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) == 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 %llu\n",
229 (unsigned long long)search_start);
235 last = cache->key.objectid + cache->key.offset;
237 cache = btrfs_lookup_block_group(root->fs_info, last);
247 if (cache_miss && !cache->cached) {
248 cache_block_group(root, cache);
250 cache = btrfs_lookup_block_group(root->fs_info, last);
252 cache = btrfs_find_block_group(root, cache, last, data, 0);
260 static u64 div_factor(u64 num, int factor)
269 static int block_group_state_bits(u64 flags)
272 if (flags & BTRFS_BLOCK_GROUP_DATA)
273 bits |= BLOCK_GROUP_DATA;
274 if (flags & BTRFS_BLOCK_GROUP_METADATA)
275 bits |= BLOCK_GROUP_METADATA;
276 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
277 bits |= BLOCK_GROUP_SYSTEM;
281 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
282 struct btrfs_block_group_cache
283 *hint, u64 search_start,
286 struct btrfs_block_group_cache *cache;
287 struct extent_io_tree *block_group_cache;
288 struct btrfs_block_group_cache *found_group = NULL;
289 struct btrfs_fs_info *info = root->fs_info;
302 block_group_cache = &info->block_group_cache;
307 bit = block_group_state_bits(data);
310 struct btrfs_block_group_cache *shint;
311 shint = btrfs_lookup_block_group(info, search_start);
312 if (shint && block_group_bits(shint, data)) {
313 used = btrfs_block_group_used(&shint->item);
314 if (used + shint->pinned <
315 div_factor(shint->key.offset, factor)) {
320 if (hint && block_group_bits(hint, data)) {
321 used = btrfs_block_group_used(&hint->item);
322 if (used + hint->pinned <
323 div_factor(hint->key.offset, factor)) {
326 last = hint->key.objectid + hint->key.offset;
330 hint_last = max(hint->key.objectid, search_start);
332 hint_last = search_start;
338 ret = find_first_extent_bit(block_group_cache, last,
343 ret = get_state_private(block_group_cache, start, &ptr);
347 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
348 last = cache->key.objectid + cache->key.offset;
349 used = btrfs_block_group_used(&cache->item);
351 if (block_group_bits(cache, data)) {
353 free_check = cache->key.offset;
355 free_check = div_factor(cache->key.offset,
358 if (used + cache->pinned < free_check) {
374 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
375 u64 owner, u64 owner_offset)
377 u32 high_crc = ~(u32)0;
378 u32 low_crc = ~(u32)0;
381 lenum = cpu_to_le64(root_objectid);
382 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
383 lenum = cpu_to_le64(ref_generation);
384 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
385 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
386 lenum = cpu_to_le64(owner);
387 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
388 lenum = cpu_to_le64(owner_offset);
389 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
391 return ((u64)high_crc << 32) | (u64)low_crc;
394 static int match_extent_ref(struct extent_buffer *leaf,
395 struct btrfs_extent_ref *disk_ref,
396 struct btrfs_extent_ref *cpu_ref)
401 if (cpu_ref->objectid)
402 len = sizeof(*cpu_ref);
404 len = 2 * sizeof(u64);
405 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
410 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
411 struct btrfs_root *root,
412 struct btrfs_path *path, u64 bytenr,
414 u64 ref_generation, u64 owner,
415 u64 owner_offset, int del)
418 struct btrfs_key key;
419 struct btrfs_key found_key;
420 struct btrfs_extent_ref ref;
421 struct extent_buffer *leaf;
422 struct btrfs_extent_ref *disk_ref;
426 btrfs_set_stack_ref_root(&ref, root_objectid);
427 btrfs_set_stack_ref_generation(&ref, ref_generation);
428 btrfs_set_stack_ref_objectid(&ref, owner);
429 btrfs_set_stack_ref_offset(&ref, owner_offset);
431 hash = hash_extent_ref(root_objectid, ref_generation, owner,
434 key.objectid = bytenr;
435 key.type = BTRFS_EXTENT_REF_KEY;
438 ret = btrfs_search_slot(trans, root, &key, path,
442 leaf = path->nodes[0];
444 u32 nritems = btrfs_header_nritems(leaf);
445 if (path->slots[0] >= nritems) {
446 ret2 = btrfs_next_leaf(root, path);
449 leaf = path->nodes[0];
451 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
452 if (found_key.objectid != bytenr ||
453 found_key.type != BTRFS_EXTENT_REF_KEY)
455 key.offset = found_key.offset;
457 btrfs_release_path(root, path);
461 disk_ref = btrfs_item_ptr(path->nodes[0],
463 struct btrfs_extent_ref);
464 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
468 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
469 key.offset = found_key.offset + 1;
470 btrfs_release_path(root, path);
477 * Back reference rules. Back refs have three main goals:
479 * 1) differentiate between all holders of references to an extent so that
480 * when a reference is dropped we can make sure it was a valid reference
481 * before freeing the extent.
483 * 2) Provide enough information to quickly find the holders of an extent
484 * if we notice a given block is corrupted or bad.
486 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
487 * maintenance. This is actually the same as #2, but with a slightly
488 * different use case.
490 * File extents can be referenced by:
492 * - multiple snapshots, subvolumes, or different generations in one subvol
493 * - different files inside a single subvolume (in theory, not implemented yet)
494 * - different offsets inside a file (bookend extents in file.c)
496 * The extent ref structure has fields for:
498 * - Objectid of the subvolume root
499 * - Generation number of the tree holding the reference
500 * - objectid of the file holding the reference
501 * - offset in the file corresponding to the key holding the reference
503 * When a file extent is allocated the fields are filled in:
504 * (root_key.objectid, trans->transid, inode objectid, offset in file)
506 * When a leaf is cow'd new references are added for every file extent found
507 * in the leaf. It looks the same as the create case, but trans->transid
508 * will be different when the block is cow'd.
