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.
19 #include <linux/sched.h>
20 #include <linux/crc32c.h>
21 #include <linux/pagemap.h>
25 #include "print-tree.h"
26 #include "transaction.h"
28 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
29 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
30 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
32 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
33 btrfs_root *extent_root);
34 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
35 btrfs_root *extent_root);
36 static int find_previous_extent(struct btrfs_root *root,
37 struct btrfs_path *path)
39 struct btrfs_key found_key;
40 struct extent_buffer *leaf;
44 if (path->slots[0] == 0) {
45 ret = btrfs_prev_leaf(root, path);
51 leaf = path->nodes[0];
52 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
53 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
59 static int cache_block_group(struct btrfs_root *root,
60 struct btrfs_block_group_cache *block_group)
62 struct btrfs_path *path;
65 struct extent_buffer *leaf;
66 struct extent_io_tree *free_space_cache;
76 root = root->fs_info->extent_root;
77 free_space_cache = &root->fs_info->free_space_cache;
79 if (block_group->cached)
82 path = btrfs_alloc_path();
87 first_free = block_group->key.objectid;
88 key.objectid = block_group->key.objectid;
90 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
91 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
94 ret = find_previous_extent(root, path);
98 leaf = path->nodes[0];
99 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
100 if (key.objectid + key.offset > first_free)
101 first_free = key.objectid + key.offset;
104 leaf = path->nodes[0];
105 slot = path->slots[0];
106 if (slot >= btrfs_header_nritems(leaf)) {
107 ret = btrfs_next_leaf(root, path);
116 btrfs_item_key_to_cpu(leaf, &key, slot);
117 if (key.objectid < block_group->key.objectid) {
120 if (key.objectid >= block_group->key.objectid +
121 block_group->key.offset) {
125 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
130 if (key.objectid > last) {
131 hole_size = key.objectid - last;
132 set_extent_dirty(free_space_cache, last,
133 last + hole_size - 1,
136 last = key.objectid + key.offset;
144 if (block_group->key.objectid +
145 block_group->key.offset > last) {
146 hole_size = block_group->key.objectid +
147 block_group->key.offset - last;
148 set_extent_dirty(free_space_cache, last,
149 last + hole_size - 1, GFP_NOFS);
151 block_group->cached = 1;
153 btrfs_free_path(path);
157 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
161 struct extent_io_tree *block_group_cache;
162 struct btrfs_block_group_cache *block_group = NULL;
168 block_group_cache = &info->block_group_cache;
169 ret = find_first_extent_bit(block_group_cache,
170 bytenr, &start, &end,
171 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
175 ret = get_state_private(block_group_cache, start, &ptr);
179 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
180 if (block_group->key.objectid <= bytenr && bytenr <
181 block_group->key.objectid + block_group->key.offset)
185 static u64 noinline find_search_start(struct btrfs_root *root,
186 struct btrfs_block_group_cache **cache_ret,
187 u64 search_start, int num, int data)
190 struct btrfs_block_group_cache *cache = *cache_ret;
191 struct extent_io_tree *free_space_cache;
192 struct extent_state *state;
202 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
203 free_space_cache = &root->fs_info->free_space_cache;
206 ret = cache_block_group(root, cache);
210 last = max(search_start, cache->key.objectid);
213 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
214 last, &start, &end, EXTENT_DIRTY);
221 start = max(last, start);
223 if (last - start < num) {
224 if (last == cache->key.objectid + cache->key.offset)
228 if (data != BTRFS_BLOCK_GROUP_MIXED &&
229 start + num > cache->key.objectid + cache->key.offset)
231 if (start + num > total_fs_bytes)
236 cache = btrfs_lookup_block_group(root->fs_info, search_start);
238 printk("Unable to find block group for %Lu\n",
246 last = cache->key.objectid + cache->key.offset;
248 cache = btrfs_lookup_block_group(root->fs_info, last);
249 if (!cache || cache->key.objectid >= total_fs_bytes) {
254 data = BTRFS_BLOCK_GROUP_MIXED;
259 if (cache_miss && !cache->cached) {
260 cache_block_group(root, cache);
262 cache = btrfs_lookup_block_group(root->fs_info, last);
264 cache = btrfs_find_block_group(root, cache, last, data, 0);
272 static u64 div_factor(u64 num, int factor)
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;
304 block_group_cache = &info->block_group_cache;
305 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
310 if (data == BTRFS_BLOCK_GROUP_MIXED) {
311 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
314 bit = BLOCK_GROUP_DATA;
316 bit = BLOCK_GROUP_METADATA;
318 if (search_start && search_start < total_fs_bytes) {
319 struct btrfs_block_group_cache *shint;
320 shint = btrfs_lookup_block_group(info, search_start);
321 if (shint && (shint->data == data ||
322 shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
323 used = btrfs_block_group_used(&shint->item);
324 if (used + shint->pinned <
325 div_factor(shint->key.offset, factor)) {
330 if (hint && hint->key.objectid < total_fs_bytes &&
331 (hint->data == data || hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
332 used = btrfs_block_group_used(&hint->item);
333 if (used + hint->pinned <
334 div_factor(hint->key.offset, factor)) {
337 last = hint->key.objectid + hint->key.offset;
341 hint_last = max(hint->key.objectid, search_start);
343 hint_last = search_start;
345 if (hint_last >= total_fs_bytes)
346 hint_last = search_start;
351 ret = find_first_extent_bit(block_group_cache, last,
356 ret = get_state_private(block_group_cache, start, &ptr);
360 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
361 last = cache->key.objectid + cache->key.offset;
362 used = btrfs_block_group_used(&cache->item);
364 if (cache->key.objectid > total_fs_bytes)
368 free_check = cache->key.offset;
370 free_check = div_factor(cache->key.offset, factor);
371 if (used + cache->pinned < free_check) {
384 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
392 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
393 u64 owner, u64 owner_offset)
395 u32 high_crc = ~(u32)0;
396 u32 low_crc = ~(u32)0;
399 lenum = cpu_to_le64(root_objectid);
400 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
401 lenum = cpu_to_le64(ref_generation);
402 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
403 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
404 lenum = cpu_to_le64(owner);
405 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
406 lenum = cpu_to_le64(owner_offset);
407 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
409 return ((u64)high_crc << 32) | (u64)low_crc;
412 static int match_extent_ref(struct extent_buffer *leaf,
413 struct btrfs_extent_ref *disk_ref,
414 struct btrfs_extent_ref *cpu_ref)
419 if (cpu_ref->objectid)
420 len = sizeof(*cpu_ref);
422 len = 2 * sizeof(u64);
423 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
428 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
429 struct btrfs_root *root,
430 struct btrfs_path *path, u64 bytenr,
432 u64 ref_generation, u64 owner,
433 u64 owner_offset, int del)
436 struct btrfs_key key;
437 struct btrfs_key found_key;
438 struct btrfs_extent_ref ref;
439 struct extent_buffer *leaf;
440 struct btrfs_extent_ref *disk_ref;
444 btrfs_set_stack_ref_root(&ref, root_objectid);
445 btrfs_set_stack_ref_generation(&ref, ref_generation);
446 btrfs_set_stack_ref_objectid(&ref, owner);
447 btrfs_set_stack_ref_offset(&ref, owner_offset);
449 hash = hash_extent_ref(root_objectid, ref_generation, owner,
452 key.objectid = bytenr;
453 key.type = BTRFS_EXTENT_REF_KEY;
456 ret = btrfs_search_slot(trans, root, &key, path,
460 leaf = path->nodes[0];
462 u32 nritems = btrfs_header_nritems(leaf);
463 if (path->slots[0] >= nritems) {
464 ret2 = btrfs_next_leaf(root, path);
467 leaf = path->nodes[0];
469 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
470 if (found_key.objectid != bytenr ||
471 found_key.type != BTRFS_EXTENT_REF_KEY)
473 key.offset = found_key.offset;
475 btrfs_release_path(root, path);
479 disk_ref = btrfs_item_ptr(path->nodes[0],
481 struct btrfs_extent_ref);
482 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
487 key.offset = found_key.offset + 1;
488 btrfs_release_path(root, path);
495 * Back reference rules. Back refs have three main goals:
497 * 1) differentiate between all holders of references to an extent so that
498 * when a reference is dropped we can make sure it was a valid reference
499 * before freeing the extent.
