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.
23 #include "kerncompat.h"
24 #include "radix-tree.h"
27 #include "print-tree.h"
28 #include "transaction.h"
31 #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->nodesize;
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(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 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_path *path,
861 struct btrfs_extent_inline_ref *iref)
863 struct btrfs_key key;
864 struct extent_buffer *leaf;
865 struct btrfs_extent_data_ref *ref1;
866 struct btrfs_shared_data_ref *ref2;
869 leaf = path->nodes[0];
870 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
872 if (btrfs_extent_inline_ref_type(leaf, iref) ==
873 BTRFS_EXTENT_DATA_REF_KEY) {
874 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
875 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
877 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
878 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
880 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
881 ref1 = btrfs_item_ptr(leaf, path->slots[0],
882 struct btrfs_extent_data_ref);
883 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
884 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
885 ref2 = btrfs_item_ptr(leaf, path->slots[0],
886 struct btrfs_shared_data_ref);
887 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
889 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
890 struct btrfs_extent_ref_v0 *ref0;
891 ref0 = btrfs_item_ptr(leaf, path->slots[0],
892 struct btrfs_extent_ref_v0);
893 num_refs = btrfs_ref_count_v0(leaf, ref0);
901 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
902 struct btrfs_root *root,
903 struct btrfs_path *path,
904 u64 bytenr, u64 parent,
907 struct btrfs_key key;
910 key.objectid = bytenr;
912 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
915 key.type = BTRFS_TREE_BLOCK_REF_KEY;
916 key.offset = root_objectid;
919 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
922 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
923 if (ret == -ENOENT && parent) {
924 btrfs_release_path(path);
925 key.type = BTRFS_EXTENT_REF_V0_KEY;
926 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
934 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
935 struct btrfs_root *root,
936 struct btrfs_path *path,
937 u64 bytenr, u64 parent,
940 struct btrfs_key key;
943 key.objectid = bytenr;
945 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
948 key.type = BTRFS_TREE_BLOCK_REF_KEY;
949 key.offset = root_objectid;
952 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
954 btrfs_release_path(path);
958 static inline int extent_ref_type(u64 parent, u64 owner)
961 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
963 type = BTRFS_SHARED_BLOCK_REF_KEY;
965 type = BTRFS_TREE_BLOCK_REF_KEY;
968 type = BTRFS_SHARED_DATA_REF_KEY;
970 type = BTRFS_EXTENT_DATA_REF_KEY;
975 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root,
977 struct btrfs_path *path,
978 struct btrfs_extent_inline_ref **ref_ret,
979 u64 bytenr, u64 num_bytes,
980 u64 parent, u64 root_objectid,
981 u64 owner, u64 offset, int insert)
983 struct btrfs_key key;
984 struct extent_buffer *leaf;
985 struct btrfs_extent_item *ei;
986 struct btrfs_extent_inline_ref *iref;
996 int skinny_metadata =
997 btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
999 key.objectid = bytenr;
1000 key.type = BTRFS_EXTENT_ITEM_KEY;
1001 key.offset = num_bytes;
1003 want = extent_ref_type(parent, owner);
1005 extra_size = btrfs_extent_inline_ref_size(want);
1009 if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1010 skinny_metadata = 1;
1011 key.type = BTRFS_METADATA_ITEM_KEY;
1013 } else if (skinny_metadata) {
1014 skinny_metadata = 0;
1018 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1025 * We may be a newly converted file system which still has the old fat
1026 * extent entries for metadata, so try and see if we have one of those.
1028 if (ret > 0 && skinny_metadata) {
1029 skinny_metadata = 0;
1030 if (path->slots[0]) {
1032 btrfs_item_key_to_cpu(path->nodes[0], &key,
1034 if (key.objectid == bytenr &&
1035 key.type == BTRFS_EXTENT_ITEM_KEY &&
1036 key.offset == num_bytes)
1040 key.type = BTRFS_EXTENT_ITEM_KEY;
1041 key.offset = num_bytes;
1042 btrfs_release_path(path);
1048 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1054 leaf = path->nodes[0];
1055 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1057 if (item_size < sizeof(*ei)) {
1062 ret = convert_extent_item_v0(trans, root, path, owner,
1068 leaf = path->nodes[0];
1069 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1072 if (item_size < sizeof(*ei)) {
1073 printf("Size is %u, needs to be %u, slot %d\n",
1074 (unsigned)item_size,
1075 (unsigned)sizeof(*ei), path->slots[0]);
1076 btrfs_print_leaf(root, leaf);
1079 BUG_ON(item_size < sizeof(*ei));
1081 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1082 flags = btrfs_extent_flags(leaf, ei);
1084 ptr = (unsigned long)(ei + 1);
1085 end = (unsigned long)ei + item_size;
1087 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1088 ptr += sizeof(struct btrfs_tree_block_info);
1090 } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1091 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1102 iref = (struct btrfs_extent_inline_ref *)ptr;
1103 type = btrfs_extent_inline_ref_type(leaf, iref);
1107 ptr += btrfs_extent_inline_ref_size(type);
1111 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1112 struct btrfs_extent_data_ref *dref;
1113 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1114 if (match_extent_data_ref(leaf, dref, root_objectid,
1119 if (hash_extent_data_ref_item(leaf, dref) <
1120 hash_extent_data_ref(root_objectid, owner, offset))
1124 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1126 if (parent == ref_offset) {
1130 if (ref_offset < parent)
1133 if (root_objectid == ref_offset) {
1137 if (ref_offset < root_objectid)
1141 ptr += btrfs_extent_inline_ref_size(type);
1143 if (err == -ENOENT && insert) {
1144 if (item_size + extra_size >=
1145 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1150 * To add new inline back ref, we have to make sure
1151 * there is no corresponding back ref item.
1152 * For simplicity, we just do not add new inline back
1153 * ref if there is any back ref item.
1155 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1156 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1161 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1166 static int setup_inline_extent_backref(struct btrfs_root *root,
1167 struct btrfs_path *path,
1168 struct btrfs_extent_inline_ref *iref,
1169 u64 parent, u64 root_objectid,
1170 u64 owner, u64 offset, int refs_to_add)
1172 struct extent_buffer *leaf;
1173 struct btrfs_extent_item *ei;
1176 unsigned long item_offset;
1182 leaf = path->nodes[0];
1183 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1184 item_offset = (unsigned long)iref - (unsigned long)ei;
1186 type = extent_ref_type(parent, owner);
1187 size = btrfs_extent_inline_ref_size(type);
1189 ret = btrfs_extend_item(root, path, size);
1192 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1193 refs = btrfs_extent_refs(leaf, ei);
1194 refs += refs_to_add;
1195 btrfs_set_extent_refs(leaf, ei, refs);
1197 ptr = (unsigned long)ei + item_offset;
1198 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1199 if (ptr < end - size)
1200 memmove_extent_buffer(leaf, ptr + size, ptr,
1203 iref = (struct btrfs_extent_inline_ref *)ptr;
1204 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1205 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1206 struct btrfs_extent_data_ref *dref;
1207 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1208 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1209 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1210 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1211 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1212 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1213 struct btrfs_shared_data_ref *sref;
1214 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1215 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1216 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1217 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1218 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1220 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1222 btrfs_mark_buffer_dirty(leaf);
1226 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1227 struct btrfs_root *root,
1228 struct btrfs_path *path,
1229 struct btrfs_extent_inline_ref **ref_ret,
1230 u64 bytenr, u64 num_bytes, u64 parent,
1231 u64 root_objectid, u64 owner, u64 offset)
1235 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1236 bytenr, num_bytes, parent,
1237 root_objectid, owner, offset, 0);
1241 btrfs_release_path(path);
1244 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1245 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1248 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1249 root_objectid, owner, offset);