510 * (root_key.objectid, trans->transid, inode objectid, offset in file)
512 * When a file extent is removed either during snapshot deletion or file
513 * truncation, the corresponding back reference is found
516 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
517 * inode objectid, offset in file)
519 * Btree extents can be referenced by:
521 * - Different subvolumes
522 * - Different generations of the same subvolume
524 * Storing sufficient information for a full reverse mapping of a btree
525 * block would require storing the lowest key of the block in the backref,
526 * and it would require updating that lowest key either before write out or
527 * every time it changed. Instead, the objectid of the lowest key is stored
528 * along with the level of the tree block. This provides a hint
529 * about where in the btree the block can be found. Searches through the
530 * btree only need to look for a pointer to that block, so they stop one
531 * level higher than the level recorded in the backref.
533 * Some btrees do not do reference counting on their extents. These
534 * include the extent tree and the tree of tree roots. Backrefs for these
535 * trees always have a generation of zero.
537 * When a tree block is created, back references are inserted:
539 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
541 * When a tree block is cow'd in a reference counted root,
542 * new back references are added for all the blocks it points to.
543 * These are of the form (trans->transid will have increased since creation):
545 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
547 * Because the lowest_key_objectid and the level are just hints
548 * they are not used when backrefs are deleted. When a backref is deleted:
550 * if backref was for a tree root:
551 * root_objectid = root->root_key.objectid
553 * root_objectid = btrfs_header_owner(parent)
555 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
557 * Back Reference Key hashing:
559 * Back references have four fields, each 64 bits long. Unfortunately,
560 * This is hashed into a single 64 bit number and placed into the key offset.
561 * The key objectid corresponds to the first byte in the extent, and the
562 * key type is set to BTRFS_EXTENT_REF_KEY
564 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
565 struct btrfs_root *root,
566 struct btrfs_path *path, u64 bytenr,
567 u64 root_objectid, u64 ref_generation,
568 u64 owner, u64 owner_offset)
571 struct btrfs_key key;
572 struct btrfs_extent_ref ref;
573 struct btrfs_extent_ref *disk_ref;
576 btrfs_set_stack_ref_root(&ref, root_objectid);
577 btrfs_set_stack_ref_generation(&ref, ref_generation);
578 btrfs_set_stack_ref_objectid(&ref, owner);
579 btrfs_set_stack_ref_offset(&ref, owner_offset);
581 hash = hash_extent_ref(root_objectid, ref_generation, owner,
584 key.objectid = bytenr;
585 key.type = BTRFS_EXTENT_REF_KEY;
587 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
588 while (ret == -EEXIST) {
589 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
590 struct btrfs_extent_ref);
591 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
594 btrfs_release_path(root, path);
595 ret = btrfs_insert_empty_item(trans, root, path, &key,
600 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
601 struct btrfs_extent_ref);
602 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
604 btrfs_mark_buffer_dirty(path->nodes[0]);
606 btrfs_release_path(root, path);
610 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
611 struct btrfs_root *root,
612 u64 bytenr, u64 num_bytes,
613 u64 root_objectid, u64 ref_generation,
614 u64 owner, u64 owner_offset)
616 struct btrfs_path *path;
618 struct btrfs_key key;
619 struct extent_buffer *l;
620 struct btrfs_extent_item *item;
623 WARN_ON(num_bytes < root->sectorsize);
624 path = btrfs_alloc_path();
628 key.objectid = bytenr;
629 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
630 key.offset = num_bytes;
631 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
640 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
641 refs = btrfs_extent_refs(l, item);
642 btrfs_set_extent_refs(l, item, refs + 1);
643 btrfs_mark_buffer_dirty(path->nodes[0]);
645 btrfs_release_path(root->fs_info->extent_root, path);
647 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
648 path, bytenr, root_objectid,
649 ref_generation, owner, owner_offset);
651 finish_current_insert(trans, root->fs_info->extent_root);
652 del_pending_extents(trans, root->fs_info->extent_root);
654 btrfs_free_path(path);
658 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
659 struct btrfs_root *root)
661 finish_current_insert(trans, root->fs_info->extent_root);
662 del_pending_extents(trans, root->fs_info->extent_root);
666 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
667 struct btrfs_root *root, u64 bytenr,
668 u64 num_bytes, u32 *refs)
670 struct btrfs_path *path;
672 struct btrfs_key key;
673 struct extent_buffer *l;
674 struct btrfs_extent_item *item;
676 WARN_ON(num_bytes < root->sectorsize);
677 path = btrfs_alloc_path();
678 key.objectid = bytenr;
679 key.offset = num_bytes;
680 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
681 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
686 btrfs_print_leaf(root, path->nodes[0]);
687 printk("failed to find block number %llu\n",
688 (unsigned long long)bytenr);
692 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
693 *refs = btrfs_extent_refs(l, item);
695 btrfs_free_path(path);
699 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
700 struct btrfs_path *count_path,
703 struct btrfs_root *extent_root = root->fs_info->extent_root;
704 struct btrfs_path *path;
707 u64 root_objectid = root->root_key.objectid;
713 struct btrfs_key key;
714 struct btrfs_key found_key;
715 struct extent_buffer *l;
716 struct btrfs_extent_item *item;
717 struct btrfs_extent_ref *ref_item;
720 path = btrfs_alloc_path();
723 bytenr = first_extent;
725 bytenr = count_path->nodes[level]->start;
728 key.objectid = bytenr;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
738 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
