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);
1044 leaf = path->nodes[0];
1045 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047 if (item_size < sizeof(*ei)) {
1052 ret = convert_extent_item_v0(trans, root, path, owner,
1058 leaf = path->nodes[0];
1059 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1062 BUG_ON(item_size < sizeof(*ei));
1064 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1065 flags = btrfs_extent_flags(leaf, ei);
1067 ptr = (unsigned long)(ei + 1);
1068 end = (unsigned long)ei + item_size;
1070 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1071 ptr += sizeof(struct btrfs_tree_block_info);
1074 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1083 iref = (struct btrfs_extent_inline_ref *)ptr;
1084 type = btrfs_extent_inline_ref_type(leaf, iref);
1088 ptr += btrfs_extent_inline_ref_size(type);
1092 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1093 struct btrfs_extent_data_ref *dref;
1094 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1095 if (match_extent_data_ref(leaf, dref, root_objectid,
1100 if (hash_extent_data_ref_item(leaf, dref) <
1101 hash_extent_data_ref(root_objectid, owner, offset))
1105 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1107 if (parent == ref_offset) {
1111 if (ref_offset < parent)
1114 if (root_objectid == ref_offset) {
1118 if (ref_offset < root_objectid)
1122 ptr += btrfs_extent_inline_ref_size(type);
1124 if (err == -ENOENT && insert) {
1125 if (item_size + extra_size >=
1126 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1131 * To add new inline back ref, we have to make sure
1132 * there is no corresponding back ref item.
1133 * For simplicity, we just do not add new inline back
1134 * ref if there is any back ref item.
1136 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1137 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1142 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1147 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1148 struct btrfs_root *root,
1149 struct btrfs_path *path,
1150 struct btrfs_extent_inline_ref *iref,
1151 u64 parent, u64 root_objectid,
1152 u64 owner, u64 offset, int refs_to_add)
1154 struct extent_buffer *leaf;
1155 struct btrfs_extent_item *ei;
1158 unsigned long item_offset;
1164 leaf = path->nodes[0];
1165 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1166 item_offset = (unsigned long)iref - (unsigned long)ei;
1168 type = extent_ref_type(parent, owner);
1169 size = btrfs_extent_inline_ref_size(type);
1171 ret = btrfs_extend_item(trans, root, path, size);
1174 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1175 refs = btrfs_extent_refs(leaf, ei);
1176 refs += refs_to_add;
1177 btrfs_set_extent_refs(leaf, ei, refs);
1179 ptr = (unsigned long)ei + item_offset;
1180 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1181 if (ptr < end - size)
1182 memmove_extent_buffer(leaf, ptr + size, ptr,
1185 iref = (struct btrfs_extent_inline_ref *)ptr;
1186 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1187 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1188 struct btrfs_extent_data_ref *dref;
1189 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1190 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1191 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1192 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1193 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1194 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1195 struct btrfs_shared_data_ref *sref;
1196 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1197 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1198 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1199 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1200 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1202 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1204 btrfs_mark_buffer_dirty(leaf);
1208 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root,
1210 struct btrfs_path *path,
1211 struct btrfs_extent_inline_ref **ref_ret,
1212 u64 bytenr, u64 num_bytes, u64 parent,
1213 u64 root_objectid, u64 owner, u64 offset)
1217 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1218 bytenr, num_bytes, parent,
1219 root_objectid, owner, offset, 0);
1223 btrfs_release_path(root, path);
1226 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1227 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1230 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1231 root_objectid, owner, offset);
1236 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 struct btrfs_path *path,
1239 struct btrfs_extent_inline_ref *iref,
1242 struct extent_buffer *leaf;
1243 struct btrfs_extent_item *ei;
1244 struct btrfs_extent_data_ref *dref = NULL;
1245 struct btrfs_shared_data_ref *sref = NULL;
1254 leaf = path->nodes[0];
1255 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1256 refs = btrfs_extent_refs(leaf, ei);
1257 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1258 refs += refs_to_mod;
1259 btrfs_set_extent_refs(leaf, ei, refs);
1261 type = btrfs_extent_inline_ref_type(leaf, iref);
1263 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1264 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1265 refs = btrfs_extent_data_ref_count(leaf, dref);
1266 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1267 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1268 refs = btrfs_shared_data_ref_count(leaf, sref);
1271 BUG_ON(refs_to_mod != -1);
1274 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1275 refs += refs_to_mod;
1278 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1281 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1283 size = btrfs_extent_inline_ref_size(type);
1284 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1285 ptr = (unsigned long)iref;
1286 end = (unsigned long)ei + item_size;
1287 if (ptr + size < end)
1288 memmove_extent_buffer(leaf, ptr, ptr + size,
1291 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1294 btrfs_mark_buffer_dirty(leaf);
1298 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1299 struct btrfs_root *root,
1300 struct btrfs_path *path,
1301 u64 bytenr, u64 num_bytes, u64 parent,
1302 u64 root_objectid, u64 owner,
1303 u64 offset, int refs_to_add)
1305 struct btrfs_extent_inline_ref *iref;
1308 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1309 bytenr, num_bytes, parent,
1310 root_objectid, owner, offset, 1);
1312 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1313 ret = update_inline_extent_backref(trans, root, path, iref,
1315 } else if (ret == -ENOENT) {
1316 ret = setup_inline_extent_backref(trans, root, path, iref,
1317 parent, root_objectid,
1318 owner, offset, refs_to_add);
1323 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1324 struct btrfs_root *root,
1325 struct btrfs_path *path,
1326 u64 bytenr, u64 parent, u64 root_objectid,
1327 u64 owner, u64 offset, int refs_to_add)
