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 u64 header_transid =
1199 btrfs_header_generation(buf);
1200 if (header_transid == transid) {
1201 free_extent_buffer(buf);
1205 free_extent_buffer(buf);
1207 update_pinned_extents(root, bytenr, num_bytes, 1);
1209 set_extent_bits(&root->fs_info->pending_del,
1210 bytenr, bytenr + num_bytes - 1,
1211 EXTENT_LOCKED, GFP_NOFS);
1218 * remove an extent from the root, returns 0 on success
1220 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1221 *root, u64 bytenr, u64 num_bytes,
1222 u64 root_objectid, u64 ref_generation,
1223 u64 owner_objectid, u64 owner_offset, int pin,
1226 struct btrfs_path *path;
1227 struct btrfs_key key;
1228 struct btrfs_fs_info *info = root->fs_info;
1229 struct btrfs_root *extent_root = info->extent_root;
1230 struct extent_buffer *leaf;
1232 struct btrfs_extent_item *ei;
1235 key.objectid = bytenr;
1236 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1237 key.offset = num_bytes;
1239 path = btrfs_alloc_path();
1243 ret = lookup_extent_backref(trans, extent_root, path,
1244 bytenr, root_objectid,
1246 owner_objectid, owner_offset, 1);
1248 ret = btrfs_del_item(trans, extent_root, path);
1250 btrfs_print_leaf(extent_root, path->nodes[0]);
1252 printk("Unable to find ref byte nr %Lu root %Lu "
1253 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1254 root_objectid, ref_generation, owner_objectid,
1257 btrfs_release_path(extent_root, path);
1258 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1263 leaf = path->nodes[0];
1264 ei = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_item);
1266 refs = btrfs_extent_refs(leaf, ei);
1269 btrfs_set_extent_refs(leaf, ei, refs);
1270 btrfs_mark_buffer_dirty(leaf);
1277 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1283 /* block accounting for super block */
1284 super_used = btrfs_super_bytes_used(&info->super_copy);
1285 btrfs_set_super_bytes_used(&info->super_copy,
1286 super_used - num_bytes);
1288 /* block accounting for root item */
1289 root_used = btrfs_root_used(&root->root_item);
1290 btrfs_set_root_used(&root->root_item,
1291 root_used - num_bytes);
1293 ret = btrfs_del_item(trans, extent_root, path);
1297 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1301 btrfs_free_path(path);
1302 finish_current_insert(trans, extent_root);
1307 * find all the blocks marked as pending in the radix tree and remove
1308 * them from the extent map
1310 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1311 btrfs_root *extent_root)
1317 struct extent_map_tree *pending_del;
1318 struct extent_map_tree *pinned_extents;
1320 pending_del = &extent_root->fs_info->pending_del;
1321 pinned_extents = &extent_root->fs_info->pinned_extents;
1324 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1328 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1329 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1331 ret = __free_extent(trans, extent_root,
1332 start, end + 1 - start,
1333 extent_root->root_key.objectid,
1342 * remove an extent from the root, returns 0 on success
1344 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1345 *root, u64 bytenr, u64 num_bytes,
1346 u64 root_objectid, u64 ref_generation,
1347 u64 owner_objectid, u64 owner_offset, int pin)
1349 struct btrfs_root *extent_root = root->fs_info->extent_root;
1353 WARN_ON(num_bytes < root->sectorsize);
1354 if (!root->ref_cows)
1357 if (root == extent_root) {
1358 pin_down_bytes(root, bytenr, num_bytes, 1);
1361 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1362 ref_generation, owner_objectid, owner_offset,
1364 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1365 return ret ? ret : pending_ret;
1368 static u64 stripe_align(struct btrfs_root *root, u64 val)
1370 u64 mask = ((u64)root->stripesize - 1);
1371 u64 ret = (val + mask) & ~mask;
1376 * walks the btree of allocated extents and find a hole of a given size.
1377 * The key ins is changed to record the hole:
1378 * ins->objectid == block start
1379 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1380 * ins->offset == number of blocks
1381 * Any available blocks before search_start are skipped.
