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.
21 #include "kerncompat.h"
22 #include "radix-tree.h"
25 #include "print-tree.h"
26 #include "transaction.h"
30 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
36 #define PENDING_EXTENT_INSERT 0
37 #define PENDING_EXTENT_DELETE 1
38 #define PENDING_BACKREF_UPDATE 2
40 struct pending_extent_op {
51 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
52 btrfs_root *extent_root);
53 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
54 btrfs_root *extent_root);
56 void maybe_lock_mutex(struct btrfs_root *root)
60 void maybe_unlock_mutex(struct btrfs_root *root)
64 static int cache_block_group(struct btrfs_root *root,
65 struct btrfs_block_group_cache *block_group)
67 struct btrfs_path *path;
70 struct extent_buffer *leaf;
71 struct extent_io_tree *free_space_cache;
81 root = root->fs_info->extent_root;
82 free_space_cache = &root->fs_info->free_space_cache;
84 if (block_group->cached)
87 path = btrfs_alloc_path();
92 first_free = block_group->key.objectid;
93 key.objectid = block_group->key.objectid;
95 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
96 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
99 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
103 leaf = path->nodes[0];
104 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
105 if (key.objectid + key.offset > first_free)
106 first_free = key.objectid + key.offset;
109 leaf = path->nodes[0];
110 slot = path->slots[0];
111 if (slot >= btrfs_header_nritems(leaf)) {
112 ret = btrfs_next_leaf(root, path);
121 btrfs_item_key_to_cpu(leaf, &key, slot);
122 if (key.objectid < block_group->key.objectid) {
125 if (key.objectid >= block_group->key.objectid +
126 block_group->key.offset) {
130 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
135 if (key.objectid > last) {
136 hole_size = key.objectid - last;
137 set_extent_dirty(free_space_cache, last,
138 last + hole_size - 1,
141 last = key.objectid + key.offset;
149 if (block_group->key.objectid +
150 block_group->key.offset > last) {
151 hole_size = block_group->key.objectid +
152 block_group->key.offset - last;
153 set_extent_dirty(free_space_cache, last,
154 last + hole_size - 1, GFP_NOFS);
156 block_group->cached = 1;
158 btrfs_free_path(path);
162 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
166 struct extent_io_tree *block_group_cache;
167 struct btrfs_block_group_cache *block_group = NULL;
173 bytenr = max_t(u64, bytenr,
174 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
175 block_group_cache = &info->block_group_cache;
176 ret = find_first_extent_bit(block_group_cache,
177 bytenr, &start, &end,
178 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
183 ret = get_state_private(block_group_cache, start, &ptr);
187 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
191 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
195 struct extent_io_tree *block_group_cache;
196 struct btrfs_block_group_cache *block_group = NULL;
202 block_group_cache = &info->block_group_cache;
203 ret = find_first_extent_bit(block_group_cache,
204 bytenr, &start, &end,
205 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
210 ret = get_state_private(block_group_cache, start, &ptr);
214 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
215 if (block_group->key.objectid <= bytenr && bytenr <
216 block_group->key.objectid + block_group->key.offset)
221 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
223 return (cache->flags & bits) == bits;
226 static int noinline find_search_start(struct btrfs_root *root,
227 struct btrfs_block_group_cache **cache_ret,
228 u64 *start_ret, int num, int data)
231 struct btrfs_block_group_cache *cache = *cache_ret;
235 u64 search_start = *start_ret;
242 ret = cache_block_group(root, cache);
246 last = max(search_start, cache->key.objectid);
247 if (cache->ro || !block_group_bits(cache, data)) {
252 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
253 last, &start, &end, EXTENT_DIRTY);
258 start = max(last, start);
260 if (last - start < num) {
263 if (start + num > cache->key.objectid + cache->key.offset) {
270 cache = btrfs_lookup_block_group(root->fs_info, search_start);
272 printk("Unable to find block group for %llu\n",
273 (unsigned long long)search_start);
279 last = cache->key.objectid + cache->key.offset;
281 cache = btrfs_lookup_first_block_group(root->fs_info, last);
291 cache = btrfs_find_block_group(root, cache, last, data, 0);
292 cache = btrfs_find_block_group(root, cache, last, data, 0);
300 static u64 div_factor(u64 num, int factor)
309 static int block_group_state_bits(u64 flags)
312 if (flags & BTRFS_BLOCK_GROUP_DATA)
313 bits |= BLOCK_GROUP_DATA;
314 if (flags & BTRFS_BLOCK_GROUP_METADATA)
315 bits |= BLOCK_GROUP_METADATA;
316 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
317 bits |= BLOCK_GROUP_SYSTEM;
321 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
322 struct btrfs_block_group_cache
323 *hint, u64 search_start,
326 struct btrfs_block_group_cache *cache;
327 struct extent_io_tree *block_group_cache;
328 struct btrfs_block_group_cache *found_group = NULL;
329 struct btrfs_fs_info *info = root->fs_info;
342 block_group_cache = &info->block_group_cache;
347 bit = block_group_state_bits(data);
350 struct btrfs_block_group_cache *shint;
351 shint = btrfs_lookup_block_group(info, search_start);
352 if (shint && !shint->ro && block_group_bits(shint, data)) {
353 used = btrfs_block_group_used(&shint->item);
354 if (used + shint->pinned <
355 div_factor(shint->key.offset, factor)) {
360 if (hint && !hint->ro && block_group_bits(hint, data)) {
361 used = btrfs_block_group_used(&hint->item);
362 if (used + hint->pinned <
363 div_factor(hint->key.offset, factor)) {
366 last = hint->key.objectid + hint->key.offset;
370 hint_last = max(hint->key.objectid, search_start);
372 hint_last = search_start;
378 ret = find_first_extent_bit(block_group_cache, last,
383 ret = get_state_private(block_group_cache, start, &ptr);
387 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
388 last = cache->key.objectid + cache->key.offset;
389 used = btrfs_block_group_used(&cache->item);
391 if (!cache->ro && block_group_bits(cache, data)) {
393 free_check = cache->key.offset;
395 free_check = div_factor(cache->key.offset,
398 if (used + cache->pinned < free_check) {
415 * Back reference rules. Back refs have three main goals:
417 * 1) differentiate between all holders of references to an extent so that
418 * when a reference is dropped we can make sure it was a valid reference
419 * before freeing the extent.
421 * 2) Provide enough information to quickly find the holders of an extent
422 * if we notice a given block is corrupted or bad.
424 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
425 * maintenance. This is actually the same as #2, but with a slightly
426 * different use case.
428 * File extents can be referenced by:
430 * - multiple snapshots, subvolumes, or different generations in one subvol
431 * - different files inside a single subvolume
432 * - different offsets inside a file (bookend extents in file.c)
434 * The extent ref structure has fields for:
436 * - Objectid of the subvolume root
437 * - Generation number of the tree holding the reference
438 * - objectid of the file holding the reference
439 * - offset in the file corresponding to the key holding the reference
440 * - number of references holding by parent node (alway 1 for tree blocks)
442 * Btree leaf may hold multiple references to a file extent. In most cases,
443 * these references are from same file and the corresponding offsets inside
444 * the file are close together. So inode objectid and offset in file are
445 * just hints, they provide hints about where in the btree the references
446 * can be found and when we can stop searching.