501 * 2) Provide enough information to quickly find the holders of an extent
502 * if we notice a given block is corrupted or bad.
504 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
505 * maintenance. This is actually the same as #2, but with a slightly
506 * different use case.
508 * File extents can be referenced by:
510 * - multiple snapshots, subvolumes, or different generations in one subvol
511 * - different files inside a single subvolume (in theory, not implemented yet)
512 * - different offsets inside a file (bookend extents in file.c)
514 * The extent ref structure has fields for:
516 * - Objectid of the subvolume root
517 * - Generation number of the tree holding the reference
518 * - objectid of the file holding the reference
519 * - offset in the file corresponding to the key holding the reference
521 * When a file extent is allocated the fields are filled in:
522 * (root_key.objectid, trans->transid, inode objectid, offset in file)
524 * When a leaf is cow'd new references are added for every file extent found
525 * in the leaf. It looks the same as the create case, but trans->transid
526 * will be different when the block is cow'd.
528 * (root_key.objectid, trans->transid, inode objectid, offset in file)
530 * When a file extent is removed either during snapshot deletion or file
531 * truncation, the corresponding back reference is found
534 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
535 * inode objectid, offset in file)
537 * Btree extents can be referenced by:
539 * - Different subvolumes
540 * - Different generations of the same subvolume
542 * Storing sufficient information for a full reverse mapping of a btree
543 * block would require storing the lowest key of the block in the backref,
544 * and it would require updating that lowest key either before write out or
545 * every time it changed. Instead, the objectid of the lowest key is stored
546 * along with the level of the tree block. This provides a hint
547 * about where in the btree the block can be found. Searches through the
548 * btree only need to look for a pointer to that block, so they stop one
549 * level higher than the level recorded in the backref.
551 * Some btrees do not do reference counting on their extents. These
552 * include the extent tree and the tree of tree roots. Backrefs for these
553 * trees always have a generation of zero.
555 * When a tree block is created, back references are inserted:
557 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
559 * When a tree block is cow'd in a reference counted root,
560 * new back references are added for all the blocks it points to.
561 * These are of the form (trans->transid will have increased since creation):
563 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
565 * Because the lowest_key_objectid and the level are just hints
566 * they are not used when backrefs are deleted. When a backref is deleted:
568 * if backref was for a tree root:
569 * root_objectid = root->root_key.objectid
571 * root_objectid = btrfs_header_owner(parent)
573 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
575 * Back Reference Key hashing:
577 * Back references have four fields, each 64 bits long. Unfortunately,
578 * This is hashed into a single 64 bit number and placed into the key offset.
579 * The key objectid corresponds to the first byte in the extent, and the
580 * key type is set to BTRFS_EXTENT_REF_KEY
582 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
583 struct btrfs_root *root,
584 struct btrfs_path *path, u64 bytenr,
585 u64 root_objectid, u64 ref_generation,
586 u64 owner, u64 owner_offset)
589 struct btrfs_key key;
590 struct btrfs_extent_ref ref;
591 struct btrfs_extent_ref *disk_ref;
594 btrfs_set_stack_ref_root(&ref, root_objectid);
595 btrfs_set_stack_ref_generation(&ref, ref_generation);
596 btrfs_set_stack_ref_objectid(&ref, owner);
597 btrfs_set_stack_ref_offset(&ref, owner_offset);
599 hash = hash_extent_ref(root_objectid, ref_generation, owner,
602 key.objectid = bytenr;
603 key.type = BTRFS_EXTENT_REF_KEY;
605 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
606 while (ret == -EEXIST) {
607 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
608 struct btrfs_extent_ref);
609 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
612 btrfs_release_path(root, path);
613 ret = btrfs_insert_empty_item(trans, root, path, &key,
618 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
619 struct btrfs_extent_ref);
620 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
622 btrfs_mark_buffer_dirty(path->nodes[0]);
624 btrfs_release_path(root, path);
628 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
629 struct btrfs_root *root,
630 u64 bytenr, u64 num_bytes,
631 u64 root_objectid, u64 ref_generation,
632 u64 owner, u64 owner_offset)
634 struct btrfs_path *path;
636 struct btrfs_key key;
637 struct extent_buffer *l;
638 struct btrfs_extent_item *item;
641 WARN_ON(num_bytes < root->sectorsize);
642 path = btrfs_alloc_path();
647 key.objectid = bytenr;
648 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
649 key.offset = num_bytes;
650 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
659 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
660 refs = btrfs_extent_refs(l, item);
661 btrfs_set_extent_refs(l, item, refs + 1);
662 btrfs_mark_buffer_dirty(path->nodes[0]);
664 btrfs_release_path(root->fs_info->extent_root, path);
667 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
668 path, bytenr, root_objectid,
669 ref_generation, owner, owner_offset);
671 finish_current_insert(trans, root->fs_info->extent_root);
672 del_pending_extents(trans, root->fs_info->extent_root);
674 btrfs_free_path(path);
678 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
679 struct btrfs_root *root)
681 finish_current_insert(trans, root->fs_info->extent_root);
682 del_pending_extents(trans, root->fs_info->extent_root);
686 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
687 struct btrfs_root *root, u64 bytenr,
688 u64 num_bytes, u32 *refs)
690 struct btrfs_path *path;
692 struct btrfs_key key;
693 struct extent_buffer *l;
694 struct btrfs_extent_item *item;
696 WARN_ON(num_bytes < root->sectorsize);
697 path = btrfs_alloc_path();
699 key.objectid = bytenr;
700 key.offset = num_bytes;
701 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
702 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
707 btrfs_print_leaf(root, path->nodes[0]);
708 printk("failed to find block number %Lu\n", bytenr);
712 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
713 *refs = btrfs_extent_refs(l, item);
715 btrfs_free_path(path);
719 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
720 struct btrfs_path *count_path,
723 struct btrfs_root *extent_root = root->fs_info->extent_root;
724 struct btrfs_path *path;
727 u64 root_objectid = root->root_key.objectid;
732 struct btrfs_key key;
733 struct btrfs_key found_key;
734 struct extent_buffer *l;
735 struct btrfs_extent_item *item;
736 struct btrfs_extent_ref *ref_item;
739 path = btrfs_alloc_path();
742 bytenr = first_extent;