1254 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1255 struct btrfs_root *root,
1256 struct btrfs_path *path,
1257 struct btrfs_extent_inline_ref *iref,
1260 struct extent_buffer *leaf;
1261 struct btrfs_extent_item *ei;
1262 struct btrfs_extent_data_ref *dref = NULL;
1263 struct btrfs_shared_data_ref *sref = NULL;
1272 leaf = path->nodes[0];
1273 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1274 refs = btrfs_extent_refs(leaf, ei);
1275 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1276 refs += refs_to_mod;
1277 btrfs_set_extent_refs(leaf, ei, refs);
1279 type = btrfs_extent_inline_ref_type(leaf, iref);
1281 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1282 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1283 refs = btrfs_extent_data_ref_count(leaf, dref);
1284 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1285 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1286 refs = btrfs_shared_data_ref_count(leaf, sref);
1289 BUG_ON(refs_to_mod != -1);
1292 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1293 refs += refs_to_mod;
1296 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1297 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1299 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1301 size = btrfs_extent_inline_ref_size(type);
1302 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1303 ptr = (unsigned long)iref;
1304 end = (unsigned long)ei + item_size;
1305 if (ptr + size < end)
1306 memmove_extent_buffer(leaf, ptr, ptr + size,
1309 ret = btrfs_truncate_item(root, path, item_size, 1);
1312 btrfs_mark_buffer_dirty(leaf);
1316 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root,
1318 struct btrfs_path *path,
1319 u64 bytenr, u64 num_bytes, u64 parent,
1320 u64 root_objectid, u64 owner,
1321 u64 offset, int refs_to_add)
1323 struct btrfs_extent_inline_ref *iref;
1326 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1327 bytenr, num_bytes, parent,
1328 root_objectid, owner, offset, 1);
1330 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1331 ret = update_inline_extent_backref(trans, root, path, iref,
1333 } else if (ret == -ENOENT) {
1334 ret = setup_inline_extent_backref(root, path, iref,
1335 parent, root_objectid,
1336 owner, offset, refs_to_add);
1341 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1342 struct btrfs_root *root,
1343 struct btrfs_path *path,
1344 u64 bytenr, u64 parent, u64 root_objectid,
1345 u64 owner, u64 offset, int refs_to_add)
1349 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1350 ret = insert_extent_data_ref(trans, root, path, bytenr,
1351 parent, root_objectid,
1352 owner, offset, refs_to_add);
1354 BUG_ON(refs_to_add != 1);
1355 ret = insert_tree_block_ref(trans, root, path, bytenr,
1356 parent, root_objectid);
1361 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 struct btrfs_extent_inline_ref *iref,
1365 int refs_to_drop, int is_data)
1369 BUG_ON(!is_data && refs_to_drop != 1);
1371 ret = update_inline_extent_backref(trans, root, path, iref,
1373 } else if (is_data) {
1374 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1376 ret = btrfs_del_item(trans, root, path);
1381 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1382 struct btrfs_root *root,
1383 u64 bytenr, u64 num_bytes, u64 parent,
1384 u64 root_objectid, u64 owner, u64 offset)
1386 struct btrfs_path *path;
1387 struct extent_buffer *leaf;
1388 struct btrfs_extent_item *item;
1393 path = btrfs_alloc_path();
1399 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1400 path, bytenr, num_bytes, parent,
1401 root_objectid, owner, offset, 1);
1405 if (ret != -EAGAIN) {
1410 leaf = path->nodes[0];
1411 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1412 refs = btrfs_extent_refs(leaf, item);
1413 btrfs_set_extent_refs(leaf, item, refs + 1);
1415 btrfs_mark_buffer_dirty(leaf);
1416 btrfs_release_path(path);
1420 /* now insert the actual backref */
1421 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1422 path, bytenr, parent, root_objectid,
1427 btrfs_free_path(path);
1428 finish_current_insert(trans, root->fs_info->extent_root);
1429 del_pending_extents(trans, root->fs_info->extent_root);
1434 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1435 struct btrfs_root *root)
1437 finish_current_insert(trans, root->fs_info->extent_root);
1438 del_pending_extents(trans, root->fs_info->extent_root);
1442 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root, u64 bytenr,
1444 u64 offset, int metadata, u64 *refs, u64 *flags)
1446 struct btrfs_path *path;
1448 struct btrfs_key key;
1449 struct extent_buffer *l;
1450 struct btrfs_extent_item *item;
1456 !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
1457 offset = root->nodesize;
1461 path = btrfs_alloc_path();
1466 key.objectid = bytenr;
1467 key.offset = offset;
1469 key.type = BTRFS_METADATA_ITEM_KEY;
1471 key.type = BTRFS_EXTENT_ITEM_KEY;
1474 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1480 * Deal with the fact that we may have mixed SKINNY and normal refs. If
1481 * we didn't find what we wanted check and see if we have a normal ref
1482 * right next to us, or re-search if we are on the edge of the leaf just
1485 if (ret > 0 && metadata) {
1486 if (path->slots[0]) {
1488 btrfs_item_key_to_cpu(path->nodes[0], &key,
1490 if (key.objectid == bytenr &&
1491 key.type == BTRFS_EXTENT_ITEM_KEY &&
1492 key.offset == root->nodesize)
1497 btrfs_release_path(path);
1498 key.type = BTRFS_EXTENT_ITEM_KEY;
1499 key.offset = root->nodesize;
1511 item_size = btrfs_item_size_nr(l, path->slots[0]);
1512 if (item_size >= sizeof(*item)) {
1513 item = btrfs_item_ptr(l, path->slots[0],
1514 struct btrfs_extent_item);
1515 num_refs = btrfs_extent_refs(l, item);
1516 extent_flags = btrfs_extent_flags(l, item);
1518 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1519 struct btrfs_extent_item_v0 *ei0;
1520 BUG_ON(item_size != sizeof(*ei0));
1521 ei0 = btrfs_item_ptr(l, path->slots[0],
1522 struct btrfs_extent_item_v0);
1523 num_refs = btrfs_extent_refs_v0(l, ei0);
1524 /* FIXME: this isn't correct for data */
1525 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1530 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1534 *flags = extent_flags;
1536 btrfs_free_path(path);
1540 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1541 struct btrfs_root *root,
1542 u64 bytenr, int level, u64 flags)
1544 struct btrfs_path *path;
1546 struct btrfs_key key;
1547 struct extent_buffer *l;
1548 struct btrfs_extent_item *item;
1550 int skinny_metadata =
1551 btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
1553 path = btrfs_alloc_path();
1558 key.objectid = bytenr;
1559 if (skinny_metadata) {
1561 key.type = BTRFS_METADATA_ITEM_KEY;
1563 key.offset = root->nodesize;
1564 key.type = BTRFS_EXTENT_ITEM_KEY;
1568 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1573 if (ret > 0 && skinny_metadata) {
1574 skinny_metadata = 0;
1575 if (path->slots[0]) {
1577 btrfs_item_key_to_cpu(path->nodes[0], &key,
1579 if (key.objectid == bytenr &&
1580 key.offset == root->nodesize &&
1581 key.type == BTRFS_EXTENT_ITEM_KEY)
1585 btrfs_release_path(path);
1586 key.offset = root->nodesize;
1587 key.type = BTRFS_EXTENT_ITEM_KEY;
1593 btrfs_print_leaf(root, path->nodes[0]);
1594 printk("failed to find block number %Lu\n",
1595 (unsigned long long)bytenr);
1599 item_size = btrfs_item_size_nr(l, path->slots[0]);
1600 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1601 if (item_size < sizeof(*item)) {
1602 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1608 item_size = btrfs_item_size_nr(l, path->slots[0]);
1611 BUG_ON(item_size < sizeof(*item));
1612 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1613 flags |= btrfs_extent_flags(l, item);
1614 btrfs_set_extent_flags(l, item, flags);
1616 btrfs_free_path(path);
1617 finish_current_insert(trans, root->fs_info->extent_root);
1618 del_pending_extents(trans, root->fs_info->extent_root);
1622 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1623 struct btrfs_root *root,
1624 struct extent_buffer *buf,
1625 int record_parent, int inc)
1632 struct btrfs_key key;
1633 struct btrfs_file_extent_item *fi;
1637 int (*process_func)(struct btrfs_trans_handle *trans,
1638 struct btrfs_root *root,
1639 u64, u64, u64, u64, u64, u64);
1641 ref_root = btrfs_header_owner(buf);
1642 nritems = btrfs_header_nritems(buf);
1643 level = btrfs_header_level(buf);
1645 if (!root->ref_cows && level == 0)
1649 process_func = btrfs_inc_extent_ref;
1651 process_func = btrfs_free_extent;
1654 parent = buf->start;
1658 for (i = 0; i < nritems; i++) {
1661 btrfs_item_key_to_cpu(buf, &key, i);
1662 if (key.type != BTRFS_EXTENT_DATA_KEY)
1664 fi = btrfs_item_ptr(buf, i,
1665 struct btrfs_file_extent_item);
1666 if (btrfs_file_extent_type(buf, fi) ==
1667 BTRFS_FILE_EXTENT_INLINE)
1669 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1673 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1674 key.offset -= btrfs_file_extent_offset(buf, fi);
1675 ret = process_func(trans, root, bytenr, num_bytes,
1676 parent, ref_root, key.objectid,
1683 bytenr = btrfs_node_blockptr(buf, i);
