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 {
45 struct btrfs_disk_key key;
49 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
50 struct btrfs_root *root,
51 u64 root_objectid, u64 generation,
52 u64 flags, struct btrfs_disk_key *key,
53 int level, struct btrfs_key *ins);
54 static int __free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_root *root,
56 u64 bytenr, u64 num_bytes, u64 parent,
57 u64 root_objectid, u64 owner_objectid,
58 u64 owner_offset, int refs_to_drop);
59 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
60 btrfs_root *extent_root);
61 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
62 btrfs_root *extent_root);
64 static int remove_sb_from_cache(struct btrfs_root *root,
65 struct btrfs_block_group_cache *cache)
71 struct extent_io_tree *free_space_cache;
73 free_space_cache = &root->fs_info->free_space_cache;
74 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
75 bytenr = btrfs_sb_offset(i);
76 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
77 cache->key.objectid, bytenr, 0,
78 &logical, &nr, &stripe_len);
81 clear_extent_dirty(free_space_cache, logical[nr],
82 logical[nr] + stripe_len - 1, GFP_NOFS);
89 static int cache_block_group(struct btrfs_root *root,
90 struct btrfs_block_group_cache *block_group)
92 struct btrfs_path *path;
95 struct extent_buffer *leaf;
96 struct extent_io_tree *free_space_cache;
104 root = root->fs_info->extent_root;
105 free_space_cache = &root->fs_info->free_space_cache;
107 if (block_group->cached)
110 path = btrfs_alloc_path();
115 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
118 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
119 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
124 leaf = path->nodes[0];
125 slot = path->slots[0];
126 if (slot >= btrfs_header_nritems(leaf)) {
127 ret = btrfs_next_leaf(root, path);
136 btrfs_item_key_to_cpu(leaf, &key, slot);
137 if (key.objectid < block_group->key.objectid) {
140 if (key.objectid >= block_group->key.objectid +
141 block_group->key.offset) {
145 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
146 if (key.objectid > last) {
147 hole_size = key.objectid - last;
148 set_extent_dirty(free_space_cache, last,
149 last + hole_size - 1,
152 last = key.objectid + key.offset;
158 if (block_group->key.objectid +
159 block_group->key.offset > last) {
160 hole_size = block_group->key.objectid +
161 block_group->key.offset - last;
162 set_extent_dirty(free_space_cache, last,
163 last + hole_size - 1, GFP_NOFS);
165 remove_sb_from_cache(root, block_group);
166 block_group->cached = 1;
168 btrfs_free_path(path);
172 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
176 struct extent_io_tree *block_group_cache;
177 struct btrfs_block_group_cache *block_group = NULL;
183 bytenr = max_t(u64, bytenr,
184 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
185 block_group_cache = &info->block_group_cache;
186 ret = find_first_extent_bit(block_group_cache,
187 bytenr, &start, &end,
188 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
193 ret = get_state_private(block_group_cache, start, &ptr);
197 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
201 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
205 struct extent_io_tree *block_group_cache;
206 struct btrfs_block_group_cache *block_group = NULL;
212 block_group_cache = &info->block_group_cache;
213 ret = find_first_extent_bit(block_group_cache,
214 bytenr, &start, &end,
215 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
220 ret = get_state_private(block_group_cache, start, &ptr);
224 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
225 if (block_group->key.objectid <= bytenr && bytenr <
226 block_group->key.objectid + block_group->key.offset)
231 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
233 return (cache->flags & bits) == bits;
236 static int noinline find_search_start(struct btrfs_root *root,
237 struct btrfs_block_group_cache **cache_ret,
238 u64 *start_ret, int num, int data)
241 struct btrfs_block_group_cache *cache = *cache_ret;
245 u64 search_start = *start_ret;
252 ret = cache_block_group(root, cache);
256 last = max(search_start, cache->key.objectid);
257 if (cache->ro || !block_group_bits(cache, data)) {
262 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
263 last, &start, &end, EXTENT_DIRTY);
268 start = max(last, start);
270 if (last - start < num) {
273 if (start + num > cache->key.objectid + cache->key.offset) {
280 cache = btrfs_lookup_block_group(root->fs_info, search_start);
282 printk("Unable to find block group for %llu\n",
283 (unsigned long long)search_start);
289 last = cache->key.objectid + cache->key.offset;
291 cache = btrfs_lookup_first_block_group(root->fs_info, last);
301 cache = btrfs_find_block_group(root, cache, last, data, 0);
302 cache = btrfs_find_block_group(root, cache, last, data, 0);
310 static u64 div_factor(u64 num, int factor)
319 static int block_group_state_bits(u64 flags)
322 if (flags & BTRFS_BLOCK_GROUP_DATA)
323 bits |= BLOCK_GROUP_DATA;
324 if (flags & BTRFS_BLOCK_GROUP_METADATA)
325 bits |= BLOCK_GROUP_METADATA;
326 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
327 bits |= BLOCK_GROUP_SYSTEM;
331 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
332 struct btrfs_block_group_cache
333 *hint, u64 search_start,
336 struct btrfs_block_group_cache *cache;
337 struct extent_io_tree *block_group_cache;
338 struct btrfs_block_group_cache *found_group = NULL;
339 struct btrfs_fs_info *info = root->fs_info;
352 block_group_cache = &info->block_group_cache;
357 bit = block_group_state_bits(data);
360 struct btrfs_block_group_cache *shint;
361 shint = btrfs_lookup_block_group(info, search_start);
362 if (shint && !shint->ro && block_group_bits(shint, data)) {
363 used = btrfs_block_group_used(&shint->item);
364 if (used + shint->pinned <
365 div_factor(shint->key.offset, factor)) {
370 if (hint && !hint->ro && block_group_bits(hint, data)) {
371 used = btrfs_block_group_used(&hint->item);
372 if (used + hint->pinned <
373 div_factor(hint->key.offset, factor)) {
376 last = hint->key.objectid + hint->key.offset;
380 hint_last = max(hint->key.objectid, search_start);
382 hint_last = search_start;
388 ret = find_first_extent_bit(block_group_cache, last,
393 ret = get_state_private(block_group_cache, start, &ptr);
397 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
398 last = cache->key.objectid + cache->key.offset;
399 used = btrfs_block_group_used(&cache->item);
401 if (!cache->ro && block_group_bits(cache, data)) {
403 free_check = cache->key.offset;
405 free_check = div_factor(cache->key.offset,
408 if (used + cache->pinned < free_check) {
425 * Back reference rules. Back refs have three main goals:
427 * 1) differentiate between all holders of references to an extent so that
428 * when a reference is dropped we can make sure it was a valid reference
429 * before freeing the extent.
431 * 2) Provide enough information to quickly find the holders of an extent
432 * if we notice a given block is corrupted or bad.
434 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
435 * maintenance. This is actually the same as #2, but with a slightly
436 * different use case.
438 * There are two kinds of back refs. The implicit back refs is optimized
439 * for pointers in non-shared tree blocks. For a given pointer in a block,
440 * back refs of this kind provide information about the block's owner tree
441 * and the pointer's key. These information allow us to find the block by
442 * b-tree searching. The full back refs is for pointers in tree blocks not
443 * referenced by their owner trees. The location of tree block is recorded
444 * in the back refs. Actually the full back refs is generic, and can be
445 * used in all cases the implicit back refs is used. The major shortcoming
446 * of the full back refs is its overhead. Every time a tree block gets
447 * COWed, we have to update back refs entry for all pointers in it.
449 * For a newly allocated tree block, we use implicit back refs for
450 * pointers in it. This means most tree related operations only involve
451 * implicit back refs. For a tree block created in old transaction, the
452 * only way to drop a reference to it is COW it. So we can detect the
453 * event that tree block loses its owner tree's reference and do the
454 * back refs conversion.
456 * When a tree block is COW'd through a tree, there are four cases:
458 * The reference count of the block is one and the tree is the block's
459 * owner tree. Nothing to do in this case.
461 * The reference count of the block is one and the tree is not the
462 * block's owner tree. In this case, full back refs is used for pointers
463 * in the block. Remove these full back refs, add implicit back refs for
464 * every pointers in the new block.
466 * The reference count of the block is greater than one and the tree is
467 * the block's owner tree. In this case, implicit back refs is used for
468 * pointers in the block. Add full back refs for every pointers in the
469 * block, increase lower level extents' reference counts. The original
470 * implicit back refs are entailed to the new block.
472 * The reference count of the block is greater than one and the tree is
473 * not the block's owner tree. Add implicit back refs for every pointer in
474 * the new block, increase lower level extents' reference count.
476 * Back Reference Key composing:
478 * The key objectid corresponds to the first byte in the extent,
479 * The key type is used to differentiate between types of back refs.
