2 * Copyright (C) 2015 Facebook. 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/kernel.h>
20 #include <linux/vmalloc.h>
24 #include "free-space-tree.h"
25 #include "transaction.h"
27 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
28 struct btrfs_fs_info *fs_info,
29 struct btrfs_block_group_cache *block_group,
30 struct btrfs_path *path);
32 void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
36 u64 num_bitmaps, total_bitmap_size;
39 * We convert to bitmaps when the disk space required for using extents
40 * exceeds that required for using bitmaps.
42 bitmap_range = cache->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
43 num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
45 bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
46 total_bitmap_size = num_bitmaps * bitmap_size;
47 cache->bitmap_high_thresh = div_u64(total_bitmap_size,
48 sizeof(struct btrfs_item));
51 * We allow for a small buffer between the high threshold and low
52 * threshold to avoid thrashing back and forth between the two formats.
54 if (cache->bitmap_high_thresh > 100)
55 cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
57 cache->bitmap_low_thresh = 0;
60 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
61 struct btrfs_fs_info *fs_info,
62 struct btrfs_block_group_cache *block_group,
63 struct btrfs_path *path)
65 struct btrfs_root *root = fs_info->free_space_root;
66 struct btrfs_free_space_info *info;
68 struct extent_buffer *leaf;
71 key.objectid = block_group->key.objectid;
72 key.type = BTRFS_FREE_SPACE_INFO_KEY;
73 key.offset = block_group->key.offset;
75 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
79 leaf = path->nodes[0];
80 info = btrfs_item_ptr(leaf, path->slots[0],
81 struct btrfs_free_space_info);
82 btrfs_set_free_space_extent_count(leaf, info, 0);
83 btrfs_set_free_space_flags(leaf, info, 0);
84 btrfs_mark_buffer_dirty(leaf);
88 btrfs_release_path(path);
92 struct btrfs_free_space_info *
93 search_free_space_info(struct btrfs_trans_handle *trans,
94 struct btrfs_fs_info *fs_info,
95 struct btrfs_block_group_cache *block_group,
96 struct btrfs_path *path, int cow)
98 struct btrfs_root *root = fs_info->free_space_root;
102 key.objectid = block_group->key.objectid;
103 key.type = BTRFS_FREE_SPACE_INFO_KEY;
104 key.offset = block_group->key.offset;
106 ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
110 btrfs_warn(fs_info, "missing free space info for %llu",
111 block_group->key.objectid);
113 return ERR_PTR(-ENOENT);
116 return btrfs_item_ptr(path->nodes[0], path->slots[0],
117 struct btrfs_free_space_info);
121 * btrfs_search_slot() but we're looking for the greatest key less than the
124 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
125 struct btrfs_root *root,
126 struct btrfs_key *key, struct btrfs_path *p,
127 int ins_len, int cow)
131 ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
140 if (p->slots[0] == 0) {
149 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
151 return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
154 static unsigned long *alloc_bitmap(u32 bitmap_size)
159 * The allocation size varies, observed numbers were < 4K up to 16K.
160 * Using vmalloc unconditionally would be too heavy, we'll try
161 * contiguous allocations first.
