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_trans_handle *trans,
1167 struct btrfs_root *root,
1168 struct btrfs_path *path,
1169 struct btrfs_extent_inline_ref *iref,
1170 u64 parent, u64 root_objectid,
1171 u64 owner, u64 offset, int refs_to_add)
1173 struct extent_buffer *leaf;
1174 struct btrfs_extent_item *ei;
1177 unsigned long item_offset;
1183 leaf = path->nodes[0];
1184 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1185 item_offset = (unsigned long)iref - (unsigned long)ei;
1187 type = extent_ref_type(parent, owner);
1188 size = btrfs_extent_inline_ref_size(type);
1190 ret = btrfs_extend_item(root, path, size);
1193 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1194 refs = btrfs_extent_refs(leaf, ei);
1195 refs += refs_to_add;
1196 btrfs_set_extent_refs(leaf, ei, refs);
1198 ptr = (unsigned long)ei + item_offset;
1199 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1200 if (ptr < end - size)
1201 memmove_extent_buffer(leaf, ptr + size, ptr,
1204 iref = (struct btrfs_extent_inline_ref *)ptr;
1205 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1206 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1207 struct btrfs_extent_data_ref *dref;
1208 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1209 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1210 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1211 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1212 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1213 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1214 struct btrfs_shared_data_ref *sref;
1215 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1216 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1217 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1218 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1219 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1221 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1223 btrfs_mark_buffer_dirty(leaf);
1227 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root,
1229 struct btrfs_path *path,
1230 struct btrfs_extent_inline_ref **ref_ret,
1231 u64 bytenr, u64 num_bytes, u64 parent,
1232 u64 root_objectid, u64 owner, u64 offset)
1236 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1237 bytenr, num_bytes, parent,
1238 root_objectid, owner, offset, 0);
1242 btrfs_release_path(path);
1245 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1246 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1249 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1250 root_objectid, owner, offset);
1255 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1256 struct btrfs_root *root,
1257 struct btrfs_path *path,
1258 struct btrfs_extent_inline_ref *iref,
1261 struct extent_buffer *leaf;
1262 struct btrfs_extent_item *ei;
1263 struct btrfs_extent_data_ref *dref = NULL;
1264 struct btrfs_shared_data_ref *sref = NULL;
1273 leaf = path->nodes[0];
1274 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1275 refs = btrfs_extent_refs(leaf, ei);
1276 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1277 refs += refs_to_mod;
1278 btrfs_set_extent_refs(leaf, ei, refs);
1280 type = btrfs_extent_inline_ref_type(leaf, iref);
1282 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1283 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1284 refs = btrfs_extent_data_ref_count(leaf, dref);
1285 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1286 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1287 refs = btrfs_shared_data_ref_count(leaf, sref);
1290 BUG_ON(refs_to_mod != -1);
1293 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1294 refs += refs_to_mod;
1297 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1298 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1300 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1302 size = btrfs_extent_inline_ref_size(type);
1303 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1304 ptr = (unsigned long)iref;
1305 end = (unsigned long)ei + item_size;
1306 if (ptr + size < end)
1307 memmove_extent_buffer(leaf, ptr, ptr + size,
1310 ret = btrfs_truncate_item(root, path, item_size, 1);
1313 btrfs_mark_buffer_dirty(leaf);
1317 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1318 struct btrfs_root *root,
1319 struct btrfs_path *path,
1320 u64 bytenr, u64 num_bytes, u64 parent,
1321 u64 root_objectid, u64 owner,
1322 u64 offset, int refs_to_add)
1324 struct btrfs_extent_inline_ref *iref;
1327 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1328 bytenr, num_bytes, parent,
1329 root_objectid, owner, offset, 1);
1331 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1332 ret = update_inline_extent_backref(trans, root, path, iref,
1334 } else if (ret == -ENOENT) {
1335 ret = setup_inline_extent_backref(trans, root, path, iref,
1336 parent, root_objectid,
1337 owner, offset, refs_to_add);
1342 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1343 struct btrfs_root *root,
1344 struct btrfs_path *path,
1345 u64 bytenr, u64 parent, u64 root_objectid,
1346 u64 owner, u64 offset, int refs_to_add)
1350 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1351 ret = insert_extent_data_ref(trans, root, path, bytenr,
1352 parent, root_objectid,
1353 owner, offset, refs_to_add);
1355 BUG_ON(refs_to_add != 1);
1356 ret = insert_tree_block_ref(trans, root, path, bytenr,
1357 parent, root_objectid);
1362 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root,
1364 struct btrfs_path *path,
1365 struct btrfs_extent_inline_ref *iref,
1366 int refs_to_drop, int is_data)
1370 BUG_ON(!is_data && refs_to_drop != 1);
1372 ret = update_inline_extent_backref(trans, root, path, iref,
1374 } else if (is_data) {
1375 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1377 ret = btrfs_del_item(trans, root, path);
1382 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1383 struct btrfs_root *root,
1384 u64 bytenr, u64 num_bytes, u64 parent,
1385 u64 root_objectid, u64 owner, u64 offset)
1387 struct btrfs_path *path;
1388 struct extent_buffer *leaf;
1389 struct btrfs_extent_item *item;
1394 path = btrfs_alloc_path();
1400 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1401 path, bytenr, num_bytes, parent,
1402 root_objectid, owner, offset, 1);
1406 if (ret != -EAGAIN) {
1411 leaf = path->nodes[0];
1412 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1413 refs = btrfs_extent_refs(leaf, item);
1414 btrfs_set_extent_refs(leaf, item, refs + 1);
1416 btrfs_mark_buffer_dirty(leaf);
1417 btrfs_release_path(path);
1421 /* now insert the actual backref */
1422 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1423 path, bytenr, parent, root_objectid,
1428 btrfs_free_path(path);
1429 finish_current_insert(trans, root->fs_info->extent_root);
1430 del_pending_extents(trans, root->fs_info->extent_root);
1435 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1436 struct btrfs_root *root)
1438 finish_current_insert(trans, root->fs_info->extent_root);
1439 del_pending_extents(trans, root->fs_info->extent_root);
1443 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1444 struct btrfs_root *root, u64 bytenr,
1445 u64 offset, int metadata, u64 *refs, u64 *flags)
1447 struct btrfs_path *path;
1449 struct btrfs_key key;
1450 struct extent_buffer *l;
1451 struct btrfs_extent_item *item;
1457 !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
1458 offset = root->nodesize;
1462 path = btrfs_alloc_path();
1467 key.objectid = bytenr;
1468 key.offset = offset;
1470 key.type = BTRFS_METADATA_ITEM_KEY;
1472 key.type = BTRFS_EXTENT_ITEM_KEY;
1475 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1481 * Deal with the fact that we may have mixed SKINNY and normal refs. If
1482 * we didn't find what we wanted check and see if we have a normal ref
1483 * right next to us, or re-search if we are on the edge of the leaf just
1486 if (ret > 0 && metadata) {
1487 if (path->slots[0]) {
1489 btrfs_item_key_to_cpu(path->nodes[0], &key,
1491 if (key.objectid == bytenr &&
1492 key.type == BTRFS_EXTENT_ITEM_KEY &&
1493 key.offset == root->nodesize)
1498 btrfs_release_path(path);
1499 key.type = BTRFS_EXTENT_ITEM_KEY;
1500 key.offset = root->nodesize;
1512 item_size = btrfs_item_size_nr(l, path->slots[0]);
1513 if (item_size >= sizeof(*item)) {
1514 item = btrfs_item_ptr(l, path->slots[0],
1515 struct btrfs_extent_item);
1516 num_refs = btrfs_extent_refs(l, item);
1517 extent_flags = btrfs_extent_flags(l, item);
1519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1520 struct btrfs_extent_item_v0 *ei0;
1521 BUG_ON(item_size != sizeof(*ei0));
1522 ei0 = btrfs_item_ptr(l, path->slots[0],
1523 struct btrfs_extent_item_v0);
1524 num_refs = btrfs_extent_refs_v0(l, ei0);
1525 /* FIXME: this isn't correct for data */
1526 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1531 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1535 *flags = extent_flags;
1537 btrfs_free_path(path);
1541 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1542 struct btrfs_root *root,
1543 u64 bytenr, int level, u64 flags)
1545 struct btrfs_path *path;
1547 struct btrfs_key key;
1548 struct extent_buffer *l;
1549 struct btrfs_extent_item *item;
1551 int skinny_metadata =