448 * When a file extent is allocated the fields are filled in:
449 * (root_key.objectid, trans->transid, inode objectid, offset in file, 1)
451 * When a leaf is cow'd new references are added for every file extent found
452 * in the leaf. It looks similar to the create case, but trans->transid will
453 * be different when the block is cow'd.
455 * (root_key.objectid, trans->transid, inode objectid, offset in file,
456 * number of references in the leaf)
458 * Because inode objectid and offset in file are just hints, they are not
459 * used when backrefs are deleted. When a file extent is removed either
460 * during snapshot deletion or file truncation, we find the corresponding
461 * back back reference and check the following fields.
463 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf))
465 * Btree extents can be referenced by:
467 * - Different subvolumes
468 * - Different generations of the same subvolume
470 * When a tree block is created, back references are inserted:
472 * (root->root_key.objectid, trans->transid, level, 0, 1)
474 * When a tree block is cow'd, new back references are added for all the
475 * blocks it points to. If the tree block isn't in reference counted root,
476 * the old back references are removed. These new back references are of
477 * the form (trans->transid will have increased since creation):
479 * (root->root_key.objectid, trans->transid, level, 0, 1)
481 * When a backref is in deleting, the following fields are checked:
483 * if backref was for a tree root:
484 * (btrfs_header_owner(itself), btrfs_header_generation(itself))
486 * (btrfs_header_owner(parent), btrfs_header_generation(parent))
488 * Back Reference Key composing:
490 * The key objectid corresponds to the first byte in the extent, the key
491 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
492 * byte of parent extent. If a extent is tree root, the key offset is set
493 * to the key objectid.
496 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
497 struct btrfs_root *root,
498 struct btrfs_path *path,
499 u64 bytenr, u64 parent,
500 u64 ref_root, u64 ref_generation,
501 u64 owner_objectid, int del)
503 struct btrfs_key key;
504 struct btrfs_extent_ref *ref;
505 struct extent_buffer *leaf;
509 key.objectid = bytenr;
510 key.type = BTRFS_EXTENT_REF_KEY;
513 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
521 leaf = path->nodes[0];
522 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
523 ref_objectid = btrfs_ref_objectid(leaf, ref);
524 if (btrfs_ref_root(leaf, ref) != ref_root ||
525 btrfs_ref_generation(leaf, ref) != ref_generation ||
526 (ref_objectid != owner_objectid &&
527 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
537 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
538 struct btrfs_root *root,
539 struct btrfs_path *path,
540 u64 bytenr, u64 parent,
541 u64 ref_root, u64 ref_generation,
544 struct btrfs_key key;
545 struct extent_buffer *leaf;
546 struct btrfs_extent_ref *ref;
550 key.objectid = bytenr;
551 key.type = BTRFS_EXTENT_REF_KEY;
554 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
556 leaf = path->nodes[0];
557 ref = btrfs_item_ptr(leaf, path->slots[0],
558 struct btrfs_extent_ref);
559 btrfs_set_ref_root(leaf, ref, ref_root);
560 btrfs_set_ref_generation(leaf, ref, ref_generation);
561 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
562 btrfs_set_ref_num_refs(leaf, ref, 1);
563 } else if (ret == -EEXIST) {
565 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
566 leaf = path->nodes[0];
567 ref = btrfs_item_ptr(leaf, path->slots[0],
568 struct btrfs_extent_ref);
569 if (btrfs_ref_root(leaf, ref) != ref_root ||
570 btrfs_ref_generation(leaf, ref) != ref_generation) {
576 num_refs = btrfs_ref_num_refs(leaf, ref);
577 BUG_ON(num_refs == 0);
578 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
580 existing_owner = btrfs_ref_objectid(leaf, ref);
581 if (existing_owner != owner_objectid &&
582 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
583 btrfs_set_ref_objectid(leaf, ref,
584 BTRFS_MULTIPLE_OBJECTIDS);
590 btrfs_mark_buffer_dirty(path->nodes[0]);
592 btrfs_release_path(root, path);
596 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
597 struct btrfs_root *root,
598 struct btrfs_path *path)
600 struct extent_buffer *leaf;
601 struct btrfs_extent_ref *ref;
605 leaf = path->nodes[0];
606 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
607 num_refs = btrfs_ref_num_refs(leaf, ref);
608 BUG_ON(num_refs == 0);
611 ret = btrfs_del_item(trans, root, path);
613 btrfs_set_ref_num_refs(leaf, ref, num_refs);
614 btrfs_mark_buffer_dirty(leaf);
616 btrfs_release_path(root, path);
620 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
621 struct btrfs_root *root, u64 bytenr,
622 u64 orig_parent, u64 parent,
623 u64 orig_root, u64 ref_root,
624 u64 orig_generation, u64 ref_generation,
628 struct btrfs_root *extent_root = root->fs_info->extent_root;
629 struct btrfs_path *path;
631 if (root == root->fs_info->extent_root) {
632 struct pending_extent_op *extent_op;
635 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
636 num_bytes = btrfs_level_size(root, (int)owner_objectid);
637 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
638 bytenr + num_bytes - 1, EXTENT_LOCKED, 0)) {
640 ret = get_state_private(&root->fs_info->extent_ins,
643 extent_op = (struct pending_extent_op *)
645 BUG_ON(extent_op->parent != orig_parent);
646 BUG_ON(extent_op->generation != orig_generation);
647 extent_op->parent = parent;
648 extent_op->generation = ref_generation;
650 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
653 extent_op->type = PENDING_BACKREF_UPDATE;
654 extent_op->bytenr = bytenr;
655 extent_op->num_bytes = num_bytes;
656 extent_op->parent = parent;
657 extent_op->orig_parent = orig_parent;
658 extent_op->generation = ref_generation;
659 extent_op->orig_generation = orig_generation;
660 extent_op->level = (int)owner_objectid;
662 set_extent_bits(&root->fs_info->extent_ins,
663 bytenr, bytenr + num_bytes - 1,
664 EXTENT_LOCKED, GFP_NOFS);
665 set_state_private(&root->fs_info->extent_ins,
666 bytenr, (unsigned long)extent_op);
671 path = btrfs_alloc_path();
674 ret = lookup_extent_backref(trans, extent_root, path,
675 bytenr, orig_parent, orig_root,
676 orig_generation, owner_objectid, 1);
679 ret = remove_extent_backref(trans, extent_root, path);
682 ret = insert_extent_backref(trans, extent_root, path, bytenr,
683 parent, ref_root, ref_generation,
686 finish_current_insert(trans, extent_root);
687 del_pending_extents(trans, extent_root);
689 btrfs_free_path(path);
693 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root, u64 bytenr,
695 u64 orig_parent, u64 parent,
696 u64 ref_root, u64 ref_generation,
700 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
701 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
703 maybe_lock_mutex(root);
704 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
705 parent, ref_root, ref_root,
706 ref_generation, ref_generation,
708 maybe_unlock_mutex(root);
712 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
713 struct btrfs_root *root, u64 bytenr,
714 u64 orig_parent, u64 parent,
