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/module.h>
22 #include "print-tree.h"
23 #include "transaction.h"
25 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
26 *orig_root, u64 num_blocks, u64 search_start,
27 u64 search_end, u64 hint_block,
28 struct btrfs_key *ins, u64 exclude_start,
29 u64 exclude_nr, int data);
30 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
31 btrfs_root *extent_root);
32 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
33 btrfs_root *extent_root);
35 static void reada_extent_leaves(struct btrfs_root *root,
36 struct btrfs_path *path, u64 limit)
38 struct btrfs_node *node;
48 node = btrfs_buffer_node(path->nodes[1]);
49 slot = path->slots[1] + 1;
50 nritems = btrfs_header_nritems(&node->header);
51 for (i = slot; i < nritems && i < slot + 8; i++) {
52 item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
53 if (item_objectid > limit)
55 blocknr = btrfs_node_blockptr(node, i);
56 ret = readahead_tree_block(root, blocknr);
62 static int cache_block_group(struct btrfs_root *root,
63 struct btrfs_block_group_cache *block_group)
65 struct btrfs_path *path;
68 struct btrfs_leaf *leaf;
69 struct radix_tree_root *extent_radix;
77 root = root->fs_info->extent_root;
78 extent_radix = &root->fs_info->extent_map_radix;
80 if (block_group->cached)
82 if (block_group->data)
84 path = btrfs_alloc_path();
87 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 if (ret && path->slots[0] > 0)
96 limit = block_group->key.objectid + block_group->key.offset;
97 reada_extent_leaves(root, path, limit);
99 leaf = btrfs_buffer_leaf(path->nodes[0]);
100 slot = path->slots[0];
101 if (slot >= btrfs_header_nritems(&leaf->header)) {
102 reada_extent_leaves(root, path, limit);
103 ret = btrfs_next_leaf(root, path);
110 hole_size = block_group->key.objectid +
111 block_group->key.offset - last;
113 last = block_group->key.objectid;
114 hole_size = block_group->key.offset;
116 for (i = 0; i < hole_size; i++) {
117 set_radix_bit(extent_radix,
123 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
124 if (key.objectid >= block_group->key.objectid +
125 block_group->key.offset) {
127 hole_size = block_group->key.objectid +
128 block_group->key.offset - last;
130 last = block_group->key.objectid;
131 hole_size = block_group->key.offset;
133 for (i = 0; i < hole_size; i++) {
134 set_radix_bit(extent_radix, last + i);
138 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
140 last = key.objectid + key.offset;
143 hole_size = key.objectid - last;
144 for (i = 0; i < hole_size; i++) {
145 set_radix_bit(extent_radix, last + i);
147 last = key.objectid + key.offset;
153 block_group->cached = 1;
155 btrfs_free_path(path);
159 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
163 struct btrfs_block_group_cache *block_group;
166 ret = radix_tree_gang_lookup(&info->block_group_radix,
167 (void **)&block_group,
170 if (block_group->key.objectid <= blocknr && blocknr <=
171 block_group->key.objectid + block_group->key.offset)
174 ret = radix_tree_gang_lookup(&info->block_group_data_radix,
175 (void **)&block_group,
178 if (block_group->key.objectid <= blocknr && blocknr <=
179 block_group->key.objectid + block_group->key.offset)
185 static u64 leaf_range(struct btrfs_root *root)
187 u64 size = BTRFS_LEAF_DATA_SIZE(root);
188 do_div(size, sizeof(struct btrfs_extent_item) +
189 sizeof(struct btrfs_item));
193 static u64 find_search_start(struct btrfs_root *root,
194 struct btrfs_block_group_cache **cache_ret,
195 u64 search_start, int num)
197 unsigned long gang[8];
199 struct btrfs_block_group_cache *cache = *cache_ret;
200 u64 last = max(search_start, cache->key.objectid);
205 last = max(last, cache->last_prealloc);
208 ret = cache_block_group(root, cache);
212 ret = find_first_radix_bit(&root->fs_info->extent_map_radix,
213 gang, last, ARRAY_SIZE(gang));
216 last = gang[ret-1] + 1;
218 if (ret != ARRAY_SIZE(gang)) {
221 if (gang[ret-1] - gang[0] > leaf_range(root)) {
225 if (gang[0] >= cache->key.objectid + cache->key.offset) {
231 return max(cache->last_alloc, search_start);
234 cache = btrfs_lookup_block_group(root->fs_info,
235 last + cache->key.offset - 1);
237 return max((*cache_ret)->last_alloc, search_start);
239 cache = btrfs_find_block_group(root, cache,
240 last + cache->key.offset - 1, 0, 0);
245 static u64 div_factor(u64 num, int factor)
252 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