1552 btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
1554 path = btrfs_alloc_path();
1559 key.objectid = bytenr;
1560 if (skinny_metadata) {
1562 key.type = BTRFS_METADATA_ITEM_KEY;
1564 key.offset = root->nodesize;
1565 key.type = BTRFS_EXTENT_ITEM_KEY;
1569 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1574 if (ret > 0 && skinny_metadata) {
1575 skinny_metadata = 0;
1576 if (path->slots[0]) {
1578 btrfs_item_key_to_cpu(path->nodes[0], &key,
1580 if (key.objectid == bytenr &&
1581 key.offset == root->nodesize &&
1582 key.type == BTRFS_EXTENT_ITEM_KEY)
1586 btrfs_release_path(path);
1587 key.offset = root->nodesize;
1588 key.type = BTRFS_EXTENT_ITEM_KEY;
1594 btrfs_print_leaf(root, path->nodes[0]);
1595 printk("failed to find block number %Lu\n",
1596 (unsigned long long)bytenr);
1600 item_size = btrfs_item_size_nr(l, path->slots[0]);
1601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1602 if (item_size < sizeof(*item)) {
1603 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1609 item_size = btrfs_item_size_nr(l, path->slots[0]);
1612 BUG_ON(item_size < sizeof(*item));
1613 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1614 flags |= btrfs_extent_flags(l, item);
1615 btrfs_set_extent_flags(l, item, flags);
1617 btrfs_free_path(path);
1618 finish_current_insert(trans, root->fs_info->extent_root);
1619 del_pending_extents(trans, root->fs_info->extent_root);
1623 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1624 struct btrfs_root *root,
1625 struct extent_buffer *buf,
1626 int record_parent, int inc)
1633 struct btrfs_key key;
1634 struct btrfs_file_extent_item *fi;
1638 int (*process_func)(struct btrfs_trans_handle *trans,
1639 struct btrfs_root *root,
1640 u64, u64, u64, u64, u64, u64);
1642 ref_root = btrfs_header_owner(buf);
1643 nritems = btrfs_header_nritems(buf);
1644 level = btrfs_header_level(buf);
1646 if (!root->ref_cows && level == 0)
1650 process_func = btrfs_inc_extent_ref;
1652 process_func = btrfs_free_extent;
1655 parent = buf->start;
1659 for (i = 0; i < nritems; i++) {
1662 btrfs_item_key_to_cpu(buf, &key, i);
1663 if (key.type != BTRFS_EXTENT_DATA_KEY)
1665 fi = btrfs_item_ptr(buf, i,
1666 struct btrfs_file_extent_item);
1667 if (btrfs_file_extent_type(buf, fi) ==
1668 BTRFS_FILE_EXTENT_INLINE)
1670 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1674 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1675 key.offset -= btrfs_file_extent_offset(buf, fi);
1676 ret = process_func(trans, root, bytenr, num_bytes,
1677 parent, ref_root, key.objectid,
1684 bytenr = btrfs_node_blockptr(buf, i);
1685 num_bytes = root->nodesize;
1686 ret = process_func(trans, root, bytenr, num_bytes,
1687 parent, ref_root, level - 1, 0);
1700 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1701 struct extent_buffer *buf, int record_parent)
1703 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1706 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1707 struct extent_buffer *buf, int record_parent)
1709 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1712 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1713 struct btrfs_root *root,
1714 struct btrfs_path *path,
1715 struct btrfs_block_group_cache *cache)
1719 struct btrfs_root *extent_root = root->fs_info->extent_root;
1721 struct extent_buffer *leaf;
1723 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1728 leaf = path->nodes[0];
1729 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1730 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1731 btrfs_mark_buffer_dirty(leaf);
1732 btrfs_release_path(path);
1734 finish_current_insert(trans, extent_root);
1735 pending_ret = del_pending_extents(trans, extent_root);
1744 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1745 struct btrfs_root *root)
1747 struct extent_io_tree *block_group_cache;
1748 struct btrfs_block_group_cache *cache;
1750 struct btrfs_path *path;
1756 block_group_cache = &root->fs_info->block_group_cache;
1757 path = btrfs_alloc_path();
1762 ret = find_first_extent_bit(block_group_cache, last,
1763 &start, &end, BLOCK_GROUP_DIRTY);
1772 ret = get_state_private(block_group_cache, start, &ptr);
1775 clear_extent_bits(block_group_cache, start, end,
1776 BLOCK_GROUP_DIRTY, GFP_NOFS);
1778 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1779 ret = write_one_cache_group(trans, root, path, cache);
1781 btrfs_free_path(path);
1785 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1788 struct btrfs_space_info *found;
1790 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1792 list_for_each_entry(found, &info->space_info, list) {
1793 if (found->flags & flags)
1800 static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
1801 u64 total_bytes, u64 bytes_used,
1802 struct btrfs_space_info **space_info)
1804 struct btrfs_space_info *found;
1806 /* only support free block group which is empty */
1810 found = __find_space_info(fs_info, flags);
1813 if (found->total_bytes < total_bytes) {
1815 "WARNING: bad space info to free %llu only have %llu\n",
1816 total_bytes, found->total_bytes);
1819 found->total_bytes -= total_bytes;
1821 *space_info = found;
1825 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1826 u64 total_bytes, u64 bytes_used,
1827 struct btrfs_space_info **space_info)
1829 struct btrfs_space_info *found;
1831 found = __find_space_info(info, flags);
1833 found->total_bytes += total_bytes;
1834 found->bytes_used += bytes_used;
1835 if (found->total_bytes < found->bytes_used) {
1836 fprintf(stderr, "warning, bad space info total_bytes "
1838 (unsigned long long)found->total_bytes,
1839 (unsigned long long)found->bytes_used);
1841 *space_info = found;
1844 found = kmalloc(sizeof(*found), GFP_NOFS);
1848 list_add(&found->list, &info->space_info);
1849 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1850 found->total_bytes = total_bytes;
1851 found->bytes_used = bytes_used;
1852 found->bytes_pinned = 0;
1854 *space_info = found;
1859 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1861 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1862 BTRFS_BLOCK_GROUP_RAID1 |
1863 BTRFS_BLOCK_GROUP_RAID10 |
1864 BTRFS_BLOCK_GROUP_RAID5 |
1865 BTRFS_BLOCK_GROUP_RAID6 |
1866 BTRFS_BLOCK_GROUP_DUP);
1868 if (flags & BTRFS_BLOCK_GROUP_DATA)
1869 fs_info->avail_data_alloc_bits |= extra_flags;
1870 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1871 fs_info->avail_metadata_alloc_bits |= extra_flags;
1872 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1873 fs_info->avail_system_alloc_bits |= extra_flags;
1877 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1878 struct btrfs_root *extent_root, u64 alloc_bytes,
1881 struct btrfs_space_info *space_info;
1887 space_info = __find_space_info(extent_root->fs_info, flags);
1889 ret = update_space_info(extent_root->fs_info, flags,
1893 BUG_ON(!space_info);
1895 if (space_info->full)
1898 thresh = div_factor(space_info->total_bytes, 7);
1899 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1904 * Avoid allocating given chunk type
1906 if (extent_root->fs_info->avoid_meta_chunk_alloc &&
1907 (flags & BTRFS_BLOCK_GROUP_METADATA))
1909 if (extent_root->fs_info->avoid_sys_chunk_alloc &&
1910 (flags & BTRFS_BLOCK_GROUP_SYSTEM))
1913 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1915 if (ret == -ENOSPC) {
1916 space_info->full = 1;
1922 ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1923 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1928 static int update_block_group(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
1930 u64 bytenr, u64 num_bytes, int alloc,
1933 struct btrfs_block_group_cache *cache;
1934 struct btrfs_fs_info *info = root->fs_info;
1935 u64 total = num_bytes;
1941 /* block accounting for super block */
1942 old_val = btrfs_super_bytes_used(info->super_copy);
1944 old_val += num_bytes;
1946 old_val -= num_bytes;
1947 btrfs_set_super_bytes_used(info->super_copy, old_val);
1949 /* block accounting for root item */
1950 old_val = btrfs_root_used(&root->root_item);
1952 old_val += num_bytes;
1954 old_val -= num_bytes;
1955 btrfs_set_root_used(&root->root_item, old_val);
1958 cache = btrfs_lookup_block_group(info, bytenr);
1962 byte_in_group = bytenr - cache->key.objectid;
1963 WARN_ON(byte_in_group > cache->key.offset);
1964 start = cache->key.objectid;
1965 end = start + cache->key.offset - 1;
1966 set_extent_bits(&info->block_group_cache, start, end,
1967 BLOCK_GROUP_DIRTY, GFP_NOFS);
1969 old_val = btrfs_block_group_used(&cache->item);
1970 num_bytes = min(total, cache->key.offset - byte_in_group);
1973 old_val += num_bytes;
1974 cache->space_info->bytes_used += num_bytes;
1976 old_val -= num_bytes;
1977 cache->space_info->bytes_used -= num_bytes;
1979 set_extent_dirty(&info->free_space_cache,
1980 bytenr, bytenr + num_bytes - 1,
1984 btrfs_set_block_group_used(&cache->item, old_val);
1986 bytenr += num_bytes;
1991 static int update_pinned_extents(struct btrfs_root *root,
1992 u64 bytenr, u64 num, int pin)
1995 struct btrfs_block_group_cache *cache;
1996 struct btrfs_fs_info *fs_info = root->fs_info;
1999 set_extent_dirty(&fs_info->pinned_extents,
2000 bytenr, bytenr + num - 1, GFP_NOFS);
2002 clear_extent_dirty(&fs_info->pinned_extents,