744 bytenr = count_path->nodes[level]->start;
747 key.objectid = bytenr;
750 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
751 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
757 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
759 if (found_key.objectid != bytenr ||
760 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
764 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
767 nritems = btrfs_header_nritems(l);
768 if (path->slots[0] >= nritems) {
769 ret = btrfs_next_leaf(extent_root, path);
774 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
775 if (found_key.objectid != bytenr)
778 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
784 ref_item = btrfs_item_ptr(l, path->slots[0],
785 struct btrfs_extent_ref);
786 found_objectid = btrfs_ref_root(l, ref_item);
788 if (found_objectid != root_objectid) {
795 if (cur_count == 0) {
799 if (level >= 0 && root->node == count_path->nodes[level])
802 btrfs_release_path(root, path);
806 btrfs_free_path(path);
809 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root, u64 owner_objectid)
816 struct btrfs_disk_key disk_key;
818 level = btrfs_header_level(root->node);
819 generation = trans->transid;
820 nritems = btrfs_header_nritems(root->node);
823 btrfs_item_key(root->node, &disk_key, 0);
825 btrfs_node_key(root->node, &disk_key, 0);
826 key_objectid = btrfs_disk_key_objectid(&disk_key);
830 return btrfs_inc_extent_ref(trans, root, root->node->start,
831 root->node->len, owner_objectid,
832 generation, level, key_objectid);
835 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
836 struct extent_buffer *buf)
840 struct btrfs_key key;
841 struct btrfs_file_extent_item *fi;
850 level = btrfs_header_level(buf);
851 nritems = btrfs_header_nritems(buf);
852 for (i = 0; i < nritems; i++) {
855 btrfs_item_key_to_cpu(buf, &key, i);
856 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
858 fi = btrfs_item_ptr(buf, i,
859 struct btrfs_file_extent_item);
860 if (btrfs_file_extent_type(buf, fi) ==
861 BTRFS_FILE_EXTENT_INLINE)
863 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
864 if (disk_bytenr == 0)
866 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
867 btrfs_file_extent_disk_num_bytes(buf, fi),
868 root->root_key.objectid, trans->transid,
869 key.objectid, key.offset);
875 bytenr = btrfs_node_blockptr(buf, i);
876 btrfs_node_key_to_cpu(buf, &key, i);
877 ret = btrfs_inc_extent_ref(trans, root, bytenr,
878 btrfs_level_size(root, level - 1),
879 root->root_key.objectid,
881 level - 1, key.objectid);
892 for (i =0; i < faili; i++) {
895 btrfs_item_key_to_cpu(buf, &key, i);
896 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
898 fi = btrfs_item_ptr(buf, i,
899 struct btrfs_file_extent_item);
900 if (btrfs_file_extent_type(buf, fi) ==
901 BTRFS_FILE_EXTENT_INLINE)
903 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
904 if (disk_bytenr == 0)
906 err = btrfs_free_extent(trans, root, disk_bytenr,
907 btrfs_file_extent_disk_num_bytes(buf,
911 bytenr = btrfs_node_blockptr(buf, i);
912 err = btrfs_free_extent(trans, root, bytenr,
913 btrfs_level_size(root, level - 1), 0);
921 static int write_one_cache_group(struct btrfs_trans_handle *trans,
922 struct btrfs_root *root,
923 struct btrfs_path *path,
924 struct btrfs_block_group_cache *cache)
928 struct btrfs_root *extent_root = root->fs_info->extent_root;
930 struct extent_buffer *leaf;
932 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
937 leaf = path->nodes[0];
938 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
939 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
940 btrfs_mark_buffer_dirty(leaf);
941 btrfs_release_path(extent_root, path);
943 finish_current_insert(trans, extent_root);
944 pending_ret = del_pending_extents(trans, extent_root);
953 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
954 struct btrfs_root *root)
956 struct extent_io_tree *block_group_cache;
957 struct btrfs_block_group_cache *cache;
961 struct btrfs_path *path;
967 block_group_cache = &root->fs_info->block_group_cache;
968 path = btrfs_alloc_path();
973 ret = find_first_extent_bit(block_group_cache, last,
974 &start, &end, BLOCK_GROUP_DIRTY);
979 ret = get_state_private(block_group_cache, start, &ptr);
983 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
984 err = write_one_cache_group(trans, root,
987 * if we fail to write the cache group, we want
988 * to keep it marked dirty in hopes that a later
995 clear_extent_bits(block_group_cache, start, end,
996 BLOCK_GROUP_DIRTY, GFP_NOFS);
998 btrfs_free_path(path);
1002 static int update_block_group(struct btrfs_trans_handle *trans,
1003 struct btrfs_root *root,
1004 u64 bytenr, u64 num_bytes, int alloc,
1005 int mark_free, int data)
1007 struct btrfs_block_group_cache *cache;
1008 struct btrfs_fs_info *info = root->fs_info;
1009 u64 total = num_bytes;
1016 cache = btrfs_lookup_block_group(info, bytenr);
1020 byte_in_group = bytenr - cache->key.objectid;
1021 WARN_ON(byte_in_group > cache->key.offset);
1022 start = cache->key.objectid;
1023 end = start + cache->key.offset - 1;
1024 set_extent_bits(&info->block_group_cache, start, end,
1025 BLOCK_GROUP_DIRTY, GFP_NOFS);
1027 old_val = btrfs_block_group_used(&cache->item);
1028 num_bytes = min(total, cache->key.offset - byte_in_group);
1030 if (cache->data != data &&
1031 old_val < (cache->key.offset >> 1)) {
1036 bit_to_clear = BLOCK_GROUP_METADATA;
1037 bit_to_set = BLOCK_GROUP_DATA;
1038 cache->item.flags &=
1039 ~BTRFS_BLOCK_GROUP_MIXED;
1040 cache->item.flags |=
1041 BTRFS_BLOCK_GROUP_DATA;
1043 bit_to_clear = BLOCK_GROUP_DATA;
1044 bit_to_set = BLOCK_GROUP_METADATA;
1045 cache->item.flags &=
1046 ~BTRFS_BLOCK_GROUP_MIXED;
1047 cache->item.flags &=
1048 ~BTRFS_BLOCK_GROUP_DATA;
1050 clear_extent_bits(&info->block_group_cache,
1051 start, end, bit_to_clear,
1053 set_extent_bits(&info->block_group_cache,
1054 start, end, bit_to_set,
1056 } else if (cache->data != data &&
1057 cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1058 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1059 set_extent_bits(&info->block_group_cache,
1062 BLOCK_GROUP_METADATA,
1065 old_val += num_bytes;
1067 old_val -= num_bytes;
1069 set_extent_dirty(&info->free_space_cache,
1070 bytenr, bytenr + num_bytes - 1,
1074 btrfs_set_block_group_used(&cache->item, old_val);
1076 bytenr += num_bytes;
1080 static int update_pinned_extents(struct btrfs_root *root,
1081 u64 bytenr, u64 num, int pin)
1084 struct btrfs_block_group_cache *cache;
1085 struct btrfs_fs_info *fs_info = root->fs_info;
1088 set_extent_dirty(&fs_info->pinned_extents,
1089 bytenr, bytenr + num - 1, GFP_NOFS);
1091 clear_extent_dirty(&fs_info->pinned_extents,
1092 bytenr, bytenr + num - 1, GFP_NOFS);
1095 cache = btrfs_lookup_block_group(fs_info, bytenr);
1097 len = min(num, cache->key.offset -
1098 (bytenr - cache->key.objectid));
1100 cache->pinned += len;
1101 fs_info->total_pinned += len;
1103 cache->pinned -= len;
1104 fs_info->total_pinned -= len;
1112 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1117 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1121 ret = find_first_extent_bit(pinned_extents, last,