253 struct btrfs_block_group_cache
254 *hint, u64 search_start,
257 struct btrfs_block_group_cache *cache[8];
258 struct btrfs_block_group_cache *found_group = NULL;
259 struct btrfs_fs_info *info = root->fs_info;
260 struct radix_tree_root *radix;
261 struct radix_tree_root *swap_radix;
275 radix = &info->block_group_data_radix;
276 swap_radix = &info->block_group_radix;
278 radix = &info->block_group_radix;
279 swap_radix = &info->block_group_data_radix;
283 struct btrfs_block_group_cache *shint;
284 shint = btrfs_lookup_block_group(info, search_start);
285 if (shint->data == data) {
286 used = btrfs_block_group_used(&shint->item);
287 if (used + shint->pinned <
288 div_factor(shint->key.offset, factor)) {
293 if (hint && hint->data == data) {
294 used = btrfs_block_group_used(&hint->item);
295 if (used + hint->pinned <
296 div_factor(hint->key.offset, factor)) {
299 if (used >= div_factor(hint->key.offset, 8)) {
300 radix_tree_tag_clear(radix,
302 hint->key.offset - 1,
303 BTRFS_BLOCK_GROUP_AVAIL);
305 last = hint->key.offset * 3;
306 if (hint->key.objectid >= last)
307 last = max(search_start + hint->key.offset - 1,
308 hint->key.objectid - last);
310 last = hint->key.objectid + hint->key.offset;
314 hint_last = max(hint->key.objectid, search_start);
316 hint_last = search_start;
321 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
322 last, ARRAY_SIZE(cache),
323 BTRFS_BLOCK_GROUP_AVAIL);
326 for (i = 0; i < ret; i++) {
327 last = cache[i]->key.objectid +
328 cache[i]->key.offset;
329 used = btrfs_block_group_used(&cache[i]->item);
330 if (used + cache[i]->pinned <
331 div_factor(cache[i]->key.offset, factor)) {
332 found_group = cache[i];
335 if (used >= div_factor(cache[i]->key.offset, 8)) {
336 radix_tree_tag_clear(radix,
337 cache[i]->key.objectid +
338 cache[i]->key.offset - 1,
339 BTRFS_BLOCK_GROUP_AVAIL);
347 ret = radix_tree_gang_lookup(radix, (void **)cache,
348 last, ARRAY_SIZE(cache));
351 for (i = 0; i < ret; i++) {
352 last = cache[i]->key.objectid +
353 cache[i]->key.offset;
354 used = btrfs_block_group_used(&cache[i]->item);
355 if (used + cache[i]->pinned < cache[i]->key.offset) {
356 found_group = cache[i];
359 if (used >= cache[i]->key.offset) {
360 radix_tree_tag_clear(radix,
361 cache[i]->key.objectid +
362 cache[i]->key.offset - 1,
363 BTRFS_BLOCK_GROUP_AVAIL);
374 struct radix_tree_root *tmp = radix;
382 ret = radix_tree_gang_lookup(radix,
383 (void **)&found_group, 0, 1);
385 ret = radix_tree_gang_lookup(swap_radix,
386 (void **)&found_group,
395 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
396 struct btrfs_root *root,
397 u64 blocknr, u64 num_blocks)
399 struct btrfs_path *path;
401 struct btrfs_key key;
402 struct btrfs_leaf *l;
403 struct btrfs_extent_item *item;
404 struct btrfs_key ins;
407 path = btrfs_alloc_path();
410 ret = find_free_extent(trans, root->fs_info->extent_root, 0, 0,
411 (u64)-1, 0, &ins, 0, 0, 0);
413 btrfs_free_path(path);
416 key.objectid = blocknr;
418 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
419 key.offset = num_blocks;
420 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
428 l = btrfs_buffer_leaf(path->nodes[0]);
429 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
430 refs = btrfs_extent_refs(item);
431 btrfs_set_extent_refs(item, refs + 1);
432 btrfs_mark_buffer_dirty(path->nodes[0]);
434 btrfs_release_path(root->fs_info->extent_root, path);
435 btrfs_free_path(path);
436 finish_current_insert(trans, root->fs_info->extent_root);
437 del_pending_extents(trans, root->fs_info->extent_root);
441 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
442 struct btrfs_root *root, u64 blocknr,
443 u64 num_blocks, u32 *refs)
445 struct btrfs_path *path;
447 struct btrfs_key key;
448 struct btrfs_leaf *l;
449 struct btrfs_extent_item *item;
451 path = btrfs_alloc_path();
452 key.objectid = blocknr;
453 key.offset = num_blocks;
455 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
456 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
462 l = btrfs_buffer_leaf(path->nodes[0]);
463 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
464 *refs = btrfs_extent_refs(item);
466 btrfs_free_path(path);
470 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
471 struct btrfs_root *root)
473 return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
476 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