715 u64 orig_root, u64 ref_root,
716 u64 orig_generation, u64 ref_generation,
719 struct btrfs_path *path;
721 struct btrfs_key key;
722 struct extent_buffer *l;
723 struct btrfs_extent_item *item;
726 path = btrfs_alloc_path();
731 key.objectid = bytenr;
732 key.type = BTRFS_EXTENT_ITEM_KEY;
733 key.offset = (u64)-1;
735 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
739 BUG_ON(ret == 0 || path->slots[0] == 0);
744 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
745 BUG_ON(key.objectid != bytenr);
746 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
748 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
749 refs = btrfs_extent_refs(l, item);
750 btrfs_set_extent_refs(l, item, refs + 1);
751 btrfs_mark_buffer_dirty(path->nodes[0]);
753 btrfs_release_path(root->fs_info->extent_root, path);
756 ret = insert_extent_backref(trans, root->fs_info->extent_root,
757 path, bytenr, parent,
758 ref_root, ref_generation,
761 finish_current_insert(trans, root->fs_info->extent_root);
762 del_pending_extents(trans, root->fs_info->extent_root);
764 btrfs_free_path(path);
768 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
769 struct btrfs_root *root,
770 u64 bytenr, u64 num_bytes, u64 parent,
771 u64 ref_root, u64 ref_generation,
775 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
776 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
778 maybe_lock_mutex(root);
779 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
780 0, ref_root, 0, ref_generation,
782 maybe_unlock_mutex(root);
786 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
787 struct btrfs_root *root)
789 finish_current_insert(trans, root->fs_info->extent_root);
790 del_pending_extents(trans, root->fs_info->extent_root);
794 int lookup_extent_ref(struct btrfs_trans_handle *trans,
795 struct btrfs_root *root, u64 bytenr,
796 u64 num_bytes, u32 *refs)
798 struct btrfs_path *path;
800 struct btrfs_key key;
801 struct extent_buffer *l;
802 struct btrfs_extent_item *item;
804 WARN_ON(num_bytes < root->sectorsize);
805 path = btrfs_alloc_path();
807 key.objectid = bytenr;
808 key.offset = num_bytes;
809 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
810 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
815 btrfs_print_leaf(root, path->nodes[0]);
816 printk("failed to find block number %Lu\n", bytenr);
820 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
821 *refs = btrfs_extent_refs(l, item);
823 btrfs_free_path(path);
827 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
828 struct extent_buffer *orig_buf, struct extent_buffer *buf,
837 u32 nr_file_extents = 0;
838 struct btrfs_key key;
839 struct btrfs_file_extent_item *fi;
844 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
845 u64, u64, u64, u64, u64, u64, u64, u64);
847 ref_root = btrfs_header_owner(buf);
848 ref_generation = btrfs_header_generation(buf);
849 orig_root = btrfs_header_owner(orig_buf);
850 orig_generation = btrfs_header_generation(orig_buf);
852 nritems = btrfs_header_nritems(buf);
853 level = btrfs_header_level(buf);
855 if (root->ref_cows) {
856 process_func = __btrfs_inc_extent_ref;
859 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
861 process_func = __btrfs_update_extent_ref;
864 for (i = 0; i < nritems; i++) {
867 btrfs_item_key_to_cpu(buf, &key, i);
868 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
870 fi = btrfs_item_ptr(buf, i,
871 struct btrfs_file_extent_item);
872 if (btrfs_file_extent_type(buf, fi) ==
873 BTRFS_FILE_EXTENT_INLINE)
875 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
881 maybe_lock_mutex(root);
882 ret = process_func(trans, root, bytenr,
883 orig_buf->start, buf->start,
885 orig_generation, ref_generation,
887 maybe_unlock_mutex(root);
895 bytenr = btrfs_node_blockptr(buf, i);
896 maybe_lock_mutex(root);
897 ret = process_func(trans, root, bytenr,
898 orig_buf->start, buf->start,
900 orig_generation, ref_generation,
902 maybe_unlock_mutex(root);
913 *nr_extents = nr_file_extents;
915 *nr_extents = nritems;
921 for (i =0; i < faili; i++) {
924 btrfs_item_key_to_cpu(buf, &key, i);
925 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
927 fi = btrfs_item_ptr(buf, i,
928 struct btrfs_file_extent_item);
929 if (btrfs_file_extent_type(buf, fi) ==
930 BTRFS_FILE_EXTENT_INLINE)
932 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
933 if (disk_bytenr == 0)
935 err = btrfs_free_extent(trans, root, disk_bytenr,
936 btrfs_file_extent_disk_num_bytes(buf,
940 bytenr = btrfs_node_blockptr(buf, i);
941 err = btrfs_free_extent(trans, root, bytenr,
942 btrfs_level_size(root, level - 1), 0);
950 int btrfs_update_ref(struct btrfs_trans_handle *trans,
951 struct btrfs_root *root, struct extent_buffer *orig_buf,
952 struct extent_buffer *buf, int start_slot, int nr)
960 struct btrfs_key key;
961 struct btrfs_file_extent_item *fi;
967 BUG_ON(start_slot < 0);
968 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
970 ref_root = btrfs_header_owner(buf);
971 ref_generation = btrfs_header_generation(buf);
972 orig_root = btrfs_header_owner(orig_buf);
973 orig_generation = btrfs_header_generation(orig_buf);
974 level = btrfs_header_level(buf);
976 if (!root->ref_cows) {
978 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
982 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
985 btrfs_item_key_to_cpu(buf, &key, slot);
986 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
988 fi = btrfs_item_ptr(buf, slot,
989 struct btrfs_file_extent_item);
990 if (btrfs_file_extent_type(buf, fi) ==
991 BTRFS_FILE_EXTENT_INLINE)
993 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
997 maybe_lock_mutex(root);
998 ret = __btrfs_update_extent_ref(trans, root, bytenr,
999 orig_buf->start, buf->start,
1000 orig_root, ref_root,
1001 orig_generation, ref_generation,
1003 maybe_unlock_mutex(root);
1007 bytenr = btrfs_node_blockptr(buf, slot);
1008 maybe_lock_mutex(root);
1009 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1010 orig_buf->start, buf->start,
1011 orig_root, ref_root,
1012 orig_generation, ref_generation,
1014 maybe_unlock_mutex(root);
1025 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1026 struct btrfs_root *root,
1027 struct btrfs_path *path,
1028 struct btrfs_block_group_cache *cache)
1032 struct btrfs_root *extent_root = root->fs_info->extent_root;
1034 struct extent_buffer *leaf;
1036 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1041 leaf = path->nodes[0];
1042 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1043 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1044 btrfs_mark_buffer_dirty(leaf);
1045 btrfs_release_path(extent_root, path);
1047 finish_current_insert(trans, extent_root);
1048 pending_ret = del_pending_extents(trans, extent_root);
1057 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1058 struct btrfs_root *root)
1060 struct extent_io_tree *block_group_cache;
1061 struct btrfs_block_group_cache *cache;
1065 struct btrfs_path *path;
1071 block_group_cache = &root->fs_info->block_group_cache;
1072 path = btrfs_alloc_path();
1077 ret = find_first_extent_bit(block_group_cache, last,
1078 &start, &end, BLOCK_GROUP_DIRTY);
1083 ret = get_state_private(block_group_cache, start, &ptr);