1383 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1384 struct btrfs_root *orig_root,
1385 u64 num_bytes, u64 empty_size,
1386 u64 search_start, u64 search_end,
1387 u64 hint_byte, struct btrfs_key *ins,
1388 u64 exclude_start, u64 exclude_nr,
1391 struct btrfs_path *path;
1392 struct btrfs_key key;
1398 u64 orig_search_start = search_start;
1400 struct extent_buffer *l;
1401 struct btrfs_root * root = orig_root->fs_info->extent_root;
1402 struct btrfs_fs_info *info = root->fs_info;
1403 u64 total_needed = num_bytes;
1405 struct btrfs_block_group_cache *block_group;
1410 WARN_ON(num_bytes < root->sectorsize);
1411 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1413 level = btrfs_header_level(root->node);
1415 if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1416 data = BTRFS_BLOCK_GROUP_MIXED;
1419 if (search_end == (u64)-1)
1420 search_end = btrfs_super_total_bytes(&info->super_copy);
1422 block_group = btrfs_lookup_block_group(info, hint_byte);
1424 hint_byte = search_start;
1425 block_group = btrfs_find_block_group(root, block_group,
1426 hint_byte, data, 1);
1428 block_group = btrfs_find_block_group(root,
1430 search_start, data, 1);
1433 total_needed += empty_size;
1434 path = btrfs_alloc_path();
1437 block_group = btrfs_lookup_block_group(info, search_start);
1439 block_group = btrfs_lookup_block_group(info,
1442 search_start = find_search_start(root, &block_group, search_start,
1443 total_needed, data);
1444 search_start = stripe_align(root, search_start);
1445 cached_start = search_start;
1446 btrfs_init_path(path);
1447 ins->objectid = search_start;
1452 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1455 ret = find_previous_extent(root, path);
1459 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1462 slot = path->slots[0];
1463 if (slot >= btrfs_header_nritems(l)) {
1464 ret = btrfs_next_leaf(root, path);
1470 search_start = max(search_start,
1471 block_group->key.objectid);
1473 aligned = stripe_align(root, search_start);
1474 ins->objectid = aligned;
1475 if (aligned >= search_end) {
1479 ins->offset = search_end - aligned;
1483 ins->objectid = stripe_align(root,
1484 last_byte > search_start ?
1485 last_byte : search_start);
1486 if (search_end <= ins->objectid) {
1490 ins->offset = search_end - ins->objectid;
1491 BUG_ON(ins->objectid >= search_end);
1494 btrfs_item_key_to_cpu(l, &key, slot);
1496 if (key.objectid >= search_start && key.objectid > last_byte &&
1498 if (last_byte < search_start)
1499 last_byte = search_start;
1500 aligned = stripe_align(root, last_byte);
1501 hole_size = key.objectid - aligned;
1502 if (key.objectid > aligned && hole_size >= num_bytes) {
1503 ins->objectid = aligned;
1504 ins->offset = hole_size;
1508 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1509 if (!start_found && btrfs_key_type(&key) ==
1510 BTRFS_BLOCK_GROUP_ITEM_KEY) {
1511 last_byte = key.objectid;
1519 last_byte = key.objectid + key.offset;
1521 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1522 last_byte >= block_group->key.objectid +
1523 block_group->key.offset) {
1524 btrfs_release_path(root, path);
1525 search_start = block_group->key.objectid +
1526 block_group->key.offset;
1534 /* we have to make sure we didn't find an extent that has already
1535 * been allocated by the map tree or the original allocation
1537 btrfs_release_path(root, path);
1538 BUG_ON(ins->objectid < search_start);
1540 if (ins->objectid + num_bytes >= search_end)
1542 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1543 ins->objectid + num_bytes > block_group->
1544 key.objectid + block_group->key.offset) {
1545 search_start = block_group->key.objectid +
1546 block_group->key.offset;
1549 if (test_range_bit(&info->extent_ins, ins->objectid,
1550 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1551 search_start = ins->objectid + num_bytes;
1554 if (test_range_bit(&info->pinned_extents, ins->objectid,
1555 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1556 search_start = ins->objectid + num_bytes;
1559 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1560 ins->objectid < exclude_start + exclude_nr)) {
1561 search_start = exclude_start + exclude_nr;
1565 block_group = btrfs_lookup_block_group(info, ins->objectid);
1567 trans->block_group = block_group;
1569 ins->offset = num_bytes;
1570 btrfs_free_path(path);
1574 if (search_start + num_bytes >= search_end) {
1576 search_start = orig_search_start;
1583 total_needed -= empty_size;
1585 data = BTRFS_BLOCK_GROUP_MIXED;
1589 block_group = btrfs_lookup_block_group(info, search_start);
1591 block_group = btrfs_find_block_group(root, block_group,
1592 search_start, data, 0);
1596 btrfs_release_path(root, path);
1597 btrfs_free_path(path);
1601 * finds a free extent and does all the dirty work required for allocation
1602 * returns the key for the extent through ins, and a tree buffer for
1603 * the first block of the extent through buf.
