2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
21 #include "kerncompat.h"
22 #include "radix-tree.h"
25 #include "print-tree.h"
26 #include "transaction.h"
30 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
37 btrfs_root *extent_root);
38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
39 btrfs_root *extent_root);
40 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
41 struct btrfs_root *root, u64 bytes_used,
42 u64 type, u64 chunk_tree, u64 chunk_objectid,
45 static int cache_block_group(struct btrfs_root *root,
46 struct btrfs_block_group_cache *block_group)
48 struct btrfs_path *path;
51 struct extent_buffer *leaf;
52 struct extent_io_tree *free_space_cache;
62 root = root->fs_info->extent_root;
63 free_space_cache = &root->fs_info->free_space_cache;
65 if (block_group->cached)
68 path = btrfs_alloc_path();
73 first_free = block_group->key.objectid;
74 key.objectid = block_group->key.objectid;
76 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
77 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
80 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
84 leaf = path->nodes[0];
85 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
86 if (key.objectid + key.offset > first_free)
87 first_free = key.objectid + key.offset;
90 leaf = path->nodes[0];
91 slot = path->slots[0];
92 if (slot >= btrfs_header_nritems(leaf)) {
93 ret = btrfs_next_leaf(root, path);
102 btrfs_item_key_to_cpu(leaf, &key, slot);
103 if (key.objectid < block_group->key.objectid) {
106 if (key.objectid >= block_group->key.objectid +
107 block_group->key.offset) {
111 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
116 if (key.objectid > last) {
117 hole_size = key.objectid - last;
118 set_extent_dirty(free_space_cache, last,
119 last + hole_size - 1,
122 last = key.objectid + key.offset;
130 if (block_group->key.objectid +
131 block_group->key.offset > last) {
132 hole_size = block_group->key.objectid +
133 block_group->key.offset - last;
134 set_extent_dirty(free_space_cache, last,
135 last + hole_size - 1, GFP_NOFS);
137 block_group->cached = 1;
139 btrfs_free_path(path);
143 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
147 struct extent_io_tree *block_group_cache;
148 struct btrfs_block_group_cache *block_group = NULL;
154 block_group_cache = &info->block_group_cache;
155 ret = find_first_extent_bit(block_group_cache,
156 bytenr, &start, &end,
157 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
162 ret = get_state_private(block_group_cache, start, &ptr);
166 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
167 if (block_group->key.objectid <= bytenr && bytenr <
168 block_group->key.objectid + block_group->key.offset)
173 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
175 return (cache->flags & bits);
178 static int noinline find_search_start(struct btrfs_root *root,
179 struct btrfs_block_group_cache **cache_ret,
180 u64 *start_ret, int num, int data)
183 struct btrfs_block_group_cache *cache = *cache_ret;
188 u64 search_start = *start_ret;
195 ret = cache_block_group(root, cache);
199 last = max(search_start, cache->key.objectid);
200 if (!block_group_bits(cache, data)) {
205 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
206 last, &start, &end, EXTENT_DIRTY);
213 start = max(last, start);
215 if (last - start < num) {
216 if (last == cache->key.objectid + cache->key.offset)
220 if (start + num > cache->key.objectid + cache->key.offset)
226 cache = btrfs_lookup_block_group(root->fs_info, search_start);
228 printk("Unable to find block group for %Lu\n", search_start);
234 last = cache->key.objectid + cache->key.offset;
236 cache = btrfs_lookup_block_group(root->fs_info, last);
246 if (cache_miss && !cache->cached) {
247 cache_block_group(root, cache);
249 cache = btrfs_lookup_block_group(root->fs_info, last);
251 cache = btrfs_find_block_group(root, cache, last, data, 0);
259 static u64 div_factor(u64 num, int factor)
268 static int block_group_state_bits(u64 flags)
271 if (flags & BTRFS_BLOCK_GROUP_DATA)
272 bits |= BLOCK_GROUP_DATA;
273 if (flags & BTRFS_BLOCK_GROUP_METADATA)
274 bits |= BLOCK_GROUP_METADATA;
275 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
276 bits |= BLOCK_GROUP_SYSTEM;
280 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
281 struct btrfs_block_group_cache
282 *hint, u64 search_start,
285 struct btrfs_block_group_cache *cache;
286 struct extent_io_tree *block_group_cache;
287 struct btrfs_block_group_cache *found_group = NULL;
288 struct btrfs_fs_info *info = root->fs_info;
301 block_group_cache = &info->block_group_cache;
306 bit = block_group_state_bits(data);
309 struct btrfs_block_group_cache *shint;
310 shint = btrfs_lookup_block_group(info, search_start);
311 if (shint && block_group_bits(shint, data)) {
312 used = btrfs_block_group_used(&shint->item);
313 if (used + shint->pinned <
314 div_factor(shint->key.offset, factor)) {
319 if (hint && block_group_bits(hint, data)) {
320 used = btrfs_block_group_used(&hint->item);
321 if (used + hint->pinned <
322 div_factor(hint->key.offset, factor)) {
325 last = hint->key.objectid + hint->key.offset;
329 hint_last = max(hint->key.objectid, search_start);
331 hint_last = search_start;
337 ret = find_first_extent_bit(block_group_cache, last,
342 ret = get_state_private(block_group_cache, start, &ptr);
346 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
347 last = cache->key.objectid + cache->key.offset;
348 used = btrfs_block_group_used(&cache->item);
351 free_check = cache->key.offset;
353 free_check = div_factor(cache->key.offset, factor);
355 if (used + cache->pinned < free_check) {
370 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
371 u64 owner, u64 owner_offset)
373 u32 high_crc = ~(u32)0;