1086 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1087 err = write_one_cache_group(trans, root,
1090 * if we fail to write the cache group, we want
1091 * to keep it marked dirty in hopes that a later
1098 clear_extent_bits(block_group_cache, start, end,
1099 BLOCK_GROUP_DIRTY, GFP_NOFS);
1101 btrfs_free_path(path);
1105 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1108 struct list_head *head = &info->space_info;
1109 struct list_head *cur;
1110 struct btrfs_space_info *found;
1111 list_for_each(cur, head) {
1112 found = list_entry(cur, struct btrfs_space_info, list);
1113 if (found->flags == flags)
1120 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1121 u64 total_bytes, u64 bytes_used,
1122 struct btrfs_space_info **space_info)
1124 struct btrfs_space_info *found;
1126 found = __find_space_info(info, flags);
1128 found->total_bytes += total_bytes;
1129 found->bytes_used += bytes_used;
1130 WARN_ON(found->total_bytes < found->bytes_used);
1131 *space_info = found;
1134 found = kmalloc(sizeof(*found), GFP_NOFS);
1138 list_add(&found->list, &info->space_info);
1139 found->flags = flags;
1140 found->total_bytes = total_bytes;
1141 found->bytes_used = bytes_used;
1142 found->bytes_pinned = 0;
1144 *space_info = found;
1149 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1151 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1152 BTRFS_BLOCK_GROUP_RAID1 |
1153 BTRFS_BLOCK_GROUP_DUP);
1155 if (flags & BTRFS_BLOCK_GROUP_DATA)
1156 fs_info->avail_data_alloc_bits |= extra_flags;
1157 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1158 fs_info->avail_metadata_alloc_bits |= extra_flags;
1159 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1160 fs_info->avail_system_alloc_bits |= extra_flags;
1164 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1165 struct btrfs_root *extent_root, u64 alloc_bytes,
1168 struct btrfs_space_info *space_info;
1174 space_info = __find_space_info(extent_root->fs_info, flags);
1176 ret = update_space_info(extent_root->fs_info, flags,
1180 BUG_ON(!space_info);
1182 if (space_info->full)
1185 thresh = div_factor(space_info->total_bytes, 7);
1186 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1190 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1191 if (ret == -ENOSPC) {
1192 space_info->full = 1;
1198 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1199 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1204 static int update_block_group(struct btrfs_trans_handle *trans,
1205 struct btrfs_root *root,
1206 u64 bytenr, u64 num_bytes, int alloc,
1209 struct btrfs_block_group_cache *cache;
1210 struct btrfs_fs_info *info = root->fs_info;
1211 u64 total = num_bytes;
1218 cache = btrfs_lookup_block_group(info, bytenr);
1222 byte_in_group = bytenr - cache->key.objectid;
1223 WARN_ON(byte_in_group > cache->key.offset);
1224 start = cache->key.objectid;
1225 end = start + cache->key.offset - 1;
1226 set_extent_bits(&info->block_group_cache, start, end,
1227 BLOCK_GROUP_DIRTY, GFP_NOFS);
1229 old_val = btrfs_block_group_used(&cache->item);
1230 num_bytes = min(total, cache->key.offset - byte_in_group);
1232 old_val += num_bytes;
1233 cache->space_info->bytes_used += num_bytes;
1235 old_val -= num_bytes;
1236 cache->space_info->bytes_used -= num_bytes;
1238 set_extent_dirty(&info->free_space_cache,
1239 bytenr, bytenr + num_bytes - 1,
1243 btrfs_set_block_group_used(&cache->item, old_val);
1245 bytenr += num_bytes;
1250 static int update_pinned_extents(struct btrfs_root *root,
1251 u64 bytenr, u64 num, int pin)
1254 struct btrfs_block_group_cache *cache;
1255 struct btrfs_fs_info *fs_info = root->fs_info;
1258 set_extent_dirty(&fs_info->pinned_extents,
1259 bytenr, bytenr + num - 1, GFP_NOFS);
1261 clear_extent_dirty(&fs_info->pinned_extents,
1262 bytenr, bytenr + num - 1, GFP_NOFS);
1265 cache = btrfs_lookup_block_group(fs_info, bytenr);
1267 len = min(num, cache->key.offset -
1268 (bytenr - cache->key.objectid));
1270 cache->pinned += len;
1271 cache->space_info->bytes_pinned += len;
1272 fs_info->total_pinned += len;
1274 cache->pinned -= len;
1275 cache->space_info->bytes_pinned -= len;
1276 fs_info->total_pinned -= len;
1284 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1289 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1293 ret = find_first_extent_bit(pinned_extents, last,
1294 &start, &end, EXTENT_DIRTY);
1297 set_extent_dirty(copy, start, end, GFP_NOFS);
1303 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1304 struct btrfs_root *root,
1305 struct extent_io_tree *unpin)
1310 struct extent_io_tree *free_space_cache;
1311 free_space_cache = &root->fs_info->free_space_cache;
1314 ret = find_first_extent_bit(unpin, 0, &start, &end,
1318 update_pinned_extents(root, start, end + 1 - start, 0);
1319 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1320 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1325 static int finish_current_insert(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *extent_root)
1331 struct btrfs_fs_info *info = extent_root->fs_info;
1332 struct btrfs_path *path;
1333 struct btrfs_extent_ref *ref;
1334 struct pending_extent_op *extent_op;
1335 struct btrfs_key key;
1336 struct btrfs_extent_item extent_item;
1340 btrfs_set_stack_extent_refs(&extent_item, 1);
1341 path = btrfs_alloc_path();
1344 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1345 &end, EXTENT_LOCKED);
1349 ret = get_state_private(&info->extent_ins, start, &priv);
1351 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1353 if (extent_op->type == PENDING_EXTENT_INSERT) {
1354 key.objectid = start;
1355 key.offset = end + 1 - start;
1356 key.type = BTRFS_EXTENT_ITEM_KEY;
1357 err = btrfs_insert_item(trans, extent_root, &key,
1358 &extent_item, sizeof(extent_item));
1361 clear_extent_bits(&info->extent_ins, start, end,
1362 EXTENT_LOCKED, GFP_NOFS);
1364 err = insert_extent_backref(trans, extent_root, path,
1365 start, extent_op->parent,
1366 extent_root->root_key.objectid,
1367 extent_op->generation,
1370 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1371 err = lookup_extent_backref(trans, extent_root, path,
1372 start, extent_op->orig_parent,
1373 extent_root->root_key.objectid,
1374 extent_op->orig_generation,
1375 extent_op->level, 0);
1378 clear_extent_bits(&info->extent_ins, start, end,
1379 EXTENT_LOCKED, GFP_NOFS);
1381 key.objectid = start;
1382 key.offset = extent_op->parent;
1383 key.type = BTRFS_EXTENT_REF_KEY;
1384 err = btrfs_set_item_key_safe(trans, extent_root, path,
1387 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1388 struct btrfs_extent_ref);
1389 btrfs_set_ref_generation(path->nodes[0], ref,
1390 extent_op->generation);
1391 btrfs_mark_buffer_dirty(path->nodes[0]);
1392 btrfs_release_path(extent_root, path);
1398 btrfs_free_path(path);
1402 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 u64 bytenr, u64 num_bytes, int is_data)
1407 struct extent_buffer *buf;
1412 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1416 /* we can reuse a block if it hasn't been written
1417 * and it is from this transaction. We can't
1418 * reuse anything from the tree log root because
1419 * it has tiny sub-transactions.