2003 bytenr, bytenr + num - 1, GFP_NOFS);
2006 cache = btrfs_lookup_block_group(fs_info, bytenr);
2008 len = min((u64)root->sectorsize, num);
2012 len = min(num, cache->key.offset -
2013 (bytenr - cache->key.objectid));
2015 cache->pinned += len;
2016 cache->space_info->bytes_pinned += len;
2017 fs_info->total_pinned += len;
2019 cache->pinned -= len;
2020 cache->space_info->bytes_pinned -= len;
2021 fs_info->total_pinned -= len;
2030 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2031 struct btrfs_root *root,
2032 struct extent_io_tree *unpin)
2037 struct extent_io_tree *free_space_cache;
2038 free_space_cache = &root->fs_info->free_space_cache;
2041 ret = find_first_extent_bit(unpin, 0, &start, &end,
2045 update_pinned_extents(root, start, end + 1 - start, 0);
2046 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2047 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2052 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2058 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2059 &end, EXTENT_LOCKED);
2061 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2067 static int finish_current_insert(struct btrfs_trans_handle *trans,
2068 struct btrfs_root *extent_root)
2073 struct btrfs_fs_info *info = extent_root->fs_info;
2074 struct pending_extent_op *extent_op;
2075 struct btrfs_key key;
2077 int skinny_metadata =
2078 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2081 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2082 &end, EXTENT_LOCKED);
2086 ret = get_state_private(&info->extent_ins, start, &priv);
2088 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2090 if (extent_op->type == PENDING_EXTENT_INSERT) {
2091 key.objectid = start;
2092 if (skinny_metadata) {
2093 key.offset = extent_op->level;
2094 key.type = BTRFS_METADATA_ITEM_KEY;
2096 key.offset = extent_op->num_bytes;
2097 key.type = BTRFS_EXTENT_ITEM_KEY;
2099 ret = alloc_reserved_tree_block(trans, extent_root,
2100 extent_root->root_key.objectid,
2104 extent_op->level, &key);
2110 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2117 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2118 struct btrfs_root *root,
2119 u64 bytenr, u64 num_bytes, int is_data)
2122 struct extent_buffer *buf;
2127 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2131 /* we can reuse a block if it hasn't been written
2132 * and it is from this transaction. We can't
2133 * reuse anything from the tree log root because
2134 * it has tiny sub-transactions.
2136 if (btrfs_buffer_uptodate(buf, 0)) {
2137 u64 header_owner = btrfs_header_owner(buf);
2138 u64 header_transid = btrfs_header_generation(buf);
2139 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2140 header_transid == trans->transid &&
2141 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2142 clean_tree_block(NULL, root, buf);
2143 free_extent_buffer(buf);
2147 free_extent_buffer(buf);
2149 update_pinned_extents(root, bytenr, num_bytes, 1);
2155 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2156 u64 bytenr, u64 num_bytes)
2158 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2161 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2162 u64 bytenr, u64 num_bytes)
2164 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2168 * remove an extent from the root, returns 0 on success
2170 static int __free_extent(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root,
2172 u64 bytenr, u64 num_bytes, u64 parent,
2173 u64 root_objectid, u64 owner_objectid,
2174 u64 owner_offset, int refs_to_drop)
2177 struct btrfs_key key;
2178 struct btrfs_path *path;
2179 struct btrfs_root *extent_root = root->fs_info->extent_root;
2180 struct extent_buffer *leaf;
2181 struct btrfs_extent_item *ei;
2182 struct btrfs_extent_inline_ref *iref;
2185 int extent_slot = 0;
2186 int found_extent = 0;
2190 int skinny_metadata =
2191 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2193 if (root->fs_info->free_extent_hook) {
2194 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2195 parent, root_objectid, owner_objectid,
2196 owner_offset, refs_to_drop);
2199 path = btrfs_alloc_path();
2205 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2207 skinny_metadata = 0;
2208 BUG_ON(!is_data && refs_to_drop != 1);
2210 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2211 bytenr, num_bytes, parent,
2212 root_objectid, owner_objectid,
2215 extent_slot = path->slots[0];
2216 while (extent_slot >= 0) {
2217 btrfs_item_key_to_cpu(path->nodes[0], &key,
2219 if (key.objectid != bytenr)
2221 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2222 key.offset == num_bytes) {
2226 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2227 key.offset == owner_objectid) {
2231 if (path->slots[0] - extent_slot > 5)
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2237 if (found_extent && item_size < sizeof(*ei))
2240 if (!found_extent) {
2242 ret = remove_extent_backref(trans, extent_root, path,
2246 btrfs_release_path(path);
2248 key.objectid = bytenr;
2250 if (skinny_metadata) {
2251 key.type = BTRFS_METADATA_ITEM_KEY;
2252 key.offset = owner_objectid;
2254 key.type = BTRFS_EXTENT_ITEM_KEY;
2255 key.offset = num_bytes;
2258 ret = btrfs_search_slot(trans, extent_root,
2260 if (ret > 0 && skinny_metadata && path->slots[0]) {
2262 btrfs_item_key_to_cpu(path->nodes[0],
2265 if (key.objectid == bytenr &&
2266 key.type == BTRFS_EXTENT_ITEM_KEY &&
2267 key.offset == num_bytes)
2271 if (ret > 0 && skinny_metadata) {
2272 skinny_metadata = 0;
2273 btrfs_release_path(path);
2274 key.type = BTRFS_EXTENT_ITEM_KEY;
2275 key.offset = num_bytes;
2276 ret = btrfs_search_slot(trans, extent_root,
2281 printk(KERN_ERR "umm, got %d back from search"
2282 ", was looking for %llu\n", ret,
2283 (unsigned long long)bytenr);
2284 btrfs_print_leaf(extent_root, path->nodes[0]);
2287 extent_slot = path->slots[0];
2290 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2291 "parent %llu root %llu owner %llu offset %llu\n",
2292 (unsigned long long)bytenr,
2293 (unsigned long long)parent,
2294 (unsigned long long)root_objectid,
2295 (unsigned long long)owner_objectid,
2296 (unsigned long long)owner_offset);
2301 leaf = path->nodes[0];
2302 item_size = btrfs_item_size_nr(leaf, extent_slot);
2303 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2304 if (item_size < sizeof(*ei)) {
2305 BUG_ON(found_extent || extent_slot != path->slots[0]);
2306 ret = convert_extent_item_v0(trans, extent_root, path,
2310 btrfs_release_path(path);
2312 key.objectid = bytenr;
2313 key.type = BTRFS_EXTENT_ITEM_KEY;
2314 key.offset = num_bytes;
2316 ret = btrfs_search_slot(trans, extent_root, &key, path,
2319 printk(KERN_ERR "umm, got %d back from search"
2320 ", was looking for %llu\n", ret,
2321 (unsigned long long)bytenr);
2322 btrfs_print_leaf(extent_root, path->nodes[0]);
2325 extent_slot = path->slots[0];
2326 leaf = path->nodes[0];
2327 item_size = btrfs_item_size_nr(leaf, extent_slot);
2330 BUG_ON(item_size < sizeof(*ei));
2331 ei = btrfs_item_ptr(leaf, extent_slot,
2332 struct btrfs_extent_item);
2333 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2334 key.type == BTRFS_EXTENT_ITEM_KEY) {
2335 struct btrfs_tree_block_info *bi;
2336 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2337 bi = (struct btrfs_tree_block_info *)(ei + 1);
2338 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2341 refs = btrfs_extent_refs(leaf, ei);
2342 BUG_ON(refs < refs_to_drop);
2343 refs -= refs_to_drop;
2347 * In the case of inline back ref, reference count will
2348 * be updated by remove_extent_backref
2351 BUG_ON(!found_extent);
2353 btrfs_set_extent_refs(leaf, ei, refs);
2354 btrfs_mark_buffer_dirty(leaf);
2357 ret = remove_extent_backref(trans, extent_root, path,
2367 BUG_ON(is_data && refs_to_drop !=
2368 extent_data_ref_count(path, iref));
2370 BUG_ON(path->slots[0] != extent_slot);
2372 BUG_ON(path->slots[0] != extent_slot + 1);
2373 path->slots[0] = extent_slot;
2379 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2386 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2389 btrfs_release_path(path);
2392 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2396 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2399 btrfs_free_path(path);
2400 finish_current_insert(trans, extent_root);
2405 * find all the blocks marked as pending in the radix tree and remove
2406 * them from the extent map
2408 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2409 btrfs_root *extent_root)
2416 struct extent_io_tree *pending_del;
2417 struct extent_io_tree *extent_ins;
2418 struct pending_extent_op *extent_op;
2420 extent_ins = &extent_root->fs_info->extent_ins;
2421 pending_del = &extent_root->fs_info->pending_del;
2424 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2429 ret = get_state_private(pending_del, start, &priv);
2431 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2433 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2436 if (!test_range_bit(extent_ins, start, end,
2437 EXTENT_LOCKED, 0)) {
2438 ret = __free_extent(trans, extent_root,
2439 start, end + 1 - start, 0,
2440 extent_root->root_key.objectid,
2441 extent_op->level, 0, 1);
2445 ret = get_state_private(extent_ins, start, &priv);