374 u32 low_crc = ~(u32)0;
377 lenum = cpu_to_le64(root_objectid);
378 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
379 lenum = cpu_to_le64(ref_generation);
380 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
381 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
382 lenum = cpu_to_le64(owner);
383 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
384 lenum = cpu_to_le64(owner_offset);
385 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
387 return ((u64)high_crc << 32) | (u64)low_crc;
390 static int match_extent_ref(struct extent_buffer *leaf,
391 struct btrfs_extent_ref *disk_ref,
392 struct btrfs_extent_ref *cpu_ref)
397 if (cpu_ref->objectid)
398 len = sizeof(*cpu_ref);
400 len = 2 * sizeof(u64);
401 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
406 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
407 struct btrfs_root *root,
408 struct btrfs_path *path, u64 bytenr,
410 u64 ref_generation, u64 owner,
411 u64 owner_offset, int del)
414 struct btrfs_key key;
415 struct btrfs_key found_key;
416 struct btrfs_extent_ref ref;
417 struct extent_buffer *leaf;
418 struct btrfs_extent_ref *disk_ref;
422 btrfs_set_stack_ref_root(&ref, root_objectid);
423 btrfs_set_stack_ref_generation(&ref, ref_generation);
424 btrfs_set_stack_ref_objectid(&ref, owner);
425 btrfs_set_stack_ref_offset(&ref, owner_offset);
427 hash = hash_extent_ref(root_objectid, ref_generation, owner,
430 key.objectid = bytenr;
431 key.type = BTRFS_EXTENT_REF_KEY;
434 ret = btrfs_search_slot(trans, root, &key, path,
438 leaf = path->nodes[0];
440 u32 nritems = btrfs_header_nritems(leaf);
441 if (path->slots[0] >= nritems) {
442 ret2 = btrfs_next_leaf(root, path);
445 leaf = path->nodes[0];
447 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
448 if (found_key.objectid != bytenr ||
449 found_key.type != BTRFS_EXTENT_REF_KEY)
451 key.offset = found_key.offset;
453 btrfs_release_path(root, path);
457 disk_ref = btrfs_item_ptr(path->nodes[0],
459 struct btrfs_extent_ref);
460 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
464 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
465 key.offset = found_key.offset + 1;
466 btrfs_release_path(root, path);
473 * Back reference rules. Back refs have three main goals:
475 * 1) differentiate between all holders of references to an extent so that
476 * when a reference is dropped we can make sure it was a valid reference
477 * before freeing the extent.
479 * 2) Provide enough information to quickly find the holders of an extent
480 * if we notice a given block is corrupted or bad.
482 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
483 * maintenance. This is actually the same as #2, but with a slightly
484 * different use case.
486 * File extents can be referenced by:
488 * - multiple snapshots, subvolumes, or different generations in one subvol
489 * - different files inside a single subvolume (in theory, not implemented yet)
490 * - different offsets inside a file (bookend extents in file.c)
492 * The extent ref structure has fields for:
494 * - Objectid of the subvolume root
495 * - Generation number of the tree holding the reference
496 * - objectid of the file holding the reference
497 * - offset in the file corresponding to the key holding the reference
499 * When a file extent is allocated the fields are filled in:
500 * (root_key.objectid, trans->transid, inode objectid, offset in file)
502 * When a leaf is cow'd new references are added for every file extent found
503 * in the leaf. It looks the same as the create case, but trans->transid
504 * will be different when the block is cow'd.
506 * (root_key.objectid, trans->transid, inode objectid, offset in file)
508 * When a file extent is removed either during snapshot deletion or file
509 * truncation, the corresponding back reference is found
512 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
513 * inode objectid, offset in file)
515 * Btree extents can be referenced by:
517 * - Different subvolumes
518 * - Different generations of the same subvolume
520 * Storing sufficient information for a full reverse mapping of a btree
521 * block would require storing the lowest key of the block in the backref,
522 * and it would require updating that lowest key either before write out or
523 * every time it changed. Instead, the objectid of the lowest key is stored
524 * along with the level of the tree block. This provides a hint
525 * about where in the btree the block can be found. Searches through the
526 * btree only need to look for a pointer to that block, so they stop one
527 * level higher than the level recorded in the backref.
529 * Some btrees do not do reference counting on their extents. These
530 * include the extent tree and the tree of tree roots. Backrefs for these
531 * trees always have a generation of zero.
533 * When a tree block is created, back references are inserted:
535 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
537 * When a tree block is cow'd in a reference counted root,
538 * new back references are added for all the blocks it points to.
539 * These are of the form (trans->transid will have increased since creation):
541 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
543 * Because the lowest_key_objectid and the level are just hints
544 * they are not used when backrefs are deleted. When a backref is deleted:
546 * if backref was for a tree root:
547 * root_objectid = root->root_key.objectid
549 * root_objectid = btrfs_header_owner(parent)
551 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
553 * Back Reference Key hashing:
555 * Back references have four fields, each 64 bits long. Unfortunately,
556 * This is hashed into a single 64 bit number and placed into the key offset.