480 * There are different meanings of the key offset for different types
483 * File extents can be referenced by:
485 * - multiple snapshots, subvolumes, or different generations in one subvol
486 * - different files inside a single subvolume
487 * - different offsets inside a file (bookend extents in file.c)
489 * The extent ref structure for the implicit back refs has fields for:
491 * - Objectid of the subvolume root
492 * - objectid of the file holding the reference
493 * - original offset in the file
494 * - how many bookend extents
496 * The key offset for the implicit back refs is hash of the first
499 * The extent ref structure for the full back refs has field for:
501 * - number of pointers in the tree leaf
503 * The key offset for the implicit back refs is the first byte of
506 * When a file extent is allocated, The implicit back refs is used.
507 * the fields are filled in:
509 * (root_key.objectid, inode objectid, offset in file, 1)
511 * When a file extent is removed file truncation, we find the
512 * corresponding implicit back refs and check the following fields:
514 * (btrfs_header_owner(leaf), inode objectid, offset in file)
516 * Btree extents can be referenced by:
518 * - Different subvolumes
520 * Both the implicit back refs and the full back refs for tree blocks
521 * only consist of key. The key offset for the implicit back refs is
522 * objectid of block's owner tree. The key offset for the full back refs
523 * is the first byte of parent block.
525 * When implicit back refs is used, information about the lowest key and
526 * level of the tree block are required. These information are stored in
527 * tree block info structure.
530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
531 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
532 struct btrfs_root *root,
533 struct btrfs_path *path,
534 u64 owner, u32 extra_size)
536 struct btrfs_extent_item *item;
537 struct btrfs_extent_item_v0 *ei0;
538 struct btrfs_extent_ref_v0 *ref0;
539 struct btrfs_tree_block_info *bi;
540 struct extent_buffer *leaf;
541 struct btrfs_key key;
542 struct btrfs_key found_key;
543 u32 new_size = sizeof(*item);
547 leaf = path->nodes[0];
548 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
550 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
551 ei0 = btrfs_item_ptr(leaf, path->slots[0],
552 struct btrfs_extent_item_v0);
553 refs = btrfs_extent_refs_v0(leaf, ei0);
555 if (owner == (u64)-1) {
557 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
558 ret = btrfs_next_leaf(root, path);
562 leaf = path->nodes[0];
564 btrfs_item_key_to_cpu(leaf, &found_key,
566 BUG_ON(key.objectid != found_key.objectid);
567 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
571 ref0 = btrfs_item_ptr(leaf, path->slots[0],
572 struct btrfs_extent_ref_v0);
573 owner = btrfs_ref_objectid_v0(leaf, ref0);
577 btrfs_release_path(root, path);
579 if (owner < BTRFS_FIRST_FREE_OBJECTID)
580 new_size += sizeof(*bi);
582 new_size -= sizeof(*ei0);
583 ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
588 ret = btrfs_extend_item(trans, root, path, new_size);
591 leaf = path->nodes[0];
592 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
593 btrfs_set_extent_refs(leaf, item, refs);
594 /* FIXME: get real generation */
595 btrfs_set_extent_generation(leaf, item, 0);
596 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
597 btrfs_set_extent_flags(leaf, item,
598 BTRFS_EXTENT_FLAG_TREE_BLOCK |
599 BTRFS_BLOCK_FLAG_FULL_BACKREF);
600 bi = (struct btrfs_tree_block_info *)(item + 1);
601 /* FIXME: get first key of the block */
602 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
603 btrfs_set_tree_block_level(leaf, bi, (int)owner);
605 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
607 btrfs_mark_buffer_dirty(leaf);
612 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
614 u32 high_crc = ~(u32)0;
615 u32 low_crc = ~(u32)0;
618 lenum = cpu_to_le64(root_objectid);
619 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
620 lenum = cpu_to_le64(owner);
621 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
622 lenum = cpu_to_le64(offset);
623 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
625 return ((u64)high_crc << 31) ^ (u64)low_crc;
628 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
629 struct btrfs_extent_data_ref *ref)
631 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
632 btrfs_extent_data_ref_objectid(leaf, ref),
633 btrfs_extent_data_ref_offset(leaf, ref));
636 static int match_extent_data_ref(struct extent_buffer *leaf,
637 struct btrfs_extent_data_ref *ref,
638 u64 root_objectid, u64 owner, u64 offset)
640 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
641 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
642 btrfs_extent_data_ref_offset(leaf, ref) != offset)
647 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
648 struct btrfs_root *root,
649 struct btrfs_path *path,
650 u64 bytenr, u64 parent,
652 u64 owner, u64 offset)
654 struct btrfs_key key;
655 struct btrfs_extent_data_ref *ref;
656 struct extent_buffer *leaf;
662 key.objectid = bytenr;
664 key.type = BTRFS_SHARED_DATA_REF_KEY;
667 key.type = BTRFS_EXTENT_DATA_REF_KEY;
668 key.offset = hash_extent_data_ref(root_objectid,
673 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
682 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
683 key.type = BTRFS_EXTENT_REF_V0_KEY;
684 btrfs_release_path(root, path);
685 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
696 leaf = path->nodes[0];
697 nritems = btrfs_header_nritems(leaf);
699 if (path->slots[0] >= nritems) {
700 ret = btrfs_next_leaf(root, path);
706 leaf = path->nodes[0];
707 nritems = btrfs_header_nritems(leaf);
711 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
712 if (key.objectid != bytenr ||
713 key.type != BTRFS_EXTENT_DATA_REF_KEY)
716 ref = btrfs_item_ptr(leaf, path->slots[0],
717 struct btrfs_extent_data_ref);
719 if (match_extent_data_ref(leaf, ref, root_objectid,
722 btrfs_release_path(root, path);
734 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root,
736 struct btrfs_path *path,
737 u64 bytenr, u64 parent,
738 u64 root_objectid, u64 owner,
739 u64 offset, int refs_to_add)
741 struct btrfs_key key;
742 struct extent_buffer *leaf;
747 key.objectid = bytenr;
749 key.type = BTRFS_SHARED_DATA_REF_KEY;
751 size = sizeof(struct btrfs_shared_data_ref);
753 key.type = BTRFS_EXTENT_DATA_REF_KEY;
754 key.offset = hash_extent_data_ref(root_objectid,
756 size = sizeof(struct btrfs_extent_data_ref);
759 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
760 if (ret && ret != -EEXIST)
763 leaf = path->nodes[0];
765 struct btrfs_shared_data_ref *ref;
766 ref = btrfs_item_ptr(leaf, path->slots[0],
767 struct btrfs_shared_data_ref);
769 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
771 num_refs = btrfs_shared_data_ref_count(leaf, ref);
772 num_refs += refs_to_add;
773 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
776 struct btrfs_extent_data_ref *ref;
777 while (ret == -EEXIST) {
778 ref = btrfs_item_ptr(leaf, path->slots[0],
779 struct btrfs_extent_data_ref);
780 if (match_extent_data_ref(leaf, ref, root_objectid,
783 btrfs_release_path(root, path);
786 ret = btrfs_insert_empty_item(trans, root, path, &key,
788 if (ret && ret != -EEXIST)
791 leaf = path->nodes[0];
793 ref = btrfs_item_ptr(leaf, path->slots[0],
794 struct btrfs_extent_data_ref);
796 btrfs_set_extent_data_ref_root(leaf, ref,
798 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
799 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
800 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
802 num_refs = btrfs_extent_data_ref_count(leaf, ref);
803 num_refs += refs_to_add;
804 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
807 btrfs_mark_buffer_dirty(leaf);
810 btrfs_release_path(root, path);
814 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 struct btrfs_path *path,
819 struct btrfs_key key;
820 struct btrfs_extent_data_ref *ref1 = NULL;
821 struct btrfs_shared_data_ref *ref2 = NULL;
822 struct extent_buffer *leaf;
826 leaf = path->nodes[0];
827 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
829 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
830 ref1 = btrfs_item_ptr(leaf, path->slots[0],
831 struct btrfs_extent_data_ref);
832 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
833 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
834 ref2 = btrfs_item_ptr(leaf, path->slots[0],
835 struct btrfs_shared_data_ref);
836 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
