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;
201 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
203 ret = cache_block_group(root, cache);
207 last = max(search_start, cache->key.objectid);
210 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
211 last, &start, &end, EXTENT_DIRTY);
218 start = max(last, start);
220 if (last - start < num) {
221 if (last == cache->key.objectid + cache->key.offset)
225 if (data != BTRFS_BLOCK_GROUP_MIXED &&
226 start + num > cache->key.objectid + cache->key.offset)
228 if (start + num > total_fs_bytes)
233 cache = btrfs_lookup_block_group(root->fs_info, search_start);
235 printk("Unable to find block group for %Lu\n",
243 last = cache->key.objectid + cache->key.offset;
245 cache = btrfs_lookup_block_group(root->fs_info, last);
246 if (!cache || cache->key.objectid >= total_fs_bytes) {
251 data = BTRFS_BLOCK_GROUP_MIXED;
256 if (cache_miss && !cache->cached) {
257 cache_block_group(root, cache);
259 cache = btrfs_lookup_block_group(root->fs_info, last);
261 cache = btrfs_find_block_group(root, cache, last, data, 0);
269 static u64 div_factor(u64 num, int factor)
278 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
279 struct btrfs_block_group_cache
280 *hint, u64 search_start,
283 struct btrfs_block_group_cache *cache;
284 struct extent_io_tree *block_group_cache;
285 struct btrfs_block_group_cache *found_group = NULL;
286 struct btrfs_fs_info *info = root->fs_info;
301 block_group_cache = &info->block_group_cache;
302 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
307 if (data == BTRFS_BLOCK_GROUP_MIXED) {
308 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
311 bit = BLOCK_GROUP_DATA;
313 bit = BLOCK_GROUP_METADATA;
315 if (search_start && search_start < total_fs_bytes) {
316 struct btrfs_block_group_cache *shint;
317 shint = btrfs_lookup_block_group(info, search_start);
318 if (shint && (shint->data == data ||
319 shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
320 used = btrfs_block_group_used(&shint->item);
321 if (used + shint->pinned <
322 div_factor(shint->key.offset, factor)) {
327 if (hint && hint->key.objectid < total_fs_bytes &&
328 (hint->data == data || hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
329 used = btrfs_block_group_used(&hint->item);
330 if (used + hint->pinned <
331 div_factor(hint->key.offset, factor)) {
334 last = hint->key.objectid + hint->key.offset;
338 hint_last = max(hint->key.objectid, search_start);
340 hint_last = search_start;
342 if (hint_last >= total_fs_bytes)
343 hint_last = search_start;
348 ret = find_first_extent_bit(block_group_cache, last,
353 ret = get_state_private(block_group_cache, start, &ptr);
357 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
358 last = cache->key.objectid + cache->key.offset;
359 used = btrfs_block_group_used(&cache->item);
361 if (cache->key.objectid > total_fs_bytes)
365 free_check = cache->key.offset;
367 free_check = div_factor(cache->key.offset, factor);
368 if (used + cache->pinned < free_check) {
381 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
389 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
390 u64 owner, u64 owner_offset)
392 u32 high_crc = ~(u32)0;
393 u32 low_crc = ~(u32)0;
396 lenum = cpu_to_le64(root_objectid);
397 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
398 lenum = cpu_to_le64(ref_generation);
399 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
400 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
401 lenum = cpu_to_le64(owner);
402 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
403 lenum = cpu_to_le64(owner_offset);
404 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
406 return ((u64)high_crc << 32) | (u64)low_crc;
409 static int match_extent_ref(struct extent_buffer *leaf,
410 struct btrfs_extent_ref *disk_ref,
411 struct btrfs_extent_ref *cpu_ref)
416 if (cpu_ref->objectid)
417 len = sizeof(*cpu_ref);
419 len = 2 * sizeof(u64);
420 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
425 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
426 struct btrfs_root *root,
427 struct btrfs_path *path, u64 bytenr,
429 u64 ref_generation, u64 owner,
430 u64 owner_offset, int del)
433 struct btrfs_key key;
434 struct btrfs_key found_key;
435 struct btrfs_extent_ref ref;
436 struct extent_buffer *leaf;
437 struct btrfs_extent_ref *disk_ref;
441 btrfs_set_stack_ref_root(&ref, root_objectid);
442 btrfs_set_stack_ref_generation(&ref, ref_generation);
443 btrfs_set_stack_ref_objectid(&ref, owner);
444 btrfs_set_stack_ref_offset(&ref, owner_offset);
446 hash = hash_extent_ref(root_objectid, ref_generation, owner,
449 key.objectid = bytenr;
450 key.type = BTRFS_EXTENT_REF_KEY;
453 ret = btrfs_search_slot(trans, root, &key, path,
457 leaf = path->nodes[0];
459 u32 nritems = btrfs_header_nritems(leaf);
460 if (path->slots[0] >= nritems) {
461 ret2 = btrfs_next_leaf(root, path);
464 leaf = path->nodes[0];
466 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
467 if (found_key.objectid != bytenr ||
468 found_key.type != BTRFS_EXTENT_REF_KEY)
470 key.offset = found_key.offset;
472 btrfs_release_path(root, path);
476 disk_ref = btrfs_item_ptr(path->nodes[0],
478 struct btrfs_extent_ref);
479 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
483 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
484 key.offset = found_key.offset + 1;
485 btrfs_release_path(root, path);
492 * Back reference rules. Back refs have three main goals:
494 * 1) differentiate between all holders of references to an extent so that
495 * when a reference is dropped we can make sure it was a valid reference
496 * before freeing the extent.
498 * 2) Provide enough information to quickly find the holders of an extent
499 * if we notice a given block is corrupted or bad.
501 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
502 * maintenance. This is actually the same as #2, but with a slightly
503 * different use case.
505 * File extents can be referenced by:
507 * - multiple snapshots, subvolumes, or different generations in one subvol
508 * - different files inside a single subvolume (in theory, not implemented yet)
509 * - different offsets inside a file (bookend extents in file.c)
511 * The extent ref structure has fields for:
513 * - Objectid of the subvolume root
514 * - Generation number of the tree holding the reference
515 * - objectid of the file holding the reference
516 * - offset in the file corresponding to the key holding the reference
518 * When a file extent is allocated the fields are filled in:
519 * (root_key.objectid, trans->transid, inode objectid, offset in file)
521 * When a leaf is cow'd new references are added for every file extent found
522 * in the leaf. It looks the same as the create case, but trans->transid
523 * will be different when the block is cow'd.
525 * (root_key.objectid, trans->transid, inode objectid, offset in file)
527 * When a file extent is removed either during snapshot deletion or file
528 * truncation, the corresponding back reference is found
531 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
532 * inode objectid, offset in file)
534 * Btree extents can be referenced by:
536 * - Different subvolumes
537 * - Different generations of the same subvolume
539 * Storing sufficient information for a full reverse mapping of a btree
540 * block would require storing the lowest key of the block in the backref,
541 * and it would require updating that lowest key either before write out or
542 * every time it changed. Instead, the objectid of the lowest key is stored
543 * along with the level of the tree block. This provides a hint
544 * about where in the btree the block can be found. Searches through the
545 * btree only need to look for a pointer to that block, so they stop one
546 * level higher than the level recorded in the backref.
