = parent;
2555 extent_op->orig_parent = 0;
2556 extent_op->generation = ref_generation;
2557 extent_op->orig_generation = 0;
2558 extent_op->level = (int)owner;
2560 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2561 ins->objectid + ins->offset - 1,
2562 EXTENT_LOCKED, GFP_NOFS);
2563 set_state_private(&root->fs_info->extent_ins,
2564 ins->objectid, (unsigned long)extent_op);
2568 memcpy(&keys[0], ins, sizeof(*ins));
2569 keys[1].objectid = ins->objectid;
2570 keys[1].type = BTRFS_EXTENT_REF_KEY;
2571 keys[1].offset = parent;
2572 sizes[0] = sizeof(*extent_item);
2573 sizes[1] = sizeof(*ref);
2575 path = btrfs_alloc_path();
2578 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2582 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2583 struct btrfs_extent_item);
2584 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2585 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2586 struct btrfs_extent_ref);
2588 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2589 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2590 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2591 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2592 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2594 btrfs_mark_buffer_dirty(path->nodes[0]);
2596 trans->alloc_exclude_start = 0;
2597 trans->alloc_exclude_nr = 0;
2598 btrfs_free_path(path);
2599 finish_current_insert(trans, extent_root);
2600 pending_ret = del_pending_extents(trans, extent_root);
2610 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2612 printk("update block group failed for %Lu %Lu\n",
2613 ins->objectid, ins->offset);
2620 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2621 struct btrfs_root *root, u64 parent,
2622 u64 root_objectid, u64 ref_generation,
2623 u64 owner, u64 owner_offset,
2624 struct btrfs_key *ins)
2628 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2630 maybe_lock_mutex(root);
2631 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2632 root_objectid, ref_generation,
2633 owner, owner_offset, ins);
2634 update_reserved_extents(root, ins->objectid, ins->offset, 0);
2635 maybe_unlock_mutex(root);
2640 * this is used by the tree logging recovery code. It records that
2641 * an extent has been allocated and makes sure to clear the free
2642 * space cache bits as well
2644 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2645 struct btrfs_root *root, u64 parent,
2646 u64 root_objectid, u64 ref_generation,
2647 u64 owner, u64 owner_offset,
2648 struct btrfs_key *ins)
2651 struct btrfs_block_group_cache *block_group;
2653 maybe_lock_mutex(root);
2654 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2655 cache_block_group(root, block_group);
2657 ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
2659 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2660 root_objectid, ref_generation,
2661 owner, owner_offset, ins);
2662 maybe_unlock_mutex(root);
2667 * finds a free extent and does all the dirty work required for allocation
2668 * returns the key for the extent through ins, and a tree buffer for
2669 * the first block of the extent through buf.
2671 * returns 0 if everything worked, non-zero otherwise.
2673 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2674 struct btrfs_root *root,
2675 u64 num_bytes, u64 parent, u64 min_alloc_size,
2676 u64 root_objectid, u64 ref_generation,
2677 u64 owner_objectid, u64 owner_offset,
2678 u64 empty_size, u64 hint_byte,
2679 u64 search_end, struct btrfs_key *ins, u64 data)
2683 maybe_lock_mutex(root);
2685 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2686 min_alloc_size, empty_size, hint_byte,
2687 search_end, ins, data);
2689 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2690 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2691 root_objectid, ref_generation,
2692 owner_objectid, owner_offset, ins);
2696 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2698 maybe_unlock_mutex(root);
2702 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2703 struct btrfs_root *root,
2704 u64 bytenr, u32 blocksize)
2706 struct extent_buffer *buf;
2708 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2710 return ERR_PTR(-ENOMEM);
2711 btrfs_set_header_generation(buf, trans->transid);
2712 btrfs_tree_lock(buf);
2713 clean_tree_block(trans, root, buf);
2714 btrfs_set_buffer_uptodate(buf);
2715 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2716 set_extent_dirty(&root->dirty_log_pages, buf->start,
2717 buf->start + buf->len - 1, GFP_NOFS);
2719 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2720 buf->start + buf->len - 1, GFP_NOFS);
2722 trans->blocks_used++;
2727 * helper function to allocate a block for a given tree
2728 * returns the tree buffer or NULL.
2730 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root,
2732 u32 blocksize, u64 parent,
2739 struct btrfs_key ins;
2741 struct extent_buffer *buf;
2743 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2744 root_objectid, ref_generation, level, 0,
2745 empty_size, hint, (u64)-1, &ins, 0);
2748 return ERR_PTR(ret);
2751 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2755 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2756 struct btrfs_root *root, struct extent_buffer *leaf)
2759 u64 leaf_generation;
2760 struct btrfs_key key;
2761 struct btrfs_file_extent_item *fi;
2766 BUG_ON(!btrfs_is_leaf(leaf));
2767 nritems = btrfs_header_nritems(leaf);
2768 leaf_owner = btrfs_header_owner(leaf);
2769 leaf_generation = btrfs_header_generation(leaf);
2771 for (i = 0; i < nritems; i++) {
2775 btrfs_item_key_to_cpu(leaf, &key, i);
2776 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2778 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2779 if (btrfs_file_extent_type(leaf, fi) ==
2780 BTRFS_FILE_EXTENT_INLINE)
2783 * FIXME make sure to insert a trans record that
2784 * repeats the snapshot del on crash
2786 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2787 if (disk_bytenr == 0)
2790 mutex_lock(&root->fs_info->alloc_mutex);
2791 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2792 btrfs_file_extent_disk_num_bytes(leaf, fi),
2793 leaf->start, leaf_owner, leaf_generation,
2794 key.objectid, key.offset, 0);
2795 mutex_unlock(&root->fs_info->alloc_mutex);
2798 atomic_inc(&root->fs_info->throttle_gen);
2799 wake_up(&root->fs_info->transaction_throttle);
2805 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2806 struct btrfs_root *root,
2807 struct btrfs_leaf_ref *ref)
2811 struct btrfs_extent_info *info = ref->extents;
2813 for (i = 0; i < ref->nritems; i++) {
2814 mutex_lock(&root->fs_info->alloc_mutex);
2815 ret = __btrfs_free_extent(trans, root, info->bytenr,
2816 info->num_bytes, ref->bytenr,
2817 ref->owner, ref->generation,
2818 info->objectid, info->offset, 0);
2819 mutex_unlock(&root->fs_info->alloc_mutex);
2821 atomic_inc(&root->fs_info->throttle_gen);
2822 wake_up(&root->fs_info->transaction_throttle);
2832 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2837 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2840 #if 0 // some debugging code in case we see problems here
2841 /* if the refs count is one, it won't get increased again. But
2842 * if the ref count is > 1, someone may be decreasing it at
2843 * the same time we are.
2846 struct extent_buffer *eb = NULL;
2847 eb = btrfs_find_create_tree_block(root, start, len);
2849 btrfs_tree_lock(eb);
2851 mutex_lock(&root->fs_info->alloc_mutex);
2852 ret = lookup_extent_ref(NULL, root, start, len, refs);
2854 mutex_unlock(&root->fs_info->alloc_mutex);
2857 btrfs_tree_unlock(eb);
2858 free_extent_buffer(eb);
2861 printk("block %llu went down to one during drop_snap\n",
2862 (unsigned long long)start);
2873 * helper function for drop_snapshot, this walks down the tree dropping ref
2874 * counts as it goes.