557 * The key objectid corresponds to the first byte in the extent, and the
558 * key type is set to BTRFS_EXTENT_REF_KEY
560 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
561 struct btrfs_root *root,
562 struct btrfs_path *path, u64 bytenr,
563 u64 root_objectid, u64 ref_generation,
564 u64 owner, u64 owner_offset)
567 struct btrfs_key key;
568 struct btrfs_extent_ref ref;
569 struct btrfs_extent_ref *disk_ref;
572 btrfs_set_stack_ref_root(&ref, root_objectid);
573 btrfs_set_stack_ref_generation(&ref, ref_generation);
574 btrfs_set_stack_ref_objectid(&ref, owner);
575 btrfs_set_stack_ref_offset(&ref, owner_offset);
577 hash = hash_extent_ref(root_objectid, ref_generation, owner,
580 key.objectid = bytenr;
581 key.type = BTRFS_EXTENT_REF_KEY;
583 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
584 while (ret == -EEXIST) {
585 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
586 struct btrfs_extent_ref);
587 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
590 btrfs_release_path(root, path);
591 ret = btrfs_insert_empty_item(trans, root, path, &key,
596 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
597 struct btrfs_extent_ref);
598 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
600 btrfs_mark_buffer_dirty(path->nodes[0]);
602 btrfs_release_path(root, path);
606 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
607 struct btrfs_root *root,
608 u64 bytenr, u64 num_bytes,
609 u64 root_objectid, u64 ref_generation,
610 u64 owner, u64 owner_offset)
612 struct btrfs_path *path;
614 struct btrfs_key key;
615 struct extent_buffer *l;
616 struct btrfs_extent_item *item;
619 WARN_ON(num_bytes < root->sectorsize);
620 path = btrfs_alloc_path();
624 key.objectid = bytenr;
625 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
626 key.offset = num_bytes;
627 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
636 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
637 refs = btrfs_extent_refs(l, item);
638 btrfs_set_extent_refs(l, item, refs + 1);
639 btrfs_mark_buffer_dirty(path->nodes[0]);
641 btrfs_release_path(root->fs_info->extent_root, path);
643 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
644 path, bytenr, root_objectid,
645 ref_generation, owner, owner_offset);
647 finish_current_insert(trans, root->fs_info->extent_root);
648 del_pending_extents(trans, root->fs_info->extent_root);
650 btrfs_free_path(path);
654 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
655 struct btrfs_root *root)
657 finish_current_insert(trans, root->fs_info->extent_root);
658 del_pending_extents(trans, root->fs_info->extent_root);
662 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
663 struct btrfs_root *root, u64 bytenr,
664 u64 num_bytes, u32 *refs)
666 struct btrfs_path *path;
668 struct btrfs_key key;
669 struct extent_buffer *l;
670 struct btrfs_extent_item *item;
672 WARN_ON(num_bytes < root->sectorsize);
673 path = btrfs_alloc_path();
674 key.objectid = bytenr;
675 key.offset = num_bytes;
676 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
677 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
682 btrfs_print_leaf(root, path->nodes[0]);
683 printk("failed to find block number %Lu\n", bytenr);
687 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
688 *refs = btrfs_extent_refs(l, item);
690 btrfs_free_path(path);
694 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
695 struct btrfs_path *count_path,
698 struct btrfs_root *extent_root = root->fs_info->extent_root;
699 struct btrfs_path *path;
702 u64 root_objectid = root->root_key.objectid;
708 struct btrfs_key key;
709 struct btrfs_key found_key;
710 struct extent_buffer *l;
711 struct btrfs_extent_item *item;
712 struct btrfs_extent_ref *ref_item;
715 path = btrfs_alloc_path();
718 bytenr = first_extent;
720 bytenr = count_path->nodes[level]->start;
723 key.objectid = bytenr;
726 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
727 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
733 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
735 if (found_key.objectid != bytenr ||
736 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
740 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
741 refs = btrfs_extent_refs(l, item);
743 nritems = btrfs_header_nritems(l);
744 if (path->slots[0] >= nritems) {
745 ret = btrfs_next_leaf(extent_root, path);
750 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
751 if (found_key.objectid != bytenr)
753 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
759 ref_item = btrfs_item_ptr(l, path->slots[0],
760 struct btrfs_extent_ref);
761 found_objectid = btrfs_ref_root(l, ref_item);
763 if (found_objectid != root_objectid) {
770 if (cur_count == 0) {
774 if (level >= 0 && root->node == count_path->nodes[level])
777 btrfs_release_path(root, path);
781 btrfs_free_path(path);
784 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
785 struct btrfs_root *root, u64 owner_objectid)
791 struct btrfs_disk_key disk_key;
793 level = btrfs_header_level(root->node);
794 generation = trans->transid;
795 nritems = btrfs_header_nritems(root->node);
798 btrfs_item_key(root->node, &disk_key, 0);
800 btrfs_node_key(root->node, &disk_key, 0);
801 key_objectid = btrfs_disk_key_objectid(&disk_key);
805 return btrfs_inc_extent_ref(trans, root, root->node->start,
806 root->node->len, owner_objectid,
807 generation, level, key_objectid);
810 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
811 struct extent_buffer *buf)
815 struct btrfs_key key;
816 struct btrfs_file_extent_item *fi;
825 level = btrfs_header_level(buf);
826 nritems = btrfs_header_nritems(buf);
827 for (i = 0; i < nritems; i++) {
830 btrfs_item_key_to_cpu(buf, &key, i);
831 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
833 fi = btrfs_item_ptr(buf, i,
834 struct btrfs_file_extent_item);
835 if (btrfs_file_extent_type(buf, fi) ==
836 BTRFS_FILE_EXTENT_INLINE)
838 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
839 if (disk_bytenr == 0)
841 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
842 btrfs_file_extent_disk_num_bytes(buf, fi),
843 root->root_key.objectid, trans->transid,
844 key.objectid, key.offset);
850 bytenr = btrfs_node_blockptr(buf, i);
851 btrfs_node_key_to_cpu(buf, &key, i);
852 ret = btrfs_inc_extent_ref(trans, root, bytenr,
853 btrfs_level_size(root, level - 1),