740 if (found_key.objectid != bytenr ||
741 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
745 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
746 refs = btrfs_extent_refs(l, item);
748 nritems = btrfs_header_nritems(l);
749 if (path->slots[0] >= nritems) {
750 ret = btrfs_next_leaf(extent_root, path);
755 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
756 if (found_key.objectid != bytenr)
758 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
764 ref_item = btrfs_item_ptr(l, path->slots[0],
765 struct btrfs_extent_ref);
766 found_objectid = btrfs_ref_root(l, ref_item);
768 if (found_objectid != root_objectid) {
775 if (cur_count == 0) {
779 if (level >= 0 && root->node == count_path->nodes[level])
782 btrfs_release_path(root, path);
786 btrfs_free_path(path);
789 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
790 struct btrfs_root *root, u64 owner_objectid)
796 struct btrfs_disk_key disk_key;
798 level = btrfs_header_level(root->node);
799 generation = trans->transid;
800 nritems = btrfs_header_nritems(root->node);
803 btrfs_item_key(root->node, &disk_key, 0);
805 btrfs_node_key(root->node, &disk_key, 0);
806 key_objectid = btrfs_disk_key_objectid(&disk_key);
810 return btrfs_inc_extent_ref(trans, root, root->node->start,
811 root->node->len, owner_objectid,
812 generation, level, key_objectid);
815 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
816 struct extent_buffer *buf)
820 struct btrfs_key key;
821 struct btrfs_file_extent_item *fi;
830 level = btrfs_header_level(buf);
831 nritems = btrfs_header_nritems(buf);
832 for (i = 0; i < nritems; i++) {
835 btrfs_item_key_to_cpu(buf, &key, i);
836 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
838 fi = btrfs_item_ptr(buf, i,
839 struct btrfs_file_extent_item);
840 if (btrfs_file_extent_type(buf, fi) ==
841 BTRFS_FILE_EXTENT_INLINE)
843 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
844 if (disk_bytenr == 0)
846 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
847 btrfs_file_extent_disk_num_bytes(buf, fi),
848 root->root_key.objectid, trans->transid,
849 key.objectid, key.offset);
855 bytenr = btrfs_node_blockptr(buf, i);
856 btrfs_node_key_to_cpu(buf, &key, i);
857 ret = btrfs_inc_extent_ref(trans, root, bytenr,
858 btrfs_level_size(root, level - 1),
859 root->root_key.objectid,
861 level - 1, key.objectid);
872 for (i =0; i < faili; i++) {
875 btrfs_item_key_to_cpu(buf, &key, i);
876 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
878 fi = btrfs_item_ptr(buf, i,
879 struct btrfs_file_extent_item);
880 if (btrfs_file_extent_type(buf, fi) ==
881 BTRFS_FILE_EXTENT_INLINE)
883 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
884 if (disk_bytenr == 0)
886 err = btrfs_free_extent(trans, root, disk_bytenr,
887 btrfs_file_extent_disk_num_bytes(buf,
891 bytenr = btrfs_node_blockptr(buf, i);
892 err = btrfs_free_extent(trans, root, bytenr,
893 btrfs_level_size(root, level - 1), 0);
901 static int write_one_cache_group(struct btrfs_trans_handle *trans,
902 struct btrfs_root *root,
903 struct btrfs_path *path,
904 struct btrfs_block_group_cache *cache)
908 struct btrfs_root *extent_root = root->fs_info->extent_root;
910 struct extent_buffer *leaf;
912 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
917 leaf = path->nodes[0];
918 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
919 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
920 btrfs_mark_buffer_dirty(leaf);
921 btrfs_release_path(extent_root, path);
923 finish_current_insert(trans, extent_root);
924 pending_ret = del_pending_extents(trans, extent_root);
933 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
934 struct btrfs_root *root)
936 struct extent_io_tree *block_group_cache;
937 struct btrfs_block_group_cache *cache;
941 struct btrfs_path *path;
947 block_group_cache = &root->fs_info->block_group_cache;
948 path = btrfs_alloc_path();
953 ret = find_first_extent_bit(block_group_cache, last,
954 &start, &end, BLOCK_GROUP_DIRTY);
959 ret = get_state_private(block_group_cache, start, &ptr);
963 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
964 err = write_one_cache_group(trans, root,
967 * if we fail to write the cache group, we want
968 * to keep it marked dirty in hopes that a later
975 clear_extent_bits(block_group_cache, start, end,
976 BLOCK_GROUP_DIRTY, GFP_NOFS);
978 btrfs_free_path(path);
982 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
985 struct list_head *head = &info->space_info;
986 struct list_head *cur;
987 struct btrfs_space_info *found;
988 list_for_each(cur, head) {
989 found = list_entry(cur, struct btrfs_space_info, list);
990 if (found->flags == flags)
997 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
998 u64 total_bytes, u64 bytes_used,
999 struct btrfs_space_info **space_info)
1001 struct btrfs_space_info *found;
1003 found = __find_space_info(info, flags);
1005 found->total_bytes += total_bytes;
1006 found->bytes_used += bytes_used;
1007 WARN_ON(found->total_bytes < found->bytes_used);
1008 *space_info = found;
1011 found = kmalloc(sizeof(*found), GFP_NOFS);
1015 list_add(&found->list, &info->space_info);
1016 found->flags = flags;
1017 found->total_bytes = total_bytes;
1018 found->bytes_used = bytes_used;
1019 found->bytes_pinned = 0;
1021 *space_info = found;
1026 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1028 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1029 BTRFS_BLOCK_GROUP_RAID1 |
1030 BTRFS_BLOCK_GROUP_DUP);
1032 if (flags & BTRFS_BLOCK_GROUP_DATA)
1033 fs_info->avail_data_alloc_bits |= extra_flags;
1034 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1035 fs_info->avail_metadata_alloc_bits |= extra_flags;
1036 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1037 fs_info->avail_system_alloc_bits |= extra_flags;
1041 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1042 struct btrfs_root *extent_root, u64 alloc_bytes,
1045 struct btrfs_space_info *space_info;
1051 space_info = __find_space_info(extent_root->fs_info, flags);
1053 ret = update_space_info(extent_root->fs_info, flags,
1057 BUG_ON(!space_info);