1331 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1332 ret = insert_extent_data_ref(trans, root, path, bytenr,
1333 parent, root_objectid,
1334 owner, offset, refs_to_add);
1336 BUG_ON(refs_to_add != 1);
1337 ret = insert_tree_block_ref(trans, root, path, bytenr,
1338 parent, root_objectid);
1343 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *root,
1345 struct btrfs_path *path,
1346 struct btrfs_extent_inline_ref *iref,
1347 int refs_to_drop, int is_data)
1351 BUG_ON(!is_data && refs_to_drop != 1);
1353 ret = update_inline_extent_backref(trans, root, path, iref,
1355 } else if (is_data) {
1356 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1358 ret = btrfs_del_item(trans, root, path);
1363 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1364 struct btrfs_root *root,
1365 u64 bytenr, u64 num_bytes, u64 parent,
1366 u64 root_objectid, u64 owner, u64 offset)
1368 struct btrfs_path *path;
1369 struct extent_buffer *leaf;
1370 struct btrfs_extent_item *item;
1375 path = btrfs_alloc_path();
1380 path->leave_spinning = 1;
1382 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1383 path, bytenr, num_bytes, parent,
1384 root_objectid, owner, offset, 1);
1388 if (ret != -EAGAIN) {
1393 leaf = path->nodes[0];
1394 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1395 refs = btrfs_extent_refs(leaf, item);
1396 btrfs_set_extent_refs(leaf, item, refs + 1);
1398 btrfs_mark_buffer_dirty(leaf);
1399 btrfs_release_path(root->fs_info->extent_root, path);
1402 path->leave_spinning = 1;
1404 /* now insert the actual backref */
1405 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1406 path, bytenr, parent, root_objectid,
1411 btrfs_free_path(path);
1412 finish_current_insert(trans, root->fs_info->extent_root);
1413 del_pending_extents(trans, root->fs_info->extent_root);
1418 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1419 struct btrfs_root *root)
1421 finish_current_insert(trans, root->fs_info->extent_root);
1422 del_pending_extents(trans, root->fs_info->extent_root);
1426 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1427 struct btrfs_root *root, u64 bytenr,
1428 u64 num_bytes, u64 *refs, u64 *flags)
1430 struct btrfs_path *path;
1432 struct btrfs_key key;
1433 struct extent_buffer *l;
1434 struct btrfs_extent_item *item;
1439 WARN_ON(num_bytes < root->sectorsize);
1440 path = btrfs_alloc_path();
1442 key.objectid = bytenr;
1443 key.offset = num_bytes;
1444 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1445 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1450 btrfs_print_leaf(root, path->nodes[0]);
1451 printk("failed to find block number %Lu\n", bytenr);
1456 item_size = btrfs_item_size_nr(l, path->slots[0]);
1457 if (item_size >= sizeof(*item)) {
1458 item = btrfs_item_ptr(l, path->slots[0],
1459 struct btrfs_extent_item);
1460 num_refs = btrfs_extent_refs(l, item);
1461 extent_flags = btrfs_extent_flags(l, item);
1463 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1464 struct btrfs_extent_item_v0 *ei0;
1465 BUG_ON(item_size != sizeof(*ei0));
1466 ei0 = btrfs_item_ptr(l, path->slots[0],
1467 struct btrfs_extent_item_v0);
1468 num_refs = btrfs_extent_refs_v0(l, ei0);
1469 /* FIXME: this isn't correct for data */
1470 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1475 BUG_ON(num_refs == 0);
1476 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1480 *flags = extent_flags;
1482 btrfs_free_path(path);
1486 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1487 struct btrfs_root *root,
1488 u64 bytenr, u64 num_bytes, u64 flags)
1490 struct btrfs_path *path;
1492 struct btrfs_key key;
1493 struct extent_buffer *l;
1494 struct btrfs_extent_item *item;
1497 WARN_ON(num_bytes < root->sectorsize);
1498 path = btrfs_alloc_path();
1500 key.objectid = bytenr;
1501 key.offset = num_bytes;
1502 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1503 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1508 btrfs_print_leaf(root, path->nodes[0]);
1509 printk("failed to find block number %Lu\n",
1510 (unsigned long long)bytenr);
1514 item_size = btrfs_item_size_nr(l, path->slots[0]);
1515 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1516 if (item_size < sizeof(*item)) {
1517 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1523 item_size = btrfs_item_size_nr(l, path->slots[0]);
1526 BUG_ON(item_size < sizeof(*item));
1527 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1528 flags |= btrfs_extent_flags(l, item);
1529 btrfs_set_extent_flags(l, item, flags);
1531 btrfs_free_path(path);
1532 finish_current_insert(trans, root->fs_info->extent_root);
1533 del_pending_extents(trans, root->fs_info->extent_root);
1537 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1538 struct btrfs_root *root,
1539 struct extent_buffer *buf,
1540 int record_parent, int inc)
1547 struct btrfs_key key;
1548 struct btrfs_file_extent_item *fi;
1552 int (*process_func)(struct btrfs_trans_handle *trans,
1553 struct btrfs_root *root,
1554 u64, u64, u64, u64, u64, u64);
1556 ref_root = btrfs_header_owner(buf);
1557 nritems = btrfs_header_nritems(buf);
1558 level = btrfs_header_level(buf);
1560 if (!root->ref_cows && level == 0)
1564 process_func = btrfs_inc_extent_ref;
1566 process_func = btrfs_free_extent;
1569 parent = buf->start;
1573 for (i = 0; i < nritems; i++) {
1576 btrfs_item_key_to_cpu(buf, &key, i);
1577 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1579 fi = btrfs_item_ptr(buf, i,
1580 struct btrfs_file_extent_item);
1581 if (btrfs_file_extent_type(buf, fi) ==
1582 BTRFS_FILE_EXTENT_INLINE)
1584 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1588 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1589 key.offset -= btrfs_file_extent_offset(buf, fi);
1590 ret = process_func(trans, root, bytenr, num_bytes,
1591 parent, ref_root, key.objectid,
1598 bytenr = btrfs_node_blockptr(buf, i);
1599 num_bytes = btrfs_level_size(root, level - 1);
1600 ret = process_func(trans, root, bytenr, num_bytes,
1601 parent, ref_root, level - 1, 0);
1614 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1615 struct extent_buffer *buf, int record_parent)
1617 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1620 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1621 struct extent_buffer *buf, int record_parent)
1623 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1626 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root,
1628 struct btrfs_path *path,
1629 struct btrfs_block_group_cache *cache)
1633 struct btrfs_root *extent_root = root->fs_info->extent_root;
1635 struct extent_buffer *leaf;
1637 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1642 leaf = path->nodes[0];
1643 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1644 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1645 btrfs_mark_buffer_dirty(leaf);
1646 btrfs_release_path(extent_root, path);
1648 finish_current_insert(trans, extent_root);
1649 pending_ret = del_pending_extents(trans, extent_root);