2876 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *root,
2878 struct btrfs_path *path, int *level)
2884 struct extent_buffer *next;
2885 struct extent_buffer *cur;
2886 struct extent_buffer *parent;
2887 struct btrfs_leaf_ref *ref;
2892 WARN_ON(*level < 0);
2893 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2894 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2895 path->nodes[*level]->len, &refs);
2901 * walk down to the last node level and free all the leaves
2903 while(*level >= 0) {
2904 WARN_ON(*level < 0);
2905 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2906 cur = path->nodes[*level];
2908 if (btrfs_header_level(cur) != *level)
2911 if (path->slots[*level] >=
2912 btrfs_header_nritems(cur))
2915 ret = btrfs_drop_leaf_ref(trans, root, cur);
2919 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2920 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2921 blocksize = btrfs_level_size(root, *level - 1);
2923 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2926 parent = path->nodes[*level];
2927 root_owner = btrfs_header_owner(parent);
2928 root_gen = btrfs_header_generation(parent);
2929 path->slots[*level]++;
2931 mutex_lock(&root->fs_info->alloc_mutex);
2932 ret = __btrfs_free_extent(trans, root, bytenr,
2933 blocksize, parent->start,
2934 root_owner, root_gen, 0, 0, 1);
2936 mutex_unlock(&root->fs_info->alloc_mutex);
2938 atomic_inc(&root->fs_info->throttle_gen);
2939 wake_up(&root->fs_info->transaction_throttle);
2945 * at this point, we have a single ref, and since the
2946 * only place referencing this extent is a dead root
2947 * the reference count should never go higher.
2948 * So, we don't need to check it again
2951 ref = btrfs_lookup_leaf_ref(root, bytenr);
2952 if (ref && ref->generation != ptr_gen) {
2953 btrfs_free_leaf_ref(root, ref);
2957 ret = cache_drop_leaf_ref(trans, root, ref);
2959 btrfs_remove_leaf_ref(root, ref);
2960 btrfs_free_leaf_ref(root, ref);
2964 if (printk_ratelimit())
2965 printk("leaf ref miss for bytenr %llu\n",
2966 (unsigned long long)bytenr);
2968 next = btrfs_find_tree_block(root, bytenr, blocksize);
2969 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2970 free_extent_buffer(next);
2972 next = read_tree_block(root, bytenr, blocksize,
2977 * this is a debugging check and can go away
2978 * the ref should never go all the way down to 1
2981 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2987 WARN_ON(*level <= 0);
2988 if (path->nodes[*level-1])
2989 free_extent_buffer(path->nodes[*level-1]);
2990 path->nodes[*level-1] = next;
2991 *level = btrfs_header_level(next);
2992 path->slots[*level] = 0;
2996 WARN_ON(*level < 0);
2997 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2999 if (path->nodes[*level] == root->node) {
3000 parent = path->nodes[*level];
3001 bytenr = path->nodes[*level]->start;
3003 parent = path->nodes[*level + 1];
3004 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3007 blocksize = btrfs_level_size(root, *level);
3008 root_owner = btrfs_header_owner(parent);
3009 root_gen = btrfs_header_generation(parent);
3011 mutex_lock(&root->fs_info->alloc_mutex);
3012 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3013 parent->start, root_owner, root_gen,
3015 mutex_unlock(&root->fs_info->alloc_mutex);
3016 free_extent_buffer(path->nodes[*level]);
3017 path->nodes[*level] = NULL;
3026 * helper for dropping snapshots. This walks back up the tree in the path
3027 * to find the first node higher up where we haven't yet gone through
3030 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3031 struct btrfs_root *root,
3032 struct btrfs_path *path, int *level)
3036 struct btrfs_root_item *root_item = &root->root_item;
3041 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3042 slot = path->slots[i];
3043 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3044 struct extent_buffer *node;
3045 struct btrfs_disk_key disk_key;
3046 node = path->nodes[i];
3049 WARN_ON(*level == 0);
3050 btrfs_node_key(node, &disk_key, path->slots[i]);
3051 memcpy(&root_item->drop_progress,
3052 &disk_key, sizeof(disk_key));
3053 root_item->drop_level = i;
3056 struct extent_buffer *parent;
3057 if (path->nodes[*level] == root->node)
3058 parent = path->nodes[*level];
3060 parent = path->nodes[*level + 1];
3062 root_owner = btrfs_header_owner(parent);
3063 root_gen = btrfs_header_generation(parent);
3064 ret = btrfs_free_extent(trans, root,
3065 path->nodes[*level]->start,
3066 path->nodes[*level]->len,
3068 root_owner, root_gen, 0, 0, 1);
3070 free_extent_buffer(path->nodes[*level]);
3071 path->nodes[*level] = NULL;
3079 * drop the reference count on the tree rooted at 'snap'. This traverses
3080 * the tree freeing any blocks that have a ref count of zero after being
3083 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3089 struct btrfs_path *path;
3092 struct btrfs_root_item *root_item = &root->root_item;
3094 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3095 path = btrfs_alloc_path();
3098 level = btrfs_header_level(root->node);
3100 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3101 path->nodes[level] = root->node;
3102 extent_buffer_get(root->node);
3103 path->slots[level] = 0;
3105 struct btrfs_key key;
3106 struct btrfs_disk_key found_key;
3107 struct extent_buffer *node;
3109 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3110 level = root_item->drop_level;
3111 path->lowest_level = level;
3112 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3117 node = path->nodes[level];
3118 btrfs_node_key(node, &found_key, path->slots[level]);
3119 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3120 sizeof(found_key)));
3122 * unlock our path, this is safe because only this
3123 * function is allowed to delete this snapshot
3125 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3126 if (path->nodes[i] && path->locks[i]) {
3128 btrfs_tree_unlock(path->nodes[i]);
3133 wret = walk_down_tree(trans, root, path, &level);
3139 wret = walk_up_tree(trans, root, path, &level);
3144 if (trans->transaction->in_commit) {
3148 atomic_inc(&root->fs_info->throttle_gen);
3149 wake_up(&root->fs_info->transaction_throttle);
3151 for (i = 0; i <= orig_level; i++) {
3152 if (path->nodes[i]) {
3153 free_extent_buffer(path->nodes[i]);
3154 path->nodes[i] = NULL;
3158 btrfs_free_path(path);
3162 static unsigned long calc_ra(unsigned long start, unsigned long last,
3165 return min(last, start + nr - 1);
3168 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3173 unsigned long first_index;
3174 unsigned long last_index;
3177 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3178 struct file_ra_state *ra;
3179 struct btrfs_ordered_extent *ordered;
3180 unsigned int total_read = 0;
3181 unsigned int total_dirty = 0;
3184 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3186 mutex_lock(&inode->i_mutex);
3187 first_index = start >> PAGE_CACHE_SHIFT;
3188 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3190 /* make sure the dirty trick played by the caller work */
3191 ret = invalidate_inode_pages2_range(inode->i_mapping,
3192 first_index, last_index);
3196 file_ra_state_init(ra, inode->i_mapping);
3198 for (i = first_index ; i <= last_index; i++) {
3199 if (total_read % ra->ra_pages == 0) {
3200 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3201 calc_ra(i, last_index, ra->ra_pages));
3205 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3207 page = grab_cache_page(inode->i_mapping, i);
3212 if (!PageUptodate(page)) {
3213 btrfs_readpage(NULL, page);
3215 if (!PageUptodate(page)) {
3217 page_cache_release(page);
3222 wait_on_page_writeback(page);
3224 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3225 page_end = page_start + PAGE_CACHE_SIZE - 1;
3226 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3228 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3230 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3232 page_cache_release(page);
3233 btrfs_start_ordered_extent(inode, ordered, 1);
3234 btrfs_put_ordered_extent(ordered);
3237 set_page_extent_mapped(page);
3239 btrfs_set_extent_delalloc(inode, page_start, page_end);
3240 if (i == first_index)
3241 set_extent_bits(io_tree, page_start, page_end,
3242 EXTENT_BOUNDARY, GFP_NOFS);
3244 set_page_dirty(page);
3247 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3249 page_cache_release(page);
3254 mutex_unlock(&inode->i_mutex);
3255 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3259 static int noinline relocate_data_extent(struct inode *reloc_inode,
3260 struct btrfs_key *extent_key,
3263 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3264 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3265 struct extent_map *em;
3267 em = alloc_extent_map(GFP_NOFS);
3268 BUG_ON(!em || IS_ERR(em));
3270 em->start = extent_key->objectid - offset;
3271 em->len = extent_key->offset;