1059 if (space_info->full)
1062 thresh = div_factor(space_info->total_bytes, 7);
1063 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1067 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1068 if (ret == -ENOSPC) {
1069 printk("space info full %llu\n", (unsigned long long)flags);
1070 space_info->full = 1;
1076 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1077 extent_root->fs_info->chunk_root->root_key.objectid,
1083 static int update_block_group(struct btrfs_trans_handle *trans,
1084 struct btrfs_root *root,
1085 u64 bytenr, u64 num_bytes, int alloc,
1088 struct btrfs_block_group_cache *cache;
1089 struct btrfs_fs_info *info = root->fs_info;
1090 u64 total = num_bytes;
1097 cache = btrfs_lookup_block_group(info, bytenr);
1101 byte_in_group = bytenr - cache->key.objectid;
1102 WARN_ON(byte_in_group > cache->key.offset);
1103 start = cache->key.objectid;
1104 end = start + cache->key.offset - 1;
1105 set_extent_bits(&info->block_group_cache, start, end,
1106 BLOCK_GROUP_DIRTY, GFP_NOFS);
1108 old_val = btrfs_block_group_used(&cache->item);
1109 num_bytes = min(total, cache->key.offset - byte_in_group);
1111 old_val += num_bytes;
1112 cache->space_info->bytes_used += num_bytes;
1114 old_val -= num_bytes;
1115 cache->space_info->bytes_used -= num_bytes;
1117 set_extent_dirty(&info->free_space_cache,
1118 bytenr, bytenr + num_bytes - 1,
1122 btrfs_set_block_group_used(&cache->item, old_val);
1124 bytenr += num_bytes;
1129 static int update_pinned_extents(struct btrfs_root *root,
1130 u64 bytenr, u64 num, int pin)
1133 struct btrfs_block_group_cache *cache;
1134 struct btrfs_fs_info *fs_info = root->fs_info;
1137 set_extent_dirty(&fs_info->pinned_extents,
1138 bytenr, bytenr + num - 1, GFP_NOFS);
1140 clear_extent_dirty(&fs_info->pinned_extents,
1141 bytenr, bytenr + num - 1, GFP_NOFS);
1144 cache = btrfs_lookup_block_group(fs_info, bytenr);
1146 len = min(num, cache->key.offset -
1147 (bytenr - cache->key.objectid));
1149 cache->pinned += len;
1150 cache->space_info->bytes_pinned += len;
1151 fs_info->total_pinned += len;
1153 cache->pinned -= len;
1154 cache->space_info->bytes_pinned -= len;
1155 fs_info->total_pinned -= len;
1163 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1168 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1172 ret = find_first_extent_bit(pinned_extents, last,
1173 &start, &end, EXTENT_DIRTY);
1176 set_extent_dirty(copy, start, end, GFP_NOFS);
1182 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct extent_io_tree *unpin)
1189 struct extent_io_tree *free_space_cache;
1190 free_space_cache = &root->fs_info->free_space_cache;
1193 ret = find_first_extent_bit(unpin, 0, &start, &end,
1197 update_pinned_extents(root, start, end + 1 - start, 0);
1198 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1199 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1204 static int finish_current_insert(struct btrfs_trans_handle *trans,
1205 struct btrfs_root *extent_root)
1209 struct btrfs_fs_info *info = extent_root->fs_info;
1210 struct extent_buffer *eb;
1211 struct btrfs_path *path;
1212 struct btrfs_key ins;
1213 struct btrfs_disk_key first;
1214 struct btrfs_extent_item extent_item;
1219 btrfs_set_stack_extent_refs(&extent_item, 1);
1220 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1221 path = btrfs_alloc_path();
1224 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1225 &end, EXTENT_LOCKED);
1229 ins.objectid = start;
1230 ins.offset = end + 1 - start;
1231 err = btrfs_insert_item(trans, extent_root, &ins,
1232 &extent_item, sizeof(extent_item));
1233 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1235 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1236 level = btrfs_header_level(eb);
1238 btrfs_item_key(eb, &first, 0);
1240 btrfs_node_key(eb, &first, 0);
1242 err = btrfs_insert_extent_backref(trans, extent_root, path,
1243 start, extent_root->root_key.objectid,
1245 btrfs_disk_key_objectid(&first));
1247 free_extent_buffer(eb);
1249 btrfs_free_path(path);
1253 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1257 struct extent_buffer *buf;
1260 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1262 if (btrfs_buffer_uptodate(buf)) {
1264 root->fs_info->running_transaction->transid;
1265 if (btrfs_header_generation(buf) == transid) {
1266 free_extent_buffer(buf);
1270 free_extent_buffer(buf);
1272 update_pinned_extents(root, bytenr, num_bytes, 1);
1274 set_extent_bits(&root->fs_info->pending_del,
1275 bytenr, bytenr + num_bytes - 1,
1276 EXTENT_LOCKED, GFP_NOFS);
1283 * remove an extent from the root, returns 0 on success
1285 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1286 *root, u64 bytenr, u64 num_bytes,
1287 u64 root_objectid, u64 ref_generation,
1288 u64 owner_objectid, u64 owner_offset, int pin,
1291 struct btrfs_path *path;
1292 struct btrfs_key key;
1293 struct btrfs_fs_info *info = root->fs_info;
1294 struct btrfs_extent_ops *ops = info->extent_ops;
1295 struct btrfs_root *extent_root = info->extent_root;
1296 struct extent_buffer *leaf;
1298 int extent_slot = 0;
1299 int found_extent = 0;
1301 struct btrfs_extent_item *ei;
1304 key.objectid = bytenr;
1305 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1306 key.offset = num_bytes;
1308 path = btrfs_alloc_path();
1312 ret = lookup_extent_backref(trans, extent_root, path,
1313 bytenr, root_objectid,
1315 owner_objectid, owner_offset, 1);
1317 struct btrfs_key found_key;
1318 extent_slot = path->slots[0];
1319 while(extent_slot > 0) {
1321 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1323 if (found_key.objectid != bytenr)
1325 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1326 found_key.offset == num_bytes) {
1330 if (path->slots[0] - extent_slot > 5)
1334 ret = btrfs_del_item(trans, extent_root, path);
1336 btrfs_print_leaf(extent_root, path->nodes[0]);
1338 printk("Unable to find ref byte nr %llu root %llu "
1339 " gen %llu owner %llu offset %llu\n",
1340 (unsigned long long)bytenr,
1341 (unsigned long long)root_objectid,
1342 (unsigned long long)ref_generation,
1343 (unsigned long long)owner_objectid,