1658 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root)
1661 struct extent_io_tree *block_group_cache;
1662 struct btrfs_block_group_cache *cache;
1664 struct btrfs_path *path;
1670 block_group_cache = &root->fs_info->block_group_cache;
1671 path = btrfs_alloc_path();
1676 ret = find_first_extent_bit(block_group_cache, last,
1677 &start, &end, BLOCK_GROUP_DIRTY);
1686 ret = get_state_private(block_group_cache, start, &ptr);
1689 clear_extent_bits(block_group_cache, start, end,
1690 BLOCK_GROUP_DIRTY, GFP_NOFS);
1692 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1693 ret = write_one_cache_group(trans, root, path, cache);
1696 btrfs_free_path(path);
1700 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1703 struct list_head *head = &info->space_info;
1704 struct list_head *cur;
1705 struct btrfs_space_info *found;
1706 list_for_each(cur, head) {
1707 found = list_entry(cur, struct btrfs_space_info, list);
1708 if (found->flags == flags)
1715 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1716 u64 total_bytes, u64 bytes_used,
1717 struct btrfs_space_info **space_info)
1719 struct btrfs_space_info *found;
1721 found = __find_space_info(info, flags);
1723 found->total_bytes += total_bytes;
1724 found->bytes_used += bytes_used;
1725 WARN_ON(found->total_bytes < found->bytes_used);
1726 *space_info = found;
1729 found = kmalloc(sizeof(*found), GFP_NOFS);
1733 list_add(&found->list, &info->space_info);
1734 found->flags = flags;
1735 found->total_bytes = total_bytes;
1736 found->bytes_used = bytes_used;
1737 found->bytes_pinned = 0;
1739 *space_info = found;
1744 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1746 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1747 BTRFS_BLOCK_GROUP_RAID1 |
1748 BTRFS_BLOCK_GROUP_RAID10 |
1749 BTRFS_BLOCK_GROUP_DUP);
1751 if (flags & BTRFS_BLOCK_GROUP_DATA)
1752 fs_info->avail_data_alloc_bits |= extra_flags;
1753 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1754 fs_info->avail_metadata_alloc_bits |= extra_flags;
1755 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1756 fs_info->avail_system_alloc_bits |= extra_flags;
1760 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1761 struct btrfs_root *extent_root, u64 alloc_bytes,
1764 struct btrfs_space_info *space_info;
1770 space_info = __find_space_info(extent_root->fs_info, flags);
1772 ret = update_space_info(extent_root->fs_info, flags,
1776 BUG_ON(!space_info);
1778 if (space_info->full)
1781 thresh = div_factor(space_info->total_bytes, 7);
1782 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1786 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1787 if (ret == -ENOSPC) {
1788 space_info->full = 1;
1794 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1795 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1800 static int update_block_group(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 u64 bytenr, u64 num_bytes, int alloc,
1805 struct btrfs_block_group_cache *cache;
1806 struct btrfs_fs_info *info = root->fs_info;
1807 u64 total = num_bytes;
1813 /* block accounting for super block */
1814 old_val = btrfs_super_bytes_used(&info->super_copy);
1816 old_val += num_bytes;
1818 old_val -= num_bytes;
1819 btrfs_set_super_bytes_used(&info->super_copy, old_val);
1821 /* block accounting for root item */
1822 old_val = btrfs_root_used(&root->root_item);
1824 old_val += num_bytes;
1826 old_val -= num_bytes;
1827 btrfs_set_root_used(&root->root_item, old_val);
1830 cache = btrfs_lookup_block_group(info, bytenr);
1834 byte_in_group = bytenr - cache->key.objectid;
1835 WARN_ON(byte_in_group > cache->key.offset);
1836 start = cache->key.objectid;
1837 end = start + cache->key.offset - 1;
1838 set_extent_bits(&info->block_group_cache, start, end,
1839 BLOCK_GROUP_DIRTY, GFP_NOFS);
1841 old_val = btrfs_block_group_used(&cache->item);
1842 num_bytes = min(total, cache->key.offset - byte_in_group);
1844 old_val += num_bytes;
1845 cache->space_info->bytes_used += num_bytes;
1847 old_val -= num_bytes;
1848 cache->space_info->bytes_used -= num_bytes;
1850 set_extent_dirty(&info->free_space_cache,
1851 bytenr, bytenr + num_bytes - 1,
1855 btrfs_set_block_group_used(&cache->item, old_val);
1857 bytenr += num_bytes;
1862 static int update_pinned_extents(struct btrfs_root *root,
1863 u64 bytenr, u64 num, int pin)
1866 struct btrfs_block_group_cache *cache;
1867 struct btrfs_fs_info *fs_info = root->fs_info;
1870 set_extent_dirty(&fs_info->pinned_extents,
1871 bytenr, bytenr + num - 1, GFP_NOFS);
1873 clear_extent_dirty(&fs_info->pinned_extents,
1874 bytenr, bytenr + num - 1, GFP_NOFS);
1877 cache = btrfs_lookup_block_group(fs_info, bytenr);
1879 len = min(num, cache->key.offset -
1880 (bytenr - cache->key.objectid));
1882 cache->pinned += len;
1883 cache->space_info->bytes_pinned += len;
1884 fs_info->total_pinned += len;
1886 cache->pinned -= len;
1887 cache->space_info->bytes_pinned -= len;
1888 fs_info->total_pinned -= len;
1896 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1901 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1905 ret = find_first_extent_bit(pinned_extents, last,
1906 &start, &end, EXTENT_DIRTY);
1909 set_extent_dirty(copy, start, end, GFP_NOFS);
1915 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1916 struct btrfs_root *root,
1917 struct extent_io_tree *unpin)
1922 struct extent_io_tree *free_space_cache;
1923 free_space_cache = &root->fs_info->free_space_cache;
1926 ret = find_first_extent_bit(unpin, 0, &start, &end,
1930 update_pinned_extents(root, start, end + 1 - start, 0);
1931 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1932 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1937 static int finish_current_insert(struct btrfs_trans_handle *trans,
1938 struct btrfs_root *extent_root)
1943 struct btrfs_fs_info *info = extent_root->fs_info;
1944 struct btrfs_path *path;
1945 struct pending_extent_op *extent_op;
1946 struct btrfs_key key;
1949 path = btrfs_alloc_path();
1952 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1953 &end, EXTENT_LOCKED);
1957 ret = get_state_private(&info->extent_ins, start, &priv);
1959 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1961 if (extent_op->type == PENDING_EXTENT_INSERT) {
1962 key.objectid = start;
1963 key.offset = end + 1 - start;
1964 key.type = BTRFS_EXTENT_ITEM_KEY;
1965 ret = alloc_reserved_tree_block(trans, extent_root,
1966 extent_root->root_key.objectid,
1970 extent_op->level, &key);
1975 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1979 btrfs_free_path(path);
1983 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1984 struct btrfs_root *root,
1985 u64 bytenr, u64 num_bytes, int is_data)
1988 struct extent_buffer *buf;
1993 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1997 /* we can reuse a block if it hasn't been written
1998 * and it is from this transaction. We can't
1999 * reuse anything from the tree log root because
2000 * it has tiny sub-transactions.
