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_map_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_map_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 ret = cache_block_group(root, cache);
205 last = max(search_start, cache->key.objectid);
208 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
209 last, &start, &end, EXTENT_DIRTY);
216 start = max(last, start);
218 if (last - start < num) {
219 if (last == cache->key.objectid + cache->key.offset)
223 if (data != BTRFS_BLOCK_GROUP_MIXED &&
224 start + num > cache->key.objectid + cache->key.offset)
229 cache = btrfs_lookup_block_group(root->fs_info, search_start);
231 printk("Unable to find block group for %Lu\n",
239 last = cache->key.objectid + cache->key.offset;
241 cache = btrfs_lookup_block_group(root->fs_info, last);
247 data = BTRFS_BLOCK_GROUP_MIXED;
252 if (cache_miss && !cache->cached) {
253 cache_block_group(root, cache);
255 cache = btrfs_lookup_block_group(root->fs_info, last);
257 cache = btrfs_find_block_group(root, cache, last, data, 0);
265 static u64 div_factor(u64 num, int factor)
274 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
275 struct btrfs_block_group_cache
276 *hint, u64 search_start,
279 struct btrfs_block_group_cache *cache;
280 struct extent_map_tree *block_group_cache;
281 struct btrfs_block_group_cache *found_group = NULL;
282 struct btrfs_fs_info *info = root->fs_info;
296 block_group_cache = &info->block_group_cache;
301 if (data == BTRFS_BLOCK_GROUP_MIXED) {
302 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
305 bit = BLOCK_GROUP_DATA;
307 bit = BLOCK_GROUP_METADATA;
310 struct btrfs_block_group_cache *shint;
311 shint = btrfs_lookup_block_group(info, search_start);
312 if (shint && (shint->data == data ||
313 shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
314 used = btrfs_block_group_used(&shint->item);
315 if (used + shint->pinned <
316 div_factor(shint->key.offset, factor)) {
321 if (hint && (hint->data == data ||
322 hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
323 used = btrfs_block_group_used(&hint->item);
324 if (used + hint->pinned <
325 div_factor(hint->key.offset, factor)) {
328 last = hint->key.objectid + hint->key.offset;
332 hint_last = max(hint->key.objectid, search_start);
334 hint_last = search_start;
340 ret = find_first_extent_bit(block_group_cache, last,
345 ret = get_state_private(block_group_cache, start, &ptr);
349 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
350 last = cache->key.objectid + cache->key.offset;
351 used = btrfs_block_group_used(&cache->item);
354 free_check = cache->key.offset;
356 free_check = div_factor(cache->key.offset, factor);
357 if (used + cache->pinned < free_check) {
370 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
378 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
379 u64 owner, u64 owner_offset)
381 u32 high_crc = ~(u32)0;
382 u32 low_crc = ~(u32)0;
385 lenum = cpu_to_le64(root_objectid);
386 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
387 lenum = cpu_to_le64(ref_generation);
388 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
391 lenum = cpu_to_le64(owner);
392 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
393 lenum = cpu_to_le64(owner_offset);
394 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
396 return ((u64)high_crc << 32) | (u64)low_crc;
399 static int match_extent_ref(struct extent_buffer *leaf,
400 struct btrfs_extent_ref *disk_ref,
401 struct btrfs_extent_ref *cpu_ref)
406 if (cpu_ref->objectid)
407 len = sizeof(*cpu_ref);
409 len = 2 * sizeof(u64);
410 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
415 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
416 struct btrfs_root *root,
417 struct btrfs_path *path, u64 bytenr,
419 u64 ref_generation, u64 owner,
420 u64 owner_offset, int del)
423 struct btrfs_key key;
424 struct btrfs_key found_key;
425 struct btrfs_extent_ref ref;
426 struct extent_buffer *leaf;
427 struct btrfs_extent_ref *disk_ref;
431 btrfs_set_stack_ref_root(&ref, root_objectid);
432 btrfs_set_stack_ref_generation(&ref, ref_generation);
433 btrfs_set_stack_ref_objectid(&ref, owner);
434 btrfs_set_stack_ref_offset(&ref, owner_offset);
436 hash = hash_extent_ref(root_objectid, ref_generation, owner,
439 key.objectid = bytenr;
440 key.type = BTRFS_EXTENT_REF_KEY;
443 ret = btrfs_search_slot(trans, root, &key, path,
447 leaf = path->nodes[0];
449 u32 nritems = btrfs_header_nritems(leaf);
450 if (path->slots[0] >= nritems) {
451 ret2 = btrfs_next_leaf(root, path);
454 leaf = path->nodes[0];
456 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
457 if (found_key.objectid != bytenr ||
458 found_key.type != BTRFS_EXTENT_REF_KEY)
460 key.offset = found_key.offset;
462 btrfs_release_path(root, path);
466 disk_ref = btrfs_item_ptr(path->nodes[0],
468 struct btrfs_extent_ref);
469 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
473 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
474 key.offset = found_key.offset + 1;
475 btrfs_release_path(root, path);
482 * Back reference rules. Back refs have three main goals:
484 * 1) differentiate between all holders of references to an extent so that
485 * when a reference is dropped we can make sure it was a valid reference
486 * before freeing the extent.
488 * 2) Provide enough information to quickly find the holders of an extent
489 * if we notice a given block is corrupted or bad.
491 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
492 * maintenance. This is actually the same as #2, but with a slightly
493 * different use case.
495 * File extents can be referenced by:
497 * - multiple snapshots, subvolumes, or different generations in one subvol
498 * - different files inside a single subvolume (in theory, not implemented yet)
499 * - different offsets inside a file (bookend extents in file.c)
501 * The extent ref structure has fields for:
503 * - Objectid of the subvolume root
504 * - Generation number of the tree holding the reference
505 * - objectid of the file holding the reference
506 * - offset in the file corresponding to the key holding the reference
508 * When a file extent is allocated the fields are filled in:
509 * (root_key.objectid, trans->transid, inode objectid, offset in file)
511 * When a leaf is cow'd new references are added for every file extent found
512 * in the leaf. It looks the same as the create case, but trans->transid
513 * will be different when the block is cow'd.
515 * (root_key.objectid, trans->transid, inode objectid, offset in file)
517 * When a file extent is removed either during snapshot deletion or file
518 * truncation, the corresponding back reference is found
521 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
522 * inode objectid, offset in file)
524 * Btree extents can be referenced by:
526 * - Different subvolumes
527 * - Different generations of the same subvolume
529 * Storing sufficient information for a full reverse mapping of a btree
530 * block would require storing the lowest key of the block in the backref,
531 * and it would require updating that lowest key either before write out or
532 * every time it changed. Instead, the objectid of the lowest key is stored
533 * along with the level of the tree block. This provides a hint
534 * about where in the btree the block can be found. Searches through the
535 * btree only need to look for a pointer to that block, so they stop one
536 * level higher than the level recorded in the backref.