548 * Some btrees do not do reference counting on their extents. These
549 * include the extent tree and the tree of tree roots. Backrefs for these
550 * trees always have a generation of zero.
552 * When a tree block is created, back references are inserted:
554 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
556 * When a tree block is cow'd in a reference counted root,
557 * new back references are added for all the blocks it points to.
558 * These are of the form (trans->transid will have increased since creation):
560 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
562 * Because the lowest_key_objectid and the level are just hints
563 * they are not used when backrefs are deleted. When a backref is deleted:
565 * if backref was for a tree root:
566 * root_objectid = root->root_key.objectid
568 * root_objectid = btrfs_header_owner(parent)
570 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
572 * Back Reference Key hashing:
574 * Back references have four fields, each 64 bits long. Unfortunately,
575 * This is hashed into a single 64 bit number and placed into the key offset.
576 * The key objectid corresponds to the first byte in the extent, and the
577 * key type is set to BTRFS_EXTENT_REF_KEY
579 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
580 struct btrfs_root *root,
581 struct btrfs_path *path, u64 bytenr,
582 u64 root_objectid, u64 ref_generation,
583 u64 owner, u64 owner_offset)
586 struct btrfs_key key;
587 struct btrfs_extent_ref ref;
588 struct btrfs_extent_ref *disk_ref;
591 btrfs_set_stack_ref_root(&ref, root_objectid);
592 btrfs_set_stack_ref_generation(&ref, ref_generation);
593 btrfs_set_stack_ref_objectid(&ref, owner);
594 btrfs_set_stack_ref_offset(&ref, owner_offset);
596 hash = hash_extent_ref(root_objectid, ref_generation, owner,
599 key.objectid = bytenr;
600 key.type = BTRFS_EXTENT_REF_KEY;
602 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
603 while (ret == -EEXIST) {
604 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
605 struct btrfs_extent_ref);
606 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
609 btrfs_release_path(root, path);
610 ret = btrfs_insert_empty_item(trans, root, path, &key,
615 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
616 struct btrfs_extent_ref);
617 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
619 btrfs_mark_buffer_dirty(path->nodes[0]);
621 btrfs_release_path(root, path);
625 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root,
627 u64 bytenr, u64 num_bytes,
628 u64 root_objectid, u64 ref_generation,
629 u64 owner, u64 owner_offset)
631 struct btrfs_path *path;
633 struct btrfs_key key;
634 struct extent_buffer *l;
635 struct btrfs_extent_item *item;
638 WARN_ON(num_bytes < root->sectorsize);
639 path = btrfs_alloc_path();
644 key.objectid = bytenr;
645 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
646 key.offset = num_bytes;
647 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
656 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
657 refs = btrfs_extent_refs(l, item);
658 btrfs_set_extent_refs(l, item, refs + 1);
659 btrfs_mark_buffer_dirty(path->nodes[0]);
661 btrfs_release_path(root->fs_info->extent_root, path);
664 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
665 path, bytenr, root_objectid,
666 ref_generation, owner, owner_offset);
668 finish_current_insert(trans, root->fs_info->extent_root);
669 del_pending_extents(trans, root->fs_info->extent_root);
671 btrfs_free_path(path);
675 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
676 struct btrfs_root *root)
678 finish_current_insert(trans, root->fs_info->extent_root);
679 del_pending_extents(trans, root->fs_info->extent_root);
683 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
684 struct btrfs_root *root, u64 bytenr,
685 u64 num_bytes, u32 *refs)
687 struct btrfs_path *path;
689 struct btrfs_key key;
690 struct extent_buffer *l;
691 struct btrfs_extent_item *item;
693 WARN_ON(num_bytes < root->sectorsize);
694 path = btrfs_alloc_path();
696 key.objectid = bytenr;
697 key.offset = num_bytes;
698 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
699 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
704 btrfs_print_leaf(root, path->nodes[0]);
705 printk("failed to find block number %Lu\n", bytenr);
709 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
710 *refs = btrfs_extent_refs(l, item);
712 btrfs_free_path(path);
716 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
717 struct btrfs_path *count_path,
720 struct btrfs_root *extent_root = root->fs_info->extent_root;
721 struct btrfs_path *path;
724 u64 root_objectid = root->root_key.objectid;
729 struct btrfs_key key;
730 struct btrfs_key found_key;
731 struct extent_buffer *l;
732 struct btrfs_extent_item *item;
733 struct btrfs_extent_ref *ref_item;
736 path = btrfs_alloc_path();
739 bytenr = first_extent;
741 bytenr = count_path->nodes[level]->start;
744 key.objectid = bytenr;
747 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
748 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
754 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
756 if (found_key.objectid != bytenr ||
757 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
761 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
764 nritems = btrfs_header_nritems(l);
765 if (path->slots[0] >= nritems) {
766 ret = btrfs_next_leaf(extent_root, path);
771 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
772 if (found_key.objectid != bytenr)
775 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
781 ref_item = btrfs_item_ptr(l, path->slots[0],
782 struct btrfs_extent_ref);
783 found_objectid = btrfs_ref_root(l, ref_item);
785 if (found_objectid != root_objectid) {
792 if (cur_count == 0) {
796 if (level >= 0 && root->node == count_path->nodes[level])
799 btrfs_release_path(root, path);
803 btrfs_free_path(path);
806 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
807 struct btrfs_root *root, u64 owner_objectid)
813 struct btrfs_disk_key disk_key;
815 level = btrfs_header_level(root->node);
816 generation = trans->transid;
817 nritems = btrfs_header_nritems(root->node);
820 btrfs_item_key(root->node, &disk_key, 0);
822 btrfs_node_key(root->node, &disk_key, 0);
823 key_objectid = btrfs_disk_key_objectid(&disk_key);
827 return btrfs_inc_extent_ref(trans, root, root->node->start,
828 root->node->len, owner_objectid,
829 generation, level, key_objectid);
832 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
833 struct extent_buffer *buf)
837 struct btrfs_key key;
838 struct btrfs_file_extent_item *fi;
847 level = btrfs_header_level(buf);
848 nritems = btrfs_header_nritems(buf);
849 for (i = 0; i < nritems; i++) {
852 btrfs_item_key_to_cpu(buf, &key, i);
853 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
855 fi = btrfs_item_ptr(buf, i,
856 struct btrfs_file_extent_item);
857 if (btrfs_file_extent_type(buf, fi) ==
858 BTRFS_FILE_EXTENT_INLINE)