163 if (bitmap_size <= PAGE_SIZE)
164 return kzalloc(bitmap_size, GFP_NOFS);
166 mem = kzalloc(bitmap_size, GFP_NOFS | __GFP_NOWARN);
170 return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO,
174 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
175 struct btrfs_fs_info *fs_info,
176 struct btrfs_block_group_cache *block_group,
177 struct btrfs_path *path)
179 struct btrfs_root *root = fs_info->free_space_root;
180 struct btrfs_free_space_info *info;
181 struct btrfs_key key, found_key;
182 struct extent_buffer *leaf;
183 unsigned long *bitmap;
187 u32 bitmap_size, flags, expected_extent_count;
188 u32 extent_count = 0;
192 bitmap_size = free_space_bitmap_size(block_group->key.offset,
193 block_group->sectorsize);
194 bitmap = alloc_bitmap(bitmap_size);
200 start = block_group->key.objectid;
201 end = block_group->key.objectid + block_group->key.offset;
203 key.objectid = end - 1;
205 key.offset = (u64)-1;
208 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
212 leaf = path->nodes[0];
215 while (path->slots[0] > 0) {
216 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
218 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
219 ASSERT(found_key.objectid == block_group->key.objectid);
220 ASSERT(found_key.offset == block_group->key.offset);
223 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
226 ASSERT(found_key.objectid >= start);
227 ASSERT(found_key.objectid < end);
228 ASSERT(found_key.objectid + found_key.offset <= end);
230 first = div_u64(found_key.objectid - start,
231 block_group->sectorsize);
232 last = div_u64(found_key.objectid + found_key.offset - start,
233 block_group->sectorsize);
234 bitmap_set(bitmap, first, last - first);
244 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
247 btrfs_release_path(path);
250 info = search_free_space_info(trans, fs_info, block_group, path, 1);
255 leaf = path->nodes[0];
256 flags = btrfs_free_space_flags(leaf, info);
257 flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
258 btrfs_set_free_space_flags(leaf, info, flags);
259 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
260 btrfs_mark_buffer_dirty(leaf);
261 btrfs_release_path(path);
263 if (extent_count != expected_extent_count) {
265 "incorrect extent count for %llu; counted %u, expected %u",
266 block_group->key.objectid, extent_count,
267 expected_extent_count);
273 bitmap_cursor = (char *)bitmap;
274 bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
281 extent_size = min(end - i, bitmap_range);
282 data_size = free_space_bitmap_size(extent_size,
283 block_group->sectorsize);
286 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
287 key.offset = extent_size;
289 ret = btrfs_insert_empty_item(trans, root, path, &key,
294 leaf = path->nodes[0];
295 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
296 write_extent_buffer(leaf, bitmap_cursor, ptr,
298 btrfs_mark_buffer_dirty(leaf);
299 btrfs_release_path(path);
302 bitmap_cursor += data_size;
309 btrfs_abort_transaction(trans, ret);
313 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
314 struct btrfs_fs_info *fs_info,
315 struct btrfs_block_group_cache *block_group,
316 struct btrfs_path *path)
318 struct btrfs_root *root = fs_info->free_space_root;
319 struct btrfs_free_space_info *info;
320 struct btrfs_key key, found_key;
321 struct extent_buffer *leaf;
322 unsigned long *bitmap;
324 /* Initialize to silence GCC. */
325 u64 extent_start = 0;
327 u32 bitmap_size, flags, expected_extent_count;
328 int prev_bit = 0, bit, bitnr;
329 u32 extent_count = 0;
333 bitmap_size = free_space_bitmap_size(block_group->key.offset,
334 block_group->sectorsize);
335 bitmap = alloc_bitmap(bitmap_size);
341 start = block_group->key.objectid;
342 end = block_group->key.objectid + block_group->key.offset;
344 key.objectid = end - 1;
346 key.offset = (u64)-1;
349 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
353 leaf = path->nodes[0];
356 while (path->slots[0] > 0) {
357 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
359 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
360 ASSERT(found_key.objectid == block_group->key.objectid);
361 ASSERT(found_key.offset == block_group->key.offset);
364 } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
367 u32 bitmap_pos, data_size;
369 ASSERT(found_key.objectid >= start);
370 ASSERT(found_key.objectid < end);
371 ASSERT(found_key.objectid + found_key.offset <= end);
373 bitmap_pos = div_u64(found_key.objectid - start,
374 block_group->sectorsize *
376 bitmap_cursor = ((char *)bitmap) + bitmap_pos;
377 data_size = free_space_bitmap_size(found_key.offset,
378 block_group->sectorsize);
380 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
381 read_extent_buffer(leaf, bitmap_cursor, ptr,
391 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
394 btrfs_release_path(path);
397 info = search_free_space_info(trans, fs_info, block_group, path, 1);
402 leaf = path->nodes[0];
