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>
22 #include "print-tree.h"
23 #include "transaction.h"
25 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
26 btrfs_root *extent_root);
27 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
28 btrfs_root *extent_root);
30 static int cache_block_group(struct btrfs_root *root,
31 struct btrfs_block_group_cache *block_group)
33 struct btrfs_path *path;
36 struct btrfs_leaf *leaf;
37 struct radix_tree_root *extent_radix;
45 root = root->fs_info->extent_root;
46 extent_radix = &root->fs_info->extent_map_radix;
48 if (block_group->cached)
50 if (block_group->data)
52 path = btrfs_alloc_path();
56 key.objectid = block_group->key.objectid;
59 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
60 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
63 if (ret && path->slots[0] > 0)
65 limit = block_group->key.objectid + block_group->key.offset;
67 leaf = btrfs_buffer_leaf(path->nodes[0]);
68 slot = path->slots[0];
69 if (slot >= btrfs_header_nritems(&leaf->header)) {
70 ret = btrfs_next_leaf(root, path);
77 hole_size = block_group->key.objectid +
78 block_group->key.offset - last;
80 last = block_group->key.objectid;
81 hole_size = block_group->key.offset;
83 for (i = 0; i < hole_size; i++) {
84 set_radix_bit(extent_radix,
90 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
91 if (key.objectid >= block_group->key.objectid +
92 block_group->key.offset) {
94 hole_size = block_group->key.objectid +
95 block_group->key.offset - last;
97 last = block_group->key.objectid;
98 hole_size = block_group->key.offset;
100 for (i = 0; i < hole_size; i++) {
101 set_radix_bit(extent_radix, last + i);
105 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
107 last = key.objectid + key.offset;
110 hole_size = key.objectid - last;
111 for (i = 0; i < hole_size; i++) {
112 set_radix_bit(extent_radix, last + i);
114 last = key.objectid + key.offset;
120 block_group->cached = 1;
122 btrfs_free_path(path);
126 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
130 struct btrfs_block_group_cache *block_group;
133 ret = radix_tree_gang_lookup(&info->block_group_radix,
134 (void **)&block_group,
137 if (block_group->key.objectid <= blocknr && blocknr <=
138 block_group->key.objectid + block_group->key.offset)
141 ret = radix_tree_gang_lookup(&info->block_group_data_radix,
142 (void **)&block_group,
145 if (block_group->key.objectid <= blocknr && blocknr <=
146 block_group->key.objectid + block_group->key.offset)
152 static u64 leaf_range(struct btrfs_root *root)
154 u64 size = BTRFS_LEAF_DATA_SIZE(root);
155 do_div(size, sizeof(struct btrfs_extent_item) +
156 sizeof(struct btrfs_item));
160 static u64 find_search_start(struct btrfs_root *root,
161 struct btrfs_block_group_cache **cache_ret,
162 u64 search_start, int num)
164 unsigned long gang[8];
166 struct btrfs_block_group_cache *cache = *cache_ret;
167 u64 last = max(search_start, cache->key.objectid);
172 ret = cache_block_group(root, cache);
176 ret = find_first_radix_bit(&root->fs_info->extent_map_radix,
177 gang, last, ARRAY_SIZE(gang));
180 last = gang[ret-1] + 1;
182 if (ret != ARRAY_SIZE(gang)) {
185 if (gang[ret-1] - gang[0] > leaf_range(root)) {
189 if (gang[0] >= cache->key.objectid + cache->key.offset) {
195 return max(cache->last_alloc, search_start);
198 cache = btrfs_lookup_block_group(root->fs_info,
199 last + cache->key.offset - 1);
201 return max((*cache_ret)->last_alloc, search_start);
203 cache = btrfs_find_block_group(root, cache,
204 last + cache->key.offset - 1, 0, 0);
209 static u64 div_factor(u64 num, int factor)
216 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
217 struct btrfs_block_group_cache
218 *hint, u64 search_start,
221 struct btrfs_block_group_cache *cache[8];
222 struct btrfs_block_group_cache *found_group = NULL;
223 struct btrfs_fs_info *info = root->fs_info;
224 struct radix_tree_root *radix;
225 struct radix_tree_root *swap_radix;
239 radix = &info->block_group_data_radix;
240 swap_radix = &info->block_group_radix;
242 radix = &info->block_group_radix;