2002 if (btrfs_buffer_uptodate(buf, 0)) {
2003 u64 header_owner = btrfs_header_owner(buf);
2004 u64 header_transid = btrfs_header_generation(buf);
2005 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2006 header_transid == trans->transid &&
2007 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2008 clean_tree_block(NULL, root, buf);
2009 free_extent_buffer(buf);
2013 free_extent_buffer(buf);
2015 update_pinned_extents(root, bytenr, num_bytes, 1);
2022 * remove an extent from the root, returns 0 on success
2024 static int __free_extent(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 u64 bytenr, u64 num_bytes, u64 parent,
2027 u64 root_objectid, u64 owner_objectid,
2028 u64 owner_offset, int refs_to_drop)
2031 struct btrfs_key key;
2032 struct btrfs_path *path;
2033 struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2034 struct btrfs_root *extent_root = root->fs_info->extent_root;
2035 struct extent_buffer *leaf;
2036 struct btrfs_extent_item *ei;
2037 struct btrfs_extent_inline_ref *iref;
2040 int extent_slot = 0;
2041 int found_extent = 0;
2046 path = btrfs_alloc_path();
2051 path->leave_spinning = 1;
2053 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2054 BUG_ON(!is_data && refs_to_drop != 1);
2056 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2057 bytenr, num_bytes, parent,
2058 root_objectid, owner_objectid,
2061 extent_slot = path->slots[0];
2062 while (extent_slot >= 0) {
2063 btrfs_item_key_to_cpu(path->nodes[0], &key,
2065 if (key.objectid != bytenr)
2067 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2068 key.offset == num_bytes) {
2072 if (path->slots[0] - extent_slot > 5)
2076 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2077 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2078 if (found_extent && item_size < sizeof(*ei))
2081 if (!found_extent) {
2083 ret = remove_extent_backref(trans, extent_root, path,
2087 btrfs_release_path(extent_root, path);
2088 path->leave_spinning = 1;
2090 key.objectid = bytenr;
2091 key.type = BTRFS_EXTENT_ITEM_KEY;
2092 key.offset = num_bytes;
2094 ret = btrfs_search_slot(trans, extent_root,
2097 printk(KERN_ERR "umm, got %d back from search"
2098 ", was looking for %llu\n", ret,
2099 (unsigned long long)bytenr);
2100 btrfs_print_leaf(extent_root, path->nodes[0]);
2103 extent_slot = path->slots[0];
2106 btrfs_print_leaf(extent_root, path->nodes[0]);
2108 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2109 "parent %llu root %llu owner %llu offset %llu\n",
2110 (unsigned long long)bytenr,
2111 (unsigned long long)parent,
2112 (unsigned long long)root_objectid,
2113 (unsigned long long)owner_objectid,
2114 (unsigned long long)owner_offset);
2117 leaf = path->nodes[0];
2118 item_size = btrfs_item_size_nr(leaf, extent_slot);
2119 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2120 if (item_size < sizeof(*ei)) {
2121 BUG_ON(found_extent || extent_slot != path->slots[0]);
2122 ret = convert_extent_item_v0(trans, extent_root, path,
2126 btrfs_release_path(extent_root, path);
2127 path->leave_spinning = 1;
2129 key.objectid = bytenr;
2130 key.type = BTRFS_EXTENT_ITEM_KEY;
2131 key.offset = num_bytes;
2133 ret = btrfs_search_slot(trans, extent_root, &key, path,
2136 printk(KERN_ERR "umm, got %d back from search"
2137 ", was looking for %llu\n", ret,
2138 (unsigned long long)bytenr);
2139 btrfs_print_leaf(extent_root, path->nodes[0]);
2142 extent_slot = path->slots[0];
2143 leaf = path->nodes[0];
2144 item_size = btrfs_item_size_nr(leaf, extent_slot);
2147 BUG_ON(item_size < sizeof(*ei));
2148 ei = btrfs_item_ptr(leaf, extent_slot,
2149 struct btrfs_extent_item);
2150 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2151 struct btrfs_tree_block_info *bi;
2152 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2153 bi = (struct btrfs_tree_block_info *)(ei + 1);
2154 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2157 refs = btrfs_extent_refs(leaf, ei);
2158 BUG_ON(refs < refs_to_drop);
2159 refs -= refs_to_drop;
2163 * In the case of inline back ref, reference count will
2164 * be updated by remove_extent_backref
2167 BUG_ON(!found_extent);
2169 btrfs_set_extent_refs(leaf, ei, refs);
2170 btrfs_mark_buffer_dirty(leaf);
2173 ret = remove_extent_backref(trans, extent_root, path,
2183 BUG_ON(is_data && refs_to_drop !=
2184 extent_data_ref_count(root, path, iref));
2186 BUG_ON(path->slots[0] != extent_slot);
2188 BUG_ON(path->slots[0] != extent_slot + 1);
2189 path->slots[0] = extent_slot;
2194 if (ops && ops->free_extent) {
2195 ret = ops->free_extent(root, bytenr, num_bytes);
2203 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2210 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2213 btrfs_release_path(extent_root, path);
2216 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2220 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2224 btrfs_free_path(path);
2225 finish_current_insert(trans, extent_root);
2230 * find all the blocks marked as pending in the radix tree and remove
2231 * them from the extent map
2233 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2234 btrfs_root *extent_root)
2241 struct extent_io_tree *pending_del;
2242 struct extent_io_tree *extent_ins;
2243 struct pending_extent_op *extent_op;
2245 extent_ins = &extent_root->fs_info->extent_ins;
2246 pending_del = &extent_root->fs_info->pending_del;
2249 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2254 ret = get_state_private(pending_del, start, &priv);
2256 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2258 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2261 if (!test_range_bit(extent_ins, start, end,