1605 * returns 0 if everything worked, non-zero otherwise.
1607 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1608 struct btrfs_root *root,
1609 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1610 u64 owner, u64 owner_offset,
1611 u64 empty_size, u64 hint_byte,
1612 u64 search_end, struct btrfs_key *ins, int data)
1616 u64 super_used, root_used;
1617 u64 search_start = 0;
1619 struct btrfs_fs_info *info = root->fs_info;
1620 struct btrfs_root *extent_root = info->extent_root;
1621 struct btrfs_extent_item extent_item;
1622 struct btrfs_path *path;
1624 btrfs_set_stack_extent_refs(&extent_item, 1);
1626 new_hint = max(hint_byte, root->fs_info->alloc_start);
1627 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1628 hint_byte = new_hint;
1630 WARN_ON(num_bytes < root->sectorsize);
1631 ret = find_free_extent(trans, root, num_bytes, empty_size,
1632 search_start, search_end, hint_byte, ins,
1633 trans->alloc_exclude_start,
1634 trans->alloc_exclude_nr, data);
1639 /* block accounting for super block */
1640 super_used = btrfs_super_bytes_used(&info->super_copy);
1641 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1643 /* block accounting for root item */
1644 root_used = btrfs_root_used(&root->root_item);
1645 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1647 clear_extent_dirty(&root->fs_info->free_space_cache,
1648 ins->objectid, ins->objectid + ins->offset - 1,
1651 if (root == extent_root) {
1652 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1653 ins->objectid + ins->offset - 1,
1654 EXTENT_LOCKED, GFP_NOFS);
1659 WARN_ON(trans->alloc_exclude_nr);
1660 trans->alloc_exclude_start = ins->objectid;
1661 trans->alloc_exclude_nr = ins->offset;
1662 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1663 sizeof(extent_item));
1665 trans->alloc_exclude_start = 0;
1666 trans->alloc_exclude_nr = 0;
1669 path = btrfs_alloc_path();
1671 ret = btrfs_insert_extent_backref(trans, extent_root, path,
1672 ins->objectid, root_objectid,
1673 ref_generation, owner, owner_offset);
1676 btrfs_free_path(path);
1677 finish_current_insert(trans, extent_root);
1678 pending_ret = del_pending_extents(trans, extent_root);
1688 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1695 * helper function to allocate a block for a given tree
1696 * returns the tree buffer or NULL.
1698 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1699 struct btrfs_root *root,
1701 u64 root_objectid, u64 hint,
1707 ref_generation = trans->transid;
1712 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1713 ref_generation, 0, 0, hint, empty_size);
1717 * helper function to allocate a block for a given tree
1718 * returns the tree buffer or NULL.