1421 if (btrfs_buffer_uptodate(buf, 0)) {
1422 u64 header_owner = btrfs_header_owner(buf);
1423 u64 header_transid = btrfs_header_generation(buf);
1424 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1425 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1426 header_transid == trans->transid &&
1427 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1428 clean_tree_block(NULL, root, buf);
1429 free_extent_buffer(buf);
1433 free_extent_buffer(buf);
1435 update_pinned_extents(root, bytenr, num_bytes, 1);
1442 * remove an extent from the root, returns 0 on success
1444 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1445 *root, u64 bytenr, u64 num_bytes, u64 parent,
1446 u64 root_objectid, u64 ref_generation,
1447 u64 owner_objectid, int pin, int mark_free)
1449 struct btrfs_path *path;
1450 struct btrfs_key key;
1451 struct btrfs_fs_info *info = root->fs_info;
1452 struct btrfs_extent_ops *ops = info->extent_ops;
1453 struct btrfs_root *extent_root = info->extent_root;
1454 struct extent_buffer *leaf;
1456 int extent_slot = 0;
1457 int found_extent = 0;
1459 struct btrfs_extent_item *ei;
1462 key.objectid = bytenr;
1463 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1464 key.offset = num_bytes;
1466 path = btrfs_alloc_path();
1470 ret = lookup_extent_backref(trans, extent_root, path,
1471 bytenr, parent, root_objectid,
1472 ref_generation, owner_objectid, 1);
1474 struct btrfs_key found_key;
1475 extent_slot = path->slots[0];
1476 while(extent_slot > 0) {
1478 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1480 if (found_key.objectid != bytenr)
1482 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1483 found_key.offset == num_bytes) {
1487 if (path->slots[0] - extent_slot > 5)
1490 if (!found_extent) {
1491 ret = remove_extent_backref(trans, extent_root, path);
1493 btrfs_release_path(extent_root, path);
1494 ret = btrfs_search_slot(trans, extent_root,
1497 extent_slot = path->slots[0];
1500 btrfs_print_leaf(extent_root, path->nodes[0]);
1501 printk("Unable to find ref byte nr %llu root %llu "
1502 " gen %llu owner %llu\n",
1503 (unsigned long long)bytenr,
1504 (unsigned long long)root_objectid,
1505 (unsigned long long)ref_generation,
1506 (unsigned long long)owner_objectid);
1510 leaf = path->nodes[0];
1511 ei = btrfs_item_ptr(leaf, extent_slot,
1512 struct btrfs_extent_item);
1513 refs = btrfs_extent_refs(leaf, ei);
1516 btrfs_set_extent_refs(leaf, ei, refs);
1518 btrfs_mark_buffer_dirty(leaf);
1520 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1521 struct btrfs_extent_ref *ref;
1522 ref = btrfs_item_ptr(leaf, path->slots[0],
1523 struct btrfs_extent_ref);
1524 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1525 /* if the back ref and the extent are next to each other
1526 * they get deleted below in one shot
1528 path->slots[0] = extent_slot;
1530 } else if (found_extent) {
1531 /* otherwise delete the extent back ref */
1532 ret = remove_extent_backref(trans, extent_root, path);
1534 /* if refs are 0, we need to setup the path for deletion */
1536 btrfs_release_path(extent_root, path);
1537 ret = btrfs_search_slot(trans, extent_root, &key, path,
1550 ret = pin_down_bytes(trans, root, bytenr, num_bytes, 0);
1556 /* block accounting for super block */
1557 super_used = btrfs_super_bytes_used(&info->super_copy);
1558 btrfs_set_super_bytes_used(&info->super_copy,
1559 super_used - num_bytes);
1561 /* block accounting for root item */
1562 root_used = btrfs_root_used(&root->root_item);
1563 btrfs_set_root_used(&root->root_item,
1564 root_used - num_bytes);
1565 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1570 if (ops && ops->free_extent)
1571 ops->free_extent(root, bytenr, num_bytes);
1573 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1577 btrfs_free_path(path);
1578 finish_current_insert(trans, extent_root);
1583 * find all the blocks marked as pending in the radix tree and remove
1584 * them from the extent map
1586 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1587 btrfs_root *extent_root)
1595 struct extent_io_tree *pending_del;
1596 struct extent_io_tree *extent_ins;
1597 struct pending_extent_op *extent_op;
1599 extent_ins = &extent_root->fs_info->extent_ins;
1600 pending_del = &extent_root->fs_info->pending_del;
1603 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1608 ret = get_state_private(pending_del, start, &priv);
1610 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1612 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1615 ret = pin_down_bytes(trans, extent_root, start,
1616 end + 1 - start, 0);
1617 mark_free = ret > 0;
1618 if (!test_range_bit(extent_ins, start, end,
1619 EXTENT_LOCKED, 0)) {
1621 ret = __free_extent(trans, extent_root,
1622 start, end + 1 - start,
1623 extent_op->orig_parent,
1624 extent_root->root_key.objectid,
1625 extent_op->orig_generation,
1626 extent_op->level, 0, mark_free);
1630 ret = get_state_private(extent_ins, start, &priv);
1632 extent_op = (struct pending_extent_op *)
1633 (unsigned long)priv;
1635 clear_extent_bits(extent_ins, start, end,
1636 EXTENT_LOCKED, GFP_NOFS);
1638 if (extent_op->type == PENDING_BACKREF_UPDATE)
1641 ret = update_block_group(trans, extent_root, start,
1642 end + 1 - start, 0, mark_free);
1653 * remove an extent from the root, returns 0 on success
1655 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1656 *root, u64 bytenr, u64 num_bytes, u64 parent,
1657 u64 root_objectid, u64 ref_generation,
1658 u64 owner_objectid, int pin)
1660 struct btrfs_root *extent_root = root->fs_info->extent_root;
1664 WARN_ON(num_bytes < root->sectorsize);
1665 if (root == extent_root) {
1666 struct pending_extent_op *extent_op;
1668 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1671 extent_op->type = PENDING_EXTENT_DELETE;
1672 extent_op->bytenr = bytenr;
1673 extent_op->num_bytes = num_bytes;
1674 extent_op->parent = parent;
1675 extent_op->orig_parent = parent;
1676 extent_op->generation = ref_generation;
1677 extent_op->orig_generation = ref_generation;
1678 extent_op->level = (int)owner_objectid;
1680 set_extent_bits(&root->fs_info->pending_del,
1681 bytenr, bytenr + num_bytes - 1,
1682 EXTENT_LOCKED, GFP_NOFS);
1683 set_state_private(&root->fs_info->pending_del,
1684 bytenr, (unsigned long)extent_op);
1687 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
1688 root_objectid, ref_generation,
1689 owner_objectid, pin, pin == 0);
1690 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1691 return ret ? ret : pending_ret;
1694 static u64 stripe_align(struct btrfs_root *root, u64 val)
1696 u64 mask = ((u64)root->stripesize - 1);
1697 u64 ret = (val + mask) & ~mask;
1702 * walks the btree of allocated extents and find a hole of a given size.