1122 &start, &end, EXTENT_DIRTY);
1125 set_extent_dirty(copy, start, end, GFP_NOFS);
1131 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1132 struct btrfs_root *root,
1133 struct extent_io_tree *unpin)
1138 struct extent_io_tree *free_space_cache;
1139 free_space_cache = &root->fs_info->free_space_cache;
1142 ret = find_first_extent_bit(unpin, 0, &start, &end,
1146 update_pinned_extents(root, start, end + 1 - start, 0);
1147 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1148 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1153 static int finish_current_insert(struct btrfs_trans_handle *trans,
1154 struct btrfs_root *extent_root)
1158 struct btrfs_fs_info *info = extent_root->fs_info;
1159 struct extent_buffer *eb;
1160 struct btrfs_path *path;
1161 struct btrfs_key ins;
1162 struct btrfs_disk_key first;
1163 struct btrfs_extent_item extent_item;
1168 btrfs_set_stack_extent_refs(&extent_item, 1);
1169 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1170 path = btrfs_alloc_path();
1173 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1174 &end, EXTENT_LOCKED);
1178 ins.objectid = start;
1179 ins.offset = end + 1 - start;
1180 err = btrfs_insert_item(trans, extent_root, &ins,
1181 &extent_item, sizeof(extent_item));
1182 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1184 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1185 level = btrfs_header_level(eb);
1187 btrfs_item_key(eb, &first, 0);
1189 btrfs_node_key(eb, &first, 0);
1191 err = btrfs_insert_extent_backref(trans, extent_root, path,
1192 start, extent_root->root_key.objectid,
1194 btrfs_disk_key_objectid(&first));
1196 free_extent_buffer(eb);
1198 btrfs_free_path(path);
1202 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1206 struct extent_buffer *buf;
1209 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1211 if (btrfs_buffer_uptodate(buf)) {
1213 root->fs_info->running_transaction->transid;
1214 u64 header_transid =
1215 btrfs_header_generation(buf);
1216 if (header_transid == transid) {
1217 clean_tree_block(NULL, root, buf);
1218 free_extent_buffer(buf);
1222 free_extent_buffer(buf);
1224 update_pinned_extents(root, bytenr, num_bytes, 1);
1226 set_extent_bits(&root->fs_info->pending_del,
1227 bytenr, bytenr + num_bytes - 1,
1228 EXTENT_LOCKED, GFP_NOFS);
1235 * remove an extent from the root, returns 0 on success
1237 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1238 *root, u64 bytenr, u64 num_bytes,
1239 u64 root_objectid, u64 ref_generation,
1240 u64 owner_objectid, u64 owner_offset, int pin,
1243 struct btrfs_path *path;
1244 struct btrfs_key key;
1245 struct btrfs_fs_info *info = root->fs_info;
1246 struct btrfs_root *extent_root = info->extent_root;
1247 struct extent_buffer *leaf;
1249 int extent_slot = 0;
1250 int found_extent = 0;
1252 struct btrfs_extent_item *ei;
1255 key.objectid = bytenr;
1256 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1257 key.offset = num_bytes;
1258 path = btrfs_alloc_path();
1263 ret = lookup_extent_backref(trans, extent_root, path,
1264 bytenr, root_objectid,
1266 owner_objectid, owner_offset, 1);
1268 struct btrfs_key found_key;
1269 extent_slot = path->slots[0];
1270 while(extent_slot > 0) {
1272 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1274 if (found_key.objectid != bytenr)
1276 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1277 found_key.offset == num_bytes) {
1281 if (path->slots[0] - extent_slot > 5)
1285 ret = btrfs_del_item(trans, extent_root, path);
1287 btrfs_print_leaf(extent_root, path->nodes[0]);
1289 printk("Unable to find ref byte nr %Lu root %Lu "
1290 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1291 root_objectid, ref_generation, owner_objectid,
1294 if (!found_extent) {
1295 btrfs_release_path(extent_root, path);
1296 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1300 extent_slot = path->slots[0];
1303 leaf = path->nodes[0];
1304 ei = btrfs_item_ptr(leaf, extent_slot,
1305 struct btrfs_extent_item);
1306 refs = btrfs_extent_refs(leaf, ei);
1309 btrfs_set_extent_refs(leaf, ei, refs);
1311 btrfs_mark_buffer_dirty(leaf);
1313 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1314 /* if the back ref and the extent are next to each other
1315 * they get deleted below in one shot
1317 path->slots[0] = extent_slot;
1319 } else if (found_extent) {
1320 /* otherwise delete the extent back ref */
1321 ret = btrfs_del_item(trans, extent_root, path);
1323 /* if refs are 0, we need to setup the path for deletion */
1325 btrfs_release_path(extent_root, path);
1326 ret = btrfs_search_slot(trans, extent_root, &key, path,
1339 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1345 /* block accounting for super block */
1346 super_used = btrfs_super_bytes_used(&info->super_copy);
1347 btrfs_set_super_bytes_used(&info->super_copy,
1348 super_used - num_bytes);
1350 /* block accounting for root item */
1351 root_used = btrfs_root_used(&root->root_item);
1352 btrfs_set_root_used(&root->root_item,
1353 root_used - num_bytes);
1354 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1359 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1363 btrfs_free_path(path);
1364 finish_current_insert(trans, extent_root);
1369 * find all the blocks marked as pending in the radix tree and remove
1370 * them from the extent map
1372 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1373 btrfs_root *extent_root)
1379 struct extent_io_tree *pending_del;
1380 struct extent_io_tree *pinned_extents;
1382 pending_del = &extent_root->fs_info->pending_del;
1383 pinned_extents = &extent_root->fs_info->pinned_extents;
1386 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1390 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1391 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1393 ret = __free_extent(trans, extent_root,
1394 start, end + 1 - start,
1395 extent_root->root_key.objectid,
1404 * remove an extent from the root, returns 0 on success
1406 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1407 *root, u64 bytenr, u64 num_bytes,
1408 u64 root_objectid, u64 ref_generation,
1409 u64 owner_objectid, u64 owner_offset, int pin)
1411 struct btrfs_root *extent_root = root->fs_info->extent_root;
1415 WARN_ON(num_bytes < root->sectorsize);
1416 if (!root->ref_cows)
1419 if (root == extent_root) {
1420 pin_down_bytes(root, bytenr, num_bytes, 1);
1423 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1424 ref_generation, owner_objectid, owner_offset,
1426 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1427 return ret ? ret : pending_ret;
1430 static u64 stripe_align(struct btrfs_root *root, u64 val)
1432 u64 mask = ((u64)root->stripesize - 1);
1433 u64 ret = (val + mask) & ~mask;
1438 * walks the btree of allocated extents and find a hole of a given size.
1439 * The key ins is changed to record the hole:
1440 * ins->objectid == block start
1441 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1442 * ins->offset == number of blocks
1443 * Any available blocks before search_start are skipped.