538 * Some btrees do not do reference counting on their extents. These
539 * include the extent tree and the tree of tree roots. Backrefs for these
540 * trees always have a generation of zero.
542 * When a tree block is created, back references are inserted:
544 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
546 * When a tree block is cow'd in a reference counted root,
547 * new back references are added for all the blocks it points to.
548 * These are of the form (trans->transid will have increased since creation):
550 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
552 * Because the lowest_key_objectid and the level are just hints
553 * they are not used when backrefs are deleted. When a backref is deleted:
555 * if backref was for a tree root:
556 * root_objectid = root->root_key.objectid
558 * root_objectid = btrfs_header_owner(parent)
560 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
562 * Back Reference Key hashing:
564 * Back references have four fields, each 64 bits long. Unfortunately,
565 * This is hashed into a single 64 bit number and placed into the key offset.
566 * The key objectid corresponds to the first byte in the extent, and the
567 * key type is set to BTRFS_EXTENT_REF_KEY
569 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
570 struct btrfs_root *root,
571 struct btrfs_path *path, u64 bytenr,
572 u64 root_objectid, u64 ref_generation,
573 u64 owner, u64 owner_offset)
576 struct btrfs_key key;
577 struct btrfs_extent_ref ref;
578 struct btrfs_extent_ref *disk_ref;
581 btrfs_set_stack_ref_root(&ref, root_objectid);
582 btrfs_set_stack_ref_generation(&ref, ref_generation);
583 btrfs_set_stack_ref_objectid(&ref, owner);
584 btrfs_set_stack_ref_offset(&ref, owner_offset);
586 hash = hash_extent_ref(root_objectid, ref_generation, owner,
589 key.objectid = bytenr;
590 key.type = BTRFS_EXTENT_REF_KEY;
592 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
593 while (ret == -EEXIST) {
594 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
595 struct btrfs_extent_ref);
596 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
599 btrfs_release_path(root, path);
600 ret = btrfs_insert_empty_item(trans, root, path, &key,
605 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
606 struct btrfs_extent_ref);
607 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
609 btrfs_mark_buffer_dirty(path->nodes[0]);
611 btrfs_release_path(root, path);
615 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
616 struct btrfs_root *root,
617 u64 bytenr, u64 num_bytes,
618 u64 root_objectid, u64 ref_generation,
619 u64 owner, u64 owner_offset)
621 struct btrfs_path *path;
623 struct btrfs_key key;
624 struct extent_buffer *l;
625 struct btrfs_extent_item *item;
628 WARN_ON(num_bytes < root->sectorsize);
629 path = btrfs_alloc_path();
633 key.objectid = bytenr;
634 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
635 key.offset = num_bytes;
636 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
645 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
646 refs = btrfs_extent_refs(l, item);
647 btrfs_set_extent_refs(l, item, refs + 1);
648 btrfs_mark_buffer_dirty(path->nodes[0]);
650 btrfs_release_path(root->fs_info->extent_root, path);
652 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
653 path, bytenr, root_objectid,
654 ref_generation, owner, owner_offset);
656 finish_current_insert(trans, root->fs_info->extent_root);
657 del_pending_extents(trans, root->fs_info->extent_root);
659 btrfs_free_path(path);
663 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
664 struct btrfs_root *root)
666 finish_current_insert(trans, root->fs_info->extent_root);
667 del_pending_extents(trans, root->fs_info->extent_root);
671 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
672 struct btrfs_root *root, u64 bytenr,
673 u64 num_bytes, u32 *refs)
675 struct btrfs_path *path;
677 struct btrfs_key key;
678 struct extent_buffer *l;
679 struct btrfs_extent_item *item;
681 WARN_ON(num_bytes < root->sectorsize);
682 path = btrfs_alloc_path();
683 key.objectid = bytenr;
684 key.offset = num_bytes;
685 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
686 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
691 btrfs_print_leaf(root, path->nodes[0]);
692 printk("failed to find block number %Lu\n", bytenr);
696 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
697 *refs = btrfs_extent_refs(l, item);
699 btrfs_free_path(path);
703 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
704 struct btrfs_path *count_path,
707 struct btrfs_root *extent_root = root->fs_info->extent_root;
708 struct btrfs_path *path;
711 u64 root_objectid = root->root_key.objectid;
716 struct btrfs_key key;
717 struct btrfs_key found_key;
718 struct extent_buffer *l;
719 struct btrfs_extent_item *item;
720 struct btrfs_extent_ref *ref_item;
723 path = btrfs_alloc_path();
726 bytenr = first_extent;
728 bytenr = count_path->nodes[level]->start;
731 key.objectid = bytenr;
734 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
735 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
741 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
743 if (found_key.objectid != bytenr ||
744 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
748 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
751 nritems = btrfs_header_nritems(l);
752 if (path->slots[0] >= nritems) {
753 ret = btrfs_next_leaf(extent_root, path);
758 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
759 if (found_key.objectid != bytenr)
762 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
768 ref_item = btrfs_item_ptr(l, path->slots[0],
769 struct btrfs_extent_ref);
770 found_objectid = btrfs_ref_root(l, ref_item);
772 if (found_objectid != root_objectid) {
779 if (cur_count == 0) {
783 if (level >= 0 && root->node == count_path->nodes[level])
786 btrfs_release_path(root, path);
790 btrfs_free_path(path);
793 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
794 struct btrfs_root *root, u64 owner_objectid)
800 struct btrfs_disk_key disk_key;
802 level = btrfs_header_level(root->node);
803 generation = trans->transid;
804 nritems = btrfs_header_nritems(root->node);
807 btrfs_item_key(root->node, &disk_key, 0);
809 btrfs_node_key(root->node, &disk_key, 0);
810 key_objectid = btrfs_disk_key_objectid(&disk_key);
814 return btrfs_inc_extent_ref(trans, root, root->node->start,
815 root->node->len, owner_objectid,
816 generation, level, key_objectid);
819 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
820 struct extent_buffer *buf)
824 struct btrfs_key key;
825 struct btrfs_file_extent_item *fi;
834 level = btrfs_header_level(buf);
835 nritems = btrfs_header_nritems(buf);
836 for (i = 0; i < nritems; i++) {