3272 em->block_start = extent_key->objectid;
3273 em->bdev = root->fs_info->fs_devices->latest_bdev;
3274 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3276 /* setup extent map to cheat btrfs_readpage */
3277 mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex);
3280 spin_lock(&em_tree->lock);
3281 ret = add_extent_mapping(em_tree, em);
3282 spin_unlock(&em_tree->lock);
3283 if (ret != -EEXIST) {
3284 free_extent_map(em);
3287 btrfs_drop_extent_cache(reloc_inode, em->start,
3288 em->start + em->len - 1, 0);
3290 mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex);
3292 return relocate_inode_pages(reloc_inode, extent_key->objectid - offset,
3293 extent_key->offset);
3296 struct btrfs_ref_path {
3298 u64 nodes[BTRFS_MAX_LEVEL];
3300 u64 root_generation;
3308 struct disk_extent {
3315 static int is_cowonly_root(u64 root_objectid)
3317 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3318 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3319 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3320 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3321 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3326 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3327 struct btrfs_root *extent_root,
3328 struct btrfs_ref_path *ref_path,
3331 struct extent_buffer *leaf;
3332 struct btrfs_path *path;
3333 struct btrfs_extent_ref *ref;
3334 struct btrfs_key key;
3335 struct btrfs_key found_key;
3341 path = btrfs_alloc_path();
3345 mutex_lock(&extent_root->fs_info->alloc_mutex);
3348 ref_path->lowest_level = -1;
3349 ref_path->current_level = -1;
3353 level = ref_path->current_level - 1;
3354 while (level >= -1) {
3356 if (level < ref_path->lowest_level)
3360 bytenr = ref_path->nodes[level];
3362 bytenr = ref_path->extent_start;
3364 BUG_ON(bytenr == 0);
3366 parent = ref_path->nodes[level + 1];
3367 ref_path->nodes[level + 1] = 0;
3368 ref_path->current_level = level;
3369 BUG_ON(parent == 0);
3371 key.objectid = bytenr;
3372 key.offset = parent + 1;
3373 key.type = BTRFS_EXTENT_REF_KEY;
3375 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3380 leaf = path->nodes[0];
3381 nritems = btrfs_header_nritems(leaf);
3382 if (path->slots[0] >= nritems) {
3383 ret = btrfs_next_leaf(extent_root, path);
3388 leaf = path->nodes[0];
3391 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3392 if (found_key.objectid == bytenr &&
3393 found_key.type == BTRFS_EXTENT_REF_KEY)
3397 btrfs_release_path(extent_root, path);
3398 if (need_resched()) {
3399 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3401 mutex_lock(&extent_root->fs_info->alloc_mutex);
3404 /* reached lowest level */
3408 level = ref_path->current_level;
3409 while (level < BTRFS_MAX_LEVEL - 1) {
3412 bytenr = ref_path->nodes[level];
3414 bytenr = ref_path->extent_start;
3416 BUG_ON(bytenr == 0);
3418 key.objectid = bytenr;
3420 key.type = BTRFS_EXTENT_REF_KEY;
3422 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3426 leaf = path->nodes[0];
3427 nritems = btrfs_header_nritems(leaf);
3428 if (path->slots[0] >= nritems) {
3429 ret = btrfs_next_leaf(extent_root, path);
3433 /* the extent was freed by someone */
3434 if (ref_path->lowest_level == level)
3436 btrfs_release_path(extent_root, path);
3439 leaf = path->nodes[0];
3442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3443 if (found_key.objectid != bytenr ||
3444 found_key.type != BTRFS_EXTENT_REF_KEY) {
3445 /* the extent was freed by someone */
3446 if (ref_path->lowest_level == level) {
3450 btrfs_release_path(extent_root, path);
3454 ref = btrfs_item_ptr(leaf, path->slots[0],
3455 struct btrfs_extent_ref);
3456 ref_objectid = btrfs_ref_objectid(leaf, ref);
3457 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3459 level = (int)ref_objectid;
3460 BUG_ON(level >= BTRFS_MAX_LEVEL);
3461 ref_path->lowest_level = level;
3462 ref_path->current_level = level;
3463 ref_path->nodes[level] = bytenr;
3465 WARN_ON(ref_objectid != level);
3468 WARN_ON(level != -1);
3472 if (ref_path->lowest_level == level) {
3473 ref_path->owner_objectid = ref_objectid;
3474 ref_path->owner_offset = btrfs_ref_offset(leaf, ref);
3475 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3479 * the block is tree root or the block isn't in reference
3482 if (found_key.objectid == found_key.offset ||
3483 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3484 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3485 ref_path->root_generation =
3486 btrfs_ref_generation(leaf, ref);
3488 /* special reference from the tree log */
3489 ref_path->nodes[0] = found_key.offset;
3490 ref_path->current_level = 0;
3497 BUG_ON(ref_path->nodes[level] != 0);
3498 ref_path->nodes[level] = found_key.offset;
3499 ref_path->current_level = level;
3502 * the reference was created in the running transaction,
3503 * no need to continue walking up.
3505 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3506 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3507 ref_path->root_generation =
3508 btrfs_ref_generation(leaf, ref);
3513 btrfs_release_path(extent_root, path);
3514 if (need_resched()) {
3515 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3517 mutex_lock(&extent_root->fs_info->alloc_mutex);
3520 /* reached max tree level, but no tree root found. */
3523 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3524 btrfs_free_path(path);
3528 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *extent_root,
3530 struct btrfs_ref_path *ref_path,
3533 memset(ref_path, 0, sizeof(*ref_path));
3534 ref_path->extent_start = extent_start;
3536 return __next_ref_path(trans, extent_root, ref_path, 1);
3539 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3540 struct btrfs_root *extent_root,
3541 struct btrfs_ref_path *ref_path)
3543 return __next_ref_path(trans, extent_root, ref_path, 0);
3546 static int noinline get_new_locations(struct inode *reloc_inode,
3547 struct btrfs_key *extent_key,
3548 u64 offset, int no_fragment,
3549 struct disk_extent **extents,
3552 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3553 struct btrfs_path *path;
3554 struct btrfs_file_extent_item *fi;
3555 struct extent_buffer *leaf;
3556 struct disk_extent *exts = *extents;
3557 struct btrfs_key found_key;
3562 int max = *nr_extents;
3565 WARN_ON(!no_fragment && *extents);
3568 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3573 path = btrfs_alloc_path();
3576 cur_pos = extent_key->objectid - offset;
3577 last_byte = extent_key->objectid + extent_key->offset;
3578 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3588 leaf = path->nodes[0];
3589 nritems = btrfs_header_nritems(leaf);
3590 if (path->slots[0] >= nritems) {
3591 ret = btrfs_next_leaf(root, path);
3596 leaf = path->nodes[0];
3599 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3600 if (found_key.offset != cur_pos ||
3601 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3602 found_key.objectid != reloc_inode->i_ino)
3605 fi = btrfs_item_ptr(leaf, path->slots[0],
3606 struct btrfs_file_extent_item);
3607 if (btrfs_file_extent_type(leaf, fi) !=
3608 BTRFS_FILE_EXTENT_REG ||
3609 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3613 struct disk_extent *old = exts;
3615 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3616 memcpy(exts, old, sizeof(*exts) * nr);
3617 if (old != *extents)
3621 exts[nr].disk_bytenr =
3622 btrfs_file_extent_disk_bytenr(leaf, fi);
3623 exts[nr].disk_num_bytes =
3624 btrfs_file_extent_disk_num_bytes(leaf, fi);
3625 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3626 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3627 WARN_ON(exts[nr].offset > 0);
3628 WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3630 cur_pos += exts[nr].num_bytes;
3633 if (cur_pos + offset >= last_byte)
3643 WARN_ON(cur_pos + offset > last_byte);
3644 if (cur_pos + offset < last_byte) {
3650 btrfs_free_path(path);
3652 if (exts != *extents)
3661 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3662 struct btrfs_root *root,
3663 struct btrfs_path *path,
3664 struct btrfs_key *extent_key,
3665 struct btrfs_key *leaf_key,
3666 struct btrfs_ref_path *ref_path,
3667 struct disk_extent *new_extents,
3670 struct extent_buffer *leaf;
3671 struct btrfs_file_extent_item *fi;
3672 struct inode *inode = NULL;
3673 struct btrfs_key key;
3680 int extent_locked = 0;
3683 first_pos = ref_path->owner_offset;
3684 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3685 key.objectid = ref_path->owner_objectid;
3686 key.offset = ref_path->owner_offset;
3687 key.type = BTRFS_EXTENT_DATA_KEY;
3689 memcpy(&key, leaf_key, sizeof(key));
3693 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3697 leaf = path->nodes[0];
3698 nritems = btrfs_header_nritems(leaf);
3700 if (extent_locked && ret > 0) {
3702 * the file extent item was modified by someone
3703 * before the extent got locked.