1445 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *orig_root,
1447 u64 num_bytes, u64 empty_size,
1448 u64 search_start, u64 search_end,
1449 u64 hint_byte, struct btrfs_key *ins,
1450 u64 exclude_start, u64 exclude_nr,
1453 struct btrfs_path *path;
1454 struct btrfs_key key;
1460 u64 *last_ptr = NULL;
1461 u64 orig_search_start = search_start;
1463 struct extent_buffer *l;
1464 struct btrfs_root * root = orig_root->fs_info->extent_root;
1465 struct btrfs_fs_info *info = root->fs_info;
1466 u64 total_needed = num_bytes;
1468 struct btrfs_block_group_cache *block_group;
1474 WARN_ON(num_bytes < root->sectorsize);
1475 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1477 level = btrfs_header_level(root->node);
1479 if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1480 data = BTRFS_BLOCK_GROUP_MIXED;
1483 if (search_end == (u64)-1)
1484 search_end = btrfs_super_total_bytes(&info->super_copy);
1486 block_group = btrfs_lookup_block_group(info, hint_byte);
1488 hint_byte = search_start;
1489 block_group = btrfs_find_block_group(root, block_group,
1490 hint_byte, data, 1);
1492 block_group = btrfs_find_block_group(root,
1494 search_start, data, 1);
1497 total_needed += empty_size;
1498 path = btrfs_alloc_path();
1501 block_group = btrfs_lookup_block_group(info, search_start);
1503 block_group = btrfs_lookup_block_group(info,
1506 search_start = find_search_start(root, &block_group, search_start,
1507 total_needed, data);
1508 search_start = stripe_align(root, search_start);
1509 cached_start = search_start;
1510 btrfs_init_path(path);
1511 ins->objectid = search_start;
1516 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1519 ret = find_previous_extent(root, path);
1523 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1526 slot = path->slots[0];
1527 if (slot >= btrfs_header_nritems(l)) {
1528 ret = btrfs_next_leaf(root, path);
1534 search_start = max(search_start,
1535 block_group->key.objectid);
1537 aligned = stripe_align(root, search_start);
1538 ins->objectid = aligned;
1539 if (aligned >= search_end) {
1543 ins->offset = search_end - aligned;
1547 ins->objectid = stripe_align(root,
1548 last_byte > search_start ?
1549 last_byte : search_start);
1550 if (search_end <= ins->objectid) {
1554 ins->offset = search_end - ins->objectid;
1555 BUG_ON(ins->objectid >= search_end);
1558 btrfs_item_key_to_cpu(l, &key, slot);
1560 if (key.objectid >= search_start && key.objectid > last_byte &&
1562 if (last_byte < search_start)
1563 last_byte = search_start;
1564 aligned = stripe_align(root, last_byte);
1565 hole_size = key.objectid - aligned;
1566 if (key.objectid > aligned && hole_size >= num_bytes) {
1567 ins->objectid = aligned;
1568 ins->offset = hole_size;
1572 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1573 if (!start_found && btrfs_key_type(&key) ==
1574 BTRFS_BLOCK_GROUP_ITEM_KEY) {
1575 last_byte = key.objectid;
1583 last_byte = key.objectid + key.offset;
1585 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1586 last_byte >= block_group->key.objectid +
1587 block_group->key.offset) {
1588 btrfs_release_path(root, path);
1589 search_start = block_group->key.objectid +
1590 block_group->key.offset;
1598 /* we have to make sure we didn't find an extent that has already
1599 * been allocated by the map tree or the original allocation
1601 btrfs_release_path(root, path);
1602 BUG_ON(ins->objectid < search_start);
1604 if (ins->objectid + num_bytes >= search_end)
1606 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1607 ins->objectid + num_bytes > block_group->
1608 key.objectid + block_group->key.offset) {
1609 search_start = block_group->key.objectid +
1610 block_group->key.offset;
1613 if (test_range_bit(&info->extent_ins, ins->objectid,
1614 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1615 search_start = ins->objectid + num_bytes;
1618 if (test_range_bit(&info->pinned_extents, ins->objectid,
1619 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1620 search_start = ins->objectid + num_bytes;
1623 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1624 ins->objectid < exclude_start + exclude_nr)) {
1625 search_start = exclude_start + exclude_nr;
1629 block_group = btrfs_lookup_block_group(info, ins->objectid);
1631 trans->block_group = block_group;
1633 ins->offset = num_bytes;
1634 btrfs_free_path(path);
1638 if (search_start + num_bytes >= search_end) {
1640 search_start = orig_search_start;
1647 total_needed -= empty_size;
1649 data = BTRFS_BLOCK_GROUP_MIXED;
1653 block_group = btrfs_lookup_block_group(info, search_start);
1655 block_group = btrfs_find_block_group(root, block_group,
1656 search_start, data, 0);
1660 btrfs_release_path(root, path);
1661 btrfs_free_path(path);
1665 * finds a free extent and does all the dirty work required for allocation
1666 * returns the key for the extent through ins, and a tree buffer for
1667 * the first block of the extent through buf.
1669 * returns 0 if everything worked, non-zero otherwise.
1671 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1672 struct btrfs_root *root,
1673 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1674 u64 owner, u64 owner_offset,
1675 u64 empty_size, u64 hint_byte,
1676 u64 search_end, struct btrfs_key *ins, int data)
1682 u64 search_start = 0;
1685 struct btrfs_fs_info *info = root->fs_info;
1686 struct btrfs_root *extent_root = info->extent_root;
1687 struct btrfs_extent_item *extent_item;
1688 struct btrfs_extent_ref *ref;
1689 struct btrfs_path *path;
1690 struct btrfs_key keys[2];
1692 new_hint = max(hint_byte, root->fs_info->alloc_start);
1693 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1694 hint_byte = new_hint;
1696 WARN_ON(num_bytes < root->sectorsize);
1697 ret = find_free_extent(trans, root, num_bytes, empty_size,
1698 search_start, search_end, hint_byte, ins,
1699 trans->alloc_exclude_start,
1700 trans->alloc_exclude_nr, data);
1705 /* block accounting for super block */
1706 super_used = btrfs_super_bytes_used(&info->super_copy);
1707 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1709 /* block accounting for root item */
1710 root_used = btrfs_root_used(&root->root_item);
1711 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1713 clear_extent_dirty(&root->fs_info->free_space_cache,
1714 ins->objectid, ins->objectid + ins->offset - 1,
1717 if (root == extent_root) {
1718 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1719 ins->objectid + ins->offset - 1,
1720 EXTENT_LOCKED, GFP_NOFS);
1725 WARN_ON(trans->alloc_exclude_nr);
1726 trans->alloc_exclude_start = ins->objectid;
1727 trans->alloc_exclude_nr = ins->offset;
1729 memcpy(&keys[0], ins, sizeof(*ins));
1730 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1731 owner, owner_offset);
1732 keys[1].objectid = ins->objectid;
1733 keys[1].type = BTRFS_EXTENT_REF_KEY;
1734 sizes[0] = sizeof(*extent_item);
1735 sizes[1] = sizeof(*ref);
1737 path = btrfs_alloc_path();
1740 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1744 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1745 struct btrfs_extent_item);
1746 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1747 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1748 struct btrfs_extent_ref);
1750 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1751 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1752 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1753 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1755 btrfs_mark_buffer_dirty(path->nodes[0]);
1757 trans->alloc_exclude_start = 0;
1758 trans->alloc_exclude_nr = 0;
1759 btrfs_free_path(path);
1760 finish_current_insert(trans, extent_root);
1761 pending_ret = del_pending_extents(trans, extent_root);
1771 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1774 printk("update block group failed for %Lu %Lu\n",
1775 ins->objectid, ins->offset);
1782 * helper function to allocate a block for a given tree
1783 * returns the tree buffer or NULL.