477 struct buffer_head *buf)
480 struct btrfs_node *buf_node;
481 struct btrfs_leaf *buf_leaf;
482 struct btrfs_disk_key *key;
483 struct btrfs_file_extent_item *fi;
492 buf_node = btrfs_buffer_node(buf);
493 leaf = btrfs_is_leaf(buf_node);
494 buf_leaf = btrfs_buffer_leaf(buf);
495 for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
498 key = &buf_leaf->items[i].key;
499 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
501 fi = btrfs_item_ptr(buf_leaf, i,
502 struct btrfs_file_extent_item);
503 if (btrfs_file_extent_type(fi) ==
504 BTRFS_FILE_EXTENT_INLINE)
506 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
507 if (disk_blocknr == 0)
509 ret = btrfs_inc_extent_ref(trans, root, disk_blocknr,
510 btrfs_file_extent_disk_num_blocks(fi));
516 blocknr = btrfs_node_blockptr(buf_node, i);
517 ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
526 for (i =0; i < faili; i++) {
529 key = &buf_leaf->items[i].key;
530 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
532 fi = btrfs_item_ptr(buf_leaf, i,
533 struct btrfs_file_extent_item);
534 if (btrfs_file_extent_type(fi) ==
535 BTRFS_FILE_EXTENT_INLINE)
537 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
538 if (disk_blocknr == 0)
540 err = btrfs_free_extent(trans, root, disk_blocknr,
541 btrfs_file_extent_disk_num_blocks(fi), 0);
544 blocknr = btrfs_node_blockptr(buf_node, i);
545 err = btrfs_free_extent(trans, root, blocknr, 1, 0);
552 static int write_one_cache_group(struct btrfs_trans_handle *trans,
553 struct btrfs_root *root,
554 struct btrfs_path *path,
555 struct btrfs_block_group_cache *cache)
559 struct btrfs_root *extent_root = root->fs_info->extent_root;
560 struct btrfs_block_group_item *bi;
561 struct btrfs_key ins;
563 ret = find_free_extent(trans, extent_root, 0, 0, (u64)-1, 0, &ins,
565 /* FIXME, set bit to recalc cache groups on next mount */
568 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
572 bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
573 struct btrfs_block_group_item);
574 memcpy(bi, &cache->item, sizeof(*bi));
575 mark_buffer_dirty(path->nodes[0]);
576 btrfs_release_path(extent_root, path);
578 finish_current_insert(trans, extent_root);
579 pending_ret = del_pending_extents(trans, extent_root);
585 cache->last_alloc = cache->first_free;
590 static int write_dirty_block_radix(struct btrfs_trans_handle *trans,
591 struct btrfs_root *root,
592 struct radix_tree_root *radix)
594 struct btrfs_block_group_cache *cache[8];
599 struct btrfs_path *path;
600 unsigned long off = 0;
602 path = btrfs_alloc_path();
607 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
608 off, ARRAY_SIZE(cache),
609 BTRFS_BLOCK_GROUP_DIRTY);
612 for (i = 0; i < ret; i++) {
613 err = write_one_cache_group(trans, root,
616 * if we fail to write the cache group, we want
617 * to keep it marked dirty in hopes that a later
622 off = cache[i]->key.objectid +
623 cache[i]->key.offset;
627 radix_tree_tag_clear(radix, cache[i]->key.objectid +
628 cache[i]->key.offset - 1,
629 BTRFS_BLOCK_GROUP_DIRTY);
632 btrfs_free_path(path);
636 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
637 struct btrfs_root *root)
641 ret = write_dirty_block_radix(trans, root,
642 &root->fs_info->block_group_radix);
643 ret2 = write_dirty_block_radix(trans, root,
644 &root->fs_info->block_group_data_radix);
652 static int update_block_group(struct btrfs_trans_handle *trans,
653 struct btrfs_root *root,
654 u64 blocknr, u64 num, int alloc, int mark_free,
657 struct btrfs_block_group_cache *cache;
658 struct btrfs_fs_info *info = root->fs_info;
666 cache = btrfs_lookup_block_group(info, blocknr);
670 block_in_group = blocknr - cache->key.objectid;
671 WARN_ON(block_in_group > cache->key.offset);
672 radix_tree_tag_set(cache->radix, cache->key.objectid +
673 cache->key.offset - 1,
674 BTRFS_BLOCK_GROUP_DIRTY);
676 old_val = btrfs_block_group_used(&cache->item);
677 num = min(total, cache->key.offset - block_in_group);
679 if (blocknr > cache->last_alloc)
680 cache->last_alloc = blocknr;
682 for (i = 0; i < num; i++) {
683 clear_radix_bit(&info->extent_map_radix,
687 if (cache->data != data &&
688 old_val < (cache->key.offset >> 1)) {
690 radix_tree_delete(cache->radix,
691 cache->key.objectid +
692 cache->key.offset - 1);
696 &info->block_group_data_radix;
698 BTRFS_BLOCK_GROUP_DATA;
700 cache->radix = &info->block_group_radix;