837 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
838 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
839 struct btrfs_extent_ref_v0 *ref0;
840 ref0 = btrfs_item_ptr(leaf, path->slots[0],
841 struct btrfs_extent_ref_v0);
842 num_refs = btrfs_ref_count_v0(leaf, ref0);
848 BUG_ON(num_refs < refs_to_drop);
849 num_refs -= refs_to_drop;
852 ret = btrfs_del_item(trans, root, path);
854 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
855 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
856 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
857 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
858 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
860 struct btrfs_extent_ref_v0 *ref0;
861 ref0 = btrfs_item_ptr(leaf, path->slots[0],
862 struct btrfs_extent_ref_v0);
863 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
866 btrfs_mark_buffer_dirty(leaf);
871 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
872 struct btrfs_path *path,
873 struct btrfs_extent_inline_ref *iref)
875 struct btrfs_key key;
876 struct extent_buffer *leaf;
877 struct btrfs_extent_data_ref *ref1;
878 struct btrfs_shared_data_ref *ref2;
881 leaf = path->nodes[0];
882 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884 if (btrfs_extent_inline_ref_type(leaf, iref) ==
885 BTRFS_EXTENT_DATA_REF_KEY) {
886 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
887 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
889 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
890 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
892 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
893 ref1 = btrfs_item_ptr(leaf, path->slots[0],
894 struct btrfs_extent_data_ref);
895 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
896 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
897 ref2 = btrfs_item_ptr(leaf, path->slots[0],
898 struct btrfs_shared_data_ref);
899 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
900 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
901 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
902 struct btrfs_extent_ref_v0 *ref0;
903 ref0 = btrfs_item_ptr(leaf, path->slots[0],
904 struct btrfs_extent_ref_v0);
905 num_refs = btrfs_ref_count_v0(leaf, ref0);
913 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root,
915 struct btrfs_path *path,
916 u64 bytenr, u64 parent,
919 struct btrfs_key key;
922 key.objectid = bytenr;
924 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
927 key.type = BTRFS_TREE_BLOCK_REF_KEY;
928 key.offset = root_objectid;
931 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
934 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
935 if (ret == -ENOENT && parent) {
936 btrfs_release_path(root, path);
937 key.type = BTRFS_EXTENT_REF_V0_KEY;
938 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
946 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
947 struct btrfs_root *root,
948 struct btrfs_path *path,
949 u64 bytenr, u64 parent,
952 struct btrfs_key key;
955 key.objectid = bytenr;
957 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
960 key.type = BTRFS_TREE_BLOCK_REF_KEY;
961 key.offset = root_objectid;
964 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
966 btrfs_release_path(root, path);
970 static inline int extent_ref_type(u64 parent, u64 owner)
972 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
974 return BTRFS_SHARED_BLOCK_REF_KEY;
976 return BTRFS_TREE_BLOCK_REF_KEY;
979 return BTRFS_SHARED_DATA_REF_KEY;
981 return BTRFS_EXTENT_DATA_REF_KEY;
985 static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
989 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
990 if (!path->nodes[level])
992 if (path->slots[level] + 1 >=
993 btrfs_header_nritems(path->nodes[level]))
996 btrfs_item_key_to_cpu(path->nodes[level], key,
997 path->slots[level] + 1);
999 btrfs_node_key_to_cpu(path->nodes[level], key,
1000 path->slots[level] + 1);
1006 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1007 struct btrfs_root *root,
1008 struct btrfs_path *path,
1009 struct btrfs_extent_inline_ref **ref_ret,
1010 u64 bytenr, u64 num_bytes,
1011 u64 parent, u64 root_objectid,
1012 u64 owner, u64 offset, int insert)
1014 struct btrfs_key key;
1015 struct extent_buffer *leaf;
1016 struct btrfs_extent_item *ei;
1017 struct btrfs_extent_inline_ref *iref;
1028 key.objectid = bytenr;
1029 key.type = BTRFS_EXTENT_ITEM_KEY;
1030 key.offset = num_bytes;
1032 want = extent_ref_type(parent, owner);
1034 extra_size = btrfs_extent_inline_ref_size(want);
1037 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1043 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1044 btrfs_print_leaf(root, path->nodes[0]);
1045 btrfs_free_path(path);
1051 leaf = path->nodes[0];
1052 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1054 if (item_size < sizeof(*ei)) {
1059 ret = convert_extent_item_v0(trans, root, path, owner,
1065 leaf = path->nodes[0];
1066 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1069 if (item_size < sizeof(*ei)) {
1070 printf("Size is %u, needs to be %u, slot %d\n", item_size,
1071 sizeof(*ei), path->slots[0]);
1072 btrfs_print_leaf(root, leaf);
1075 BUG_ON(item_size < sizeof(*ei));
1077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1078 flags = btrfs_extent_flags(leaf, ei);
1080 ptr = (unsigned long)(ei + 1);
1081 end = (unsigned long)ei + item_size;
1083 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1084 ptr += sizeof(struct btrfs_tree_block_info);
1087 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1096 iref = (struct btrfs_extent_inline_ref *)ptr;
1097 type = btrfs_extent_inline_ref_type(leaf, iref);
1101 ptr += btrfs_extent_inline_ref_size(type);
1105 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1106 struct btrfs_extent_data_ref *dref;
1107 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1108 if (match_extent_data_ref(leaf, dref, root_objectid,
1113 if (hash_extent_data_ref_item(leaf, dref) <
1114 hash_extent_data_ref(root_objectid, owner, offset))
1118 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1120 if (parent == ref_offset) {
1124 if (ref_offset < parent)
1127 if (root_objectid == ref_offset) {
1131 if (ref_offset < root_objectid)
1135 ptr += btrfs_extent_inline_ref_size(type);
1137 if (err == -ENOENT && insert) {
1138 if (item_size + extra_size >=
1139 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1144 * To add new inline back ref, we have to make sure
1145 * there is no corresponding back ref item.
1146 * For simplicity, we just do not add new inline back
1147 * ref if there is any back ref item.
1149 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1150 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1155 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1160 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 struct btrfs_extent_inline_ref *iref,
1164 u64 parent, u64 root_objectid,
1165 u64 owner, u64 offset, int refs_to_add)
1167 struct extent_buffer *leaf;
1168 struct btrfs_extent_item *ei;
1171 unsigned long item_offset;
1177 leaf = path->nodes[0];
1178 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1179 item_offset = (unsigned long)iref - (unsigned long)ei;
1181 type = extent_ref_type(parent, owner);
1182 size = btrfs_extent_inline_ref_size(type);
1184 ret = btrfs_extend_item(trans, root, path, size);
1187 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1188 refs = btrfs_extent_refs(leaf, ei);
1189 refs += refs_to_add;
1190 btrfs_set_extent_refs(leaf, ei, refs);
1192 ptr = (unsigned long)ei + item_offset;
1193 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1194 if (ptr < end - size)
1195 memmove_extent_buffer(leaf, ptr + size, ptr,
1198 iref = (struct btrfs_extent_inline_ref *)ptr;
1199 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1200 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1201 struct btrfs_extent_data_ref *dref;
1202 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1203 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1204 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1205 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1206 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1207 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1208 struct btrfs_shared_data_ref *sref;
1209 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1210 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1211 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1212 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1213 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1215 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1217 btrfs_mark_buffer_dirty(leaf);