243 swap_radix = &info->block_group_data_radix;
247 struct btrfs_block_group_cache *shint;
248 shint = btrfs_lookup_block_group(info, search_start);
249 if (shint->data == data) {
250 used = btrfs_block_group_used(&shint->item);
251 if (used + shint->pinned <
252 div_factor(shint->key.offset, factor)) {
257 if (hint && hint->data == data) {
258 used = btrfs_block_group_used(&hint->item);
259 if (used + hint->pinned <
260 div_factor(hint->key.offset, factor)) {
263 if (used >= div_factor(hint->key.offset, 8)) {
264 radix_tree_tag_clear(radix,
266 hint->key.offset - 1,
267 BTRFS_BLOCK_GROUP_AVAIL);
269 last = hint->key.offset * 3;
270 if (hint->key.objectid >= last)
271 last = max(search_start + hint->key.offset - 1,
272 hint->key.objectid - last);
274 last = hint->key.objectid + hint->key.offset;
278 hint_last = max(hint->key.objectid, search_start);
280 hint_last = search_start;
285 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
286 last, ARRAY_SIZE(cache),
287 BTRFS_BLOCK_GROUP_AVAIL);
290 for (i = 0; i < ret; i++) {
291 last = cache[i]->key.objectid +
292 cache[i]->key.offset;
293 used = btrfs_block_group_used(&cache[i]->item);
294 if (used + cache[i]->pinned <
295 div_factor(cache[i]->key.offset, factor)) {
296 found_group = cache[i];
299 if (used >= div_factor(cache[i]->key.offset, 8)) {
300 radix_tree_tag_clear(radix,
301 cache[i]->key.objectid +
302 cache[i]->key.offset - 1,
303 BTRFS_BLOCK_GROUP_AVAIL);
311 ret = radix_tree_gang_lookup(radix, (void **)cache,
312 last, ARRAY_SIZE(cache));
315 for (i = 0; i < ret; i++) {
316 last = cache[i]->key.objectid +
317 cache[i]->key.offset;
318 used = btrfs_block_group_used(&cache[i]->item);
319 if (used + cache[i]->pinned < cache[i]->key.offset) {
320 found_group = cache[i];
323 if (used >= cache[i]->key.offset) {
324 radix_tree_tag_clear(radix,
325 cache[i]->key.objectid +
326 cache[i]->key.offset - 1,
327 BTRFS_BLOCK_GROUP_AVAIL);
338 struct radix_tree_root *tmp = radix;
346 ret = radix_tree_gang_lookup(radix,
347 (void **)&found_group, 0, 1);
349 ret = radix_tree_gang_lookup(swap_radix,
350 (void **)&found_group,
359 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
360 struct btrfs_root *root,
361 u64 blocknr, u64 num_blocks)
363 struct btrfs_path *path;
365 struct btrfs_key key;
366 struct btrfs_leaf *l;
367 struct btrfs_extent_item *item;
370 path = btrfs_alloc_path();
374 key.objectid = blocknr;
376 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
377 key.offset = num_blocks;
378 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
386 l = btrfs_buffer_leaf(path->nodes[0]);
387 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
388 refs = btrfs_extent_refs(item);
389 btrfs_set_extent_refs(item, refs + 1);
390 btrfs_mark_buffer_dirty(path->nodes[0]);
392 btrfs_release_path(root->fs_info->extent_root, path);
393 btrfs_free_path(path);
394 finish_current_insert(trans, root->fs_info->extent_root);
395 del_pending_extents(trans, root->fs_info->extent_root);
399 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
400 struct btrfs_root *root)
402 finish_current_insert(trans, root->fs_info->extent_root);
403 del_pending_extents(trans, root->fs_info->extent_root);
407 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
408 struct btrfs_root *root, u64 blocknr,
409 u64 num_blocks, u32 *refs)
411 struct btrfs_path *path;
413 struct btrfs_key key;
414 struct btrfs_leaf *l;
415 struct btrfs_extent_item *item;
417 path = btrfs_alloc_path();
418 key.objectid = blocknr;
419 key.offset = num_blocks;
421 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
422 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);
432 btrfs_free_path(path);
436 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
437 struct btrfs_root *root)
439 return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
442 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
443 struct buffer_head *buf)
446 struct btrfs_node *buf_node;
447 struct btrfs_leaf *buf_leaf;
448 struct btrfs_disk_key *key;
449 struct btrfs_file_extent_item *fi;