702 ~BTRFS_BLOCK_GROUP_DATA;
704 ret = radix_tree_insert(cache->radix,
705 cache->key.objectid +
706 cache->key.offset - 1,
712 if (blocknr < cache->first_free)
713 cache->first_free = blocknr;
714 if (!cache->data && mark_free) {
715 for (i = 0; i < num; i++) {
716 set_radix_bit(&info->extent_map_radix,
720 if (old_val < (cache->key.offset >> 1) &&
721 old_val + num >= (cache->key.offset >> 1)) {
722 radix_tree_tag_set(cache->radix,
723 cache->key.objectid +
724 cache->key.offset - 1,
725 BTRFS_BLOCK_GROUP_AVAIL);
728 btrfs_set_block_group_used(&cache->item, old_val);
735 static int try_remove_page(struct address_space *mapping, unsigned long index)
738 ret = invalidate_mapping_pages(mapping, index, index);
742 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
745 unsigned long gang[8];
746 struct inode *btree_inode = root->fs_info->btree_inode;
747 struct btrfs_block_group_cache *block_group;
751 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
752 struct radix_tree_root *extent_radix = &root->fs_info->extent_map_radix;
755 ret = find_first_radix_bit(pinned_radix, gang, 0,
761 for (i = 0; i < ret; i++) {
762 clear_radix_bit(pinned_radix, gang[i]);
763 block_group = btrfs_lookup_block_group(root->fs_info,
766 WARN_ON(block_group->pinned == 0);
767 block_group->pinned--;
768 if (gang[i] < block_group->last_alloc)
769 block_group->last_alloc = gang[i];
770 if (gang[i] < block_group->last_prealloc)
771 block_group->last_prealloc = gang[i];
772 if (!block_group->data)
773 set_radix_bit(extent_radix, gang[i]);
775 try_remove_page(btree_inode->i_mapping,
776 gang[i] << (PAGE_CACHE_SHIFT -
777 btree_inode->i_blkbits));
783 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
784 btrfs_root *extent_root)
786 struct btrfs_key ins;
787 struct btrfs_extent_item extent_item;
790 u64 super_blocks_used;
791 struct btrfs_fs_info *info = extent_root->fs_info;
793 btrfs_set_extent_refs(&extent_item, 1);
796 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
797 btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);
799 for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) {
800 ins.objectid = extent_root->fs_info->extent_tree_insert[i];
801 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
802 btrfs_set_super_blocks_used(&info->super_copy,
803 super_blocks_used + 1);
804 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
805 sizeof(extent_item));
808 extent_root->fs_info->extent_tree_insert_nr = 0;
812 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
815 struct btrfs_header *header;
816 struct buffer_head *bh;
819 bh = btrfs_find_tree_block(root, blocknr);
821 if (buffer_uptodate(bh)) {
823 root->fs_info->running_transaction->transid;
824 header = btrfs_buffer_header(bh);
825 if (btrfs_header_generation(header) ==
827 btrfs_block_release(root, bh);
831 btrfs_block_release(root, bh);
833 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
835 struct btrfs_block_group_cache *cache;
836 cache = btrfs_lookup_block_group(root->fs_info,
842 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
849 * remove an extent from the root, returns 0 on success
851 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
852 *root, u64 blocknr, u64 num_blocks, int pin,
855 struct btrfs_path *path;
856 struct btrfs_key key;
857 struct btrfs_fs_info *info = root->fs_info;
858 struct btrfs_root *extent_root = info->extent_root;
860 struct btrfs_extent_item *ei;
861 struct btrfs_key ins;
864 key.objectid = blocknr;
866 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
867 key.offset = num_blocks;
869 path = btrfs_alloc_path();
873 ret = find_free_extent(trans, root, 0, 0, (u64)-1, 0, &ins, 0, 0, 0);
875 btrfs_free_path(path);
879 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
883 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
884 struct btrfs_extent_item);
885 BUG_ON(ei->refs == 0);
886 refs = btrfs_extent_refs(ei) - 1;
887 btrfs_set_extent_refs(ei, refs);
888 btrfs_mark_buffer_dirty(path->nodes[0]);
890 u64 super_blocks_used;
893 ret = pin_down_block(root, blocknr, 0);
897 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
898 btrfs_set_super_blocks_used(&info->super_copy,
899 super_blocks_used - num_blocks);
900 ret = btrfs_del_item(trans, extent_root, path);
904 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
908 btrfs_free_path(path);
909 finish_current_insert(trans, extent_root);