860 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
861 if (disk_bytenr == 0)
863 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
864 btrfs_file_extent_disk_num_bytes(buf, fi),
865 root->root_key.objectid, trans->transid,
866 key.objectid, key.offset);
872 bytenr = btrfs_node_blockptr(buf, i);
873 btrfs_node_key_to_cpu(buf, &key, i);
874 ret = btrfs_inc_extent_ref(trans, root, bytenr,
875 btrfs_level_size(root, level - 1),
876 root->root_key.objectid,
878 level - 1, key.objectid);
889 for (i =0; i < faili; i++) {
892 btrfs_item_key_to_cpu(buf, &key, i);
893 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
895 fi = btrfs_item_ptr(buf, i,
896 struct btrfs_file_extent_item);
897 if (btrfs_file_extent_type(buf, fi) ==
898 BTRFS_FILE_EXTENT_INLINE)
900 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
901 if (disk_bytenr == 0)
903 err = btrfs_free_extent(trans, root, disk_bytenr,
904 btrfs_file_extent_disk_num_bytes(buf,
908 bytenr = btrfs_node_blockptr(buf, i);
909 err = btrfs_free_extent(trans, root, bytenr,
910 btrfs_level_size(root, level - 1), 0);
918 static int write_one_cache_group(struct btrfs_trans_handle *trans,
919 struct btrfs_root *root,
920 struct btrfs_path *path,
921 struct btrfs_block_group_cache *cache)
925 struct btrfs_root *extent_root = root->fs_info->extent_root;
927 struct extent_buffer *leaf;
929 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
934 leaf = path->nodes[0];
935 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
936 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
937 btrfs_mark_buffer_dirty(leaf);
938 btrfs_release_path(extent_root, path);
940 finish_current_insert(trans, extent_root);
941 pending_ret = del_pending_extents(trans, extent_root);
950 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
951 struct btrfs_root *root)
953 struct extent_io_tree *block_group_cache;
954 struct btrfs_block_group_cache *cache;
958 struct btrfs_path *path;
964 block_group_cache = &root->fs_info->block_group_cache;
965 path = btrfs_alloc_path();
970 ret = find_first_extent_bit(block_group_cache, last,
971 &start, &end, BLOCK_GROUP_DIRTY);
976 ret = get_state_private(block_group_cache, start, &ptr);
980 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
981 err = write_one_cache_group(trans, root,
984 * if we fail to write the cache group, we want
985 * to keep it marked dirty in hopes that a later
992 clear_extent_bits(block_group_cache, start, end,
993 BLOCK_GROUP_DIRTY, GFP_NOFS);
995 btrfs_free_path(path);
999 static int update_block_group(struct btrfs_trans_handle *trans,
1000 struct btrfs_root *root,
1001 u64 bytenr, u64 num_bytes, int alloc,
1002 int mark_free, int data)
1004 struct btrfs_block_group_cache *cache;
1005 struct btrfs_fs_info *info = root->fs_info;
1006 u64 total = num_bytes;
1013 cache = btrfs_lookup_block_group(info, bytenr);
1017 byte_in_group = bytenr - cache->key.objectid;
1018 WARN_ON(byte_in_group > cache->key.offset);
1019 start = cache->key.objectid;
1020 end = start + cache->key.offset - 1;
1021 set_extent_bits(&info->block_group_cache, start, end,
1022 BLOCK_GROUP_DIRTY, GFP_NOFS);
1024 old_val = btrfs_block_group_used(&cache->item);
1025 num_bytes = min(total, cache->key.offset - byte_in_group);
1027 if (cache->data != data &&
1028 old_val < (cache->key.offset >> 1)) {
1033 bit_to_clear = BLOCK_GROUP_METADATA;
1034 bit_to_set = BLOCK_GROUP_DATA;
1035 cache->item.flags &=
1036 ~BTRFS_BLOCK_GROUP_MIXED;
1037 cache->item.flags |=
1038 BTRFS_BLOCK_GROUP_DATA;
1040 bit_to_clear = BLOCK_GROUP_DATA;
1041 bit_to_set = BLOCK_GROUP_METADATA;
1042 cache->item.flags &=
1043 ~BTRFS_BLOCK_GROUP_MIXED;
1044 cache->item.flags &=
1045 ~BTRFS_BLOCK_GROUP_DATA;
1047 clear_extent_bits(&info->block_group_cache,
1048 start, end, bit_to_clear,
1050 set_extent_bits(&info->block_group_cache,
1051 start, end, bit_to_set,
1053 } else if (cache->data != data &&
1054 cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1055 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1056 set_extent_bits(&info->block_group_cache,
1059 BLOCK_GROUP_METADATA,
1062 old_val += num_bytes;
1064 old_val -= num_bytes;
1066 set_extent_dirty(&info->free_space_cache,
1067 bytenr, bytenr + num_bytes - 1,
1071 btrfs_set_block_group_used(&cache->item, old_val);
1073 bytenr += num_bytes;
1077 static int update_pinned_extents(struct btrfs_root *root,
1078 u64 bytenr, u64 num, int pin)
1081 struct btrfs_block_group_cache *cache;
1082 struct btrfs_fs_info *fs_info = root->fs_info;
1085 set_extent_dirty(&fs_info->pinned_extents,
1086 bytenr, bytenr + num - 1, GFP_NOFS);
1088 clear_extent_dirty(&fs_info->pinned_extents,
1089 bytenr, bytenr + num - 1, GFP_NOFS);
1092 cache = btrfs_lookup_block_group(fs_info, bytenr);
1094 len = min(num, cache->key.offset -
1095 (bytenr - cache->key.objectid));
1097 cache->pinned += len;
1098 fs_info->total_pinned += len;
1100 cache->pinned -= len;
1101 fs_info->total_pinned -= len;
1109 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1114 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1118 ret = find_first_extent_bit(pinned_extents, last,
1119 &start, &end, EXTENT_DIRTY);
1122 set_extent_dirty(copy, start, end, GFP_NOFS);
1128 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1129 struct btrfs_root *root,
1130 struct extent_io_tree *unpin)
1135 struct extent_io_tree *free_space_cache;
1136 free_space_cache = &root->fs_info->free_space_cache;
1139 ret = find_first_extent_bit(unpin, 0, &start, &end,
1143 update_pinned_extents(root, start, end + 1 - start, 0);
1144 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1145 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1150 static int finish_current_insert(struct btrfs_trans_handle *trans,
1151 struct btrfs_root *extent_root)
1155 struct btrfs_fs_info *info = extent_root->fs_info;
1156 struct extent_buffer *eb;
1157 struct btrfs_path *path;
1158 struct btrfs_key ins;
1159 struct btrfs_disk_key first;
1160 struct btrfs_extent_item extent_item;
1165 btrfs_set_stack_extent_refs(&extent_item, 1);
1166 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1167 path = btrfs_alloc_path();
1170 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1171 &end, EXTENT_LOCKED);
1175 ins.objectid = start;
1176 ins.offset = end + 1 - start;
1177 err = btrfs_insert_item(trans, extent_root, &ins,
1178 &extent_item, sizeof(extent_item));
1179 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1181 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1182 level = btrfs_header_level(eb);
1184 btrfs_item_key(eb, &first, 0);
1186 btrfs_node_key(eb, &first, 0);
1188 err = btrfs_insert_extent_backref(trans, extent_root, path,
1189 start, extent_root->root_key.objectid,
1191 btrfs_disk_key_objectid(&first));
1193 free_extent_buffer(eb);
1195 btrfs_free_path(path);
1199 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1203 struct extent_buffer *buf;