1684 num_bytes = root->nodesize;
1685 ret = process_func(trans, root, bytenr, num_bytes,
1686 parent, ref_root, level - 1, 0);
1699 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1700 struct extent_buffer *buf, int record_parent)
1702 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1705 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1706 struct extent_buffer *buf, int record_parent)
1708 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1711 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1712 struct btrfs_root *root,
1713 struct btrfs_path *path,
1714 struct btrfs_block_group_cache *cache)
1718 struct btrfs_root *extent_root = root->fs_info->extent_root;
1720 struct extent_buffer *leaf;
1722 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1727 leaf = path->nodes[0];
1728 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1729 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1730 btrfs_mark_buffer_dirty(leaf);
1731 btrfs_release_path(path);
1733 finish_current_insert(trans, extent_root);
1734 pending_ret = del_pending_extents(trans, extent_root);
1743 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1744 struct btrfs_root *root)
1746 struct extent_io_tree *block_group_cache;
1747 struct btrfs_block_group_cache *cache;
1749 struct btrfs_path *path;
1755 block_group_cache = &root->fs_info->block_group_cache;
1756 path = btrfs_alloc_path();
1761 ret = find_first_extent_bit(block_group_cache, last,
1762 &start, &end, BLOCK_GROUP_DIRTY);
1771 ret = get_state_private(block_group_cache, start, &ptr);
1774 clear_extent_bits(block_group_cache, start, end,
1775 BLOCK_GROUP_DIRTY, GFP_NOFS);
1777 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1778 ret = write_one_cache_group(trans, root, path, cache);
1780 btrfs_free_path(path);
1784 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1787 struct btrfs_space_info *found;
1789 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1791 list_for_each_entry(found, &info->space_info, list) {
1792 if (found->flags & flags)
1799 static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
1800 u64 total_bytes, u64 bytes_used,
1801 struct btrfs_space_info **space_info)
1803 struct btrfs_space_info *found;
1805 /* only support free block group which is empty */
1809 found = __find_space_info(fs_info, flags);
1812 if (found->total_bytes < total_bytes) {
1814 "WARNING: bad space info to free %llu only have %llu\n",
1815 total_bytes, found->total_bytes);
1818 found->total_bytes -= total_bytes;
1820 *space_info = found;
1824 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1825 u64 total_bytes, u64 bytes_used,
1826 struct btrfs_space_info **space_info)
1828 struct btrfs_space_info *found;
1830 found = __find_space_info(info, flags);
1832 found->total_bytes += total_bytes;
1833 found->bytes_used += bytes_used;
1834 if (found->total_bytes < found->bytes_used) {
1835 fprintf(stderr, "warning, bad space info total_bytes "
1837 (unsigned long long)found->total_bytes,
1838 (unsigned long long)found->bytes_used);
1840 *space_info = found;
1843 found = kmalloc(sizeof(*found), GFP_NOFS);
1847 list_add(&found->list, &info->space_info);
1848 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1849 found->total_bytes = total_bytes;
1850 found->bytes_used = bytes_used;
1851 found->bytes_pinned = 0;
1853 *space_info = found;
1858 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1860 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1861 BTRFS_BLOCK_GROUP_RAID1 |
1862 BTRFS_BLOCK_GROUP_RAID10 |
1863 BTRFS_BLOCK_GROUP_RAID5 |
1864 BTRFS_BLOCK_GROUP_RAID6 |
1865 BTRFS_BLOCK_GROUP_DUP);
1867 if (flags & BTRFS_BLOCK_GROUP_DATA)
1868 fs_info->avail_data_alloc_bits |= extra_flags;
1869 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1870 fs_info->avail_metadata_alloc_bits |= extra_flags;
1871 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1872 fs_info->avail_system_alloc_bits |= extra_flags;
1876 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1877 struct btrfs_root *extent_root, u64 alloc_bytes,
1880 struct btrfs_space_info *space_info;
1886 space_info = __find_space_info(extent_root->fs_info, flags);
1888 ret = update_space_info(extent_root->fs_info, flags,
1892 BUG_ON(!space_info);
1894 if (space_info->full)
1897 thresh = div_factor(space_info->total_bytes, 7);
1898 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1903 * Avoid allocating given chunk type
1905 if (extent_root->fs_info->avoid_meta_chunk_alloc &&
1906 (flags & BTRFS_BLOCK_GROUP_METADATA))
1908 if (extent_root->fs_info->avoid_sys_chunk_alloc &&
1909 (flags & BTRFS_BLOCK_GROUP_SYSTEM))
1912 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1914 if (ret == -ENOSPC) {
1915 space_info->full = 1;
1921 ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1922 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1927 static int update_block_group(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *root,
1929 u64 bytenr, u64 num_bytes, int alloc,
1932 struct btrfs_block_group_cache *cache;
1933 struct btrfs_fs_info *info = root->fs_info;
1934 u64 total = num_bytes;
1940 /* block accounting for super block */
1941 old_val = btrfs_super_bytes_used(info->super_copy);
1943 old_val += num_bytes;
1945 old_val -= num_bytes;
1946 btrfs_set_super_bytes_used(info->super_copy, old_val);
1948 /* block accounting for root item */
1949 old_val = btrfs_root_used(&root->root_item);
1951 old_val += num_bytes;
1953 old_val -= num_bytes;
1954 btrfs_set_root_used(&root->root_item, old_val);
1957 cache = btrfs_lookup_block_group(info, bytenr);
1961 byte_in_group = bytenr - cache->key.objectid;
1962 WARN_ON(byte_in_group > cache->key.offset);
1963 start = cache->key.objectid;
1964 end = start + cache->key.offset - 1;
1965 set_extent_bits(&info->block_group_cache, start, end,
1966 BLOCK_GROUP_DIRTY, GFP_NOFS);
1968 old_val = btrfs_block_group_used(&cache->item);
1969 num_bytes = min(total, cache->key.offset - byte_in_group);
1972 old_val += num_bytes;
1973 cache->space_info->bytes_used += num_bytes;
1975 old_val -= num_bytes;
1976 cache->space_info->bytes_used -= num_bytes;
1978 set_extent_dirty(&info->free_space_cache,
1979 bytenr, bytenr + num_bytes - 1,
1983 btrfs_set_block_group_used(&cache->item, old_val);
1985 bytenr += num_bytes;
1990 static int update_pinned_extents(struct btrfs_root *root,
1991 u64 bytenr, u64 num, int pin)
1994 struct btrfs_block_group_cache *cache;
1995 struct btrfs_fs_info *fs_info = root->fs_info;
1998 set_extent_dirty(&fs_info->pinned_extents,
1999 bytenr, bytenr + num - 1, GFP_NOFS);
2001 clear_extent_dirty(&fs_info->pinned_extents,
2002 bytenr, bytenr + num - 1, GFP_NOFS);
2005 cache = btrfs_lookup_block_group(fs_info, bytenr);
2007 len = min((u64)root->sectorsize, num);
2011 len = min(num, cache->key.offset -
2012 (bytenr - cache->key.objectid));
2014 cache->pinned += len;
2015 cache->space_info->bytes_pinned += len;
2016 fs_info->total_pinned += len;
2018 cache->pinned -= len;
2019 cache->space_info->bytes_pinned -= len;
2020 fs_info->total_pinned -= len;
2029 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2030 struct btrfs_root *root,
2031 struct extent_io_tree *unpin)
2036 struct extent_io_tree *free_space_cache;
2037 free_space_cache = &root->fs_info->free_space_cache;
2040 ret = find_first_extent_bit(unpin, 0, &start, &end,
2044 update_pinned_extents(root, start, end + 1 - start, 0);
2045 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2046 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2051 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2057 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2058 &end, EXTENT_LOCKED);
2060 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2066 static int finish_current_insert(struct btrfs_trans_handle *trans,
2067 struct btrfs_root *extent_root)
2072 struct btrfs_fs_info *info = extent_root->fs_info;
2073 struct pending_extent_op *extent_op;
2074 struct btrfs_key key;
2076 int skinny_metadata =
2077 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2080 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2081 &end, EXTENT_LOCKED);
2085 ret = get_state_private(&info->extent_ins, start, &priv);
2087 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2089 if (extent_op->type == PENDING_EXTENT_INSERT) {
2090 key.objectid = start;
2091 if (skinny_metadata) {
2092 key.offset = extent_op->level;
2093 key.type = BTRFS_METADATA_ITEM_KEY;
2095 key.offset = extent_op->num_bytes;
2096 key.type = BTRFS_EXTENT_ITEM_KEY;
2098 ret = alloc_reserved_tree_block(trans, extent_root,
2099 extent_root->root_key.objectid,
2103 extent_op->level, &key);
2109 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2116 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2117 struct btrfs_root *root,
2118 u64 bytenr, u64 num_bytes, int is_data)
2121 struct extent_buffer *buf;
2126 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2130 /* we can reuse a block if it hasn't been written
2131 * and it is from this transaction. We can't
2132 * reuse anything from the tree log root because
2133 * it has tiny sub-transactions.