839 btrfs_item_key_to_cpu(buf, &key, i);
840 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
842 fi = btrfs_item_ptr(buf, i,
843 struct btrfs_file_extent_item);
844 if (btrfs_file_extent_type(buf, fi) ==
845 BTRFS_FILE_EXTENT_INLINE)
847 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
848 if (disk_bytenr == 0)
850 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
851 btrfs_file_extent_disk_num_bytes(buf, fi),
852 root->root_key.objectid, trans->transid,
853 key.objectid, key.offset);
859 bytenr = btrfs_node_blockptr(buf, i);
860 btrfs_node_key_to_cpu(buf, &key, i);
861 ret = btrfs_inc_extent_ref(trans, root, bytenr,
862 btrfs_level_size(root, level - 1),
863 root->root_key.objectid,
865 level - 1, key.objectid);
876 for (i =0; i < faili; i++) {
879 btrfs_item_key_to_cpu(buf, &key, i);
880 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
882 fi = btrfs_item_ptr(buf, i,
883 struct btrfs_file_extent_item);
884 if (btrfs_file_extent_type(buf, fi) ==
885 BTRFS_FILE_EXTENT_INLINE)
887 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
888 if (disk_bytenr == 0)
890 err = btrfs_free_extent(trans, root, disk_bytenr,
891 btrfs_file_extent_disk_num_bytes(buf,
895 bytenr = btrfs_node_blockptr(buf, i);
896 err = btrfs_free_extent(trans, root, bytenr,
897 btrfs_level_size(root, level - 1), 0);
905 static int write_one_cache_group(struct btrfs_trans_handle *trans,
906 struct btrfs_root *root,
907 struct btrfs_path *path,
908 struct btrfs_block_group_cache *cache)
912 struct btrfs_root *extent_root = root->fs_info->extent_root;
914 struct extent_buffer *leaf;
916 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
921 leaf = path->nodes[0];
922 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
923 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
924 btrfs_mark_buffer_dirty(leaf);
925 btrfs_release_path(extent_root, path);
927 finish_current_insert(trans, extent_root);
928 pending_ret = del_pending_extents(trans, extent_root);
937 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
938 struct btrfs_root *root)
940 struct extent_map_tree *block_group_cache;
941 struct btrfs_block_group_cache *cache;
945 struct btrfs_path *path;
951 block_group_cache = &root->fs_info->block_group_cache;
952 path = btrfs_alloc_path();
957 ret = find_first_extent_bit(block_group_cache, last,
958 &start, &end, BLOCK_GROUP_DIRTY);
963 ret = get_state_private(block_group_cache, start, &ptr);
967 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
968 err = write_one_cache_group(trans, root,
971 * if we fail to write the cache group, we want
972 * to keep it marked dirty in hopes that a later
979 clear_extent_bits(block_group_cache, start, end,
980 BLOCK_GROUP_DIRTY, GFP_NOFS);
982 btrfs_free_path(path);
986 static int update_block_group(struct btrfs_trans_handle *trans,
987 struct btrfs_root *root,
988 u64 bytenr, u64 num_bytes, int alloc,
989 int mark_free, int data)
991 struct btrfs_block_group_cache *cache;
992 struct btrfs_fs_info *info = root->fs_info;
993 u64 total = num_bytes;
1000 cache = btrfs_lookup_block_group(info, bytenr);
1004 byte_in_group = bytenr - cache->key.objectid;
1005 WARN_ON(byte_in_group > cache->key.offset);
1006 start = cache->key.objectid;
1007 end = start + cache->key.offset - 1;
1008 set_extent_bits(&info->block_group_cache, start, end,
1009 BLOCK_GROUP_DIRTY, GFP_NOFS);
1011 old_val = btrfs_block_group_used(&cache->item);
1012 num_bytes = min(total, cache->key.offset - byte_in_group);
1014 if (cache->data != data &&
1015 old_val < (cache->key.offset >> 1)) {
1020 bit_to_clear = BLOCK_GROUP_METADATA;
1021 bit_to_set = BLOCK_GROUP_DATA;
1022 cache->item.flags &=
1023 ~BTRFS_BLOCK_GROUP_MIXED;
1024 cache->item.flags |=
1025 BTRFS_BLOCK_GROUP_DATA;
1027 bit_to_clear = BLOCK_GROUP_DATA;
1028 bit_to_set = BLOCK_GROUP_METADATA;
1029 cache->item.flags &=
1030 ~BTRFS_BLOCK_GROUP_MIXED;
1031 cache->item.flags &=
1032 ~BTRFS_BLOCK_GROUP_DATA;
1034 clear_extent_bits(&info->block_group_cache,
1035 start, end, bit_to_clear,
1037 set_extent_bits(&info->block_group_cache,
1038 start, end, bit_to_set,
1040 } else if (cache->data != data &&
1041 cache->data != BTRFS_BLOCK_GROUP_MIXED) {
1042 cache->data = BTRFS_BLOCK_GROUP_MIXED;
1043 set_extent_bits(&info->block_group_cache,
1046 BLOCK_GROUP_METADATA,
1049 old_val += num_bytes;
1051 old_val -= num_bytes;
1053 set_extent_dirty(&info->free_space_cache,
1054 bytenr, bytenr + num_bytes - 1,
1058 btrfs_set_block_group_used(&cache->item, old_val);
1060 bytenr += num_bytes;
1064 static int update_pinned_extents(struct btrfs_root *root,
1065 u64 bytenr, u64 num, int pin)
1068 struct btrfs_block_group_cache *cache;
1069 struct btrfs_fs_info *fs_info = root->fs_info;
1072 set_extent_dirty(&fs_info->pinned_extents,
1073 bytenr, bytenr + num - 1, GFP_NOFS);
1075 clear_extent_dirty(&fs_info->pinned_extents,
1076 bytenr, bytenr + num - 1, GFP_NOFS);
1079 cache = btrfs_lookup_block_group(fs_info, bytenr);
1081 len = min(num, cache->key.offset -
1082 (bytenr - cache->key.objectid));
1084 cache->pinned += len;
1085 fs_info->total_pinned += len;
1087 cache->pinned -= len;
1088 fs_info->total_pinned -= len;
1096 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
1101 struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
1105 ret = find_first_extent_bit(pinned_extents, last,
1106 &start, &end, EXTENT_DIRTY);
1109 set_extent_dirty(copy, start, end, GFP_NOFS);
1115 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1116 struct btrfs_root *root,
1117 struct extent_map_tree *unpin)
1122 struct extent_map_tree *free_space_cache;
1123 free_space_cache = &root->fs_info->free_space_cache;
1126 ret = find_first_extent_bit(unpin, 0, &start, &end,
1130 update_pinned_extents(root, start, end + 1 - start, 0);
1131 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1132 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1137 static int finish_current_insert(struct btrfs_trans_handle *trans,
1138 struct btrfs_root *extent_root)
1142 struct btrfs_fs_info *info = extent_root->fs_info;
1143 struct extent_buffer *eb;
1144 struct btrfs_path *path;
1145 struct btrfs_key ins;
1146 struct btrfs_disk_key first;
1147 struct btrfs_extent_item extent_item;
1152 btrfs_set_stack_extent_refs(&extent_item, 1);
1153 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1154 path = btrfs_alloc_path();
1157 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1158 &end, EXTENT_LOCKED);
1162 ins.objectid = start;
1163 ins.offset = end + 1 - start;
1164 err = btrfs_insert_item(trans, extent_root, &ins,
1165 &extent_item, sizeof(extent_item));
1166 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1168 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1169 level = btrfs_header_level(eb);