458 buf_node = btrfs_buffer_node(buf);
459 leaf = btrfs_is_leaf(buf_node);
460 buf_leaf = btrfs_buffer_leaf(buf);
461 for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
464 key = &buf_leaf->items[i].key;
465 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
467 fi = btrfs_item_ptr(buf_leaf, i,
468 struct btrfs_file_extent_item);
469 if (btrfs_file_extent_type(fi) ==
470 BTRFS_FILE_EXTENT_INLINE)
472 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
473 if (disk_blocknr == 0)
475 ret = btrfs_inc_extent_ref(trans, root, disk_blocknr,
476 btrfs_file_extent_disk_num_blocks(fi));
482 blocknr = btrfs_node_blockptr(buf_node, i);
483 ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
493 for (i =0; i < faili; i++) {
496 key = &buf_leaf->items[i].key;
497 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
499 fi = btrfs_item_ptr(buf_leaf, i,
500 struct btrfs_file_extent_item);
501 if (btrfs_file_extent_type(fi) ==
502 BTRFS_FILE_EXTENT_INLINE)
504 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
505 if (disk_blocknr == 0)
507 err = btrfs_free_extent(trans, root, disk_blocknr,
508 btrfs_file_extent_disk_num_blocks(fi), 0);
511 blocknr = btrfs_node_blockptr(buf_node, i);
512 err = btrfs_free_extent(trans, root, blocknr, 1, 0);
519 static int write_one_cache_group(struct btrfs_trans_handle *trans,
520 struct btrfs_root *root,
521 struct btrfs_path *path,
522 struct btrfs_block_group_cache *cache)
526 struct btrfs_root *extent_root = root->fs_info->extent_root;
527 struct btrfs_block_group_item *bi;
529 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
533 bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
534 struct btrfs_block_group_item);
535 memcpy(bi, &cache->item, sizeof(*bi));
536 btrfs_mark_buffer_dirty(path->nodes[0]);
537 btrfs_release_path(extent_root, path);
539 finish_current_insert(trans, extent_root);
540 pending_ret = del_pending_extents(trans, extent_root);
546 cache->last_alloc = cache->first_free;
551 static int write_dirty_block_radix(struct btrfs_trans_handle *trans,
552 struct btrfs_root *root,
553 struct radix_tree_root *radix)
555 struct btrfs_block_group_cache *cache[8];
560 struct btrfs_path *path;
561 unsigned long off = 0;
563 path = btrfs_alloc_path();
568 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
569 off, ARRAY_SIZE(cache),
570 BTRFS_BLOCK_GROUP_DIRTY);
573 for (i = 0; i < ret; i++) {
574 err = write_one_cache_group(trans, root,
577 * if we fail to write the cache group, we want
578 * to keep it marked dirty in hopes that a later
583 off = cache[i]->key.objectid +
584 cache[i]->key.offset;
588 radix_tree_tag_clear(radix, cache[i]->key.objectid +
589 cache[i]->key.offset - 1,
590 BTRFS_BLOCK_GROUP_DIRTY);
593 btrfs_free_path(path);
597 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
598 struct btrfs_root *root)
602 ret = write_dirty_block_radix(trans, root,
603 &root->fs_info->block_group_radix);
604 ret2 = write_dirty_block_radix(trans, root,
605 &root->fs_info->block_group_data_radix);
613 static int update_block_group(struct btrfs_trans_handle *trans,
614 struct btrfs_root *root,
615 u64 blocknr, u64 num, int alloc, int mark_free,
618 struct btrfs_block_group_cache *cache;
619 struct btrfs_fs_info *info = root->fs_info;
627 cache = btrfs_lookup_block_group(info, blocknr);
631 block_in_group = blocknr - cache->key.objectid;
632 WARN_ON(block_in_group > cache->key.offset);
633 radix_tree_tag_set(cache->radix, cache->key.objectid +
634 cache->key.offset - 1,
635 BTRFS_BLOCK_GROUP_DIRTY);
637 old_val = btrfs_block_group_used(&cache->item);
638 num = min(total, cache->key.offset - block_in_group);
640 if (blocknr > cache->last_alloc)
641 cache->last_alloc = blocknr;
643 for (i = 0; i < num; i++) {
644 clear_radix_bit(&info->extent_map_radix,
648 if (cache->data != data &&
649 old_val < (cache->key.offset >> 1)) {
651 radix_tree_delete(cache->radix,
652 cache->key.objectid +
653 cache->key.offset - 1);
657 &info->block_group_data_radix;
659 BTRFS_BLOCK_GROUP_DATA;
661 cache->radix = &info->block_group_radix;