2135 if (btrfs_buffer_uptodate(buf, 0)) {
2136 u64 header_owner = btrfs_header_owner(buf);
2137 u64 header_transid = btrfs_header_generation(buf);
2138 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2139 header_transid == trans->transid &&
2140 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2141 clean_tree_block(NULL, root, buf);
2142 free_extent_buffer(buf);
2146 free_extent_buffer(buf);
2148 update_pinned_extents(root, bytenr, num_bytes, 1);
2154 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2155 u64 bytenr, u64 num_bytes)
2157 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2160 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2161 u64 bytenr, u64 num_bytes)
2163 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2167 * remove an extent from the root, returns 0 on success
2169 static int __free_extent(struct btrfs_trans_handle *trans,
2170 struct btrfs_root *root,
2171 u64 bytenr, u64 num_bytes, u64 parent,
2172 u64 root_objectid, u64 owner_objectid,
2173 u64 owner_offset, int refs_to_drop)
2176 struct btrfs_key key;
2177 struct btrfs_path *path;
2178 struct btrfs_root *extent_root = root->fs_info->extent_root;
2179 struct extent_buffer *leaf;
2180 struct btrfs_extent_item *ei;
2181 struct btrfs_extent_inline_ref *iref;
2184 int extent_slot = 0;
2185 int found_extent = 0;
2189 int skinny_metadata =
2190 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2192 if (root->fs_info->free_extent_hook) {
2193 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2194 parent, root_objectid, owner_objectid,
2195 owner_offset, refs_to_drop);
2198 path = btrfs_alloc_path();
2204 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2206 skinny_metadata = 0;
2207 BUG_ON(!is_data && refs_to_drop != 1);
2209 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2210 bytenr, num_bytes, parent,
2211 root_objectid, owner_objectid,
2214 extent_slot = path->slots[0];
2215 while (extent_slot >= 0) {
2216 btrfs_item_key_to_cpu(path->nodes[0], &key,
2218 if (key.objectid != bytenr)
2220 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2221 key.offset == num_bytes) {
2225 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2226 key.offset == owner_objectid) {
2230 if (path->slots[0] - extent_slot > 5)
2234 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2235 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2236 if (found_extent && item_size < sizeof(*ei))
2239 if (!found_extent) {
2241 ret = remove_extent_backref(trans, extent_root, path,
2245 btrfs_release_path(path);
2247 key.objectid = bytenr;
2249 if (skinny_metadata) {
2250 key.type = BTRFS_METADATA_ITEM_KEY;
2251 key.offset = owner_objectid;
2253 key.type = BTRFS_EXTENT_ITEM_KEY;
2254 key.offset = num_bytes;
2257 ret = btrfs_search_slot(trans, extent_root,
2259 if (ret > 0 && skinny_metadata && path->slots[0]) {
2261 btrfs_item_key_to_cpu(path->nodes[0],
2264 if (key.objectid == bytenr &&
2265 key.type == BTRFS_EXTENT_ITEM_KEY &&
2266 key.offset == num_bytes)
2270 if (ret > 0 && skinny_metadata) {
2271 skinny_metadata = 0;
2272 btrfs_release_path(path);
2273 key.type = BTRFS_EXTENT_ITEM_KEY;
2274 key.offset = num_bytes;
2275 ret = btrfs_search_slot(trans, extent_root,
2280 printk(KERN_ERR "umm, got %d back from search"
2281 ", was looking for %llu\n", ret,
2282 (unsigned long long)bytenr);
2283 btrfs_print_leaf(extent_root, path->nodes[0]);
2286 extent_slot = path->slots[0];
2289 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2290 "parent %llu root %llu owner %llu offset %llu\n",
2291 (unsigned long long)bytenr,
2292 (unsigned long long)parent,
2293 (unsigned long long)root_objectid,
2294 (unsigned long long)owner_objectid,
2295 (unsigned long long)owner_offset);
2300 leaf = path->nodes[0];
2301 item_size = btrfs_item_size_nr(leaf, extent_slot);
2302 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2303 if (item_size < sizeof(*ei)) {
2304 BUG_ON(found_extent || extent_slot != path->slots[0]);
2305 ret = convert_extent_item_v0(trans, extent_root, path,
2309 btrfs_release_path(path);
2311 key.objectid = bytenr;
2312 key.type = BTRFS_EXTENT_ITEM_KEY;
2313 key.offset = num_bytes;
2315 ret = btrfs_search_slot(trans, extent_root, &key, path,
2318 printk(KERN_ERR "umm, got %d back from search"
2319 ", was looking for %llu\n", ret,
2320 (unsigned long long)bytenr);
2321 btrfs_print_leaf(extent_root, path->nodes[0]);
2324 extent_slot = path->slots[0];
2325 leaf = path->nodes[0];
2326 item_size = btrfs_item_size_nr(leaf, extent_slot);
2329 BUG_ON(item_size < sizeof(*ei));
2330 ei = btrfs_item_ptr(leaf, extent_slot,
2331 struct btrfs_extent_item);
2332 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2333 key.type == BTRFS_EXTENT_ITEM_KEY) {
2334 struct btrfs_tree_block_info *bi;
2335 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2336 bi = (struct btrfs_tree_block_info *)(ei + 1);
2337 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2340 refs = btrfs_extent_refs(leaf, ei);
2341 BUG_ON(refs < refs_to_drop);
2342 refs -= refs_to_drop;
2346 * In the case of inline back ref, reference count will
2347 * be updated by remove_extent_backref
2350 BUG_ON(!found_extent);
2352 btrfs_set_extent_refs(leaf, ei, refs);
2353 btrfs_mark_buffer_dirty(leaf);
2356 ret = remove_extent_backref(trans, extent_root, path,
2366 BUG_ON(is_data && refs_to_drop !=
2367 extent_data_ref_count(path, iref));
2369 BUG_ON(path->slots[0] != extent_slot);
2371 BUG_ON(path->slots[0] != extent_slot + 1);
2372 path->slots[0] = extent_slot;
2378 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2385 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2388 btrfs_release_path(path);
2391 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2395 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2398 btrfs_free_path(path);
2399 finish_current_insert(trans, extent_root);
2404 * find all the blocks marked as pending in the radix tree and remove
2405 * them from the extent map
2407 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2408 btrfs_root *extent_root)
2415 struct extent_io_tree *pending_del;
2416 struct extent_io_tree *extent_ins;
2417 struct pending_extent_op *extent_op;
2419 extent_ins = &extent_root->fs_info->extent_ins;
2420 pending_del = &extent_root->fs_info->pending_del;
2423 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2428 ret = get_state_private(pending_del, start, &priv);
2430 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2432 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2435 if (!test_range_bit(extent_ins, start, end,
2436 EXTENT_LOCKED, 0)) {
2437 ret = __free_extent(trans, extent_root,
2438 start, end + 1 - start, 0,
2439 extent_root->root_key.objectid,
2440 extent_op->level, 0, 1);
2444 ret = get_state_private(extent_ins, start, &priv);
2446 extent_op = (struct pending_extent_op *)
2447 (unsigned long)priv;
2449 clear_extent_bits(extent_ins, start, end,
2450 EXTENT_LOCKED, GFP_NOFS);
2452 if (extent_op->type == PENDING_BACKREF_UPDATE)
2464 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
2465 struct btrfs_root *root,
2466 struct extent_buffer *buf,
2467 u64 parent, int last_ref)
2469 return btrfs_free_extent(trans, root, buf->start, buf->len, parent,
2470 root->root_key.objectid,
2471 btrfs_header_level(buf), 0);
2475 * remove an extent from the root, returns 0 on success
2478 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2479 struct btrfs_root *root,
2480 u64 bytenr, u64 num_bytes, u64 parent,
2481 u64 root_objectid, u64 owner, u64 offset)
2483 struct btrfs_root *extent_root = root->fs_info->extent_root;
2487 WARN_ON(num_bytes < root->sectorsize);
2488 if (root == extent_root) {
2489 struct pending_extent_op *extent_op;
2491 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2494 extent_op->type = PENDING_EXTENT_DELETE;
2495 extent_op->bytenr = bytenr;
2496 extent_op->num_bytes = num_bytes;
2497 extent_op->level = (int)owner;
2499 set_extent_bits(&root->fs_info->pending_del,
2500 bytenr, bytenr + num_bytes - 1,
2501 EXTENT_LOCKED, GFP_NOFS);
2502 set_state_private(&root->fs_info->pending_del,
2503 bytenr, (unsigned long)extent_op);
2506 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2507 root_objectid, owner, offset, 1);
2508 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2509 return ret ? ret : pending_ret;
2512 static u64 stripe_align(struct btrfs_root *root, u64 val)
2514 u64 mask = ((u64)root->stripesize - 1);
2515 u64 ret = (val + mask) & ~mask;
2520 * walks the btree of allocated extents and find a hole of a given size.
2521 * The key ins is changed to record the hole:
2522 * ins->objectid == block start
2523 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2524 * ins->offset == number of blocks
2525 * Any available blocks before search_start are skipped.
