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\n",
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 u8 *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 u8 *bitmap, *bitmap_cursor;
186 u32 bitmap_size, flags, expected_extent_count;
187 u32 extent_count = 0;
191 bitmap_size = free_space_bitmap_size(block_group->key.offset,
192 block_group->sectorsize);
193 bitmap = alloc_bitmap(bitmap_size);
199 start = block_group->key.objectid;
200 end = block_group->key.objectid + block_group->key.offset;
202 key.objectid = end - 1;
204 key.offset = (u64)-1;
207 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
211 leaf = path->nodes[0];
214 while (path->slots[0] > 0) {
215 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
217 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
218 ASSERT(found_key.objectid == block_group->key.objectid);
219 ASSERT(found_key.offset == block_group->key.offset);
222 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
225 ASSERT(found_key.objectid >= start);
226 ASSERT(found_key.objectid < end);
227 ASSERT(found_key.objectid + found_key.offset <= end);
229 first = div_u64(found_key.objectid - start,
230 block_group->sectorsize);
231 last = div_u64(found_key.objectid + found_key.offset - start,
232 block_group->sectorsize);
233 le_bitmap_set(bitmap, first, last - first);
243 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
246 btrfs_release_path(path);
249 info = search_free_space_info(trans, fs_info, block_group, path, 1);
254 leaf = path->nodes[0];
255 flags = btrfs_free_space_flags(leaf, info);
256 flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
257 btrfs_set_free_space_flags(leaf, info, flags);
258 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
259 btrfs_mark_buffer_dirty(leaf);
260 btrfs_release_path(path);
262 if (extent_count != expected_extent_count) {
263 btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
264 block_group->key.objectid, extent_count,
265 expected_extent_count);
271 bitmap_cursor = bitmap;
272 bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
279 extent_size = min(end - i, bitmap_range);
280 data_size = free_space_bitmap_size(extent_size,
281 block_group->sectorsize);
284 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
285 key.offset = extent_size;
287 ret = btrfs_insert_empty_item(trans, root, path, &key,
292 leaf = path->nodes[0];
293 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
294 write_extent_buffer(leaf, bitmap_cursor, ptr,
296 btrfs_mark_buffer_dirty(leaf);
297 btrfs_release_path(path);
300 bitmap_cursor += data_size;
307 btrfs_abort_transaction(trans, ret);
311 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
312 struct btrfs_fs_info *fs_info,
313 struct btrfs_block_group_cache *block_group,
314 struct btrfs_path *path)
316 struct btrfs_root *root = fs_info->free_space_root;
317 struct btrfs_free_space_info *info;
318 struct btrfs_key key, found_key;
319 struct extent_buffer *leaf;
322 /* Initialize to silence GCC. */
323 u64 extent_start = 0;
325 u32 bitmap_size, flags, expected_extent_count;
326 int prev_bit = 0, bit, bitnr;
327 u32 extent_count = 0;
331 bitmap_size = free_space_bitmap_size(block_group->key.offset,
332 block_group->sectorsize);
333 bitmap = alloc_bitmap(bitmap_size);
339 start = block_group->key.objectid;
340 end = block_group->key.objectid + block_group->key.offset;
342 key.objectid = end - 1;
344 key.offset = (u64)-1;
347 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
351 leaf = path->nodes[0];
354 while (path->slots[0] > 0) {
355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
357 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
358 ASSERT(found_key.objectid == block_group->key.objectid);
359 ASSERT(found_key.offset == block_group->key.offset);
362 } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
365 u32 bitmap_pos, data_size;
367 ASSERT(found_key.objectid >= start);
368 ASSERT(found_key.objectid < end);
369 ASSERT(found_key.objectid + found_key.offset <= end);
371 bitmap_pos = div_u64(found_key.objectid - start,
372 block_group->sectorsize *
374 bitmap_cursor = bitmap + bitmap_pos;
375 data_size = free_space_bitmap_size(found_key.offset,
376 block_group->sectorsize);
378 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
379 read_extent_buffer(leaf, bitmap_cursor, ptr,
389 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
392 btrfs_release_path(path);
395 info = search_free_space_info(trans, fs_info, block_group, path, 1);
400 leaf = path->nodes[0];
401 flags = btrfs_free_space_flags(leaf, info);
402 flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
403 btrfs_set_free_space_flags(leaf, info, flags);
404 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
