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.
22 #include "kerncompat.h"
23 #include "radix-tree.h"
26 #include "print-tree.h"
27 #include "transaction.h"
30 #include "free-space-cache.h"
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
38 struct pending_extent_op {
43 struct btrfs_disk_key key;
47 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
48 struct btrfs_root *root,
49 u64 root_objectid, u64 generation,
50 u64 flags, struct btrfs_disk_key *key,
51 int level, struct btrfs_key *ins);
52 static int __free_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *root,
54 u64 bytenr, u64 num_bytes, u64 parent,
55 u64 root_objectid, u64 owner_objectid,
56 u64 owner_offset, int refs_to_drop);
57 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
58 btrfs_root *extent_root);
59 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
60 btrfs_root *extent_root);
61 static struct btrfs_block_group_cache *
62 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
63 *hint, u64 search_start, int data, int owner);
65 static int remove_sb_from_cache(struct btrfs_root *root,
66 struct btrfs_block_group_cache *cache)
72 struct extent_io_tree *free_space_cache;
74 free_space_cache = &root->fs_info->free_space_cache;
75 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
76 bytenr = btrfs_sb_offset(i);
77 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
78 cache->key.objectid, bytenr, 0,
79 &logical, &nr, &stripe_len);
82 clear_extent_dirty(free_space_cache, logical[nr],
83 logical[nr] + stripe_len - 1, GFP_NOFS);
90 static int cache_block_group(struct btrfs_root *root,
91 struct btrfs_block_group_cache *block_group)
93 struct btrfs_path *path;
96 struct extent_buffer *leaf;
97 struct extent_io_tree *free_space_cache;
105 root = root->fs_info->extent_root;
106 free_space_cache = &root->fs_info->free_space_cache;
108 if (block_group->cached)
111 path = btrfs_alloc_path();
116 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
126 leaf = path->nodes[0];
127 slot = path->slots[0];
128 if (slot >= btrfs_header_nritems(leaf)) {
129 ret = btrfs_next_leaf(root, path);
138 btrfs_item_key_to_cpu(leaf, &key, slot);
139 if (key.objectid < block_group->key.objectid) {
142 if (key.objectid >= block_group->key.objectid +
143 block_group->key.offset) {
147 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
148 key.type == BTRFS_METADATA_ITEM_KEY) {
149 if (key.objectid > last) {
150 hole_size = key.objectid - last;
151 set_extent_dirty(free_space_cache, last,
152 last + hole_size - 1,
155 if (key.type == BTRFS_METADATA_ITEM_KEY)
156 last = key.objectid + root->leafsize;
158 last = key.objectid + key.offset;
164 if (block_group->key.objectid +
165 block_group->key.offset > last) {
166 hole_size = block_group->key.objectid +
167 block_group->key.offset - last;
168 set_extent_dirty(free_space_cache, last,
169 last + hole_size - 1, GFP_NOFS);
171 remove_sb_from_cache(root, block_group);
172 block_group->cached = 1;
174 btrfs_free_path(path);
178 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
182 struct extent_io_tree *block_group_cache;
183 struct btrfs_block_group_cache *block_group = NULL;
189 bytenr = max_t(u64, bytenr,
190 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
191 block_group_cache = &info->block_group_cache;
192 ret = find_first_extent_bit(block_group_cache,
193 bytenr, &start, &end,
194 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
199 ret = get_state_private(block_group_cache, start, &ptr);
203 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
207 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
211 struct extent_io_tree *block_group_cache;
212 struct btrfs_block_group_cache *block_group = NULL;
218 block_group_cache = &info->block_group_cache;
219 ret = find_first_extent_bit(block_group_cache,
220 bytenr, &start, &end,
221 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
226 ret = get_state_private(block_group_cache, start, &ptr);
230 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
231 if (block_group->key.objectid <= bytenr && bytenr <
232 block_group->key.objectid + block_group->key.offset)
237 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
239 return (cache->flags & bits) == bits;
242 static int noinline find_search_start(struct btrfs_root *root,
243 struct btrfs_block_group_cache **cache_ret,
244 u64 *start_ret, int num, int data)
247 struct btrfs_block_group_cache *cache = *cache_ret;
248 u64 last = *start_ret;
251 u64 search_start = *start_ret;
257 ret = cache_block_group(root, cache);
261 last = max(search_start, cache->key.objectid);
262 if (cache->ro || !block_group_bits(cache, data))
266 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
267 last, &start, &end, EXTENT_DIRTY);
272 start = max(last, start);
274 if (last - start < num) {
277 if (start + num > cache->key.objectid + cache->key.offset) {
285 cache = btrfs_lookup_block_group(root->fs_info, search_start);
287 printk("Unable to find block group for %llu\n",
288 (unsigned long long)search_start);
294 last = cache->key.objectid + cache->key.offset;
296 cache = btrfs_lookup_first_block_group(root->fs_info, last);
309 static int block_group_state_bits(u64 flags)
312 if (flags & BTRFS_BLOCK_GROUP_DATA)
313 bits |= BLOCK_GROUP_DATA;
314 if (flags & BTRFS_BLOCK_GROUP_METADATA)
315 bits |= BLOCK_GROUP_METADATA;
316 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
317 bits |= BLOCK_GROUP_SYSTEM;
321 static struct btrfs_block_group_cache *
322 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
323 *hint, u64 search_start, int data, int owner)
325 struct btrfs_block_group_cache *cache;
326 struct extent_io_tree *block_group_cache;
327 struct btrfs_block_group_cache *found_group = NULL;
328 struct btrfs_fs_info *info = root->fs_info;
341 block_group_cache = &info->block_group_cache;
346 bit = block_group_state_bits(data);
349 struct btrfs_block_group_cache *shint;
350 shint = btrfs_lookup_block_group(info, search_start);
351 if (shint && !shint->ro && block_group_bits(shint, data)) {
352 used = btrfs_block_group_used(&shint->item);
353 if (used + shint->pinned <
354 div_factor(shint->key.offset, factor)) {
359 if (hint && !hint->ro && block_group_bits(hint, data)) {
360 used = btrfs_block_group_used(&hint->item);
361 if (used + hint->pinned <
362 div_factor(hint->key.offset, factor)) {
365 last = hint->key.objectid + hint->key.offset;
369 hint_last = max(hint->key.objectid, search_start);
371 hint_last = search_start;
377 ret = find_first_extent_bit(block_group_cache, last,
382 ret = get_state_private(block_group_cache, start, &ptr);
386 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
387 last = cache->key.objectid + cache->key.offset;
388 used = btrfs_block_group_used(&cache->item);
390 if (!cache->ro && block_group_bits(cache, data)) {
392 free_check = cache->key.offset;
394 free_check = div_factor(cache->key.offset,
397 if (used + cache->pinned < free_check) {
414 * Back reference rules. Back refs have three main goals:
416 * 1) differentiate between all holders of references to an extent so that
417 * when a reference is dropped we can make sure it was a valid reference
418 * before freeing the extent.
420 * 2) Provide enough information to quickly find the holders of an extent
421 * if we notice a given block is corrupted or bad.
423 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
424 * maintenance. This is actually the same as #2, but with a slightly
425 * different use case.
427 * There are two kinds of back refs. The implicit back refs is optimized
428 * for pointers in non-shared tree blocks. For a given pointer in a block,
429 * back refs of this kind provide information about the block's owner tree
430 * and the pointer's key. These information allow us to find the block by
431 * b-tree searching. The full back refs is for pointers in tree blocks not
432 * referenced by their owner trees. The location of tree block is recorded
433 * in the back refs. Actually the full back refs is generic, and can be
434 * used in all cases the implicit back refs is used. The major shortcoming
435 * of the full back refs is its overhead. Every time a tree block gets
436 * COWed, we have to update back refs entry for all pointers in it.
438 * For a newly allocated tree block, we use implicit back refs for
439 * pointers in it. This means most tree related operations only involve
440 * implicit back refs. For a tree block created in old transaction, the
441 * only way to drop a reference to it is COW it. So we can detect the
442 * event that tree block loses its owner tree's reference and do the
443 * back refs conversion.
445 * When a tree block is COW'd through a tree, there are four cases:
447 * The reference count of the block is one and the tree is the block's
448 * owner tree. Nothing to do in this case.
450 * The reference count of the block is one and the tree is not the
451 * block's owner tree. In this case, full back refs is used for pointers
452 * in the block. Remove these full back refs, add implicit back refs for
453 * every pointers in the new block.
455 * The reference count of the block is greater than one and the tree is
456 * the block's owner tree. In this case, implicit back refs is used for
457 * pointers in the block. Add full back refs for every pointers in the
458 * block, increase lower level extents' reference counts. The original
459 * implicit back refs are entailed to the new block.
461 * The reference count of the block is greater than one and the tree is
462 * not the block's owner tree. Add implicit back refs for every pointer in
463 * the new block, increase lower level extents' reference count.
465 * Back Reference Key composing:
467 * The key objectid corresponds to the first byte in the extent,
468 * The key type is used to differentiate between types of back refs.
469 * There are different meanings of the key offset for different types
472 * File extents can be referenced by:
474 * - multiple snapshots, subvolumes, or different generations in one subvol
475 * - different files inside a single subvolume
476 * - different offsets inside a file (bookend extents in file.c)
478 * The extent ref structure for the implicit back refs has fields for:
480 * - Objectid of the subvolume root
481 * - objectid of the file holding the reference
482 * - original offset in the file
483 * - how many bookend extents
485 * The key offset for the implicit back refs is hash of the first
488 * The extent ref structure for the full back refs has field for:
490 * - number of pointers in the tree leaf
492 * The key offset for the implicit back refs is the first byte of
495 * When a file extent is allocated, The implicit back refs is used.
496 * the fields are filled in:
498 * (root_key.objectid, inode objectid, offset in file, 1)
500 * When a file extent is removed file truncation, we find the
501 * corresponding implicit back refs and check the following fields:
503 * (btrfs_header_owner(leaf), inode objectid, offset in file)
505 * Btree extents can be referenced by:
507 * - Different subvolumes
509 * Both the implicit back refs and the full back refs for tree blocks
510 * only consist of key. The key offset for the implicit back refs is
511 * objectid of block's owner tree. The key offset for the full back refs
512 * is the first byte of parent block.
514 * When implicit back refs is used, information about the lowest key and
515 * level of the tree block are required. These information are stored in
516 * tree block info structure.