1703 * The key ins is changed to record the hole:
1704 * ins->objectid == block start
1705 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1706 * ins->offset == number of blocks
1707 * Any available blocks before search_start are skipped.
1709 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1710 struct btrfs_root *orig_root,
1711 u64 num_bytes, u64 empty_size,
1712 u64 search_start, u64 search_end,
1713 u64 hint_byte, struct btrfs_key *ins,
1714 u64 exclude_start, u64 exclude_nr,
1718 u64 orig_search_start = search_start;
1719 struct btrfs_root * root = orig_root->fs_info->extent_root;
1720 struct btrfs_fs_info *info = root->fs_info;
1721 u64 total_needed = num_bytes;
1722 struct btrfs_block_group_cache *block_group;
1726 WARN_ON(num_bytes < root->sectorsize);
1727 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1730 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1732 hint_byte = search_start;
1733 block_group = btrfs_find_block_group(root, block_group,
1734 hint_byte, data, 1);
1736 block_group = btrfs_find_block_group(root,
1738 search_start, data, 1);
1741 total_needed += empty_size;
1745 block_group = btrfs_lookup_first_block_group(info,
1748 block_group = btrfs_lookup_first_block_group(info,
1751 ret = find_search_start(root, &block_group, &search_start,
1752 total_needed, data);
1756 search_start = stripe_align(root, search_start);
1757 ins->objectid = search_start;
1758 ins->offset = num_bytes;
1760 if (ins->objectid + num_bytes >
1761 block_group->key.objectid + block_group->key.offset) {
1762 search_start = block_group->key.objectid +
1763 block_group->key.offset;
1767 if (test_range_bit(&info->extent_ins, ins->objectid,
1768 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1769 search_start = ins->objectid + num_bytes;
1773 if (test_range_bit(&info->pinned_extents, ins->objectid,
1774 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1775 search_start = ins->objectid + num_bytes;
1779 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1780 ins->objectid < exclude_start + exclude_nr)) {
1781 search_start = exclude_start + exclude_nr;
1785 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1786 block_group = btrfs_lookup_block_group(info, ins->objectid);
1788 trans->block_group = block_group;
1790 ins->offset = num_bytes;
1794 block_group = btrfs_lookup_first_block_group(info, search_start);
1796 search_start = orig_search_start;
1803 total_needed -= empty_size;
1809 block_group = btrfs_find_block_group(root, block_group,
1810 search_start, data, 0);
1817 * finds a free extent and does all the dirty work required for allocation
1818 * returns the key for the extent through ins, and a tree buffer for
1819 * the first block of the extent through buf.
1821 * returns 0 if everything worked, non-zero otherwise.
1823 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root,
1825 u64 num_bytes, u64 parent,
1826 u64 root_objectid, u64 ref_generation,
1827 u64 owner, u64 empty_size, u64 hint_byte,
1828 u64 search_end, struct btrfs_key *ins, int data)
1832 u64 super_used, root_used;
1833 u64 search_start = 0;
1836 struct btrfs_fs_info *info = root->fs_info;
1837 struct btrfs_root *extent_root = info->extent_root;
1838 struct btrfs_path *path;
1839 struct btrfs_extent_item *extent_item;
1840 struct btrfs_extent_ref *ref;
1841 struct btrfs_key keys[2];
1843 if (info->extent_ops) {
1844 struct btrfs_extent_ops *ops = info->extent_ops;
1845 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
1851 alloc_profile = info->avail_data_alloc_bits &
1852 info->data_alloc_profile;
1853 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1854 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
1855 info->system_allocs >= 0) {
1856 alloc_profile = info->avail_system_alloc_bits &
1857 info->system_alloc_profile;
1858 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1860 alloc_profile = info->avail_metadata_alloc_bits &
1861 info->metadata_alloc_profile;
1862 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1865 if (root->ref_cows) {
1866 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1867 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1869 BTRFS_BLOCK_GROUP_METADATA);
1872 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1873 num_bytes + 2 * 1024 * 1024, data);
1877 WARN_ON(num_bytes < root->sectorsize);
1878 ret = find_free_extent(trans, root, num_bytes, empty_size,
1879 search_start, search_end, hint_byte, ins,
1880 trans->alloc_exclude_start,
1881 trans->alloc_exclude_nr, data);
1888 parent = ins->objectid;
1890 /* block accounting for super block */
1891 super_used = btrfs_super_bytes_used(&info->super_copy);
1892 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1894 /* block accounting for root item */
1895 root_used = btrfs_root_used(&root->root_item);
1896 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1898 clear_extent_dirty(&root->fs_info->free_space_cache,
1899 ins->objectid, ins->objectid + ins->offset - 1,
1902 if (root == extent_root) {
1903 struct pending_extent_op *extent_op;
1905 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1908 extent_op->type = PENDING_EXTENT_INSERT;
1909 extent_op->bytenr = ins->objectid;
1910 extent_op->num_bytes = ins->offset;
1911 extent_op->parent = parent;
1912 extent_op->orig_parent = 0;
1913 extent_op->generation = ref_generation;
1914 extent_op->orig_generation = 0;
1915 extent_op->level = (int)owner;
1917 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1918 ins->objectid + ins->offset - 1,
1919 EXTENT_LOCKED, GFP_NOFS);
1920 set_state_private(&root->fs_info->extent_ins,
1921 ins->objectid, (unsigned long)extent_op);
1925 WARN_ON(trans->alloc_exclude_nr);
1926 trans->alloc_exclude_start = ins->objectid;
1927 trans->alloc_exclude_nr = ins->offset;
1929 memcpy(&keys[0], ins, sizeof(*ins));
1930 keys[1].objectid = ins->objectid;
1931 keys[1].type = BTRFS_EXTENT_REF_KEY;
1932 keys[1].offset = parent;
1933 sizes[0] = sizeof(*extent_item);
1934 sizes[1] = sizeof(*ref);
1936 path = btrfs_alloc_path();
1939 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1943 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1944 struct btrfs_extent_item);
1945 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1946 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1947 struct btrfs_extent_ref);
1949 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1950 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1951 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1952 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
1954 btrfs_mark_buffer_dirty(path->nodes[0]);
1956 trans->alloc_exclude_start = 0;
1957 trans->alloc_exclude_nr = 0;
1958 btrfs_free_path(path);
1959 finish_current_insert(trans, extent_root);
1960 pending_ret = del_pending_extents(trans, extent_root);
1970 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1972 printk("update block group failed for %llu %llu\n",
1973 (unsigned long long)ins->objectid,
1974 (unsigned long long)ins->offset);
1981 * helper function to allocate a block for a given tree
1982 * returns the tree buffer or NULL.