663 ~BTRFS_BLOCK_GROUP_DATA;
665 ret = radix_tree_insert(cache->radix,
666 cache->key.objectid +
667 cache->key.offset - 1,
673 if (blocknr < cache->first_free)
674 cache->first_free = blocknr;
675 if (!cache->data && mark_free) {
676 for (i = 0; i < num; i++) {
677 set_radix_bit(&info->extent_map_radix,
681 if (old_val < (cache->key.offset >> 1) &&
682 old_val + num >= (cache->key.offset >> 1)) {
683 radix_tree_tag_set(cache->radix,
684 cache->key.objectid +
685 cache->key.offset - 1,
686 BTRFS_BLOCK_GROUP_AVAIL);
689 btrfs_set_block_group_used(&cache->item, old_val);
696 int btrfs_copy_pinned(struct btrfs_root *root, struct radix_tree_root *copy)
698 unsigned long gang[8];
700 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
705 ret = find_first_radix_bit(pinned_radix, gang, last,
709 for (i = 0 ; i < ret; i++) {
710 set_radix_bit(copy, gang[i]);
714 ret = find_first_radix_bit(&root->fs_info->extent_ins_radix, gang, 0,
720 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
721 struct btrfs_root *root,
722 struct radix_tree_root *unpin_radix)
724 unsigned long gang[8];
725 struct btrfs_block_group_cache *block_group;
729 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
730 struct radix_tree_root *extent_radix = &root->fs_info->extent_map_radix;
733 ret = find_first_radix_bit(unpin_radix, gang, 0,
739 for (i = 0; i < ret; i++) {
740 clear_radix_bit(pinned_radix, gang[i]);
741 clear_radix_bit(unpin_radix, gang[i]);
742 block_group = btrfs_lookup_block_group(root->fs_info,
745 WARN_ON(block_group->pinned == 0);
746 block_group->pinned--;
747 if (gang[i] < block_group->last_alloc)
748 block_group->last_alloc = gang[i];
749 if (!block_group->data)
750 set_radix_bit(extent_radix, gang[i]);
757 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
758 btrfs_root *extent_root)
760 struct btrfs_key ins;
761 struct btrfs_extent_item extent_item;
765 unsigned long gang[8];
766 struct btrfs_fs_info *info = extent_root->fs_info;
768 btrfs_set_extent_refs(&extent_item, 1);
771 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
772 btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);
775 ret = find_first_radix_bit(&info->extent_ins_radix, gang, 0,
780 for (i = 0; i < ret; i++) {
781 ins.objectid = gang[i];
782 err = btrfs_insert_item(trans, extent_root, &ins,
784 sizeof(extent_item));
785 clear_radix_bit(&info->extent_ins_radix, gang[i]);
792 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
795 struct btrfs_header *header;
796 struct buffer_head *bh;
799 bh = btrfs_find_tree_block(root, blocknr);
801 if (buffer_uptodate(bh)) {
803 root->fs_info->running_transaction->transid;
804 header = btrfs_buffer_header(bh);
805 if (btrfs_header_generation(header) ==
807 btrfs_block_release(root, bh);
811 btrfs_block_release(root, bh);
813 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
815 struct btrfs_block_group_cache *cache;
816 cache = btrfs_lookup_block_group(root->fs_info,
822 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
829 * remove an extent from the root, returns 0 on success
831 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
832 *root, u64 blocknr, u64 num_blocks, int pin,
835 struct btrfs_path *path;
836 struct btrfs_key key;
837 struct btrfs_fs_info *info = root->fs_info;
838 struct btrfs_root *extent_root = info->extent_root;
840 struct btrfs_extent_item *ei;
843 key.objectid = blocknr;
845 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
846 key.offset = num_blocks;
848 path = btrfs_alloc_path();
852 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
856 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
857 struct btrfs_extent_item);
858 BUG_ON(ei->refs == 0);
859 refs = btrfs_extent_refs(ei) - 1;
860 btrfs_set_extent_refs(ei, refs);
861 btrfs_mark_buffer_dirty(path->nodes[0]);
863 u64 super_blocks_used;
866 ret = pin_down_block(root, blocknr, 0);
870 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
871 btrfs_set_super_blocks_used(&info->super_copy,
872 super_blocks_used - num_blocks);
873 ret = btrfs_del_item(trans, extent_root, path);