519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
520 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
521 struct btrfs_root *root,
522 struct btrfs_path *path,
523 u64 owner, u32 extra_size)
525 struct btrfs_extent_item *item;
526 struct btrfs_extent_item_v0 *ei0;
527 struct btrfs_extent_ref_v0 *ref0;
528 struct btrfs_tree_block_info *bi;
529 struct extent_buffer *leaf;
530 struct btrfs_key key;
531 struct btrfs_key found_key;
532 u32 new_size = sizeof(*item);
536 leaf = path->nodes[0];
537 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
539 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
540 ei0 = btrfs_item_ptr(leaf, path->slots[0],
541 struct btrfs_extent_item_v0);
542 refs = btrfs_extent_refs_v0(leaf, ei0);
544 if (owner == (u64)-1) {
546 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
547 ret = btrfs_next_leaf(root, path);
551 leaf = path->nodes[0];
553 btrfs_item_key_to_cpu(leaf, &found_key,
555 BUG_ON(key.objectid != found_key.objectid);
556 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
560 ref0 = btrfs_item_ptr(leaf, path->slots[0],
561 struct btrfs_extent_ref_v0);
562 owner = btrfs_ref_objectid_v0(leaf, ref0);
566 btrfs_release_path(path);
568 if (owner < BTRFS_FIRST_FREE_OBJECTID)
569 new_size += sizeof(*bi);
571 new_size -= sizeof(*ei0);
572 ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
577 ret = btrfs_extend_item(trans, root, path, new_size);
580 leaf = path->nodes[0];
581 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
582 btrfs_set_extent_refs(leaf, item, refs);
583 /* FIXME: get real generation */
584 btrfs_set_extent_generation(leaf, item, 0);
585 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
586 btrfs_set_extent_flags(leaf, item,
587 BTRFS_EXTENT_FLAG_TREE_BLOCK |
588 BTRFS_BLOCK_FLAG_FULL_BACKREF);
589 bi = (struct btrfs_tree_block_info *)(item + 1);
590 /* FIXME: get first key of the block */
591 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
592 btrfs_set_tree_block_level(leaf, bi, (int)owner);
594 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
596 btrfs_mark_buffer_dirty(leaf);
601 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
603 u32 high_crc = ~(u32)0;
604 u32 low_crc = ~(u32)0;
607 lenum = cpu_to_le64(root_objectid);
608 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
609 lenum = cpu_to_le64(owner);
610 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
611 lenum = cpu_to_le64(offset);
612 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
614 return ((u64)high_crc << 31) ^ (u64)low_crc;
617 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
618 struct btrfs_extent_data_ref *ref)
620 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
621 btrfs_extent_data_ref_objectid(leaf, ref),
622 btrfs_extent_data_ref_offset(leaf, ref));
625 static int match_extent_data_ref(struct extent_buffer *leaf,
626 struct btrfs_extent_data_ref *ref,
627 u64 root_objectid, u64 owner, u64 offset)
629 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
630 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
631 btrfs_extent_data_ref_offset(leaf, ref) != offset)
636 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
637 struct btrfs_root *root,
638 struct btrfs_path *path,
639 u64 bytenr, u64 parent,
641 u64 owner, u64 offset)
643 struct btrfs_key key;
644 struct btrfs_extent_data_ref *ref;
645 struct extent_buffer *leaf;
651 key.objectid = bytenr;
653 key.type = BTRFS_SHARED_DATA_REF_KEY;
656 key.type = BTRFS_EXTENT_DATA_REF_KEY;
657 key.offset = hash_extent_data_ref(root_objectid,
662 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
671 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
672 key.type = BTRFS_EXTENT_REF_V0_KEY;
673 btrfs_release_path(path);
674 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
685 leaf = path->nodes[0];
686 nritems = btrfs_header_nritems(leaf);
688 if (path->slots[0] >= nritems) {
689 ret = btrfs_next_leaf(root, path);
695 leaf = path->nodes[0];
696 nritems = btrfs_header_nritems(leaf);
700 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
701 if (key.objectid != bytenr ||
702 key.type != BTRFS_EXTENT_DATA_REF_KEY)
705 ref = btrfs_item_ptr(leaf, path->slots[0],
706 struct btrfs_extent_data_ref);
708 if (match_extent_data_ref(leaf, ref, root_objectid,
711 btrfs_release_path(path);
723 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
724 struct btrfs_root *root,
725 struct btrfs_path *path,
726 u64 bytenr, u64 parent,
727 u64 root_objectid, u64 owner,
728 u64 offset, int refs_to_add)
730 struct btrfs_key key;
731 struct extent_buffer *leaf;
736 key.objectid = bytenr;
738 key.type = BTRFS_SHARED_DATA_REF_KEY;
740 size = sizeof(struct btrfs_shared_data_ref);
742 key.type = BTRFS_EXTENT_DATA_REF_KEY;
743 key.offset = hash_extent_data_ref(root_objectid,
745 size = sizeof(struct btrfs_extent_data_ref);
748 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
749 if (ret && ret != -EEXIST)
752 leaf = path->nodes[0];
754 struct btrfs_shared_data_ref *ref;
755 ref = btrfs_item_ptr(leaf, path->slots[0],
756 struct btrfs_shared_data_ref);
758 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
760 num_refs = btrfs_shared_data_ref_count(leaf, ref);
761 num_refs += refs_to_add;
762 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
765 struct btrfs_extent_data_ref *ref;
766 while (ret == -EEXIST) {
767 ref = btrfs_item_ptr(leaf, path->slots[0],
768 struct btrfs_extent_data_ref);
769 if (match_extent_data_ref(leaf, ref, root_objectid,
772 btrfs_release_path(path);
775 ret = btrfs_insert_empty_item(trans, root, path, &key,
777 if (ret && ret != -EEXIST)
780 leaf = path->nodes[0];
782 ref = btrfs_item_ptr(leaf, path->slots[0],
783 struct btrfs_extent_data_ref);
785 btrfs_set_extent_data_ref_root(leaf, ref,
787 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
788 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
789 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
791 num_refs = btrfs_extent_data_ref_count(leaf, ref);
792 num_refs += refs_to_add;
793 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
796 btrfs_mark_buffer_dirty(leaf);
799 btrfs_release_path(path);
803 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
804 struct btrfs_root *root,
805 struct btrfs_path *path,
808 struct btrfs_key key;
809 struct btrfs_extent_data_ref *ref1 = NULL;
810 struct btrfs_shared_data_ref *ref2 = NULL;
811 struct extent_buffer *leaf;
815 leaf = path->nodes[0];
816 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
818 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
819 ref1 = btrfs_item_ptr(leaf, path->slots[0],
820 struct btrfs_extent_data_ref);
821 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
822 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
823 ref2 = btrfs_item_ptr(leaf, path->slots[0],
824 struct btrfs_shared_data_ref);
825 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
826 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
827 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
828 struct btrfs_extent_ref_v0 *ref0;
829 ref0 = btrfs_item_ptr(leaf, path->slots[0],
830 struct btrfs_extent_ref_v0);
831 num_refs = btrfs_ref_count_v0(leaf, ref0);
837 BUG_ON(num_refs < refs_to_drop);
838 num_refs -= refs_to_drop;
841 ret = btrfs_del_item(trans, root, path);
843 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
844 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
845 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
846 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
847 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
849 struct btrfs_extent_ref_v0 *ref0;
850 ref0 = btrfs_item_ptr(leaf, path->slots[0],
851 struct btrfs_extent_ref_v0);
852 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
855 btrfs_mark_buffer_dirty(leaf);
860 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
861 struct btrfs_path *path,
862 struct btrfs_extent_inline_ref *iref)
864 struct btrfs_key key;
865 struct extent_buffer *leaf;
866 struct btrfs_extent_data_ref *ref1;
867 struct btrfs_shared_data_ref *ref2;
870 leaf = path->nodes[0];
871 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
873 if (btrfs_extent_inline_ref_type(leaf, iref) ==
874 BTRFS_EXTENT_DATA_REF_KEY) {
875 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
876 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
878 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
879 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
881 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
882 ref1 = btrfs_item_ptr(leaf, path->slots[0],
883 struct btrfs_extent_data_ref);
884 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
885 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
886 ref2 = btrfs_item_ptr(leaf, path->slots[0],
887 struct btrfs_shared_data_ref);
888 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
889 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
890 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
891 struct btrfs_extent_ref_v0 *ref0;
892 ref0 = btrfs_item_ptr(leaf, path->slots[0],
893 struct btrfs_extent_ref_v0);
894 num_refs = btrfs_ref_count_v0(leaf, ref0);
902 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
903 struct btrfs_root *root,
904 struct btrfs_path *path,
905 u64 bytenr, u64 parent,
908 struct btrfs_key key;
911 key.objectid = bytenr;
913 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
916 key.type = BTRFS_TREE_BLOCK_REF_KEY;
917 key.offset = root_objectid;
920 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
923 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
924 if (ret == -ENOENT && parent) {
925 btrfs_release_path(path);
926 key.type = BTRFS_EXTENT_REF_V0_KEY;
927 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
935 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
936 struct btrfs_root *root,
937 struct btrfs_path *path,
938 u64 bytenr, u64 parent,
941 struct btrfs_key key;
944 key.objectid = bytenr;
946 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
949 key.type = BTRFS_TREE_BLOCK_REF_KEY;
950 key.offset = root_objectid;
953 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
955 btrfs_release_path(path);
959 static inline int extent_ref_type(u64 parent, u64 owner)
962 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
964 type = BTRFS_SHARED_BLOCK_REF_KEY;
966 type = BTRFS_TREE_BLOCK_REF_KEY;
969 type = BTRFS_SHARED_DATA_REF_KEY;
971 type = BTRFS_EXTENT_DATA_REF_KEY;
976 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
977 struct btrfs_root *root,
978 struct btrfs_path *path,
979 struct btrfs_extent_inline_ref **ref_ret,
980 u64 bytenr, u64 num_bytes,
981 u64 parent, u64 root_objectid,
982 u64 owner, u64 offset, int insert)
984 struct btrfs_key key;
985 struct extent_buffer *leaf;
986 struct btrfs_extent_item *ei;
987 struct btrfs_extent_inline_ref *iref;
997 int skinny_metadata =
998 btrfs_fs_incompat(root->fs_info,
999 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1001 key.objectid = bytenr;
1002 key.type = BTRFS_EXTENT_ITEM_KEY;
1003 key.offset = num_bytes;
1005 want = extent_ref_type(parent, owner);
1007 extra_size = btrfs_extent_inline_ref_size(want);
1011 if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1012 skinny_metadata = 1;
1013 key.type = BTRFS_METADATA_ITEM_KEY;
1015 } else if (skinny_metadata) {
1016 skinny_metadata = 0;
1020 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1027 * We may be a newly converted file system which still has the old fat
1028 * extent entries for metadata, so try and see if we have one of those.