914 * find all the blocks marked as pending in the radix tree and remove
915 * them from the extent map
917 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
918 btrfs_root *extent_root)
923 unsigned long gang[4];
925 struct radix_tree_root *pending_radix;
926 struct radix_tree_root *pinned_radix;
927 struct btrfs_block_group_cache *cache;
929 pending_radix = &extent_root->fs_info->pending_del_radix;
930 pinned_radix = &extent_root->fs_info->pinned_radix;
933 ret = find_first_radix_bit(pending_radix, gang, 0,
937 for (i = 0; i < ret; i++) {
938 wret = set_radix_bit(pinned_radix, gang[i]);
941 btrfs_lookup_block_group(extent_root->fs_info,
947 printk(KERN_CRIT "set_radix_bit, err %d\n",
951 wret = clear_radix_bit(pending_radix, gang[i]);
953 wret = __free_extent(trans, extent_root,
963 * remove an extent from the root, returns 0 on success
965 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
966 *root, u64 blocknr, u64 num_blocks, int pin)
968 struct btrfs_root *extent_root = root->fs_info->extent_root;
972 if (root == extent_root) {
973 pin_down_block(root, blocknr, 1);
976 ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
977 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
978 return ret ? ret : pending_ret;
982 * walks the btree of allocated extents and find a hole of a given size.
983 * The key ins is changed to record the hole:
984 * ins->objectid == block start
985 * ins->flags = BTRFS_EXTENT_ITEM_KEY
986 * ins->offset == number of blocks
987 * Any available blocks before search_start are skipped.
989 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
990 *orig_root, u64 num_blocks, u64 search_start, u64
991 search_end, u64 hint_block,
992 struct btrfs_key *ins, u64 exclude_start,
993 u64 exclude_nr, int data)
995 struct btrfs_path *path;
996 struct btrfs_key key;
1002 u64 orig_search_start = search_start;
1004 struct btrfs_leaf *l;
1005 struct btrfs_root * root = orig_root->fs_info->extent_root;
1006 struct btrfs_fs_info *info = root->fs_info;
1007 int total_needed = num_blocks;
1008 int total_found = 0;
1009 int fill_prealloc = 0;
1011 struct btrfs_block_group_cache *block_group;
1017 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1019 level = btrfs_header_level(btrfs_buffer_header(root->node));
1020 if (num_blocks == 0) {
1023 total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3;
1025 if (fill_prealloc) {
1027 int nr = info->extent_tree_prealloc_nr;
1028 first = info->extent_tree_prealloc[nr - 1];
1029 if (info->extent_tree_prealloc_nr >= total_needed &&
1030 first >= search_start) {
1031 ins->objectid = info->extent_tree_prealloc[0];
1035 info->extent_tree_prealloc_nr = 0;
1037 if (search_end == (u64)-1)
1038 search_end = btrfs_super_total_blocks(&info->super_copy);
1040 block_group = btrfs_lookup_block_group(info, hint_block);
1041 block_group = btrfs_find_block_group(root, block_group,
1042 hint_block, data, 1);
1044 block_group = btrfs_find_block_group(root,
1045 trans->block_group, 0,
1049 path = btrfs_alloc_path();
1052 if (!block_group->data)
1053 search_start = find_search_start(root, &block_group,
1054 search_start, total_needed);
1055 else if (!full_scan)
1056 search_start = max(block_group->last_alloc, search_start);
1058 btrfs_init_path(path);
1059 ins->objectid = search_start;
1063 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1067 if (path->slots[0] > 0) {
1071 l = btrfs_buffer_leaf(path->nodes[0]);
1072 btrfs_disk_key_to_cpu(&key, &l->items[path->slots[0]].key);
1074 * a rare case, go back one key if we hit a block group item
1075 * instead of an extent item
1077 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
1078 key.objectid + key.offset >= search_start) {
1079 ins->objectid = key.objectid;
1080 ins->offset = key.offset - 1;
1081 btrfs_release_path(root, path);
1082 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1086 if (path->slots[0] > 0) {
1092 l = btrfs_buffer_leaf(path->nodes[0]);
1093 slot = path->slots[0];
1094 if (slot >= btrfs_header_nritems(&l->header)) {
1095 if (fill_prealloc) {
1096 info->extent_tree_prealloc_nr = 0;
1100 limit = last_block +
1101 (block_group->key.offset >> 1);
1103 limit = search_start +
1104 (block_group->key.offset >> 1);
1105 ret = btrfs_next_leaf(root, path);
1111 ins->objectid = search_start;
1112 ins->offset = search_end - search_start;
1116 ins->objectid = last_block > search_start ?