403 flags = btrfs_free_space_flags(leaf, info);
404 flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
405 btrfs_set_free_space_flags(leaf, info, flags);
406 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
407 btrfs_mark_buffer_dirty(leaf);
408 btrfs_release_path(path);
412 while (offset < end) {
413 bit = !!test_bit(bitnr, bitmap);
414 if (prev_bit == 0 && bit == 1) {
415 extent_start = offset;
416 } else if (prev_bit == 1 && bit == 0) {
417 key.objectid = extent_start;
418 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
419 key.offset = offset - extent_start;
421 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
424 btrfs_release_path(path);
429 offset += block_group->sectorsize;
433 key.objectid = extent_start;
434 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
435 key.offset = end - extent_start;
437 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
440 btrfs_release_path(path);
445 if (extent_count != expected_extent_count) {
447 "incorrect extent count for %llu; counted %u, expected %u",
448 block_group->key.objectid, extent_count,
449 expected_extent_count);
459 btrfs_abort_transaction(trans, ret);
463 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
464 struct btrfs_fs_info *fs_info,
465 struct btrfs_block_group_cache *block_group,
466 struct btrfs_path *path,
469 struct btrfs_free_space_info *info;
474 if (new_extents == 0)
477 info = search_free_space_info(trans, fs_info, block_group, path, 1);
482 flags = btrfs_free_space_flags(path->nodes[0], info);
483 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
485 extent_count += new_extents;
486 btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
487 btrfs_mark_buffer_dirty(path->nodes[0]);
488 btrfs_release_path(path);
490 if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
491 extent_count > block_group->bitmap_high_thresh) {
492 ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
494 } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
495 extent_count < block_group->bitmap_low_thresh) {
496 ret = convert_free_space_to_extents(trans, fs_info, block_group,
504 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
505 struct btrfs_path *path, u64 offset)
507 struct extent_buffer *leaf;
508 struct btrfs_key key;
509 u64 found_start, found_end;
510 unsigned long ptr, i;
512 leaf = path->nodes[0];
513 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
514 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
516 found_start = key.objectid;
517 found_end = key.objectid + key.offset;
518 ASSERT(offset >= found_start && offset < found_end);
520 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
521 i = div_u64(offset - found_start, block_group->sectorsize);
522 return !!extent_buffer_test_bit(leaf, ptr, i);
525 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
526 struct btrfs_path *path, u64 *start, u64 *size,
529 struct extent_buffer *leaf;
530 struct btrfs_key key;
531 u64 end = *start + *size;
532 u64 found_start, found_end;
533 unsigned long ptr, first, last;
535 leaf = path->nodes[0];
536 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
537 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
539 found_start = key.objectid;
540 found_end = key.objectid + key.offset;
541 ASSERT(*start >= found_start && *start < found_end);
542 ASSERT(end > found_start);
547 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
548 first = div_u64(*start - found_start, block_group->sectorsize);
549 last = div_u64(end - found_start, block_group->sectorsize);
551 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
553 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
554 btrfs_mark_buffer_dirty(leaf);
556 *size -= end - *start;
561 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
562 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
563 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
566 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
567 struct btrfs_root *root, struct btrfs_path *p)
569 struct btrfs_key key;
571 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
576 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
577 btrfs_release_path(p);
579 key.objectid += key.offset;
581 key.offset = (u64)-1;
583 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
587 * If remove is 1, then we are removing free space, thus clearing bits in the
588 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
591 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
592 struct btrfs_fs_info *fs_info,
593 struct btrfs_block_group_cache *block_group,
594 struct btrfs_path *path,
595 u64 start, u64 size, int remove)
597 struct btrfs_root *root = fs_info->free_space_root;
598 struct btrfs_key key;
599 u64 end = start + size;
600 u64 cur_start, cur_size;
601 int prev_bit, next_bit;
606 * Read the bit for the block immediately before the extent of space if
607 * that block is within the block group.