1030 if (ret > 0 && skinny_metadata) {
1031 skinny_metadata = 0;
1032 if (path->slots[0]) {
1034 btrfs_item_key_to_cpu(path->nodes[0], &key,
1036 if (key.objectid == bytenr &&
1037 key.type == BTRFS_EXTENT_ITEM_KEY &&
1038 key.offset == num_bytes)
1042 key.type = BTRFS_EXTENT_ITEM_KEY;
1043 key.offset = num_bytes;
1044 btrfs_release_path(path);
1050 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1056 leaf = path->nodes[0];
1057 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1058 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1059 if (item_size < sizeof(*ei)) {
1064 ret = convert_extent_item_v0(trans, root, path, owner,
1070 leaf = path->nodes[0];
1071 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1074 if (item_size < sizeof(*ei)) {
1075 printf("Size is %u, needs to be %u, slot %d\n",
1076 (unsigned)item_size,
1077 (unsigned)sizeof(*ei), path->slots[0]);
1078 btrfs_print_leaf(root, leaf);
1081 BUG_ON(item_size < sizeof(*ei));
1083 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1084 flags = btrfs_extent_flags(leaf, ei);
1086 ptr = (unsigned long)(ei + 1);
1087 end = (unsigned long)ei + item_size;
1089 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1090 ptr += sizeof(struct btrfs_tree_block_info);
1092 } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1093 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1104 iref = (struct btrfs_extent_inline_ref *)ptr;
1105 type = btrfs_extent_inline_ref_type(leaf, iref);
1109 ptr += btrfs_extent_inline_ref_size(type);
1113 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1114 struct btrfs_extent_data_ref *dref;
1115 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1116 if (match_extent_data_ref(leaf, dref, root_objectid,
1121 if (hash_extent_data_ref_item(leaf, dref) <
1122 hash_extent_data_ref(root_objectid, owner, offset))
1126 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1128 if (parent == ref_offset) {
1132 if (ref_offset < parent)
1135 if (root_objectid == ref_offset) {
1139 if (ref_offset < root_objectid)
1143 ptr += btrfs_extent_inline_ref_size(type);
1145 if (err == -ENOENT && insert) {
1146 if (item_size + extra_size >=
1147 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1152 * To add new inline back ref, we have to make sure
1153 * there is no corresponding back ref item.
1154 * For simplicity, we just do not add new inline back
1155 * ref if there is any back ref item.
1157 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1158 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1163 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1168 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1169 struct btrfs_root *root,
1170 struct btrfs_path *path,
1171 struct btrfs_extent_inline_ref *iref,
1172 u64 parent, u64 root_objectid,
1173 u64 owner, u64 offset, int refs_to_add)
1175 struct extent_buffer *leaf;
1176 struct btrfs_extent_item *ei;
1179 unsigned long item_offset;
1185 leaf = path->nodes[0];
1186 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1187 item_offset = (unsigned long)iref - (unsigned long)ei;
1189 type = extent_ref_type(parent, owner);
1190 size = btrfs_extent_inline_ref_size(type);
1192 ret = btrfs_extend_item(trans, root, path, size);
1195 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1196 refs = btrfs_extent_refs(leaf, ei);
1197 refs += refs_to_add;
1198 btrfs_set_extent_refs(leaf, ei, refs);
1200 ptr = (unsigned long)ei + item_offset;
1201 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1202 if (ptr < end - size)
1203 memmove_extent_buffer(leaf, ptr + size, ptr,
1206 iref = (struct btrfs_extent_inline_ref *)ptr;
1207 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1208 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1209 struct btrfs_extent_data_ref *dref;
1210 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1211 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1212 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1213 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1214 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1215 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1216 struct btrfs_shared_data_ref *sref;
1217 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1218 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1219 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1220 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1221 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1223 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1225 btrfs_mark_buffer_dirty(leaf);
1229 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1230 struct btrfs_root *root,
1231 struct btrfs_path *path,
1232 struct btrfs_extent_inline_ref **ref_ret,
1233 u64 bytenr, u64 num_bytes, u64 parent,
1234 u64 root_objectid, u64 owner, u64 offset)
1238 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1239 bytenr, num_bytes, parent,
1240 root_objectid, owner, offset, 0);
1244 btrfs_release_path(path);
1247 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1248 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1251 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1252 root_objectid, owner, offset);
1257 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1258 struct btrfs_root *root,
1259 struct btrfs_path *path,
1260 struct btrfs_extent_inline_ref *iref,
1263 struct extent_buffer *leaf;
1264 struct btrfs_extent_item *ei;
1265 struct btrfs_extent_data_ref *dref = NULL;
1266 struct btrfs_shared_data_ref *sref = NULL;
1275 leaf = path->nodes[0];
1276 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1277 refs = btrfs_extent_refs(leaf, ei);
1278 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1279 refs += refs_to_mod;
1280 btrfs_set_extent_refs(leaf, ei, refs);
1282 type = btrfs_extent_inline_ref_type(leaf, iref);
1284 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1285 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1286 refs = btrfs_extent_data_ref_count(leaf, dref);
1287 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1288 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1289 refs = btrfs_shared_data_ref_count(leaf, sref);
1292 BUG_ON(refs_to_mod != -1);
1295 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1296 refs += refs_to_mod;
1299 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1300 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1302 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1304 size = btrfs_extent_inline_ref_size(type);
1305 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1306 ptr = (unsigned long)iref;
1307 end = (unsigned long)ei + item_size;
1308 if (ptr + size < end)
1309 memmove_extent_buffer(leaf, ptr, ptr + size,
1312 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1315 btrfs_mark_buffer_dirty(leaf);
1319 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1320 struct btrfs_root *root,
1321 struct btrfs_path *path,
1322 u64 bytenr, u64 num_bytes, u64 parent,
1323 u64 root_objectid, u64 owner,
1324 u64 offset, int refs_to_add)
1326 struct btrfs_extent_inline_ref *iref;
1329 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1330 bytenr, num_bytes, parent,
1331 root_objectid, owner, offset, 1);
1333 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1334 ret = update_inline_extent_backref(trans, root, path, iref,
1336 } else if (ret == -ENOENT) {
1337 ret = setup_inline_extent_backref(trans, root, path, iref,
1338 parent, root_objectid,
1339 owner, offset, refs_to_add);
1344 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1345 struct btrfs_root *root,
1346 struct btrfs_path *path,
1347 u64 bytenr, u64 parent, u64 root_objectid,
1348 u64 owner, u64 offset, int refs_to_add)
1352 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1353 ret = insert_extent_data_ref(trans, root, path, bytenr,
1354 parent, root_objectid,
1355 owner, offset, refs_to_add);
1357 BUG_ON(refs_to_add != 1);
1358 ret = insert_tree_block_ref(trans, root, path, bytenr,
1359 parent, root_objectid);
1364 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1365 struct btrfs_root *root,
1366 struct btrfs_path *path,
1367 struct btrfs_extent_inline_ref *iref,
1368 int refs_to_drop, int is_data)
1372 BUG_ON(!is_data && refs_to_drop != 1);
1374 ret = update_inline_extent_backref(trans, root, path, iref,
1376 } else if (is_data) {
1377 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1379 ret = btrfs_del_item(trans, root, path);
1384 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1385 struct btrfs_root *root,
1386 u64 bytenr, u64 num_bytes, u64 parent,
1387 u64 root_objectid, u64 owner, u64 offset)
1389 struct btrfs_path *path;
1390 struct extent_buffer *leaf;
1391 struct btrfs_extent_item *item;
1396 path = btrfs_alloc_path();
1402 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1403 path, bytenr, num_bytes, parent,
1404 root_objectid, owner, offset, 1);
1408 if (ret != -EAGAIN) {
1413 leaf = path->nodes[0];
1414 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1415 refs = btrfs_extent_refs(leaf, item);
1416 btrfs_set_extent_refs(leaf, item, refs + 1);
1418 btrfs_mark_buffer_dirty(leaf);
1419 btrfs_release_path(path);
1423 /* now insert the actual backref */
1424 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1425 path, bytenr, parent, root_objectid,
1430 btrfs_free_path(path);
1431 finish_current_insert(trans, root->fs_info->extent_root);
1432 del_pending_extents(trans, root->fs_info->extent_root);
1437 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1438 struct btrfs_root *root)
1440 finish_current_insert(trans, root->fs_info->extent_root);
1441 del_pending_extents(trans, root->fs_info->extent_root);
1445 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root, u64 bytenr,
1447 u64 offset, int metadata, u64 *refs, u64 *flags)
1449 struct btrfs_path *path;
1451 struct btrfs_key key;
1452 struct extent_buffer *l;
1453 struct btrfs_extent_item *item;
1459 !btrfs_fs_incompat(root->fs_info,
1460 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
1461 offset = root->leafsize;
1465 path = btrfs_alloc_path();
1470 key.objectid = bytenr;
1471 key.offset = offset;
1473 key.type = BTRFS_METADATA_ITEM_KEY;
1475 key.type = BTRFS_EXTENT_ITEM_KEY;
1478 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1484 * Deal with the fact that we may have mixed SKINNY and normal refs. If
1485 * we didn't find what we wanted check and see if we have a normal ref
1486 * right next to us, or re-search if we are on the edge of the leaf just
1489 if (ret > 0 && metadata) {
1490 if (path->slots[0]) {
1492 btrfs_item_key_to_cpu(path->nodes[0], &key,
1494 if (key.objectid == bytenr &&
1495 key.type == BTRFS_EXTENT_ITEM_KEY &&
1496 key.offset == root->leafsize)
1501 btrfs_release_path(path);
1502 key.type = BTRFS_EXTENT_ITEM_KEY;
1503 key.offset = root->leafsize;
1515 item_size = btrfs_item_size_nr(l, path->slots[0]);
1516 if (item_size >= sizeof(*item)) {
1517 item = btrfs_item_ptr(l, path->slots[0],
1518 struct btrfs_extent_item);
1519 num_refs = btrfs_extent_refs(l, item);
1520 extent_flags = btrfs_extent_flags(l, item);
1522 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1523 struct btrfs_extent_item_v0 *ei0;
1524 BUG_ON(item_size != sizeof(*ei0));
1525 ei0 = btrfs_item_ptr(l, path->slots[0],
1526 struct btrfs_extent_item_v0);
1527 num_refs = btrfs_extent_refs_v0(l, ei0);
1528 /* FIXME: this isn't correct for data */
1529 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1534 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1538 *flags = extent_flags;
1540 btrfs_free_path(path);
1544 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1545 struct btrfs_root *root,
1546 u64 bytenr, int level, u64 flags)