1221 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1222 struct btrfs_root *root,
1223 struct btrfs_path *path,
1224 struct btrfs_extent_inline_ref **ref_ret,
1225 u64 bytenr, u64 num_bytes, u64 parent,
1226 u64 root_objectid, u64 owner, u64 offset)
1230 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1231 bytenr, num_bytes, parent,
1232 root_objectid, owner, offset, 0);
1236 btrfs_release_path(root, path);
1239 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1240 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1243 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1244 root_objectid, owner, offset);
1249 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1250 struct btrfs_root *root,
1251 struct btrfs_path *path,
1252 struct btrfs_extent_inline_ref *iref,
1255 struct extent_buffer *leaf;
1256 struct btrfs_extent_item *ei;
1257 struct btrfs_extent_data_ref *dref = NULL;
1258 struct btrfs_shared_data_ref *sref = NULL;
1267 leaf = path->nodes[0];
1268 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1269 refs = btrfs_extent_refs(leaf, ei);
1270 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1271 refs += refs_to_mod;
1272 btrfs_set_extent_refs(leaf, ei, refs);
1274 type = btrfs_extent_inline_ref_type(leaf, iref);
1276 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1278 refs = btrfs_extent_data_ref_count(leaf, dref);
1279 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1280 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1281 refs = btrfs_shared_data_ref_count(leaf, sref);
1284 BUG_ON(refs_to_mod != -1);
1287 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1288 refs += refs_to_mod;
1291 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1292 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1294 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1296 size = btrfs_extent_inline_ref_size(type);
1297 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1298 ptr = (unsigned long)iref;
1299 end = (unsigned long)ei + item_size;
1300 if (ptr + size < end)
1301 memmove_extent_buffer(leaf, ptr, ptr + size,
1304 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1307 btrfs_mark_buffer_dirty(leaf);
1311 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1312 struct btrfs_root *root,
1313 struct btrfs_path *path,
1314 u64 bytenr, u64 num_bytes, u64 parent,
1315 u64 root_objectid, u64 owner,
1316 u64 offset, int refs_to_add)
1318 struct btrfs_extent_inline_ref *iref;
1321 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1322 bytenr, num_bytes, parent,
1323 root_objectid, owner, offset, 1);
1325 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1326 ret = update_inline_extent_backref(trans, root, path, iref,
1328 } else if (ret == -ENOENT) {
1329 ret = setup_inline_extent_backref(trans, root, path, iref,
1330 parent, root_objectid,
1331 owner, offset, refs_to_add);
1336 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1337 struct btrfs_root *root,
1338 struct btrfs_path *path,
1339 u64 bytenr, u64 parent, u64 root_objectid,
1340 u64 owner, u64 offset, int refs_to_add)
1344 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1345 ret = insert_extent_data_ref(trans, root, path, bytenr,
1346 parent, root_objectid,
1347 owner, offset, refs_to_add);
1349 BUG_ON(refs_to_add != 1);
1350 ret = insert_tree_block_ref(trans, root, path, bytenr,
1351 parent, root_objectid);
1356 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1357 struct btrfs_root *root,
1358 struct btrfs_path *path,
1359 struct btrfs_extent_inline_ref *iref,
1360 int refs_to_drop, int is_data)
1364 BUG_ON(!is_data && refs_to_drop != 1);
1366 ret = update_inline_extent_backref(trans, root, path, iref,
1368 } else if (is_data) {
1369 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1371 ret = btrfs_del_item(trans, root, path);
1376 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1377 struct btrfs_root *root,
1378 u64 bytenr, u64 num_bytes, u64 parent,
1379 u64 root_objectid, u64 owner, u64 offset)
1381 struct btrfs_path *path;
1382 struct extent_buffer *leaf;
1383 struct btrfs_extent_item *item;
1388 path = btrfs_alloc_path();
1393 path->leave_spinning = 1;
1395 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1396 path, bytenr, num_bytes, parent,
1397 root_objectid, owner, offset, 1);
1401 if (ret != -EAGAIN) {
1406 leaf = path->nodes[0];
1407 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1408 refs = btrfs_extent_refs(leaf, item);
1409 btrfs_set_extent_refs(leaf, item, refs + 1);
1411 btrfs_mark_buffer_dirty(leaf);
1412 btrfs_release_path(root->fs_info->extent_root, path);
1415 path->leave_spinning = 1;
1417 /* now insert the actual backref */
1418 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1419 path, bytenr, parent, root_objectid,
1424 btrfs_free_path(path);
1425 finish_current_insert(trans, root->fs_info->extent_root);
1426 del_pending_extents(trans, root->fs_info->extent_root);
1431 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1432 struct btrfs_root *root)
1434 finish_current_insert(trans, root->fs_info->extent_root);
1435 del_pending_extents(trans, root->fs_info->extent_root);
1439 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root, u64 bytenr,
1441 u64 num_bytes, u64 *refs, u64 *flags)
1443 struct btrfs_path *path;
1445 struct btrfs_key key;
1446 struct extent_buffer *l;
1447 struct btrfs_extent_item *item;
1452 WARN_ON(num_bytes < root->sectorsize);
1453 path = btrfs_alloc_path();
1455 key.objectid = bytenr;
1456 key.offset = num_bytes;
1457 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1458 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1463 btrfs_print_leaf(root, path->nodes[0]);
1464 printk("failed to find block number %Lu\n",
1465 (unsigned long long)bytenr);
1470 item_size = btrfs_item_size_nr(l, path->slots[0]);
1471 if (item_size >= sizeof(*item)) {
1472 item = btrfs_item_ptr(l, path->slots[0],
1473 struct btrfs_extent_item);
1474 num_refs = btrfs_extent_refs(l, item);
1475 extent_flags = btrfs_extent_flags(l, item);
1477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1478 struct btrfs_extent_item_v0 *ei0;
1479 BUG_ON(item_size != sizeof(*ei0));
1480 ei0 = btrfs_item_ptr(l, path->slots[0],
1481 struct btrfs_extent_item_v0);
1482 num_refs = btrfs_extent_refs_v0(l, ei0);
1483 /* FIXME: this isn't correct for data */
1484 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1489 BUG_ON(num_refs == 0);
1490 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1494 *flags = extent_flags;
1496 btrfs_free_path(path);
1500 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1501 struct btrfs_root *root,
1502 u64 bytenr, u64 num_bytes, u64 flags)
1504 struct btrfs_path *path;
1506 struct btrfs_key key;
1507 struct extent_buffer *l;
1508 struct btrfs_extent_item *item;
1511 WARN_ON(num_bytes < root->sectorsize);
1512 path = btrfs_alloc_path();
1514 key.objectid = bytenr;
1515 key.offset = num_bytes;
1516 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1517 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1522 btrfs_print_leaf(root, path->nodes[0]);
1523 printk("failed to find block number %Lu\n",
1524 (unsigned long long)bytenr);
1528 item_size = btrfs_item_size_nr(l, path->slots[0]);
1529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1530 if (item_size < sizeof(*item)) {
1531 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1537 item_size = btrfs_item_size_nr(l, path->slots[0]);
1540 BUG_ON(item_size < sizeof(*item));
1541 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1542 flags |= btrfs_extent_flags(l, item);
1543 btrfs_set_extent_flags(l, item, flags);
1545 btrfs_free_path(path);
1546 finish_current_insert(trans, root->fs_info->extent_root);
1547 del_pending_extents(trans, root->fs_info->extent_root);
1551 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1552 struct btrfs_root *root,
1553 struct extent_buffer *buf,
1554 int record_parent, int inc)
1561 struct btrfs_key key;
1562 struct btrfs_file_extent_item *fi;
1566 int (*process_func)(struct btrfs_trans_handle *trans,
1567 struct btrfs_root *root,
1568 u64, u64, u64, u64, u64, u64);
1570 ref_root = btrfs_header_owner(buf);
1571 nritems = btrfs_header_nritems(buf);
1572 level = btrfs_header_level(buf);
1574 if (!root->ref_cows && level == 0)
1578 process_func = btrfs_inc_extent_ref;
1580 process_func = btrfs_free_extent;
1583 parent = buf->start;
1587 for (i = 0; i < nritems; i++) {
1590 btrfs_item_key_to_cpu(buf, &key, i);
1591 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1593 fi = btrfs_item_ptr(buf, i,