2447 extent_op = (struct pending_extent_op *)
2448 (unsigned long)priv;
2450 clear_extent_bits(extent_ins, start, end,
2451 EXTENT_LOCKED, GFP_NOFS);
2453 if (extent_op->type == PENDING_BACKREF_UPDATE)
2465 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
2466 struct btrfs_root *root,
2467 struct extent_buffer *buf,
2468 u64 parent, int last_ref)
2470 return btrfs_free_extent(trans, root, buf->start, buf->len, parent,
2471 root->root_key.objectid,
2472 btrfs_header_level(buf), 0);
2476 * remove an extent from the root, returns 0 on success
2479 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2480 struct btrfs_root *root,
2481 u64 bytenr, u64 num_bytes, u64 parent,
2482 u64 root_objectid, u64 owner, u64 offset)
2484 struct btrfs_root *extent_root = root->fs_info->extent_root;
2488 WARN_ON(num_bytes < root->sectorsize);
2489 if (root == extent_root) {
2490 struct pending_extent_op *extent_op;
2492 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2495 extent_op->type = PENDING_EXTENT_DELETE;
2496 extent_op->bytenr = bytenr;
2497 extent_op->num_bytes = num_bytes;
2498 extent_op->level = (int)owner;
2500 set_extent_bits(&root->fs_info->pending_del,
2501 bytenr, bytenr + num_bytes - 1,
2502 EXTENT_LOCKED, GFP_NOFS);
2503 set_state_private(&root->fs_info->pending_del,
2504 bytenr, (unsigned long)extent_op);
2507 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2508 root_objectid, owner, offset, 1);
2509 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2510 return ret ? ret : pending_ret;
2513 static u64 stripe_align(struct btrfs_root *root, u64 val)
2515 u64 mask = ((u64)root->stripesize - 1);
2516 u64 ret = (val + mask) & ~mask;
2521 * walks the btree of allocated extents and find a hole of a given size.
2522 * The key ins is changed to record the hole:
2523 * ins->objectid == block start
2524 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2525 * ins->offset == number of blocks
2526 * Any available blocks before search_start are skipped.
2528 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2529 struct btrfs_root *orig_root,
2530 u64 num_bytes, u64 empty_size,
2531 u64 search_start, u64 search_end,
2532 u64 hint_byte, struct btrfs_key *ins,
2533 u64 exclude_start, u64 exclude_nr,
2537 u64 orig_search_start = search_start;
2538 struct btrfs_root * root = orig_root->fs_info->extent_root;
2539 struct btrfs_fs_info *info = root->fs_info;
2540 u64 total_needed = num_bytes;
2541 struct btrfs_block_group_cache *block_group;
2545 WARN_ON(num_bytes < root->sectorsize);
2546 ins->type = BTRFS_EXTENT_ITEM_KEY;
2548 search_start = stripe_align(root, search_start);
2551 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2553 hint_byte = search_start;
2554 block_group = btrfs_find_block_group(root, block_group,
2555 hint_byte, data, 1);
2557 block_group = btrfs_find_block_group(root,
2559 search_start, data, 1);
2562 total_needed += empty_size;
2565 search_start = stripe_align(root, search_start);
2567 block_group = btrfs_lookup_first_block_group(info,
2570 block_group = btrfs_lookup_first_block_group(info,
2573 ret = find_search_start(root, &block_group, &search_start,
2574 total_needed, data);
2578 ins->objectid = search_start;
2579 ins->offset = num_bytes;
2581 if (ins->objectid + num_bytes >
2582 block_group->key.objectid + block_group->key.offset) {
2583 search_start = block_group->key.objectid +
2584 block_group->key.offset;
2588 if (test_range_bit(&info->extent_ins, ins->objectid,
2589 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2590 search_start = ins->objectid + num_bytes;
2594 if (test_range_bit(&info->pinned_extents, ins->objectid,
2595 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2596 search_start = ins->objectid + num_bytes;
2600 if (info->excluded_extents &&
2601 test_range_bit(info->excluded_extents, ins->objectid,
2602 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2603 search_start = ins->objectid + num_bytes;
2607 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2608 ins->objectid < exclude_start + exclude_nr)) {
2609 search_start = exclude_start + exclude_nr;
2613 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2614 if (check_crossing_stripes(info, ins->objectid, num_bytes)) {
2615 struct btrfs_block_group_cache *bg_cache;
2618 bg_cache = btrfs_lookup_block_group(info, ins->objectid);
2621 bg_offset = ins->objectid - bg_cache->key.objectid;
2623 search_start = round_up(bg_offset + num_bytes,
2624 BTRFS_STRIPE_LEN) + bg_offset;
2628 block_group = btrfs_lookup_block_group(info, ins->objectid);
2630 trans->block_group = block_group;
2632 ins->offset = num_bytes;
2636 block_group = btrfs_lookup_first_block_group(info, search_start);
2638 search_start = orig_search_start;
2645 total_needed -= empty_size;
2651 block_group = btrfs_find_block_group(root, block_group,
2652 search_start, data, 0);
2659 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2660 struct btrfs_root *root,
2661 u64 num_bytes, u64 empty_size,
2662 u64 hint_byte, u64 search_end,
2663 struct btrfs_key *ins, int data)
2666 u64 search_start = 0;
2668 struct btrfs_fs_info *info = root->fs_info;
2671 alloc_profile = info->avail_data_alloc_bits &
2672 info->data_alloc_profile;
2673 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2674 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2675 info->system_allocs >= 0) {
2676 alloc_profile = info->avail_system_alloc_bits &
2677 info->system_alloc_profile;
2678 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2680 alloc_profile = info->avail_metadata_alloc_bits &
2681 info->metadata_alloc_profile;
2682 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2685 if (root->ref_cows) {
2686 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2687 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2689 BTRFS_BLOCK_GROUP_METADATA);
2692 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2693 num_bytes + SZ_2M, data);
2697 WARN_ON(num_bytes < root->sectorsize);
2698 ret = find_free_extent(trans, root, num_bytes, empty_size,
2699 search_start, search_end, hint_byte, ins,
2700 trans->alloc_exclude_start,
2701 trans->alloc_exclude_nr, data);
2703 clear_extent_dirty(&root->fs_info->free_space_cache,
2704 ins->objectid, ins->objectid + ins->offset - 1,
2709 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *root,
2711 u64 root_objectid, u64 generation,
2712 u64 flags, struct btrfs_disk_key *key,
2713 int level, struct btrfs_key *ins)
2716 struct btrfs_fs_info *fs_info = root->fs_info;
2717 struct btrfs_extent_item *extent_item;
2718 struct btrfs_tree_block_info *block_info;
2719 struct btrfs_extent_inline_ref *iref;
2720 struct btrfs_path *path;
2721 struct extent_buffer *leaf;
2722 u32 size = sizeof(*extent_item) + sizeof(*iref);
2723 int skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2725 if (!skinny_metadata)
2726 size += sizeof(*block_info);
2728 path = btrfs_alloc_path();
2732 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2736 leaf = path->nodes[0];
2737 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2738 struct btrfs_extent_item);
2739 btrfs_set_extent_refs(leaf, extent_item, 1);
2740 btrfs_set_extent_generation(leaf, extent_item, generation);
2741 btrfs_set_extent_flags(leaf, extent_item,
2742 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2744 if (skinny_metadata) {
2745 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2747 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2748 btrfs_set_tree_block_key(leaf, block_info, key);
2749 btrfs_set_tree_block_level(leaf, block_info, level);
2750 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2753 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2754 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2756 btrfs_mark_buffer_dirty(leaf);
2757 btrfs_free_path(path);
2759 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
2764 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root, u64 num_bytes,
2766 u64 root_objectid, u64 generation,
2767 u64 flags, struct btrfs_disk_key *key,
2768 int level, u64 empty_size, u64 hint_byte,
2769 u64 search_end, struct btrfs_key *ins)
2772 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2773 hint_byte, search_end, ins, 0);
2776 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2777 struct pending_extent_op *extent_op;
2779 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2782 extent_op->type = PENDING_EXTENT_INSERT;
2783 extent_op->bytenr = ins->objectid;
2784 extent_op->num_bytes = ins->offset;
2785 extent_op->level = level;
2786 extent_op->flags = flags;
2787 memcpy(&extent_op->key, key, sizeof(*key));
2789 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2790 ins->objectid + ins->offset - 1,
2791 EXTENT_LOCKED, GFP_NOFS);
2792 set_state_private(&root->fs_info->extent_ins,
2793 ins->objectid, (unsigned long)extent_op);
2795 if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
2796 ins->offset = level;
2797 ins->type = BTRFS_METADATA_ITEM_KEY;
2799 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2802 finish_current_insert(trans, root->fs_info->extent_root);
2803 del_pending_extents(trans, root->fs_info->extent_root);
2809 * helper function to allocate a block for a given tree
2810 * returns the tree buffer or NULL.