1171 btrfs_item_key(eb, &first, 0);
1173 btrfs_node_key(eb, &first, 0);
1175 err = btrfs_insert_extent_backref(trans, extent_root, path,
1176 start, extent_root->root_key.objectid,
1178 btrfs_disk_key_objectid(&first));
1180 free_extent_buffer(eb);
1182 btrfs_free_path(path);
1186 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1190 struct extent_buffer *buf;
1193 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1195 if (btrfs_buffer_uptodate(buf)) {
1197 root->fs_info->running_transaction->transid;
1198 if (btrfs_header_generation(buf) == transid) {
1199 free_extent_buffer(buf);
1203 free_extent_buffer(buf);
1205 update_pinned_extents(root, bytenr, num_bytes, 1);
1207 set_extent_bits(&root->fs_info->pending_del,
1208 bytenr, bytenr + num_bytes - 1,
1209 EXTENT_LOCKED, GFP_NOFS);
1216 * remove an extent from the root, returns 0 on success
1218 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1219 *root, u64 bytenr, u64 num_bytes,
1220 u64 root_objectid, u64 ref_generation,
1221 u64 owner_objectid, u64 owner_offset, int pin,
1224 struct btrfs_path *path;
1225 struct btrfs_key key;
1226 struct btrfs_fs_info *info = root->fs_info;
1227 struct btrfs_root *extent_root = info->extent_root;
1228 struct extent_buffer *leaf;
1230 struct btrfs_extent_item *ei;
1233 key.objectid = bytenr;
1234 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1235 key.offset = num_bytes;
1237 path = btrfs_alloc_path();
1241 ret = lookup_extent_backref(trans, extent_root, path,
1242 bytenr, root_objectid,
1244 owner_objectid, owner_offset, 1);
1246 ret = btrfs_del_item(trans, extent_root, path);
1248 btrfs_print_leaf(extent_root, path->nodes[0]);
1250 printk("Unable to find ref byte nr %Lu root %Lu "
1251 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1252 root_objectid, ref_generation, owner_objectid,
1255 btrfs_release_path(extent_root, path);
1256 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1261 leaf = path->nodes[0];
1262 ei = btrfs_item_ptr(leaf, path->slots[0],
1263 struct btrfs_extent_item);
1264 refs = btrfs_extent_refs(leaf, ei);
1267 btrfs_set_extent_refs(leaf, ei, refs);
1268 btrfs_mark_buffer_dirty(leaf);
1275 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1281 /* block accounting for super block */
1282 super_used = btrfs_super_bytes_used(&info->super_copy);
1283 btrfs_set_super_bytes_used(&info->super_copy,
1284 super_used - num_bytes);
1286 /* block accounting for root item */
1287 root_used = btrfs_root_used(&root->root_item);
1288 btrfs_set_root_used(&root->root_item,
1289 root_used - num_bytes);
1291 ret = btrfs_del_item(trans, extent_root, path);
1295 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1299 btrfs_free_path(path);
1300 finish_current_insert(trans, extent_root);
1305 * find all the blocks marked as pending in the radix tree and remove
1306 * them from the extent map
1308 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1309 btrfs_root *extent_root)
1315 struct extent_map_tree *pending_del;
1316 struct extent_map_tree *pinned_extents;
1318 pending_del = &extent_root->fs_info->pending_del;
1319 pinned_extents = &extent_root->fs_info->pinned_extents;
1322 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1326 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1327 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1329 ret = __free_extent(trans, extent_root,
1330 start, end + 1 - start,
1331 extent_root->root_key.objectid,
1340 * remove an extent from the root, returns 0 on success
1342 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1343 *root, u64 bytenr, u64 num_bytes,
1344 u64 root_objectid, u64 ref_generation,
1345 u64 owner_objectid, u64 owner_offset, int pin)
1347 struct btrfs_root *extent_root = root->fs_info->extent_root;
1351 WARN_ON(num_bytes < root->sectorsize);
1352 if (!root->ref_cows)
1355 if (root == extent_root) {
1356 pin_down_bytes(root, bytenr, num_bytes, 1);
1359 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1360 ref_generation, owner_objectid, owner_offset,
1362 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1363 return ret ? ret : pending_ret;
1366 static u64 stripe_align(struct btrfs_root *root, u64 val)
1368 u64 mask = ((u64)root->stripesize - 1);
1369 u64 ret = (val + mask) & ~mask;
1374 * walks the btree of allocated extents and find a hole of a given size.
1375 * The key ins is changed to record the hole:
1376 * ins->objectid == block start
1377 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1378 * ins->offset == number of blocks
1379 * Any available blocks before search_start are skipped.
1381 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1382 struct btrfs_root *orig_root,
1383 u64 num_bytes, u64 empty_size,
1384 u64 search_start, u64 search_end,
1385 u64 hint_byte, struct btrfs_key *ins,
1386 u64 exclude_start, u64 exclude_nr,
1389 struct btrfs_path *path;
1390 struct btrfs_key key;
1396 u64 orig_search_start = search_start;
1398 struct extent_buffer *l;
1399 struct btrfs_root * root = orig_root->fs_info->extent_root;
1400 struct btrfs_fs_info *info = root->fs_info;
1401 u64 total_needed = num_bytes;
1403 struct btrfs_block_group_cache *block_group;
1408 WARN_ON(num_bytes < root->sectorsize);
1409 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1411 level = btrfs_header_level(root->node);
1413 if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1414 data = BTRFS_BLOCK_GROUP_MIXED;
1417 if (search_end == (u64)-1)
1418 search_end = btrfs_super_total_bytes(&info->super_copy);
1420 block_group = btrfs_lookup_block_group(info, hint_byte);
1422 hint_byte = search_start;
1423 block_group = btrfs_find_block_group(root, block_group,
1424 hint_byte, data, 1);
1426 block_group = btrfs_find_block_group(root,
1428 search_start, data, 1);
1431 total_needed += empty_size;
1432 path = btrfs_alloc_path();
1435 block_group = btrfs_lookup_block_group(info, search_start);
1437 block_group = btrfs_lookup_block_group(info,
1440 search_start = find_search_start(root, &block_group, search_start,
1441 total_needed, data);
1442 search_start = stripe_align(root, search_start);
1443 cached_start = search_start;
1444 btrfs_init_path(path);
1445 ins->objectid = search_start;
1450 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1453 ret = find_previous_extent(root, path);
1457 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1460 slot = path->slots[0];
1461 if (slot >= btrfs_header_nritems(l)) {
1462 ret = btrfs_next_leaf(root, path);
1468 search_start = max(search_start,
1469 block_group->key.objectid);
1471 aligned = stripe_align(root, search_start);
1472 ins->objectid = aligned;
1473 if (aligned >= search_end) {
1477 ins->offset = search_end - aligned;
1481 ins->objectid = stripe_align(root,
1482 last_byte > search_start ?