1720 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1721 struct btrfs_root *root,
1730 struct btrfs_key ins;
1732 struct extent_buffer *buf;
1734 ret = btrfs_alloc_extent(trans, root, blocksize,
1735 root_objectid, ref_generation,
1736 level, first_objectid, empty_size, hint,
1740 return ERR_PTR(ret);
1742 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1744 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1745 root->root_key.objectid, ref_generation,
1747 return ERR_PTR(-ENOMEM);
1749 btrfs_set_buffer_uptodate(buf);
1750 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1751 buf->start + buf->len - 1, GFP_NOFS);
1752 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1753 buf->start, buf->start + buf->len - 1,
1754 EXTENT_CSUM, GFP_NOFS);
1755 buf->flags |= EXTENT_CSUM;
1756 btrfs_set_buffer_defrag(buf);
1757 trans->blocks_used++;
1761 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1762 struct btrfs_root *root,
1763 struct extent_buffer *leaf)
1766 u64 leaf_generation;
1767 struct btrfs_key key;
1768 struct btrfs_file_extent_item *fi;
1773 BUG_ON(!btrfs_is_leaf(leaf));
1774 nritems = btrfs_header_nritems(leaf);
1775 leaf_owner = btrfs_header_owner(leaf);
1776 leaf_generation = btrfs_header_generation(leaf);
1778 for (i = 0; i < nritems; i++) {
1781 btrfs_item_key_to_cpu(leaf, &key, i);
1782 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1784 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1785 if (btrfs_file_extent_type(leaf, fi) ==
1786 BTRFS_FILE_EXTENT_INLINE)
1789 * FIXME make sure to insert a trans record that
1790 * repeats the snapshot del on crash
1792 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1793 if (disk_bytenr == 0)
1795 ret = btrfs_free_extent(trans, root, disk_bytenr,
1796 btrfs_file_extent_disk_num_bytes(leaf, fi),
1797 leaf_owner, leaf_generation,
1798 key.objectid, key.offset, 0);
1804 static void noinline reada_walk_down(struct btrfs_root *root,
1805 struct extent_buffer *node)
1815 nritems = btrfs_header_nritems(node);
1816 level = btrfs_header_level(node);
1817 for (i = 0; i < nritems; i++) {
1818 bytenr = btrfs_node_blockptr(node, i);
1819 blocksize = btrfs_level_size(root, level - 1);
1820 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1824 mutex_unlock(&root->fs_info->fs_mutex);
1825 ret = readahead_tree_block(root, bytenr, blocksize);
1827 mutex_lock(&root->fs_info->fs_mutex);
1834 * helper function for drop_snapshot, this walks down the tree dropping ref
1835 * counts as it goes.
1837 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
1838 struct btrfs_root *root,
1839 struct btrfs_path *path, int *level)
1844 struct extent_buffer *next;
1845 struct extent_buffer *cur;
1846 struct extent_buffer *parent;
1851 WARN_ON(*level < 0);
1852 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1853 ret = lookup_extent_ref(trans, root,
1854 path->nodes[*level]->start,
1855 path->nodes[*level]->len, &refs);
1861 * walk down to the last node level and free all the leaves
1863 while(*level >= 0) {
1864 WARN_ON(*level < 0);
1865 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1866 cur = path->nodes[*level];
1868 if (*level > 0 && path->slots[*level] == 0)
1869 reada_walk_down(root, cur);
1871 if (btrfs_header_level(cur) != *level)
1874 if (path->slots[*level] >=
1875 btrfs_header_nritems(cur))
1878 ret = drop_leaf_ref(trans, root, cur);
1882 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1883 blocksize = btrfs_level_size(root, *level - 1);
1884 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1887 parent = path->nodes[*level];
1888 root_owner = btrfs_header_owner(parent);
1889 root_gen = btrfs_header_generation(parent);
1890 path->slots[*level]++;
1891 ret = btrfs_free_extent(trans, root, bytenr,
1892 blocksize, root_owner,
1897 next = btrfs_find_tree_block(root, bytenr, blocksize);
1898 if (!next || !btrfs_buffer_uptodate(next)) {
1899 free_extent_buffer(next);
1900 mutex_unlock(&root->fs_info->fs_mutex);
1901 next = read_tree_block(root, bytenr, blocksize);
1902 mutex_lock(&root->fs_info->fs_mutex);
1904 /* we dropped the lock, check one more time */
1905 ret = lookup_extent_ref(trans, root, bytenr,