1594 struct btrfs_file_extent_item);
1595 if (btrfs_file_extent_type(buf, fi) ==
1596 BTRFS_FILE_EXTENT_INLINE)
1598 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1602 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1603 key.offset -= btrfs_file_extent_offset(buf, fi);
1604 ret = process_func(trans, root, bytenr, num_bytes,
1605 parent, ref_root, key.objectid,
1612 bytenr = btrfs_node_blockptr(buf, i);
1613 num_bytes = btrfs_level_size(root, level - 1);
1614 ret = process_func(trans, root, bytenr, num_bytes,
1615 parent, ref_root, level - 1, 0);
1628 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1629 struct extent_buffer *buf, int record_parent)
1631 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1634 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1635 struct extent_buffer *buf, int record_parent)
1637 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1640 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1641 struct btrfs_root *root,
1642 struct btrfs_path *path,
1643 struct btrfs_block_group_cache *cache)
1647 struct btrfs_root *extent_root = root->fs_info->extent_root;
1649 struct extent_buffer *leaf;
1651 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1656 leaf = path->nodes[0];
1657 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1658 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1659 btrfs_mark_buffer_dirty(leaf);
1660 btrfs_release_path(extent_root, path);
1662 finish_current_insert(trans, extent_root);
1663 pending_ret = del_pending_extents(trans, extent_root);
1672 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1673 struct btrfs_root *root)
1675 struct extent_io_tree *block_group_cache;
1676 struct btrfs_block_group_cache *cache;
1678 struct btrfs_path *path;
1684 block_group_cache = &root->fs_info->block_group_cache;
1685 path = btrfs_alloc_path();
1690 ret = find_first_extent_bit(block_group_cache, last,
1691 &start, &end, BLOCK_GROUP_DIRTY);
1700 ret = get_state_private(block_group_cache, start, &ptr);
1703 clear_extent_bits(block_group_cache, start, end,
1704 BLOCK_GROUP_DIRTY, GFP_NOFS);
1706 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1707 ret = write_one_cache_group(trans, root, path, cache);
1710 btrfs_free_path(path);
1714 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1717 struct list_head *head = &info->space_info;
1718 struct list_head *cur;
1719 struct btrfs_space_info *found;
1720 list_for_each(cur, head) {
1721 found = list_entry(cur, struct btrfs_space_info, list);
1722 if (found->flags & flags)
1729 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1730 u64 total_bytes, u64 bytes_used,
1731 struct btrfs_space_info **space_info)
1733 struct btrfs_space_info *found;
1735 found = __find_space_info(info, flags);
1737 found->total_bytes += total_bytes;
1738 found->bytes_used += bytes_used;
1739 WARN_ON(found->total_bytes < found->bytes_used);
1740 *space_info = found;
1743 found = kmalloc(sizeof(*found), GFP_NOFS);
1747 list_add(&found->list, &info->space_info);
1748 found->flags = flags;
1749 found->total_bytes = total_bytes;
1750 found->bytes_used = bytes_used;
1751 found->bytes_pinned = 0;
1753 *space_info = found;
1758 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1760 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1761 BTRFS_BLOCK_GROUP_RAID1 |
1762 BTRFS_BLOCK_GROUP_RAID10 |
1763 BTRFS_BLOCK_GROUP_DUP);
1765 if (flags & BTRFS_BLOCK_GROUP_DATA)
1766 fs_info->avail_data_alloc_bits |= extra_flags;
1767 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1768 fs_info->avail_metadata_alloc_bits |= extra_flags;
1769 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1770 fs_info->avail_system_alloc_bits |= extra_flags;
1774 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1775 struct btrfs_root *extent_root, u64 alloc_bytes,
1778 struct btrfs_space_info *space_info;
1784 space_info = __find_space_info(extent_root->fs_info, flags);
1786 ret = update_space_info(extent_root->fs_info, flags,
1790 BUG_ON(!space_info);
1792 if (space_info->full)
1795 thresh = div_factor(space_info->total_bytes, 7);
1796 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1800 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1802 if (ret == -ENOSPC) {
1803 space_info->full = 1;
1809 ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1810 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1815 static int update_block_group(struct btrfs_trans_handle *trans,
1816 struct btrfs_root *root,
1817 u64 bytenr, u64 num_bytes, int alloc,
1820 struct btrfs_block_group_cache *cache;
1821 struct btrfs_fs_info *info = root->fs_info;
1822 u64 total = num_bytes;
1828 /* block accounting for super block */
1829 old_val = btrfs_super_bytes_used(&info->super_copy);
1831 old_val += num_bytes;
1833 old_val -= num_bytes;
1834 btrfs_set_super_bytes_used(&info->super_copy, old_val);
1836 /* block accounting for root item */
1837 old_val = btrfs_root_used(&root->root_item);
1839 old_val += num_bytes;
1841 old_val -= num_bytes;
1842 btrfs_set_root_used(&root->root_item, old_val);
1845 cache = btrfs_lookup_block_group(info, bytenr);
1849 byte_in_group = bytenr - cache->key.objectid;
1850 WARN_ON(byte_in_group > cache->key.offset);
1851 start = cache->key.objectid;
1852 end = start + cache->key.offset - 1;
1853 set_extent_bits(&info->block_group_cache, start, end,
1854 BLOCK_GROUP_DIRTY, GFP_NOFS);
1856 old_val = btrfs_block_group_used(&cache->item);
1857 num_bytes = min(total, cache->key.offset - byte_in_group);
1859 old_val += num_bytes;
1860 cache->space_info->bytes_used += num_bytes;
1862 old_val -= num_bytes;
1863 cache->space_info->bytes_used -= num_bytes;
1865 set_extent_dirty(&info->free_space_cache,
1866 bytenr, bytenr + num_bytes - 1,
1870 btrfs_set_block_group_used(&cache->item, old_val);
1872 bytenr += num_bytes;
1877 static int update_pinned_extents(struct btrfs_root *root,
1878 u64 bytenr, u64 num, int pin)
1881 struct btrfs_block_group_cache *cache;
1882 struct btrfs_fs_info *fs_info = root->fs_info;
1885 set_extent_dirty(&fs_info->pinned_extents,
1886 bytenr, bytenr + num - 1, GFP_NOFS);
1888 clear_extent_dirty(&fs_info->pinned_extents,
1889 bytenr, bytenr + num - 1, GFP_NOFS);
1892 cache = btrfs_lookup_block_group(fs_info, bytenr);
1894 len = min(num, cache->key.offset -
1895 (bytenr - cache->key.objectid));
1897 cache->pinned += len;
1898 cache->space_info->bytes_pinned += len;
1899 fs_info->total_pinned += len;
1901 cache->pinned -= len;
1902 cache->space_info->bytes_pinned -= len;
1903 fs_info->total_pinned -= len;
1911 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1916 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1920 ret = find_first_extent_bit(pinned_extents, last,
1921 &start, &end, EXTENT_DIRTY);
1924 set_extent_dirty(copy, start, end, GFP_NOFS);
1930 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1931 struct btrfs_root *root,
1932 struct extent_io_tree *unpin)
1937 struct extent_io_tree *free_space_cache;
1938 free_space_cache = &root->fs_info->free_space_cache;
1941 ret = find_first_extent_bit(unpin, 0, &start, &end,
1945 update_pinned_extents(root, start, end + 1 - start, 0);
1946 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1947 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1952 static int finish_current_insert(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *extent_root)
1958 struct btrfs_fs_info *info = extent_root->fs_info;
1959 struct btrfs_path *path;
1960 struct pending_extent_op *extent_op;
1961 struct btrfs_key key;
1964 path = btrfs_alloc_path();
1967 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1968 &end, EXTENT_LOCKED);
1972 ret = get_state_private(&info->extent_ins, start, &priv);
1974 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1976 if (extent_op->type == PENDING_EXTENT_INSERT) {
1977 key.objectid = start;
1978 key.offset = end + 1 - start;
1979 key.type = BTRFS_EXTENT_ITEM_KEY;
1980 ret = alloc_reserved_tree_block(trans, extent_root,
1981 extent_root->root_key.objectid,
1985 extent_op->level, &key);
1990 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1994 btrfs_free_path(path);
1998 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1999 struct btrfs_root *root,
2000 u64 bytenr, u64 num_bytes, int is_data)
2003 struct extent_buffer *buf;
2008 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2012 /* we can reuse a block if it hasn't been written
2013 * and it is from this transaction. We can't
2014 * reuse anything from the tree log root because
2015 * it has tiny sub-transactions.