2262 EXTENT_LOCKED, 0)) {
2263 ret = __free_extent(trans, extent_root,
2264 start, end + 1 - start, 0,
2265 extent_root->root_key.objectid,
2266 extent_op->level, 0, 1);
2270 ret = get_state_private(extent_ins, start, &priv);
2272 extent_op = (struct pending_extent_op *)
2273 (unsigned long)priv;
2275 clear_extent_bits(extent_ins, start, end,
2276 EXTENT_LOCKED, GFP_NOFS);
2278 if (extent_op->type == PENDING_BACKREF_UPDATE)
2290 * remove an extent from the root, returns 0 on success
2293 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2294 struct btrfs_root *root,
2295 u64 bytenr, u64 num_bytes, u64 parent,
2296 u64 root_objectid, u64 owner, u64 offset)
2298 struct btrfs_root *extent_root = root->fs_info->extent_root;
2302 WARN_ON(num_bytes < root->sectorsize);
2303 if (root == extent_root) {
2304 struct pending_extent_op *extent_op;
2306 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2309 extent_op->type = PENDING_EXTENT_DELETE;
2310 extent_op->bytenr = bytenr;
2311 extent_op->num_bytes = num_bytes;
2312 extent_op->level = (int)owner;
2314 set_extent_bits(&root->fs_info->pending_del,
2315 bytenr, bytenr + num_bytes - 1,
2316 EXTENT_LOCKED, GFP_NOFS);
2317 set_state_private(&root->fs_info->pending_del,
2318 bytenr, (unsigned long)extent_op);
2321 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2322 root_objectid, owner, offset, 1);
2323 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2324 return ret ? ret : pending_ret;
2327 static u64 stripe_align(struct btrfs_root *root, u64 val)
2329 u64 mask = ((u64)root->stripesize - 1);
2330 u64 ret = (val + mask) & ~mask;
2335 * walks the btree of allocated extents and find a hole of a given size.
2336 * The key ins is changed to record the hole:
2337 * ins->objectid == block start
2338 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2339 * ins->offset == number of blocks
2340 * Any available blocks before search_start are skipped.
2342 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2343 struct btrfs_root *orig_root,
2344 u64 num_bytes, u64 empty_size,
2345 u64 search_start, u64 search_end,
2346 u64 hint_byte, struct btrfs_key *ins,
2347 u64 exclude_start, u64 exclude_nr,
2351 u64 orig_search_start = search_start;
2352 struct btrfs_root * root = orig_root->fs_info->extent_root;
2353 struct btrfs_fs_info *info = root->fs_info;
2354 u64 total_needed = num_bytes;
2355 struct btrfs_block_group_cache *block_group;
2359 WARN_ON(num_bytes < root->sectorsize);
2360 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2363 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2365 hint_byte = search_start;
2366 block_group = btrfs_find_block_group(root, block_group,
2367 hint_byte, data, 1);
2369 block_group = btrfs_find_block_group(root,
2371 search_start, data, 1);
2374 total_needed += empty_size;
2378 block_group = btrfs_lookup_first_block_group(info,
2381 block_group = btrfs_lookup_first_block_group(info,
2384 ret = find_search_start(root, &block_group, &search_start,
2385 total_needed, data);
2389 search_start = stripe_align(root, search_start);
2390 ins->objectid = search_start;
2391 ins->offset = num_bytes;
2393 if (ins->objectid + num_bytes >
2394 block_group->key.objectid + block_group->key.offset) {
2395 search_start = block_group->key.objectid +
2396 block_group->key.offset;
2400 if (test_range_bit(&info->extent_ins, ins->objectid,
2401 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2402 search_start = ins->objectid + num_bytes;
2406 if (test_range_bit(&info->pinned_extents, ins->objectid,
2407 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2408 search_start = ins->objectid + num_bytes;
2412 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2413 ins->objectid < exclude_start + exclude_nr)) {
2414 search_start = exclude_start + exclude_nr;
2418 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2419 block_group = btrfs_lookup_block_group(info, ins->objectid);
2421 trans->block_group = block_group;
2423 ins->offset = num_bytes;
2427 block_group = btrfs_lookup_first_block_group(info, search_start);
2429 search_start = orig_search_start;
2436 total_needed -= empty_size;
2442 block_group = btrfs_find_block_group(root, block_group,
2443 search_start, data, 0);
2450 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2451 struct btrfs_root *root,
2452 u64 num_bytes, u64 empty_size,
2453 u64 hint_byte, u64 search_end,
2454 struct btrfs_key *ins, int data)
2457 u64 search_start = 0;
2459 struct btrfs_fs_info *info = root->fs_info;
2461 if (info->extent_ops) {
2462 struct btrfs_extent_ops *ops = info->extent_ops;
2463 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2469 alloc_profile = info->avail_data_alloc_bits &
2470 info->data_alloc_profile;
2471 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2472 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2473 info->system_allocs >= 0) {
2474 alloc_profile = info->avail_system_alloc_bits &
2475 info->system_alloc_profile;
2476 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2478 alloc_profile = info->avail_metadata_alloc_bits &
2479 info->metadata_alloc_profile;
2480 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2483 if (root->ref_cows) {
2484 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2485 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2487 BTRFS_BLOCK_GROUP_METADATA);
2490 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2491 num_bytes + 2 * 1024 * 1024, data);
2495 WARN_ON(num_bytes < root->sectorsize);
2496 ret = find_free_extent(trans, root, num_bytes, empty_size,