2527 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2528 struct btrfs_root *orig_root,
2529 u64 num_bytes, u64 empty_size,
2530 u64 search_start, u64 search_end,
2531 u64 hint_byte, struct btrfs_key *ins,
2532 u64 exclude_start, u64 exclude_nr,
2536 u64 orig_search_start = search_start;
2537 struct btrfs_root * root = orig_root->fs_info->extent_root;
2538 struct btrfs_fs_info *info = root->fs_info;
2539 u64 total_needed = num_bytes;
2540 struct btrfs_block_group_cache *block_group;
2544 WARN_ON(num_bytes < root->sectorsize);
2545 ins->type = BTRFS_EXTENT_ITEM_KEY;
2547 search_start = stripe_align(root, search_start);
2550 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2552 hint_byte = search_start;
2553 block_group = btrfs_find_block_group(root, block_group,
2554 hint_byte, data, 1);
2556 block_group = btrfs_find_block_group(root,
2558 search_start, data, 1);
2561 total_needed += empty_size;
2564 search_start = stripe_align(root, search_start);
2566 block_group = btrfs_lookup_first_block_group(info,
2569 block_group = btrfs_lookup_first_block_group(info,
2572 ret = find_search_start(root, &block_group, &search_start,
2573 total_needed, data);
2577 ins->objectid = search_start;
2578 ins->offset = num_bytes;
2580 if (ins->objectid + num_bytes >
2581 block_group->key.objectid + block_group->key.offset) {
2582 search_start = block_group->key.objectid +
2583 block_group->key.offset;
2587 if (test_range_bit(&info->extent_ins, ins->objectid,
2588 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2589 search_start = ins->objectid + num_bytes;
2593 if (test_range_bit(&info->pinned_extents, ins->objectid,
2594 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2595 search_start = ins->objectid + num_bytes;
2599 if (info->excluded_extents &&
2600 test_range_bit(info->excluded_extents, ins->objectid,
2601 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2602 search_start = ins->objectid + num_bytes;
2606 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2607 ins->objectid < exclude_start + exclude_nr)) {
2608 search_start = exclude_start + exclude_nr;
2612 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2613 if (check_crossing_stripes(info, ins->objectid, num_bytes)) {
2614 struct btrfs_block_group_cache *bg_cache;
2617 bg_cache = btrfs_lookup_block_group(info, ins->objectid);
2620 bg_offset = ins->objectid - bg_cache->key.objectid;
2622 search_start = round_up(bg_offset + num_bytes,
2623 BTRFS_STRIPE_LEN) + bg_offset;
2627 block_group = btrfs_lookup_block_group(info, ins->objectid);
2629 trans->block_group = block_group;
2631 ins->offset = num_bytes;
2635 block_group = btrfs_lookup_first_block_group(info, search_start);
2637 search_start = orig_search_start;
2644 total_needed -= empty_size;
2650 block_group = btrfs_find_block_group(root, block_group,
2651 search_start, data, 0);
2658 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root,
2660 u64 num_bytes, u64 empty_size,
2661 u64 hint_byte, u64 search_end,
2662 struct btrfs_key *ins, int data)
2665 u64 search_start = 0;
2667 struct btrfs_fs_info *info = root->fs_info;
2670 alloc_profile = info->avail_data_alloc_bits &
2671 info->data_alloc_profile;
2672 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2673 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2674 info->system_allocs >= 0) {
2675 alloc_profile = info->avail_system_alloc_bits &
2676 info->system_alloc_profile;
2677 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2679 alloc_profile = info->avail_metadata_alloc_bits &
2680 info->metadata_alloc_profile;
2681 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2684 if (root->ref_cows) {
2685 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2686 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2688 BTRFS_BLOCK_GROUP_METADATA);
2691 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2692 num_bytes + SZ_2M, data);
2696 WARN_ON(num_bytes < root->sectorsize);
2697 ret = find_free_extent(trans, root, num_bytes, empty_size,
2698 search_start, search_end, hint_byte, ins,
2699 trans->alloc_exclude_start,
2700 trans->alloc_exclude_nr, data);
2702 clear_extent_dirty(&root->fs_info->free_space_cache,
2703 ins->objectid, ins->objectid + ins->offset - 1,
2708 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2709 struct btrfs_root *root,
2710 u64 root_objectid, u64 generation,
2711 u64 flags, struct btrfs_disk_key *key,
2712 int level, struct btrfs_key *ins)
2715 struct btrfs_fs_info *fs_info = root->fs_info;
2716 struct btrfs_extent_item *extent_item;
2717 struct btrfs_tree_block_info *block_info;
2718 struct btrfs_extent_inline_ref *iref;
2719 struct btrfs_path *path;
2720 struct extent_buffer *leaf;
2721 u32 size = sizeof(*extent_item) + sizeof(*iref);
2722 int skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2724 if (!skinny_metadata)
2725 size += sizeof(*block_info);
2727 path = btrfs_alloc_path();
2731 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2735 leaf = path->nodes[0];
2736 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2737 struct btrfs_extent_item);
2738 btrfs_set_extent_refs(leaf, extent_item, 1);
2739 btrfs_set_extent_generation(leaf, extent_item, generation);
2740 btrfs_set_extent_flags(leaf, extent_item,
2741 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2743 if (skinny_metadata) {
2744 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2746 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2747 btrfs_set_tree_block_key(leaf, block_info, key);
2748 btrfs_set_tree_block_level(leaf, block_info, level);
2749 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2752 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2753 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2755 btrfs_mark_buffer_dirty(leaf);
2756 btrfs_free_path(path);
2758 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
2763 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root, u64 num_bytes,
2765 u64 root_objectid, u64 generation,
2766 u64 flags, struct btrfs_disk_key *key,
2767 int level, u64 empty_size, u64 hint_byte,
2768 u64 search_end, struct btrfs_key *ins)
2771 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2772 hint_byte, search_end, ins, 0);
2775 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2776 struct pending_extent_op *extent_op;
2778 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2781 extent_op->type = PENDING_EXTENT_INSERT;
2782 extent_op->bytenr = ins->objectid;
2783 extent_op->num_bytes = ins->offset;
2784 extent_op->level = level;
2785 extent_op->flags = flags;
2786 memcpy(&extent_op->key, key, sizeof(*key));
2788 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2789 ins->objectid + ins->offset - 1,
2790 EXTENT_LOCKED, GFP_NOFS);
2791 set_state_private(&root->fs_info->extent_ins,
2792 ins->objectid, (unsigned long)extent_op);
2794 if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
2795 ins->offset = level;
2796 ins->type = BTRFS_METADATA_ITEM_KEY;
2798 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2801 finish_current_insert(trans, root->fs_info->extent_root);
2802 del_pending_extents(trans, root->fs_info->extent_root);
2808 * helper function to allocate a block for a given tree
2809 * returns the tree buffer or NULL.
2811 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2812 struct btrfs_root *root,
2813 u32 blocksize, u64 root_objectid,
2814 struct btrfs_disk_key *key, int level,
2815 u64 hint, u64 empty_size)
2817 struct btrfs_key ins;
2819 struct extent_buffer *buf;
2821 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2822 trans->transid, 0, key, level,
2823 empty_size, hint, (u64)-1, &ins);
2826 return ERR_PTR(ret);
2829 buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid,
2832 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2833 0, root->root_key.objectid, level, 0);
2835 return ERR_PTR(-ENOMEM);
2837 btrfs_set_buffer_uptodate(buf);
2838 trans->blocks_used++;
2845 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2846 struct btrfs_root *root,
2847 struct extent_buffer *leaf)
2850 u64 leaf_generation;
2851 struct btrfs_key key;
2852 struct btrfs_file_extent_item *fi;
2857 BUG_ON(!btrfs_is_leaf(leaf));
2858 nritems = btrfs_header_nritems(leaf);
2859 leaf_owner = btrfs_header_owner(leaf);
2860 leaf_generation = btrfs_header_generation(leaf);
2862 for (i = 0; i < nritems; i++) {
2865 btrfs_item_key_to_cpu(leaf, &key, i);
2866 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2868 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2869 if (btrfs_file_extent_type(leaf, fi) ==
2870 BTRFS_FILE_EXTENT_INLINE)
2873 * FIXME make sure to insert a trans record that
2874 * repeats the snapshot del on crash
2876 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2877 if (disk_bytenr == 0)
2879 ret = btrfs_free_extent(trans, root, disk_bytenr,
2880 btrfs_file_extent_disk_num_bytes(leaf, fi),
2881 leaf->start, leaf_owner, leaf_generation,
2888 static void noinline reada_walk_down(struct btrfs_root *root,
2889 struct extent_buffer *node,
2902 nritems = btrfs_header_nritems(node);
2903 level = btrfs_header_level(node);
2907 for (i = slot; i < nritems && skipped < 32; i++) {
2908 bytenr = btrfs_node_blockptr(node, i);
2909 if (last && ((bytenr > last && bytenr - last > SZ_32K) ||
2910 (last > bytenr && last - bytenr > SZ_32K))) {
2914 blocksize = btrfs_level_size(root, level - 1);
2916 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2924 mutex_unlock(&root->fs_info->fs_mutex);
2925 ret = readahead_tree_block(root, bytenr, blocksize,
2926 btrfs_node_ptr_generation(node, i));
2927 last = bytenr + blocksize;
2929 mutex_lock(&root->fs_info->fs_mutex);
2936 * helper function for drop_snapshot, this walks down the tree dropping ref
2937 * counts as it goes.