1483 last_byte : search_start);
1484 if (search_end <= ins->objectid) {
1488 ins->offset = search_end - ins->objectid;
1489 BUG_ON(ins->objectid >= search_end);
1492 btrfs_item_key_to_cpu(l, &key, slot);
1494 if (key.objectid >= search_start && key.objectid > last_byte &&
1496 if (last_byte < search_start)
1497 last_byte = search_start;
1498 aligned = stripe_align(root, last_byte);
1499 hole_size = key.objectid - aligned;
1500 if (key.objectid > aligned && hole_size >= num_bytes) {
1501 ins->objectid = aligned;
1502 ins->offset = hole_size;
1506 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1507 if (!start_found && btrfs_key_type(&key) ==
1508 BTRFS_BLOCK_GROUP_ITEM_KEY) {
1509 last_byte = key.objectid;
1517 last_byte = key.objectid + key.offset;
1519 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1520 last_byte >= block_group->key.objectid +
1521 block_group->key.offset) {
1522 btrfs_release_path(root, path);
1523 search_start = block_group->key.objectid +
1524 block_group->key.offset;
1532 /* we have to make sure we didn't find an extent that has already
1533 * been allocated by the map tree or the original allocation
1535 btrfs_release_path(root, path);
1536 BUG_ON(ins->objectid < search_start);
1538 if (ins->objectid + num_bytes >= search_end)
1540 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1541 ins->objectid + num_bytes > block_group->
1542 key.objectid + block_group->key.offset) {
1543 search_start = block_group->key.objectid +
1544 block_group->key.offset;
1547 if (test_range_bit(&info->extent_ins, ins->objectid,
1548 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1549 search_start = ins->objectid + num_bytes;
1552 if (test_range_bit(&info->pinned_extents, ins->objectid,
1553 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1554 search_start = ins->objectid + num_bytes;
1557 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1558 ins->objectid < exclude_start + exclude_nr)) {
1559 search_start = exclude_start + exclude_nr;
1563 block_group = btrfs_lookup_block_group(info, ins->objectid);
1565 trans->block_group = block_group;
1567 ins->offset = num_bytes;
1568 btrfs_free_path(path);
1572 if (search_start + num_bytes >= search_end) {
1574 search_start = orig_search_start;
1581 total_needed -= empty_size;
1583 data = BTRFS_BLOCK_GROUP_MIXED;
1587 block_group = btrfs_lookup_block_group(info, search_start);
1589 block_group = btrfs_find_block_group(root, block_group,
1590 search_start, data, 0);
1594 btrfs_release_path(root, path);
1595 btrfs_free_path(path);
1599 * finds a free extent and does all the dirty work required for allocation
1600 * returns the key for the extent through ins, and a tree buffer for
1601 * the first block of the extent through buf.
1603 * returns 0 if everything worked, non-zero otherwise.
1605 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1606 struct btrfs_root *root,
1607 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1608 u64 owner, u64 owner_offset,
1609 u64 empty_size, u64 hint_byte,
1610 u64 search_end, struct btrfs_key *ins, int data)
1614 u64 super_used, root_used;
1615 u64 search_start = 0;
1617 struct btrfs_fs_info *info = root->fs_info;
1618 struct btrfs_root *extent_root = info->extent_root;
1619 struct btrfs_extent_item extent_item;
1620 struct btrfs_path *path;
1622 btrfs_set_stack_extent_refs(&extent_item, 1);
1624 new_hint = max(hint_byte, root->fs_info->alloc_start);
1625 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1626 hint_byte = new_hint;
1628 WARN_ON(num_bytes < root->sectorsize);
1629 ret = find_free_extent(trans, root, num_bytes, empty_size,
1630 search_start, search_end, hint_byte, ins,
1631 trans->alloc_exclude_start,
1632 trans->alloc_exclude_nr, data);
1637 /* block accounting for super block */
1638 super_used = btrfs_super_bytes_used(&info->super_copy);
1639 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1641 /* block accounting for root item */
1642 root_used = btrfs_root_used(&root->root_item);
1643 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1645 clear_extent_dirty(&root->fs_info->free_space_cache,
1646 ins->objectid, ins->objectid + ins->offset - 1,
1649 if (root == extent_root) {
1650 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1651 ins->objectid + ins->offset - 1,
1652 EXTENT_LOCKED, GFP_NOFS);
1657 WARN_ON(trans->alloc_exclude_nr);
1658 trans->alloc_exclude_start = ins->objectid;
1659 trans->alloc_exclude_nr = ins->offset;
1660 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1661 sizeof(extent_item));
1663 trans->alloc_exclude_start = 0;
1664 trans->alloc_exclude_nr = 0;
1667 path = btrfs_alloc_path();
1669 ret = btrfs_insert_extent_backref(trans, extent_root, path,
1670 ins->objectid, root_objectid,
1671 ref_generation, owner, owner_offset);
1674 btrfs_free_path(path);
1675 finish_current_insert(trans, extent_root);
1676 pending_ret = del_pending_extents(trans, extent_root);
1686 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1693 * helper function to allocate a block for a given tree
1694 * returns the tree buffer or NULL.
1696 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1697 struct btrfs_root *root,
1699 u64 root_objectid, u64 hint,
1705 ref_generation = trans->transid;
1710 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1711 ref_generation, 0, 0, hint, empty_size);
1715 * helper function to allocate a block for a given tree
1716 * returns the tree buffer or NULL.