609 if (start > block_group->key.objectid) {
610 u64 prev_block = start - block_group->sectorsize;
612 key.objectid = prev_block;
614 key.offset = (u64)-1;
616 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
620 prev_bit = free_space_test_bit(block_group, path, prev_block);
622 /* The previous block may have been in the previous bitmap. */
623 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
624 if (start >= key.objectid + key.offset) {
625 ret = free_space_next_bitmap(trans, root, path);
630 key.objectid = start;
632 key.offset = (u64)-1;
634 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
642 * Iterate over all of the bitmaps overlapped by the extent of space,
643 * clearing/setting bits as required.
648 free_space_set_bits(block_group, path, &cur_start, &cur_size,
652 ret = free_space_next_bitmap(trans, root, path);
658 * Read the bit for the block immediately after the extent of space if
659 * that block is within the block group.
661 if (end < block_group->key.objectid + block_group->key.offset) {
662 /* The next block may be in the next bitmap. */
663 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
664 if (end >= key.objectid + key.offset) {
665 ret = free_space_next_bitmap(trans, root, path);
670 next_bit = free_space_test_bit(block_group, path, end);
678 /* Leftover on the left. */
682 /* Leftover on the right. */
688 /* Merging with neighbor on the left. */
692 /* Merging with neighbor on the right. */
697 btrfs_release_path(path);
698 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
705 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
706 struct btrfs_fs_info *fs_info,
707 struct btrfs_block_group_cache *block_group,
708 struct btrfs_path *path,
711 struct btrfs_root *root = fs_info->free_space_root;
712 struct btrfs_key key;
713 u64 found_start, found_end;
714 u64 end = start + size;
715 int new_extents = -1;
718 key.objectid = start;
720 key.offset = (u64)-1;
722 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
726 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
728 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
730 found_start = key.objectid;
731 found_end = key.objectid + key.offset;
732 ASSERT(start >= found_start && end <= found_end);
735 * Okay, now that we've found the free space extent which contains the
736 * free space that we are removing, there are four cases:
738 * 1. We're using the whole extent: delete the key we found and
739 * decrement the free space extent count.
740 * 2. We are using part of the extent starting at the beginning: delete
741 * the key we found and insert a new key representing the leftover at
742 * the end. There is no net change in the number of extents.
743 * 3. We are using part of the extent ending at the end: delete the key
744 * we found and insert a new key representing the leftover at the
745 * beginning. There is no net change in the number of extents.