3705 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3706 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3707 lock_end, GFP_NOFS);
3711 if (path->slots[0] >= nritems) {
3712 if (ref_path->owner_objectid ==
3713 BTRFS_MULTIPLE_OBJECTIDS)
3716 BUG_ON(extent_locked);
3717 ret = btrfs_next_leaf(root, path);
3722 leaf = path->nodes[0];
3723 nritems = btrfs_header_nritems(leaf);
3726 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3728 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3729 if ((key.objectid > ref_path->owner_objectid) ||
3730 (key.objectid == ref_path->owner_objectid &&
3731 key.type > BTRFS_EXTENT_DATA_KEY) ||
3732 (key.offset >= first_pos + extent_key->offset))
3736 if (inode && key.objectid != inode->i_ino) {
3737 BUG_ON(extent_locked);
3738 btrfs_release_path(root, path);
3739 mutex_unlock(&inode->i_mutex);
3745 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3750 fi = btrfs_item_ptr(leaf, path->slots[0],
3751 struct btrfs_file_extent_item);
3752 if ((btrfs_file_extent_type(leaf, fi) !=
3753 BTRFS_FILE_EXTENT_REG) ||
3754 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3755 extent_key->objectid)) {
3761 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3762 ext_offset = btrfs_file_extent_offset(leaf, fi);
3764 if (first_pos > key.offset - ext_offset)
3765 first_pos = key.offset - ext_offset;
3767 if (!extent_locked) {
3768 lock_start = key.offset;
3769 lock_end = lock_start + num_bytes - 1;
3771 BUG_ON(lock_start != key.offset);
3772 BUG_ON(lock_end - lock_start + 1 < num_bytes);
3776 btrfs_release_path(root, path);
3778 inode = btrfs_iget_locked(root->fs_info->sb,
3779 key.objectid, root);
3780 if (inode->i_state & I_NEW) {
3781 BTRFS_I(inode)->root = root;
3782 BTRFS_I(inode)->location.objectid =
3784 BTRFS_I(inode)->location.type =
3785 BTRFS_INODE_ITEM_KEY;
3786 BTRFS_I(inode)->location.offset = 0;
3787 btrfs_read_locked_inode(inode);
3788 unlock_new_inode(inode);
3791 * some code call btrfs_commit_transaction while
3792 * holding the i_mutex, so we can't use mutex_lock
3795 if (is_bad_inode(inode) ||
3796 !mutex_trylock(&inode->i_mutex)) {
3799 key.offset = (u64)-1;
3804 if (!extent_locked) {
3805 struct btrfs_ordered_extent *ordered;
3807 btrfs_release_path(root, path);
3809 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3810 lock_end, GFP_NOFS);
3811 ordered = btrfs_lookup_first_ordered_extent(inode,
3814 ordered->file_offset <= lock_end &&
3815 ordered->file_offset + ordered->len > lock_start) {
3816 unlock_extent(&BTRFS_I(inode)->io_tree,
3817 lock_start, lock_end, GFP_NOFS);
3818 btrfs_start_ordered_extent(inode, ordered, 1);
3819 btrfs_put_ordered_extent(ordered);
3820 key.offset += num_bytes;
3824 btrfs_put_ordered_extent(ordered);
3826 mutex_lock(&BTRFS_I(inode)->extent_mutex);
3831 if (nr_extents == 1) {
3832 /* update extent pointer in place */
3833 btrfs_set_file_extent_generation(leaf, fi,
3835 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3836 new_extents[0].disk_bytenr);
3837 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3838 new_extents[0].disk_num_bytes);
3839 ext_offset += new_extents[0].offset;
3840 btrfs_set_file_extent_offset(leaf, fi, ext_offset);
3841 btrfs_mark_buffer_dirty(leaf);
3843 btrfs_drop_extent_cache(inode, key.offset,
3844 key.offset + num_bytes - 1, 0);
3846 ret = btrfs_inc_extent_ref(trans, root,
3847 new_extents[0].disk_bytenr,
3848 new_extents[0].disk_num_bytes,
3850 root->root_key.objectid,
3852 key.objectid, key.offset);
3855 ret = btrfs_free_extent(trans, root,
3856 extent_key->objectid,
3859 btrfs_header_owner(leaf),
3860 btrfs_header_generation(leaf),
3861 key.objectid, key.offset, 0);
3864 btrfs_release_path(root, path);
3865 key.offset += num_bytes;
3871 * drop old extent pointer at first, then insert the
3872 * new pointers one bye one
3874 btrfs_release_path(root, path);
3875 ret = btrfs_drop_extents(trans, root, inode, key.offset,
3876 key.offset + num_bytes,
3877 key.offset, &alloc_hint);
3880 for (i = 0; i < nr_extents; i++) {
3881 if (ext_offset >= new_extents[i].num_bytes) {
3882 ext_offset -= new_extents[i].num_bytes;
3885 extent_len = min(new_extents[i].num_bytes -
3886 ext_offset, num_bytes);
3888 ret = btrfs_insert_empty_item(trans, root,
3893 leaf = path->nodes[0];
3894 fi = btrfs_item_ptr(leaf, path->slots[0],
3895 struct btrfs_file_extent_item);
3896 btrfs_set_file_extent_generation(leaf, fi,
3898 btrfs_set_file_extent_type(leaf, fi,
3899 BTRFS_FILE_EXTENT_REG);
3900 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3901 new_extents[i].disk_bytenr);
3902 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3903 new_extents[i].disk_num_bytes);
3904 btrfs_set_file_extent_num_bytes(leaf, fi,
3906 ext_offset += new_extents[i].offset;
3907 btrfs_set_file_extent_offset(leaf, fi,
3909 btrfs_mark_buffer_dirty(leaf);
3911 btrfs_drop_extent_cache(inode, key.offset,
3912 key.offset + extent_len - 1, 0);
3914 ret = btrfs_inc_extent_ref(trans, root,
3915 new_extents[i].disk_bytenr,
3916 new_extents[i].disk_num_bytes,
3918 root->root_key.objectid,
3920 key.objectid, key.offset);
3922 btrfs_release_path(root, path);
3924 inode->i_blocks += extent_len >> 9;
3927 num_bytes -= extent_len;
3928 key.offset += extent_len;
3933 BUG_ON(i >= nr_extents);
3936 if (extent_locked) {
3937 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3938 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3939 lock_end, GFP_NOFS);
3943 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
3944 key.offset >= first_pos + extent_key->offset)
3951 btrfs_release_path(root, path);
3953 mutex_unlock(&inode->i_mutex);
3954 if (extent_locked) {
3955 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3956 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3957 lock_end, GFP_NOFS);
3964 int btrfs_add_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3965 u64 num_bytes, u64 new_bytenr)
3967 set_extent_bits(&root->fs_info->reloc_mapping_tree,
3968 orig_bytenr, orig_bytenr + num_bytes - 1,
3969 EXTENT_LOCKED, GFP_NOFS);
3970 set_state_private(&root->fs_info->reloc_mapping_tree,
3971 orig_bytenr, new_bytenr);
3975 int btrfs_get_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3976 u64 num_bytes, u64 *new_bytenr)
3979 u64 cur_bytenr = orig_bytenr;
3980 u64 prev_bytenr = orig_bytenr;
3984 ret = get_state_private(&root->fs_info->reloc_mapping_tree,
3985 cur_bytenr, &bytenr);
3988 prev_bytenr = cur_bytenr;
3989 cur_bytenr = bytenr;
3992 if (orig_bytenr == cur_bytenr)
3995 if (prev_bytenr != orig_bytenr) {
3996 set_state_private(&root->fs_info->reloc_mapping_tree,
3997 orig_bytenr, cur_bytenr);
3999 *new_bytenr = cur_bytenr;
4003 void btrfs_free_reloc_mappings(struct btrfs_root *root)
4005 clear_extent_bits(&root->fs_info->reloc_mapping_tree,
4006 0, (u64)-1, -1, GFP_NOFS);
4009 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4010 struct btrfs_root *root,
4011 struct extent_buffer *buf, u64 orig_start)
4016 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4017 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4019 level = btrfs_header_level(buf);
4021 struct btrfs_leaf_ref *ref;
4022 struct btrfs_leaf_ref *orig_ref;