1548 struct btrfs_path *path;
1550 struct btrfs_key key;
1551 struct extent_buffer *l;
1552 struct btrfs_extent_item *item;
1554 int skinny_metadata =
1555 btrfs_fs_incompat(root->fs_info,
1556 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1558 path = btrfs_alloc_path();
1563 key.objectid = bytenr;
1564 if (skinny_metadata) {
1566 key.type = BTRFS_METADATA_ITEM_KEY;
1568 key.offset = root->leafsize;
1569 key.type = BTRFS_EXTENT_ITEM_KEY;
1573 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1578 if (ret > 0 && skinny_metadata) {
1579 skinny_metadata = 0;
1580 if (path->slots[0]) {
1582 btrfs_item_key_to_cpu(path->nodes[0], &key,
1584 if (key.objectid == bytenr &&
1585 key.offset == root->leafsize &&
1586 key.type == BTRFS_EXTENT_ITEM_KEY)
1590 btrfs_release_path(path);
1591 key.offset = root->leafsize;
1592 key.type = BTRFS_EXTENT_ITEM_KEY;
1598 btrfs_print_leaf(root, path->nodes[0]);
1599 printk("failed to find block number %Lu\n",
1600 (unsigned long long)bytenr);
1604 item_size = btrfs_item_size_nr(l, path->slots[0]);
1605 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1606 if (item_size < sizeof(*item)) {
1607 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1613 item_size = btrfs_item_size_nr(l, path->slots[0]);
1616 BUG_ON(item_size < sizeof(*item));
1617 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1618 flags |= btrfs_extent_flags(l, item);
1619 btrfs_set_extent_flags(l, item, flags);
1621 btrfs_free_path(path);
1622 finish_current_insert(trans, root->fs_info->extent_root);
1623 del_pending_extents(trans, root->fs_info->extent_root);
1627 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1628 struct btrfs_root *root,
1629 struct extent_buffer *buf,
1630 int record_parent, int inc)
1637 struct btrfs_key key;
1638 struct btrfs_file_extent_item *fi;
1642 int (*process_func)(struct btrfs_trans_handle *trans,
1643 struct btrfs_root *root,
1644 u64, u64, u64, u64, u64, u64);
1646 ref_root = btrfs_header_owner(buf);
1647 nritems = btrfs_header_nritems(buf);
1648 level = btrfs_header_level(buf);
1650 if (!root->ref_cows && level == 0)
1654 process_func = btrfs_inc_extent_ref;
1656 process_func = btrfs_free_extent;
1659 parent = buf->start;
1663 for (i = 0; i < nritems; i++) {
1666 btrfs_item_key_to_cpu(buf, &key, i);
1667 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1669 fi = btrfs_item_ptr(buf, i,
1670 struct btrfs_file_extent_item);
1671 if (btrfs_file_extent_type(buf, fi) ==
1672 BTRFS_FILE_EXTENT_INLINE)
1674 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1678 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1679 key.offset -= btrfs_file_extent_offset(buf, fi);
1680 ret = process_func(trans, root, bytenr, num_bytes,
1681 parent, ref_root, key.objectid,
1688 bytenr = btrfs_node_blockptr(buf, i);
1689 num_bytes = btrfs_level_size(root, level - 1);
1690 ret = process_func(trans, root, bytenr, num_bytes,
1691 parent, ref_root, level - 1, 0);
1704 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1705 struct extent_buffer *buf, int record_parent)
1707 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1710 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1711 struct extent_buffer *buf, int record_parent)
1713 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1716 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1717 struct btrfs_root *root,
1718 struct btrfs_path *path,
1719 struct btrfs_block_group_cache *cache)
1723 struct btrfs_root *extent_root = root->fs_info->extent_root;
1725 struct extent_buffer *leaf;
1727 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1732 leaf = path->nodes[0];
1733 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1734 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1735 btrfs_mark_buffer_dirty(leaf);
1736 btrfs_release_path(path);
1738 finish_current_insert(trans, extent_root);
1739 pending_ret = del_pending_extents(trans, extent_root);
1748 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1749 struct btrfs_root *root)
1751 struct extent_io_tree *block_group_cache;
1752 struct btrfs_block_group_cache *cache;
1754 struct btrfs_path *path;
1760 block_group_cache = &root->fs_info->block_group_cache;
1761 path = btrfs_alloc_path();
1766 ret = find_first_extent_bit(block_group_cache, last,
1767 &start, &end, BLOCK_GROUP_DIRTY);
1776 ret = get_state_private(block_group_cache, start, &ptr);
1779 clear_extent_bits(block_group_cache, start, end,
1780 BLOCK_GROUP_DIRTY, GFP_NOFS);
1782 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1783 ret = write_one_cache_group(trans, root, path, cache);
1785 btrfs_free_path(path);
1789 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1792 struct btrfs_space_info *found;
1794 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1796 list_for_each_entry(found, &info->space_info, list) {
1797 if (found->flags & flags)
1804 static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
1805 u64 total_bytes, u64 bytes_used,
1806 struct btrfs_space_info **space_info)
1808 struct btrfs_space_info *found;
1810 /* only support free block group which is empty */
1814 found = __find_space_info(fs_info, flags);
1817 if (found->total_bytes < total_bytes) {
1819 "WARNING: bad space info to free %llu only have %llu\n",
1820 total_bytes, found->total_bytes);
1823 found->total_bytes -= total_bytes;
1825 *space_info = found;
1829 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1830 u64 total_bytes, u64 bytes_used,
1831 struct btrfs_space_info **space_info)
1833 struct btrfs_space_info *found;
1835 found = __find_space_info(info, flags);
1837 found->total_bytes += total_bytes;
1838 found->bytes_used += bytes_used;
1839 if (found->total_bytes < found->bytes_used) {
1840 fprintf(stderr, "warning, bad space info total_bytes "
1842 (unsigned long long)found->total_bytes,
1843 (unsigned long long)found->bytes_used);
1845 *space_info = found;
1848 found = kmalloc(sizeof(*found), GFP_NOFS);
1852 list_add(&found->list, &info->space_info);
1853 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1854 found->total_bytes = total_bytes;
1855 found->bytes_used = bytes_used;
1856 found->bytes_pinned = 0;
1858 *space_info = found;
1863 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1865 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1866 BTRFS_BLOCK_GROUP_RAID1 |
1867 BTRFS_BLOCK_GROUP_RAID10 |
1868 BTRFS_BLOCK_GROUP_RAID5 |
1869 BTRFS_BLOCK_GROUP_RAID6 |
1870 BTRFS_BLOCK_GROUP_DUP);
1872 if (flags & BTRFS_BLOCK_GROUP_DATA)
1873 fs_info->avail_data_alloc_bits |= extra_flags;
1874 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1875 fs_info->avail_metadata_alloc_bits |= extra_flags;
1876 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1877 fs_info->avail_system_alloc_bits |= extra_flags;
1881 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1882 struct btrfs_root *extent_root, u64 alloc_bytes,
1885 struct btrfs_space_info *space_info;
1891 space_info = __find_space_info(extent_root->fs_info, flags);
1893 ret = update_space_info(extent_root->fs_info, flags,
1897 BUG_ON(!space_info);
1899 if (space_info->full)
1902 thresh = div_factor(space_info->total_bytes, 7);
1903 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1907 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1909 if (ret == -ENOSPC) {
1910 space_info->full = 1;
1916 ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1917 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1922 static int update_block_group(struct btrfs_trans_handle *trans,
1923 struct btrfs_root *root,
1924 u64 bytenr, u64 num_bytes, int alloc,
1927 struct btrfs_block_group_cache *cache;
1928 struct btrfs_fs_info *info = root->fs_info;
1929 u64 total = num_bytes;
1935 /* block accounting for super block */
1936 old_val = btrfs_super_bytes_used(info->super_copy);
1938 old_val += num_bytes;
1940 old_val -= num_bytes;
1941 btrfs_set_super_bytes_used(info->super_copy, old_val);
1943 /* block accounting for root item */
1944 old_val = btrfs_root_used(&root->root_item);
1946 old_val += num_bytes;
1948 old_val -= num_bytes;
1949 btrfs_set_root_used(&root->root_item, old_val);
1952 cache = btrfs_lookup_block_group(info, bytenr);
1956 byte_in_group = bytenr - cache->key.objectid;
1957 WARN_ON(byte_in_group > cache->key.offset);
1958 start = cache->key.objectid;
1959 end = start + cache->key.offset - 1;
1960 set_extent_bits(&info->block_group_cache, start, end,
1961 BLOCK_GROUP_DIRTY, GFP_NOFS);
1963 old_val = btrfs_block_group_used(&cache->item);
1964 num_bytes = min(total, cache->key.offset - byte_in_group);
1967 old_val += num_bytes;
1968 cache->space_info->bytes_used += num_bytes;
1970 old_val -= num_bytes;
1971 cache->space_info->bytes_used -= num_bytes;
1973 set_extent_dirty(&info->free_space_cache,
1974 bytenr, bytenr + num_bytes - 1,
1978 btrfs_set_block_group_used(&cache->item, old_val);
1980 bytenr += num_bytes;
1985 static int update_pinned_extents(struct btrfs_root *root,
1986 u64 bytenr, u64 num, int pin)
1989 struct btrfs_block_group_cache *cache;
1990 struct btrfs_fs_info *fs_info = root->fs_info;
1993 set_extent_dirty(&fs_info->pinned_extents,
1994 bytenr, bytenr + num - 1, GFP_NOFS);
1996 clear_extent_dirty(&fs_info->pinned_extents,
1997 bytenr, bytenr + num - 1, GFP_NOFS);
2000 cache = btrfs_lookup_block_group(fs_info, bytenr);
2002 len = min((u64)root->sectorsize, num);
2006 len = min(num, cache->key.offset -
2007 (bytenr - cache->key.objectid));
2009 cache->pinned += len;
2010 cache->space_info->bytes_pinned += len;
2011 fs_info->total_pinned += len;
2013 cache->pinned -= len;
2014 cache->space_info->bytes_pinned -= len;
2015 fs_info->total_pinned -= len;
2024 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 struct extent_io_tree *unpin)
2031 struct extent_io_tree *free_space_cache;
2032 free_space_cache = &root->fs_info->free_space_cache;
2035 ret = find_first_extent_bit(unpin, 0, &start, &end,
2039 update_pinned_extents(root, start, end + 1 - start, 0);
2040 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2041 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2046 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2052 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2053 &end, EXTENT_LOCKED);
2055 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2061 static int finish_current_insert(struct btrfs_trans_handle *trans,
2062 struct btrfs_root *extent_root)
2067 struct btrfs_fs_info *info = extent_root->fs_info;
2068 struct pending_extent_op *extent_op;
2069 struct btrfs_key key;
2071 int skinny_metadata =
2072 btrfs_fs_incompat(extent_root->fs_info,
2073 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2076 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2077 &end, EXTENT_LOCKED);
2081 ret = get_state_private(&info->extent_ins, start, &priv);
2083 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2085 if (extent_op->type == PENDING_EXTENT_INSERT) {
2086 key.objectid = start;
2087 if (skinny_metadata) {
2088 key.offset = extent_op->level;
2089 key.type = BTRFS_METADATA_ITEM_KEY;
2091 key.offset = extent_op->num_bytes;
2092 key.type = BTRFS_EXTENT_ITEM_KEY;
2094 ret = alloc_reserved_tree_block(trans, extent_root,
2095 extent_root->root_key.objectid,
2099 extent_op->level, &key);
2105 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2112 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2113 struct btrfs_root *root,
2114 u64 bytenr, u64 num_bytes, int is_data)
2117 struct extent_buffer *buf;
2122 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2126 /* we can reuse a block if it hasn't been written
2127 * and it is from this transaction. We can't
2128 * reuse anything from the tree log root because
2129 * it has tiny sub-transactions.