854 root->root_key.objectid,
856 level - 1, key.objectid);
867 for (i =0; i < faili; i++) {
870 btrfs_item_key_to_cpu(buf, &key, i);
871 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
873 fi = btrfs_item_ptr(buf, i,
874 struct btrfs_file_extent_item);
875 if (btrfs_file_extent_type(buf, fi) ==
876 BTRFS_FILE_EXTENT_INLINE)
878 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
879 if (disk_bytenr == 0)
881 err = btrfs_free_extent(trans, root, disk_bytenr,
882 btrfs_file_extent_disk_num_bytes(buf,
886 bytenr = btrfs_node_blockptr(buf, i);
887 err = btrfs_free_extent(trans, root, bytenr,
888 btrfs_level_size(root, level - 1), 0);
896 static int write_one_cache_group(struct btrfs_trans_handle *trans,
897 struct btrfs_root *root,
898 struct btrfs_path *path,
899 struct btrfs_block_group_cache *cache)
903 struct btrfs_root *extent_root = root->fs_info->extent_root;
905 struct extent_buffer *leaf;
907 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
912 leaf = path->nodes[0];
913 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
914 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
915 btrfs_mark_buffer_dirty(leaf);
916 btrfs_release_path(extent_root, path);
918 finish_current_insert(trans, extent_root);
919 pending_ret = del_pending_extents(trans, extent_root);
928 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
929 struct btrfs_root *root)
931 struct extent_io_tree *block_group_cache;
932 struct btrfs_block_group_cache *cache;
936 struct btrfs_path *path;
942 block_group_cache = &root->fs_info->block_group_cache;
943 path = btrfs_alloc_path();
948 ret = find_first_extent_bit(block_group_cache, last,
949 &start, &end, BLOCK_GROUP_DIRTY);
954 ret = get_state_private(block_group_cache, start, &ptr);
958 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
959 err = write_one_cache_group(trans, root,
962 * if we fail to write the cache group, we want
963 * to keep it marked dirty in hopes that a later
970 clear_extent_bits(block_group_cache, start, end,
971 BLOCK_GROUP_DIRTY, GFP_NOFS);
973 btrfs_free_path(path);
977 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
980 struct list_head *head = &info->space_info;
981 struct list_head *cur;
982 struct btrfs_space_info *found;
983 list_for_each(cur, head) {
984 found = list_entry(cur, struct btrfs_space_info, list);
985 if (found->flags == flags)
992 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
993 u64 total_bytes, u64 bytes_used,
994 struct btrfs_space_info **space_info)
996 struct btrfs_space_info *found;
998 found = __find_space_info(info, flags);
1000 found->total_bytes += total_bytes;
1001 found->bytes_used += bytes_used;
1002 WARN_ON(found->total_bytes < found->bytes_used);
1003 *space_info = found;
1006 found = kmalloc(sizeof(*found), GFP_NOFS);
1010 list_add(&found->list, &info->space_info);
1011 found->flags = flags;
1012 found->total_bytes = total_bytes;
1013 found->bytes_used = bytes_used;
1014 found->bytes_pinned = 0;
1016 *space_info = found;
1021 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1022 struct btrfs_root *extent_root, u64 alloc_bytes,
1025 struct btrfs_space_info *space_info;
1031 space_info = __find_space_info(extent_root->fs_info, flags);
1033 ret = update_space_info(extent_root->fs_info, flags,
1037 BUG_ON(!space_info);
1039 if (space_info->full)
1042 thresh = div_factor(space_info->total_bytes, 7);
1043 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1047 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1048 if (ret == -ENOSPC) {
1049 printk("space info full %Lu\n", flags);
1050 space_info->full = 1;
1056 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1057 extent_root->fs_info->chunk_root->root_key.objectid,
1063 static int update_block_group(struct btrfs_trans_handle *trans,
1064 struct btrfs_root *root,
1065 u64 bytenr, u64 num_bytes, int alloc,
1068 struct btrfs_block_group_cache *cache;
1069 struct btrfs_fs_info *info = root->fs_info;
1070 u64 total = num_bytes;
1077 cache = btrfs_lookup_block_group(info, bytenr);
1081 byte_in_group = bytenr - cache->key.objectid;
1082 WARN_ON(byte_in_group > cache->key.offset);
1083 start = cache->key.objectid;
1084 end = start + cache->key.offset - 1;
1085 set_extent_bits(&info->block_group_cache, start, end,
1086 BLOCK_GROUP_DIRTY, GFP_NOFS);
1088 old_val = btrfs_block_group_used(&cache->item);
1089 num_bytes = min(total, cache->key.offset - byte_in_group);
1091 old_val += num_bytes;
1092 cache->space_info->bytes_used += num_bytes;
1094 old_val -= num_bytes;
1095 cache->space_info->bytes_used -= num_bytes;
1097 set_extent_dirty(&info->free_space_cache,
1098 bytenr, bytenr + num_bytes - 1,
1102 btrfs_set_block_group_used(&cache->item, old_val);
1104 bytenr += num_bytes;
1109 static int update_pinned_extents(struct btrfs_root *root,
1110 u64 bytenr, u64 num, int pin)
1113 struct btrfs_block_group_cache *cache;
1114 struct btrfs_fs_info *fs_info = root->fs_info;
1117 set_extent_dirty(&fs_info->pinned_extents,
1118 bytenr, bytenr + num - 1, GFP_NOFS);
1120 clear_extent_dirty(&fs_info->pinned_extents,
1121 bytenr, bytenr + num - 1, GFP_NOFS);
1124 cache = btrfs_lookup_block_group(fs_info, bytenr);
1126 len = min(num, cache->key.offset -
1127 (bytenr - cache->key.objectid));
1129 cache->pinned += len;
1130 cache->space_info->bytes_pinned += len;
1131 fs_info->total_pinned += len;
1133 cache->pinned -= len;
1134 cache->space_info->bytes_pinned -= len;
1135 fs_info->total_pinned -= len;
1143 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1148 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1152 ret = find_first_extent_bit(pinned_extents, last,
1153 &start, &end, EXTENT_DIRTY);
1156 set_extent_dirty(copy, start, end, GFP_NOFS);
1162 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1163 struct btrfs_root *root,
1164 struct extent_io_tree *unpin)
1169 struct extent_io_tree *free_space_cache;
1170 free_space_cache = &root->fs_info->free_space_cache;
1173 ret = find_first_extent_bit(unpin, 0, &start, &end,
1177 update_pinned_extents(root, start, end + 1 - start, 0);
1178 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1179 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1184 static int finish_current_insert(struct btrfs_trans_handle *trans,
1185 struct btrfs_root *extent_root)
1189 struct btrfs_fs_info *info = extent_root->fs_info;
1190 struct extent_buffer *eb;