1785 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1786 struct btrfs_root *root,
1788 u64 root_objectid, u64 hint,
1794 ref_generation = trans->transid;
1799 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1800 ref_generation, 0, 0, hint, empty_size);
1804 * helper function to allocate a block for a given tree
1805 * returns the tree buffer or NULL.
1807 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *root,
1817 struct btrfs_key ins;
1819 struct extent_buffer *buf;
1821 ret = btrfs_alloc_extent(trans, root, blocksize,
1822 root_objectid, ref_generation,
1823 level, first_objectid, empty_size, hint,
1827 return ERR_PTR(ret);
1829 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1831 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1832 root->root_key.objectid, ref_generation,
1834 return ERR_PTR(-ENOMEM);
1836 btrfs_set_header_generation(buf, trans->transid);
1837 clean_tree_block(trans, root, buf);
1838 wait_on_tree_block_writeback(root, buf);
1839 btrfs_set_buffer_uptodate(buf);
1841 if (PageDirty(buf->first_page)) {
1842 printk("page %lu dirty\n", buf->first_page->index);
1846 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1847 buf->start + buf->len - 1, GFP_NOFS);
1848 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1849 buf->start, buf->start + buf->len - 1,
1850 EXTENT_CSUM, GFP_NOFS);
1851 buf->flags |= EXTENT_CSUM;
1852 if (!btrfs_test_opt(root, SSD))
1853 btrfs_set_buffer_defrag(buf);
1854 trans->blocks_used++;
1858 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1859 struct btrfs_root *root,
1860 struct extent_buffer *leaf)
1863 u64 leaf_generation;
1864 struct btrfs_key key;
1865 struct btrfs_file_extent_item *fi;
1870 BUG_ON(!btrfs_is_leaf(leaf));
1871 nritems = btrfs_header_nritems(leaf);
1872 leaf_owner = btrfs_header_owner(leaf);
1873 leaf_generation = btrfs_header_generation(leaf);
1875 for (i = 0; i < nritems; i++) {
1878 btrfs_item_key_to_cpu(leaf, &key, i);
1879 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1881 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1882 if (btrfs_file_extent_type(leaf, fi) ==
1883 BTRFS_FILE_EXTENT_INLINE)
1886 * FIXME make sure to insert a trans record that
1887 * repeats the snapshot del on crash
1889 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1890 if (disk_bytenr == 0)
1892 ret = btrfs_free_extent(trans, root, disk_bytenr,
1893 btrfs_file_extent_disk_num_bytes(leaf, fi),
1894 leaf_owner, leaf_generation,
1895 key.objectid, key.offset, 0);
1901 static void noinline reada_walk_down(struct btrfs_root *root,
1902 struct extent_buffer *node,
1915 nritems = btrfs_header_nritems(node);
1916 level = btrfs_header_level(node);
1920 for (i = slot; i < nritems && skipped < 32; i++) {
1921 bytenr = btrfs_node_blockptr(node, i);
1922 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1923 (last > bytenr && last - bytenr > 32 * 1024))) {
1927 blocksize = btrfs_level_size(root, level - 1);
1929 ret = lookup_extent_ref(NULL, root, bytenr,
1937 mutex_unlock(&root->fs_info->fs_mutex);
1938 ret = readahead_tree_block(root, bytenr, blocksize);
1939 last = bytenr + blocksize;
1941 mutex_lock(&root->fs_info->fs_mutex);
1948 * helper function for drop_snapshot, this walks down the tree dropping ref
1949 * counts as it goes.
1951 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root,
1953 struct btrfs_path *path, int *level)
1958 struct extent_buffer *next;
1959 struct extent_buffer *cur;
1960 struct extent_buffer *parent;
1965 WARN_ON(*level < 0);
1966 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1967 ret = lookup_extent_ref(trans, root,
1968 path->nodes[*level]->start,
1969 path->nodes[*level]->len, &refs);
1975 * walk down to the last node level and free all the leaves
1977 while(*level >= 0) {
1978 WARN_ON(*level < 0);
1979 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1980 cur = path->nodes[*level];
1982 if (btrfs_header_level(cur) != *level)
1985 if (path->slots[*level] >=
1986 btrfs_header_nritems(cur))
1989 ret = drop_leaf_ref(trans, root, cur);
1993 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1994 blocksize = btrfs_level_size(root, *level - 1);
1995 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1998 parent = path->nodes[*level];
1999 root_owner = btrfs_header_owner(parent);
2000 root_gen = btrfs_header_generation(parent);
2001 path->slots[*level]++;
2002 ret = btrfs_free_extent(trans, root, bytenr,
2003 blocksize, root_owner,
2008 next = btrfs_find_tree_block(root, bytenr, blocksize);
2009 if (!next || !btrfs_buffer_uptodate(next)) {
2010 free_extent_buffer(next);
2011 reada_walk_down(root, cur, path->slots[*level]);
2012 mutex_unlock(&root->fs_info->fs_mutex);
2013 next = read_tree_block(root, bytenr, blocksize);
2014 mutex_lock(&root->fs_info->fs_mutex);
2016 /* we dropped the lock, check one more time */
2017 ret = lookup_extent_ref(trans, root, bytenr,
2021 parent = path->nodes[*level];
2022 root_owner = btrfs_header_owner(parent);
2023 root_gen = btrfs_header_generation(parent);
2025 path->slots[*level]++;
2026 free_extent_buffer(next);
2027 ret = btrfs_free_extent(trans, root, bytenr,
2035 WARN_ON(*level <= 0);
2036 if (path->nodes[*level-1])
2037 free_extent_buffer(path->nodes[*level-1]);
2038 path->nodes[*level-1] = next;
2039 *level = btrfs_header_level(next);
2040 path->slots[*level] = 0;
2043 WARN_ON(*level < 0);
2044 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2046 if (path->nodes[*level] == root->node) {
2047 root_owner = root->root_key.objectid;
2048 parent = path->nodes[*level];
2050 parent = path->nodes[*level + 1];
2051 root_owner = btrfs_header_owner(parent);
2054 root_gen = btrfs_header_generation(parent);
2055 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2056 path->nodes[*level]->len,
2057 root_owner, root_gen, 0, 0, 1);
2058 free_extent_buffer(path->nodes[*level]);
2059 path->nodes[*level] = NULL;
2066 * helper for dropping snapshots. This walks back up the tree in the path
2067 * to find the first node higher up where we haven't yet gone through
2070 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2071 struct btrfs_root *root,
2072 struct btrfs_path *path, int *level)
2076 struct btrfs_root_item *root_item = &root->root_item;
2081 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2082 slot = path->slots[i];
2083 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2084 struct extent_buffer *node;
2085 struct btrfs_disk_key disk_key;
2086 node = path->nodes[i];
2089 WARN_ON(*level == 0);
2090 btrfs_node_key(node, &disk_key, path->slots[i]);
2091 memcpy(&root_item->drop_progress,
2092 &disk_key, sizeof(disk_key));
2093 root_item->drop_level = i;
2096 if (path->nodes[*level] == root->node) {
2097 root_owner = root->root_key.objectid;
2099 btrfs_header_generation(path->nodes[*level]);
2101 struct extent_buffer *node;
2102 node = path->nodes[*level + 1];
2103 root_owner = btrfs_header_owner(node);
2104 root_gen = btrfs_header_generation(node);
2106 ret = btrfs_free_extent(trans, root,
2107 path->nodes[*level]->start,
2108 path->nodes[*level]->len,
2109 root_owner, root_gen, 0, 0, 1);