1909 parent = path->nodes[*level];
1910 root_owner = btrfs_header_owner(parent);
1911 root_gen = btrfs_header_generation(parent);
1913 path->slots[*level]++;
1914 free_extent_buffer(next);
1915 ret = btrfs_free_extent(trans, root, bytenr,
1923 WARN_ON(*level <= 0);
1924 if (path->nodes[*level-1])
1925 free_extent_buffer(path->nodes[*level-1]);
1926 path->nodes[*level-1] = next;
1927 *level = btrfs_header_level(next);
1928 path->slots[*level] = 0;
1931 WARN_ON(*level < 0);
1932 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1934 if (path->nodes[*level] == root->node) {
1935 root_owner = root->root_key.objectid;
1936 parent = path->nodes[*level];
1938 parent = path->nodes[*level + 1];
1939 root_owner = btrfs_header_owner(parent);
1942 root_gen = btrfs_header_generation(parent);
1943 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1944 path->nodes[*level]->len,
1945 root_owner, root_gen, 0, 0, 1);
1946 free_extent_buffer(path->nodes[*level]);
1947 path->nodes[*level] = NULL;
1954 * helper for dropping snapshots. This walks back up the tree in the path
1955 * to find the first node higher up where we haven't yet gone through
1958 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
1959 struct btrfs_root *root,
1960 struct btrfs_path *path, int *level)
1964 struct btrfs_root_item *root_item = &root->root_item;
1969 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1970 slot = path->slots[i];
1971 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1972 struct extent_buffer *node;
1973 struct btrfs_disk_key disk_key;
1974 node = path->nodes[i];
1977 WARN_ON(*level == 0);
1978 btrfs_node_key(node, &disk_key, path->slots[i]);
1979 memcpy(&root_item->drop_progress,
1980 &disk_key, sizeof(disk_key));
1981 root_item->drop_level = i;
1984 if (path->nodes[*level] == root->node) {
1985 root_owner = root->root_key.objectid;
1987 btrfs_header_generation(path->nodes[*level]);
1989 struct extent_buffer *node;
1990 node = path->nodes[*level + 1];
1991 root_owner = btrfs_header_owner(node);
1992 root_gen = btrfs_header_generation(node);
1994 ret = btrfs_free_extent(trans, root,
1995 path->nodes[*level]->start,
1996 path->nodes[*level]->len,
1997 root_owner, root_gen, 0, 0, 1);
1999 free_extent_buffer(path->nodes[*level]);
2000 path->nodes[*level] = NULL;
2008 * drop the reference count on the tree rooted at 'snap'. This traverses
2009 * the tree freeing any blocks that have a ref count of zero after being
2012 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2018 struct btrfs_path *path;
2021 struct btrfs_root_item *root_item = &root->root_item;
2023 path = btrfs_alloc_path();
2026 level = btrfs_header_level(root->node);
2028 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2029 path->nodes[level] = root->node;
2030 extent_buffer_get(root->node);
2031 path->slots[level] = 0;
2033 struct btrfs_key key;
2034 struct btrfs_disk_key found_key;
2035 struct extent_buffer *node;
2037 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2038 level = root_item->drop_level;
2039 path->lowest_level = level;
2040 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2045 node = path->nodes[level];
2046 btrfs_node_key(node, &found_key, path->slots[level]);
2047 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2048 sizeof(found_key)));
2051 wret = walk_down_tree(trans, root, path, &level);
2057 wret = walk_up_tree(trans, root, path, &level);
2065 for (i = 0; i <= orig_level; i++) {
2066 if (path->nodes[i]) {
2067 free_extent_buffer(path->nodes[i]);
2068 path->nodes[i] = NULL;
2072 btrfs_free_path(path);
2076 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2083 ret = find_first_extent_bit(&info->block_group_cache, 0,
2084 &start, &end, (unsigned int)-1);
2087 ret = get_state_private(&info->block_group_cache, start, &ptr);
2089 kfree((void *)(unsigned long)ptr);
2090 clear_extent_bits(&info->block_group_cache, start,
2091 end, (unsigned int)-1, GFP_NOFS);
2094 ret = find_first_extent_bit(&info->free_space_cache, 0,
2095 &start, &end, EXTENT_DIRTY);
2098 clear_extent_dirty(&info->free_space_cache, start,
2104 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2110 u64 existing_delalloc;