2017 if (btrfs_buffer_uptodate(buf, 0)) {
2018 u64 header_owner = btrfs_header_owner(buf);
2019 u64 header_transid = btrfs_header_generation(buf);
2020 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2021 header_transid == trans->transid &&
2022 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2023 clean_tree_block(NULL, root, buf);
2024 free_extent_buffer(buf);
2028 free_extent_buffer(buf);
2030 update_pinned_extents(root, bytenr, num_bytes, 1);
2037 * remove an extent from the root, returns 0 on success
2039 static int __free_extent(struct btrfs_trans_handle *trans,
2040 struct btrfs_root *root,
2041 u64 bytenr, u64 num_bytes, u64 parent,
2042 u64 root_objectid, u64 owner_objectid,
2043 u64 owner_offset, int refs_to_drop)
2046 struct btrfs_key key;
2047 struct btrfs_path *path;
2048 struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2049 struct btrfs_root *extent_root = root->fs_info->extent_root;
2050 struct extent_buffer *leaf;
2051 struct btrfs_extent_item *ei;
2052 struct btrfs_extent_inline_ref *iref;
2055 int extent_slot = 0;
2056 int found_extent = 0;
2061 path = btrfs_alloc_path();
2066 path->leave_spinning = 1;
2068 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2069 BUG_ON(!is_data && refs_to_drop != 1);
2071 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2072 bytenr, num_bytes, parent,
2073 root_objectid, owner_objectid,
2076 extent_slot = path->slots[0];
2077 while (extent_slot >= 0) {
2078 btrfs_item_key_to_cpu(path->nodes[0], &key,
2080 if (key.objectid != bytenr)
2082 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2083 key.offset == num_bytes) {
2087 if (path->slots[0] - extent_slot > 5)
2091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2092 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2093 if (found_extent && item_size < sizeof(*ei))
2096 if (!found_extent) {
2098 ret = remove_extent_backref(trans, extent_root, path,
2102 btrfs_release_path(extent_root, path);
2103 path->leave_spinning = 1;
2105 key.objectid = bytenr;
2106 key.type = BTRFS_EXTENT_ITEM_KEY;
2107 key.offset = num_bytes;
2109 ret = btrfs_search_slot(trans, extent_root,
2112 printk(KERN_ERR "umm, got %d back from search"
2113 ", was looking for %llu\n", ret,
2114 (unsigned long long)bytenr);
2115 btrfs_print_leaf(extent_root, path->nodes[0]);
2118 extent_slot = path->slots[0];
2121 btrfs_print_leaf(extent_root, path->nodes[0]);
2123 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2124 "parent %llu root %llu owner %llu offset %llu\n",
2125 (unsigned long long)bytenr,
2126 (unsigned long long)parent,
2127 (unsigned long long)root_objectid,
2128 (unsigned long long)owner_objectid,
2129 (unsigned long long)owner_offset);
2132 leaf = path->nodes[0];
2133 item_size = btrfs_item_size_nr(leaf, extent_slot);
2134 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2135 if (item_size < sizeof(*ei)) {
2136 BUG_ON(found_extent || extent_slot != path->slots[0]);
2137 ret = convert_extent_item_v0(trans, extent_root, path,
2141 btrfs_release_path(extent_root, path);
2142 path->leave_spinning = 1;
2144 key.objectid = bytenr;
2145 key.type = BTRFS_EXTENT_ITEM_KEY;
2146 key.offset = num_bytes;
2148 ret = btrfs_search_slot(trans, extent_root, &key, path,
2151 printk(KERN_ERR "umm, got %d back from search"
2152 ", was looking for %llu\n", ret,
2153 (unsigned long long)bytenr);
2154 btrfs_print_leaf(extent_root, path->nodes[0]);
2157 extent_slot = path->slots[0];
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, extent_slot);
2162 BUG_ON(item_size < sizeof(*ei));
2163 ei = btrfs_item_ptr(leaf, extent_slot,
2164 struct btrfs_extent_item);
2165 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2166 struct btrfs_tree_block_info *bi;
2167 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2168 bi = (struct btrfs_tree_block_info *)(ei + 1);
2169 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2172 refs = btrfs_extent_refs(leaf, ei);
2173 BUG_ON(refs < refs_to_drop);
2174 refs -= refs_to_drop;
2178 * In the case of inline back ref, reference count will
2179 * be updated by remove_extent_backref
2182 BUG_ON(!found_extent);
2184 btrfs_set_extent_refs(leaf, ei, refs);
2185 btrfs_mark_buffer_dirty(leaf);
2188 ret = remove_extent_backref(trans, extent_root, path,
2198 BUG_ON(is_data && refs_to_drop !=
2199 extent_data_ref_count(root, path, iref));
2201 BUG_ON(path->slots[0] != extent_slot);
2203 BUG_ON(path->slots[0] != extent_slot + 1);
2204 path->slots[0] = extent_slot;
2209 if (ops && ops->free_extent) {
2210 ret = ops->free_extent(root, bytenr, num_bytes);
2218 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2225 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2228 btrfs_release_path(extent_root, path);
2231 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2235 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2239 btrfs_free_path(path);
2240 finish_current_insert(trans, extent_root);
2245 * find all the blocks marked as pending in the radix tree and remove
2246 * them from the extent map
2248 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2249 btrfs_root *extent_root)
2256 struct extent_io_tree *pending_del;
2257 struct extent_io_tree *extent_ins;
2258 struct pending_extent_op *extent_op;
2260 extent_ins = &extent_root->fs_info->extent_ins;
2261 pending_del = &extent_root->fs_info->pending_del;
2264 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2269 ret = get_state_private(pending_del, start, &priv);
2271 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2273 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2276 if (!test_range_bit(extent_ins, start, end,
2277 EXTENT_LOCKED, 0)) {
2278 ret = __free_extent(trans, extent_root,
2279 start, end + 1 - start, 0,
2280 extent_root->root_key.objectid,
2281 extent_op->level, 0, 1);
2285 ret = get_state_private(extent_ins, start, &priv);
2287 extent_op = (struct pending_extent_op *)
2288 (unsigned long)priv;
2290 clear_extent_bits(extent_ins, start, end,
2291 EXTENT_LOCKED, GFP_NOFS);
2293 if (extent_op->type == PENDING_BACKREF_UPDATE)
2305 * remove an extent from the root, returns 0 on success
2308 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2309 struct btrfs_root *root,
2310 u64 bytenr, u64 num_bytes, u64 parent,
2311 u64 root_objectid, u64 owner, u64 offset)
2313 struct btrfs_root *extent_root = root->fs_info->extent_root;
2317 WARN_ON(num_bytes < root->sectorsize);
2318 if (root == extent_root) {
2319 struct pending_extent_op *extent_op;
2321 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2324 extent_op->type = PENDING_EXTENT_DELETE;
2325 extent_op->bytenr = bytenr;
2326 extent_op->num_bytes = num_bytes;
2327 extent_op->level = (int)owner;
2329 set_extent_bits(&root->fs_info->pending_del,
2330 bytenr, bytenr + num_bytes - 1,
2331 EXTENT_LOCKED, GFP_NOFS);
2332 set_state_private(&root->fs_info->pending_del,
2333 bytenr, (unsigned long)extent_op);
2336 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2337 root_objectid, owner, offset, 1);
2338 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2339 return ret ? ret : pending_ret;
2342 static u64 stripe_align(struct btrfs_root *root, u64 val)
2344 u64 mask = ((u64)root->stripesize - 1);
2345 u64 ret = (val + mask) & ~mask;
2350 * walks the btree of allocated extents and find a hole of a given size.
2351 * The key ins is changed to record the hole:
2352 * ins->objectid == block start
2353 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2354 * ins->offset == number of blocks
2355 * Any available blocks before search_start are skipped.
2357 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2358 struct btrfs_root *orig_root,
2359 u64 num_bytes, u64 empty_size,
2360 u64 search_start, u64 search_end,
2361 u64 hint_byte, struct btrfs_key *ins,
2362 u64 exclude_start, u64 exclude_nr,
2366 u64 orig_search_start = search_start;
2367 struct btrfs_root * root = orig_root->fs_info->extent_root;
2368 struct btrfs_fs_info *info = root->fs_info;
2369 u64 total_needed = num_bytes;
2370 struct btrfs_block_group_cache *block_group;
2374 WARN_ON(num_bytes < root->sectorsize);
2375 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2378 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2380 hint_byte = search_start;
2381 block_group = btrfs_find_block_group(root, block_group,
2382 hint_byte, data, 1);
2384 block_group = btrfs_find_block_group(root,
2386 search_start, data, 1);
2389 total_needed += empty_size;
2393 block_group = btrfs_lookup_first_block_group(info,
2396 block_group = btrfs_lookup_first_block_group(info,
2399 ret = find_search_start(root, &block_group, &search_start,
2400 total_needed, data);
2404 search_start = stripe_align(root, search_start);
2405 ins->objectid = search_start;
2406 ins->offset = num_bytes;
2408 if (ins->objectid + num_bytes >
2409 block_group->key.objectid + block_group->key.offset) {
2410 search_start = block_group->key.objectid +
2411 block_group->key.offset;
2415 if (test_range_bit(&info->extent_ins, ins->objectid,
2416 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2417 search_start = ins->objectid + num_bytes;
2421 if (test_range_bit(&info->pinned_extents, ins->objectid,
2422 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2423 search_start = ins->objectid + num_bytes;
2427 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2428 ins->objectid < exclude_start + exclude_nr)) {
2429 search_start = exclude_start + exclude_nr;
2433 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2434 block_group = btrfs_lookup_block_group(info, ins->objectid);
2436 trans->block_group = block_group;
2438 ins->offset = num_bytes;
2442 block_group = btrfs_lookup_first_block_group(info, search_start);