1117 last_block : search_start;
1118 ins->offset = search_end - ins->objectid;
1122 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
1123 if (key.objectid >= search_start && key.objectid > last_block &&
1125 if (last_block < search_start)
1126 last_block = search_start;
1127 hole_size = key.objectid - last_block;
1128 if (hole_size >= num_blocks) {
1129 ins->objectid = last_block;
1130 ins->offset = hole_size;
1135 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1139 last_block = key.objectid + key.offset;
1140 if (!full_scan && last_block >= block_group->key.objectid +
1141 block_group->key.offset) {
1142 btrfs_release_path(root, path);
1143 search_start = block_group->key.objectid +
1144 block_group->key.offset * 2;
1152 /* we have to make sure we didn't find an extent that has already
1153 * been allocated by the map tree or the original allocation
1155 btrfs_release_path(root, path);
1156 BUG_ON(ins->objectid < search_start);
1158 if (ins->objectid + num_blocks >= search_end) {
1163 search_start = orig_search_start;
1170 for (test_block = ins->objectid;
1171 test_block < ins->objectid + num_blocks; test_block++) {
1172 if (test_radix_bit(&info->pinned_radix, test_block)) {
1173 search_start = test_block + 1;
1177 if (!fill_prealloc && info->extent_tree_insert_nr) {
1179 info->extent_tree_insert[info->extent_tree_insert_nr - 1];
1180 if (ins->objectid + num_blocks >
1181 info->extent_tree_insert[0] &&
1182 ins->objectid <= last) {
1183 search_start = last + 1;
1184 WARN_ON(!full_scan);
1188 if (!fill_prealloc && info->extent_tree_prealloc_nr) {
1190 info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1];
1191 if (ins->objectid + num_blocks > first &&
1192 ins->objectid <= info->extent_tree_prealloc[0]) {
1193 search_start = info->extent_tree_prealloc[0] + 1;
1197 if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1198 ins->objectid < exclude_start + exclude_nr)) {
1199 search_start = exclude_start + exclude_nr;
1202 if (fill_prealloc) {
1204 test_block = ins->objectid;
1205 if (test_block - info->extent_tree_prealloc[total_needed - 1] >=
1208 info->extent_tree_prealloc_nr = total_found;
1210 while(test_block < ins->objectid + ins->offset &&
1211 total_found < total_needed) {
1212 nr = total_needed - total_found - 1;
1214 info->extent_tree_prealloc[nr] = test_block;
1218 if (total_found < total_needed) {
1219 search_start = test_block;
1222 info->extent_tree_prealloc_nr = total_found;
1225 block_group = btrfs_lookup_block_group(info, ins->objectid);
1228 block_group->last_prealloc =
1229 info->extent_tree_prealloc[total_needed-1];
1231 trans->block_group = block_group;
1234 ins->offset = num_blocks;
1235 btrfs_free_path(path);
1239 if (search_start + num_blocks >= search_end) {
1240 search_start = orig_search_start;
1250 block_group = btrfs_lookup_block_group(info, search_start);
1253 block_group = btrfs_find_block_group(root, block_group,
1254 search_start, data, 0);
1258 btrfs_release_path(root, path);
1259 btrfs_free_path(path);
1263 * finds a free extent and does all the dirty work required for allocation
1264 * returns the key for the extent through ins, and a tree buffer for
1265 * the first block of the extent through buf.
1267 * returns 0 if everything worked, non-zero otherwise.
1269 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root, u64 owner,
1271 u64 num_blocks, u64 hint_block,
1272 u64 search_end, struct btrfs_key *ins, int data)
1276 u64 super_blocks_used;
1277 u64 search_start = 0;
1278 u64 exclude_start = 0;
1280 struct btrfs_fs_info *info = root->fs_info;
1281 struct btrfs_root *extent_root = info->extent_root;
1282 struct btrfs_extent_item extent_item;
1283 struct btrfs_key prealloc_key;
1285 btrfs_set_extent_refs(&extent_item, 1);
1286 btrfs_set_extent_owner(&extent_item, owner);
1288 if (root == extent_root) {
1290 BUG_ON(info->extent_tree_prealloc_nr == 0);
1291 BUG_ON(num_blocks != 1);
1293 info->extent_tree_prealloc_nr--;
1294 nr = info->extent_tree_prealloc_nr;
1295 ins->objectid = info->extent_tree_prealloc[nr];
1296 info->extent_tree_insert[info->extent_tree_insert_nr++] =
1298 ret = update_block_group(trans, root,
1299 ins->objectid, ins->offset, 1, 0, 0);
1305 * if we're doing a data allocation, preallocate room in the
1306 * extent tree first. This way the extent tree blocks end up
1307 * in the correct block group.