1344 (unsigned long long)owner_offset);
1346 if (!found_extent) {
1347 btrfs_release_path(extent_root, path);
1348 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1352 extent_slot = path->slots[0];
1355 leaf = path->nodes[0];
1356 ei = btrfs_item_ptr(leaf, extent_slot,
1357 struct btrfs_extent_item);
1358 refs = btrfs_extent_refs(leaf, ei);
1361 btrfs_set_extent_refs(leaf, ei, refs);
1363 btrfs_mark_buffer_dirty(leaf);
1365 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1366 /* if the back ref and the extent are next to each other
1367 * they get deleted below in one shot
1369 path->slots[0] = extent_slot;
1371 } else if (found_extent) {
1372 /* otherwise delete the extent back ref */
1373 ret = btrfs_del_item(trans, extent_root, path);
1375 /* if refs are 0, we need to setup the path for deletion */
1377 btrfs_release_path(extent_root, path);
1378 ret = btrfs_search_slot(trans, extent_root, &key, path,
1391 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1397 /* block accounting for super block */
1398 super_used = btrfs_super_bytes_used(&info->super_copy);
1399 btrfs_set_super_bytes_used(&info->super_copy,
1400 super_used - num_bytes);
1402 /* block accounting for root item */
1403 root_used = btrfs_root_used(&root->root_item);
1404 btrfs_set_root_used(&root->root_item,
1405 root_used - num_bytes);
1406 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1411 if (ops && ops->free_extent)
1412 ops->free_extent(root, bytenr, num_bytes);
1414 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1418 btrfs_free_path(path);
1419 finish_current_insert(trans, extent_root);
1424 * find all the blocks marked as pending in the radix tree and remove
1425 * them from the extent map
1427 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1428 btrfs_root *extent_root)
1434 struct extent_io_tree *pending_del;
1435 struct extent_io_tree *pinned_extents;
1437 pending_del = &extent_root->fs_info->pending_del;
1438 pinned_extents = &extent_root->fs_info->pinned_extents;
1441 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1445 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1446 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1448 ret = __free_extent(trans, extent_root,
1449 start, end + 1 - start,
1450 extent_root->root_key.objectid,
1459 * remove an extent from the root, returns 0 on success
1461 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1462 *root, u64 bytenr, u64 num_bytes,
1463 u64 root_objectid, u64 ref_generation,
1464 u64 owner_objectid, u64 owner_offset, int pin)
1466 struct btrfs_root *extent_root = root->fs_info->extent_root;
1470 WARN_ON(num_bytes < root->sectorsize);
1471 if (!root->ref_cows)
1474 if (root == extent_root) {
1475 pin_down_bytes(root, bytenr, num_bytes, 1);
1478 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1479 ref_generation, owner_objectid, owner_offset,
1481 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1482 return ret ? ret : pending_ret;
1485 static u64 stripe_align(struct btrfs_root *root, u64 val)
1487 u64 mask = ((u64)root->stripesize - 1);
1488 u64 ret = (val + mask) & ~mask;
1493 * walks the btree of allocated extents and find a hole of a given size.
1494 * The key ins is changed to record the hole:
1495 * ins->objectid == block start
1496 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1497 * ins->offset == number of blocks
1498 * Any available blocks before search_start are skipped.
1500 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1501 struct btrfs_root *orig_root,
1502 u64 num_bytes, u64 empty_size,
1503 u64 search_start, u64 search_end,
1504 u64 hint_byte, struct btrfs_key *ins,
1505 u64 exclude_start, u64 exclude_nr,
1509 u64 orig_search_start = search_start;
1510 struct btrfs_root * root = orig_root->fs_info->extent_root;
1511 struct btrfs_fs_info *info = root->fs_info;
1512 u64 total_needed = num_bytes;
1513 struct btrfs_block_group_cache *block_group;
1517 WARN_ON(num_bytes < root->sectorsize);
1518 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1520 if (search_end == (u64)-1)
1521 search_end = btrfs_super_total_bytes(&info->super_copy);
1524 block_group = btrfs_lookup_block_group(info, hint_byte);
1526 hint_byte = search_start;
1527 block_group = btrfs_find_block_group(root, block_group,
1528 hint_byte, data, 1);
1530 block_group = btrfs_find_block_group(root,
1532 search_start, data, 1);
1535 total_needed += empty_size;
1539 block_group = btrfs_lookup_block_group(info, search_start);
1541 block_group = btrfs_lookup_block_group(info,
1544 ret = find_search_start(root, &block_group, &search_start,
1545 total_needed, data);
1549 search_start = stripe_align(root, search_start);
1550 ins->objectid = search_start;
1551 ins->offset = num_bytes;
1553 if (ins->objectid + num_bytes >= search_end)
1556 if (ins->objectid + num_bytes >
1557 block_group->key.objectid + block_group->key.offset) {
1558 search_start = block_group->key.objectid +
1559 block_group->key.offset;
1563 if (test_range_bit(&info->extent_ins, ins->objectid,
1564 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1565 search_start = ins->objectid + num_bytes;
1569 if (test_range_bit(&info->pinned_extents, ins->objectid,
1570 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1571 search_start = ins->objectid + num_bytes;
1575 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1576 ins->objectid < exclude_start + exclude_nr)) {
1577 search_start = exclude_start + exclude_nr;
1581 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1582 block_group = btrfs_lookup_block_group(info, ins->objectid);
1584 trans->block_group = block_group;
1586 ins->offset = num_bytes;
1590 if (search_start + num_bytes >= search_end) {
1592 search_start = orig_search_start;
1599 total_needed -= empty_size;
1604 block_group = btrfs_lookup_block_group(info, search_start);
1606 block_group = btrfs_find_block_group(root, block_group,
1607 search_start, data, 0);
1614 * finds a free extent and does all the dirty work required for allocation
1615 * returns the key for the extent through ins, and a tree buffer for
1616 * the first block of the extent through buf.