2111 unsigned long last_index;
2114 struct btrfs_root *root = BTRFS_I(inode)->root;
2115 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2116 struct file_ra_state *ra;
2118 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2120 mutex_lock(&inode->i_mutex);
2121 i = start >> PAGE_CACHE_SHIFT;
2122 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2124 file_ra_state_init(ra, inode->i_mapping);
2125 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2128 for (; i <= last_index; i++) {
2129 page = grab_cache_page(inode->i_mapping, i);
2132 if (!PageUptodate(page)) {
2133 btrfs_readpage(NULL, page);
2135 if (!PageUptodate(page)) {
2137 page_cache_release(page);
2141 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2142 page_end = page_start + PAGE_CACHE_SIZE - 1;
2144 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2146 delalloc_start = page_start;
2148 count_range_bits(&BTRFS_I(inode)->extent_tree,
2149 &delalloc_start, page_end,
2150 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2152 set_extent_delalloc(em_tree, page_start,
2153 page_end, GFP_NOFS);
2155 spin_lock(&root->fs_info->delalloc_lock);
2156 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2158 spin_unlock(&root->fs_info->delalloc_lock);
2160 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2161 set_page_dirty(page);
2163 page_cache_release(page);
2167 mutex_unlock(&inode->i_mutex);
2172 * note, this releases the path
2174 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2175 struct btrfs_path *path,
2176 struct btrfs_key *extent_key)
2178 struct inode *inode;
2179 struct btrfs_root *found_root;
2180 struct btrfs_key *root_location;
2181 struct btrfs_extent_ref *ref;
2188 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2189 struct btrfs_extent_ref);
2190 ref_root = btrfs_ref_root(path->nodes[0], ref);
2191 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2192 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2193 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2194 btrfs_release_path(extent_root, path);
2196 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2197 root_location->objectid = ref_root;
2199 root_location->offset = 0;
2201 root_location->offset = (u64)-1;
2202 root_location->type = BTRFS_ROOT_ITEM_KEY;
2204 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2206 BUG_ON(!found_root);
2207 kfree(root_location);
2209 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2210 mutex_unlock(&extent_root->fs_info->fs_mutex);
2211 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2212 ref_objectid, found_root);
2213 if (inode->i_state & I_NEW) {
2214 /* the inode and parent dir are two different roots */
2215 BTRFS_I(inode)->root = found_root;
2216 BTRFS_I(inode)->location.objectid = ref_objectid;
2217 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2218 BTRFS_I(inode)->location.offset = 0;
2219 btrfs_read_locked_inode(inode);
2220 unlock_new_inode(inode);
2223 /* this can happen if the reference is not against
2224 * the latest version of the tree root
2226 if (is_bad_inode(inode)) {
2227 mutex_lock(&extent_root->fs_info->fs_mutex);
2230 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2231 /* FIXME, data=ordered will help get rid of this */
2232 filemap_fdatawrite(inode->i_mapping);
2234 mutex_lock(&extent_root->fs_info->fs_mutex);
2236 struct btrfs_trans_handle *trans;
2237 struct btrfs_key found_key;
2238 struct extent_buffer *eb;
2242 trans = btrfs_start_transaction(found_root, 1);
2243 eb = read_tree_block(found_root, extent_key->objectid,
2244 extent_key->offset);
2245 level = btrfs_header_level(eb);
2248 btrfs_item_key_to_cpu(eb, &found_key, 0);
2250 btrfs_node_key_to_cpu(eb, &found_key, 0);
2252 free_extent_buffer(eb);
2254 path->lowest_level = level;
2256 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2258 path->lowest_level = 0;
2259 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2260 if (!path->nodes[i])
2262 free_extent_buffer(path->nodes[i]);
2263 path->nodes[i] = NULL;
2265 btrfs_release_path(found_root, path);
2266 btrfs_end_transaction(trans, found_root);