877 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
881 btrfs_free_path(path);
882 finish_current_insert(trans, extent_root);
887 * find all the blocks marked as pending in the radix tree and remove
888 * them from the extent map
890 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
891 btrfs_root *extent_root)
896 unsigned long gang[4];
898 struct radix_tree_root *pending_radix;
899 struct radix_tree_root *pinned_radix;
900 struct btrfs_block_group_cache *cache;
902 pending_radix = &extent_root->fs_info->pending_del_radix;
903 pinned_radix = &extent_root->fs_info->pinned_radix;
906 ret = find_first_radix_bit(pending_radix, gang, 0,
910 for (i = 0; i < ret; i++) {
911 wret = set_radix_bit(pinned_radix, gang[i]);
914 btrfs_lookup_block_group(extent_root->fs_info,
920 printk(KERN_CRIT "set_radix_bit, err %d\n",
924 wret = clear_radix_bit(pending_radix, gang[i]);
926 wret = __free_extent(trans, extent_root,
936 * remove an extent from the root, returns 0 on success
938 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
939 *root, u64 blocknr, u64 num_blocks, int pin)
941 struct btrfs_root *extent_root = root->fs_info->extent_root;
945 if (root == extent_root) {
946 pin_down_block(root, blocknr, 1);
949 ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
950 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
951 return ret ? ret : pending_ret;
955 * walks the btree of allocated extents and find a hole of a given size.
956 * The key ins is changed to record the hole:
957 * ins->objectid == block start
958 * ins->flags = BTRFS_EXTENT_ITEM_KEY
959 * ins->offset == number of blocks
960 * Any available blocks before search_start are skipped.
962 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
963 *orig_root, u64 num_blocks, u64 empty_size,
964 u64 search_start, u64 search_end, u64 hint_block,
965 struct btrfs_key *ins, u64 exclude_start,
966 u64 exclude_nr, int data)
968 struct btrfs_path *path;
969 struct btrfs_key key;
975 u64 orig_search_start = search_start;
977 struct btrfs_leaf *l;
978 struct btrfs_root * root = orig_root->fs_info->extent_root;
979 struct btrfs_fs_info *info = root->fs_info;
980 int total_needed = num_blocks;
982 struct btrfs_block_group_cache *block_group;
987 WARN_ON(num_blocks < 1);
989 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
991 level = btrfs_header_level(btrfs_buffer_header(root->node));
992 if (search_end == (u64)-1)
993 search_end = btrfs_super_total_blocks(&info->super_copy);
995 block_group = btrfs_lookup_block_group(info, hint_block);
996 block_group = btrfs_find_block_group(root, block_group,
997 hint_block, data, 1);
999 block_group = btrfs_find_block_group(root,
1000 trans->block_group, 0,
1004 total_needed += empty_size;
1005 path = btrfs_alloc_path();
1008 if (!block_group->data)
1009 search_start = find_search_start(root, &block_group,
1010 search_start, total_needed);
1011 else if (!full_scan)
1012 search_start = max(block_group->last_alloc, search_start);
1014 btrfs_init_path(path);
1015 ins->objectid = search_start;
1019 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1023 if (path->slots[0] > 0) {
1027 l = btrfs_buffer_leaf(path->nodes[0]);
1028 btrfs_disk_key_to_cpu(&key, &l->items[path->slots[0]].key);
1030 * a rare case, go back one key if we hit a block group item
1031 * instead of an extent item
1033 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
1034 key.objectid + key.offset >= search_start) {
1035 ins->objectid = key.objectid;
1036 ins->offset = key.offset - 1;
1037 btrfs_release_path(root, path);
1038 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1042 if (path->slots[0] > 0) {
1048 l = btrfs_buffer_leaf(path->nodes[0]);
1049 slot = path->slots[0];
1050 if (slot >= btrfs_header_nritems(&l->header)) {
1052 limit = last_block +
1053 (block_group->key.offset >> 1);
1055 limit = search_start +
1056 (block_group->key.offset >> 1);
1057 ret = btrfs_next_leaf(root, path);
1063 ins->objectid = search_start;
1064 ins->offset = search_end - search_start;
1068 ins->objectid = last_block > search_start ?