1618 * returns 0 if everything worked, non-zero otherwise.
1620 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1621 struct btrfs_root *root,
1622 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1623 u64 owner, u64 owner_offset,
1624 u64 empty_size, u64 hint_byte,
1625 u64 search_end, struct btrfs_key *ins, int data)
1629 u64 super_used, root_used;
1630 u64 search_start = 0;
1632 struct btrfs_fs_info *info = root->fs_info;
1633 struct btrfs_extent_ops *ops = info->extent_ops;
1635 struct btrfs_root *extent_root = info->extent_root;
1636 struct btrfs_path *path;
1637 struct btrfs_extent_item *extent_item;
1638 struct btrfs_extent_ref *ref;
1639 struct btrfs_key keys[2];
1642 alloc_profile = info->avail_data_alloc_bits &
1643 info->data_alloc_profile;
1644 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1645 } else if (root == info->chunk_root || info->force_system_allocs) {
1646 alloc_profile = info->avail_system_alloc_bits &
1647 info->system_alloc_profile;
1648 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1650 alloc_profile = info->avail_metadata_alloc_bits &
1651 info->metadata_alloc_profile;
1652 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1655 if (root->ref_cows) {
1656 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1657 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1659 BTRFS_BLOCK_GROUP_METADATA);
1662 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1663 num_bytes + 2 * 1024 * 1024, data);
1667 WARN_ON(num_bytes < root->sectorsize);
1668 if (ops && ops->alloc_extent) {
1669 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
1671 ret = find_free_extent(trans, root, num_bytes, empty_size,
1672 search_start, search_end, hint_byte,
1673 ins, trans->alloc_exclude_start,
1674 trans->alloc_exclude_nr, data);
1680 /* block accounting for super block */
1681 super_used = btrfs_super_bytes_used(&info->super_copy);
1682 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1684 /* block accounting for root item */
1685 root_used = btrfs_root_used(&root->root_item);
1686 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1688 clear_extent_dirty(&root->fs_info->free_space_cache,
1689 ins->objectid, ins->objectid + ins->offset - 1,
1692 if (root == extent_root) {
1693 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1694 ins->objectid + ins->offset - 1,
1695 EXTENT_LOCKED, GFP_NOFS);
1699 WARN_ON(trans->alloc_exclude_nr);
1700 trans->alloc_exclude_start = ins->objectid;
1701 trans->alloc_exclude_nr = ins->offset;
1703 memcpy(&keys[0], ins, sizeof(*ins));
1704 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1705 owner, owner_offset);
1706 keys[1].objectid = ins->objectid;
1707 keys[1].type = BTRFS_EXTENT_REF_KEY;
1708 sizes[0] = sizeof(*extent_item);
1709 sizes[1] = sizeof(*ref);
1711 path = btrfs_alloc_path();
1714 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1718 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1719 struct btrfs_extent_item);
1720 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1721 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1722 struct btrfs_extent_ref);
1724 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1725 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1726 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1727 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1729 btrfs_mark_buffer_dirty(path->nodes[0]);
1731 trans->alloc_exclude_start = 0;
1732 trans->alloc_exclude_nr = 0;
1733 btrfs_free_path(path);
1734 finish_current_insert(trans, extent_root);
1735 pending_ret = del_pending_extents(trans, extent_root);
1745 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1747 printk("update block group failed for %llu %llu\n",
1748 (unsigned long long)ins->objectid,
1749 (unsigned long long)ins->offset);
1756 * helper function to allocate a block for a given tree
1757 * returns the tree buffer or NULL.
1759 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1760 struct btrfs_root *root,
1762 u64 root_objectid, u64 hint,
1768 ref_generation = trans->transid;
1773 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1774 ref_generation, 0, 0, hint, empty_size);
1778 * helper function to allocate a block for a given tree
1779 * returns the tree buffer or NULL.