1984 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1985 struct btrfs_root *root,
1986 u32 blocksize, u64 parent,
1993 struct btrfs_key ins;
1995 struct extent_buffer *buf;
1997 ret = btrfs_alloc_extent(trans, root, blocksize, parent,
1998 root_objectid, ref_generation,
1999 level, empty_size, hint,
2003 return ERR_PTR(ret);
2005 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2008 parent = ins.objectid;
2009 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2010 parent, root->root_key.objectid,
2011 ref_generation, level, 0);
2013 return ERR_PTR(-ENOMEM);
2015 btrfs_set_buffer_uptodate(buf);
2016 trans->blocks_used++;
2020 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2021 struct btrfs_root *root,
2022 struct extent_buffer *leaf)
2025 u64 leaf_generation;
2026 struct btrfs_key key;
2027 struct btrfs_file_extent_item *fi;
2032 BUG_ON(!btrfs_is_leaf(leaf));
2033 nritems = btrfs_header_nritems(leaf);
2034 leaf_owner = btrfs_header_owner(leaf);
2035 leaf_generation = btrfs_header_generation(leaf);
2037 for (i = 0; i < nritems; i++) {
2040 btrfs_item_key_to_cpu(leaf, &key, i);
2041 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2043 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2044 if (btrfs_file_extent_type(leaf, fi) ==
2045 BTRFS_FILE_EXTENT_INLINE)
2048 * FIXME make sure to insert a trans record that
2049 * repeats the snapshot del on crash
2051 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2052 if (disk_bytenr == 0)
2054 ret = btrfs_free_extent(trans, root, disk_bytenr,
2055 btrfs_file_extent_disk_num_bytes(leaf, fi),
2056 leaf->start, leaf_owner, leaf_generation,
2063 static void noinline reada_walk_down(struct btrfs_root *root,
2064 struct extent_buffer *node,
2077 nritems = btrfs_header_nritems(node);
2078 level = btrfs_header_level(node);
2082 for (i = slot; i < nritems && skipped < 32; i++) {
2083 bytenr = btrfs_node_blockptr(node, i);
2084 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2085 (last > bytenr && last - bytenr > 32 * 1024))) {
2089 blocksize = btrfs_level_size(root, level - 1);
2091 ret = lookup_extent_ref(NULL, root, bytenr,
2099 mutex_unlock(&root->fs_info->fs_mutex);
2100 ret = readahead_tree_block(root, bytenr, blocksize,
2101 btrfs_node_ptr_generation(node, i));
2102 last = bytenr + blocksize;
2104 mutex_lock(&root->fs_info->fs_mutex);
2111 * helper function for drop_snapshot, this walks down the tree dropping ref
2112 * counts as it goes.
2114 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2115 struct btrfs_root *root,
2116 struct btrfs_path *path, int *level)
2122 struct extent_buffer *next;
2123 struct extent_buffer *cur;
2124 struct extent_buffer *parent;
2129 WARN_ON(*level < 0);
2130 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2131 ret = lookup_extent_ref(trans, root,
2132 path->nodes[*level]->start,
2133 path->nodes[*level]->len, &refs);
2139 * walk down to the last node level and free all the leaves
2141 while(*level >= 0) {
2142 WARN_ON(*level < 0);
2143 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2144 cur = path->nodes[*level];
2146 if (btrfs_header_level(cur) != *level)
2149 if (path->slots[*level] >=
2150 btrfs_header_nritems(cur))
2153 ret = drop_leaf_ref(trans, root, cur);
2157 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2158 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2159 blocksize = btrfs_level_size(root, *level - 1);
2160 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2163 parent = path->nodes[*level];
2164 root_owner = btrfs_header_owner(parent);
2165 root_gen = btrfs_header_generation(parent);
2166 path->slots[*level]++;
2167 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2168 parent->start, root_owner,
2169 root_gen, *level - 1, 1);
2173 next = btrfs_find_tree_block(root, bytenr, blocksize);
2174 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2175 free_extent_buffer(next);
2176 reada_walk_down(root, cur, path->slots[*level]);
2177 mutex_unlock(&root->fs_info->fs_mutex);
2178 next = read_tree_block(root, bytenr, blocksize,
2180 mutex_lock(&root->fs_info->fs_mutex);
2182 WARN_ON(*level <= 0);
2183 if (path->nodes[*level-1])
2184 free_extent_buffer(path->nodes[*level-1]);
2185 path->nodes[*level-1] = next;
2186 *level = btrfs_header_level(next);
2187 path->slots[*level] = 0;
2190 WARN_ON(*level < 0);
2191 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2193 if (path->nodes[*level] == root->node) {
2194 root_owner = root->root_key.objectid;
2195 parent = path->nodes[*level];
2197 parent = path->nodes[*level + 1];
2198 root_owner = btrfs_header_owner(parent);
2201 root_gen = btrfs_header_generation(parent);
2202 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2203 path->nodes[*level]->len, parent->start,
2204 root_owner, root_gen, *level, 1);
2205 free_extent_buffer(path->nodes[*level]);
2206 path->nodes[*level] = NULL;
2213 * helper for dropping snapshots. This walks back up the tree in the path
2214 * to find the first node higher up where we haven't yet gone through
2217 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2218 struct btrfs_root *root,
2219 struct btrfs_path *path, int *level)
2223 struct btrfs_root_item *root_item = &root->root_item;
2228 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2229 slot = path->slots[i];
2230 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2231 struct extent_buffer *node;
2232 struct btrfs_disk_key disk_key;
2233 node = path->nodes[i];
2236 WARN_ON(*level == 0);
2237 btrfs_node_key(node, &disk_key, path->slots[i]);
2238 memcpy(&root_item->drop_progress,
2239 &disk_key, sizeof(disk_key));
2240 root_item->drop_level = i;
2243 struct extent_buffer *parent;
2244 if (path->nodes[*level] == root->node)
2245 parent = path->nodes[*level];
2247 parent = path->nodes[*level + 1];
2249 root_owner = btrfs_header_owner(parent);
2250 root_gen = btrfs_header_generation(parent);
2251 ret = btrfs_free_extent(trans, root,
2252 path->nodes[*level]->start,
2253 path->nodes[*level]->len,
2254 parent->start, root_owner,
2255 root_gen, *level, 1);
2257 free_extent_buffer(path->nodes[*level]);
2258 path->nodes[*level] = NULL;
2266 * drop the reference count on the tree rooted at 'snap'. This traverses
2267 * the tree freeing any blocks that have a ref count of zero after being
2270 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2276 struct btrfs_path *path;
2279 struct btrfs_root_item *root_item = &root->root_item;
2281 path = btrfs_alloc_path();
2284 level = btrfs_header_level(root->node);
2286 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2287 path->nodes[level] = root->node;
2288 extent_buffer_get(root->node);
2289 path->slots[level] = 0;
2291 struct btrfs_key key;
2292 struct btrfs_disk_key found_key;
2293 struct extent_buffer *node;
2295 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2296 level = root_item->drop_level;
2297 path->lowest_level = level;
2298 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2303 node = path->nodes[level];
2304 btrfs_node_key(node, &found_key, path->slots[level]);
2305 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2306 sizeof(found_key)));
2309 wret = walk_down_tree(trans, root, path, &level);
2315 wret = walk_up_tree(trans, root, path, &level);
2325 for (i = 0; i <= orig_level; i++) {
2326 if (path->nodes[i]) {