2939 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2940 struct btrfs_root *root,
2941 struct btrfs_path *path, int *level)
2947 struct extent_buffer *next;
2948 struct extent_buffer *cur;
2949 struct extent_buffer *parent;
2954 WARN_ON(*level < 0);
2955 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2956 ret = btrfs_lookup_extent_ref(trans, root,
2957 path->nodes[*level]->start,
2958 path->nodes[*level]->len, &refs);
2964 * walk down to the last node level and free all the leaves
2966 while(*level >= 0) {
2967 WARN_ON(*level < 0);
2968 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2969 cur = path->nodes[*level];
2971 if (btrfs_header_level(cur) != *level)
2974 if (path->slots[*level] >=
2975 btrfs_header_nritems(cur))
2978 ret = drop_leaf_ref(trans, root, cur);
2982 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2983 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2984 blocksize = btrfs_level_size(root, *level - 1);
2985 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2989 parent = path->nodes[*level];
2990 root_owner = btrfs_header_owner(parent);
2991 root_gen = btrfs_header_generation(parent);
2992 path->slots[*level]++;
2993 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2994 parent->start, root_owner,
2995 root_gen, *level - 1, 1);
2999 next = btrfs_find_tree_block(root, bytenr, blocksize);
3000 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3001 free_extent_buffer(next);
3002 reada_walk_down(root, cur, path->slots[*level]);
3003 mutex_unlock(&root->fs_info->fs_mutex);
3004 next = read_tree_block(root, bytenr, blocksize,
3006 mutex_lock(&root->fs_info->fs_mutex);
3007 if (!extent_buffer_uptodate(next)) {
3009 ret = PTR_ERR(next);
3015 WARN_ON(*level <= 0);
3016 if (path->nodes[*level-1])
3017 free_extent_buffer(path->nodes[*level-1]);
3018 path->nodes[*level-1] = next;
3019 *level = btrfs_header_level(next);
3020 path->slots[*level] = 0;
3023 WARN_ON(*level < 0);
3024 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3026 if (path->nodes[*level] == root->node) {
3027 root_owner = root->root_key.objectid;
3028 parent = path->nodes[*level];
3030 parent = path->nodes[*level + 1];
3031 root_owner = btrfs_header_owner(parent);
3034 root_gen = btrfs_header_generation(parent);
3035 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3036 path->nodes[*level]->len, parent->start,
3037 root_owner, root_gen, *level, 1);
3038 free_extent_buffer(path->nodes[*level]);
3039 path->nodes[*level] = NULL;
3046 * helper for dropping snapshots. This walks back up the tree in the path
3047 * to find the first node higher up where we haven't yet gone through
3050 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3051 struct btrfs_root *root,
3052 struct btrfs_path *path, int *level)
3056 struct btrfs_root_item *root_item = &root->root_item;
3061 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3062 slot = path->slots[i];
3063 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3064 struct extent_buffer *node;
3065 struct btrfs_disk_key disk_key;
3066 node = path->nodes[i];
3069 WARN_ON(*level == 0);
3070 btrfs_node_key(node, &disk_key, path->slots[i]);
3071 memcpy(&root_item->drop_progress,
3072 &disk_key, sizeof(disk_key));
3073 root_item->drop_level = i;
3076 struct extent_buffer *parent;
3077 if (path->nodes[*level] == root->node)
3078 parent = path->nodes[*level];
3080 parent = path->nodes[*level + 1];
3082 root_owner = btrfs_header_owner(parent);
3083 root_gen = btrfs_header_generation(parent);
3084 ret = btrfs_free_extent(trans, root,
3085 path->nodes[*level]->start,
3086 path->nodes[*level]->len,
3087 parent->start, root_owner,
3088 root_gen, *level, 1);
3090 free_extent_buffer(path->nodes[*level]);
3091 path->nodes[*level] = NULL;
3100 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3102 struct btrfs_space_info *sinfo;
3103 struct btrfs_block_group_cache *cache;
3110 ret = find_first_extent_bit(&info->block_group_cache, 0,
3111 &start, &end, (unsigned int)-1);
3114 ret = get_state_private(&info->block_group_cache, start, &ptr);
3116 cache = u64_to_ptr(ptr);
3117 if (cache->free_space_ctl) {
3118 btrfs_remove_free_space_cache(cache);
3119 kfree(cache->free_space_ctl);
3123 clear_extent_bits(&info->block_group_cache, start,
3124 end, (unsigned int)-1, GFP_NOFS);
3127 ret = find_first_extent_bit(&info->free_space_cache, 0,
3128 &start, &end, EXTENT_DIRTY);
3131 clear_extent_dirty(&info->free_space_cache, start,
3135 while (!list_empty(&info->space_info)) {
3136 sinfo = list_entry(info->space_info.next,
3137 struct btrfs_space_info, list);
3138 list_del_init(&sinfo->list);
3144 static int find_first_block_group(struct btrfs_root *root,
3145 struct btrfs_path *path, struct btrfs_key *key)
3148 struct btrfs_key found_key;
3149 struct extent_buffer *leaf;
3152 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3156 slot = path->slots[0];
3157 leaf = path->nodes[0];
3158 if (slot >= btrfs_header_nritems(leaf)) {
3159 ret = btrfs_next_leaf(root, path);
3166 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3168 if (found_key.objectid >= key->objectid &&
3169 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3178 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3179 struct btrfs_block_group_cache *cache)
3186 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3187 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3188 cache->bytes_super += stripe_len;
3191 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3192 bytenr = btrfs_sb_offset(i);
3193 ret = btrfs_rmap_block(&fs_info->mapping_tree,
3194 cache->key.objectid, bytenr,
3195 0, &logical, &nr, &stripe_len);
3202 if (logical[nr] > cache->key.objectid +
3206 if (logical[nr] + stripe_len <= cache->key.objectid)
3209 start = logical[nr];
3210 if (start < cache->key.objectid) {
3211 start = cache->key.objectid;
3212 len = (logical[nr] + stripe_len) - start;
3214 len = min_t(u64, stripe_len,
3215 cache->key.objectid +
3216 cache->key.offset - start);
3219 cache->bytes_super += len;
3226 int btrfs_read_block_groups(struct btrfs_root *root)
3228 struct btrfs_path *path;
3231 struct btrfs_block_group_cache *cache;
3232 struct btrfs_fs_info *info = root->fs_info;
3233 struct btrfs_space_info *space_info;
3234 struct extent_io_tree *block_group_cache;
3235 struct btrfs_key key;
3236 struct btrfs_key found_key;
3237 struct extent_buffer *leaf;
3239 block_group_cache = &info->block_group_cache;
3241 root = info->extent_root;
3244 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3245 path = btrfs_alloc_path();
3250 ret = find_first_block_group(root, path, &key);
3258 leaf = path->nodes[0];
3259 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3260 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3266 read_extent_buffer(leaf, &cache->item,
3267 btrfs_item_ptr_offset(leaf, path->slots[0]),
3268 sizeof(cache->item));
3269 memcpy(&cache->key, &found_key, sizeof(found_key));
3272 key.objectid = found_key.objectid + found_key.offset;
3273 if (found_key.offset == 0)
3275 btrfs_release_path(path);
3276 cache->flags = btrfs_block_group_flags(&cache->item);
3278 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3279 bit = BLOCK_GROUP_DATA;
3280 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3281 bit = BLOCK_GROUP_SYSTEM;
3282 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3283 bit = BLOCK_GROUP_METADATA;
3285 set_avail_alloc_bits(info, cache->flags);
3286 if (btrfs_chunk_readonly(root, cache->key.objectid))
3289 account_super_bytes(info, cache);
3291 ret = update_space_info(info, cache->flags, found_key.offset,
3292 btrfs_block_group_used(&cache->item),
3295 cache->space_info = space_info;
3297 /* use EXTENT_LOCKED to prevent merging */
3298 set_extent_bits(block_group_cache, found_key.objectid,
3299 found_key.objectid + found_key.offset - 1,
3300 bit | EXTENT_LOCKED, GFP_NOFS);
3301 set_state_private(block_group_cache, found_key.objectid,
3302 (unsigned long)cache);
3306 btrfs_free_path(path);
3310 struct btrfs_block_group_cache *
3311 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3312 u64 chunk_objectid, u64 chunk_offset, u64 size)
3316 struct btrfs_block_group_cache *cache;
3317 struct extent_io_tree *block_group_cache;
3319 block_group_cache = &fs_info->block_group_cache;
3321 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3323 cache->key.objectid = chunk_offset;
3324 cache->key.offset = size;
3326 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3327 btrfs_set_block_group_used(&cache->item, bytes_used);
3328 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3329 cache->flags = type;
3330 btrfs_set_block_group_flags(&cache->item, type);
3332 account_super_bytes(fs_info, cache);
3333 ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3334 &cache->space_info);
3337 bit = block_group_state_bits(type);
3338 ret = set_extent_bits(block_group_cache, chunk_offset,
3339 chunk_offset + size - 1,
3340 bit | EXTENT_LOCKED, GFP_NOFS);
3343 ret = set_state_private(block_group_cache, chunk_offset,
3344 (unsigned long)cache);
3346 set_avail_alloc_bits(fs_info, type);
3351 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3352 struct btrfs_root *root, u64 bytes_used,
3353 u64 type, u64 chunk_objectid, u64 chunk_offset,
3357 struct btrfs_root *extent_root;
3358 struct btrfs_block_group_cache *cache;
3360 cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3361 chunk_objectid, chunk_offset, size);
3362 extent_root = root->fs_info->extent_root;
3363 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3364 sizeof(cache->item));
3367 ret = finish_current_insert(trans, extent_root);
3369 ret = del_pending_extents(trans, extent_root);
3376 * This is for converter use only.
3378 * In that case, we don't know where are free blocks located.
3379 * Therefore all block group cache entries must be setup properly
3380 * before doing any block allocation.