2812 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2813 struct btrfs_root *root,
2814 u32 blocksize, u64 root_objectid,
2815 struct btrfs_disk_key *key, int level,
2816 u64 hint, u64 empty_size)
2818 struct btrfs_key ins;
2820 struct extent_buffer *buf;
2822 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2823 trans->transid, 0, key, level,
2824 empty_size, hint, (u64)-1, &ins);
2827 return ERR_PTR(ret);
2830 buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid,
2833 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2834 0, root->root_key.objectid, level, 0);
2836 return ERR_PTR(-ENOMEM);
2838 btrfs_set_buffer_uptodate(buf);
2839 trans->blocks_used++;
2846 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2847 struct btrfs_root *root,
2848 struct extent_buffer *leaf)
2851 u64 leaf_generation;
2852 struct btrfs_key key;
2853 struct btrfs_file_extent_item *fi;
2858 BUG_ON(!btrfs_is_leaf(leaf));
2859 nritems = btrfs_header_nritems(leaf);
2860 leaf_owner = btrfs_header_owner(leaf);
2861 leaf_generation = btrfs_header_generation(leaf);
2863 for (i = 0; i < nritems; i++) {
2866 btrfs_item_key_to_cpu(leaf, &key, i);
2867 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2869 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2870 if (btrfs_file_extent_type(leaf, fi) ==
2871 BTRFS_FILE_EXTENT_INLINE)
2874 * FIXME make sure to insert a trans record that
2875 * repeats the snapshot del on crash
2877 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2878 if (disk_bytenr == 0)
2880 ret = btrfs_free_extent(trans, root, disk_bytenr,
2881 btrfs_file_extent_disk_num_bytes(leaf, fi),
2882 leaf->start, leaf_owner, leaf_generation,
2889 static void noinline reada_walk_down(struct btrfs_root *root,
2890 struct extent_buffer *node,
2903 nritems = btrfs_header_nritems(node);
2904 level = btrfs_header_level(node);
2908 for (i = slot; i < nritems && skipped < 32; i++) {
2909 bytenr = btrfs_node_blockptr(node, i);
2910 if (last && ((bytenr > last && bytenr - last > SZ_32K) ||
2911 (last > bytenr && last - bytenr > SZ_32K))) {
2915 blocksize = btrfs_level_size(root, level - 1);
2917 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2925 mutex_unlock(&root->fs_info->fs_mutex);
2926 ret = readahead_tree_block(root, bytenr, blocksize,
2927 btrfs_node_ptr_generation(node, i));
2928 last = bytenr + blocksize;
2930 mutex_lock(&root->fs_info->fs_mutex);
2937 * helper function for drop_snapshot, this walks down the tree dropping ref
2938 * counts as it goes.
2940 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2941 struct btrfs_root *root,
2942 struct btrfs_path *path, int *level)
2948 struct extent_buffer *next;
2949 struct extent_buffer *cur;
2950 struct extent_buffer *parent;
2955 WARN_ON(*level < 0);
2956 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2957 ret = btrfs_lookup_extent_ref(trans, root,
2958 path->nodes[*level]->start,
2959 path->nodes[*level]->len, &refs);
2965 * walk down to the last node level and free all the leaves
2967 while(*level >= 0) {
2968 WARN_ON(*level < 0);
2969 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2970 cur = path->nodes[*level];
2972 if (btrfs_header_level(cur) != *level)
2975 if (path->slots[*level] >=
2976 btrfs_header_nritems(cur))
2979 ret = drop_leaf_ref(trans, root, cur);
2983 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2984 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2985 blocksize = btrfs_level_size(root, *level - 1);
2986 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2990 parent = path->nodes[*level];
2991 root_owner = btrfs_header_owner(parent);
2992 root_gen = btrfs_header_generation(parent);
2993 path->slots[*level]++;
2994 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2995 parent->start, root_owner,
2996 root_gen, *level - 1, 1);
3000 next = btrfs_find_tree_block(root, bytenr, blocksize);
3001 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3002 free_extent_buffer(next);
3003 reada_walk_down(root, cur, path->slots[*level]);
3004 mutex_unlock(&root->fs_info->fs_mutex);
3005 next = read_tree_block(root, bytenr, blocksize,
3007 mutex_lock(&root->fs_info->fs_mutex);
3008 if (!extent_buffer_uptodate(next)) {
3010 ret = PTR_ERR(next);
3016 WARN_ON(*level <= 0);
3017 if (path->nodes[*level-1])
3018 free_extent_buffer(path->nodes[*level-1]);
3019 path->nodes[*level-1] = next;
3020 *level = btrfs_header_level(next);
3021 path->slots[*level] = 0;
3024 WARN_ON(*level < 0);
3025 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3027 if (path->nodes[*level] == root->node) {
3028 root_owner = root->root_key.objectid;
3029 parent = path->nodes[*level];
3031 parent = path->nodes[*level + 1];
3032 root_owner = btrfs_header_owner(parent);
3035 root_gen = btrfs_header_generation(parent);
3036 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3037 path->nodes[*level]->len, parent->start,
3038 root_owner, root_gen, *level, 1);
3039 free_extent_buffer(path->nodes[*level]);
3040 path->nodes[*level] = NULL;
3047 * helper for dropping snapshots. This walks back up the tree in the path
3048 * to find the first node higher up where we haven't yet gone through
3051 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3052 struct btrfs_root *root,
3053 struct btrfs_path *path, int *level)
3057 struct btrfs_root_item *root_item = &root->root_item;
3062 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3063 slot = path->slots[i];
3064 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3065 struct extent_buffer *node;
3066 struct btrfs_disk_key disk_key;
3067 node = path->nodes[i];
3070 WARN_ON(*level == 0);
3071 btrfs_node_key(node, &disk_key, path->slots[i]);
3072 memcpy(&root_item->drop_progress,
3073 &disk_key, sizeof(disk_key));
3074 root_item->drop_level = i;
3077 struct extent_buffer *parent;
3078 if (path->nodes[*level] == root->node)
3079 parent = path->nodes[*level];
3081 parent = path->nodes[*level + 1];
3083 root_owner = btrfs_header_owner(parent);
3084 root_gen = btrfs_header_generation(parent);
3085 ret = btrfs_free_extent(trans, root,
3086 path->nodes[*level]->start,
3087 path->nodes[*level]->len,
3088 parent->start, root_owner,
3089 root_gen, *level, 1);
3091 free_extent_buffer(path->nodes[*level]);
3092 path->nodes[*level] = NULL;
3101 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3103 struct btrfs_space_info *sinfo;
3104 struct btrfs_block_group_cache *cache;
3111 ret = find_first_extent_bit(&info->block_group_cache, 0,
3112 &start, &end, (unsigned int)-1);
3115 ret = get_state_private(&info->block_group_cache, start, &ptr);
3117 cache = u64_to_ptr(ptr);
3118 if (cache->free_space_ctl) {
3119 btrfs_remove_free_space_cache(cache);
3120 kfree(cache->free_space_ctl);
3124 clear_extent_bits(&info->block_group_cache, start,
3125 end, (unsigned int)-1, GFP_NOFS);
3128 ret = find_first_extent_bit(&info->free_space_cache, 0,
3129 &start, &end, EXTENT_DIRTY);
3132 clear_extent_dirty(&info->free_space_cache, start,
3136 while (!list_empty(&info->space_info)) {
3137 sinfo = list_entry(info->space_info.next,
3138 struct btrfs_space_info, list);
3139 list_del_init(&sinfo->list);
3145 static int find_first_block_group(struct btrfs_root *root,
3146 struct btrfs_path *path, struct btrfs_key *key)
3149 struct btrfs_key found_key;
3150 struct extent_buffer *leaf;
3153 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3157 slot = path->slots[0];
3158 leaf = path->nodes[0];
3159 if (slot >= btrfs_header_nritems(leaf)) {
3160 ret = btrfs_next_leaf(root, path);
3167 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3169 if (found_key.objectid >= key->objectid &&