746 * 4. We are using part of the extent in the middle: delete the key we
747 * found and insert two new keys representing the leftovers on each
748 * side. Where we used to have one extent, we now have two, so increment
749 * the extent count. We may need to convert the block group to bitmaps
753 /* Delete the existing key (cases 1-4). */
754 ret = btrfs_del_item(trans, root, path);
758 /* Add a key for leftovers at the beginning (cases 3 and 4). */
759 if (start > found_start) {
760 key.objectid = found_start;
761 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
762 key.offset = start - found_start;
764 btrfs_release_path(path);
765 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
771 /* Add a key for leftovers at the end (cases 2 and 4). */
772 if (end < found_end) {
774 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
775 key.offset = found_end - end;
777 btrfs_release_path(path);
778 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
784 btrfs_release_path(path);
785 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
792 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
793 struct btrfs_fs_info *fs_info,
794 struct btrfs_block_group_cache *block_group,
795 struct btrfs_path *path, u64 start, u64 size)
797 struct btrfs_free_space_info *info;
801 if (block_group->needs_free_space) {
802 ret = __add_block_group_free_space(trans, fs_info, block_group,
808 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
810 return PTR_ERR(info);
811 flags = btrfs_free_space_flags(path->nodes[0], info);
812 btrfs_release_path(path);
814 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
815 return modify_free_space_bitmap(trans, fs_info, block_group,
816 path, start, size, 1);
818 return remove_free_space_extent(trans, fs_info, block_group,
823 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info,
827 struct btrfs_block_group_cache *block_group;
828 struct btrfs_path *path;
831 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
834 path = btrfs_alloc_path();
840 block_group = btrfs_lookup_block_group(fs_info, start);
847 mutex_lock(&block_group->free_space_lock);
848 ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
850 mutex_unlock(&block_group->free_space_lock);
852 btrfs_put_block_group(block_group);
854 btrfs_free_path(path);
856 btrfs_abort_transaction(trans, ret);
860 static int add_free_space_extent(struct btrfs_trans_handle *trans,
861 struct btrfs_fs_info *fs_info,
862 struct btrfs_block_group_cache *block_group,
863 struct btrfs_path *path,
866 struct btrfs_root *root = fs_info->free_space_root;
867 struct btrfs_key key, new_key;
868 u64 found_start, found_end;
869 u64 end = start + size;
874 * We are adding a new extent of free space, but we need to merge
875 * extents. There are four cases here:
877 * 1. The new extent does not have any immediate neighbors to merge
878 * with: add the new key and increment the free space extent count. We
879 * may need to convert the block group to bitmaps as a result.
880 * 2. The new extent has an immediate neighbor before it: remove the
881 * previous key and insert a new key combining both of them. There is no
882 * net change in the number of extents.
883 * 3. The new extent has an immediate neighbor after it: remove the next
884 * key and insert a new key combining both of them. There is no net
885 * change in the number of extents.
886 * 4. The new extent has immediate neighbors on both sides: remove both
887 * of the keys and insert a new key combining all of them. Where we used
888 * to have two extents, we now have one, so decrement the extent count.
891 new_key.objectid = start;
892 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
893 new_key.offset = size;
895 /* Search for a neighbor on the left. */
896 if (start == block_group->key.objectid)
898 key.objectid = start - 1;
900 key.offset = (u64)-1;
902 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
906 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
908 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
909 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
910 btrfs_release_path(path);
914 found_start = key.objectid;
915 found_end = key.objectid + key.offset;
916 ASSERT(found_start >= block_group->key.objectid &&
917 found_end > block_group->key.objectid);
918 ASSERT(found_start < start && found_end <= start);
921 * Delete the neighbor on the left and absorb it into the new key (cases
924 if (found_end == start) {
925 ret = btrfs_del_item(trans, root, path);
928 new_key.objectid = found_start;