3382 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3383 struct btrfs_root *root)
3391 u64 total_metadata = 0;
3395 struct btrfs_root *extent_root;
3396 struct btrfs_block_group_cache *cache;
3397 struct extent_io_tree *block_group_cache;
3399 extent_root = root->fs_info->extent_root;
3400 block_group_cache = &root->fs_info->block_group_cache;
3401 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3402 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3403 group_align = 64 * root->sectorsize;
3406 while (cur_start < total_bytes) {
3407 group_size = total_bytes / 12;
3408 group_size = min_t(u64, group_size, total_bytes - cur_start);
3409 if (cur_start == 0) {
3410 bit = BLOCK_GROUP_SYSTEM;
3411 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3413 group_size &= ~(group_align - 1);
3414 group_size = max_t(u64, group_size, SZ_8M);
3415 group_size = min_t(u64, group_size, SZ_32M);
3417 group_size &= ~(group_align - 1);
3418 if (total_data >= total_metadata * 2) {
3419 group_type = BTRFS_BLOCK_GROUP_METADATA;
3420 group_size = min_t(u64, group_size, SZ_1G);
3421 total_metadata += group_size;
3423 group_type = BTRFS_BLOCK_GROUP_DATA;
3424 group_size = min_t(u64, group_size,
3426 total_data += group_size;
3428 if ((total_bytes - cur_start) * 4 < group_size * 5)
3429 group_size = total_bytes - cur_start;
3432 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3435 cache->key.objectid = cur_start;
3436 cache->key.offset = group_size;
3437 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3439 btrfs_set_block_group_used(&cache->item, 0);
3440 btrfs_set_block_group_chunk_objectid(&cache->item,
3442 btrfs_set_block_group_flags(&cache->item, group_type);
3444 cache->flags = group_type;
3446 ret = update_space_info(root->fs_info, group_type, group_size,
3447 0, &cache->space_info);
3449 set_avail_alloc_bits(extent_root->fs_info, group_type);
3451 set_extent_bits(block_group_cache, cur_start,
3452 cur_start + group_size - 1,
3453 bit | EXTENT_LOCKED, GFP_NOFS);
3454 set_state_private(block_group_cache, cur_start,
3455 (unsigned long)cache);
3456 cur_start += group_size;
3458 /* then insert all the items */
3460 while(cur_start < total_bytes) {
3461 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3464 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3465 sizeof(cache->item));
3468 finish_current_insert(trans, extent_root);
3469 ret = del_pending_extents(trans, extent_root);
3472 cur_start = cache->key.objectid + cache->key.offset;
3477 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3478 struct btrfs_root *root,
3479 u64 bytenr, u64 num_bytes, int alloc,
3482 return update_block_group(trans, root, bytenr, num_bytes,
3487 * Just remove a block group item in extent tree
3488 * Caller should ensure the block group is empty and all space is pinned.
3489 * Or new tree block/data may be allocated into it.
3491 static int free_block_group_item(struct btrfs_trans_handle *trans,
3492 struct btrfs_fs_info *fs_info,
3493 u64 bytenr, u64 len)
3495 struct btrfs_path *path;
3496 struct btrfs_key key;
3497 struct btrfs_root *root = fs_info->extent_root;
3500 key.objectid = bytenr;
3502 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3504 path = btrfs_alloc_path();
3508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3516 ret = btrfs_del_item(trans, root, path);
3518 btrfs_free_path(path);
3522 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3523 struct btrfs_fs_info *fs_info,
3524 u64 devid, u64 dev_offset)
3526 struct btrfs_root *root = fs_info->dev_root;
3527 struct btrfs_path *path;
3528 struct btrfs_key key;
3531 path = btrfs_alloc_path();
3535 key.objectid = devid;
3536 key.type = BTRFS_DEV_EXTENT_KEY;
3537 key.offset = dev_offset;
3539 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3547 ret = btrfs_del_item(trans, root, path);
3549 btrfs_free_path(path);
3553 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3554 struct btrfs_fs_info *fs_info,
3557 struct btrfs_chunk *chunk = NULL;
3558 struct btrfs_root *root= fs_info->chunk_root;
3559 struct btrfs_path *path;
3560 struct btrfs_key key;
3565 path = btrfs_alloc_path();
3569 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3570 key.type = BTRFS_CHUNK_ITEM_KEY;
3571 key.offset = chunk_offset;
3573 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3580 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3581 struct btrfs_chunk);
3582 num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3583 for (i = 0; i < num_stripes; i++) {
3584 ret = free_dev_extent_item(trans, fs_info,
3585 btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3586 btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3591 btrfs_free_path(path);
3595 static int free_system_chunk_item(struct btrfs_super_block *super,
3596 struct btrfs_key *key)
3598 struct btrfs_disk_key *disk_key;
3599 struct btrfs_key cpu_key;
3600 u32 array_size = btrfs_super_sys_array_size(super);
3601 char *ptr = (char *)super->sys_chunk_array;
3605 while (cur < btrfs_super_sys_array_size(super)) {
3606 struct btrfs_chunk *chunk;
3610 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3611 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3612 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3618 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3619 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3620 chunk_len = btrfs_chunk_item_size(num_stripes) +
3623 if (key->objectid == cpu_key.objectid &&
3624 key->offset == cpu_key.offset &&
3625 key->type == cpu_key.type) {
3626 memmove(ptr + cur, ptr + cur + chunk_len,
3627 array_size - cur - chunk_len);
3628 array_size -= chunk_len;
3629 btrfs_set_super_sys_array_size(super, array_size);
3640 static int free_chunk_item(struct btrfs_trans_handle *trans,
3641 struct btrfs_fs_info *fs_info,
3644 struct btrfs_path *path;
3645 struct btrfs_key key;
3646 struct btrfs_root *root = fs_info->chunk_root;
3647 struct btrfs_chunk *chunk;
3651 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3652 key.offset = bytenr;
3653 key.type = BTRFS_CHUNK_ITEM_KEY;
3655 path = btrfs_alloc_path();
3659 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3666 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3667 struct btrfs_chunk);
3668 chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3670 ret = btrfs_del_item(trans, root, path);
3674 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3675 ret = free_system_chunk_item(fs_info->super_copy, &key);
3677 btrfs_free_path(path);
3681 static u64 get_dev_extent_len(struct map_lookup *map)
3685 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3686 case 0: /* Single */
3687 case BTRFS_BLOCK_GROUP_DUP:
3688 case BTRFS_BLOCK_GROUP_RAID1:
3691 case BTRFS_BLOCK_GROUP_RAID5:
3692 div = (map->num_stripes - 1);
3694 case BTRFS_BLOCK_GROUP_RAID6:
3695 div = (map->num_stripes - 2);
3697 case BTRFS_BLOCK_GROUP_RAID10:
3698 div = (map->num_stripes / map->sub_stripes);
3701 /* normally, read chunk security hook should handled it */
3704 return map->ce.size / div;
3707 /* free block group/chunk related caches */
3708 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3709 struct btrfs_fs_info *fs_info,
3710 u64 bytenr, u64 len)
3712 struct btrfs_block_group_cache *cache;
3713 struct cache_extent *ce;
3714 struct map_lookup *map;
3719 /* Free block group cache first */
3720 cache = btrfs_lookup_block_group(fs_info, bytenr);
3723 flags = cache->flags;
3724 if (cache->free_space_ctl) {
3725 btrfs_remove_free_space_cache(cache);
3726 kfree(cache->free_space_ctl);
3728 clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len,
3729 (unsigned int)-1, GFP_NOFS);
3730 ret = free_space_info(fs_info, flags, len, 0, NULL);
3735 /* Then free mapping info and dev usage info */
3736 ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3737 if (!ce || ce->start != bytenr) {
3741 map = container_of(ce, struct map_lookup, ce);
3742 for (i = 0; i < map->num_stripes; i++) {
3743 struct btrfs_device *device;
3745 device = map->stripes[i].dev;
3746 device->bytes_used -= get_dev_extent_len(map);
3747 ret = btrfs_update_device(trans, device);
3751 remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3757 int btrfs_free_block_group(struct btrfs_trans_handle *trans,
3758 struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
3760 struct btrfs_root *extent_root = fs_info->extent_root;
3761 struct btrfs_path *path;
3762 struct btrfs_block_group_item *bgi;
3763 struct btrfs_key key;
3766 path = btrfs_alloc_path();
3770 key.objectid = bytenr;
3771 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3774 /* Double check the block group to ensure it's empty */
3775 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3783 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3784 struct btrfs_block_group_item);
3785 if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
3787 "WARNING: block group [%llu,%llu) is not empty\n",
3788 bytenr, bytenr + len);
3792 btrfs_release_path(path);
3795 * Now pin all space in the block group, to prevent further transaction
3796 * allocate space from it.
3797 * Every operation needs a transaction must be in the range.
3799 btrfs_pin_extent(fs_info, bytenr, len);
3801 /* delete block group item and chunk item */
3802 ret = free_block_group_item(trans, fs_info, bytenr, len);
3805 "failed to free block group item for [%llu,%llu)\n",
3806 bytenr, bytenr + len);
3807 btrfs_unpin_extent(fs_info, bytenr, len);
3811 ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
3814 "failed to dev extents belongs to [%llu,%llu)\n",
3815 bytenr, bytenr + len);
3816 btrfs_unpin_extent(fs_info, bytenr, len);
3819 ret = free_chunk_item(trans, fs_info, bytenr);
3822 "failed to free chunk for [%llu,%llu)\n",
3823 bytenr, bytenr + len);
3824 btrfs_unpin_extent(fs_info, bytenr, len);
3828 /* Now release the block_group_cache */
3829 ret = free_block_group_cache(trans, fs_info, bytenr, len);
3830 btrfs_unpin_extent(fs_info, bytenr, len);
3833 btrfs_free_path(path);
3838 * Fixup block accounting. The initial block accounting created by
3839 * make_block_groups isn't accuracy in this case.