4024 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4028 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4030 btrfs_free_leaf_ref(root, orig_ref);
4034 ref->nritems = orig_ref->nritems;
4035 memcpy(ref->extents, orig_ref->extents,
4036 sizeof(ref->extents[0]) * ref->nritems);
4038 btrfs_free_leaf_ref(root, orig_ref);
4040 ref->root_gen = trans->transid;
4041 ref->bytenr = buf->start;
4042 ref->owner = btrfs_header_owner(buf);
4043 ref->generation = btrfs_header_generation(buf);
4044 ret = btrfs_add_leaf_ref(root, ref, 0);
4046 btrfs_free_leaf_ref(root, ref);
4051 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4052 struct extent_buffer *leaf,
4053 struct btrfs_block_group_cache *group,
4054 struct btrfs_root *target_root)
4056 struct btrfs_key key;
4057 struct inode *inode = NULL;
4058 struct btrfs_file_extent_item *fi;
4060 u64 skip_objectid = 0;
4064 nritems = btrfs_header_nritems(leaf);
4065 for (i = 0; i < nritems; i++) {
4066 btrfs_item_key_to_cpu(leaf, &key, i);
4067 if (key.objectid == skip_objectid ||
4068 key.type != BTRFS_EXTENT_DATA_KEY)
4070 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4071 if (btrfs_file_extent_type(leaf, fi) ==
4072 BTRFS_FILE_EXTENT_INLINE)
4074 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4076 if (!inode || inode->i_ino != key.objectid) {
4078 inode = btrfs_ilookup(target_root->fs_info->sb,
4079 key.objectid, target_root, 1);
4082 skip_objectid = key.objectid;
4085 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4087 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4088 key.offset + num_bytes - 1, GFP_NOFS);
4089 mutex_lock(&BTRFS_I(inode)->extent_mutex);
4090 btrfs_drop_extent_cache(inode, key.offset,
4091 key.offset + num_bytes - 1, 1);
4092 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
4093 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4094 key.offset + num_bytes - 1, GFP_NOFS);
4101 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4102 struct btrfs_root *root,
4103 struct extent_buffer *leaf,
4104 struct btrfs_block_group_cache *group,
4105 struct inode *reloc_inode)
4107 struct btrfs_key key;
4108 struct btrfs_key extent_key;
4109 struct btrfs_file_extent_item *fi;
4110 struct btrfs_leaf_ref *ref;
4111 struct disk_extent *new_extent;
4120 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4121 BUG_ON(!new_extent);
4123 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4127 nritems = btrfs_header_nritems(leaf);
4128 for (i = 0; i < nritems; i++) {
4129 btrfs_item_key_to_cpu(leaf, &key, i);
4130 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4132 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4133 if (btrfs_file_extent_type(leaf, fi) ==
4134 BTRFS_FILE_EXTENT_INLINE)
4136 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4137 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4142 if (bytenr >= group->key.objectid + group->key.offset ||
4143 bytenr + num_bytes <= group->key.objectid)
4146 extent_key.objectid = bytenr;
4147 extent_key.offset = num_bytes;
4148 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4150 ret = get_new_locations(reloc_inode, &extent_key,
4151 group->key.objectid, 1,
4152 &new_extent, &nr_extent);
4157 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4158 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4159 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4160 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4162 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4163 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4164 new_extent->disk_bytenr);
4165 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4166 new_extent->disk_num_bytes);
4167 new_extent->offset += btrfs_file_extent_offset(leaf, fi);
4168 btrfs_set_file_extent_offset(leaf, fi, new_extent->offset);
4169 btrfs_mark_buffer_dirty(leaf);
4171 ret = btrfs_inc_extent_ref(trans, root,
4172 new_extent->disk_bytenr,
4173 new_extent->disk_num_bytes,
4175 root->root_key.objectid,
4177 key.objectid, key.offset);
4179 ret = btrfs_free_extent(trans, root,
4180 bytenr, num_bytes, leaf->start,
4181 btrfs_header_owner(leaf),
4182 btrfs_header_generation(leaf),
4183 key.objectid, key.offset, 0);
4188 BUG_ON(ext_index + 1 != ref->nritems);
4189 btrfs_free_leaf_ref(root, ref);
4193 int btrfs_free_reloc_root(struct btrfs_root *root)
4195 struct btrfs_root *reloc_root;
4197 if (root->reloc_root) {
4198 reloc_root = root->reloc_root;
4199 root->reloc_root = NULL;
4200 list_add(&reloc_root->dead_list,
4201 &root->fs_info->dead_reloc_roots);
4206 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4208 struct btrfs_trans_handle *trans;
4209 struct btrfs_root *reloc_root;
4210 struct btrfs_root *prev_root = NULL;
4211 struct list_head dead_roots;
4215 INIT_LIST_HEAD(&dead_roots);
4216 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4218 while (!list_empty(&dead_roots)) {
4219 reloc_root = list_entry(dead_roots.prev,
4220 struct btrfs_root, dead_list);
4221 list_del_init(&reloc_root->dead_list);
4223 BUG_ON(reloc_root->commit_root != NULL);
4225 trans = btrfs_join_transaction(root, 1);
4228 mutex_lock(&root->fs_info->drop_mutex);
4229 ret = btrfs_drop_snapshot(trans, reloc_root);
4232 mutex_unlock(&root->fs_info->drop_mutex);
4234 nr = trans->blocks_used;
4235 ret = btrfs_end_transaction(trans, root);
4237 btrfs_btree_balance_dirty(root, nr);
4240 free_extent_buffer(reloc_root->node);
4242 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4243 &reloc_root->root_key);
4245 mutex_unlock(&root->fs_info->drop_mutex);
4247 nr = trans->blocks_used;
4248 ret = btrfs_end_transaction(trans, root);
4250 btrfs_btree_balance_dirty(root, nr);
4253 prev_root = reloc_root;
4256 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4262 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4264 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4268 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4270 struct btrfs_root *reloc_root;
4271 struct btrfs_trans_handle *trans;
4272 struct btrfs_key location;
4276 mutex_lock(&root->fs_info->tree_reloc_mutex);
4277 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4279 found = !list_empty(&root->fs_info->dead_reloc_roots);
4280 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4283 trans = btrfs_start_transaction(root, 1);
4285 ret = btrfs_commit_transaction(trans, root);
4289 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4290 location.offset = (u64)-1;
4291 location.type = BTRFS_ROOT_ITEM_KEY;
4293 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4294 BUG_ON(!reloc_root);
4295 btrfs_orphan_cleanup(reloc_root);
4299 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4300 struct btrfs_root *root)
4302 struct btrfs_root *reloc_root;
4303 struct extent_buffer *eb;
4304 struct btrfs_root_item *root_item;
4305 struct btrfs_key root_key;
4308 BUG_ON(!root->ref_cows);
4309 if (root->reloc_root)
4312 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4315 ret = btrfs_copy_root(trans, root, root->commit_root,
4316 &eb, BTRFS_TREE_RELOC_OBJECTID);
4319 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4320 root_key.offset = root->root_key.objectid;
4321 root_key.type = BTRFS_ROOT_ITEM_KEY;
4323 memcpy(root_item, &root->root_item, sizeof(root_item));
4324 btrfs_set_root_refs(root_item, 0);