2497 search_start, search_end, hint_byte, ins,
2498 trans->alloc_exclude_start,
2499 trans->alloc_exclude_nr, data);
2502 clear_extent_dirty(&root->fs_info->free_space_cache,
2503 ins->objectid, ins->objectid + ins->offset - 1,
2508 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 u64 root_objectid, u64 generation,
2511 u64 flags, struct btrfs_disk_key *key,
2512 int level, struct btrfs_key *ins)
2515 struct btrfs_fs_info *fs_info = root->fs_info;
2516 struct btrfs_extent_item *extent_item;
2517 struct btrfs_tree_block_info *block_info;
2518 struct btrfs_extent_inline_ref *iref;
2519 struct btrfs_path *path;
2520 struct extent_buffer *leaf;
2521 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2523 path = btrfs_alloc_path();
2526 path->leave_spinning = 1;
2527 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2531 leaf = path->nodes[0];
2532 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2533 struct btrfs_extent_item);
2534 btrfs_set_extent_refs(leaf, extent_item, 1);
2535 btrfs_set_extent_generation(leaf, extent_item, generation);
2536 btrfs_set_extent_flags(leaf, extent_item,
2537 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2538 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2540 btrfs_set_tree_block_key(leaf, block_info, key);
2541 btrfs_set_tree_block_level(leaf, block_info, level);
2543 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2544 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2545 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2547 btrfs_mark_buffer_dirty(leaf);
2548 btrfs_free_path(path);
2550 ret = update_block_group(trans, root, ins->objectid, ins->offset,
2553 printk(KERN_ERR "btrfs update block group failed for %llu "
2554 "%llu\n", (unsigned long long)ins->objectid,
2555 (unsigned long long)ins->offset);
2561 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2562 struct btrfs_root *root, u64 num_bytes,
2563 u64 root_objectid, u64 generation,
2564 u64 flags, struct btrfs_disk_key *key,
2565 int level, u64 empty_size, u64 hint_byte,
2566 u64 search_end, struct btrfs_key *ins)
2569 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2570 hint_byte, search_end, ins, 0);
2573 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2574 struct pending_extent_op *extent_op;
2576 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2579 extent_op->type = PENDING_EXTENT_INSERT;
2580 extent_op->bytenr = ins->objectid;
2581 extent_op->num_bytes = ins->offset;
2582 extent_op->level = level;
2583 extent_op->flags = flags;
2584 memcpy(&extent_op->key, key, sizeof(*key));
2586 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2587 ins->objectid + ins->offset - 1,
2588 EXTENT_LOCKED, GFP_NOFS);
2589 set_state_private(&root->fs_info->extent_ins,
2590 ins->objectid, (unsigned long)extent_op);
2592 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2595 finish_current_insert(trans, root->fs_info->extent_root);
2596 del_pending_extents(trans, root->fs_info->extent_root);
2602 * helper function to allocate a block for a given tree
2603 * returns the tree buffer or NULL.
2605 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root,
2607 u32 blocksize, u64 root_objectid,
2608 struct btrfs_disk_key *key, int level,
2609 u64 hint, u64 empty_size)
2611 struct btrfs_key ins;
2613 struct extent_buffer *buf;
2615 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2616 trans->transid, 0, key, level,
2617 empty_size, hint, (u64)-1, &ins);
2620 return ERR_PTR(ret);
2623 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2625 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2626 0, root->root_key.objectid, level, 0);
2628 return ERR_PTR(-ENOMEM);
2630 btrfs_set_buffer_uptodate(buf);
2631 trans->blocks_used++;
2638 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2639 struct btrfs_root *root,
2640 struct extent_buffer *leaf)
2643 u64 leaf_generation;
2644 struct btrfs_key key;
2645 struct btrfs_file_extent_item *fi;
2650 BUG_ON(!btrfs_is_leaf(leaf));
2651 nritems = btrfs_header_nritems(leaf);
2652 leaf_owner = btrfs_header_owner(leaf);
2653 leaf_generation = btrfs_header_generation(leaf);
2655 for (i = 0; i < nritems; i++) {
2658 btrfs_item_key_to_cpu(leaf, &key, i);
2659 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2661 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2662 if (btrfs_file_extent_type(leaf, fi) ==
2663 BTRFS_FILE_EXTENT_INLINE)
2666 * FIXME make sure to insert a trans record that
2667 * repeats the snapshot del on crash
2669 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2670 if (disk_bytenr == 0)
2672 ret = btrfs_free_extent(trans, root, disk_bytenr,
2673 btrfs_file_extent_disk_num_bytes(leaf, fi),
2674 leaf->start, leaf_owner, leaf_generation,
2681 static void noinline reada_walk_down(struct btrfs_root *root,
2682 struct extent_buffer *node,
2695 nritems = btrfs_header_nritems(node);
2696 level = btrfs_header_level(node);
2700 for (i = slot; i < nritems && skipped < 32; i++) {
2701 bytenr = btrfs_node_blockptr(node, i);
2702 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2703 (last > bytenr && last - bytenr > 32 * 1024))) {
2707 blocksize = btrfs_level_size(root, level - 1);
2709 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2717 mutex_unlock(&root->fs_info->fs_mutex);
2718 ret = readahead_tree_block(root, bytenr, blocksize,
2719 btrfs_node_ptr_generation(node, i));
2720 last = bytenr + blocksize;
2722 mutex_lock(&root->fs_info->fs_mutex);
2729 * helper function for drop_snapshot, this walks down the tree dropping ref
2730 * counts as it goes.