2131 if (btrfs_buffer_uptodate(buf, 0)) {
2132 u64 header_owner = btrfs_header_owner(buf);
2133 u64 header_transid = btrfs_header_generation(buf);
2134 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2135 header_transid == trans->transid &&
2136 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2137 clean_tree_block(NULL, root, buf);
2138 free_extent_buffer(buf);
2142 free_extent_buffer(buf);
2144 update_pinned_extents(root, bytenr, num_bytes, 1);
2150 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2151 u64 bytenr, u64 num_bytes)
2153 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2156 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2157 u64 bytenr, u64 num_bytes)
2159 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2163 * remove an extent from the root, returns 0 on success
2165 static int __free_extent(struct btrfs_trans_handle *trans,
2166 struct btrfs_root *root,
2167 u64 bytenr, u64 num_bytes, u64 parent,
2168 u64 root_objectid, u64 owner_objectid,
2169 u64 owner_offset, int refs_to_drop)
2172 struct btrfs_key key;
2173 struct btrfs_path *path;
2174 struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2175 struct btrfs_root *extent_root = root->fs_info->extent_root;
2176 struct extent_buffer *leaf;
2177 struct btrfs_extent_item *ei;
2178 struct btrfs_extent_inline_ref *iref;
2181 int extent_slot = 0;
2182 int found_extent = 0;
2186 int skinny_metadata =
2187 btrfs_fs_incompat(extent_root->fs_info,
2188 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2190 if (root->fs_info->free_extent_hook) {
2191 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2192 parent, root_objectid, owner_objectid,
2193 owner_offset, refs_to_drop);
2196 path = btrfs_alloc_path();
2202 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2204 skinny_metadata = 0;
2205 BUG_ON(!is_data && refs_to_drop != 1);
2207 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2208 bytenr, num_bytes, parent,
2209 root_objectid, owner_objectid,
2212 extent_slot = path->slots[0];
2213 while (extent_slot >= 0) {
2214 btrfs_item_key_to_cpu(path->nodes[0], &key,
2216 if (key.objectid != bytenr)
2218 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2219 key.offset == num_bytes) {
2223 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2224 key.offset == owner_objectid) {
2228 if (path->slots[0] - extent_slot > 5)
2232 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2233 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2234 if (found_extent && item_size < sizeof(*ei))
2237 if (!found_extent) {
2239 ret = remove_extent_backref(trans, extent_root, path,
2243 btrfs_release_path(path);
2245 key.objectid = bytenr;
2247 if (skinny_metadata) {
2248 key.type = BTRFS_METADATA_ITEM_KEY;
2249 key.offset = owner_objectid;
2251 key.type = BTRFS_EXTENT_ITEM_KEY;
2252 key.offset = num_bytes;
2255 ret = btrfs_search_slot(trans, extent_root,
2257 if (ret > 0 && skinny_metadata && path->slots[0]) {
2259 btrfs_item_key_to_cpu(path->nodes[0],
2262 if (key.objectid == bytenr &&
2263 key.type == BTRFS_EXTENT_ITEM_KEY &&
2264 key.offset == num_bytes)
2268 if (ret > 0 && skinny_metadata) {
2269 skinny_metadata = 0;
2270 btrfs_release_path(path);
2271 key.type = BTRFS_EXTENT_ITEM_KEY;
2272 key.offset = num_bytes;
2273 ret = btrfs_search_slot(trans, extent_root,
2278 printk(KERN_ERR "umm, got %d back from search"
2279 ", was looking for %llu\n", ret,
2280 (unsigned long long)bytenr);
2281 btrfs_print_leaf(extent_root, path->nodes[0]);
2284 extent_slot = path->slots[0];
2287 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2288 "parent %llu root %llu owner %llu offset %llu\n",
2289 (unsigned long long)bytenr,
2290 (unsigned long long)parent,
2291 (unsigned long long)root_objectid,
2292 (unsigned long long)owner_objectid,
2293 (unsigned long long)owner_offset);
2298 leaf = path->nodes[0];
2299 item_size = btrfs_item_size_nr(leaf, extent_slot);
2300 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2301 if (item_size < sizeof(*ei)) {
2302 BUG_ON(found_extent || extent_slot != path->slots[0]);
2303 ret = convert_extent_item_v0(trans, extent_root, path,
2307 btrfs_release_path(path);
2309 key.objectid = bytenr;
2310 key.type = BTRFS_EXTENT_ITEM_KEY;
2311 key.offset = num_bytes;
2313 ret = btrfs_search_slot(trans, extent_root, &key, path,
2316 printk(KERN_ERR "umm, got %d back from search"
2317 ", was looking for %llu\n", ret,
2318 (unsigned long long)bytenr);
2319 btrfs_print_leaf(extent_root, path->nodes[0]);
2322 extent_slot = path->slots[0];
2323 leaf = path->nodes[0];
2324 item_size = btrfs_item_size_nr(leaf, extent_slot);
2327 BUG_ON(item_size < sizeof(*ei));
2328 ei = btrfs_item_ptr(leaf, extent_slot,
2329 struct btrfs_extent_item);
2330 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2331 key.type == BTRFS_EXTENT_ITEM_KEY) {
2332 struct btrfs_tree_block_info *bi;
2333 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2334 bi = (struct btrfs_tree_block_info *)(ei + 1);
2335 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2338 refs = btrfs_extent_refs(leaf, ei);
2339 BUG_ON(refs < refs_to_drop);
2340 refs -= refs_to_drop;
2344 * In the case of inline back ref, reference count will
2345 * be updated by remove_extent_backref
2348 BUG_ON(!found_extent);
2350 btrfs_set_extent_refs(leaf, ei, refs);
2351 btrfs_mark_buffer_dirty(leaf);
2354 ret = remove_extent_backref(trans, extent_root, path,
2364 BUG_ON(is_data && refs_to_drop !=
2365 extent_data_ref_count(root, path, iref));
2367 BUG_ON(path->slots[0] != extent_slot);
2369 BUG_ON(path->slots[0] != extent_slot + 1);
2370 path->slots[0] = extent_slot;
2375 if (ops && ops->free_extent) {
2376 ret = ops->free_extent(root, bytenr, num_bytes);
2384 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2391 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2394 btrfs_release_path(path);
2397 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2401 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2404 btrfs_free_path(path);
2405 finish_current_insert(trans, extent_root);
2410 * find all the blocks marked as pending in the radix tree and remove
2411 * them from the extent map
2413 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2414 btrfs_root *extent_root)
2421 struct extent_io_tree *pending_del;
2422 struct extent_io_tree *extent_ins;
2423 struct pending_extent_op *extent_op;
2425 extent_ins = &extent_root->fs_info->extent_ins;
2426 pending_del = &extent_root->fs_info->pending_del;
2429 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2434 ret = get_state_private(pending_del, start, &priv);
2436 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2438 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2441 if (!test_range_bit(extent_ins, start, end,
2442 EXTENT_LOCKED, 0)) {
2443 ret = __free_extent(trans, extent_root,
2444 start, end + 1 - start, 0,
2445 extent_root->root_key.objectid,
2446 extent_op->level, 0, 1);
2450 ret = get_state_private(extent_ins, start, &priv);
2452 extent_op = (struct pending_extent_op *)
2453 (unsigned long)priv;
2455 clear_extent_bits(extent_ins, start, end,
2456 EXTENT_LOCKED, GFP_NOFS);
2458 if (extent_op->type == PENDING_BACKREF_UPDATE)
2470 * remove an extent from the root, returns 0 on success
2473 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2474 struct btrfs_root *root,
2475 u64 bytenr, u64 num_bytes, u64 parent,
2476 u64 root_objectid, u64 owner, u64 offset)
2478 struct btrfs_root *extent_root = root->fs_info->extent_root;
2482 WARN_ON(num_bytes < root->sectorsize);
2483 if (root == extent_root) {
2484 struct pending_extent_op *extent_op;
2486 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2489 extent_op->type = PENDING_EXTENT_DELETE;
2490 extent_op->bytenr = bytenr;
2491 extent_op->num_bytes = num_bytes;
2492 extent_op->level = (int)owner;
2494 set_extent_bits(&root->fs_info->pending_del,
2495 bytenr, bytenr + num_bytes - 1,
2496 EXTENT_LOCKED, GFP_NOFS);
2497 set_state_private(&root->fs_info->pending_del,
2498 bytenr, (unsigned long)extent_op);
2501 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2502 root_objectid, owner, offset, 1);
2503 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2504 return ret ? ret : pending_ret;
2507 static u64 stripe_align(struct btrfs_root *root, u64 val)
2509 u64 mask = ((u64)root->stripesize - 1);
2510 u64 ret = (val + mask) & ~mask;
2515 * walks the btree of allocated extents and find a hole of a given size.
2516 * The key ins is changed to record the hole:
2517 * ins->objectid == block start
2518 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2519 * ins->offset == number of blocks
2520 * Any available blocks before search_start are skipped.