929 new_key.offset += key.offset;
932 btrfs_release_path(path);
935 /* Search for a neighbor on the right. */
936 if (end == block_group->key.objectid + block_group->key.offset)
940 key.offset = (u64)-1;
942 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
946 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
948 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
949 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
950 btrfs_release_path(path);
954 found_start = key.objectid;
955 found_end = key.objectid + key.offset;
956 ASSERT(found_start >= block_group->key.objectid &&
957 found_end > block_group->key.objectid);
958 ASSERT((found_start < start && found_end <= start) ||
959 (found_start >= end && found_end > end));
962 * Delete the neighbor on the right and absorb it into the new key
965 if (found_start == end) {
966 ret = btrfs_del_item(trans, root, path);
969 new_key.offset += key.offset;
972 btrfs_release_path(path);
975 /* Insert the new key (cases 1-4). */
976 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
980 btrfs_release_path(path);
981 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
988 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
989 struct btrfs_fs_info *fs_info,
990 struct btrfs_block_group_cache *block_group,
991 struct btrfs_path *path, u64 start, u64 size)
993 struct btrfs_free_space_info *info;
997 if (block_group->needs_free_space) {
998 ret = __add_block_group_free_space(trans, fs_info, block_group,
1004 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1006 return PTR_ERR(info);
1007 flags = btrfs_free_space_flags(path->nodes[0], info);
1008 btrfs_release_path(path);
1010 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1011 return modify_free_space_bitmap(trans, fs_info, block_group,
1012 path, start, size, 0);
1014 return add_free_space_extent(trans, fs_info, block_group, path,
1019 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1020 struct btrfs_fs_info *fs_info,
1021 u64 start, u64 size)
1023 struct btrfs_block_group_cache *block_group;
1024 struct btrfs_path *path;
1027 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1030 path = btrfs_alloc_path();
1036 block_group = btrfs_lookup_block_group(fs_info, start);
1043 mutex_lock(&block_group->free_space_lock);
1044 ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
1046 mutex_unlock(&block_group->free_space_lock);
1048 btrfs_put_block_group(block_group);
1050 btrfs_free_path(path);
1052 btrfs_abort_transaction(trans, ret);
1057 * Populate the free space tree by walking the extent tree. Operations on the
1058 * extent tree that happen as a result of writes to the free space tree will go
1059 * through the normal add/remove hooks.
1061 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1062 struct btrfs_fs_info *fs_info,
1063 struct btrfs_block_group_cache *block_group)
1065 struct btrfs_root *extent_root = fs_info->extent_root;
1066 struct btrfs_path *path, *path2;
1067 struct btrfs_key key;
1071 path = btrfs_alloc_path();
1076 path2 = btrfs_alloc_path();
1078 btrfs_free_path(path);
1082 ret = add_new_free_space_info(trans, fs_info, block_group, path2);
1086 mutex_lock(&block_group->free_space_lock);
1089 * Iterate through all of the extent and metadata items in this block
1090 * group, adding the free space between them and the free space at the
1091 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1092 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1095 key.objectid = block_group->key.objectid;
1096 key.type = BTRFS_EXTENT_ITEM_KEY;
1099 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1104 start = block_group->key.objectid;
1105 end = block_group->key.objectid + block_group->key.offset;
1107 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1109 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1110 key.type == BTRFS_METADATA_ITEM_KEY) {
1111 if (key.objectid >= end)
1114 if (start < key.objectid) {
1115 ret = __add_to_free_space_tree(trans, fs_info,
1123 start = key.objectid;
1124 if (key.type == BTRFS_METADATA_ITEM_KEY)
1125 start += fs_info->tree_root->nodesize;
1127 start += key.offset;
1128 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1129 if (key.objectid != block_group->key.objectid)
1133 ret = btrfs_next_item(extent_root, path);
1140 ret = __add_to_free_space_tree(trans, fs_info, block_group,
1141 path2, start, end - start);
1148 mutex_unlock(&block_group->free_space_lock);
1150 btrfs_free_path(path2);
1151 btrfs_free_path(path);
1155 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1157 struct btrfs_trans_handle *trans;