1069 last_block : search_start;
1070 ins->offset = search_end - ins->objectid;
1074 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
1075 if (key.objectid >= search_start && key.objectid > last_block &&
1077 if (last_block < search_start)
1078 last_block = search_start;
1079 hole_size = key.objectid - last_block;
1080 if (hole_size >= num_blocks) {
1081 ins->objectid = last_block;
1082 ins->offset = hole_size;
1087 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1091 last_block = key.objectid + key.offset;
1092 if (!full_scan && last_block >= block_group->key.objectid +
1093 block_group->key.offset) {
1094 btrfs_release_path(root, path);
1095 search_start = block_group->key.objectid +
1096 block_group->key.offset * 2;
1104 /* we have to make sure we didn't find an extent that has already
1105 * been allocated by the map tree or the original allocation
1107 btrfs_release_path(root, path);
1108 BUG_ON(ins->objectid < search_start);
1110 if (ins->objectid + num_blocks >= search_end) {
1115 search_start = orig_search_start;
1118 total_needed -= empty_size;
1124 for (test_block = ins->objectid;
1125 test_block < ins->objectid + num_blocks; test_block++) {
1126 if (test_radix_bit(&info->pinned_radix, test_block) ||
1127 test_radix_bit(&info->extent_ins_radix, test_block)) {
1128 search_start = test_block + 1;
1132 if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1133 ins->objectid < exclude_start + exclude_nr)) {
1134 search_start = exclude_start + exclude_nr;
1138 block_group = btrfs_lookup_block_group(info, ins->objectid);
1140 trans->block_group = block_group;
1142 ins->offset = num_blocks;
1143 btrfs_free_path(path);
1147 if (search_start + num_blocks >= search_end) {
1148 search_start = orig_search_start;
1155 total_needed -= empty_size;
1160 block_group = btrfs_lookup_block_group(info, search_start);
1163 block_group = btrfs_find_block_group(root, block_group,
1164 search_start, data, 0);
1168 btrfs_release_path(root, path);
1169 btrfs_free_path(path);
1173 * finds a free extent and does all the dirty work required for allocation
1174 * returns the key for the extent through ins, and a tree buffer for
1175 * the first block of the extent through buf.
1177 * returns 0 if everything worked, non-zero otherwise.
1179 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1180 struct btrfs_root *root, u64 owner,
1181 u64 num_blocks, u64 empty_size, u64 hint_block,
1182 u64 search_end, struct btrfs_key *ins, int data)
1186 u64 super_blocks_used;
1187 u64 search_start = 0;
1188 struct btrfs_fs_info *info = root->fs_info;
1189 struct btrfs_root *extent_root = info->extent_root;
1190 struct btrfs_extent_item extent_item;
1192 btrfs_set_extent_refs(&extent_item, 1);
1193 btrfs_set_extent_owner(&extent_item, owner);
1195 WARN_ON(num_blocks < 1);
1196 ret = find_free_extent(trans, root, num_blocks, empty_size,
1197 search_start, search_end, hint_block, ins,
1198 trans->alloc_exclude_start,
1199 trans->alloc_exclude_nr, data);
1204 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1205 btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1208 if (root == extent_root) {
1209 BUG_ON(num_blocks != 1);
1210 set_radix_bit(&root->fs_info->extent_ins_radix, ins->objectid);
1214 WARN_ON(trans->alloc_exclude_nr);
1215 trans->alloc_exclude_start = ins->objectid;
1216 trans->alloc_exclude_nr = ins->offset;
1217 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1218 sizeof(extent_item));
1220 trans->alloc_exclude_start = 0;
1221 trans->alloc_exclude_nr = 0;
1224 finish_current_insert(trans, extent_root);
1225 pending_ret = del_pending_extents(trans, extent_root);
1234 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1241 * helper function to allocate a block for a given tree
1242 * returns the tree buffer or NULL.