1206 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1208 if (btrfs_buffer_uptodate(buf)) {
1210 root->fs_info->running_transaction->transid;
1211 u64 header_transid =
1212 btrfs_header_generation(buf);
1213 if (header_transid == transid) {
1214 clean_tree_block(NULL, root, buf);
1215 free_extent_buffer(buf);
1219 free_extent_buffer(buf);
1221 update_pinned_extents(root, bytenr, num_bytes, 1);
1223 set_extent_bits(&root->fs_info->pending_del,
1224 bytenr, bytenr + num_bytes - 1,
1225 EXTENT_LOCKED, GFP_NOFS);
1232 * remove an extent from the root, returns 0 on success
1234 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1235 *root, u64 bytenr, u64 num_bytes,
1236 u64 root_objectid, u64 ref_generation,
1237 u64 owner_objectid, u64 owner_offset, int pin,
1240 struct btrfs_path *path;
1241 struct btrfs_key key;
1242 struct btrfs_fs_info *info = root->fs_info;
1243 struct btrfs_root *extent_root = info->extent_root;
1244 struct extent_buffer *leaf;
1246 struct btrfs_extent_item *ei;
1249 key.objectid = bytenr;
1250 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1251 key.offset = num_bytes;
1252 path = btrfs_alloc_path();
1257 ret = lookup_extent_backref(trans, extent_root, path,
1258 bytenr, root_objectid,
1260 owner_objectid, owner_offset, 1);
1262 ret = btrfs_del_item(trans, extent_root, path);
1264 btrfs_print_leaf(extent_root, path->nodes[0]);
1266 printk("Unable to find ref byte nr %Lu root %Lu "
1267 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1268 root_objectid, ref_generation, owner_objectid,
1271 btrfs_release_path(extent_root, path);
1272 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1277 leaf = path->nodes[0];
1278 ei = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_item);
1280 refs = btrfs_extent_refs(leaf, ei);
1283 btrfs_set_extent_refs(leaf, ei, refs);
1284 btrfs_mark_buffer_dirty(leaf);
1291 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1297 /* block accounting for super block */
1298 super_used = btrfs_super_bytes_used(&info->super_copy);
1299 btrfs_set_super_bytes_used(&info->super_copy,
1300 super_used - num_bytes);
1302 /* block accounting for root item */
1303 root_used = btrfs_root_used(&root->root_item);
1304 btrfs_set_root_used(&root->root_item,
1305 root_used - num_bytes);
1307 ret = btrfs_del_item(trans, extent_root, path);
1311 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1315 btrfs_free_path(path);
1316 finish_current_insert(trans, extent_root);
1321 * find all the blocks marked as pending in the radix tree and remove
1322 * them from the extent map
1324 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1325 btrfs_root *extent_root)
1331 struct extent_io_tree *pending_del;
1332 struct extent_io_tree *pinned_extents;
1334 pending_del = &extent_root->fs_info->pending_del;
1335 pinned_extents = &extent_root->fs_info->pinned_extents;
1338 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1342 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1343 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1345 ret = __free_extent(trans, extent_root,
1346 start, end + 1 - start,
1347 extent_root->root_key.objectid,
1356 * remove an extent from the root, returns 0 on success
1358 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1359 *root, u64 bytenr, u64 num_bytes,
1360 u64 root_objectid, u64 ref_generation,
1361 u64 owner_objectid, u64 owner_offset, int pin)
1363 struct btrfs_root *extent_root = root->fs_info->extent_root;
1367 WARN_ON(num_bytes < root->sectorsize);
1368 if (!root->ref_cows)
1371 if (root == extent_root) {
1372 pin_down_bytes(root, bytenr, num_bytes, 1);
1375 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1376 ref_generation, owner_objectid, owner_offset,
1378 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1379 return ret ? ret : pending_ret;
1382 static u64 stripe_align(struct btrfs_root *root, u64 val)
1384 u64 mask = ((u64)root->stripesize - 1);
1385 u64 ret = (val + mask) & ~mask;
1390 * walks the btree of allocated extents and find a hole of a given size.
1391 * The key ins is changed to record the hole:
1392 * ins->objectid == block start
1393 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1394 * ins->offset == number of blocks
1395 * Any available blocks before search_start are skipped.
1397 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1398 struct btrfs_root *orig_root,
1399 u64 num_bytes, u64 empty_size,
1400 u64 search_start, u64 search_end,
1401 u64 hint_byte, struct btrfs_key *ins,
1402 u64 exclude_start, u64 exclude_nr,
1405 struct btrfs_path *path;
1406 struct btrfs_key key;
1412 u64 *last_ptr = NULL;
1413 u64 orig_search_start = search_start;
1415 struct extent_buffer *l;
1416 struct btrfs_root * root = orig_root->fs_info->extent_root;
1417 struct btrfs_fs_info *info = root->fs_info;
1418 u64 total_needed = num_bytes;
1420 struct btrfs_block_group_cache *block_group;
1425 WARN_ON(num_bytes < root->sectorsize);
1426 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1428 level = btrfs_header_level(root->node);
1430 if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1431 data = BTRFS_BLOCK_GROUP_MIXED;
1435 last_ptr = &root->fs_info->last_alloc;
1437 if (data && btrfs_test_opt(root, SSD))
1438 last_ptr = &root->fs_info->last_data_alloc;
1442 hint_byte = *last_ptr;
1444 hint_byte = hint_byte &
1445 ~((u64)BTRFS_BLOCK_GROUP_SIZE - 1);
1446 empty_size += 2 * 1024 * 1024;
1450 search_end = min(search_end,
1451 btrfs_super_total_bytes(&info->super_copy));
1453 block_group = btrfs_lookup_block_group(info, hint_byte);
1455 hint_byte = search_start;
1456 block_group = btrfs_find_block_group(root, block_group,
1457 hint_byte, data, 1);
1459 block_group = btrfs_find_block_group(root,
1461 search_start, data, 1);
1464 total_needed += empty_size;
1465 path = btrfs_alloc_path();
1468 block_group = btrfs_lookup_block_group(info, search_start);
1470 block_group = btrfs_lookup_block_group(info,
1473 search_start = find_search_start(root, &block_group, search_start,
1474 total_needed, data);
1476 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1479 empty_size += 16 * 1024 * 1024;
1480 total_needed += empty_size;
1482 search_start = find_search_start(root, &block_group,
1483 search_start, total_needed,
1487 search_start = stripe_align(root, search_start);
1488 cached_start = search_start;
1489 btrfs_init_path(path);
1490 ins->objectid = search_start;
1495 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1498 ret = find_previous_extent(root, path);
1502 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1505 slot = path->slots[0];
1506 if (slot >= btrfs_header_nritems(l)) {
1507 ret = btrfs_next_leaf(root, path);
1513 search_start = max(search_start,
1514 block_group->key.objectid);
1516 aligned = stripe_align(root, search_start);
1517 ins->objectid = aligned;
1518 if (aligned >= search_end) {
1522 ins->offset = search_end - aligned;
1526 ins->objectid = stripe_align(root,
1527 last_byte > search_start ?