3170 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3179 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3180 struct btrfs_block_group_cache *cache)
3187 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3188 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3189 cache->bytes_super += stripe_len;
3192 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3193 bytenr = btrfs_sb_offset(i);
3194 ret = btrfs_rmap_block(&fs_info->mapping_tree,
3195 cache->key.objectid, bytenr,
3196 0, &logical, &nr, &stripe_len);
3203 if (logical[nr] > cache->key.objectid +
3207 if (logical[nr] + stripe_len <= cache->key.objectid)
3210 start = logical[nr];
3211 if (start < cache->key.objectid) {
3212 start = cache->key.objectid;
3213 len = (logical[nr] + stripe_len) - start;
3215 len = min_t(u64, stripe_len,
3216 cache->key.objectid +
3217 cache->key.offset - start);
3220 cache->bytes_super += len;
3227 int btrfs_read_block_groups(struct btrfs_root *root)
3229 struct btrfs_path *path;
3232 struct btrfs_block_group_cache *cache;
3233 struct btrfs_fs_info *info = root->fs_info;
3234 struct btrfs_space_info *space_info;
3235 struct extent_io_tree *block_group_cache;
3236 struct btrfs_key key;
3237 struct btrfs_key found_key;
3238 struct extent_buffer *leaf;
3240 block_group_cache = &info->block_group_cache;
3242 root = info->extent_root;
3245 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3246 path = btrfs_alloc_path();
3251 ret = find_first_block_group(root, path, &key);
3259 leaf = path->nodes[0];
3260 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3261 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3267 read_extent_buffer(leaf, &cache->item,
3268 btrfs_item_ptr_offset(leaf, path->slots[0]),
3269 sizeof(cache->item));
3270 memcpy(&cache->key, &found_key, sizeof(found_key));
3273 key.objectid = found_key.objectid + found_key.offset;
3274 if (found_key.offset == 0)
3276 btrfs_release_path(path);
3277 cache->flags = btrfs_block_group_flags(&cache->item);
3279 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3280 bit = BLOCK_GROUP_DATA;
3281 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3282 bit = BLOCK_GROUP_SYSTEM;
3283 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3284 bit = BLOCK_GROUP_METADATA;
3286 set_avail_alloc_bits(info, cache->flags);
3287 if (btrfs_chunk_readonly(root, cache->key.objectid))
3290 account_super_bytes(info, cache);
3292 ret = update_space_info(info, cache->flags, found_key.offset,
3293 btrfs_block_group_used(&cache->item),
3296 cache->space_info = space_info;
3298 /* use EXTENT_LOCKED to prevent merging */
3299 set_extent_bits(block_group_cache, found_key.objectid,
3300 found_key.objectid + found_key.offset - 1,
3301 bit | EXTENT_LOCKED, GFP_NOFS);
3302 set_state_private(block_group_cache, found_key.objectid,
3303 (unsigned long)cache);
3307 btrfs_free_path(path);
3311 struct btrfs_block_group_cache *
3312 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3313 u64 chunk_objectid, u64 chunk_offset, u64 size)
3317 struct btrfs_block_group_cache *cache;
3318 struct extent_io_tree *block_group_cache;
3320 block_group_cache = &fs_info->block_group_cache;
3322 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3324 cache->key.objectid = chunk_offset;
3325 cache->key.offset = size;
3327 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3328 btrfs_set_block_group_used(&cache->item, bytes_used);
3329 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3330 cache->flags = type;
3331 btrfs_set_block_group_flags(&cache->item, type);
3333 account_super_bytes(fs_info, cache);
3334 ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3335 &cache->space_info);
3338 bit = block_group_state_bits(type);
3339 ret = set_extent_bits(block_group_cache, chunk_offset,
3340 chunk_offset + size - 1,
3341 bit | EXTENT_LOCKED, GFP_NOFS);
3344 ret = set_state_private(block_group_cache, chunk_offset,
3345 (unsigned long)cache);
3347 set_avail_alloc_bits(fs_info, type);
3352 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3353 struct btrfs_root *root, u64 bytes_used,
3354 u64 type, u64 chunk_objectid, u64 chunk_offset,
3358 struct btrfs_root *extent_root;
3359 struct btrfs_block_group_cache *cache;
3361 cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3362 chunk_objectid, chunk_offset, size);
3363 extent_root = root->fs_info->extent_root;
3364 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3365 sizeof(cache->item));
3368 ret = finish_current_insert(trans, extent_root);
3370 ret = del_pending_extents(trans, extent_root);
3377 * This is for converter use only.
3379 * In that case, we don't know where are free blocks located.
3380 * Therefore all block group cache entries must be setup properly
3381 * before doing any block allocation.
3383 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3384 struct btrfs_root *root)
3392 u64 total_metadata = 0;
3396 struct btrfs_root *extent_root;
3397 struct btrfs_block_group_cache *cache;
3398 struct extent_io_tree *block_group_cache;
3400 extent_root = root->fs_info->extent_root;
3401 block_group_cache = &root->fs_info->block_group_cache;
3402 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3403 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3404 group_align = 64 * root->sectorsize;
3407 while (cur_start < total_bytes) {
3408 group_size = total_bytes / 12;
3409 group_size = min_t(u64, group_size, total_bytes - cur_start);
3410 if (cur_start == 0) {
3411 bit = BLOCK_GROUP_SYSTEM;
3412 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3414 group_size &= ~(group_align - 1);
3415 group_size = max_t(u64, group_size, SZ_8M);
3416 group_size = min_t(u64, group_size, SZ_32M);
3418 group_size &= ~(group_align - 1);
3419 if (total_data >= total_metadata * 2) {
3420 group_type = BTRFS_BLOCK_GROUP_METADATA;
3421 group_size = min_t(u64, group_size, SZ_1G);
3422 total_metadata += group_size;
3424 group_type = BTRFS_BLOCK_GROUP_DATA;
3425 group_size = min_t(u64, group_size,
3427 total_data += group_size;
3429 if ((total_bytes - cur_start) * 4 < group_size * 5)
3430 group_size = total_bytes - cur_start;
3433 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3436 cache->key.objectid = cur_start;
3437 cache->key.offset = group_size;
3438 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3440 btrfs_set_block_group_used(&cache->item, 0);
3441 btrfs_set_block_group_chunk_objectid(&cache->item,
3443 btrfs_set_block_group_flags(&cache->item, group_type);
3445 cache->flags = group_type;
3447 ret = update_space_info(root->fs_info, group_type, group_size,
3448 0, &cache->space_info);
3450 set_avail_alloc_bits(extent_root->fs_info, group_type);
3452 set_extent_bits(block_group_cache, cur_start,
3453 cur_start + group_size - 1,
3454 bit | EXTENT_LOCKED, GFP_NOFS);
3455 set_state_private(block_group_cache, cur_start,
3456 (unsigned long)cache);
3457 cur_start += group_size;
3459 /* then insert all the items */
3461 while(cur_start < total_bytes) {
3462 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3465 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3466 sizeof(cache->item));
3469 finish_current_insert(trans, extent_root);
3470 ret = del_pending_extents(trans, extent_root);
3473 cur_start = cache->key.objectid + cache->key.offset;
3478 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3479 struct btrfs_root *root,
3480 u64 bytenr, u64 num_bytes, int alloc,
3483 return update_block_group(trans, root, bytenr, num_bytes,
3488 * Just remove a block group item in extent tree
3489 * Caller should ensure the block group is empty and all space is pinned.