1718 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1719 struct btrfs_root *root,
1728 struct btrfs_key ins;
1730 struct extent_buffer *buf;
1732 ret = btrfs_alloc_extent(trans, root, blocksize,
1733 root_objectid, ref_generation,
1734 level, first_objectid, empty_size, hint,
1738 return ERR_PTR(ret);
1740 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1742 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1743 root->root_key.objectid, ref_generation,
1745 return ERR_PTR(-ENOMEM);
1747 btrfs_set_buffer_uptodate(buf);
1748 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1749 buf->start + buf->len - 1, GFP_NOFS);
1750 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1751 buf->start, buf->start + buf->len - 1,
1752 EXTENT_CSUM, GFP_NOFS);
1753 buf->flags |= EXTENT_CSUM;
1754 btrfs_set_buffer_defrag(buf);
1755 trans->blocks_used++;
1759 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1760 struct btrfs_root *root,
1761 struct extent_buffer *leaf)
1764 u64 leaf_generation;
1765 struct btrfs_key key;
1766 struct btrfs_file_extent_item *fi;
1771 BUG_ON(!btrfs_is_leaf(leaf));
1772 nritems = btrfs_header_nritems(leaf);
1773 leaf_owner = btrfs_header_owner(leaf);
1774 leaf_generation = btrfs_header_generation(leaf);
1776 for (i = 0; i < nritems; i++) {
1779 btrfs_item_key_to_cpu(leaf, &key, i);
1780 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1782 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1783 if (btrfs_file_extent_type(leaf, fi) ==
1784 BTRFS_FILE_EXTENT_INLINE)
1787 * FIXME make sure to insert a trans record that
1788 * repeats the snapshot del on crash
1790 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1791 if (disk_bytenr == 0)
1793 ret = btrfs_free_extent(trans, root, disk_bytenr,
1794 btrfs_file_extent_disk_num_bytes(leaf, fi),
1795 leaf_owner, leaf_generation,
1796 key.objectid, key.offset, 0);
1802 static void noinline reada_walk_down(struct btrfs_root *root,
1803 struct extent_buffer *node)
1813 nritems = btrfs_header_nritems(node);
1814 level = btrfs_header_level(node);
1815 for (i = 0; i < nritems; i++) {
1816 bytenr = btrfs_node_blockptr(node, i);
1817 blocksize = btrfs_level_size(root, level - 1);
1818 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1822 mutex_unlock(&root->fs_info->fs_mutex);
1823 ret = readahead_tree_block(root, bytenr, blocksize);
1825 mutex_lock(&root->fs_info->fs_mutex);
1832 * helper function for drop_snapshot, this walks down the tree dropping ref
1833 * counts as it goes.
1835 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1836 struct btrfs_root *root,
1837 struct btrfs_path *path, int *level)
1842 struct extent_buffer *next;
1843 struct extent_buffer *cur;
1844 struct extent_buffer *parent;
1849 WARN_ON(*level < 0);
1850 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1851 ret = lookup_extent_ref(trans, root,
1852 path->nodes[*level]->start,
1853 path->nodes[*level]->len, &refs);
1859 * walk down to the last node level and free all the leaves
1861 while(*level >= 0) {
1862 WARN_ON(*level < 0);
1863 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1864 cur = path->nodes[*level];
1866 if (*level > 0 && path->slots[*level] == 0)
1867 reada_walk_down(root, cur);
1869 if (btrfs_header_level(cur) != *level)
1872 if (path->slots[*level] >=
1873 btrfs_header_nritems(cur))
1876 ret = drop_leaf_ref(trans, root, cur);
1880 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1881 blocksize = btrfs_level_size(root, *level - 1);
1882 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1885 parent = path->nodes[*level];
1886 root_owner = btrfs_header_owner(parent);
1887 root_gen = btrfs_header_generation(parent);
1888 path->slots[*level]++;
1889 ret = btrfs_free_extent(trans, root, bytenr,
1890 blocksize, root_owner,
1895 next = btrfs_find_tree_block(root, bytenr, blocksize);
1896 if (!next || !btrfs_buffer_uptodate(next)) {
1897 free_extent_buffer(next);
1898 mutex_unlock(&root->fs_info->fs_mutex);
1899 next = read_tree_block(root, bytenr, blocksize);
1900 mutex_lock(&root->fs_info->fs_mutex);
1902 /* we dropped the lock, check one more time */
1903 ret = lookup_extent_ref(trans, root, bytenr,
1907 parent = path->nodes[*level];
1908 root_owner = btrfs_header_owner(parent);
1909 root_gen = btrfs_header_generation(parent);
1911 path->slots[*level]++;
1912 free_extent_buffer(next);
1913 ret = btrfs_free_extent(trans, root, bytenr,
1921 WARN_ON(*level <= 0);
1922 if (path->nodes[*level-1])
1923 free_extent_buffer(path->nodes[*level-1]);
1924 path->nodes[*level-1] = next;
1925 *level = btrfs_header_level(next);
1926 path->slots[*level] = 0;
1929 WARN_ON(*level < 0);
1930 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1932 if (path->nodes[*level] == root->node) {
1933 root_owner = root->root_key.objectid;
1934 parent = path->nodes[*level];
1936 parent = path->nodes[*level + 1];
1937 root_owner = btrfs_header_owner(parent);
1940 root_gen = btrfs_header_generation(parent);
1941 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1942 path->nodes[*level]->len,
1943 root_owner, root_gen, 0, 0, 1);
1944 free_extent_buffer(path->nodes[*level]);
1945 path->nodes[*level] = NULL;
1952 * helper for dropping snapshots. This walks back up the tree in the path
1953 * to find the first node higher up where we haven't yet gone through
1956 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
1957 struct btrfs_root *root,
1958 struct btrfs_path *path, int *level)
1962 struct btrfs_root_item *root_item = &root->root_item;
1967 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1968 slot = path->slots[i];
1969 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1970 struct extent_buffer *node;
1971 struct btrfs_disk_key disk_key;
1972 node = path->nodes[i];
1975 WARN_ON(*level == 0);
1976 btrfs_node_key(node, &disk_key, path->slots[i]);
1977 memcpy(&root_item->drop_progress,
1978 &disk_key, sizeof(disk_key));
1979 root_item->drop_level = i;
1982 if (path->nodes[*level] == root->node) {
1983 root_owner = root->root_key.objectid;
1985 btrfs_header_generation(path->nodes[*level]);
1987 struct extent_buffer *node;
1988 node = path->nodes[*level + 1];
1989 root_owner = btrfs_header_owner(node);
1990 root_gen = btrfs_header_generation(node);
1992 ret = btrfs_free_extent(trans, root,
1993 path->nodes[*level]->start,
1994 path->nodes[*level]->len,
1995 root_owner, root_gen, 0, 0, 1);
1997 free_extent_buffer(path->nodes[*level]);
1998 path->nodes[*level] = NULL;
2006 * drop the reference count on the tree rooted at 'snap'. This traverses
2007 * the tree freeing any blocks that have a ref count of zero after being
2010 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2016 struct btrfs_path *path;
2019 struct btrfs_root_item *root_item = &root->root_item;