1528 last_byte : search_start);
1529 if (search_end <= ins->objectid) {
1533 ins->offset = search_end - ins->objectid;
1534 BUG_ON(ins->objectid >= search_end);
1537 btrfs_item_key_to_cpu(l, &key, slot);
1539 if (key.objectid >= search_start && key.objectid > last_byte &&
1541 if (last_byte < search_start)
1542 last_byte = search_start;
1543 aligned = stripe_align(root, last_byte);
1544 hole_size = key.objectid - aligned;
1545 if (key.objectid > aligned && hole_size >= num_bytes) {
1546 ins->objectid = aligned;
1547 ins->offset = hole_size;
1551 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1552 if (!start_found && btrfs_key_type(&key) ==
1553 BTRFS_BLOCK_GROUP_ITEM_KEY) {
1554 last_byte = key.objectid;
1562 last_byte = key.objectid + key.offset;
1564 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1565 last_byte >= block_group->key.objectid +
1566 block_group->key.offset) {
1567 btrfs_release_path(root, path);
1568 search_start = block_group->key.objectid +
1569 block_group->key.offset;
1577 /* we have to make sure we didn't find an extent that has already
1578 * been allocated by the map tree or the original allocation
1580 btrfs_release_path(root, path);
1581 BUG_ON(ins->objectid < search_start);
1583 if (ins->objectid + num_bytes >= search_end)
1585 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1586 ins->objectid + num_bytes > block_group->
1587 key.objectid + block_group->key.offset) {
1588 search_start = block_group->key.objectid +
1589 block_group->key.offset;
1592 if (test_range_bit(&info->extent_ins, ins->objectid,
1593 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1594 search_start = ins->objectid + num_bytes;
1597 if (test_range_bit(&info->pinned_extents, ins->objectid,
1598 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1599 search_start = ins->objectid + num_bytes;
1602 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1603 ins->objectid < exclude_start + exclude_nr)) {
1604 search_start = exclude_start + exclude_nr;
1608 block_group = btrfs_lookup_block_group(info, ins->objectid);
1610 trans->block_group = block_group;
1612 ins->offset = num_bytes;
1613 btrfs_free_path(path);
1615 *last_ptr = ins->objectid + ins->offset;
1619 if (search_start + num_bytes >= search_end) {
1621 search_start = orig_search_start;
1628 total_needed -= empty_size;
1630 data = BTRFS_BLOCK_GROUP_MIXED;
1634 block_group = btrfs_lookup_block_group(info, search_start);
1636 block_group = btrfs_find_block_group(root, block_group,
1637 search_start, data, 0);
1641 btrfs_release_path(root, path);
1642 btrfs_free_path(path);
1646 * finds a free extent and does all the dirty work required for allocation
1647 * returns the key for the extent through ins, and a tree buffer for
1648 * the first block of the extent through buf.
1650 * returns 0 if everything worked, non-zero otherwise.
1652 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1653 struct btrfs_root *root,
1654 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1655 u64 owner, u64 owner_offset,
1656 u64 empty_size, u64 hint_byte,
1657 u64 search_end, struct btrfs_key *ins, int data)
1663 u64 search_start = 0;
1666 struct btrfs_fs_info *info = root->fs_info;
1667 struct btrfs_root *extent_root = info->extent_root;
1668 struct btrfs_extent_item *extent_item;
1669 struct btrfs_extent_ref *ref;
1670 struct btrfs_path *path;
1671 struct btrfs_key keys[2];
1673 new_hint = max(hint_byte, root->fs_info->alloc_start);
1674 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1675 hint_byte = new_hint;
1677 WARN_ON(num_bytes < root->sectorsize);
1678 ret = find_free_extent(trans, root, num_bytes, empty_size,
1679 search_start, search_end, hint_byte, ins,
1680 trans->alloc_exclude_start,
1681 trans->alloc_exclude_nr, data);
1686 /* block accounting for super block */
1687 super_used = btrfs_super_bytes_used(&info->super_copy);
1688 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1690 /* block accounting for root item */
1691 root_used = btrfs_root_used(&root->root_item);
1692 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1694 clear_extent_dirty(&root->fs_info->free_space_cache,
1695 ins->objectid, ins->objectid + ins->offset - 1,
1698 if (root == extent_root) {
1699 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1700 ins->objectid + ins->offset - 1,
1701 EXTENT_LOCKED, GFP_NOFS);
1706 WARN_ON(trans->alloc_exclude_nr);
1707 trans->alloc_exclude_start = ins->objectid;
1708 trans->alloc_exclude_nr = ins->offset;
1710 memcpy(&keys[0], ins, sizeof(*ins));
1711 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1712 owner, owner_offset);
1713 keys[1].objectid = ins->objectid;
1714 keys[1].type = BTRFS_EXTENT_REF_KEY;
1715 sizes[0] = sizeof(*extent_item);
1716 sizes[1] = sizeof(*ref);
1718 path = btrfs_alloc_path();
1721 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1725 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1726 struct btrfs_extent_item);
1727 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1728 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1729 struct btrfs_extent_ref);
1731 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1732 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1733 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1734 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1736 btrfs_mark_buffer_dirty(path->nodes[0]);
1738 trans->alloc_exclude_start = 0;
1739 trans->alloc_exclude_nr = 0;
1740 btrfs_free_path(path);
1741 finish_current_insert(trans, extent_root);
1742 pending_ret = del_pending_extents(trans, extent_root);
1752 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1759 * helper function to allocate a block for a given tree
1760 * returns the tree buffer or NULL.
1762 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1763 struct btrfs_root *root,
1765 u64 root_objectid, u64 hint,
1771 ref_generation = trans->transid;
1776 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1777 ref_generation, 0, 0, hint, empty_size);
1781 * helper function to allocate a block for a given tree
1782 * returns the tree buffer or NULL.