2111 free_extent_buffer(path->nodes[*level]);
2112 path->nodes[*level] = NULL;
2120 * drop the reference count on the tree rooted at 'snap'. This traverses
2121 * the tree freeing any blocks that have a ref count of zero after being
2124 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2130 struct btrfs_path *path;
2133 struct btrfs_root_item *root_item = &root->root_item;
2135 path = btrfs_alloc_path();
2138 level = btrfs_header_level(root->node);
2140 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2141 path->nodes[level] = root->node;
2142 extent_buffer_get(root->node);
2143 path->slots[level] = 0;
2145 struct btrfs_key key;
2146 struct btrfs_disk_key found_key;
2147 struct extent_buffer *node;
2149 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2150 level = root_item->drop_level;
2151 path->lowest_level = level;
2152 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2157 node = path->nodes[level];
2158 btrfs_node_key(node, &found_key, path->slots[level]);
2159 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2160 sizeof(found_key)));
2163 wret = walk_down_tree(trans, root, path, &level);
2169 wret = walk_up_tree(trans, root, path, &level);
2177 for (i = 0; i <= orig_level; i++) {
2178 if (path->nodes[i]) {
2179 free_extent_buffer(path->nodes[i]);
2180 path->nodes[i] = NULL;
2184 btrfs_free_path(path);
2188 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2195 ret = find_first_extent_bit(&info->block_group_cache, 0,
2196 &start, &end, (unsigned int)-1);
2199 ret = get_state_private(&info->block_group_cache, start, &ptr);
2201 kfree((void *)(unsigned long)ptr);
2202 clear_extent_bits(&info->block_group_cache, start,
2203 end, (unsigned int)-1, GFP_NOFS);
2206 ret = find_first_extent_bit(&info->free_space_cache, 0,
2207 &start, &end, EXTENT_DIRTY);
2210 clear_extent_dirty(&info->free_space_cache, start,
2216 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2222 u64 existing_delalloc;
2223 unsigned long last_index;
2226 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2227 struct file_ra_state *ra;
2229 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2231 mutex_lock(&inode->i_mutex);
2232 i = start >> PAGE_CACHE_SHIFT;
2233 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2235 file_ra_state_init(ra, inode->i_mapping);
2236 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2239 for (; i <= last_index; i++) {
2240 page = grab_cache_page(inode->i_mapping, i);
2243 if (!PageUptodate(page)) {
2244 btrfs_readpage(NULL, page);
2246 if (!PageUptodate(page)) {
2248 page_cache_release(page);
2252 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2253 page_end = page_start + PAGE_CACHE_SIZE - 1;
2255 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2257 delalloc_start = page_start;
2258 existing_delalloc = count_range_bits(io_tree,
2259 &delalloc_start, page_end,
2260 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2262 set_extent_delalloc(io_tree, page_start,
2263 page_end, GFP_NOFS);
2265 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2266 set_page_dirty(page);
2268 page_cache_release(page);
2272 mutex_unlock(&inode->i_mutex);
2277 * note, this releases the path
2279 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2280 struct btrfs_path *path,
2281 struct btrfs_key *extent_key)
2283 struct inode *inode;
2284 struct btrfs_root *found_root;
2285 struct btrfs_key *root_location;
2286 struct btrfs_extent_ref *ref;
2293 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2294 struct btrfs_extent_ref);
2295 ref_root = btrfs_ref_root(path->nodes[0], ref);
2296 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2297 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2298 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2299 btrfs_release_path(extent_root, path);
2301 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2302 root_location->objectid = ref_root;
2304 root_location->offset = 0;
2306 root_location->offset = (u64)-1;
2307 root_location->type = BTRFS_ROOT_ITEM_KEY;
2309 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2311 BUG_ON(!found_root);
2312 kfree(root_location);
2314 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2315 mutex_unlock(&extent_root->fs_info->fs_mutex);
2316 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2317 ref_objectid, found_root);
2318 if (inode->i_state & I_NEW) {
2319 /* the inode and parent dir are two different roots */
2320 BTRFS_I(inode)->root = found_root;
2321 BTRFS_I(inode)->location.objectid = ref_objectid;
2322 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2323 BTRFS_I(inode)->location.offset = 0;
2324 btrfs_read_locked_inode(inode);
2325 unlock_new_inode(inode);
2328 /* this can happen if the reference is not against
2329 * the latest version of the tree root
2331 if (is_bad_inode(inode)) {
2332 mutex_lock(&extent_root->fs_info->fs_mutex);
2335 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2336 /* FIXME, data=ordered will help get rid of this */
2337 filemap_fdatawrite(inode->i_mapping);
2339 mutex_lock(&extent_root->fs_info->fs_mutex);
2341 struct btrfs_trans_handle *trans;
2342 struct btrfs_key found_key;
2343 struct extent_buffer *eb;
2347 trans = btrfs_start_transaction(found_root, 1);
2348 eb = read_tree_block(found_root, extent_key->objectid,
2349 extent_key->offset);
2350 level = btrfs_header_level(eb);
2353 btrfs_item_key_to_cpu(eb, &found_key, 0);
2355 btrfs_node_key_to_cpu(eb, &found_key, 0);
2357 free_extent_buffer(eb);
2359 path->lowest_level = level;
2361 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2363 path->lowest_level = 0;
2364 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2365 if (!path->nodes[i])
2367 free_extent_buffer(path->nodes[i]);
2368 path->nodes[i] = NULL;
2370 btrfs_release_path(found_root, path);
2371 btrfs_end_transaction(trans, found_root);
2378 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2379 struct btrfs_path *path,
2380 struct btrfs_key *extent_key)
2382 struct btrfs_key key;
2383 struct btrfs_key found_key;
2384 struct extent_buffer *leaf;
2389 key.objectid = extent_key->objectid;
2390 key.type = BTRFS_EXTENT_REF_KEY;
2394 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2400 leaf = path->nodes[0];
2401 nritems = btrfs_header_nritems(leaf);
2402 if (path->slots[0] == nritems)
2405 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2406 if (found_key.objectid != extent_key->objectid)
2409 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2412 key.offset = found_key.offset + 1;
2413 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2415 ret = relocate_one_reference(extent_root, path, extent_key);
2421 btrfs_release_path(extent_root, path);
2425 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2427 struct btrfs_trans_handle *trans;
2428 struct btrfs_root *tree_root = root->fs_info->tree_root;
2429 struct btrfs_path *path;
2432 struct btrfs_fs_info *info = root->fs_info;
2433 struct extent_io_tree *block_group_cache;
2434 struct btrfs_key key;
2435 struct btrfs_key found_key;
2436 struct extent_buffer *leaf;