1244 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1245 struct btrfs_root *root, u64 hint,
1248 struct btrfs_key ins;
1250 struct buffer_head *buf;
1252 ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1253 1, empty_size, hint,
1254 (unsigned long)-1, &ins, 0);
1257 return ERR_PTR(ret);
1259 buf = btrfs_find_create_tree_block(root, ins.objectid);
1261 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1262 return ERR_PTR(-ENOMEM);
1264 WARN_ON(buffer_dirty(buf));
1265 set_buffer_uptodate(buf);
1266 set_buffer_checked(buf);
1267 set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index);
1271 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root, struct buffer_head *cur)
1274 struct btrfs_disk_key *key;
1275 struct btrfs_leaf *leaf;
1276 struct btrfs_file_extent_item *fi;
1281 BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
1282 leaf = btrfs_buffer_leaf(cur);
1283 nritems = btrfs_header_nritems(&leaf->header);
1284 for (i = 0; i < nritems; i++) {
1286 key = &leaf->items[i].key;
1287 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
1289 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1290 if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE)
1293 * FIXME make sure to insert a trans record that
1294 * repeats the snapshot del on crash
1296 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
1297 if (disk_blocknr == 0)
1299 ret = btrfs_free_extent(trans, root, disk_blocknr,
1300 btrfs_file_extent_disk_num_blocks(fi),
1307 static void reada_walk_down(struct btrfs_root *root,
1308 struct btrfs_node *node)
1316 nritems = btrfs_header_nritems(&node->header);
1317 for (i = 0; i < nritems; i++) {
1318 blocknr = btrfs_node_blockptr(node, i);
1319 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1323 mutex_unlock(&root->fs_info->fs_mutex);
1324 ret = readahead_tree_block(root, blocknr);
1326 mutex_lock(&root->fs_info->fs_mutex);
1333 * helper function for drop_snapshot, this walks down the tree dropping ref
1334 * counts as it goes.
1336 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1337 *root, struct btrfs_path *path, int *level)
1339 struct buffer_head *next;
1340 struct buffer_head *cur;
1345 WARN_ON(*level < 0);
1346 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1347 ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
1354 * walk down to the last node level and free all the leaves
1356 while(*level >= 0) {
1357 WARN_ON(*level < 0);
1358 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1359 cur = path->nodes[*level];
1361 if (*level > 0 && path->slots[*level] == 0)
1362 reada_walk_down(root, btrfs_buffer_node(cur));
1364 if (btrfs_header_level(btrfs_buffer_header(cur)) != *level)
1367 if (path->slots[*level] >=
1368 btrfs_header_nritems(btrfs_buffer_header(cur)))
1371 ret = drop_leaf_ref(trans, root, cur);
1375 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
1376 path->slots[*level]);
1377 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1380 path->slots[*level]++;
1381 ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1385 next = btrfs_find_tree_block(root, blocknr);
1386 if (!next || !buffer_uptodate(next)) {
1388 mutex_unlock(&root->fs_info->fs_mutex);
1389 next = read_tree_block(root, blocknr);
1390 mutex_lock(&root->fs_info->fs_mutex);
1392 /* we dropped the lock, check one more time */
1393 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1396 path->slots[*level]++;
1398 ret = btrfs_free_extent(trans, root,
1404 WARN_ON(*level <= 0);
1405 if (path->nodes[*level-1])
1406 btrfs_block_release(root, path->nodes[*level-1]);
1407 path->nodes[*level-1] = next;
1408 *level = btrfs_header_level(btrfs_buffer_header(next));
1409 path->slots[*level] = 0;
1412 WARN_ON(*level < 0);
1413 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1414 ret = btrfs_free_extent(trans, root,
1415 bh_blocknr(path->nodes[*level]), 1, 1);
1416 btrfs_block_release(root, path->nodes[*level]);
1417 path->nodes[*level] = NULL;
1424 * helper for dropping snapshots. This walks back up the tree in the path
1425 * to find the first node higher up where we haven't yet gone through
1428 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1429 *root, struct btrfs_path *path, int *level)
1434 struct btrfs_root_item *root_item = &root->root_item;