1784 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1794 struct btrfs_key ins;
1796 struct extent_buffer *buf;
1798 ret = btrfs_alloc_extent(trans, root, blocksize,
1799 root_objectid, ref_generation,
1800 level, first_objectid, empty_size, hint,
1804 return ERR_PTR(ret);
1806 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1808 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1809 root->root_key.objectid, ref_generation,
1811 return ERR_PTR(-ENOMEM);
1813 btrfs_set_header_generation(buf, trans->transid);
1814 clean_tree_block(trans, root, buf);
1815 wait_on_tree_block_writeback(root, buf);
1816 btrfs_set_buffer_uptodate(buf);
1818 if (PageDirty(buf->first_page)) {
1819 printk("page %lu dirty\n", buf->first_page->index);
1823 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1824 buf->start + buf->len - 1, GFP_NOFS);
1825 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1826 buf->start, buf->start + buf->len - 1,
1827 EXTENT_CSUM, GFP_NOFS);
1828 buf->flags |= EXTENT_CSUM;
1829 btrfs_set_buffer_defrag(buf);
1830 trans->blocks_used++;
1834 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1835 struct btrfs_root *root,
1836 struct extent_buffer *leaf)
1839 u64 leaf_generation;
1840 struct btrfs_key key;
1841 struct btrfs_file_extent_item *fi;
1846 BUG_ON(!btrfs_is_leaf(leaf));
1847 nritems = btrfs_header_nritems(leaf);
1848 leaf_owner = btrfs_header_owner(leaf);
1849 leaf_generation = btrfs_header_generation(leaf);
1851 for (i = 0; i < nritems; i++) {
1854 btrfs_item_key_to_cpu(leaf, &key, i);
1855 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1857 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1858 if (btrfs_file_extent_type(leaf, fi) ==
1859 BTRFS_FILE_EXTENT_INLINE)
1862 * FIXME make sure to insert a trans record that
1863 * repeats the snapshot del on crash
1865 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1866 if (disk_bytenr == 0)
1868 ret = btrfs_free_extent(trans, root, disk_bytenr,
1869 btrfs_file_extent_disk_num_bytes(leaf, fi),
1870 leaf_owner, leaf_generation,
1871 key.objectid, key.offset, 0);
1877 static void noinline reada_walk_down(struct btrfs_root *root,
1878 struct extent_buffer *node,
1891 nritems = btrfs_header_nritems(node);
1892 level = btrfs_header_level(node);
1896 for (i = slot; i < nritems && skipped < 32; i++) {
1897 bytenr = btrfs_node_blockptr(node, i);
1898 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1899 (last > bytenr && last - bytenr > 32 * 1024))) {
1903 blocksize = btrfs_level_size(root, level - 1);
1905 ret = lookup_extent_ref(NULL, root, bytenr,
1913 mutex_unlock(&root->fs_info->fs_mutex);
1914 ret = readahead_tree_block(root, bytenr, blocksize);
1915 last = bytenr + blocksize;
1917 mutex_lock(&root->fs_info->fs_mutex);
1924 * helper function for drop_snapshot, this walks down the tree dropping ref
1925 * counts as it goes.
1927 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *root,
1929 struct btrfs_path *path, int *level)
1934 struct extent_buffer *next;
1935 struct extent_buffer *cur;
1936 struct extent_buffer *parent;
1941 WARN_ON(*level < 0);
1942 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1943 ret = lookup_extent_ref(trans, root,
1944 path->nodes[*level]->start,
1945 path->nodes[*level]->len, &refs);
1951 * walk down to the last node level and free all the leaves
1953 while(*level >= 0) {
1954 WARN_ON(*level < 0);
1955 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1956 cur = path->nodes[*level];
1958 if (btrfs_header_level(cur) != *level)
1961 if (path->slots[*level] >=
1962 btrfs_header_nritems(cur))
1965 ret = drop_leaf_ref(trans, root, cur);
1969 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1970 blocksize = btrfs_level_size(root, *level - 1);
1971 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1974 parent = path->nodes[*level];
1975 root_owner = btrfs_header_owner(parent);
1976 root_gen = btrfs_header_generation(parent);
1977 path->slots[*level]++;
1978 ret = btrfs_free_extent(trans, root, bytenr,
1979 blocksize, root_owner,
1984 next = btrfs_find_tree_block(root, bytenr, blocksize);
1985 if (!next || !btrfs_buffer_uptodate(next)) {
1986 free_extent_buffer(next);
1987 reada_walk_down(root, cur, path->slots[*level]);
1988 mutex_unlock(&root->fs_info->fs_mutex);
1989 next = read_tree_block(root, bytenr, blocksize);
1990 mutex_lock(&root->fs_info->fs_mutex);
1992 /* we dropped the lock, check one more time */
1993 ret = lookup_extent_ref(trans, root, bytenr,
1997 parent = path->nodes[*level];
1998 root_owner = btrfs_header_owner(parent);
1999 root_gen = btrfs_header_generation(parent);
2001 path->slots[*level]++;
2002 free_extent_buffer(next);
2003 ret = btrfs_free_extent(trans, root, bytenr,
2011 WARN_ON(*level <= 0);
2012 if (path->nodes[*level-1])
2013 free_extent_buffer(path->nodes[*level-1]);
2014 path->nodes[*level-1] = next;
2015 *level = btrfs_header_level(next);
2016 path->slots[*level] = 0;
2019 WARN_ON(*level < 0);
2020 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2022 if (path->nodes[*level] == root->node) {
2023 root_owner = root->root_key.objectid;
2024 parent = path->nodes[*level];
2026 parent = path->nodes[*level + 1];
2027 root_owner = btrfs_header_owner(parent);
2030 root_gen = btrfs_header_generation(parent);
2031 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2032 path->nodes[*level]->len,
2033 root_owner, root_gen, 0, 0, 1);
2034 free_extent_buffer(path->nodes[*level]);
2035 path->nodes[*level] = NULL;
2042 * helper for dropping snapshots. This walks back up the tree in the path
2043 * to find the first node higher up where we haven't yet gone through
2046 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2047 struct btrfs_root *root,
2048 struct btrfs_path *path, int *level)
2052 struct btrfs_root_item *root_item = &root->root_item;
2057 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2058 slot = path->slots[i];
2059 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2060 struct extent_buffer *node;
2061 struct btrfs_disk_key disk_key;
2062 node = path->nodes[i];
2065 WARN_ON(*level == 0);
2066 btrfs_node_key(node, &disk_key, path->slots[i]);
2067 memcpy(&root_item->drop_progress,
2068 &disk_key, sizeof(disk_key));
2069 root_item->drop_level = i;
2072 if (path->nodes[*level] == root->node) {
2073 root_owner = root->root_key.objectid;
2075 btrfs_header_generation(path->nodes[*level]);
2077 struct extent_buffer *node;
2078 node = path->nodes[*level + 1];
2079 root_owner = btrfs_header_owner(node);
2080 root_gen = btrfs_header_generation(node);
2082 ret = btrfs_free_extent(trans, root,
2083 path->nodes[*level]->start,
2084 path->nodes[*level]->len,
2085 root_owner, root_gen, 0, 0, 1);
2087 free_extent_buffer(path->nodes[*level]);
2088 path->nodes[*level] = NULL;
2096 * drop the reference count on the tree rooted at 'snap'. This traverses
2097 * the tree freeing any blocks that have a ref count of zero after being