2327 free_extent_buffer(path->nodes[i]);
2328 path->nodes[i] = NULL;
2332 btrfs_free_path(path);
2336 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2343 ret = find_first_extent_bit(&info->block_group_cache, 0,
2344 &start, &end, (unsigned int)-1);
2347 ret = get_state_private(&info->block_group_cache, start, &ptr);
2349 kfree((void *)(unsigned long)ptr);
2350 clear_extent_bits(&info->block_group_cache, start,
2351 end, (unsigned int)-1, GFP_NOFS);
2354 ret = find_first_extent_bit(&info->free_space_cache, 0,
2355 &start, &end, EXTENT_DIRTY);
2358 clear_extent_dirty(&info->free_space_cache, start,
2364 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2365 struct btrfs_key *key)
2368 struct btrfs_key found_key;
2369 struct extent_buffer *leaf;
2372 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2376 slot = path->slots[0];
2377 leaf = path->nodes[0];
2378 if (slot >= btrfs_header_nritems(leaf)) {
2379 ret = btrfs_next_leaf(root, path);
2386 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2388 if (found_key.objectid >= key->objectid &&
2389 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2398 int btrfs_read_block_groups(struct btrfs_root *root)
2400 struct btrfs_path *path;
2403 struct btrfs_block_group_cache *cache;
2404 struct btrfs_fs_info *info = root->fs_info;
2405 struct btrfs_space_info *space_info;
2406 struct extent_io_tree *block_group_cache;
2407 struct btrfs_key key;
2408 struct btrfs_key found_key;
2409 struct extent_buffer *leaf;
2411 block_group_cache = &info->block_group_cache;
2413 root = info->extent_root;
2416 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2417 path = btrfs_alloc_path();
2422 ret = find_first_block_group(root, path, &key);
2430 leaf = path->nodes[0];
2431 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2432 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2438 read_extent_buffer(leaf, &cache->item,
2439 btrfs_item_ptr_offset(leaf, path->slots[0]),
2440 sizeof(cache->item));
2441 memcpy(&cache->key, &found_key, sizeof(found_key));
2444 key.objectid = found_key.objectid + found_key.offset;
2445 btrfs_release_path(root, path);
2446 cache->flags = btrfs_block_group_flags(&cache->item);
2448 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2449 bit = BLOCK_GROUP_DATA;
2450 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2451 bit = BLOCK_GROUP_SYSTEM;
2452 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2453 bit = BLOCK_GROUP_METADATA;
2455 set_avail_alloc_bits(info, cache->flags);
2456 if (btrfs_chunk_readonly(root, cache->key.objectid))
2459 ret = update_space_info(info, cache->flags, found_key.offset,
2460 btrfs_block_group_used(&cache->item),
2463 cache->space_info = space_info;
2465 /* use EXTENT_LOCKED to prevent merging */
2466 set_extent_bits(block_group_cache, found_key.objectid,
2467 found_key.objectid + found_key.offset - 1,
2468 bit | EXTENT_LOCKED, GFP_NOFS);
2469 set_state_private(block_group_cache, found_key.objectid,
2470 (unsigned long)cache);
2474 btrfs_free_path(path);
2478 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2479 struct btrfs_root *root, u64 bytes_used,
2480 u64 type, u64 chunk_objectid, u64 chunk_offset,
2485 struct btrfs_root *extent_root;
2486 struct btrfs_block_group_cache *cache;
2487 struct extent_io_tree *block_group_cache;
2489 extent_root = root->fs_info->extent_root;
2490 block_group_cache = &root->fs_info->block_group_cache;
2492 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2494 cache->key.objectid = chunk_offset;
2495 cache->key.offset = size;
2497 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2498 btrfs_set_block_group_used(&cache->item, bytes_used);
2499 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2500 cache->flags = type;
2501 btrfs_set_block_group_flags(&cache->item, type);
2503 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2504 &cache->space_info);
2507 bit = block_group_state_bits(type);
2508 set_extent_bits(block_group_cache, chunk_offset,
2509 chunk_offset + size - 1,
2510 bit | EXTENT_LOCKED, GFP_NOFS);
2512 set_state_private(block_group_cache, chunk_offset,
2513 (unsigned long)cache);
2514 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2515 sizeof(cache->item));
2518 finish_current_insert(trans, extent_root);
2519 ret = del_pending_extents(trans, extent_root);
2521 set_avail_alloc_bits(extent_root->fs_info, type);
2526 * This is for converter use only.
2528 * In that case, we don't know where are free blocks located.
2529 * Therefore all block group cache entries must be setup properly
2530 * before doing any block allocation.
2532 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root)
2541 u64 total_metadata = 0;
2545 struct btrfs_root *extent_root;
2546 struct btrfs_block_group_cache *cache;
2547 struct extent_io_tree *block_group_cache;
2549 extent_root = root->fs_info->extent_root;
2550 block_group_cache = &root->fs_info->block_group_cache;
2551 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2552 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
2553 group_align = 64 * root->sectorsize;
2556 while (cur_start < total_bytes) {
2557 group_size = total_bytes / 12;
2558 group_size = min_t(u64, group_size, total_bytes - cur_start);
2559 if (cur_start == 0) {
2560 bit = BLOCK_GROUP_SYSTEM;
2561 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
2563 group_size &= ~(group_align - 1);
2564 group_size = max_t(u64, group_size, 32 * 1024 * 1024);
2565 group_size = min_t(u64, group_size, 128 * 1024 * 1024);
2567 group_size &= ~(group_align - 1);
2568 if (total_data >= total_metadata * 2) {
2569 group_type = BTRFS_BLOCK_GROUP_METADATA;
2570 group_size = min_t(u64, group_size,
2571 1ULL * 1024 * 1024 * 1024);
2572 total_metadata += group_size;
2574 group_type = BTRFS_BLOCK_GROUP_DATA;
2575 group_size = min_t(u64, group_size,
2576 5ULL * 1024 * 1024 * 1024);
2577 total_data += group_size;
2579 if ((total_bytes - cur_start) * 4 < group_size * 5)
2580 group_size = total_bytes - cur_start;
2583 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2586 cache->key.objectid = cur_start;
2587 cache->key.offset = group_size;
2588 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2590 btrfs_set_block_group_used(&cache->item, 0);
2591 btrfs_set_block_group_chunk_objectid(&cache->item,
2593 btrfs_set_block_group_flags(&cache->item, group_type);
2595 cache->flags = group_type;
2597 ret = update_space_info(root->fs_info, group_type, group_size,
2598 0, &cache->space_info);
2600 set_avail_alloc_bits(extent_root->fs_info, group_type);
2602 set_extent_bits(block_group_cache, cur_start,
2603 cur_start + group_size - 1,
2604 bit | EXTENT_LOCKED, GFP_NOFS);
2605 set_state_private(block_group_cache, cur_start,
2606 (unsigned long)cache);
2607 cur_start += group_size;
2609 /* then insert all the items */
2611 while(cur_start < total_bytes) {
2612 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
2615 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2616 sizeof(cache->item));
2619 finish_current_insert(trans, extent_root);
2620 ret = del_pending_extents(trans, extent_root);
2623 cur_start = cache->key.objectid + cache->key.offset;
2628 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root,
2630 u64 bytenr, u64 num_bytes, int alloc,
2633 return update_block_group(trans, root, bytenr, num_bytes,