3490 * Or new tree block/data may be allocated into it.
3492 static int free_block_group_item(struct btrfs_trans_handle *trans,
3493 struct btrfs_fs_info *fs_info,
3494 u64 bytenr, u64 len)
3496 struct btrfs_path *path;
3497 struct btrfs_key key;
3498 struct btrfs_root *root = fs_info->extent_root;
3501 key.objectid = bytenr;
3503 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3505 path = btrfs_alloc_path();
3509 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3517 ret = btrfs_del_item(trans, root, path);
3519 btrfs_free_path(path);
3523 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3524 struct btrfs_fs_info *fs_info,
3525 u64 devid, u64 dev_offset)
3527 struct btrfs_root *root = fs_info->dev_root;
3528 struct btrfs_path *path;
3529 struct btrfs_key key;
3532 path = btrfs_alloc_path();
3536 key.objectid = devid;
3537 key.type = BTRFS_DEV_EXTENT_KEY;
3538 key.offset = dev_offset;
3540 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3548 ret = btrfs_del_item(trans, root, path);
3550 btrfs_free_path(path);
3554 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3555 struct btrfs_fs_info *fs_info,
3558 struct btrfs_chunk *chunk = NULL;
3559 struct btrfs_root *root= fs_info->chunk_root;
3560 struct btrfs_path *path;
3561 struct btrfs_key key;
3566 path = btrfs_alloc_path();
3570 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3571 key.type = BTRFS_CHUNK_ITEM_KEY;
3572 key.offset = chunk_offset;
3574 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3581 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3582 struct btrfs_chunk);
3583 num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3584 for (i = 0; i < num_stripes; i++) {
3585 ret = free_dev_extent_item(trans, fs_info,
3586 btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3587 btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3592 btrfs_free_path(path);
3596 static int free_system_chunk_item(struct btrfs_super_block *super,
3597 struct btrfs_key *key)
3599 struct btrfs_disk_key *disk_key;
3600 struct btrfs_key cpu_key;
3601 u32 array_size = btrfs_super_sys_array_size(super);
3602 char *ptr = (char *)super->sys_chunk_array;
3606 while (cur < btrfs_super_sys_array_size(super)) {
3607 struct btrfs_chunk *chunk;
3611 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3612 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3613 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3619 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3620 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3621 chunk_len = btrfs_chunk_item_size(num_stripes) +
3624 if (key->objectid == cpu_key.objectid &&
3625 key->offset == cpu_key.offset &&
3626 key->type == cpu_key.type) {
3627 memmove(ptr + cur, ptr + cur + chunk_len,
3628 array_size - cur - chunk_len);
3629 array_size -= chunk_len;
3630 btrfs_set_super_sys_array_size(super, array_size);
3641 static int free_chunk_item(struct btrfs_trans_handle *trans,
3642 struct btrfs_fs_info *fs_info,
3643 u64 bytenr, u64 len)
3645 struct btrfs_path *path;
3646 struct btrfs_key key;
3647 struct btrfs_root *root = fs_info->chunk_root;
3648 struct btrfs_chunk *chunk;
3652 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3653 key.offset = bytenr;
3654 key.type = BTRFS_CHUNK_ITEM_KEY;
3656 path = btrfs_alloc_path();
3660 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3667 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3668 struct btrfs_chunk);
3669 chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3671 ret = btrfs_del_item(trans, root, path);
3675 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3676 ret = free_system_chunk_item(fs_info->super_copy, &key);
3678 btrfs_free_path(path);
3682 static u64 get_dev_extent_len(struct map_lookup *map)
3686 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3687 case 0: /* Single */
3688 case BTRFS_BLOCK_GROUP_DUP:
3689 case BTRFS_BLOCK_GROUP_RAID1:
3692 case BTRFS_BLOCK_GROUP_RAID5:
3693 div = (map->num_stripes - 1);
3695 case BTRFS_BLOCK_GROUP_RAID6:
3696 div = (map->num_stripes - 2);
3698 case BTRFS_BLOCK_GROUP_RAID10:
3699 div = (map->num_stripes / map->sub_stripes);
3702 /* normally, read chunk security hook should handled it */
3705 return map->ce.size / div;
3708 /* free block group/chunk related caches */
3709 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3710 struct btrfs_fs_info *fs_info,
3711 u64 bytenr, u64 len)
3713 struct btrfs_block_group_cache *cache;
3714 struct cache_extent *ce;
3715 struct map_lookup *map;
3720 /* Free block group cache first */
3721 cache = btrfs_lookup_block_group(fs_info, bytenr);
3724 flags = cache->flags;
3725 if (cache->free_space_ctl) {
3726 btrfs_remove_free_space_cache(cache);
3727 kfree(cache->free_space_ctl);
3729 clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len,
3730 (unsigned int)-1, GFP_NOFS);
3731 ret = free_space_info(fs_info, flags, len, 0, NULL);
3736 /* Then free mapping info and dev usage info */
3737 ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3738 if (!ce || ce->start != bytenr) {
3742 map = container_of(ce, struct map_lookup, ce);
3743 for (i = 0; i < map->num_stripes; i++) {
3744 struct btrfs_device *device;
3746 device = map->stripes[i].dev;
3747 device->bytes_used -= get_dev_extent_len(map);
3748 ret = btrfs_update_device(trans, device);
3752 remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3758 int btrfs_free_block_group(struct btrfs_trans_handle *trans,
3759 struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
3761 struct btrfs_root *extent_root = fs_info->extent_root;
3762 struct btrfs_path *path;
3763 struct btrfs_block_group_item *bgi;
3764 struct btrfs_key key;
3767 path = btrfs_alloc_path();
3771 key.objectid = bytenr;
3772 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3775 /* Double check the block group to ensure it's empty */
3776 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3784 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3785 struct btrfs_block_group_item);
3786 if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
3788 "WARNING: block group [%llu,%llu) is not empty\n",
3789 bytenr, bytenr + len);
3793 btrfs_release_path(path);
3796 * Now pin all space in the block group, to prevent further transaction
3797 * allocate space from it.
3798 * Every operation needs a transaction must be in the range.
3800 btrfs_pin_extent(fs_info, bytenr, len);
3802 /* delete block group item and chunk item */
3803 ret = free_block_group_item(trans, fs_info, bytenr, len);
3806 "failed to free block group item for [%llu,%llu)\n",
3807 bytenr, bytenr + len);
3808 btrfs_unpin_extent(fs_info, bytenr, len);
3812 ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
3815 "failed to dev extents belongs to [%llu,%llu)\n",
3816 bytenr, bytenr + len);
3817 btrfs_unpin_extent(fs_info, bytenr, len);
3820 ret = free_chunk_item(trans, fs_info, bytenr, len);
3823 "failed to free chunk for [%llu,%llu)\n",
3824 bytenr, bytenr + len);
3825 btrfs_unpin_extent(fs_info, bytenr, len);
3829 /* Now release the block_group_cache */
3830 ret = free_block_group_cache(trans, fs_info, bytenr, len);
3831 btrfs_unpin_extent(fs_info, bytenr, len);
3834 btrfs_free_path(path);
3839 * Fixup block accounting. The initial block accounting created by
3840 * make_block_groups isn't accuracy in this case.