2100 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2106 struct btrfs_path *path;
2109 struct btrfs_root_item *root_item = &root->root_item;
2111 path = btrfs_alloc_path();
2114 level = btrfs_header_level(root->node);
2116 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2117 path->nodes[level] = root->node;
2118 extent_buffer_get(root->node);
2119 path->slots[level] = 0;
2121 struct btrfs_key key;
2122 struct btrfs_disk_key found_key;
2123 struct extent_buffer *node;
2125 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2126 level = root_item->drop_level;
2127 path->lowest_level = level;
2128 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2133 node = path->nodes[level];
2134 btrfs_node_key(node, &found_key, path->slots[level]);
2135 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2136 sizeof(found_key)));
2139 wret = walk_down_tree(trans, root, path, &level);
2145 wret = walk_up_tree(trans, root, path, &level);
2153 for (i = 0; i <= orig_level; i++) {
2154 if (path->nodes[i]) {
2155 free_extent_buffer(path->nodes[i]);
2156 path->nodes[i] = NULL;
2160 btrfs_free_path(path);
2164 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2171 ret = find_first_extent_bit(&info->block_group_cache, 0,
2172 &start, &end, (unsigned int)-1);
2175 ret = get_state_private(&info->block_group_cache, start, &ptr);
2177 kfree((void *)(unsigned long)ptr);
2178 clear_extent_bits(&info->block_group_cache, start,
2179 end, (unsigned int)-1, GFP_NOFS);
2182 ret = find_first_extent_bit(&info->free_space_cache, 0,
2183 &start, &end, EXTENT_DIRTY);
2186 clear_extent_dirty(&info->free_space_cache, start,
2192 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2198 u64 existing_delalloc;
2199 unsigned long last_index;
2202 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2203 struct file_ra_state *ra;
2205 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2207 mutex_lock(&inode->i_mutex);
2208 i = start >> PAGE_CACHE_SHIFT;
2209 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2211 file_ra_state_init(ra, inode->i_mapping);
2212 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2215 for (; i <= last_index; i++) {
2216 page = grab_cache_page(inode->i_mapping, i);
2219 if (!PageUptodate(page)) {
2220 btrfs_readpage(NULL, page);
2222 if (!PageUptodate(page)) {
2224 page_cache_release(page);
2228 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2229 page_end = page_start + PAGE_CACHE_SIZE - 1;
2231 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2233 delalloc_start = page_start;
2234 existing_delalloc = count_range_bits(io_tree,
2235 &delalloc_start, page_end,
2236 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2238 set_extent_delalloc(io_tree, page_start,
2239 page_end, GFP_NOFS);
2241 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2242 set_page_dirty(page);
2244 page_cache_release(page);
2248 mutex_unlock(&inode->i_mutex);
2253 * note, this releases the path
2255 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2256 struct btrfs_path *path,
2257 struct btrfs_key *extent_key)
2259 struct inode *inode;
2260 struct btrfs_root *found_root;
2261 struct btrfs_key *root_location;
2262 struct btrfs_extent_ref *ref;
2269 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2270 struct btrfs_extent_ref);
2271 ref_root = btrfs_ref_root(path->nodes[0], ref);
2272 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2273 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2274 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2275 btrfs_release_path(extent_root, path);
2277 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2278 root_location->objectid = ref_root;
2280 root_location->offset = 0;
2282 root_location->offset = (u64)-1;
2283 root_location->type = BTRFS_ROOT_ITEM_KEY;
2285 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2287 BUG_ON(!found_root);
2288 kfree(root_location);
2290 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2291 mutex_unlock(&extent_root->fs_info->fs_mutex);
2292 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2293 ref_objectid, found_root);
2294 if (inode->i_state & I_NEW) {
2295 /* the inode and parent dir are two different roots */
2296 BTRFS_I(inode)->root = found_root;
2297 BTRFS_I(inode)->location.objectid = ref_objectid;
2298 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2299 BTRFS_I(inode)->location.offset = 0;
2300 btrfs_read_locked_inode(inode);
2301 unlock_new_inode(inode);
2304 /* this can happen if the reference is not against
2305 * the latest version of the tree root
2307 if (is_bad_inode(inode)) {
2308 mutex_lock(&extent_root->fs_info->fs_mutex);
2311 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2312 /* FIXME, data=ordered will help get rid of this */
2313 filemap_fdatawrite(inode->i_mapping);
2315 mutex_lock(&extent_root->fs_info->fs_mutex);
2317 struct btrfs_trans_handle *trans;
2318 struct btrfs_key found_key;
2319 struct extent_buffer *eb;
2323 trans = btrfs_start_transaction(found_root, 1);
2324 eb = read_tree_block(found_root, extent_key->objectid,
2325 extent_key->offset);
2326 level = btrfs_header_level(eb);
2329 btrfs_item_key_to_cpu(eb, &found_key, 0);
2331 btrfs_node_key_to_cpu(eb, &found_key, 0);
2333 free_extent_buffer(eb);
2335 path->lowest_level = level;
2337 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2339 path->lowest_level = 0;
2340 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2341 if (!path->nodes[i])
2343 free_extent_buffer(path->nodes[i]);
2344 path->nodes[i] = NULL;
2346 btrfs_release_path(found_root, path);
2347 btrfs_end_transaction(trans, found_root);
2354 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2355 struct btrfs_path *path,
2356 struct btrfs_key *extent_key)
2358 struct btrfs_key key;
2359 struct btrfs_key found_key;
2360 struct extent_buffer *leaf;
2365 key.objectid = extent_key->objectid;
2366 key.type = BTRFS_EXTENT_REF_KEY;
2370 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2376 leaf = path->nodes[0];
2377 nritems = btrfs_header_nritems(leaf);
2378 if (path->slots[0] == nritems)
2381 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2382 if (found_key.objectid != extent_key->objectid)
2385 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2388 key.offset = found_key.offset + 1;
2389 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2391 ret = relocate_one_reference(extent_root, path, extent_key);
2397 btrfs_release_path(extent_root, path);
2401 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2403 struct btrfs_trans_handle *trans;
2404 struct btrfs_root *tree_root = root->fs_info->tree_root;
2405 struct btrfs_path *path;
2408 struct btrfs_fs_info *info = root->fs_info;
2409 struct extent_io_tree *block_group_cache;
2410 struct btrfs_key key;
2411 struct btrfs_key found_key;
2412 struct extent_buffer *leaf;
2417 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2418 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2420 block_group_cache = &info->block_group_cache;