1191 struct btrfs_path *path;
1192 struct btrfs_key ins;
1193 struct btrfs_disk_key first;
1194 struct btrfs_extent_item extent_item;
1199 btrfs_set_stack_extent_refs(&extent_item, 1);
1200 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1201 path = btrfs_alloc_path();
1204 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1205 &end, EXTENT_LOCKED);
1209 ins.objectid = start;
1210 ins.offset = end + 1 - start;
1211 err = btrfs_insert_item(trans, extent_root, &ins,
1212 &extent_item, sizeof(extent_item));
1213 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1215 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1216 level = btrfs_header_level(eb);
1218 btrfs_item_key(eb, &first, 0);
1220 btrfs_node_key(eb, &first, 0);
1222 err = btrfs_insert_extent_backref(trans, extent_root, path,
1223 start, extent_root->root_key.objectid,
1225 btrfs_disk_key_objectid(&first));
1227 free_extent_buffer(eb);
1229 btrfs_free_path(path);
1233 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1237 struct extent_buffer *buf;
1240 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1242 if (btrfs_buffer_uptodate(buf)) {
1244 root->fs_info->running_transaction->transid;
1245 if (btrfs_header_generation(buf) == transid) {
1246 free_extent_buffer(buf);
1250 free_extent_buffer(buf);
1252 update_pinned_extents(root, bytenr, num_bytes, 1);
1254 set_extent_bits(&root->fs_info->pending_del,
1255 bytenr, bytenr + num_bytes - 1,
1256 EXTENT_LOCKED, GFP_NOFS);
1263 * remove an extent from the root, returns 0 on success
1265 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1266 *root, u64 bytenr, u64 num_bytes,
1267 u64 root_objectid, u64 ref_generation,
1268 u64 owner_objectid, u64 owner_offset, int pin,
1271 struct btrfs_path *path;
1272 struct btrfs_key key;
1273 struct btrfs_fs_info *info = root->fs_info;
1274 struct btrfs_extent_ops *ops = info->extent_ops;
1275 struct btrfs_root *extent_root = info->extent_root;
1276 struct extent_buffer *leaf;
1278 int extent_slot = 0;
1279 int found_extent = 0;
1281 struct btrfs_extent_item *ei;
1284 key.objectid = bytenr;
1285 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1286 key.offset = num_bytes;
1288 path = btrfs_alloc_path();
1292 ret = lookup_extent_backref(trans, extent_root, path,
1293 bytenr, root_objectid,
1295 owner_objectid, owner_offset, 1);
1297 struct btrfs_key found_key;
1298 extent_slot = path->slots[0];
1299 while(extent_slot > 0) {
1301 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1303 if (found_key.objectid != bytenr)
1305 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1306 found_key.offset == num_bytes) {
1310 if (path->slots[0] - extent_slot > 5)
1314 ret = btrfs_del_item(trans, extent_root, path);
1316 btrfs_print_leaf(extent_root, path->nodes[0]);
1318 printk("Unable to find ref byte nr %Lu root %Lu "
1319 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1320 root_objectid, ref_generation, owner_objectid,
1323 if (!found_extent) {
1324 btrfs_release_path(extent_root, path);
1325 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1329 extent_slot = path->slots[0];
1332 leaf = path->nodes[0];
1333 ei = btrfs_item_ptr(leaf, extent_slot,
1334 struct btrfs_extent_item);
1335 refs = btrfs_extent_refs(leaf, ei);
1338 btrfs_set_extent_refs(leaf, ei, refs);
1340 btrfs_mark_buffer_dirty(leaf);
1342 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1343 /* if the back ref and the extent are next to each other
1344 * they get deleted below in one shot
1346 path->slots[0] = extent_slot;
1348 } else if (found_extent) {
1349 /* otherwise delete the extent back ref */
1350 ret = btrfs_del_item(trans, extent_root, path);
1352 /* if refs are 0, we need to setup the path for deletion */
1354 btrfs_release_path(extent_root, path);
1355 ret = btrfs_search_slot(trans, extent_root, &key, path,
1368 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1374 /* block accounting for super block */
1375 super_used = btrfs_super_bytes_used(&info->super_copy);
1376 btrfs_set_super_bytes_used(&info->super_copy,
1377 super_used - num_bytes);
1379 /* block accounting for root item */
1380 root_used = btrfs_root_used(&root->root_item);
1381 btrfs_set_root_used(&root->root_item,
1382 root_used - num_bytes);
1383 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1388 if (ops && ops->free_extent)
1389 ops->free_extent(root, bytenr, num_bytes);
1391 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1395 btrfs_free_path(path);
1396 finish_current_insert(trans, extent_root);
1401 * find all the blocks marked as pending in the radix tree and remove
1402 * them from the extent map
1404 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1405 btrfs_root *extent_root)
1411 struct extent_io_tree *pending_del;
1412 struct extent_io_tree *pinned_extents;
1414 pending_del = &extent_root->fs_info->pending_del;
1415 pinned_extents = &extent_root->fs_info->pinned_extents;
1418 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1422 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1423 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1425 ret = __free_extent(trans, extent_root,
1426 start, end + 1 - start,
1427 extent_root->root_key.objectid,
1436 * remove an extent from the root, returns 0 on success
1438 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1439 *root, u64 bytenr, u64 num_bytes,
1440 u64 root_objectid, u64 ref_generation,
1441 u64 owner_objectid, u64 owner_offset, int pin)
1443 struct btrfs_root *extent_root = root->fs_info->extent_root;
1447 WARN_ON(num_bytes < root->sectorsize);
1448 if (!root->ref_cows)
1451 if (root == extent_root) {
1452 pin_down_bytes(root, bytenr, num_bytes, 1);
1455 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1456 ref_generation, owner_objectid, owner_offset,
1458 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1459 return ret ? ret : pending_ret;
1462 static u64 stripe_align(struct btrfs_root *root, u64 val)
1464 u64 mask = ((u64)root->stripesize - 1);
1465 u64 ret = (val + mask) & ~mask;
1470 * walks the btree of allocated extents and find a hole of a given size.
1471 * The key ins is changed to record the hole:
1472 * ins->objectid == block start
1473 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1474 * ins->offset == number of blocks
1475 * Any available blocks before search_start are skipped.