3842 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3843 struct btrfs_root *root)
3849 struct btrfs_path path;
3850 struct btrfs_key key;
3851 struct extent_buffer *leaf;
3852 struct btrfs_block_group_cache *cache;
3853 struct btrfs_fs_info *fs_info = root->fs_info;
3855 root = root->fs_info->extent_root;
3857 while(extent_root_pending_ops(fs_info)) {
3858 ret = finish_current_insert(trans, root);
3861 ret = del_pending_extents(trans, root);
3867 cache = btrfs_lookup_first_block_group(fs_info, start);
3870 start = cache->key.objectid + cache->key.offset;
3871 btrfs_set_block_group_used(&cache->item, 0);
3872 cache->space_info->bytes_used = 0;
3873 set_extent_bits(&root->fs_info->block_group_cache,
3874 cache->key.objectid,
3875 cache->key.objectid + cache->key.offset -1,
3876 BLOCK_GROUP_DIRTY, GFP_NOFS);
3879 btrfs_init_path(&path);
3882 key.type = BTRFS_EXTENT_ITEM_KEY;
3883 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3888 leaf = path.nodes[0];
3889 slot = path.slots[0];
3890 if (slot >= btrfs_header_nritems(leaf)) {
3891 ret = btrfs_next_leaf(root, &path);
3896 leaf = path.nodes[0];
3897 slot = path.slots[0];
3899 btrfs_item_key_to_cpu(leaf, &key, slot);
3900 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3901 bytes_used += key.offset;
3902 ret = btrfs_update_block_group(trans, root,
3903 key.objectid, key.offset, 1, 0);
3905 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3906 bytes_used += root->nodesize;
3907 ret = btrfs_update_block_group(trans, root,
3908 key.objectid, root->nodesize, 1, 0);
3913 btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3914 btrfs_release_path(&path);
3918 static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
3919 u64 *start, u64 *len)
3921 struct btrfs_key key;
3923 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3924 BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
3925 key.type == BTRFS_METADATA_ITEM_KEY));
3926 *start = key.objectid;
3927 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3930 *len = root->nodesize;
3934 * Find first overlap extent for range [bytenr, bytenr + len)
3935 * Return 0 for found and point path to it.
3936 * Return >0 for not found.
3939 int btrfs_search_overlap_extent(struct btrfs_root *root,
3940 struct btrfs_path *path, u64 bytenr, u64 len)
3942 struct btrfs_key key;
3947 key.objectid = bytenr;
3948 key.type = BTRFS_EXTENT_DATA_KEY;
3949 key.offset = (u64)-1;
3951 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3956 ret = btrfs_previous_extent_item(root, path, 0);
3959 /* no previous, check next extent */
3962 __get_extent_size(root, path, &cur_start, &cur_len);
3964 if (cur_start + cur_len > bytenr)
3968 ret = btrfs_next_extent_item(root, path, bytenr + len);
3971 /* No next, prev already checked, no overlap */
3974 __get_extent_size(root, path, &cur_start, &cur_len);
3976 if (cur_start < bytenr + len)
3981 static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3982 struct btrfs_root *root, u64 objectid,
3983 struct btrfs_inode_item *inode,
3984 u64 file_pos, u64 disk_bytenr,
3988 struct btrfs_fs_info *info = root->fs_info;
3989 struct btrfs_root *extent_root = info->extent_root;
3990 struct extent_buffer *leaf;
3991 struct btrfs_file_extent_item *fi;
3992 struct btrfs_key ins_key;
3993 struct btrfs_path *path;
3994 struct btrfs_extent_item *ei;
3996 u64 extent_num_bytes;
3999 u64 num_bytes = *ret_num_bytes;
4002 * All supported file system should not use its 0 extent.
4005 * And hole extent has no size limit, no need to loop.
4007 if (disk_bytenr == 0) {
4008 ret = btrfs_insert_file_extent(trans, root, objectid,
4009 file_pos, disk_bytenr,
4010 num_bytes, num_bytes);
4013 num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
4015 path = btrfs_alloc_path();
4019 /* First to check extent overlap */
4020 ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
4029 __get_extent_size(extent_root, path, &cur_start, &cur_len);
4031 * For convert case, this extent should be a subset of
4034 BUG_ON(disk_bytenr < cur_start);
4036 extent_bytenr = cur_start;
4037 extent_num_bytes = cur_len;
4038 extent_offset = disk_bytenr - extent_bytenr;
4040 /* No overlap, create new extent */
4041 btrfs_release_path(path);
4042 ins_key.objectid = disk_bytenr;
4043 ins_key.offset = num_bytes;
4044 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4046 ret = btrfs_insert_empty_item(trans, extent_root, path,
4047 &ins_key, sizeof(*ei));
4049 leaf = path->nodes[0];
4050 ei = btrfs_item_ptr(leaf, path->slots[0],
4051 struct btrfs_extent_item);
4053 btrfs_set_extent_refs(leaf, ei, 0);
4054 btrfs_set_extent_generation(leaf, ei, 0);
4055 btrfs_set_extent_flags(leaf, ei,
4056 BTRFS_EXTENT_FLAG_DATA);
4057 btrfs_mark_buffer_dirty(leaf);
4059 ret = btrfs_update_block_group(trans, root, disk_bytenr,
4063 } else if (ret != -EEXIST) {
4066 btrfs_extent_post_op(trans, extent_root);
4067 extent_bytenr = disk_bytenr;
4068 extent_num_bytes = num_bytes;
4071 btrfs_release_path(path);
4072 ins_key.objectid = objectid;
4073 ins_key.offset = file_pos;
4074 ins_key.type = BTRFS_EXTENT_DATA_KEY;
4075 ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
4079 leaf = path->nodes[0];
4080 fi = btrfs_item_ptr(leaf, path->slots[0],
4081 struct btrfs_file_extent_item);
4082 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4083 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
4084 btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
4085 btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
4086 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
4087 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
4088 btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
4089 btrfs_set_file_extent_compression(leaf, fi, 0);
4090 btrfs_set_file_extent_encryption(leaf, fi, 0);
4091 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
4092 btrfs_mark_buffer_dirty(leaf);
4094 nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
4095 btrfs_set_stack_inode_nbytes(inode, nbytes);
4096 btrfs_release_path(path);
4098 ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
4099 0, root->root_key.objectid, objectid,
4100 file_pos - extent_offset);
4104 *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
4106 btrfs_free_path(path);
4111 * Record a file extent. Do all the required works, such as inserting
4112 * file extent item, inserting extent item and backref item into extent
4113 * tree and updating block accounting.
4115 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *root, u64 objectid,
4117 struct btrfs_inode_item *inode,
4118 u64 file_pos, u64 disk_bytenr,
4121 u64 cur_disk_bytenr = disk_bytenr;
4122 u64 cur_file_pos = file_pos;
4123 u64 cur_num_bytes = num_bytes;
4126 while (num_bytes > 0) {
4127 ret = __btrfs_record_file_extent(trans, root, objectid,
4128 inode, cur_file_pos,
4133 cur_disk_bytenr += cur_num_bytes;
4134 cur_file_pos += cur_num_bytes;
4135 num_bytes -= cur_num_bytes;
4141 static int add_excluded_extent(struct btrfs_root *root,
4142 u64 start, u64 num_bytes)
4144 u64 end = start + num_bytes - 1;
4145 set_extent_bits(&root->fs_info->pinned_extents,
4146 start, end, EXTENT_UPTODATE, GFP_NOFS);
4150 void free_excluded_extents(struct btrfs_root *root,
4151 struct btrfs_block_group_cache *cache)
4155 start = cache->key.objectid;
4156 end = start + cache->key.offset - 1;
4158 clear_extent_bits(&root->fs_info->pinned_extents,
4159 start, end, EXTENT_UPTODATE, GFP_NOFS);
4162 int exclude_super_stripes(struct btrfs_root *root,
4163 struct btrfs_block_group_cache *cache)
4170 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
4171 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
4172 cache->bytes_super += stripe_len;
4173 ret = add_excluded_extent(root, cache->key.objectid,
4179 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4180 bytenr = btrfs_sb_offset(i);
4181 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4182 cache->key.objectid, bytenr,
4183 0, &logical, &nr, &stripe_len);
4190 if (logical[nr] > cache->key.objectid +
4194 if (logical[nr] + stripe_len <= cache->key.objectid)
4197 start = logical[nr];
4198 if (start < cache->key.objectid) {
4199 start = cache->key.objectid;
4200 len = (logical[nr] + stripe_len) - start;
4202 len = min_t(u64, stripe_len,
4203 cache->key.objectid +
4204 cache->key.offset - start);
4207 cache->bytes_super += len;
4208 ret = add_excluded_extent(root, start, len);
4220 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
4221 struct btrfs_fs_info *info, u64 start, u64 end)
4223 u64 extent_start, extent_end, size, total_added = 0;
4226 while (start < end) {
4227 ret = find_first_extent_bit(&info->pinned_extents, start,
4228 &extent_start, &extent_end,
4229 EXTENT_DIRTY | EXTENT_UPTODATE);
4233 if (extent_start <= start) {
4234 start = extent_end + 1;
4235 } else if (extent_start > start && extent_start < end) {
4236 size = extent_start - start;
4237 total_added += size;
4238 ret = btrfs_add_free_space(block_group->free_space_ctl,
4240 BUG_ON(ret); /* -ENOMEM or logic error */
4241 start = extent_end + 1;
4249 total_added += size;
4250 ret = btrfs_add_free_space(block_group->free_space_ctl, start,
4252 BUG_ON(ret); /* -ENOMEM or logic error */