2444 search_start = orig_search_start;
2451 total_needed -= empty_size;
2457 block_group = btrfs_find_block_group(root, block_group,
2458 search_start, data, 0);
2465 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2466 struct btrfs_root *root,
2467 u64 num_bytes, u64 empty_size,
2468 u64 hint_byte, u64 search_end,
2469 struct btrfs_key *ins, int data)
2472 u64 search_start = 0;
2474 struct btrfs_fs_info *info = root->fs_info;
2476 if (info->extent_ops) {
2477 struct btrfs_extent_ops *ops = info->extent_ops;
2478 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2484 alloc_profile = info->avail_data_alloc_bits &
2485 info->data_alloc_profile;
2486 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2487 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2488 info->system_allocs >= 0) {
2489 alloc_profile = info->avail_system_alloc_bits &
2490 info->system_alloc_profile;
2491 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2493 alloc_profile = info->avail_metadata_alloc_bits &
2494 info->metadata_alloc_profile;
2495 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2498 if (root->ref_cows) {
2499 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2500 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2502 BTRFS_BLOCK_GROUP_METADATA);
2505 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2506 num_bytes + 2 * 1024 * 1024, data);
2510 WARN_ON(num_bytes < root->sectorsize);
2511 ret = find_free_extent(trans, root, num_bytes, empty_size,
2512 search_start, search_end, hint_byte, ins,
2513 trans->alloc_exclude_start,
2514 trans->alloc_exclude_nr, data);
2517 clear_extent_dirty(&root->fs_info->free_space_cache,
2518 ins->objectid, ins->objectid + ins->offset - 1,
2523 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2524 struct btrfs_root *root,
2525 u64 root_objectid, u64 generation,
2526 u64 flags, struct btrfs_disk_key *key,
2527 int level, struct btrfs_key *ins)
2530 struct btrfs_fs_info *fs_info = root->fs_info;
2531 struct btrfs_extent_item *extent_item;
2532 struct btrfs_tree_block_info *block_info;
2533 struct btrfs_extent_inline_ref *iref;
2534 struct btrfs_path *path;
2535 struct extent_buffer *leaf;
2536 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2538 path = btrfs_alloc_path();
2541 path->leave_spinning = 1;
2542 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2546 leaf = path->nodes[0];
2547 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2548 struct btrfs_extent_item);
2549 btrfs_set_extent_refs(leaf, extent_item, 1);
2550 btrfs_set_extent_generation(leaf, extent_item, generation);
2551 btrfs_set_extent_flags(leaf, extent_item,
2552 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2553 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2555 btrfs_set_tree_block_key(leaf, block_info, key);
2556 btrfs_set_tree_block_level(leaf, block_info, level);
2558 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2559 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2560 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2562 btrfs_mark_buffer_dirty(leaf);
2563 btrfs_free_path(path);
2565 ret = update_block_group(trans, root, ins->objectid, ins->offset,
2568 printk(KERN_ERR "btrfs update block group failed for %llu "
2569 "%llu\n", (unsigned long long)ins->objectid,
2570 (unsigned long long)ins->offset);
2576 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2577 struct btrfs_root *root, u64 num_bytes,
2578 u64 root_objectid, u64 generation,
2579 u64 flags, struct btrfs_disk_key *key,
2580 int level, u64 empty_size, u64 hint_byte,
2581 u64 search_end, struct btrfs_key *ins)
2584 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2585 hint_byte, search_end, ins, 0);
2588 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2589 struct pending_extent_op *extent_op;
2591 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2594 extent_op->type = PENDING_EXTENT_INSERT;
2595 extent_op->bytenr = ins->objectid;
2596 extent_op->num_bytes = ins->offset;
2597 extent_op->level = level;
2598 extent_op->flags = flags;
2599 memcpy(&extent_op->key, key, sizeof(*key));
2601 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2602 ins->objectid + ins->offset - 1,
2603 EXTENT_LOCKED, GFP_NOFS);
2604 set_state_private(&root->fs_info->extent_ins,
2605 ins->objectid, (unsigned long)extent_op);
2607 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2610 finish_current_insert(trans, root->fs_info->extent_root);
2611 del_pending_extents(trans, root->fs_info->extent_root);
2617 * helper function to allocate a block for a given tree
2618 * returns the tree buffer or NULL.
2620 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2621 struct btrfs_root *root,
2622 u32 blocksize, u64 root_objectid,
2623 struct btrfs_disk_key *key, int level,
2624 u64 hint, u64 empty_size)
2626 struct btrfs_key ins;
2628 struct extent_buffer *buf;
2630 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2631 trans->transid, 0, key, level,
2632 empty_size, hint, (u64)-1, &ins);
2635 return ERR_PTR(ret);
2638 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2640 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2641 0, root->root_key.objectid, level, 0);
2643 return ERR_PTR(-ENOMEM);
2645 btrfs_set_buffer_uptodate(buf);
2646 trans->blocks_used++;
2653 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2654 struct btrfs_root *root,
2655 struct extent_buffer *leaf)
2658 u64 leaf_generation;
2659 struct btrfs_key key;
2660 struct btrfs_file_extent_item *fi;
2665 BUG_ON(!btrfs_is_leaf(leaf));
2666 nritems = btrfs_header_nritems(leaf);
2667 leaf_owner = btrfs_header_owner(leaf);
2668 leaf_generation = btrfs_header_generation(leaf);
2670 for (i = 0; i < nritems; i++) {
2673 btrfs_item_key_to_cpu(leaf, &key, i);
2674 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2676 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2677 if (btrfs_file_extent_type(leaf, fi) ==
2678 BTRFS_FILE_EXTENT_INLINE)
2681 * FIXME make sure to insert a trans record that
2682 * repeats the snapshot del on crash
2684 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2685 if (disk_bytenr == 0)
2687 ret = btrfs_free_extent(trans, root, disk_bytenr,
2688 btrfs_file_extent_disk_num_bytes(leaf, fi),
2689 leaf->start, leaf_owner, leaf_generation,
2696 static void noinline reada_walk_down(struct btrfs_root *root,
2697 struct extent_buffer *node,
2710 nritems = btrfs_header_nritems(node);
2711 level = btrfs_header_level(node);
2715 for (i = slot; i < nritems && skipped < 32; i++) {
2716 bytenr = btrfs_node_blockptr(node, i);
2717 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2718 (last > bytenr && last - bytenr > 32 * 1024))) {
2722 blocksize = btrfs_level_size(root, level - 1);
2724 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2732 mutex_unlock(&root->fs_info->fs_mutex);
2733 ret = readahead_tree_block(root, bytenr, blocksize,
2734 btrfs_node_ptr_generation(node, i));
2735 last = bytenr + blocksize;
2737 mutex_lock(&root->fs_info->fs_mutex);
2744 * helper function for drop_snapshot, this walks down the tree dropping ref
2745 * counts as it goes.
2747 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 struct btrfs_path *path, int *level)
2755 struct extent_buffer *next;
2756 struct extent_buffer *cur;
2757 struct extent_buffer *parent;
2762 WARN_ON(*level < 0);
2763 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2764 ret = btrfs_lookup_extent_ref(trans, root,
2765 path->nodes[*level]->start,
2766 path->nodes[*level]->len, &refs);
2772 * walk down to the last node level and free all the leaves
2774 while(*level >= 0) {
2775 WARN_ON(*level < 0);
2776 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2777 cur = path->nodes[*level];
2779 if (btrfs_header_level(cur) != *level)
2782 if (path->slots[*level] >=
2783 btrfs_header_nritems(cur))
2786 ret = drop_leaf_ref(trans, root, cur);
2790 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2791 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2792 blocksize = btrfs_level_size(root, *level - 1);
2793 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2797 parent = path->nodes[*level];
2798 root_owner = btrfs_header_owner(parent);
2799 root_gen = btrfs_header_generation(parent);
2800 path->slots[*level]++;
2801 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2802 parent->start, root_owner,
2803 root_gen, *level - 1, 1);
2807 next = btrfs_find_tree_block(root, bytenr, blocksize);
2808 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2809 free_extent_buffer(next);
2810 reada_walk_down(root, cur, path->slots[*level]);
2811 mutex_unlock(&root->fs_info->fs_mutex);
2812 next = read_tree_block(root, bytenr, blocksize,
2814 mutex_lock(&root->fs_info->fs_mutex);
2816 WARN_ON(*level <= 0);
2817 if (path->nodes[*level-1])
2818 free_extent_buffer(path->nodes[*level-1]);
2819 path->nodes[*level-1] = next;
2820 *level = btrfs_header_level(next);
2821 path->slots[*level] = 0;
2824 WARN_ON(*level < 0);
2825 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2827 if (path->nodes[*level] == root->node) {
2828 root_owner = root->root_key.objectid;
2829 parent = path->nodes[*level];
2831 parent = path->nodes[*level + 1];
2832 root_owner = btrfs_header_owner(parent);
2835 root_gen = btrfs_header_generation(parent);
2836 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2837 path->nodes[*level]->len, parent->start,
2838 root_owner, root_gen, *level, 1);
2839 free_extent_buffer(path->nodes[*level]);
2840 path->nodes[*level] = NULL;
2847 * helper for dropping snapshots. This walks back up the tree in the path
2848 * to find the first node higher up where we haven't yet gone through