405 btrfs_mark_buffer_dirty(leaf);
406 btrfs_release_path(path);
410 while (offset < end) {
411 bit = !!le_test_bit(bitnr, bitmap);
412 if (prev_bit == 0 && bit == 1) {
413 extent_start = offset;
414 } else if (prev_bit == 1 && bit == 0) {
415 key.objectid = extent_start;
416 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
417 key.offset = offset - extent_start;
419 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
422 btrfs_release_path(path);
427 offset += block_group->sectorsize;
431 key.objectid = extent_start;
432 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
433 key.offset = end - extent_start;
435 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
438 btrfs_release_path(path);
443 if (extent_count != expected_extent_count) {
444 btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
445 block_group->key.objectid, extent_count,
446 expected_extent_count);
456 btrfs_abort_transaction(trans, ret);
460 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
461 struct btrfs_fs_info *fs_info,
462 struct btrfs_block_group_cache *block_group,
463 struct btrfs_path *path,
466 struct btrfs_free_space_info *info;
471 if (new_extents == 0)
474 info = search_free_space_info(trans, fs_info, block_group, path, 1);
479 flags = btrfs_free_space_flags(path->nodes[0], info);
480 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
482 extent_count += new_extents;
483 btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
484 btrfs_mark_buffer_dirty(path->nodes[0]);
485 btrfs_release_path(path);
487 if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
488 extent_count > block_group->bitmap_high_thresh) {
489 ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
491 } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
492 extent_count < block_group->bitmap_low_thresh) {
493 ret = convert_free_space_to_extents(trans, fs_info, block_group,
501 int free_space_test_bit(struct btrfs_block_group_cache *block_group,
502 struct btrfs_path *path, u64 offset)
504 struct extent_buffer *leaf;
505 struct btrfs_key key;
506 u64 found_start, found_end;
507 unsigned long ptr, i;
509 leaf = path->nodes[0];
510 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
511 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
513 found_start = key.objectid;
514 found_end = key.objectid + key.offset;
515 ASSERT(offset >= found_start && offset < found_end);
517 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
518 i = div_u64(offset - found_start, block_group->sectorsize);
519 return !!extent_buffer_test_bit(leaf, ptr, i);
522 static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
523 struct btrfs_path *path, u64 *start, u64 *size,
526 struct extent_buffer *leaf;
527 struct btrfs_key key;
528 u64 end = *start + *size;
529 u64 found_start, found_end;
530 unsigned long ptr, first, last;
532 leaf = path->nodes[0];
533 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
534 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
536 found_start = key.objectid;
537 found_end = key.objectid + key.offset;
538 ASSERT(*start >= found_start && *start < found_end);
539 ASSERT(end > found_start);
544 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
545 first = div_u64(*start - found_start, block_group->sectorsize);
546 last = div_u64(end - found_start, block_group->sectorsize);
548 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
550 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
551 btrfs_mark_buffer_dirty(leaf);
553 *size -= end - *start;
558 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
559 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
560 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
563 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
564 struct btrfs_root *root, struct btrfs_path *p)
566 struct btrfs_key key;
568 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
573 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
574 btrfs_release_path(p);
576 key.objectid += key.offset;
578 key.offset = (u64)-1;
580 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
584 * If remove is 1, then we are removing free space, thus clearing bits in the
585 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
588 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
589 struct btrfs_fs_info *fs_info,
590 struct btrfs_block_group_cache *block_group,
591 struct btrfs_path *path,
592 u64 start, u64 size, int remove)
594 struct btrfs_root *root = fs_info->free_space_root;
595 struct btrfs_key key;
596 u64 end = start + size;
597 u64 cur_start, cur_size;
598 int prev_bit, next_bit;
603 * Read the bit for the block immediately before the extent of space if
604 * that block is within the block group.