2732 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2733 struct btrfs_root *root,
2734 struct btrfs_path *path, int *level)
2740 struct extent_buffer *next;
2741 struct extent_buffer *cur;
2742 struct extent_buffer *parent;
2747 WARN_ON(*level < 0);
2748 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2749 ret = btrfs_lookup_extent_ref(trans, root,
2750 path->nodes[*level]->start,
2751 path->nodes[*level]->len, &refs);
2757 * walk down to the last node level and free all the leaves
2759 while(*level >= 0) {
2760 WARN_ON(*level < 0);
2761 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2762 cur = path->nodes[*level];
2764 if (btrfs_header_level(cur) != *level)
2767 if (path->slots[*level] >=
2768 btrfs_header_nritems(cur))
2771 ret = drop_leaf_ref(trans, root, cur);
2775 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2776 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2777 blocksize = btrfs_level_size(root, *level - 1);
2778 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2782 parent = path->nodes[*level];
2783 root_owner = btrfs_header_owner(parent);
2784 root_gen = btrfs_header_generation(parent);
2785 path->slots[*level]++;
2786 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2787 parent->start, root_owner,
2788 root_gen, *level - 1, 1);
2792 next = btrfs_find_tree_block(root, bytenr, blocksize);
2793 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2794 free_extent_buffer(next);
2795 reada_walk_down(root, cur, path->slots[*level]);
2796 mutex_unlock(&root->fs_info->fs_mutex);
2797 next = read_tree_block(root, bytenr, blocksize,
2799 mutex_lock(&root->fs_info->fs_mutex);
2801 WARN_ON(*level <= 0);
2802 if (path->nodes[*level-1])
2803 free_extent_buffer(path->nodes[*level-1]);
2804 path->nodes[*level-1] = next;
2805 *level = btrfs_header_level(next);
2806 path->slots[*level] = 0;
2809 WARN_ON(*level < 0);
2810 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2812 if (path->nodes[*level] == root->node) {
2813 root_owner = root->root_key.objectid;
2814 parent = path->nodes[*level];
2816 parent = path->nodes[*level + 1];
2817 root_owner = btrfs_header_owner(parent);
2820 root_gen = btrfs_header_generation(parent);
2821 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2822 path->nodes[*level]->len, parent->start,
2823 root_owner, root_gen, *level, 1);
2824 free_extent_buffer(path->nodes[*level]);
2825 path->nodes[*level] = NULL;
2832 * helper for dropping snapshots. This walks back up the tree in the path
2833 * to find the first node higher up where we haven't yet gone through
2836 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2837 struct btrfs_root *root,
2838 struct btrfs_path *path, int *level)
2842 struct btrfs_root_item *root_item = &root->root_item;
2847 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2848 slot = path->slots[i];
2849 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2850 struct extent_buffer *node;
2851 struct btrfs_disk_key disk_key;
2852 node = path->nodes[i];
2855 WARN_ON(*level == 0);
2856 btrfs_node_key(node, &disk_key, path->slots[i]);
2857 memcpy(&root_item->drop_progress,
2858 &disk_key, sizeof(disk_key));
2859 root_item->drop_level = i;
2862 struct extent_buffer *parent;
2863 if (path->nodes[*level] == root->node)
2864 parent = path->nodes[*level];
2866 parent = path->nodes[*level + 1];
2868 root_owner = btrfs_header_owner(parent);
2869 root_gen = btrfs_header_generation(parent);
2870 ret = btrfs_free_extent(trans, root,
2871 path->nodes[*level]->start,
2872 path->nodes[*level]->len,
2873 parent->start, root_owner,
2874 root_gen, *level, 1);
2876 free_extent_buffer(path->nodes[*level]);
2877 path->nodes[*level] = NULL;
2885 * drop the reference count on the tree rooted at 'snap'. This traverses
2886 * the tree freeing any blocks that have a ref count of zero after being
2889 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2895 struct btrfs_path *path;
2898 struct btrfs_root_item *root_item = &root->root_item;
2900 path = btrfs_alloc_path();
2903 level = btrfs_header_level(root->node);
2905 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2906 path->nodes[level] = root->node;
2907 extent_buffer_get(root->node);
2908 path->slots[level] = 0;
2910 struct btrfs_key key;
2911 struct btrfs_disk_key found_key;
2912 struct extent_buffer *node;
2914 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2915 level = root_item->drop_level;
2916 path->lowest_level = level;
2917 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2922 node = path->nodes[level];
2923 btrfs_node_key(node, &found_key, path->slots[level]);
2924 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2925 sizeof(found_key)));
2928 wret = walk_down_tree(trans, root, path, &level);
2934 wret = walk_up_tree(trans, root, path, &level);
2944 for (i = 0; i <= orig_level; i++) {
2945 if (path->nodes[i]) {
2946 free_extent_buffer(path->nodes[i]);
2947 path->nodes[i] = NULL;
2951 btrfs_free_path(path);
2957 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2964 ret = find_first_extent_bit(&info->block_group_cache, 0,
2965 &start, &end, (unsigned int)-1);
2968 ret = get_state_private(&info->block_group_cache, start, &ptr);
2970 kfree((void *)(unsigned long)ptr);
2971 clear_extent_bits(&info->block_group_cache, start,
2972 end, (unsigned int)-1, GFP_NOFS);
2975 ret = find_first_extent_bit(&info->free_space_cache, 0,
2976 &start, &end, EXTENT_DIRTY);
2979 clear_extent_dirty(&info->free_space_cache, start,
2985 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2986 struct btrfs_key *key)
2989 struct btrfs_key found_key;
2990 struct extent_buffer *leaf;
2993 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2997 slot = path->slots[0];
2998 leaf = path->nodes[0];
2999 if (slot >= btrfs_header_nritems(leaf)) {
3000 ret = btrfs_next_leaf(root, path);
3007 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3009 if (found_key.objectid >= key->objectid &&
3010 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3019 int btrfs_read_block_groups(struct btrfs_root *root)
3021 struct btrfs_path *path;
3024 struct btrfs_block_group_cache *cache;