1781 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1782 struct btrfs_root *root,
1791 struct btrfs_key ins;
1793 struct extent_buffer *buf;
1795 ret = btrfs_alloc_extent(trans, root, blocksize,
1796 root_objectid, ref_generation,
1797 level, first_objectid, empty_size, hint,
1801 return ERR_PTR(ret);
1803 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1805 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1806 root->root_key.objectid, ref_generation,
1809 return ERR_PTR(-ENOMEM);
1811 btrfs_set_buffer_uptodate(buf);
1812 trans->blocks_used++;
1816 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1817 struct btrfs_root *root,
1818 struct extent_buffer *leaf)
1821 u64 leaf_generation;
1822 struct btrfs_key key;
1823 struct btrfs_file_extent_item *fi;
1828 BUG_ON(!btrfs_is_leaf(leaf));
1829 nritems = btrfs_header_nritems(leaf);
1830 leaf_owner = btrfs_header_owner(leaf);
1831 leaf_generation = btrfs_header_generation(leaf);
1833 for (i = 0; i < nritems; i++) {
1836 btrfs_item_key_to_cpu(leaf, &key, i);
1837 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1839 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1840 if (btrfs_file_extent_type(leaf, fi) ==
1841 BTRFS_FILE_EXTENT_INLINE)
1844 * FIXME make sure to insert a trans record that
1845 * repeats the snapshot del on crash
1847 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1848 if (disk_bytenr == 0)
1850 ret = btrfs_free_extent(trans, root, disk_bytenr,
1851 btrfs_file_extent_disk_num_bytes(leaf, fi),
1852 leaf_owner, leaf_generation,
1853 key.objectid, key.offset, 0);
1859 static void noinline reada_walk_down(struct btrfs_root *root,
1860 struct extent_buffer *node,
1873 nritems = btrfs_header_nritems(node);
1874 level = btrfs_header_level(node);
1878 for (i = slot; i < nritems && skipped < 32; i++) {
1879 bytenr = btrfs_node_blockptr(node, i);
1880 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1881 (last > bytenr && last - bytenr > 32 * 1024))) {
1885 blocksize = btrfs_level_size(root, level - 1);
1887 ret = lookup_extent_ref(NULL, root, bytenr,
1895 mutex_unlock(&root->fs_info->fs_mutex);
1896 ret = readahead_tree_block(root, bytenr, blocksize);
1897 last = bytenr + blocksize;
1899 mutex_lock(&root->fs_info->fs_mutex);
1906 * helper function for drop_snapshot, this walks down the tree dropping ref
1907 * counts as it goes.
1909 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1910 struct btrfs_root *root,
1911 struct btrfs_path *path, int *level)
1916 struct extent_buffer *next;
1917 struct extent_buffer *cur;
1918 struct extent_buffer *parent;
1923 WARN_ON(*level < 0);
1924 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1925 ret = lookup_extent_ref(trans, root,
1926 path->nodes[*level]->start,
1927 path->nodes[*level]->len, &refs);
1933 * walk down to the last node level and free all the leaves
1935 while(*level >= 0) {
1936 WARN_ON(*level < 0);
1937 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1938 cur = path->nodes[*level];
1940 if (btrfs_header_level(cur) != *level)
1943 if (path->slots[*level] >=
1944 btrfs_header_nritems(cur))
1947 ret = drop_leaf_ref(trans, root, cur);
1951 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1952 blocksize = btrfs_level_size(root, *level - 1);
1953 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1956 parent = path->nodes[*level];
1957 root_owner = btrfs_header_owner(parent);
1958 root_gen = btrfs_header_generation(parent);
1959 path->slots[*level]++;
1960 ret = btrfs_free_extent(trans, root, bytenr,
1961 blocksize, root_owner,
1966 next = btrfs_find_tree_block(root, bytenr, blocksize);
1967 if (!next || !btrfs_buffer_uptodate(next)) {
1968 free_extent_buffer(next);
1969 reada_walk_down(root, cur, path->slots[*level]);
1970 mutex_unlock(&root->fs_info->fs_mutex);
1971 next = read_tree_block(root, bytenr, blocksize);
1972 mutex_lock(&root->fs_info->fs_mutex);
1974 /* we dropped the lock, check one more time */
1975 ret = lookup_extent_ref(trans, root, bytenr,
1979 parent = path->nodes[*level];
1980 root_owner = btrfs_header_owner(parent);
1981 root_gen = btrfs_header_generation(parent);
1983 path->slots[*level]++;
1984 free_extent_buffer(next);
1985 ret = btrfs_free_extent(trans, root, bytenr,
1993 WARN_ON(*level <= 0);
1994 if (path->nodes[*level-1])
1995 free_extent_buffer(path->nodes[*level-1]);
1996 path->nodes[*level-1] = next;
1997 *level = btrfs_header_level(next);
1998 path->slots[*level] = 0;
2001 WARN_ON(*level < 0);
2002 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2004 if (path->nodes[*level] == root->node) {
2005 root_owner = root->root_key.objectid;
2006 parent = path->nodes[*level];
2008 parent = path->nodes[*level + 1];
2009 root_owner = btrfs_header_owner(parent);
2012 root_gen = btrfs_header_generation(parent);
2013 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2014 path->nodes[*level]->len,
2015 root_owner, root_gen, 0, 0, 1);
2016 free_extent_buffer(path->nodes[*level]);
2017 path->nodes[*level] = NULL;
2024 * helper for dropping snapshots. This walks back up the tree in the path
2025 * to find the first node higher up where we haven't yet gone through
2028 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2029 struct btrfs_root *root,
2030 struct btrfs_path *path, int *level)
2034 struct btrfs_root_item *root_item = &root->root_item;
2039 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2040 slot = path->slots[i];
2041 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2042 struct extent_buffer *node;
2043 struct btrfs_disk_key disk_key;
2044 node = path->nodes[i];
2047 WARN_ON(*level == 0);
2048 btrfs_node_key(node, &disk_key, path->slots[i]);
2049 memcpy(&root_item->drop_progress,
2050 &disk_key, sizeof(disk_key));
2051 root_item->drop_level = i;
2054 if (path->nodes[*level] == root->node) {
2055 root_owner = root->root_key.objectid;
2057 btrfs_header_generation(path->nodes[*level]);
2059 struct extent_buffer *node;
2060 node = path->nodes[*level + 1];
2061 root_owner = btrfs_header_owner(node);
2062 root_gen = btrfs_header_generation(node);
2064 ret = btrfs_free_extent(trans, root,
2065 path->nodes[*level]->start,