3841 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3842 struct btrfs_root *root)
3848 struct btrfs_path path;
3849 struct btrfs_key key;
3850 struct extent_buffer *leaf;
3851 struct btrfs_block_group_cache *cache;
3852 struct btrfs_fs_info *fs_info = root->fs_info;
3854 root = root->fs_info->extent_root;
3856 while(extent_root_pending_ops(fs_info)) {
3857 ret = finish_current_insert(trans, root);
3860 ret = del_pending_extents(trans, root);
3866 cache = btrfs_lookup_first_block_group(fs_info, start);
3869 start = cache->key.objectid + cache->key.offset;
3870 btrfs_set_block_group_used(&cache->item, 0);
3871 cache->space_info->bytes_used = 0;
3872 set_extent_bits(&root->fs_info->block_group_cache,
3873 cache->key.objectid,
3874 cache->key.objectid + cache->key.offset -1,
3875 BLOCK_GROUP_DIRTY, GFP_NOFS);
3878 btrfs_init_path(&path);
3881 key.type = BTRFS_EXTENT_ITEM_KEY;
3882 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3887 leaf = path.nodes[0];
3888 slot = path.slots[0];
3889 if (slot >= btrfs_header_nritems(leaf)) {
3890 ret = btrfs_next_leaf(root, &path);
3895 leaf = path.nodes[0];
3896 slot = path.slots[0];
3898 btrfs_item_key_to_cpu(leaf, &key, slot);
3899 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3900 bytes_used += key.offset;
3901 ret = btrfs_update_block_group(trans, root,
3902 key.objectid, key.offset, 1, 0);
3904 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3905 bytes_used += root->nodesize;
3906 ret = btrfs_update_block_group(trans, root,
3907 key.objectid, root->nodesize, 1, 0);
3912 btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3913 btrfs_release_path(&path);
3917 static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
3918 u64 *start, u64 *len)
3920 struct btrfs_key key;
3922 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3923 BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
3924 key.type == BTRFS_METADATA_ITEM_KEY));
3925 *start = key.objectid;
3926 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3929 *len = root->nodesize;
3933 * Find first overlap extent for range [bytenr, bytenr + len)
3934 * Return 0 for found and point path to it.
3935 * Return >0 for not found.
3938 int btrfs_search_overlap_extent(struct btrfs_root *root,
3939 struct btrfs_path *path, u64 bytenr, u64 len)
3941 struct btrfs_key key;
3946 key.objectid = bytenr;
3947 key.type = BTRFS_EXTENT_DATA_KEY;
3948 key.offset = (u64)-1;
3950 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3955 ret = btrfs_previous_extent_item(root, path, 0);
3958 /* no previous, check next extent */
3961 __get_extent_size(root, path, &cur_start, &cur_len);
3963 if (cur_start + cur_len > bytenr)
3967 ret = btrfs_next_extent_item(root, path, bytenr + len);
3970 /* No next, prev already checked, no overlap */
3973 __get_extent_size(root, path, &cur_start, &cur_len);
3975 if (cur_start < bytenr + len)
3980 static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3981 struct btrfs_root *root, u64 objectid,
3982 struct btrfs_inode_item *inode,
3983 u64 file_pos, u64 disk_bytenr,
3987 struct btrfs_fs_info *info = root->fs_info;
3988 struct btrfs_root *extent_root = info->extent_root;
3989 struct extent_buffer *leaf;
3990 struct btrfs_file_extent_item *fi;
3991 struct btrfs_key ins_key;
3992 struct btrfs_path *path;
3993 struct btrfs_extent_item *ei;
3995 u64 extent_num_bytes;
3998 u64 num_bytes = *ret_num_bytes;
4001 * All supported file system should not use its 0 extent.
4004 * And hole extent has no size limit, no need to loop.
4006 if (disk_bytenr == 0) {
4007 ret = btrfs_insert_file_extent(trans, root, objectid,
4008 file_pos, disk_bytenr,
4009 num_bytes, num_bytes);
4012 num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
4014 path = btrfs_alloc_path();
4018 /* First to check extent overlap */
4019 ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
4028 __get_extent_size(extent_root, path, &cur_start, &cur_len);
4030 * For convert case, this extent should be a subset of
4033 BUG_ON(disk_bytenr < cur_start);
4035 extent_bytenr = cur_start;
4036 extent_num_bytes = cur_len;
4037 extent_offset = disk_bytenr - extent_bytenr;
4039 /* No overlap, create new extent */
4040 btrfs_release_path(path);
4041 ins_key.objectid = disk_bytenr;
4042 ins_key.offset = num_bytes;
4043 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4045 ret = btrfs_insert_empty_item(trans, extent_root, path,
4046 &ins_key, sizeof(*ei));
4048 leaf = path->nodes[0];
4049 ei = btrfs_item_ptr(leaf, path->slots[0],
4050 struct btrfs_extent_item);
4052 btrfs_set_extent_refs(leaf, ei, 0);
4053 btrfs_set_extent_generation(leaf, ei, 0);
4054 btrfs_set_extent_flags(leaf, ei,
4055 BTRFS_EXTENT_FLAG_DATA);
4056 btrfs_mark_buffer_dirty(leaf);
4058 ret = btrfs_update_block_group(trans, root, disk_bytenr,
4062 } else if (ret != -EEXIST) {
4065 btrfs_extent_post_op(trans, extent_root);
4066 extent_bytenr = disk_bytenr;
4067 extent_num_bytes = num_bytes;
4070 btrfs_release_path(path);
4071 ins_key.objectid = objectid;
4072 ins_key.offset = file_pos;
4073 ins_key.type = BTRFS_EXTENT_DATA_KEY;
4074 ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
4078 leaf = path->nodes[0];
4079 fi = btrfs_item_ptr(leaf, path->slots[0],
4080 struct btrfs_file_extent_item);
4081 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4082 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
4083 btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
4084 btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
4085 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
4086 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
4087 btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
4088 btrfs_set_file_extent_compression(leaf, fi, 0);
4089 btrfs_set_file_extent_encryption(leaf, fi, 0);
4090 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
4091 btrfs_mark_buffer_dirty(leaf);
4093 nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
4094 btrfs_set_stack_inode_nbytes(inode, nbytes);
4095 btrfs_release_path(path);
4097 ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
4098 0, root->root_key.objectid, objectid,
4099 file_pos - extent_offset);
4103 *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
4105 btrfs_free_path(path);
4110 * Record a file extent. Do all the required works, such as inserting
4111 * file extent item, inserting extent item and backref item into extent
4112 * tree and updating block accounting.
4114 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
4115 struct btrfs_root *root, u64 objectid,
4116 struct btrfs_inode_item *inode,
4117 u64 file_pos, u64 disk_bytenr,
4120 u64 cur_disk_bytenr = disk_bytenr;
4121 u64 cur_file_pos = file_pos;
4122 u64 cur_num_bytes = num_bytes;
4125 while (num_bytes > 0) {
4126 ret = __btrfs_record_file_extent(trans, root, objectid,
4127 inode, cur_file_pos,
4132 cur_disk_bytenr += cur_num_bytes;
4133 cur_file_pos += cur_num_bytes;
4134 num_bytes -= cur_num_bytes;
4140 static int add_excluded_extent(struct btrfs_root *root,
4141 u64 start, u64 num_bytes)
4143 u64 end = start + num_bytes - 1;
4144 set_extent_bits(&root->fs_info->pinned_extents,
4145 start, end, EXTENT_UPTODATE, GFP_NOFS);
4149 void free_excluded_extents(struct btrfs_root *root,
4150 struct btrfs_block_group_cache *cache)
4154 start = cache->key.objectid;
4155 end = start + cache->key.offset - 1;
4157 clear_extent_bits(&root->fs_info->pinned_extents,
4158 start, end, EXTENT_UPTODATE, GFP_NOFS);
4161 int exclude_super_stripes(struct btrfs_root *root,
4162 struct btrfs_block_group_cache *cache)
4169 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
4170 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
4171 cache->bytes_super += stripe_len;
4172 ret = add_excluded_extent(root, cache->key.objectid,
4178 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4179 bytenr = btrfs_sb_offset(i);
4180 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4181 cache->key.objectid, bytenr,
4182 0, &logical, &nr, &stripe_len);
4189 if (logical[nr] > cache->key.objectid +
4193 if (logical[nr] + stripe_len <= cache->key.objectid)
4196 start = logical[nr];
4197 if (start < cache->key.objectid) {
4198 start = cache->key.objectid;
4199 len = (logical[nr] + stripe_len) - start;
4201 len = min_t(u64, stripe_len,
4202 cache->key.objectid +
4203 cache->key.offset - start);
4206 cache->bytes_super += len;
4207 ret = add_excluded_extent(root, start, len);
4219 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
4220 struct btrfs_fs_info *info, u64 start, u64 end)
4222 u64 extent_start, extent_end, size, total_added = 0;
4225 while (start < end) {
4226 ret = find_first_extent_bit(&info->pinned_extents, start,
4227 &extent_start, &extent_end,
4228 EXTENT_DIRTY | EXTENT_UPTODATE);
4232 if (extent_start <= start) {
4233 start = extent_end + 1;
4234 } else if (extent_start > start && extent_start < end) {
4235 size = extent_start - start;
4236 total_added += size;
4237 ret = btrfs_add_free_space(block_group->free_space_ctl,
4239 BUG_ON(ret); /* -ENOMEM or logic error */
4240 start = extent_end + 1;
4248 total_added += size;
4249 ret = btrfs_add_free_space(block_group->free_space_ctl, start,
4251 BUG_ON(ret); /* -ENOMEM or logic error */