1477 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1478 struct btrfs_root *orig_root,
1479 u64 num_bytes, u64 empty_size,
1480 u64 search_start, u64 search_end,
1481 u64 hint_byte, struct btrfs_key *ins,
1482 u64 exclude_start, u64 exclude_nr,
1486 u64 orig_search_start = search_start;
1487 struct btrfs_root * root = orig_root->fs_info->extent_root;
1488 struct btrfs_fs_info *info = root->fs_info;
1489 u64 total_needed = num_bytes;
1490 struct btrfs_block_group_cache *block_group;
1494 WARN_ON(num_bytes < root->sectorsize);
1495 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1497 if (search_end == (u64)-1)
1498 search_end = btrfs_super_total_bytes(&info->super_copy);
1501 block_group = btrfs_lookup_block_group(info, hint_byte);
1503 hint_byte = search_start;
1504 block_group = btrfs_find_block_group(root, block_group,
1505 hint_byte, data, 1);
1507 block_group = btrfs_find_block_group(root,
1509 search_start, data, 1);
1512 total_needed += empty_size;
1516 block_group = btrfs_lookup_block_group(info, search_start);
1518 block_group = btrfs_lookup_block_group(info,
1521 ret = find_search_start(root, &block_group, &search_start,
1522 total_needed, data);
1526 search_start = stripe_align(root, search_start);
1527 ins->objectid = search_start;
1528 ins->offset = num_bytes;
1530 if (ins->objectid + num_bytes >= search_end)
1533 if (ins->objectid + num_bytes >
1534 block_group->key.objectid + block_group->key.offset) {
1535 search_start = block_group->key.objectid +
1536 block_group->key.offset;
1540 if (test_range_bit(&info->extent_ins, ins->objectid,
1541 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1542 search_start = ins->objectid + num_bytes;
1546 if (test_range_bit(&info->pinned_extents, ins->objectid,
1547 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1548 search_start = ins->objectid + num_bytes;
1552 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1553 ins->objectid < exclude_start + exclude_nr)) {
1554 search_start = exclude_start + exclude_nr;
1558 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1559 block_group = btrfs_lookup_block_group(info, ins->objectid);
1561 trans->block_group = block_group;
1563 ins->offset = num_bytes;
1567 if (search_start + num_bytes >= search_end) {
1569 search_start = orig_search_start;
1576 total_needed -= empty_size;
1581 block_group = btrfs_lookup_block_group(info, search_start);
1583 block_group = btrfs_find_block_group(root, block_group,
1584 search_start, data, 0);
1591 * finds a free extent and does all the dirty work required for allocation
1592 * returns the key for the extent through ins, and a tree buffer for
1593 * the first block of the extent through buf.
1595 * returns 0 if everything worked, non-zero otherwise.
1597 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1600 u64 owner, u64 owner_offset,
1601 u64 empty_size, u64 hint_byte,
1602 u64 search_end, struct btrfs_key *ins, int data)
1606 u64 super_used, root_used;
1607 u64 search_start = 0;
1608 struct btrfs_fs_info *info = root->fs_info;
1609 struct btrfs_extent_ops *ops = info->extent_ops;
1611 struct btrfs_root *extent_root = info->extent_root;
1612 struct btrfs_path *path;
1613 struct btrfs_extent_item *extent_item;
1614 struct btrfs_extent_ref *ref;
1615 struct btrfs_key keys[2];
1616 int extra_alloc_flags = 0;
1618 if (btrfs_super_num_devices(&info->super_copy) > 1)
1619 extra_alloc_flags = BTRFS_BLOCK_GROUP_RAID0;
1622 data = BTRFS_BLOCK_GROUP_DATA | extra_alloc_flags;
1623 } else if (root == root->fs_info->chunk_root ||
1624 info->force_system_allocs) {
1625 data = BTRFS_BLOCK_GROUP_SYSTEM;
1627 data = BTRFS_BLOCK_GROUP_METADATA | extra_alloc_flags;
1630 if (root->ref_cows) {
1631 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1632 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1634 BTRFS_BLOCK_GROUP_METADATA |
1638 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1639 num_bytes + 2 * 1024 * 1024, data);
1643 WARN_ON(num_bytes < root->sectorsize);
1644 if (ops && ops->alloc_extent) {
1645 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
1647 ret = find_free_extent(trans, root, num_bytes, empty_size,
1648 search_start, search_end, hint_byte,
1649 ins, trans->alloc_exclude_start,
1650 trans->alloc_exclude_nr, data);
1656 /* block accounting for super block */
1657 super_used = btrfs_super_bytes_used(&info->super_copy);
1658 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1660 /* block accounting for root item */
1661 root_used = btrfs_root_used(&root->root_item);
1662 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1664 clear_extent_dirty(&root->fs_info->free_space_cache,
1665 ins->objectid, ins->objectid + ins->offset - 1,
1668 if (root == extent_root) {
1669 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1670 ins->objectid + ins->offset - 1,
1671 EXTENT_LOCKED, GFP_NOFS);
1675 WARN_ON(trans->alloc_exclude_nr);
1676 trans->alloc_exclude_start = ins->objectid;
1677 trans->alloc_exclude_nr = ins->offset;
1679 memcpy(&keys[0], ins, sizeof(*ins));
1680 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1681 owner, owner_offset);
1682 keys[1].objectid = ins->objectid;
1683 keys[1].type = BTRFS_EXTENT_REF_KEY;
1684 sizes[0] = sizeof(*extent_item);
1685 sizes[1] = sizeof(*ref);
1687 path = btrfs_alloc_path();
1690 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1694 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1695 struct btrfs_extent_item);
1696 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1697 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1698 struct btrfs_extent_ref);
1700 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1701 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1702 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1703 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1705 btrfs_mark_buffer_dirty(path->nodes[0]);
1707 trans->alloc_exclude_start = 0;
1708 trans->alloc_exclude_nr = 0;
1709 btrfs_free_path(path);
1710 finish_current_insert(trans, extent_root);
1711 pending_ret = del_pending_extents(trans, extent_root);
1721 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1723 printk("update block group failed for %Lu %Lu\n",
1724 ins->objectid, ins->offset);
1731 * helper function to allocate a block for a given tree
1732 * returns the tree buffer or NULL.
1734 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1735 struct btrfs_root *root,
1737 u64 root_objectid, u64 hint,
1743 ref_generation = trans->transid;
1748 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1749 ref_generation, 0, 0, hint, empty_size);
1753 * helper function to allocate a block for a given tree
1754 * returns the tree buffer or NULL.