2273 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2274 struct btrfs_path *path,
2275 struct btrfs_key *extent_key)
2277 struct btrfs_key key;
2278 struct btrfs_key found_key;
2279 struct extent_buffer *leaf;
2284 key.objectid = extent_key->objectid;
2285 key.type = BTRFS_EXTENT_REF_KEY;
2289 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2297 leaf = path->nodes[0];
2298 nritems = btrfs_header_nritems(leaf);
2299 if (path->slots[0] == nritems)
2302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2303 if (found_key.objectid != extent_key->objectid)
2306 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2309 key.offset = found_key.offset + 1;
2310 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2312 ret = relocate_one_reference(extent_root, path, extent_key);
2318 btrfs_release_path(extent_root, path);
2322 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2324 struct btrfs_trans_handle *trans;
2325 struct btrfs_root *tree_root = root->fs_info->tree_root;
2326 struct btrfs_path *path;
2329 struct btrfs_fs_info *info = root->fs_info;
2330 struct extent_map_tree *block_group_cache;
2331 struct btrfs_key key;
2332 struct btrfs_key found_key;
2333 struct extent_buffer *leaf;
2338 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2339 block_group_cache = &info->block_group_cache;
2340 path = btrfs_alloc_path();
2341 root = root->fs_info->extent_root;
2346 key.objectid = new_size;
2349 cur_byte = key.objectid;
2351 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2355 ret = find_previous_extent(root, path);
2359 leaf = path->nodes[0];
2360 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2361 if (found_key.objectid + found_key.offset > new_size) {
2362 cur_byte = found_key.objectid;
2363 key.objectid = cur_byte;
2366 btrfs_release_path(root, path);
2369 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2373 leaf = path->nodes[0];
2374 nritems = btrfs_header_nritems(leaf);
2376 if (path->slots[0] >= nritems) {
2377 ret = btrfs_next_leaf(root, path);
2384 leaf = path->nodes[0];
2385 nritems = btrfs_header_nritems(leaf);
2388 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2390 if (progress && need_resched()) {
2391 memcpy(&key, &found_key, sizeof(key));
2392 mutex_unlock(&root->fs_info->fs_mutex);
2394 mutex_lock(&root->fs_info->fs_mutex);
2395 btrfs_release_path(root, path);
2396 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2402 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2403 found_key.objectid + found_key.offset <= cur_byte) {
2409 cur_byte = found_key.objectid + found_key.offset;
2410 key.objectid = cur_byte;
2411 btrfs_release_path(root, path);
2412 ret = relocate_one_extent(root, path, &found_key);
2415 btrfs_release_path(root, path);
2417 if (total_found > 0) {
2418 trans = btrfs_start_transaction(tree_root, 1);
2419 btrfs_commit_transaction(trans, tree_root);
2421 mutex_unlock(&root->fs_info->fs_mutex);
2422 btrfs_clean_old_snapshots(tree_root);
2423 mutex_lock(&root->fs_info->fs_mutex);
2425 trans = btrfs_start_transaction(tree_root, 1);
2426 btrfs_commit_transaction(trans, tree_root);
2430 trans = btrfs_start_transaction(root, 1);
2431 key.objectid = new_size;
2437 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2441 leaf = path->nodes[0];
2442 nritems = btrfs_header_nritems(leaf);
2444 if (path->slots[0] >= nritems) {
2445 ret = btrfs_next_leaf(root, path);
2452 leaf = path->nodes[0];
2453 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2456 * btrfs_next_leaf doesn't cow buffers, we have to
2457 * do the search again
2459 memcpy(&key, &found_key, sizeof(key));
2460 btrfs_release_path(root, path);
2464 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2465 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2466 printk("shrinker found key %Lu %u %Lu\n",
2467 found_key.objectid, found_key.type,
2472 ret = get_state_private(&info->block_group_cache,
2473 found_key.objectid, &ptr);
2475 kfree((void *)(unsigned long)ptr);
2477 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2478 found_key.objectid + found_key.offset - 1,
2479 (unsigned int)-1, GFP_NOFS);
2481 key.objectid = found_key.objectid + 1;