3025 struct btrfs_fs_info *info = root->fs_info;
3026 struct btrfs_space_info *space_info;
3027 struct extent_io_tree *block_group_cache;
3028 struct btrfs_key key;
3029 struct btrfs_key found_key;
3030 struct extent_buffer *leaf;
3032 block_group_cache = &info->block_group_cache;
3034 root = info->extent_root;
3037 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3038 path = btrfs_alloc_path();
3043 ret = find_first_block_group(root, path, &key);
3051 leaf = path->nodes[0];
3052 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3053 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3059 read_extent_buffer(leaf, &cache->item,
3060 btrfs_item_ptr_offset(leaf, path->slots[0]),
3061 sizeof(cache->item));
3062 memcpy(&cache->key, &found_key, sizeof(found_key));
3065 key.objectid = found_key.objectid + found_key.offset;
3066 btrfs_release_path(root, path);
3067 cache->flags = btrfs_block_group_flags(&cache->item);
3069 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3070 bit = BLOCK_GROUP_DATA;
3071 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3072 bit = BLOCK_GROUP_SYSTEM;
3073 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3074 bit = BLOCK_GROUP_METADATA;
3076 set_avail_alloc_bits(info, cache->flags);
3077 if (btrfs_chunk_readonly(root, cache->key.objectid))
3080 ret = update_space_info(info, cache->flags, found_key.offset,
3081 btrfs_block_group_used(&cache->item),
3084 cache->space_info = space_info;
3086 /* use EXTENT_LOCKED to prevent merging */
3087 set_extent_bits(block_group_cache, found_key.objectid,
3088 found_key.objectid + found_key.offset - 1,
3089 bit | EXTENT_LOCKED, GFP_NOFS);
3090 set_state_private(block_group_cache, found_key.objectid,
3091 (unsigned long)cache);
3095 btrfs_free_path(path);
3099 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3100 struct btrfs_root *root, u64 bytes_used,
3101 u64 type, u64 chunk_objectid, u64 chunk_offset,
3106 struct btrfs_root *extent_root;
3107 struct btrfs_block_group_cache *cache;
3108 struct extent_io_tree *block_group_cache;
3110 extent_root = root->fs_info->extent_root;
3111 block_group_cache = &root->fs_info->block_group_cache;
3113 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3115 cache->key.objectid = chunk_offset;
3116 cache->key.offset = size;
3118 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3119 btrfs_set_block_group_used(&cache->item, bytes_used);
3120 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3121 cache->flags = type;
3122 btrfs_set_block_group_flags(&cache->item, type);
3124 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3125 &cache->space_info);
3128 bit = block_group_state_bits(type);
3129 set_extent_bits(block_group_cache, chunk_offset,
3130 chunk_offset + size - 1,
3131 bit | EXTENT_LOCKED, GFP_NOFS);
3133 set_state_private(block_group_cache, chunk_offset,
3134 (unsigned long)cache);
3135 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3136 sizeof(cache->item));
3139 finish_current_insert(trans, extent_root);
3140 ret = del_pending_extents(trans, extent_root);
3142 set_avail_alloc_bits(extent_root->fs_info, type);
3147 * This is for converter use only.
3149 * In that case, we don't know where are free blocks located.
3150 * Therefore all block group cache entries must be setup properly
3151 * before doing any block allocation.
3153 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3154 struct btrfs_root *root)
3162 u64 total_metadata = 0;
3166 struct btrfs_root *extent_root;
3167 struct btrfs_block_group_cache *cache;
3168 struct extent_io_tree *block_group_cache;
3170 extent_root = root->fs_info->extent_root;
3171 block_group_cache = &root->fs_info->block_group_cache;
3172 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3173 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3174 group_align = 64 * root->sectorsize;
3177 while (cur_start < total_bytes) {
3178 group_size = total_bytes / 12;
3179 group_size = min_t(u64, group_size, total_bytes - cur_start);
3180 if (cur_start == 0) {
3181 bit = BLOCK_GROUP_SYSTEM;
3182 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3184 group_size &= ~(group_align - 1);
3185 group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3186 group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3188 group_size &= ~(group_align - 1);
3189 if (total_data >= total_metadata * 2) {
3190 group_type = BTRFS_BLOCK_GROUP_METADATA;
3191 group_size = min_t(u64, group_size,
3192 1ULL * 1024 * 1024 * 1024);
3193 total_metadata += group_size;
3195 group_type = BTRFS_BLOCK_GROUP_DATA;
3196 group_size = min_t(u64, group_size,
3197 5ULL * 1024 * 1024 * 1024);
3198 total_data += group_size;
3200 if ((total_bytes - cur_start) * 4 < group_size * 5)
3201 group_size = total_bytes - cur_start;
3204 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3207 cache->key.objectid = cur_start;
3208 cache->key.offset = group_size;
3209 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3211 btrfs_set_block_group_used(&cache->item, 0);
3212 btrfs_set_block_group_chunk_objectid(&cache->item,
3214 btrfs_set_block_group_flags(&cache->item, group_type);
3216 cache->flags = group_type;
3218 ret = update_space_info(root->fs_info, group_type, group_size,
3219 0, &cache->space_info);
3221 set_avail_alloc_bits(extent_root->fs_info, group_type);
3223 set_extent_bits(block_group_cache, cur_start,
3224 cur_start + group_size - 1,
3225 bit | EXTENT_LOCKED, GFP_NOFS);
3226 set_state_private(block_group_cache, cur_start,
3227 (unsigned long)cache);
3228 cur_start += group_size;
3230 /* then insert all the items */
3232 while(cur_start < total_bytes) {
3233 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3236 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3237 sizeof(cache->item));
3240 finish_current_insert(trans, extent_root);
3241 ret = del_pending_extents(trans, extent_root);
3244 cur_start = cache->key.objectid + cache->key.offset;
3249 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3250 struct btrfs_root *root,
3251 u64 bytenr, u64 num_bytes, int alloc,
3254 return update_block_group(trans, root, bytenr, num_bytes,