1158 struct btrfs_root *tree_root = fs_info->tree_root;
1159 struct btrfs_root *free_space_root;
1160 struct btrfs_block_group_cache *block_group;
1161 struct rb_node *node;
1164 trans = btrfs_start_transaction(tree_root, 0);
1166 return PTR_ERR(trans);
1168 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1169 free_space_root = btrfs_create_tree(trans, fs_info,
1170 BTRFS_FREE_SPACE_TREE_OBJECTID);
1171 if (IS_ERR(free_space_root)) {
1172 ret = PTR_ERR(free_space_root);
1175 fs_info->free_space_root = free_space_root;
1177 node = rb_first(&fs_info->block_group_cache_tree);
1179 block_group = rb_entry(node, struct btrfs_block_group_cache,
1181 ret = populate_free_space_tree(trans, fs_info, block_group);
1184 node = rb_next(node);
1187 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1188 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1190 ret = btrfs_commit_transaction(trans, tree_root);
1197 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1198 btrfs_abort_transaction(trans, ret);
1199 btrfs_end_transaction(trans, tree_root);
1203 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1204 struct btrfs_root *root)
1206 struct btrfs_path *path;
1207 struct btrfs_key key;
1211 path = btrfs_alloc_path();
1215 path->leave_spinning = 1;
1222 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1226 nr = btrfs_header_nritems(path->nodes[0]);
1231 ret = btrfs_del_items(trans, root, path, 0, nr);
1235 btrfs_release_path(path);
1240 btrfs_free_path(path);
1244 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1246 struct btrfs_trans_handle *trans;
1247 struct btrfs_root *tree_root = fs_info->tree_root;
1248 struct btrfs_root *free_space_root = fs_info->free_space_root;
1251 trans = btrfs_start_transaction(tree_root, 0);
1253 return PTR_ERR(trans);
1255 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1256 fs_info->free_space_root = NULL;
1258 ret = clear_free_space_tree(trans, free_space_root);
1262 ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
1266 list_del(&free_space_root->dirty_list);
1268 btrfs_tree_lock(free_space_root->node);
1269 clean_tree_block(trans, tree_root->fs_info, free_space_root->node);
1270 btrfs_tree_unlock(free_space_root->node);
1271 btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1274 free_extent_buffer(free_space_root->node);
1275 free_extent_buffer(free_space_root->commit_root);
1276 kfree(free_space_root);
1278 ret = btrfs_commit_transaction(trans, tree_root);
1285 btrfs_abort_transaction(trans, ret);
1286 btrfs_end_transaction(trans, tree_root);
1290 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1291 struct btrfs_fs_info *fs_info,
1292 struct btrfs_block_group_cache *block_group,
1293 struct btrfs_path *path)
1298 start = block_group->key.objectid;
1299 end = block_group->key.objectid + block_group->key.offset;
1301 block_group->needs_free_space = 0;
1303 ret = add_new_free_space_info(trans, fs_info, block_group, path);
1307 return __add_to_free_space_tree(trans, fs_info, block_group, path,
1308 block_group->key.objectid,
1309 block_group->key.offset);
1312 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1313 struct btrfs_fs_info *fs_info,
1314 struct btrfs_block_group_cache *block_group)
1316 struct btrfs_path *path = NULL;
1319 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1322 mutex_lock(&block_group->free_space_lock);
1323 if (!block_group->needs_free_space)
1326 path = btrfs_alloc_path();
1332 ret = __add_block_group_free_space(trans, fs_info, block_group, path);
1335 btrfs_free_path(path);
1336 mutex_unlock(&block_group->free_space_lock);
1338 btrfs_abort_transaction(trans, ret);
1342 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1343 struct btrfs_fs_info *fs_info,
1344 struct btrfs_block_group_cache *block_group)
1346 struct btrfs_root *root = fs_info->free_space_root;
1347 struct btrfs_path *path;
1348 struct btrfs_key key, found_key;
1349 struct extent_buffer *leaf;
1354 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1357 if (block_group->needs_free_space) {
1358 /* We never added this block group to the free space tree. */
1362 path = btrfs_alloc_path();
1368 start = block_group->key.objectid;
1369 end = block_group->key.objectid + block_group->key.offset;
1371 key.objectid = end - 1;
1373 key.offset = (u64)-1;
1376 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1380 leaf = path->nodes[0];
1383 while (path->slots[0] > 0) {