2851 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2852 struct btrfs_root *root,
2853 struct btrfs_path *path, int *level)
2857 struct btrfs_root_item *root_item = &root->root_item;
2862 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2863 slot = path->slots[i];
2864 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2865 struct extent_buffer *node;
2866 struct btrfs_disk_key disk_key;
2867 node = path->nodes[i];
2870 WARN_ON(*level == 0);
2871 btrfs_node_key(node, &disk_key, path->slots[i]);
2872 memcpy(&root_item->drop_progress,
2873 &disk_key, sizeof(disk_key));
2874 root_item->drop_level = i;
2877 struct extent_buffer *parent;
2878 if (path->nodes[*level] == root->node)
2879 parent = path->nodes[*level];
2881 parent = path->nodes[*level + 1];
2883 root_owner = btrfs_header_owner(parent);
2884 root_gen = btrfs_header_generation(parent);
2885 ret = btrfs_free_extent(trans, root,
2886 path->nodes[*level]->start,
2887 path->nodes[*level]->len,
2888 parent->start, root_owner,
2889 root_gen, *level, 1);
2891 free_extent_buffer(path->nodes[*level]);
2892 path->nodes[*level] = NULL;
2900 * drop the reference count on the tree rooted at 'snap'. This traverses
2901 * the tree freeing any blocks that have a ref count of zero after being
2904 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2910 struct btrfs_path *path;
2913 struct btrfs_root_item *root_item = &root->root_item;
2915 path = btrfs_alloc_path();
2918 level = btrfs_header_level(root->node);
2920 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2921 path->nodes[level] = root->node;
2922 extent_buffer_get(root->node);
2923 path->slots[level] = 0;
2925 struct btrfs_key key;
2926 struct btrfs_disk_key found_key;
2927 struct extent_buffer *node;
2929 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2930 level = root_item->drop_level;
2931 path->lowest_level = level;
2932 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2937 node = path->nodes[level];
2938 btrfs_node_key(node, &found_key, path->slots[level]);
2939 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2940 sizeof(found_key)));
2943 wret = walk_down_tree(trans, root, path, &level);
2949 wret = walk_up_tree(trans, root, path, &level);
2959 for (i = 0; i <= orig_level; i++) {
2960 if (path->nodes[i]) {
2961 free_extent_buffer(path->nodes[i]);
2962 path->nodes[i] = NULL;
2966 btrfs_free_path(path);
2972 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2979 ret = find_first_extent_bit(&info->block_group_cache, 0,
2980 &start, &end, (unsigned int)-1);
2983 ret = get_state_private(&info->block_group_cache, start, &ptr);
2985 kfree((void *)(unsigned long)ptr);
2986 clear_extent_bits(&info->block_group_cache, start,
2987 end, (unsigned int)-1, GFP_NOFS);
2990 ret = find_first_extent_bit(&info->free_space_cache, 0,
2991 &start, &end, EXTENT_DIRTY);
2994 clear_extent_dirty(&info->free_space_cache, start,
3000 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3001 struct btrfs_key *key)
3004 struct btrfs_key found_key;
3005 struct extent_buffer *leaf;
3008 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3012 slot = path->slots[0];
3013 leaf = path->nodes[0];
3014 if (slot >= btrfs_header_nritems(leaf)) {
3015 ret = btrfs_next_leaf(root, path);
3022 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3024 if (found_key.objectid >= key->objectid &&
3025 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3034 int btrfs_read_block_groups(struct btrfs_root *root)
3036 struct btrfs_path *path;
3039 struct btrfs_block_group_cache *cache;
3040 struct btrfs_fs_info *info = root->fs_info;
3041 struct btrfs_space_info *space_info;
3042 struct extent_io_tree *block_group_cache;
3043 struct btrfs_key key;
3044 struct btrfs_key found_key;
3045 struct extent_buffer *leaf;
3047 block_group_cache = &info->block_group_cache;
3049 root = info->extent_root;
3052 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3053 path = btrfs_alloc_path();
3058 ret = find_first_block_group(root, path, &key);
3066 leaf = path->nodes[0];
3067 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3068 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3074 read_extent_buffer(leaf, &cache->item,
3075 btrfs_item_ptr_offset(leaf, path->slots[0]),
3076 sizeof(cache->item));
3077 memcpy(&cache->key, &found_key, sizeof(found_key));
3080 key.objectid = found_key.objectid + found_key.offset;
3081 btrfs_release_path(root, path);
3082 cache->flags = btrfs_block_group_flags(&cache->item);
3084 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3085 bit = BLOCK_GROUP_DATA;
3086 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3087 bit = BLOCK_GROUP_SYSTEM;
3088 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3089 bit = BLOCK_GROUP_METADATA;
3091 set_avail_alloc_bits(info, cache->flags);
3092 if (btrfs_chunk_readonly(root, cache->key.objectid))
3095 ret = update_space_info(info, cache->flags, found_key.offset,
3096 btrfs_block_group_used(&cache->item),
3099 cache->space_info = space_info;
3101 /* use EXTENT_LOCKED to prevent merging */
3102 set_extent_bits(block_group_cache, found_key.objectid,
3103 found_key.objectid + found_key.offset - 1,
3104 bit | EXTENT_LOCKED, GFP_NOFS);
3105 set_state_private(block_group_cache, found_key.objectid,
3106 (unsigned long)cache);
3110 btrfs_free_path(path);
3114 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3115 struct btrfs_root *root, u64 bytes_used,
3116 u64 type, u64 chunk_objectid, u64 chunk_offset,
3121 struct btrfs_root *extent_root;
3122 struct btrfs_block_group_cache *cache;
3123 struct extent_io_tree *block_group_cache;
3125 extent_root = root->fs_info->extent_root;
3126 block_group_cache = &root->fs_info->block_group_cache;
3128 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3130 cache->key.objectid = chunk_offset;
3131 cache->key.offset = size;
3133 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3134 btrfs_set_block_group_used(&cache->item, bytes_used);
3135 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3136 cache->flags = type;
3137 btrfs_set_block_group_flags(&cache->item, type);
3139 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3140 &cache->space_info);
3143 bit = block_group_state_bits(type);
3144 set_extent_bits(block_group_cache, chunk_offset,
3145 chunk_offset + size - 1,
3146 bit | EXTENT_LOCKED, GFP_NOFS);
3148 set_state_private(block_group_cache, chunk_offset,
3149 (unsigned long)cache);
3150 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3151 sizeof(cache->item));
3154 finish_current_insert(trans, extent_root);
3155 ret = del_pending_extents(trans, extent_root);
3157 set_avail_alloc_bits(extent_root->fs_info, type);
3162 * This is for converter use only.
3164 * In that case, we don't know where are free blocks located.
3165 * Therefore all block group cache entries must be setup properly
3166 * before doing any block allocation.
3168 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3169 struct btrfs_root *root)
3177 u64 total_metadata = 0;
3181 struct btrfs_root *extent_root;
3182 struct btrfs_block_group_cache *cache;
3183 struct extent_io_tree *block_group_cache;
3185 extent_root = root->fs_info->extent_root;
3186 block_group_cache = &root->fs_info->block_group_cache;
3187 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3188 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3189 group_align = 64 * root->sectorsize;
3192 while (cur_start < total_bytes) {
3193 group_size = total_bytes / 12;
3194 group_size = min_t(u64, group_size, total_bytes - cur_start);
3195 if (cur_start == 0) {
3196 bit = BLOCK_GROUP_SYSTEM;
3197 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3199 group_size &= ~(group_align - 1);
3200 group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3201 group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3203 group_size &= ~(group_align - 1);
3204 if (total_data >= total_metadata * 2) {
3205 group_type = BTRFS_BLOCK_GROUP_METADATA;
3206 group_size = min_t(u64, group_size,
3207 1ULL * 1024 * 1024 * 1024);
3208 total_metadata += group_size;
3210 group_type = BTRFS_BLOCK_GROUP_DATA;
3211 group_size = min_t(u64, group_size,
3212 5ULL * 1024 * 1024 * 1024);
3213 total_data += group_size;
3215 if ((total_bytes - cur_start) * 4 < group_size * 5)
3216 group_size = total_bytes - cur_start;
3219 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3222 cache->key.objectid = cur_start;
3223 cache->key.offset = group_size;
3224 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3226 btrfs_set_block_group_used(&cache->item, 0);
3227 btrfs_set_block_group_chunk_objectid(&cache->item,
3229 btrfs_set_block_group_flags(&cache->item, group_type);
3231 cache->flags = group_type;
3233 ret = update_space_info(root->fs_info, group_type, group_size,
3234 0, &cache->space_info);
3236 set_avail_alloc_bits(extent_root->fs_info, group_type);
3238 set_extent_bits(block_group_cache, cur_start,
3239 cur_start + group_size - 1,
3240 bit | EXTENT_LOCKED, GFP_NOFS);
3241 set_state_private(block_group_cache, cur_start,
3242 (unsigned long)cache);
3243 cur_start += group_size;
3245 /* then insert all the items */
3247 while(cur_start < total_bytes) {
3248 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3251 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3252 sizeof(cache->item));
3255 finish_current_insert(trans, extent_root);
3256 ret = del_pending_extents(trans, extent_root);
3259 cur_start = cache->key.objectid + cache->key.offset;
3264 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3265 struct btrfs_root *root,
3266 u64 bytenr, u64 num_bytes, int alloc,
3269 return update_block_group(trans, root, bytenr, num_bytes,