2441 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2442 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2444 block_group_cache = &info->block_group_cache;
2445 path = btrfs_alloc_path();
2446 root = root->fs_info->extent_root;
2451 key.objectid = new_size;
2454 cur_byte = key.objectid;
2456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2460 ret = find_previous_extent(root, path);
2464 leaf = path->nodes[0];
2465 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2466 if (found_key.objectid + found_key.offset > new_size) {
2467 cur_byte = found_key.objectid;
2468 key.objectid = cur_byte;
2471 btrfs_release_path(root, path);
2474 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2478 leaf = path->nodes[0];
2479 nritems = btrfs_header_nritems(leaf);
2481 if (path->slots[0] >= nritems) {
2482 ret = btrfs_next_leaf(root, path);
2489 leaf = path->nodes[0];
2490 nritems = btrfs_header_nritems(leaf);
2493 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2495 if (progress && need_resched()) {
2496 memcpy(&key, &found_key, sizeof(key));
2497 mutex_unlock(&root->fs_info->fs_mutex);
2499 mutex_lock(&root->fs_info->fs_mutex);
2500 btrfs_release_path(root, path);
2501 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2507 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2508 found_key.objectid + found_key.offset <= cur_byte) {
2514 cur_byte = found_key.objectid + found_key.offset;
2515 key.objectid = cur_byte;
2516 btrfs_release_path(root, path);
2517 ret = relocate_one_extent(root, path, &found_key);
2520 btrfs_release_path(root, path);
2522 if (total_found > 0) {
2523 trans = btrfs_start_transaction(tree_root, 1);
2524 btrfs_commit_transaction(trans, tree_root);
2526 mutex_unlock(&root->fs_info->fs_mutex);
2527 btrfs_clean_old_snapshots(tree_root);
2528 mutex_lock(&root->fs_info->fs_mutex);
2530 trans = btrfs_start_transaction(tree_root, 1);
2531 btrfs_commit_transaction(trans, tree_root);
2535 trans = btrfs_start_transaction(root, 1);
2536 key.objectid = new_size;
2542 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2546 leaf = path->nodes[0];
2547 nritems = btrfs_header_nritems(leaf);
2549 if (path->slots[0] >= nritems) {
2550 ret = btrfs_next_leaf(root, path);
2557 leaf = path->nodes[0];
2558 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2561 * btrfs_next_leaf doesn't cow buffers, we have to
2562 * do the search again
2564 memcpy(&key, &found_key, sizeof(key));
2565 btrfs_release_path(root, path);
2569 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2570 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2571 printk("shrinker found key %Lu %u %Lu\n",
2572 found_key.objectid, found_key.type,
2577 ret = get_state_private(&info->block_group_cache,
2578 found_key.objectid, &ptr);
2580 kfree((void *)(unsigned long)ptr);
2582 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2583 found_key.objectid + found_key.offset - 1,
2584 (unsigned int)-1, GFP_NOFS);
2586 key.objectid = found_key.objectid + 1;
2587 btrfs_del_item(trans, root, path);
2588 btrfs_release_path(root, path);
2590 if (need_resched()) {
2591 mutex_unlock(&root->fs_info->fs_mutex);
2593 mutex_lock(&root->fs_info->fs_mutex);
2596 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2598 btrfs_commit_transaction(trans, root);
2600 btrfs_free_path(path);
2604 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root, u64 new_size)
2607 struct btrfs_path *path;
2612 struct btrfs_block_group_cache *cache;
2613 struct btrfs_block_group_item *item;
2614 struct btrfs_fs_info *info = root->fs_info;
2615 struct extent_io_tree *block_group_cache;
2616 struct btrfs_key key;
2617 struct extent_buffer *leaf;
2621 old_size = btrfs_super_total_bytes(&info->super_copy);
2622 block_group_cache = &info->block_group_cache;
2624 root = info->extent_root;
2626 cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2628 cur_byte = cache->key.objectid + cache->key.offset;
2629 if (cur_byte >= new_size)
2632 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2633 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2635 path = btrfs_alloc_path();
2639 while(cur_byte < new_size) {
2640 key.objectid = cur_byte;
2641 ret = btrfs_insert_empty_item(trans, root, path, &key,
2642 sizeof(struct btrfs_block_group_item));
2644 leaf = path->nodes[0];
2645 item = btrfs_item_ptr(leaf, path->slots[0],
2646 struct btrfs_block_group_item);
2648 btrfs_set_disk_block_group_used(leaf, item, 0);
2649 div_long_long_rem(nr, 3, &rem);
2651 btrfs_set_disk_block_group_flags(leaf, item,
2652 BTRFS_BLOCK_GROUP_DATA);
2654 btrfs_set_disk_block_group_flags(leaf, item, 0);
2658 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2661 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2662 sizeof(cache->item));
2664 memcpy(&cache->key, &key, sizeof(key));
2667 cur_byte = key.objectid + key.offset;
2668 btrfs_release_path(root, path);
2670 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2671 bit = BLOCK_GROUP_DATA;
2672 cache->data = BTRFS_BLOCK_GROUP_DATA;
2674 bit = BLOCK_GROUP_METADATA;
2678 /* use EXTENT_LOCKED to prevent merging */
2679 set_extent_bits(block_group_cache, key.objectid,
2680 key.objectid + key.offset - 1,
2681 bit | EXTENT_LOCKED, GFP_NOFS);
2682 set_state_private(block_group_cache, key.objectid,
2683 (unsigned long)cache);
2685 btrfs_free_path(path);
2687 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2691 int btrfs_read_block_groups(struct btrfs_root *root)
2693 struct btrfs_path *path;
2697 struct btrfs_block_group_cache *cache;
2698 struct btrfs_fs_info *info = root->fs_info;
2699 struct extent_io_tree *block_group_cache;
2700 struct btrfs_key key;
2701 struct btrfs_key found_key;
2702 struct extent_buffer *leaf;
2704 block_group_cache = &info->block_group_cache;
2706 root = info->extent_root;
2708 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2709 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2711 path = btrfs_alloc_path();
2716 ret = btrfs_search_slot(NULL, info->extent_root,
2722 leaf = path->nodes[0];
2723 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2724 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2730 read_extent_buffer(leaf, &cache->item,
2731 btrfs_item_ptr_offset(leaf, path->slots[0]),
2732 sizeof(cache->item));
2733 memcpy(&cache->key, &found_key, sizeof(found_key));
2736 key.objectid = found_key.objectid + found_key.offset;
2737 btrfs_release_path(root, path);
2739 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2740 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2741 cache->data = BTRFS_BLOCK_GROUP_MIXED;
2742 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2743 bit = BLOCK_GROUP_DATA;
2744 cache->data = BTRFS_BLOCK_GROUP_DATA;
2746 bit = BLOCK_GROUP_METADATA;
2750 /* use EXTENT_LOCKED to prevent merging */
2751 set_extent_bits(block_group_cache, found_key.objectid,
2752 found_key.objectid + found_key.offset - 1,
2753 bit | EXTENT_LOCKED, GFP_NOFS);
2754 set_state_private(block_group_cache, found_key.objectid,
2755 (unsigned long)cache);
2758 btrfs_super_total_bytes(&info->super_copy))
2762 btrfs_free_path(path);