1310 ret = find_free_extent(trans, root, 0, 0,
1311 search_end, 0, &prealloc_key, 0, 0, 0);
1314 exclude_nr = info->extent_tree_prealloc_nr;
1315 exclude_start = info->extent_tree_prealloc[exclude_nr - 1];
1318 /* do the real allocation */
1319 ret = find_free_extent(trans, root, num_blocks, search_start,
1320 search_end, hint_block, ins,
1321 exclude_start, exclude_nr, data);
1326 * if we're doing a metadata allocation, preallocate space in the
1327 * extent tree second. This way, we don't create a tiny hole
1328 * in the allocation map between any unused preallocation blocks
1329 * and the metadata block we're actually allocating. On disk,
1331 * [block we've allocated], [used prealloc 1], [ unused prealloc ]
1332 * The unused prealloc will get reused the next time around.
1335 exclude_start = ins->objectid;
1336 exclude_nr = ins->offset;
1337 ret = find_free_extent(trans, root, 0, search_start,
1338 search_end, hint_block,
1339 &prealloc_key, exclude_start,
1345 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1346 btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1348 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1349 sizeof(extent_item));
1351 finish_current_insert(trans, extent_root);
1352 pending_ret = del_pending_extents(trans, extent_root);
1359 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1366 * helper function to allocate a block for a given tree
1367 * returns the tree buffer or NULL.
1369 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root, u64 hint)
1372 struct btrfs_key ins;
1374 struct buffer_head *buf;
1376 ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1377 1, hint, (unsigned long)-1, &ins, 0);
1380 return ERR_PTR(ret);
1382 buf = btrfs_find_create_tree_block(root, ins.objectid);
1384 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1385 return ERR_PTR(-ENOMEM);
1387 set_buffer_uptodate(buf);
1388 set_buffer_checked(buf);
1389 set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index);
1393 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root, struct buffer_head *cur)
1396 struct btrfs_disk_key *key;
1397 struct btrfs_leaf *leaf;
1398 struct btrfs_file_extent_item *fi;
1403 BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
1404 leaf = btrfs_buffer_leaf(cur);
1405 nritems = btrfs_header_nritems(&leaf->header);
1406 for (i = 0; i < nritems; i++) {
1408 key = &leaf->items[i].key;
1409 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
1411 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1412 if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE)
1415 * FIXME make sure to insert a trans record that
1416 * repeats the snapshot del on crash
1418 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
1419 if (disk_blocknr == 0)
1421 ret = btrfs_free_extent(trans, root, disk_blocknr,
1422 btrfs_file_extent_disk_num_blocks(fi),
1429 static void reada_walk_down(struct btrfs_root *root,
1430 struct btrfs_node *node)
1438 nritems = btrfs_header_nritems(&node->header);
1439 for (i = 0; i < nritems; i++) {
1440 blocknr = btrfs_node_blockptr(node, i);
1441 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1445 ret = readahead_tree_block(root, blocknr);
1452 * helper function for drop_snapshot, this walks down the tree dropping ref
1453 * counts as it goes.