1436 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1437 slot = path->slots[i];
1438 if (slot < btrfs_header_nritems(
1439 btrfs_buffer_header(path->nodes[i])) - 1) {
1440 struct btrfs_node *node;
1441 node = btrfs_buffer_node(path->nodes[i]);
1444 WARN_ON(*level == 0);
1445 memcpy(&root_item->drop_progress,
1446 &node->ptrs[path->slots[i]].key,
1447 sizeof(root_item->drop_progress));
1448 root_item->drop_level = i;
1451 ret = btrfs_free_extent(trans, root,
1452 bh_blocknr(path->nodes[*level]),
1455 btrfs_block_release(root, path->nodes[*level]);
1456 path->nodes[*level] = NULL;
1464 * drop the reference count on the tree rooted at 'snap'. This traverses
1465 * the tree freeing any blocks that have a ref count of zero after being
1468 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1474 struct btrfs_path *path;
1477 struct btrfs_root_item *root_item = &root->root_item;
1479 path = btrfs_alloc_path();
1482 level = btrfs_header_level(btrfs_buffer_header(root->node));
1484 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1485 path->nodes[level] = root->node;
1486 path->slots[level] = 0;
1488 struct btrfs_key key;
1489 struct btrfs_disk_key *found_key;
1490 struct btrfs_node *node;
1492 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1493 level = root_item->drop_level;
1494 path->lowest_level = level;
1495 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1500 node = btrfs_buffer_node(path->nodes[level]);
1501 found_key = &node->ptrs[path->slots[level]].key;
1502 WARN_ON(memcmp(found_key, &root_item->drop_progress,
1503 sizeof(*found_key)));
1506 wret = walk_down_tree(trans, root, path, &level);
1512 wret = walk_up_tree(trans, root, path, &level);
1521 for (i = 0; i <= orig_level; i++) {
1522 if (path->nodes[i]) {
1523 btrfs_block_release(root, path->nodes[i]);
1528 btrfs_free_path(path);
1532 static int free_block_group_radix(struct radix_tree_root *radix)
1535 struct btrfs_block_group_cache *cache[8];
1539 ret = radix_tree_gang_lookup(radix, (void **)cache, 0,
1543 for (i = 0; i < ret; i++) {
1544 radix_tree_delete(radix, cache[i]->key.objectid +
1545 cache[i]->key.offset - 1);
1552 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1556 unsigned long gang[16];
1559 ret = free_block_group_radix(&info->block_group_radix);
1560 ret2 = free_block_group_radix(&info->block_group_data_radix);
1567 ret = find_first_radix_bit(&info->extent_map_radix,
1568 gang, 0, ARRAY_SIZE(gang));
1571 for (i = 0; i < ret; i++) {
1572 clear_radix_bit(&info->extent_map_radix, gang[i]);
1578 int btrfs_read_block_groups(struct btrfs_root *root)
1580 struct btrfs_path *path;
1583 struct btrfs_block_group_item *bi;
1584 struct btrfs_block_group_cache *cache;
1585 struct btrfs_fs_info *info = root->fs_info;
1586 struct radix_tree_root *radix;
1587 struct btrfs_key key;
1588 struct btrfs_key found_key;
1589 struct btrfs_leaf *leaf;
1590 u64 group_size_blocks;
1593 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1594 root->fs_info->sb->s_blocksize_bits;
1595 root = info->extent_root;
1597 key.offset = group_size_blocks;
1599 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1601 path = btrfs_alloc_path();
1606 ret = btrfs_search_slot(NULL, info->extent_root,
1612 leaf = btrfs_buffer_leaf(path->nodes[0]);
1613 btrfs_disk_key_to_cpu(&found_key,
1614 &leaf->items[path->slots[0]].key);
1615 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1621 bi = btrfs_item_ptr(leaf, path->slots[0],
1622 struct btrfs_block_group_item);
1623 if (bi->flags & BTRFS_BLOCK_GROUP_DATA) {
1624 radix = &info->block_group_data_radix;
1627 radix = &info->block_group_radix;
1631 memcpy(&cache->item, bi, sizeof(*bi));
1632 memcpy(&cache->key, &found_key, sizeof(found_key));
1633 cache->last_alloc = cache->key.objectid;
1634 cache->first_free = cache->key.objectid;
1638 cache->radix = radix;
1640 key.objectid = found_key.objectid + found_key.offset;
1641 btrfs_release_path(root, path);
1642 ret = radix_tree_insert(radix, found_key.objectid +
1643 found_key.offset - 1,
1646 used = btrfs_block_group_used(bi);
1647 if (used < div_factor(key.offset, 8)) {
1648 radix_tree_tag_set(radix, found_key.objectid +
1649 found_key.offset - 1,
1650 BTRFS_BLOCK_GROUP_AVAIL);
1653 btrfs_super_total_blocks(&info->super_copy))
1657 btrfs_free_path(path);