2522 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2523 struct btrfs_root *orig_root,
2524 u64 num_bytes, u64 empty_size,
2525 u64 search_start, u64 search_end,
2526 u64 hint_byte, struct btrfs_key *ins,
2527 u64 exclude_start, u64 exclude_nr,
2531 u64 orig_search_start = search_start;
2532 struct btrfs_root * root = orig_root->fs_info->extent_root;
2533 struct btrfs_fs_info *info = root->fs_info;
2534 u64 total_needed = num_bytes;
2535 struct btrfs_block_group_cache *block_group;
2539 WARN_ON(num_bytes < root->sectorsize);
2540 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2542 search_start = stripe_align(root, search_start);
2545 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2547 hint_byte = search_start;
2548 block_group = btrfs_find_block_group(root, block_group,
2549 hint_byte, data, 1);
2551 block_group = btrfs_find_block_group(root,
2553 search_start, data, 1);
2556 total_needed += empty_size;
2559 search_start = stripe_align(root, search_start);
2561 block_group = btrfs_lookup_first_block_group(info,
2564 block_group = btrfs_lookup_first_block_group(info,
2567 ret = find_search_start(root, &block_group, &search_start,
2568 total_needed, data);
2572 ins->objectid = search_start;
2573 ins->offset = num_bytes;
2575 if (ins->objectid + num_bytes >
2576 block_group->key.objectid + block_group->key.offset) {
2577 search_start = block_group->key.objectid +
2578 block_group->key.offset;
2582 if (test_range_bit(&info->extent_ins, ins->objectid,
2583 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2584 search_start = ins->objectid + num_bytes;
2588 if (test_range_bit(&info->pinned_extents, ins->objectid,
2589 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2590 search_start = ins->objectid + num_bytes;
2594 if (info->excluded_extents &&
2595 test_range_bit(info->excluded_extents, ins->objectid,
2596 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2597 search_start = ins->objectid + num_bytes;
2601 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2602 ins->objectid < exclude_start + exclude_nr)) {
2603 search_start = exclude_start + exclude_nr;
2607 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2608 block_group = btrfs_lookup_block_group(info, ins->objectid);
2610 trans->block_group = block_group;
2612 ins->offset = num_bytes;
2616 block_group = btrfs_lookup_first_block_group(info, search_start);
2618 search_start = orig_search_start;
2625 total_needed -= empty_size;
2631 block_group = btrfs_find_block_group(root, block_group,
2632 search_start, data, 0);
2639 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2640 struct btrfs_root *root,
2641 u64 num_bytes, u64 empty_size,
2642 u64 hint_byte, u64 search_end,
2643 struct btrfs_key *ins, int data)
2646 u64 search_start = 0;
2648 struct btrfs_fs_info *info = root->fs_info;
2650 if (info->extent_ops) {
2651 struct btrfs_extent_ops *ops = info->extent_ops;
2652 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2658 alloc_profile = info->avail_data_alloc_bits &
2659 info->data_alloc_profile;
2660 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2661 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2662 info->system_allocs >= 0) {
2663 alloc_profile = info->avail_system_alloc_bits &
2664 info->system_alloc_profile;
2665 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2667 alloc_profile = info->avail_metadata_alloc_bits &
2668 info->metadata_alloc_profile;
2669 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2672 if (root->ref_cows) {
2673 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2674 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2676 BTRFS_BLOCK_GROUP_METADATA);
2679 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2680 num_bytes + 2 * 1024 * 1024, data);
2684 WARN_ON(num_bytes < root->sectorsize);
2685 ret = find_free_extent(trans, root, num_bytes, empty_size,
2686 search_start, search_end, hint_byte, ins,
2687 trans->alloc_exclude_start,
2688 trans->alloc_exclude_nr, data);
2691 clear_extent_dirty(&root->fs_info->free_space_cache,
2692 ins->objectid, ins->objectid + ins->offset - 1,
2697 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2698 struct btrfs_root *root,
2699 u64 root_objectid, u64 generation,
2700 u64 flags, struct btrfs_disk_key *key,
2701 int level, struct btrfs_key *ins)
2704 struct btrfs_fs_info *fs_info = root->fs_info;
2705 struct btrfs_extent_item *extent_item;
2706 struct btrfs_tree_block_info *block_info;
2707 struct btrfs_extent_inline_ref *iref;
2708 struct btrfs_path *path;
2709 struct extent_buffer *leaf;
2710 u32 size = sizeof(*extent_item) + sizeof(*iref);
2711 int skinny_metadata =
2712 btrfs_fs_incompat(fs_info,
2713 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2715 if (!skinny_metadata)
2716 size += sizeof(*block_info);
2718 path = btrfs_alloc_path();
2721 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2725 leaf = path->nodes[0];
2726 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2727 struct btrfs_extent_item);
2728 btrfs_set_extent_refs(leaf, extent_item, 1);
2729 btrfs_set_extent_generation(leaf, extent_item, generation);
2730 btrfs_set_extent_flags(leaf, extent_item,
2731 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2733 if (skinny_metadata) {
2734 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2736 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2737 btrfs_set_tree_block_key(leaf, block_info, key);
2738 btrfs_set_tree_block_level(leaf, block_info, level);
2739 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2742 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2743 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2745 btrfs_mark_buffer_dirty(leaf);
2746 btrfs_free_path(path);
2748 ret = update_block_group(trans, root, ins->objectid, root->leafsize,
2753 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2754 struct btrfs_root *root, u64 num_bytes,
2755 u64 root_objectid, u64 generation,
2756 u64 flags, struct btrfs_disk_key *key,
2757 int level, u64 empty_size, u64 hint_byte,
2758 u64 search_end, struct btrfs_key *ins)
2761 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2762 hint_byte, search_end, ins, 0);
2765 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2766 struct pending_extent_op *extent_op;
2768 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2771 extent_op->type = PENDING_EXTENT_INSERT;
2772 extent_op->bytenr = ins->objectid;
2773 extent_op->num_bytes = ins->offset;
2774 extent_op->level = level;
2775 extent_op->flags = flags;
2776 memcpy(&extent_op->key, key, sizeof(*key));
2778 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2779 ins->objectid + ins->offset - 1,
2780 EXTENT_LOCKED, GFP_NOFS);
2781 set_state_private(&root->fs_info->extent_ins,
2782 ins->objectid, (unsigned long)extent_op);
2784 if (btrfs_fs_incompat(root->fs_info,
2785 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
2786 ins->offset = level;
2787 ins->type = BTRFS_METADATA_ITEM_KEY;
2789 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2792 finish_current_insert(trans, root->fs_info->extent_root);
2793 del_pending_extents(trans, root->fs_info->extent_root);
2799 * helper function to allocate a block for a given tree
2800 * returns the tree buffer or NULL.
2802 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root,
2804 u32 blocksize, u64 root_objectid,
2805 struct btrfs_disk_key *key, int level,
2806 u64 hint, u64 empty_size)
2808 struct btrfs_key ins;
2810 struct extent_buffer *buf;
2812 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2813 trans->transid, 0, key, level,
2814 empty_size, hint, (u64)-1, &ins);
2817 return ERR_PTR(ret);
2820 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2822 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2823 0, root->root_key.objectid, level, 0);
2825 return ERR_PTR(-ENOMEM);
2827 btrfs_set_buffer_uptodate(buf);
2828 trans->blocks_used++;
2835 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2836 struct btrfs_root *root,
2837 struct extent_buffer *leaf)
2840 u64 leaf_generation;
2841 struct btrfs_key key;
2842 struct btrfs_file_extent_item *fi;
2847 BUG_ON(!btrfs_is_leaf(leaf));
2848 nritems = btrfs_header_nritems(leaf);
2849 leaf_owner = btrfs_header_owner(leaf);
2850 leaf_generation = btrfs_header_generation(leaf);
2852 for (i = 0; i < nritems; i++) {
2855 btrfs_item_key_to_cpu(leaf, &key, i);
2856 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2858 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2859 if (btrfs_file_extent_type(leaf, fi) ==
2860 BTRFS_FILE_EXTENT_INLINE)
2863 * FIXME make sure to insert a trans record that
2864 * repeats the snapshot del on crash
2866 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2867 if (disk_bytenr == 0)
2869 ret = btrfs_free_extent(trans, root, disk_bytenr,
2870 btrfs_file_extent_disk_num_bytes(leaf, fi),
2871 leaf->start, leaf_owner, leaf_generation,
2878 static void noinline reada_walk_down(struct btrfs_root *root,
2879 struct extent_buffer *node,
2892 nritems = btrfs_header_nritems(node);
2893 level = btrfs_header_level(node);
2897 for (i = slot; i < nritems && skipped < 32; i++) {
2898 bytenr = btrfs_node_blockptr(node, i);
2899 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2900 (last > bytenr && last - bytenr > 32 * 1024))) {
2904 blocksize = btrfs_level_size(root, level - 1);
2906 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2914 mutex_unlock(&root->fs_info->fs_mutex);
2915 ret = readahead_tree_block(root, bytenr, blocksize,
2916 btrfs_node_ptr_generation(node, i));
2917 last = bytenr + blocksize;
2919 mutex_lock(&root->fs_info->fs_mutex);
2926 * helper function for drop_snapshot, this walks down the tree dropping ref
2927 * counts as it goes.