2421 path = btrfs_alloc_path();
2422 root = root->fs_info->extent_root;
2427 key.objectid = new_size;
2430 cur_byte = key.objectid;
2432 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2436 ret = find_previous_extent(root, path);
2440 leaf = path->nodes[0];
2441 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2442 if (found_key.objectid + found_key.offset > new_size) {
2443 cur_byte = found_key.objectid;
2444 key.objectid = cur_byte;
2447 btrfs_release_path(root, path);
2450 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2454 leaf = path->nodes[0];
2455 nritems = btrfs_header_nritems(leaf);
2457 if (path->slots[0] >= nritems) {
2458 ret = btrfs_next_leaf(root, path);
2465 leaf = path->nodes[0];
2466 nritems = btrfs_header_nritems(leaf);
2469 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2471 if (progress && need_resched()) {
2472 memcpy(&key, &found_key, sizeof(key));
2473 mutex_unlock(&root->fs_info->fs_mutex);
2475 mutex_lock(&root->fs_info->fs_mutex);
2476 btrfs_release_path(root, path);
2477 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2483 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2484 found_key.objectid + found_key.offset <= cur_byte) {
2490 cur_byte = found_key.objectid + found_key.offset;
2491 key.objectid = cur_byte;
2492 btrfs_release_path(root, path);
2493 ret = relocate_one_extent(root, path, &found_key);
2496 btrfs_release_path(root, path);
2498 if (total_found > 0) {
2499 trans = btrfs_start_transaction(tree_root, 1);
2500 btrfs_commit_transaction(trans, tree_root);
2502 mutex_unlock(&root->fs_info->fs_mutex);
2503 btrfs_clean_old_snapshots(tree_root);
2504 mutex_lock(&root->fs_info->fs_mutex);
2506 trans = btrfs_start_transaction(tree_root, 1);
2507 btrfs_commit_transaction(trans, tree_root);
2511 trans = btrfs_start_transaction(root, 1);
2512 key.objectid = new_size;
2518 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2522 leaf = path->nodes[0];
2523 nritems = btrfs_header_nritems(leaf);
2525 if (path->slots[0] >= nritems) {
2526 ret = btrfs_next_leaf(root, path);
2533 leaf = path->nodes[0];
2534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2537 * btrfs_next_leaf doesn't cow buffers, we have to
2538 * do the search again
2540 memcpy(&key, &found_key, sizeof(key));
2541 btrfs_release_path(root, path);
2545 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2546 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2547 printk("shrinker found key %Lu %u %Lu\n",
2548 found_key.objectid, found_key.type,
2553 ret = get_state_private(&info->block_group_cache,
2554 found_key.objectid, &ptr);
2556 kfree((void *)(unsigned long)ptr);
2558 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2559 found_key.objectid + found_key.offset - 1,
2560 (unsigned int)-1, GFP_NOFS);
2562 key.objectid = found_key.objectid + 1;
2563 btrfs_del_item(trans, root, path);
2564 btrfs_release_path(root, path);
2566 if (need_resched()) {
2567 mutex_unlock(&root->fs_info->fs_mutex);
2569 mutex_lock(&root->fs_info->fs_mutex);
2572 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2574 btrfs_commit_transaction(trans, root);
2576 btrfs_free_path(path);
2580 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2581 struct btrfs_root *root, u64 new_size)
2583 struct btrfs_path *path;
2588 struct btrfs_block_group_cache *cache;
2589 struct btrfs_block_group_item *item;
2590 struct btrfs_fs_info *info = root->fs_info;
2591 struct extent_io_tree *block_group_cache;
2592 struct btrfs_key key;
2593 struct extent_buffer *leaf;
2597 old_size = btrfs_super_total_bytes(&info->super_copy);
2598 block_group_cache = &info->block_group_cache;
2600 root = info->extent_root;
2602 cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2604 cur_byte = cache->key.objectid + cache->key.offset;
2605 if (cur_byte >= new_size)
2608 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2609 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2611 path = btrfs_alloc_path();
2615 while(cur_byte < new_size) {
2616 key.objectid = cur_byte;
2617 ret = btrfs_insert_empty_item(trans, root, path, &key,
2618 sizeof(struct btrfs_block_group_item));
2620 leaf = path->nodes[0];
2621 item = btrfs_item_ptr(leaf, path->slots[0],
2622 struct btrfs_block_group_item);
2624 btrfs_set_disk_block_group_used(leaf, item, 0);
2625 div_long_long_rem(nr, 3, &rem);
2627 btrfs_set_disk_block_group_flags(leaf, item,
2628 BTRFS_BLOCK_GROUP_DATA);
2630 btrfs_set_disk_block_group_flags(leaf, item, 0);
2634 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2637 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2638 sizeof(cache->item));
2640 memcpy(&cache->key, &key, sizeof(key));
2643 cur_byte = key.objectid + key.offset;
2644 btrfs_release_path(root, path);
2646 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2647 bit = BLOCK_GROUP_DATA;
2648 cache->data = BTRFS_BLOCK_GROUP_DATA;
2650 bit = BLOCK_GROUP_METADATA;
2654 /* use EXTENT_LOCKED to prevent merging */
2655 set_extent_bits(block_group_cache, key.objectid,
2656 key.objectid + key.offset - 1,
2657 bit | EXTENT_LOCKED, GFP_NOFS);
2658 set_state_private(block_group_cache, key.objectid,
2659 (unsigned long)cache);
2661 btrfs_free_path(path);
2663 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2667 int btrfs_read_block_groups(struct btrfs_root *root)
2669 struct btrfs_path *path;
2673 struct btrfs_block_group_cache *cache;
2674 struct btrfs_fs_info *info = root->fs_info;
2675 struct extent_io_tree *block_group_cache;
2676 struct btrfs_key key;
2677 struct btrfs_key found_key;
2678 struct extent_buffer *leaf;
2680 block_group_cache = &info->block_group_cache;
2682 root = info->extent_root;
2684 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2685 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2687 path = btrfs_alloc_path();
2692 ret = btrfs_search_slot(NULL, info->extent_root,
2698 leaf = path->nodes[0];
2699 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2700 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2706 read_extent_buffer(leaf, &cache->item,
2707 btrfs_item_ptr_offset(leaf, path->slots[0]),
2708 sizeof(cache->item));
2709 memcpy(&cache->key, &found_key, sizeof(found_key));
2712 key.objectid = found_key.objectid + found_key.offset;
2713 btrfs_release_path(root, path);
2715 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2716 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2717 cache->data = BTRFS_BLOCK_GROUP_MIXED;
2718 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2719 bit = BLOCK_GROUP_DATA;
2720 cache->data = BTRFS_BLOCK_GROUP_DATA;
2722 bit = BLOCK_GROUP_METADATA;
2726 /* use EXTENT_LOCKED to prevent merging */
2727 set_extent_bits(block_group_cache, found_key.objectid,
2728 found_key.objectid + found_key.offset - 1,
2729 bit | EXTENT_LOCKED, GFP_NOFS);
2730 set_state_private(block_group_cache, found_key.objectid,
2731 (unsigned long)cache);
2734 btrfs_super_total_bytes(&info->super_copy))
2738 btrfs_free_path(path);