2066 path->nodes[*level]->len,
2067 root_owner, root_gen, 0, 0, 1);
2069 free_extent_buffer(path->nodes[*level]);
2070 path->nodes[*level] = NULL;
2078 * drop the reference count on the tree rooted at 'snap'. This traverses
2079 * the tree freeing any blocks that have a ref count of zero after being
2082 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2088 struct btrfs_path *path;
2091 struct btrfs_root_item *root_item = &root->root_item;
2093 path = btrfs_alloc_path();
2096 level = btrfs_header_level(root->node);
2098 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2099 path->nodes[level] = root->node;
2100 extent_buffer_get(root->node);
2101 path->slots[level] = 0;
2103 struct btrfs_key key;
2104 struct btrfs_disk_key found_key;
2105 struct extent_buffer *node;
2107 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2108 level = root_item->drop_level;
2109 path->lowest_level = level;
2110 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2115 node = path->nodes[level];
2116 btrfs_node_key(node, &found_key, path->slots[level]);
2117 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2118 sizeof(found_key)));
2121 wret = walk_down_tree(trans, root, path, &level);
2127 wret = walk_up_tree(trans, root, path, &level);
2137 for (i = 0; i <= orig_level; i++) {
2138 if (path->nodes[i]) {
2139 free_extent_buffer(path->nodes[i]);
2140 path->nodes[i] = NULL;
2144 btrfs_free_path(path);
2148 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2155 ret = find_first_extent_bit(&info->block_group_cache, 0,
2156 &start, &end, (unsigned int)-1);
2159 ret = get_state_private(&info->block_group_cache, start, &ptr);
2161 kfree((void *)(unsigned long)ptr);
2162 clear_extent_bits(&info->block_group_cache, start,
2163 end, (unsigned int)-1, GFP_NOFS);
2166 ret = find_first_extent_bit(&info->free_space_cache, 0,
2167 &start, &end, EXTENT_DIRTY);
2170 clear_extent_dirty(&info->free_space_cache, start,
2176 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2177 struct btrfs_key *key)
2180 struct btrfs_key found_key;
2181 struct extent_buffer *leaf;
2184 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2188 slot = path->slots[0];
2189 leaf = path->nodes[0];
2190 if (slot >= btrfs_header_nritems(leaf)) {
2191 ret = btrfs_next_leaf(root, path);
2198 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2200 if (found_key.objectid >= key->objectid &&
2201 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2210 int btrfs_read_block_groups(struct btrfs_root *root)
2212 struct btrfs_path *path;
2215 struct btrfs_block_group_cache *cache;
2216 struct btrfs_fs_info *info = root->fs_info;
2217 struct btrfs_space_info *space_info;
2218 struct extent_io_tree *block_group_cache;
2219 struct btrfs_key key;
2220 struct btrfs_key found_key;
2221 struct extent_buffer *leaf;
2223 block_group_cache = &info->block_group_cache;
2225 root = info->extent_root;
2228 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2229 path = btrfs_alloc_path();
2234 ret = find_first_block_group(root, path, &key);
2242 leaf = path->nodes[0];
2243 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2244 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2250 read_extent_buffer(leaf, &cache->item,
2251 btrfs_item_ptr_offset(leaf, path->slots[0]),
2252 sizeof(cache->item));
2253 memcpy(&cache->key, &found_key, sizeof(found_key));
2256 key.objectid = found_key.objectid + found_key.offset;
2257 btrfs_release_path(root, path);
2258 cache->flags = btrfs_block_group_flags(&cache->item);
2260 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2261 bit = BLOCK_GROUP_DATA;
2262 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2263 bit = BLOCK_GROUP_SYSTEM;
2264 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2265 bit = BLOCK_GROUP_METADATA;
2267 set_avail_alloc_bits(info, cache->flags);
2269 ret = update_space_info(info, cache->flags, found_key.offset,
2270 btrfs_block_group_used(&cache->item),
2273 cache->space_info = space_info;
2275 /* use EXTENT_LOCKED to prevent merging */
2276 set_extent_bits(block_group_cache, found_key.objectid,
2277 found_key.objectid + found_key.offset - 1,
2278 bit | EXTENT_LOCKED, GFP_NOFS);
2279 set_state_private(block_group_cache, found_key.objectid,
2280 (unsigned long)cache);
2283 btrfs_super_total_bytes(&info->super_copy))
2288 btrfs_free_path(path);
2292 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2293 struct btrfs_root *root, u64 bytes_used,
2294 u64 type, u64 chunk_tree, u64 chunk_objectid,
2299 struct btrfs_root *extent_root;
2300 struct btrfs_block_group_cache *cache;
2301 struct extent_io_tree *block_group_cache;
2303 extent_root = root->fs_info->extent_root;
2304 block_group_cache = &root->fs_info->block_group_cache;
2306 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2308 cache->key.objectid = chunk_objectid;
2309 cache->key.offset = size;
2310 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2311 btrfs_set_block_group_used(&cache->item, bytes_used);
2312 btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
2313 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2314 cache->flags = type;
2315 btrfs_set_block_group_flags(&cache->item, type);
2317 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2318 &cache->space_info);
2321 bit = block_group_state_bits(type);
2322 set_extent_bits(block_group_cache, chunk_objectid,
2323 chunk_objectid + size - 1,
2324 bit | EXTENT_LOCKED, GFP_NOFS);
2325 set_state_private(block_group_cache, chunk_objectid,
2326 (unsigned long)cache);
2328 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2329 sizeof(cache->item));
2332 finish_current_insert(trans, extent_root);
2333 ret = del_pending_extents(trans, extent_root);
2335 set_avail_alloc_bits(extent_root->fs_info, type);
2339 u64 btrfs_hash_extent_ref(u64 root_objectid, u64 ref_generation,
2340 u64 owner, u64 owner_offset)
2342 return hash_extent_ref(root_objectid, ref_generation,
2343 owner, owner_offset);
2346 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
2347 struct btrfs_root *root,
2348 u64 bytenr, u64 num_bytes, int alloc,
2351 return update_block_group(trans, root, bytenr, num_bytes,