2929 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2930 struct btrfs_root *root,
2931 struct btrfs_path *path, int *level)
2937 struct extent_buffer *next;
2938 struct extent_buffer *cur;
2939 struct extent_buffer *parent;
2944 WARN_ON(*level < 0);
2945 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2946 ret = btrfs_lookup_extent_ref(trans, root,
2947 path->nodes[*level]->start,
2948 path->nodes[*level]->len, &refs);
2954 * walk down to the last node level and free all the leaves
2956 while(*level >= 0) {
2957 WARN_ON(*level < 0);
2958 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2959 cur = path->nodes[*level];
2961 if (btrfs_header_level(cur) != *level)
2964 if (path->slots[*level] >=
2965 btrfs_header_nritems(cur))
2968 ret = drop_leaf_ref(trans, root, cur);
2972 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2973 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2974 blocksize = btrfs_level_size(root, *level - 1);
2975 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2979 parent = path->nodes[*level];
2980 root_owner = btrfs_header_owner(parent);
2981 root_gen = btrfs_header_generation(parent);
2982 path->slots[*level]++;
2983 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2984 parent->start, root_owner,
2985 root_gen, *level - 1, 1);
2989 next = btrfs_find_tree_block(root, bytenr, blocksize);
2990 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2991 free_extent_buffer(next);
2992 reada_walk_down(root, cur, path->slots[*level]);
2993 mutex_unlock(&root->fs_info->fs_mutex);
2994 next = read_tree_block(root, bytenr, blocksize,
2996 mutex_lock(&root->fs_info->fs_mutex);
2997 if (!extent_buffer_uptodate(next)) {
2999 ret = PTR_ERR(next);
3005 WARN_ON(*level <= 0);
3006 if (path->nodes[*level-1])
3007 free_extent_buffer(path->nodes[*level-1]);
3008 path->nodes[*level-1] = next;
3009 *level = btrfs_header_level(next);
3010 path->slots[*level] = 0;
3013 WARN_ON(*level < 0);
3014 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3016 if (path->nodes[*level] == root->node) {
3017 root_owner = root->root_key.objectid;
3018 parent = path->nodes[*level];
3020 parent = path->nodes[*level + 1];
3021 root_owner = btrfs_header_owner(parent);
3024 root_gen = btrfs_header_generation(parent);
3025 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3026 path->nodes[*level]->len, parent->start,
3027 root_owner, root_gen, *level, 1);
3028 free_extent_buffer(path->nodes[*level]);
3029 path->nodes[*level] = NULL;
3036 * helper for dropping snapshots. This walks back up the tree in the path
3037 * to find the first node higher up where we haven't yet gone through
3040 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3041 struct btrfs_root *root,
3042 struct btrfs_path *path, int *level)
3046 struct btrfs_root_item *root_item = &root->root_item;
3051 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3052 slot = path->slots[i];
3053 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3054 struct extent_buffer *node;
3055 struct btrfs_disk_key disk_key;
3056 node = path->nodes[i];
3059 WARN_ON(*level == 0);
3060 btrfs_node_key(node, &disk_key, path->slots[i]);
3061 memcpy(&root_item->drop_progress,
3062 &disk_key, sizeof(disk_key));
3063 root_item->drop_level = i;
3066 struct extent_buffer *parent;
3067 if (path->nodes[*level] == root->node)
3068 parent = path->nodes[*level];
3070 parent = path->nodes[*level + 1];
3072 root_owner = btrfs_header_owner(parent);
3073 root_gen = btrfs_header_generation(parent);
3074 ret = btrfs_free_extent(trans, root,
3075 path->nodes[*level]->start,
3076 path->nodes[*level]->len,
3077 parent->start, root_owner,
3078 root_gen, *level, 1);
3080 free_extent_buffer(path->nodes[*level]);
3081 path->nodes[*level] = NULL;
3090 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3092 struct btrfs_space_info *sinfo;
3093 struct btrfs_block_group_cache *cache;
3100 ret = find_first_extent_bit(&info->block_group_cache, 0,
3101 &start, &end, (unsigned int)-1);
3104 ret = get_state_private(&info->block_group_cache, start, &ptr);
3106 cache = u64_to_ptr(ptr);
3107 if (cache->free_space_ctl) {
3108 btrfs_remove_free_space_cache(cache);
3109 kfree(cache->free_space_ctl);
3113 clear_extent_bits(&info->block_group_cache, start,
3114 end, (unsigned int)-1, GFP_NOFS);
3117 ret = find_first_extent_bit(&info->free_space_cache, 0,
3118 &start, &end, EXTENT_DIRTY);
3121 clear_extent_dirty(&info->free_space_cache, start,
3125 while (!list_empty(&info->space_info)) {
3126 sinfo = list_entry(info->space_info.next,
3127 struct btrfs_space_info, list);
3128 list_del_init(&sinfo->list);
3134 static int find_first_block_group(struct btrfs_root *root,
3135 struct btrfs_path *path, struct btrfs_key *key)
3138 struct btrfs_key found_key;
3139 struct extent_buffer *leaf;
3142 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3146 slot = path->slots[0];
3147 leaf = path->nodes[0];
3148 if (slot >= btrfs_header_nritems(leaf)) {
3149 ret = btrfs_next_leaf(root, path);
3156 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3158 if (found_key.objectid >= key->objectid &&
3159 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3168 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3169 struct btrfs_block_group_cache *cache)
3176 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3177 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3178 cache->bytes_super += stripe_len;
3181 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3182 bytenr = btrfs_sb_offset(i);
3183 ret = btrfs_rmap_block(&fs_info->mapping_tree,
3184 cache->key.objectid, bytenr,
3185 0, &logical, &nr, &stripe_len);
3192 if (logical[nr] > cache->key.objectid +
3196 if (logical[nr] + stripe_len <= cache->key.objectid)
3199 start = logical[nr];
3200 if (start < cache->key.objectid) {
3201 start = cache->key.objectid;
3202 len = (logical[nr] + stripe_len) - start;
3204 len = min_t(u64, stripe_len,
3205 cache->key.objectid +
3206 cache->key.offset - start);
3209 cache->bytes_super += len;
3216 int btrfs_read_block_groups(struct btrfs_root *root)
3218 struct btrfs_path *path;
3221 struct btrfs_block_group_cache *cache;
3222 struct btrfs_fs_info *info = root->fs_info;
3223 struct btrfs_space_info *space_info;
3224 struct extent_io_tree *block_group_cache;
3225 struct btrfs_key key;
3226 struct btrfs_key found_key;
3227 struct extent_buffer *leaf;
3229 block_group_cache = &info->block_group_cache;
3231 root = info->extent_root;
3234 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3235 path = btrfs_alloc_path();
3240 ret = find_first_block_group(root, path, &key);
3248 leaf = path->nodes[0];
3249 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3250 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3256 read_extent_buffer(leaf, &cache->item,
3257 btrfs_item_ptr_offset(leaf, path->slots[0]),
3258 sizeof(cache->item));
3259 memcpy(&cache->key, &found_key, sizeof(found_key));
3262 key.objectid = found_key.objectid + found_key.offset;
3263 btrfs_release_path(path);
3264 cache->flags = btrfs_block_group_flags(&cache->item);
3266 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3267 bit = BLOCK_GROUP_DATA;
3268 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3269 bit = BLOCK_GROUP_SYSTEM;
3270 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3271 bit = BLOCK_GROUP_METADATA;
3273 set_avail_alloc_bits(info, cache->flags);
3274 if (btrfs_chunk_readonly(root, cache->key.objectid))
3277 account_super_bytes(info, cache);
3279 ret = update_space_info(info, cache->flags, found_key.offset,
3280 btrfs_block_group_used(&cache->item),
3283 cache->space_info = space_info;
3285 /* use EXTENT_LOCKED to prevent merging */
3286 set_extent_bits(block_group_cache, found_key.objectid,
3287 found_key.objectid + found_key.offset - 1,
3288 bit | EXTENT_LOCKED, GFP_NOFS);
3289 set_state_private(block_group_cache, found_key.objectid,
3290 (unsigned long)cache);
3294 btrfs_free_path(path);
3298 struct btrfs_block_group_cache *
3299 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3300 u64 chunk_objectid, u64 chunk_offset, u64 size)
3304 struct btrfs_block_group_cache *cache;
3305 struct extent_io_tree *block_group_cache;
3307 block_group_cache = &fs_info->block_group_cache;
3309 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3311 cache->key.objectid = chunk_offset;
3312 cache->key.offset = size;
3314 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3315 btrfs_set_block_group_used(&cache->item, bytes_used);
3316 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3317 cache->flags = type;
3318 btrfs_set_block_group_flags(&cache->item, type);
3320 account_super_bytes(fs_info, cache);
3321 ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3322 &cache->space_info);
3325 bit = block_group_state_bits(type);
3326 ret = set_extent_bits(block_group_cache, chunk_offset,
3327 chunk_offset + size - 1,
3328 bit | EXTENT_LOCKED, GFP_NOFS);
3331 ret = set_state_private(block_group_cache, chunk_offset,
3332 (unsigned long)cache);
3334 set_avail_alloc_bits(fs_info, type);
3339 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root, u64 bytes_used,
3341 u64 type, u64 chunk_objectid, u64 chunk_offset,
3345 struct btrfs_root *extent_root;
3346 struct btrfs_block_group_cache *cache;
3348 cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3349 chunk_objectid, chunk_offset, size);
3350 extent_root = root->fs_info->extent_root;
3351 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3352 sizeof(cache->item));
3355 ret = finish_current_insert(trans, extent_root);
3357 ret = del_pending_extents(trans, extent_root);
3364 * This is for converter use only.
3366 * In that case, we don't know where are free blocks located.
3367 * Therefore all block group cache entries must be setup properly
3368 * before doing any block allocation.
3370 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root)
3379 u64 total_metadata = 0;
3383 struct btrfs_root *extent_root;
3384 struct btrfs_block_group_cache *cache;
3385 struct extent_io_tree *block_group_cache;
3387 extent_root = root->fs_info->extent_root;
3388 block_group_cache = &root->fs_info->block_group_cache;
3389 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3390 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3391 group_align = 64 * root->sectorsize;
3394 while (cur_start < total_bytes) {
3395 group_size = total_bytes / 12;
3396 group_size = min_t(u64, group_size, total_bytes - cur_start);
3397 if (cur_start == 0) {
3398 bit = BLOCK_GROUP_SYSTEM;
3399 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3401 group_size &= ~(group_align - 1);
3402 group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3403 group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3405 group_size &= ~(group_align - 1);
3406 if (total_data >= total_metadata * 2) {
3407 group_type = BTRFS_BLOCK_GROUP_METADATA;
3408 group_size = min_t(u64, group_size,
3409 1ULL * 1024 * 1024 * 1024);
3410 total_metadata += group_size;
3412 group_type = BTRFS_BLOCK_GROUP_DATA;
3413 group_size = min_t(u64, group_size,
3414 5ULL * 1024 * 1024 * 1024);
3415 total_data += group_size;
3417 if ((total_bytes - cur_start) * 4 < group_size * 5)
3418 group_size = total_bytes - cur_start;
3421 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3424 cache->key.objectid = cur_start;
3425 cache->key.offset = group_size;
3426 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3428 btrfs_set_block_group_used(&cache->item, 0);
3429 btrfs_set_block_group_chunk_objectid(&cache->item,
3431 btrfs_set_block_group_flags(&cache->item, group_type);
3433 cache->flags = group_type;
3435 ret = update_space_info(root->fs_info, group_type, group_size,
3436 0, &cache->space_info);
3438 set_avail_alloc_bits(extent_root->fs_info, group_type);
3440 set_extent_bits(block_group_cache, cur_start,
3441 cur_start + group_size - 1,
3442 bit | EXTENT_LOCKED, GFP_NOFS);
3443 set_state_private(block_group_cache, cur_start,
3444 (unsigned long)cache);
3445 cur_start += group_size;
3447 /* then insert all the items */
3449 while(cur_start < total_bytes) {
3450 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3453 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3454 sizeof(cache->item));
3457 finish_current_insert(trans, extent_root);
3458 ret = del_pending_extents(trans, extent_root);
3461 cur_start = cache->key.objectid + cache->key.offset;
3466 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3467 struct btrfs_root *root,
3468 u64 bytenr, u64 num_bytes, int alloc,
3471 return update_block_group(trans, root, bytenr, num_bytes,
3476 * Just remove a block group item in extent tree
3477 * Caller should ensure the block group is empty and all space is pinned.