1756 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1766 struct btrfs_key ins;
1768 struct extent_buffer *buf;
1770 ret = btrfs_alloc_extent(trans, root, blocksize,
1771 root_objectid, ref_generation,
1772 level, first_objectid, empty_size, hint,
1776 return ERR_PTR(ret);
1778 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1780 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1781 root->root_key.objectid, ref_generation,
1784 return ERR_PTR(-ENOMEM);
1786 btrfs_set_buffer_uptodate(buf);
1787 trans->blocks_used++;
1791 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct extent_buffer *leaf)
1796 u64 leaf_generation;
1797 struct btrfs_key key;
1798 struct btrfs_file_extent_item *fi;
1803 BUG_ON(!btrfs_is_leaf(leaf));
1804 nritems = btrfs_header_nritems(leaf);
1805 leaf_owner = btrfs_header_owner(leaf);
1806 leaf_generation = btrfs_header_generation(leaf);
1808 for (i = 0; i < nritems; i++) {
1811 btrfs_item_key_to_cpu(leaf, &key, i);
1812 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1814 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1815 if (btrfs_file_extent_type(leaf, fi) ==
1816 BTRFS_FILE_EXTENT_INLINE)
1819 * FIXME make sure to insert a trans record that
1820 * repeats the snapshot del on crash
1822 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1823 if (disk_bytenr == 0)
1825 ret = btrfs_free_extent(trans, root, disk_bytenr,
1826 btrfs_file_extent_disk_num_bytes(leaf, fi),
1827 leaf_owner, leaf_generation,
1828 key.objectid, key.offset, 0);
1834 static void noinline reada_walk_down(struct btrfs_root *root,
1835 struct extent_buffer *node,
1848 nritems = btrfs_header_nritems(node);
1849 level = btrfs_header_level(node);
1853 for (i = slot; i < nritems && skipped < 32; i++) {
1854 bytenr = btrfs_node_blockptr(node, i);
1855 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1856 (last > bytenr && last - bytenr > 32 * 1024))) {
1860 blocksize = btrfs_level_size(root, level - 1);
1862 ret = lookup_extent_ref(NULL, root, bytenr,
1870 mutex_unlock(&root->fs_info->fs_mutex);
1871 ret = readahead_tree_block(root, bytenr, blocksize);
1872 last = bytenr + blocksize;
1874 mutex_lock(&root->fs_info->fs_mutex);
1881 * helper function for drop_snapshot, this walks down the tree dropping ref
1882 * counts as it goes.
1884 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 struct btrfs_path *path, int *level)
1891 struct extent_buffer *next;
1892 struct extent_buffer *cur;
1893 struct extent_buffer *parent;
1898 WARN_ON(*level < 0);
1899 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1900 ret = lookup_extent_ref(trans, root,
1901 path->nodes[*level]->start,
1902 path->nodes[*level]->len, &refs);
1908 * walk down to the last node level and free all the leaves
1910 while(*level >= 0) {
1911 WARN_ON(*level < 0);
1912 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1913 cur = path->nodes[*level];
1915 if (btrfs_header_level(cur) != *level)
1918 if (path->slots[*level] >=
1919 btrfs_header_nritems(cur))
1922 ret = drop_leaf_ref(trans, root, cur);
1926 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1927 blocksize = btrfs_level_size(root, *level - 1);
1928 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1931 parent = path->nodes[*level];
1932 root_owner = btrfs_header_owner(parent);
1933 root_gen = btrfs_header_generation(parent);
1934 path->slots[*level]++;
1935 ret = btrfs_free_extent(trans, root, bytenr,
1936 blocksize, root_owner,
1941 next = btrfs_find_tree_block(root, bytenr, blocksize);
1942 if (!next || !btrfs_buffer_uptodate(next)) {
1943 free_extent_buffer(next);
1944 reada_walk_down(root, cur, path->slots[*level]);
1945 mutex_unlock(&root->fs_info->fs_mutex);
1946 next = read_tree_block(root, bytenr, blocksize);
1947 mutex_lock(&root->fs_info->fs_mutex);
1949 /* we dropped the lock, check one more time */
1950 ret = lookup_extent_ref(trans, root, bytenr,
1954 parent = path->nodes[*level];
1955 root_owner = btrfs_header_owner(parent);
1956 root_gen = btrfs_header_generation(parent);
1958 path->slots[*level]++;
1959 free_extent_buffer(next);
1960 ret = btrfs_free_extent(trans, root, bytenr,
1968 WARN_ON(*level <= 0);
1969 if (path->nodes[*level-1])
1970 free_extent_buffer(path->nodes[*level-1]);
1971 path->nodes[*level-1] = next;
1972 *level = btrfs_header_level(next);
1973 path->slots[*level] = 0;
1976 WARN_ON(*level < 0);
1977 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1979 if (path->nodes[*level] == root->node) {
1980 root_owner = root->root_key.objectid;
1981 parent = path->nodes[*level];
1983 parent = path->nodes[*level + 1];
1984 root_owner = btrfs_header_owner(parent);
1987 root_gen = btrfs_header_generation(parent);
1988 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1989 path->nodes[*level]->len,
1990 root_owner, root_gen, 0, 0, 1);
1991 free_extent_buffer(path->nodes[*level]);
1992 path->nodes[*level] = NULL;
1999 * helper for dropping snapshots. This walks back up the tree in the path
2000 * to find the first node higher up where we haven't yet gone through
2003 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2004 struct btrfs_root *root,
2005 struct btrfs_path *path, int *level)
2009 struct btrfs_root_item *root_item = &root->root_item;
2014 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2015 slot = path->slots[i];
2016 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2017 struct extent_buffer *node;
2018 struct btrfs_disk_key disk_key;
2019 node = path->nodes[i];
2022 WARN_ON(*level == 0);
2023 btrfs_node_key(node, &disk_key, path->slots[i]);
2024 memcpy(&root_item->drop_progress,
2025 &disk_key, sizeof(disk_key));
2026 root_item->drop_level = i;
2029 if (path->nodes[*level] == root->node) {
2030 root_owner = root->root_key.objectid;
2032 btrfs_header_generation(path->nodes[*level]);
2034 struct extent_buffer *node;
2035 node = path->nodes[*level + 1];
2036 root_owner = btrfs_header_owner(node);
2037 root_gen = btrfs_header_generation(node);
2039 ret = btrfs_free_extent(trans, root,
2040 path->nodes[*level]->start,
2041 path->nodes[*level]->len,
2042 root_owner, root_gen, 0, 0, 1);
2044 free_extent_buffer(path->nodes[*level]);