2482 btrfs_del_item(trans, root, path);
2483 btrfs_release_path(root, path);
2485 if (need_resched()) {
2486 mutex_unlock(&root->fs_info->fs_mutex);
2488 mutex_lock(&root->fs_info->fs_mutex);
2491 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2493 btrfs_commit_transaction(trans, root);
2495 btrfs_free_path(path);
2499 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2500 struct btrfs_root *root, u64 new_size)
2502 struct btrfs_path *path;
2507 struct btrfs_block_group_cache *cache;
2508 struct btrfs_block_group_item *item;
2509 struct btrfs_fs_info *info = root->fs_info;
2510 struct extent_map_tree *block_group_cache;
2511 struct btrfs_key key;
2512 struct extent_buffer *leaf;
2516 old_size = btrfs_super_total_bytes(&info->super_copy);
2517 block_group_cache = &info->block_group_cache;
2519 root = info->extent_root;
2521 cache = btrfs_lookup_block_group(root->fs_info, old_size - 1);
2523 cur_byte = cache->key.objectid + cache->key.offset;
2524 if (cur_byte >= new_size)
2527 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2528 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2530 path = btrfs_alloc_path();
2534 while(cur_byte < new_size) {
2535 key.objectid = cur_byte;
2536 ret = btrfs_insert_empty_item(trans, root, path, &key,
2537 sizeof(struct btrfs_block_group_item));
2539 leaf = path->nodes[0];
2540 item = btrfs_item_ptr(leaf, path->slots[0],
2541 struct btrfs_block_group_item);
2543 btrfs_set_disk_block_group_used(leaf, item, 0);
2544 div_long_long_rem(nr, 3, &rem);
2546 btrfs_set_disk_block_group_flags(leaf, item,
2547 BTRFS_BLOCK_GROUP_DATA);
2549 btrfs_set_disk_block_group_flags(leaf, item, 0);
2553 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2556 read_extent_buffer(leaf, &cache->item, (unsigned long)item,
2557 sizeof(cache->item));
2559 memcpy(&cache->key, &key, sizeof(key));
2562 cur_byte = key.objectid + key.offset;
2563 btrfs_release_path(root, path);
2565 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2566 bit = BLOCK_GROUP_DATA;
2567 cache->data = BTRFS_BLOCK_GROUP_DATA;
2569 bit = BLOCK_GROUP_METADATA;
2573 /* use EXTENT_LOCKED to prevent merging */
2574 set_extent_bits(block_group_cache, key.objectid,
2575 key.objectid + key.offset - 1,
2576 bit | EXTENT_LOCKED, GFP_NOFS);
2577 set_state_private(block_group_cache, key.objectid,
2578 (unsigned long)cache);
2580 btrfs_free_path(path);
2582 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2586 int btrfs_read_block_groups(struct btrfs_root *root)
2588 struct btrfs_path *path;
2592 struct btrfs_block_group_cache *cache;
2593 struct btrfs_fs_info *info = root->fs_info;
2594 struct extent_map_tree *block_group_cache;
2595 struct btrfs_key key;
2596 struct btrfs_key found_key;
2597 struct extent_buffer *leaf;
2599 block_group_cache = &info->block_group_cache;
2601 root = info->extent_root;
2603 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2604 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2606 path = btrfs_alloc_path();
2611 ret = btrfs_search_slot(NULL, info->extent_root,
2617 leaf = path->nodes[0];
2618 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2619 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2625 read_extent_buffer(leaf, &cache->item,
2626 btrfs_item_ptr_offset(leaf, path->slots[0]),
2627 sizeof(cache->item));
2628 memcpy(&cache->key, &found_key, sizeof(found_key));
2631 key.objectid = found_key.objectid + found_key.offset;
2632 btrfs_release_path(root, path);
2634 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2635 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2636 cache->data = BTRFS_BLOCK_GROUP_MIXED;
2637 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2638 bit = BLOCK_GROUP_DATA;
2639 cache->data = BTRFS_BLOCK_GROUP_DATA;
2641 bit = BLOCK_GROUP_METADATA;
2645 /* use EXTENT_LOCKED to prevent merging */
2646 set_extent_bits(block_group_cache, found_key.objectid,
2647 found_key.objectid + found_key.offset - 1,
2648 bit | EXTENT_LOCKED, GFP_NOFS);
2649 set_state_private(block_group_cache, found_key.objectid,
2650 (unsigned long)cache);
2653 btrfs_super_total_bytes(&info->super_copy))
2657 btrfs_free_path(path);