4325 btrfs_set_root_bytenr(root_item, eb->start);
4326 btrfs_set_root_level(root_item, btrfs_header_level(eb));
4327 memset(&root_item->drop_progress, 0, sizeof(root_item->drop_progress));
4328 root_item->drop_level = 0;
4330 btrfs_tree_unlock(eb);
4331 free_extent_buffer(eb);
4333 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4334 &root_key, root_item);
4338 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4340 BUG_ON(!reloc_root);
4341 reloc_root->last_trans = trans->transid;
4342 reloc_root->commit_root = NULL;
4343 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4345 root->reloc_root = reloc_root;
4350 * Core function of space balance.
4352 * The idea is using reloc trees to relocate tree blocks in reference
4353 * counted roots. There is one reloc tree for each subvol, all reloc
4354 * trees share same key objectid. Reloc trees are snapshots of the
4355 * latest committed roots (subvol root->commit_root). To relocate a tree
4356 * block referenced by a subvol, the code COW the block through the reloc
4357 * tree, then update pointer in the subvol to point to the new block.
4358 * Since all reloc trees share same key objectid, we can easily do special
4359 * handing to share tree blocks between reloc trees. Once a tree block has
4360 * been COWed in one reloc tree, we can use the result when the same block
4361 * is COWed again through other reloc trees.
4363 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4364 struct btrfs_root *root,
4365 struct btrfs_path *path,
4366 struct btrfs_key *first_key,
4367 struct btrfs_ref_path *ref_path,
4368 struct btrfs_block_group_cache *group,
4369 struct inode *reloc_inode)
4371 struct btrfs_root *reloc_root;
4372 struct extent_buffer *eb = NULL;
4373 struct btrfs_key *keys;
4377 int lowest_level = 0;
4381 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4382 lowest_level = ref_path->owner_objectid;
4384 if (is_cowonly_root(ref_path->root_objectid)) {
4385 path->lowest_level = lowest_level;
4386 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4388 path->lowest_level = 0;
4389 btrfs_release_path(root, path);
4393 keys = kzalloc(sizeof(*keys) * BTRFS_MAX_LEVEL, GFP_NOFS);
4395 nodes = kzalloc(sizeof(*nodes) * BTRFS_MAX_LEVEL, GFP_NOFS);
4398 mutex_lock(&root->fs_info->tree_reloc_mutex);
4399 ret = init_reloc_tree(trans, root);
4401 reloc_root = root->reloc_root;
4403 path->lowest_level = lowest_level;
4404 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 0);
4407 * get relocation mapping for tree blocks in the path
4409 lowest_merge = BTRFS_MAX_LEVEL;
4410 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4412 eb = path->nodes[level];
4413 if (!eb || eb == reloc_root->node)
4415 ret = btrfs_get_reloc_mapping(reloc_root, eb->start, eb->len,
4420 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4422 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4423 nodes[level] = new_bytenr;
4424 lowest_merge = level;
4428 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4429 eb = path->nodes[0];
4430 if (btrfs_header_generation(eb) < trans->transid)
4434 btrfs_release_path(reloc_root, path);
4436 * merge tree blocks that already relocated in other reloc trees
4438 if (lowest_merge != BTRFS_MAX_LEVEL) {
4439 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4444 * cow any tree blocks that still haven't been relocated
4446 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 1);
4449 * if we are relocating data block group, update extent pointers
4450 * in the newly created tree leaf.
4452 eb = path->nodes[0];
4453 if (update_refs && nodes[0] != eb->start) {
4454 ret = replace_extents_in_leaf(trans, reloc_root, eb, group,
4459 memset(keys, 0, sizeof(*keys) * BTRFS_MAX_LEVEL);
4460 memset(nodes, 0, sizeof(*nodes) * BTRFS_MAX_LEVEL);
4461 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4462 eb = path->nodes[level];
4463 if (!eb || eb == reloc_root->node)
4465 BUG_ON(btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID);
4466 nodes[level] = eb->start;
4468 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4470 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4473 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4474 eb = path->nodes[0];
4475 extent_buffer_get(eb);
4477 btrfs_release_path(reloc_root, path);
4479 * replace tree blocks in the fs tree with tree blocks in
4482 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4485 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4486 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4488 free_extent_buffer(eb);
4490 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4492 path->lowest_level = 0;
4498 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4499 struct btrfs_root *root,
4500 struct btrfs_path *path,
4501 struct btrfs_key *first_key,
4502 struct btrfs_ref_path *ref_path)
4507 if (root == root->fs_info->extent_root ||
4508 root == root->fs_info->chunk_root ||
4509 root == root->fs_info->dev_root) {
4511 mutex_lock(&root->fs_info->alloc_mutex);
4514 ret = relocate_one_path(trans, root, path, first_key,
4515 ref_path, NULL, NULL);
4518 if (root == root->fs_info->extent_root)
4519 btrfs_extent_post_op(trans, root);
4521 mutex_unlock(&root->fs_info->alloc_mutex);
4526 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4527 struct btrfs_root *extent_root,
4528 struct btrfs_path *path,
4529 struct btrfs_key *extent_key)
4533 mutex_lock(&extent_root->fs_info->alloc_mutex);
4534 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4537 ret = btrfs_del_item(trans, extent_root, path);
4539 btrfs_release_path(extent_root, path);
4540 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4544 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4545 struct btrfs_ref_path *ref_path)
4547 struct btrfs_key root_key;
4549 root_key.objectid = ref_path->root_objectid;
4550 root_key.type = BTRFS_ROOT_ITEM_KEY;
4551 if (is_cowonly_root(ref_path->root_objectid))
4552 root_key.offset = 0;
4554 root_key.offset = (u64)-1;
4556 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4559 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4560 struct btrfs_path *path,
4561 struct btrfs_key *extent_key,
4562 struct btrfs_block_group_cache *group,
4563 struct inode *reloc_inode, int pass)
4565 struct btrfs_trans_handle *trans;
4566 struct btrfs_root *found_root;
4567 struct btrfs_ref_path *ref_path = NULL;
4568 struct disk_extent *new_extents = NULL;
4573 struct btrfs_key first_key;
4576 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4578 trans = btrfs_start_transaction(extent_root, 1);
4581 if (extent_key->objectid == 0) {
4582 ret = del_extent_zero(trans, extent_root, path, extent_key);
4586 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4592 for (loops = 0; ; loops++) {
4594 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4595 extent_key->objectid);
4597 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4604 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4605 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4608 found_root = read_ref_root(extent_root->fs_info, ref_path);
4609 BUG_ON(!found_root);
4611 * for reference counted tree, only process reference paths
4612 * rooted at the latest committed root.