1384 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1386 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1387 ASSERT(found_key.objectid == block_group->key.objectid);
1388 ASSERT(found_key.offset == block_group->key.offset);
1393 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1394 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1395 ASSERT(found_key.objectid >= start);
1396 ASSERT(found_key.objectid < end);
1397 ASSERT(found_key.objectid + found_key.offset <= end);
1405 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1408 btrfs_release_path(path);
1413 btrfs_free_path(path);
1415 btrfs_abort_transaction(trans, ret);
1419 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1420 struct btrfs_path *path,
1421 u32 expected_extent_count)
1423 struct btrfs_block_group_cache *block_group;
1424 struct btrfs_fs_info *fs_info;
1425 struct btrfs_root *root;
1426 struct btrfs_key key;
1427 int prev_bit = 0, bit;
1428 /* Initialize to silence GCC. */
1429 u64 extent_start = 0;
1431 u64 total_found = 0;
1432 u32 extent_count = 0;
1435 block_group = caching_ctl->block_group;
1436 fs_info = block_group->fs_info;
1437 root = fs_info->free_space_root;
1439 end = block_group->key.objectid + block_group->key.offset;
1442 ret = btrfs_next_item(root, path);
1448 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1450 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1453 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1454 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1456 caching_ctl->progress = key.objectid;
1458 offset = key.objectid;
1459 while (offset < key.objectid + key.offset) {
1460 bit = free_space_test_bit(block_group, path, offset);
1461 if (prev_bit == 0 && bit == 1) {
1462 extent_start = offset;
1463 } else if (prev_bit == 1 && bit == 0) {
1464 total_found += add_new_free_space(block_group,
1468 if (total_found > CACHING_CTL_WAKE_UP) {
1470 wake_up(&caching_ctl->wait);
1475 offset += block_group->sectorsize;
1478 if (prev_bit == 1) {
1479 total_found += add_new_free_space(block_group, fs_info,
1484 if (extent_count != expected_extent_count) {
1486 "incorrect extent count for %llu; counted %u, expected %u",
1487 block_group->key.objectid, extent_count,
1488 expected_extent_count);
1494 caching_ctl->progress = (u64)-1;
1501 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1502 struct btrfs_path *path,
1503 u32 expected_extent_count)
1505 struct btrfs_block_group_cache *block_group;
1506 struct btrfs_fs_info *fs_info;
1507 struct btrfs_root *root;
1508 struct btrfs_key key;
1510 u64 total_found = 0;
1511 u32 extent_count = 0;
1514 block_group = caching_ctl->block_group;
1515 fs_info = block_group->fs_info;
1516 root = fs_info->free_space_root;
1518 end = block_group->key.objectid + block_group->key.offset;
1521 ret = btrfs_next_item(root, path);
1527 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1529 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1532 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1533 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1535 caching_ctl->progress = key.objectid;
1537 total_found += add_new_free_space(block_group, fs_info,
1539 key.objectid + key.offset);
1540 if (total_found > CACHING_CTL_WAKE_UP) {
1542 wake_up(&caching_ctl->wait);
1547 if (extent_count != expected_extent_count) {
1549 "incorrect extent count for %llu; counted %u, expected %u",
1550 block_group->key.objectid, extent_count,
1551 expected_extent_count);
1557 caching_ctl->progress = (u64)-1;
1564 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1566 struct btrfs_block_group_cache *block_group;
1567 struct btrfs_fs_info *fs_info;
1568 struct btrfs_free_space_info *info;
1569 struct btrfs_path *path;
1570 u32 extent_count, flags;
1573 block_group = caching_ctl->block_group;
1574 fs_info = block_group->fs_info;
1576 path = btrfs_alloc_path();
1581 * Just like caching_thread() doesn't want to deadlock on the extent
1582 * tree, we don't want to deadlock on the free space tree.
1584 path->skip_locking = 1;
1585 path->search_commit_root = 1;
1588 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1590 ret = PTR_ERR(info);
1593 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1594 flags = btrfs_free_space_flags(path->nodes[0], info);
1597 * We left path pointing to the free space info item, so now
1598 * load_free_space_foo can just iterate through the free space tree from
1601 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1602 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1604 ret = load_free_space_extents(caching_ctl, path, extent_count);
1607 btrfs_free_path(path);