2045 path->nodes[*level] = NULL;
2053 * drop the reference count on the tree rooted at 'snap'. This traverses
2054 * the tree freeing any blocks that have a ref count of zero after being
2057 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2063 struct btrfs_path *path;
2066 struct btrfs_root_item *root_item = &root->root_item;
2068 path = btrfs_alloc_path();
2071 level = btrfs_header_level(root->node);
2073 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2074 path->nodes[level] = root->node;
2075 extent_buffer_get(root->node);
2076 path->slots[level] = 0;
2078 struct btrfs_key key;
2079 struct btrfs_disk_key found_key;
2080 struct extent_buffer *node;
2082 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2083 level = root_item->drop_level;
2084 path->lowest_level = level;
2085 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2090 node = path->nodes[level];
2091 btrfs_node_key(node, &found_key, path->slots[level]);
2092 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2093 sizeof(found_key)));
2096 wret = walk_down_tree(trans, root, path, &level);
2102 wret = walk_up_tree(trans, root, path, &level);
2112 for (i = 0; i <= orig_level; i++) {
2113 if (path->nodes[i]) {
2114 free_extent_buffer(path->nodes[i]);
2115 path->nodes[i] = NULL;
2119 btrfs_free_path(path);
2123 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2130 ret = find_first_extent_bit(&info->block_group_cache, 0,
2131 &start, &end, (unsigned int)-1);
2134 ret = get_state_private(&info->block_group_cache, start, &ptr);
2136 kfree((void *)(unsigned long)ptr);
2137 clear_extent_bits(&info->block_group_cache, start,
2138 end, (unsigned int)-1, GFP_NOFS);
2141 ret = find_first_extent_bit(&info->free_space_cache, 0,
2142 &start, &end, EXTENT_DIRTY);
2145 clear_extent_dirty(&info->free_space_cache, start,
2151 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2152 struct btrfs_key *key)
2155 struct btrfs_key found_key;
2156 struct extent_buffer *leaf;
2159 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2163 slot = path->slots[0];
2164 leaf = path->nodes[0];
2165 if (slot >= btrfs_header_nritems(leaf)) {
2166 ret = btrfs_next_leaf(root, path);
2173 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2175 if (found_key.objectid >= key->objectid &&
2176 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2185 int btrfs_read_block_groups(struct btrfs_root *root)
2187 struct btrfs_path *path;
2190 struct btrfs_block_group_cache *cache;
2191 struct btrfs_fs_info *info = root->fs_info;
2192 struct btrfs_space_info *space_info;
2193 struct extent_io_tree *block_group_cache;
2194 struct btrfs_key key;
2195 struct btrfs_key found_key;
2196 struct extent_buffer *leaf;
2198 block_group_cache = &info->block_group_cache;
2200 root = info->extent_root;
2203 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2204 path = btrfs_alloc_path();
2209 ret = find_first_block_group(root, path, &key);
2217 leaf = path->nodes[0];
2218 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2219 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2225 read_extent_buffer(leaf, &cache->item,
2226 btrfs_item_ptr_offset(leaf, path->slots[0]),
2227 sizeof(cache->item));
2228 memcpy(&cache->key, &found_key, sizeof(found_key));
2231 key.objectid = found_key.objectid + found_key.offset;
2232 btrfs_release_path(root, path);
2233 cache->flags = btrfs_block_group_flags(&cache->item);
2235 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2236 bit = BLOCK_GROUP_DATA;
2237 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2238 bit = BLOCK_GROUP_SYSTEM;
2239 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2240 bit = BLOCK_GROUP_METADATA;
2243 ret = update_space_info(info, cache->flags, found_key.offset,
2244 btrfs_block_group_used(&cache->item),
2247 cache->space_info = space_info;
2249 /* use EXTENT_LOCKED to prevent merging */
2250 set_extent_bits(block_group_cache, found_key.objectid,
2251 found_key.objectid + found_key.offset - 1,
2252 bit | EXTENT_LOCKED, GFP_NOFS);
2253 set_state_private(block_group_cache, found_key.objectid,
2254 (unsigned long)cache);
2257 btrfs_super_total_bytes(&info->super_copy))
2262 btrfs_free_path(path);
2266 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2267 struct btrfs_root *root, u64 bytes_used,
2268 u64 type, u64 chunk_tree, u64 chunk_objectid,
2273 struct btrfs_root *extent_root;
2274 struct btrfs_block_group_cache *cache;
2275 struct extent_io_tree *block_group_cache;
2277 extent_root = root->fs_info->extent_root;
2278 block_group_cache = &root->fs_info->block_group_cache;
2280 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2282 cache->key.objectid = chunk_objectid;
2283 cache->key.offset = size;
2285 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2286 memset(&cache->item, 0, sizeof(cache->item));
2287 btrfs_set_block_group_used(&cache->item, bytes_used);
2288 btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
2289 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2290 cache->flags = type;
2291 btrfs_set_block_group_flags(&cache->item, type);
2293 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2294 &cache->space_info);
2297 if (type & BTRFS_BLOCK_GROUP_DATA) {
2298 bit = BLOCK_GROUP_DATA;
2299 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2300 bit = BLOCK_GROUP_SYSTEM;
2301 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2302 bit = BLOCK_GROUP_METADATA;
2304 set_extent_bits(block_group_cache, chunk_objectid,
2305 chunk_objectid + size - 1,
2306 bit | EXTENT_LOCKED, GFP_NOFS);
2307 set_state_private(block_group_cache, chunk_objectid,
2308 (unsigned long)cache);
2310 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2311 sizeof(cache->item));
2314 finish_current_insert(trans, extent_root);
2315 ret = del_pending_extents(trans, extent_root);
2320 u64 btrfs_hash_extent_ref(u64 root_objectid, u64 ref_generation,
2321 u64 owner, u64 owner_offset)
2323 return hash_extent_ref(root_objectid, ref_generation,
2324 owner, owner_offset);
2327 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
2328 struct btrfs_root *root,
2329 u64 bytenr, u64 num_bytes, int alloc,
2332 return update_block_group(trans, root, bytenr, num_bytes,