3478 * Or new tree block/data may be allocated into it.
3480 static int free_block_group_item(struct btrfs_trans_handle *trans,
3481 struct btrfs_fs_info *fs_info,
3482 u64 bytenr, u64 len)
3484 struct btrfs_path *path;
3485 struct btrfs_key key;
3486 struct btrfs_root *root = fs_info->extent_root;
3489 key.objectid = bytenr;
3491 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3493 path = btrfs_alloc_path();
3497 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3505 ret = btrfs_del_item(trans, root, path);
3507 btrfs_free_path(path);
3511 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3512 struct btrfs_fs_info *fs_info,
3513 u64 devid, u64 dev_offset)
3515 struct btrfs_root *root = fs_info->dev_root;
3516 struct btrfs_path *path;
3517 struct btrfs_key key;
3520 path = btrfs_alloc_path();
3524 key.objectid = devid;
3525 key.type = BTRFS_DEV_EXTENT_KEY;
3526 key.offset = dev_offset;
3528 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3536 ret = btrfs_del_item(trans, root, path);
3538 btrfs_free_path(path);
3542 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3543 struct btrfs_fs_info *fs_info,
3546 struct btrfs_chunk *chunk = NULL;
3547 struct btrfs_root *root= fs_info->chunk_root;
3548 struct btrfs_path *path;
3549 struct btrfs_key key;
3554 path = btrfs_alloc_path();
3558 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3559 key.type = BTRFS_CHUNK_ITEM_KEY;
3560 key.offset = chunk_offset;
3562 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3569 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3570 struct btrfs_chunk);
3571 num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3572 for (i = 0; i < num_stripes; i++) {
3573 ret = free_dev_extent_item(trans, fs_info,
3574 btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3575 btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3580 btrfs_free_path(path);
3584 static int free_system_chunk_item(struct btrfs_super_block *super,
3585 struct btrfs_key *key)
3587 struct btrfs_disk_key *disk_key;
3588 struct btrfs_key cpu_key;
3589 u32 array_size = btrfs_super_sys_array_size(super);
3590 char *ptr = (char *)super->sys_chunk_array;
3594 while (cur < btrfs_super_sys_array_size(super)) {
3595 struct btrfs_chunk *chunk;
3599 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3600 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3601 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3607 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3608 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3609 chunk_len = btrfs_chunk_item_size(num_stripes) +
3612 if (key->objectid == cpu_key.objectid &&
3613 key->offset == cpu_key.offset &&
3614 key->type == cpu_key.type) {
3615 memmove(ptr + cur, ptr + cur + chunk_len,
3616 array_size - cur - chunk_len);
3617 array_size -= chunk_len;
3618 btrfs_set_super_sys_array_size(super, array_size);
3629 static int free_chunk_item(struct btrfs_trans_handle *trans,
3630 struct btrfs_fs_info *fs_info,
3631 u64 bytenr, u64 len)
3633 struct btrfs_path *path;
3634 struct btrfs_key key;
3635 struct btrfs_root *root = fs_info->chunk_root;
3636 struct btrfs_chunk *chunk;
3640 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3641 key.offset = bytenr;
3642 key.type = BTRFS_CHUNK_ITEM_KEY;
3644 path = btrfs_alloc_path();
3648 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3655 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3656 struct btrfs_chunk);
3657 chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3659 ret = btrfs_del_item(trans, root, path);
3663 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3664 ret = free_system_chunk_item(fs_info->super_copy, &key);
3666 btrfs_free_path(path);
3670 static u64 get_dev_extent_len(struct map_lookup *map)
3674 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3675 case 0: /* Single */
3676 case BTRFS_BLOCK_GROUP_DUP:
3677 case BTRFS_BLOCK_GROUP_RAID1:
3680 case BTRFS_BLOCK_GROUP_RAID5:
3681 div = (map->num_stripes - 1);
3683 case BTRFS_BLOCK_GROUP_RAID6:
3684 div = (map->num_stripes - 2);
3686 case BTRFS_BLOCK_GROUP_RAID10:
3687 div = (map->num_stripes / map->sub_stripes);
3690 /* normally, read chunk security hook should handled it */
3693 return map->ce.size / div;
3696 /* free block group/chunk related caches */
3697 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3698 struct btrfs_fs_info *fs_info,
3699 u64 bytenr, u64 len)
3701 struct btrfs_block_group_cache *cache;
3702 struct cache_extent *ce;
3703 struct map_lookup *map;
3708 /* Free block group cache first */
3709 cache = btrfs_lookup_block_group(fs_info, bytenr);
3712 flags = cache->flags;
3713 if (cache->free_space_ctl) {
3714 btrfs_remove_free_space_cache(cache);
3715 kfree(cache->free_space_ctl);
3717 clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len,
3718 (unsigned int)-1, GFP_NOFS);
3719 ret = free_space_info(fs_info, flags, len, 0, NULL);
3724 /* Then free mapping info and dev usage info */
3725 ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3726 if (!ce || ce->start != bytenr) {
3730 map = container_of(ce, struct map_lookup, ce);
3731 for (i = 0; i < map->num_stripes; i++) {
3732 struct btrfs_device *device;
3734 device = map->stripes[i].dev;
3735 device->bytes_used -= get_dev_extent_len(map);
3736 ret = btrfs_update_device(trans, device);
3740 remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3747 * Fixup block accounting. The initial block accounting created by
3748 * make_block_groups isn't accuracy in this case.
3750 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3751 struct btrfs_root *root)
3757 struct btrfs_path path;
3758 struct btrfs_key key;
3759 struct extent_buffer *leaf;
3760 struct btrfs_block_group_cache *cache;
3761 struct btrfs_fs_info *fs_info = root->fs_info;
3763 root = root->fs_info->extent_root;
3765 while(extent_root_pending_ops(fs_info)) {
3766 ret = finish_current_insert(trans, root);
3769 ret = del_pending_extents(trans, root);
3775 cache = btrfs_lookup_first_block_group(fs_info, start);
3778 start = cache->key.objectid + cache->key.offset;
3779 btrfs_set_block_group_used(&cache->item, 0);
3780 cache->space_info->bytes_used = 0;
3781 set_extent_bits(&root->fs_info->block_group_cache,
3782 cache->key.objectid,
3783 cache->key.objectid + cache->key.offset -1,
3784 BLOCK_GROUP_DIRTY, GFP_NOFS);
3787 btrfs_init_path(&path);
3790 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
3791 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3796 leaf = path.nodes[0];
3797 slot = path.slots[0];
3798 if (slot >= btrfs_header_nritems(leaf)) {
3799 ret = btrfs_next_leaf(root, &path);
3804 leaf = path.nodes[0];
3805 slot = path.slots[0];
3807 btrfs_item_key_to_cpu(leaf, &key, slot);
3808 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3809 bytes_used += key.offset;
3810 ret = btrfs_update_block_group(trans, root,
3811 key.objectid, key.offset, 1, 0);
3813 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3814 bytes_used += root->leafsize;
3815 ret = btrfs_update_block_group(trans, root,
3816 key.objectid, root->leafsize, 1, 0);
3821 btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3822 btrfs_release_path(&path);
3827 * Record a file extent. Do all the required works, such as inserting
3828 * file extent item, inserting extent item and backref item into extent
3829 * tree and updating block accounting.
3831 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3832 struct btrfs_root *root, u64 objectid,
3833 struct btrfs_inode_item *inode,
3834 u64 file_pos, u64 disk_bytenr,
3838 struct btrfs_fs_info *info = root->fs_info;
3839 struct btrfs_root *extent_root = info->extent_root;
3840 struct extent_buffer *leaf;
3841 struct btrfs_file_extent_item *fi;
3842 struct btrfs_key ins_key;
3843 struct btrfs_path path;
3844 struct btrfs_extent_item *ei;
3847 if (disk_bytenr == 0) {
3848 ret = btrfs_insert_file_extent(trans, root, objectid,
3849 file_pos, disk_bytenr,
3850 num_bytes, num_bytes);
3854 btrfs_init_path(&path);
3856 ins_key.objectid = objectid;
3857 ins_key.offset = file_pos;
3858 btrfs_set_key_type(&ins_key, BTRFS_EXTENT_DATA_KEY);
3859 ret = btrfs_insert_empty_item(trans, root, &path, &ins_key,
3863 leaf = path.nodes[0];
3864 fi = btrfs_item_ptr(leaf, path.slots[0],
3865 struct btrfs_file_extent_item);
3866 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
3867 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
3868 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
3869 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
3870 btrfs_set_file_extent_offset(leaf, fi, 0);
3871 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
3872 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
3873 btrfs_set_file_extent_compression(leaf, fi, 0);
3874 btrfs_set_file_extent_encryption(leaf, fi, 0);
3875 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
3876 btrfs_mark_buffer_dirty(leaf);
3878 nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
3879 btrfs_set_stack_inode_nbytes(inode, nbytes);
3881 btrfs_release_path(&path);
3883 ins_key.objectid = disk_bytenr;
3884 ins_key.offset = num_bytes;
3885 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3887 ret = btrfs_insert_empty_item(trans, extent_root, &path,
3888 &ins_key, sizeof(*ei));
3890 leaf = path.nodes[0];
3891 ei = btrfs_item_ptr(leaf, path.slots[0],
3892 struct btrfs_extent_item);
3894 btrfs_set_extent_refs(leaf, ei, 0);
3895 btrfs_set_extent_generation(leaf, ei, 0);
3896 btrfs_set_extent_flags(leaf, ei, BTRFS_EXTENT_FLAG_DATA);
3898 btrfs_mark_buffer_dirty(leaf);
3900 ret = btrfs_update_block_group(trans, root, disk_bytenr,
3904 } else if (ret != -EEXIST) {
3907 btrfs_extent_post_op(trans, extent_root);
3909 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, num_bytes, 0,
3910 root->root_key.objectid,
3911 objectid, file_pos);
3916 btrfs_release_path(&path);