2021 path = btrfs_alloc_path();
2024 level = btrfs_header_level(root->node);
2026 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2027 path->nodes[level] = root->node;
2028 extent_buffer_get(root->node);
2029 path->slots[level] = 0;
2031 struct btrfs_key key;
2032 struct btrfs_disk_key found_key;
2033 struct extent_buffer *node;
2035 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2036 level = root_item->drop_level;
2037 path->lowest_level = level;
2038 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2043 node = path->nodes[level];
2044 btrfs_node_key(node, &found_key, path->slots[level]);
2045 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2046 sizeof(found_key)));
2049 wret = walk_down_tree(trans, root, path, &level);
2055 wret = walk_up_tree(trans, root, path, &level);
2063 for (i = 0; i <= orig_level; i++) {
2064 if (path->nodes[i]) {
2065 free_extent_buffer(path->nodes[i]);
2066 path->nodes[i] = NULL;
2070 btrfs_free_path(path);
2074 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2081 ret = find_first_extent_bit(&info->block_group_cache, 0,
2082 &start, &end, (unsigned int)-1);
2085 ret = get_state_private(&info->block_group_cache, start, &ptr);
2087 kfree((void *)(unsigned long)ptr);
2088 clear_extent_bits(&info->block_group_cache, start,
2089 end, (unsigned int)-1, GFP_NOFS);
2092 ret = find_first_extent_bit(&info->free_space_cache, 0,
2093 &start, &end, EXTENT_DIRTY);
2096 clear_extent_dirty(&info->free_space_cache, start,
2102 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2108 u64 existing_delalloc;
2109 unsigned long last_index;
2112 struct btrfs_root *root = BTRFS_I(inode)->root;
2113 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2114 struct file_ra_state *ra;
2116 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2118 mutex_lock(&inode->i_mutex);
2119 i = start >> PAGE_CACHE_SHIFT;
2120 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2122 file_ra_state_init(ra, inode->i_mapping);
2123 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2126 for (; i <= last_index; i++) {
2127 page = grab_cache_page(inode->i_mapping, i);
2130 if (!PageUptodate(page)) {
2131 btrfs_readpage(NULL, page);
2133 if (!PageUptodate(page)) {
2135 page_cache_release(page);
2139 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2140 page_end = page_start + PAGE_CACHE_SIZE - 1;
2142 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2144 delalloc_start = page_start;
2146 count_range_bits(&BTRFS_I(inode)->extent_tree,
2147 &delalloc_start, page_end,
2148 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2150 set_extent_delalloc(em_tree, page_start,
2151 page_end, GFP_NOFS);
2153 spin_lock(&root->fs_info->delalloc_lock);
2154 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2156 spin_unlock(&root->fs_info->delalloc_lock);
2158 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2159 set_page_dirty(page);
2161 page_cache_release(page);
2165 mutex_unlock(&inode->i_mutex);
2170 * note, this releases the path
2172 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2173 struct btrfs_path *path,
2174 struct btrfs_key *extent_key)
2176 struct inode *inode;
2177 struct btrfs_root *found_root;
2178 struct btrfs_key *root_location;
2179 struct btrfs_extent_ref *ref;
2186 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2187 struct btrfs_extent_ref);
2188 ref_root = btrfs_ref_root(path->nodes[0], ref);
2189 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2190 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2191 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2192 btrfs_release_path(extent_root, path);
2194 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2195 root_location->objectid = ref_root;
2197 root_location->offset = 0;
2199 root_location->offset = (u64)-1;
2200 root_location->type = BTRFS_ROOT_ITEM_KEY;
2202 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2204 BUG_ON(!found_root);
2205 kfree(root_location);
2207 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2208 mutex_unlock(&extent_root->fs_info->fs_mutex);
2209 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2210 ref_objectid, found_root);
2211 if (inode->i_state & I_NEW) {
2212 /* the inode and parent dir are two different roots */
2213 BTRFS_I(inode)->root = found_root;
2214 BTRFS_I(inode)->location.objectid = ref_objectid;
2215 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2216 BTRFS_I(inode)->location.offset = 0;
2217 btrfs_read_locked_inode(inode);
2218 unlock_new_inode(inode);
2221 /* this can happen if the reference is not against
2222 * the latest version of the tree root
2224 if (is_bad_inode(inode)) {
2225 mutex_lock(&extent_root->fs_info->fs_mutex);
2228 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2229 /* FIXME, data=ordered will help get rid of this */
2230 filemap_fdatawrite(inode->i_mapping);
2232 mutex_lock(&extent_root->fs_info->fs_mutex);
2234 struct btrfs_trans_handle *trans;
2235 struct btrfs_key found_key;
2236 struct extent_buffer *eb;
2240 trans = btrfs_start_transaction(found_root, 1);
2241 eb = read_tree_block(found_root, extent_key->objectid,
2242 extent_key->offset);
2243 level = btrfs_header_level(eb);
2246 btrfs_item_key_to_cpu(eb, &found_key, 0);
2248 btrfs_node_key_to_cpu(eb, &found_key, 0);
2250 free_extent_buffer(eb);
2252 path->lowest_level = level;
2254 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2256 path->lowest_level = 0;
2257 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2258 if (!path->nodes[i])
2260 free_extent_buffer(path->nodes[i]);
2261 path->nodes[i] = NULL;
2263 btrfs_release_path(found_root, path);
2264 btrfs_end_transaction(trans, found_root);
2271 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2272 struct btrfs_path *path,
2273 struct btrfs_key *extent_key)
2275 struct btrfs_key key;
2276 struct btrfs_key found_key;
2277 struct extent_buffer *leaf;
2282 key.objectid = extent_key->objectid;
2283 key.type = BTRFS_EXTENT_REF_KEY;
2287 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2295 leaf = path->nodes[0];
2296 nritems = btrfs_header_nritems(leaf);
2297 if (path->slots[0] == nritems)
2300 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2301 if (found_key.objectid != extent_key->objectid)
2304 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2307 key.offset = found_key.offset + 1;
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2310 ret = relocate_one_reference(extent_root, path, extent_key);
2316 btrfs_release_path(extent_root, path);
2320 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2322 struct btrfs_trans_handle *trans;
2323 struct btrfs_root *tree_root = root->fs_info->tree_root;
2324 struct btrfs_path *path;
2327 struct btrfs_fs_info *info = root->fs_info;
2328 struct extent_map_tree *block_group_cache;