1455 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1456 *root, struct btrfs_path *path, int *level)
1458 struct buffer_head *next;
1459 struct buffer_head *cur;
1464 WARN_ON(*level < 0);
1465 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1466 ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
1473 * walk down to the last node level and free all the leaves
1475 while(*level >= 0) {
1476 WARN_ON(*level < 0);
1477 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1478 cur = path->nodes[*level];
1480 if (*level > 0 && path->slots[*level] == 0)
1481 reada_walk_down(root, btrfs_buffer_node(cur));
1483 if (btrfs_header_level(btrfs_buffer_header(cur)) != *level)
1486 if (path->slots[*level] >=
1487 btrfs_header_nritems(btrfs_buffer_header(cur)))
1490 ret = drop_leaf_ref(trans, root, cur);
1494 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
1495 path->slots[*level]);
1496 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1499 path->slots[*level]++;
1500 ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1504 next = read_tree_block(root, blocknr);
1505 WARN_ON(*level <= 0);
1506 if (path->nodes[*level-1])
1507 btrfs_block_release(root, path->nodes[*level-1]);
1508 path->nodes[*level-1] = next;
1509 *level = btrfs_header_level(btrfs_buffer_header(next));
1510 path->slots[*level] = 0;
1513 WARN_ON(*level < 0);
1514 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1515 ret = btrfs_free_extent(trans, root,
1516 bh_blocknr(path->nodes[*level]), 1, 1);
1517 btrfs_block_release(root, path->nodes[*level]);
1518 path->nodes[*level] = NULL;
1525 * helper for dropping snapshots. This walks back up the tree in the path
1526 * to find the first node higher up where we haven't yet gone through
1529 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1530 *root, struct btrfs_path *path, int *level)
1535 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1536 slot = path->slots[i];
1537 if (slot < btrfs_header_nritems(
1538 btrfs_buffer_header(path->nodes[i])) - 1) {
1543 ret = btrfs_free_extent(trans, root,
1544 bh_blocknr(path->nodes[*level]),
1547 btrfs_block_release(root, path->nodes[*level]);
1548 path->nodes[*level] = NULL;
1556 * drop the reference count on the tree rooted at 'snap'. This traverses
1557 * the tree freeing any blocks that have a ref count of zero after being
1560 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1561 *root, struct buffer_head *snap)
1566 struct btrfs_path *path;
1570 path = btrfs_alloc_path();
1573 level = btrfs_header_level(btrfs_buffer_header(snap));
1575 path->nodes[level] = snap;
1576 path->slots[level] = 0;
1578 wret = walk_down_tree(trans, root, path, &level);
1584 wret = walk_up_tree(trans, root, path, &level);
1590 for (i = 0; i <= orig_level; i++) {
1591 if (path->nodes[i]) {
1592 btrfs_block_release(root, path->nodes[i]);
1595 btrfs_free_path(path);
1599 static int free_block_group_radix(struct radix_tree_root *radix)
1602 struct btrfs_block_group_cache *cache[8];
1606 ret = radix_tree_gang_lookup(radix, (void **)cache, 0,
1610 for (i = 0; i < ret; i++) {
1611 radix_tree_delete(radix, cache[i]->key.objectid +
1612 cache[i]->key.offset - 1);
1619 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1623 unsigned long gang[16];
1626 ret = free_block_group_radix(&info->block_group_radix);
1627 ret2 = free_block_group_radix(&info->block_group_data_radix);
1634 ret = find_first_radix_bit(&info->extent_map_radix,
1635 gang, 0, ARRAY_SIZE(gang));
1638 for (i = 0; i < ret; i++) {
1639 clear_radix_bit(&info->extent_map_radix, gang[i]);
1645 int btrfs_read_block_groups(struct btrfs_root *root)
1647 struct btrfs_path *path;
1650 struct btrfs_block_group_item *bi;
1651 struct btrfs_block_group_cache *cache;
1652 struct btrfs_fs_info *info = root->fs_info;
1653 struct radix_tree_root *radix;
1654 struct btrfs_key key;
1655 struct btrfs_key found_key;
1656 struct btrfs_leaf *leaf;
1657 u64 group_size_blocks;
1660 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1661 root->fs_info->sb->s_blocksize_bits;
1662 root = info->extent_root;
1664 key.offset = group_size_blocks;
1666 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1668 path = btrfs_alloc_path();
1673 ret = btrfs_search_slot(NULL, info->extent_root,
1679 leaf = btrfs_buffer_leaf(path->nodes[0]);
1680 btrfs_disk_key_to_cpu(&found_key,
1681 &leaf->items[path->slots[0]].key);
1682 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1688 bi = btrfs_item_ptr(leaf, path->slots[0],
1689 struct btrfs_block_group_item);
1690 if (bi->flags & BTRFS_BLOCK_GROUP_DATA) {
1691 radix = &info->block_group_data_radix;
1694 radix = &info->block_group_radix;
1698 memcpy(&cache->item, bi, sizeof(*bi));
1699 memcpy(&cache->key, &found_key, sizeof(found_key));
1700 cache->last_alloc = cache->key.objectid;
1701 cache->first_free = cache->key.objectid;
1702 cache->last_prealloc = cache->key.objectid;
1706 cache->radix = radix;
1708 key.objectid = found_key.objectid + found_key.offset;
1709 btrfs_release_path(root, path);
1710 ret = radix_tree_insert(radix, found_key.objectid +
1711 found_key.offset - 1,
1714 used = btrfs_block_group_used(bi);
1715 if (used < div_factor(key.offset, 8)) {
1716 radix_tree_tag_set(radix, found_key.objectid +
1717 found_key.offset - 1,
1718 BTRFS_BLOCK_GROUP_AVAIL);
1721 btrfs_super_total_blocks(&info->super_copy))
1725 btrfs_free_path(path);