4614 if (found_root->ref_cows &&
4615 ref_path->root_generation != found_root->root_key.offset)
4618 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4621 * copy data extents to new locations
4623 u64 group_start = group->key.objectid;
4624 ret = relocate_data_extent(reloc_inode,
4633 level = ref_path->owner_objectid;
4636 if (prev_block != ref_path->nodes[level]) {
4637 struct extent_buffer *eb;
4638 u64 block_start = ref_path->nodes[level];
4639 u64 block_size = btrfs_level_size(found_root, level);
4641 eb = read_tree_block(found_root, block_start,
4643 btrfs_tree_lock(eb);
4644 BUG_ON(level != btrfs_header_level(eb));
4647 btrfs_item_key_to_cpu(eb, &first_key, 0);
4649 btrfs_node_key_to_cpu(eb, &first_key, 0);
4651 btrfs_tree_unlock(eb);
4652 free_extent_buffer(eb);
4653 prev_block = block_start;
4656 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4659 * use fallback method to process the remaining
4663 u64 group_start = group->key.objectid;
4664 ret = get_new_locations(reloc_inode,
4672 btrfs_record_root_in_trans(found_root);
4673 ret = replace_one_extent(trans, found_root,
4675 &first_key, ref_path,
4676 new_extents, nr_extents);
4682 btrfs_record_root_in_trans(found_root);
4683 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4684 ret = relocate_tree_block(trans, found_root, path,
4685 &first_key, ref_path);
4688 * try to update data extent references while
4689 * keeping metadata shared between snapshots.
4691 ret = relocate_one_path(trans, found_root, path,
4692 &first_key, ref_path,
4693 group, reloc_inode);
4700 btrfs_end_transaction(trans, extent_root);
4703 mutex_lock(&extent_root->fs_info->alloc_mutex);
4707 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4710 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4711 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4713 num_devices = root->fs_info->fs_devices->num_devices;
4714 if (num_devices == 1) {
4715 stripped |= BTRFS_BLOCK_GROUP_DUP;
4716 stripped = flags & ~stripped;
4718 /* turn raid0 into single device chunks */
4719 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4722 /* turn mirroring into duplication */
4723 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4724 BTRFS_BLOCK_GROUP_RAID10))
4725 return stripped | BTRFS_BLOCK_GROUP_DUP;
4728 /* they already had raid on here, just return */
4729 if (flags & stripped)
4732 stripped |= BTRFS_BLOCK_GROUP_DUP;
4733 stripped = flags & ~stripped;
4735 /* switch duplicated blocks with raid1 */
4736 if (flags & BTRFS_BLOCK_GROUP_DUP)
4737 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4739 /* turn single device chunks into raid0 */
4740 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4745 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4746 struct btrfs_block_group_cache *shrink_block_group,
4749 struct btrfs_trans_handle *trans;
4750 u64 new_alloc_flags;
4753 spin_lock(&shrink_block_group->lock);
4754 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4755 spin_unlock(&shrink_block_group->lock);
4756 mutex_unlock(&root->fs_info->alloc_mutex);
4758 trans = btrfs_start_transaction(root, 1);
4759 mutex_lock(&root->fs_info->alloc_mutex);
4760 spin_lock(&shrink_block_group->lock);
4762 new_alloc_flags = update_block_group_flags(root,
4763 shrink_block_group->flags);
4764 if (new_alloc_flags != shrink_block_group->flags) {
4766 btrfs_block_group_used(&shrink_block_group->item);
4768 calc = shrink_block_group->key.offset;
4770 spin_unlock(&shrink_block_group->lock);
4772 do_chunk_alloc(trans, root->fs_info->extent_root,
4773 calc + 2 * 1024 * 1024, new_alloc_flags, force);
4775 mutex_unlock(&root->fs_info->alloc_mutex);
4776 btrfs_end_transaction(trans, root);
4777 mutex_lock(&root->fs_info->alloc_mutex);
4779 spin_unlock(&shrink_block_group->lock);
4783 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4784 struct btrfs_root *root,
4785 u64 objectid, u64 size)
4787 struct btrfs_path *path;
4788 struct btrfs_inode_item *item;
4789 struct extent_buffer *leaf;
4792 path = btrfs_alloc_path();
4796 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4800 leaf = path->nodes[0];
4801 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4802 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4803 btrfs_set_inode_generation(leaf, item, 1);
4804 btrfs_set_inode_size(leaf, item, size);
4805 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4806 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM);
4807 btrfs_mark_buffer_dirty(leaf);
4808 btrfs_release_path(root, path);
4810 btrfs_free_path(path);
4814 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4815 struct btrfs_block_group_cache *group)
4817 struct inode *inode = NULL;
4818 struct btrfs_trans_handle *trans;
4819 struct btrfs_root *root;
4820 struct btrfs_key root_key;
4821 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4824 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4825 root_key.type = BTRFS_ROOT_ITEM_KEY;
4826 root_key.offset = (u64)-1;
4827 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4829 return ERR_CAST(root);
4831 trans = btrfs_start_transaction(root, 1);
4834 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4838 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4841 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4842 group->key.offset, 0);
4845 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4846 if (inode->i_state & I_NEW) {
4847 BTRFS_I(inode)->root = root;
4848 BTRFS_I(inode)->location.objectid = objectid;
4849 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4850 BTRFS_I(inode)->location.offset = 0;
4851 btrfs_read_locked_inode(inode);
4852 unlock_new_inode(inode);
4853 BUG_ON(is_bad_inode(inode));
4858 err = btrfs_orphan_add(trans, inode);
4860 btrfs_end_transaction(trans, root);
4864 inode = ERR_PTR(err);
4869 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4871 struct btrfs_trans_handle *trans;
4872 struct btrfs_path *path;
4873 struct btrfs_fs_info *info = root->fs_info;
4874 struct extent_buffer *leaf;
4875 struct inode *reloc_inode;
4876 struct btrfs_block_group_cache *block_group;
4877 struct btrfs_key key;
4885 root = root->fs_info->extent_root;
4887 block_group = btrfs_lookup_block_group(info, group_start);
4888 BUG_ON(!block_group);
4890 printk("btrfs relocating block group %llu flags %llu\n",
4891 (unsigned long long)block_group->key.objectid,
4892 (unsigned long long)block_group->flags);
4894 path = btrfs_alloc_path();
4897 reloc_inode = create_reloc_inode(info, block_group);
4898 BUG_ON(IS_ERR(reloc_inode));
4900 mutex_lock(&root->fs_info->alloc_mutex);
4902 __alloc_chunk_for_shrink(root, block_group, 1);
4903 block_group->ro = 1;
4904 block_group->space_info->total_bytes -= block_group->key.offset;
4906 mutex_unlock(&root->fs_info->alloc_mutex);
4908 btrfs_start_delalloc_inodes(info->tree_root);
4909 btrfs_wait_ordered_extents(info->tree_root, 0);
4913 key.objectid = block_group->key.objectid;
4916 cur_byte = key.objectid;
4918 trans = btrfs_start_transaction(info->tree_root, 1);
4919 btrfs_commit_transaction(trans, info->tree_root);
4921 mutex_lock(&root->fs_info->cleaner_mutex);