2329 struct btrfs_key key;
2330 struct btrfs_key found_key;
2331 struct extent_buffer *leaf;
2335 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2336 block_group_cache = &info->block_group_cache;
2337 path = btrfs_alloc_path();
2338 root = root->fs_info->extent_root;
2343 key.objectid = new_size;
2346 cur_byte = key.objectid;
2348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2352 ret = find_previous_extent(root, path);
2356 leaf = path->nodes[0];
2357 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2358 if (found_key.objectid + found_key.offset > new_size) {
2359 cur_byte = found_key.objectid;
2360 key.objectid = cur_byte;
2363 btrfs_release_path(root, path);
2366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2370 leaf = path->nodes[0];
2371 nritems = btrfs_header_nritems(leaf);
2373 if (path->slots[0] >= nritems) {
2374 ret = btrfs_next_leaf(root, path);
2381 leaf = path->nodes[0];
2382 nritems = btrfs_header_nritems(leaf);
2385 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2386 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2387 found_key.objectid + found_key.offset <= cur_byte) {
2393 cur_byte = found_key.objectid + found_key.offset;
2394 key.objectid = cur_byte;
2395 btrfs_release_path(root, path);
2396 ret = relocate_one_extent(root, path, &found_key);
2399 btrfs_release_path(root, path);
2401 if (total_found > 0) {
2402 trans = btrfs_start_transaction(tree_root, 1);
2403 btrfs_commit_transaction(trans, tree_root);
2405 mutex_unlock(&root->fs_info->fs_mutex);
2406 btrfs_clean_old_snapshots(tree_root);
2407 mutex_lock(&root->fs_info->fs_mutex);
2409 trans = btrfs_start_transaction(tree_root, 1);
2410 btrfs_commit_transaction(trans, tree_root);
2414 trans = btrfs_start_transaction(root, 1);
2415 key.objectid = new_size;
2421 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2425 leaf = path->nodes[0];
2426 nritems = btrfs_header_nritems(leaf);
2428 if (path->slots[0] >= nritems) {
2429 ret = btrfs_next_leaf(root, path);
2436 leaf = path->nodes[0];
2437 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2440 * btrfs_next_leaf doesn't cow buffers, we have to
2441 * do the search again
2443 memcpy(&key, &found_key, sizeof(key));
2444 btrfs_release_path(root, path);
2448 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2449 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2450 printk("shrinker found key %Lu %u %Lu\n",
2451 found_key.objectid, found_key.type,
2456 ret = get_state_private(&info->block_group_cache,
2457 found_key.objectid, &ptr);
2459 kfree((void *)(unsigned long)ptr);
2461 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2462 found_key.objectid + found_key.offset - 1,
2463 (unsigned int)-1, GFP_NOFS);
2465 key.objectid = found_key.objectid + 1;
2466 btrfs_del_item(trans, root, path);
2467 btrfs_release_path(root, path);
2469 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2471 btrfs_commit_transaction(trans, root);
2473 btrfs_free_path(path);
2477 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2478 struct btrfs_root *root, u64 new_size)
2480 struct btrfs_path *path;
2485 struct btrfs_block_group_cache *cache;
2486 struct btrfs_block_group_item *item;
2487 struct btrfs_fs_info *info = root->fs_info;
2488 struct extent_map_tree *block_group_cache;
2489 struct btrfs_key key;
2490 struct extent_buffer *leaf;
2494 old_size = btrfs_super_total_bytes(&info->super_copy);
2495 block_group_cache = &info->block_group_cache;
2497 root = info->extent_root;
2499 cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2501 cur_byte = cache->key.objectid + cache->key.offset;
2502 if (cur_byte >= new_size)
2505 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2506 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2508 path = btrfs_alloc_path();
2512 while(cur_byte < new_size) {
2513 key.objectid = cur_byte;
2514 ret = btrfs_insert_empty_item(trans, root, path, &key,
2515 sizeof(struct btrfs_block_group_item));
2517 leaf = path->nodes[0];
2518 item = btrfs_item_ptr(leaf, path->slots[0],
2519 struct btrfs_block_group_item);
2521 btrfs_set_disk_block_group_used(leaf, item, 0);
2522 div_long_long_rem(nr, 3, &rem);
2524 btrfs_set_disk_block_group_flags(leaf, item,
2525 BTRFS_BLOCK_GROUP_DATA);
2527 btrfs_set_disk_block_group_flags(leaf, item, 0);
2531 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2534 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2535 sizeof(cache->item));
2537 memcpy(&cache->key, &key, sizeof(key));
2540 cur_byte = key.objectid + key.offset;
2541 btrfs_release_path(root, path);
2543 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2544 bit = BLOCK_GROUP_DATA;
2545 cache->data = BTRFS_BLOCK_GROUP_DATA;
2547 bit = BLOCK_GROUP_METADATA;
2551 /* use EXTENT_LOCKED to prevent merging */
2552 set_extent_bits(block_group_cache, key.objectid,
2553 key.objectid + key.offset - 1,
2554 bit | EXTENT_LOCKED, GFP_NOFS);
2555 set_state_private(block_group_cache, key.objectid,
2556 (unsigned long)cache);
2558 btrfs_free_path(path);
2560 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2564 int btrfs_read_block_groups(struct btrfs_root *root)
2566 struct btrfs_path *path;
2570 struct btrfs_block_group_cache *cache;
2571 struct btrfs_fs_info *info = root->fs_info;
2572 struct extent_map_tree *block_group_cache;
2573 struct btrfs_key key;
2574 struct btrfs_key found_key;
2575 struct extent_buffer *leaf;
2577 block_group_cache = &info->block_group_cache;
2579 root = info->extent_root;
2581 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2582 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2584 path = btrfs_alloc_path();
2589 ret = btrfs_search_slot(NULL, info->extent_root,
2595 leaf = path->nodes[0];
2596 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2597 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2603 read_extent_buffer(leaf, &cache->item,
2604 btrfs_item_ptr_offset(leaf, path->slots[0]),
2605 sizeof(cache->item));
2606 memcpy(&cache->key, &found_key, sizeof(found_key));
2609 key.objectid = found_key.objectid + found_key.offset;
2610 btrfs_release_path(root, path);
2612 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2613 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2614 cache->data = BTRFS_BLOCK_GROUP_MIXED;
2615 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2616 bit = BLOCK_GROUP_DATA;
2617 cache->data = BTRFS_BLOCK_GROUP_DATA;
2619 bit = BLOCK_GROUP_METADATA;
2623 /* use EXTENT_LOCKED to prevent merging */
2624 set_extent_bits(block_group_cache, found_key.objectid,
2625 found_key.objectid + found_key.offset - 1,
2626 bit | EXTENT_LOCKED, GFP_NOFS);
2627 set_state_private(block_group_cache, found_key.objectid,
2628 (unsigned long)cache);
2631 btrfs_super_total_bytes(&info->super_copy))
2635 btrfs_free_path(path);