4922 btrfs_clean_old_snapshots(info->tree_root);
4923 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4924 mutex_unlock(&root->fs_info->cleaner_mutex);
4926 mutex_lock(&root->fs_info->alloc_mutex);
4929 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4933 leaf = path->nodes[0];
4934 nritems = btrfs_header_nritems(leaf);
4935 if (path->slots[0] >= nritems) {
4936 ret = btrfs_next_leaf(root, path);
4943 leaf = path->nodes[0];
4944 nritems = btrfs_header_nritems(leaf);
4947 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4949 if (key.objectid >= block_group->key.objectid +
4950 block_group->key.offset)
4953 if (progress && need_resched()) {
4954 btrfs_release_path(root, path);
4955 mutex_unlock(&root->fs_info->alloc_mutex);
4957 mutex_lock(&root->fs_info->alloc_mutex);
4963 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4964 key.objectid + key.offset <= cur_byte) {
4970 cur_byte = key.objectid + key.offset;
4971 btrfs_release_path(root, path);
4973 __alloc_chunk_for_shrink(root, block_group, 0);
4974 ret = relocate_one_extent(root, path, &key, block_group,
4978 key.objectid = cur_byte;
4983 btrfs_release_path(root, path);
4984 mutex_unlock(&root->fs_info->alloc_mutex);
4987 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4988 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4989 WARN_ON(reloc_inode->i_mapping->nrpages);
4992 if (total_found > 0) {
4993 printk("btrfs found %llu extents in pass %d\n",
4994 (unsigned long long)total_found, pass);
4999 /* delete reloc_inode */
5002 /* unpin extents in this range */
5003 trans = btrfs_start_transaction(info->tree_root, 1);
5004 btrfs_commit_transaction(trans, info->tree_root);
5006 mutex_lock(&root->fs_info->alloc_mutex);
5008 spin_lock(&block_group->lock);
5009 WARN_ON(block_group->pinned > 0);
5010 WARN_ON(block_group->reserved > 0);
5011 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5012 spin_unlock(&block_group->lock);
5015 mutex_unlock(&root->fs_info->alloc_mutex);
5016 btrfs_free_path(path);
5020 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5021 struct btrfs_key *key)
5024 struct btrfs_key found_key;
5025 struct extent_buffer *leaf;
5028 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5033 slot = path->slots[0];
5034 leaf = path->nodes[0];
5035 if (slot >= btrfs_header_nritems(leaf)) {
5036 ret = btrfs_next_leaf(root, path);
5043 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5045 if (found_key.objectid >= key->objectid &&
5046 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5057 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5059 struct btrfs_block_group_cache *block_group;
5062 mutex_lock(&info->alloc_mutex);
5063 spin_lock(&info->block_group_cache_lock);
5064 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5065 block_group = rb_entry(n, struct btrfs_block_group_cache,
5068 spin_unlock(&info->block_group_cache_lock);
5069 btrfs_remove_free_space_cache(block_group);
5070 spin_lock(&info->block_group_cache_lock);
5072 rb_erase(&block_group->cache_node,
5073 &info->block_group_cache_tree);
5074 spin_lock(&block_group->space_info->lock);
5075 list_del(&block_group->list);
5076 spin_unlock(&block_group->space_info->lock);
5079 spin_unlock(&info->block_group_cache_lock);
5080 mutex_unlock(&info->alloc_mutex);
5084 int btrfs_read_block_groups(struct btrfs_root *root)
5086 struct btrfs_path *path;
5088 struct btrfs_block_group_cache *cache;
5089 struct btrfs_fs_info *info = root->fs_info;
5090 struct btrfs_space_info *space_info;
5091 struct btrfs_key key;
5092 struct btrfs_key found_key;
5093 struct extent_buffer *leaf;
5095 root = info->extent_root;
5098 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5099 path = btrfs_alloc_path();
5103 mutex_lock(&root->fs_info->alloc_mutex);
5105 ret = find_first_block_group(root, path, &key);
5113 leaf = path->nodes[0];
5114 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5115 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5121 spin_lock_init(&cache->lock);
5122 INIT_LIST_HEAD(&cache->list);
5123 read_extent_buffer(leaf, &cache->item,
5124 btrfs_item_ptr_offset(leaf, path->slots[0]),
5125 sizeof(cache->item));
5126 memcpy(&cache->key, &found_key, sizeof(found_key));
5128 key.objectid = found_key.objectid + found_key.offset;
5129 btrfs_release_path(root, path);
5130 cache->flags = btrfs_block_group_flags(&cache->item);
5132 ret = update_space_info(info, cache->flags, found_key.offset,
5133 btrfs_block_group_used(&cache->item),
5136 cache->space_info = space_info;
5137 spin_lock(&space_info->lock);
5138 list_add(&cache->list, &space_info->block_groups);
5139 spin_unlock(&space_info->lock);
5141 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5146 btrfs_free_path(path);
5147 mutex_unlock(&root->fs_info->alloc_mutex);
5151 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5152 struct btrfs_root *root, u64 bytes_used,
5153 u64 type, u64 chunk_objectid, u64 chunk_offset,
5157 struct btrfs_root *extent_root;
5158 struct btrfs_block_group_cache *cache;
5160 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5161 extent_root = root->fs_info->extent_root;
5163 root->fs_info->last_trans_new_blockgroup = trans->transid;
5165 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5169 cache->key.objectid = chunk_offset;
5170 cache->key.offset = size;
5171 spin_lock_init(&cache->lock);
5172 INIT_LIST_HEAD(&cache->list);
5173 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5175 btrfs_set_block_group_used(&cache->item, bytes_used);
5176 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5177 cache->flags = type;
5178 btrfs_set_block_group_flags(&cache->item, type);
5180 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5181 &cache->space_info);
5183 spin_lock(&cache->space_info->lock);
5184 list_add(&cache->list, &cache->space_info->block_groups);
5185 spin_unlock(&cache->space_info->lock);
5187 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5190 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5191 sizeof(cache->item));
5194 finish_current_insert(trans, extent_root);
5195 ret = del_pending_extents(trans, extent_root);
5197 set_avail_alloc_bits(extent_root->fs_info, type);
5202 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5203 struct btrfs_root *root, u64 group_start)
5205 struct btrfs_path *path;
5206 struct btrfs_block_group_cache *block_group;
5207 struct btrfs_key key;
5210 BUG_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5211 root = root->fs_info->extent_root;
5213 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5214 BUG_ON(!block_group);
5216 memcpy(&key, &block_group->key, sizeof(key));
5218 path = btrfs_alloc_path();
5221 btrfs_remove_free_space_cache(block_group);
5222 rb_erase(&block_group->cache_node,
5223 &root->fs_info->block_group_cache_tree);
5224 spin_lock(&block_group->space_info->lock);
5225 list_del(&block_group->list);
5226 spin_unlock(&block_group->space_info->lock);
5229 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5230 kfree(shrink_block_group);
5233 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5239 ret = btrfs_del_item(trans, root, path);
5241 btrfs_free_path(path);