606 if (start > block_group->key.objectid) {
607 u64 prev_block = start - block_group->sectorsize;
609 key.objectid = prev_block;
611 key.offset = (u64)-1;
613 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
617 prev_bit = free_space_test_bit(block_group, path, prev_block);
619 /* The previous block may have been in the previous bitmap. */
620 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
621 if (start >= key.objectid + key.offset) {
622 ret = free_space_next_bitmap(trans, root, path);
627 key.objectid = start;
629 key.offset = (u64)-1;
631 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
639 * Iterate over all of the bitmaps overlapped by the extent of space,
640 * clearing/setting bits as required.
645 free_space_set_bits(block_group, path, &cur_start, &cur_size,
649 ret = free_space_next_bitmap(trans, root, path);
655 * Read the bit for the block immediately after the extent of space if
656 * that block is within the block group.
658 if (end < block_group->key.objectid + block_group->key.offset) {
659 /* The next block may be in the next bitmap. */
660 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
661 if (end >= key.objectid + key.offset) {
662 ret = free_space_next_bitmap(trans, root, path);
667 next_bit = free_space_test_bit(block_group, path, end);
675 /* Leftover on the left. */
679 /* Leftover on the right. */
685 /* Merging with neighbor on the left. */
689 /* Merging with neighbor on the right. */
694 btrfs_release_path(path);
695 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
702 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
703 struct btrfs_fs_info *fs_info,
704 struct btrfs_block_group_cache *block_group,
705 struct btrfs_path *path,
708 struct btrfs_root *root = fs_info->free_space_root;
709 struct btrfs_key key;
710 u64 found_start, found_end;
711 u64 end = start + size;
712 int new_extents = -1;
715 key.objectid = start;
717 key.offset = (u64)-1;
719 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
723 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
725 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
727 found_start = key.objectid;
728 found_end = key.objectid + key.offset;
729 ASSERT(start >= found_start && end <= found_end);
732 * Okay, now that we've found the free space extent which contains the
733 * free space that we are removing, there are four cases:
735 * 1. We're using the whole extent: delete the key we found and
736 * decrement the free space extent count.
737 * 2. We are using part of the extent starting at the beginning: delete
738 * the key we found and insert a new key representing the leftover at
739 * the end. There is no net change in the number of extents.
740 * 3. We are using part of the extent ending at the end: delete the key
741 * we found and insert a new key representing the leftover at the
742 * beginning. There is no net change in the number of extents.
743 * 4. We are using part of the extent in the middle: delete the key we
744 * found and insert two new keys representing the leftovers on each
745 * side. Where we used to have one extent, we now have two, so increment
746 * the extent count. We may need to convert the block group to bitmaps
750 /* Delete the existing key (cases 1-4). */
751 ret = btrfs_del_item(trans, root, path);
755 /* Add a key for leftovers at the beginning (cases 3 and 4). */
756 if (start > found_start) {
757 key.objectid = found_start;
758 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
759 key.offset = start - found_start;
761 btrfs_release_path(path);
762 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
768 /* Add a key for leftovers at the end (cases 2 and 4). */
769 if (end < found_end) {
771 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
772 key.offset = found_end - end;
774 btrfs_release_path(path);
775 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
781 btrfs_release_path(path);
782 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
789 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
790 struct btrfs_fs_info *fs_info,
791 struct btrfs_block_group_cache *block_group,
792 struct btrfs_path *path, u64 start, u64 size)
794 struct btrfs_free_space_info *info;
798 if (block_group->needs_free_space) {
799 ret = __add_block_group_free_space(trans, fs_info, block_group,
805 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
807 return PTR_ERR(info);
808 flags = btrfs_free_space_flags(path->nodes[0], info);
809 btrfs_release_path(path);
811 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
812 return modify_free_space_bitmap(trans, fs_info, block_group,
813 path, start, size, 1);
815 return remove_free_space_extent(trans, fs_info, block_group,
820 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
821 struct btrfs_fs_info *fs_info,
824 struct btrfs_block_group_cache *block_group;
825 struct btrfs_path *path;
828 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
831 path = btrfs_alloc_path();
837 block_group = btrfs_lookup_block_group(fs_info, start);
844 mutex_lock(&block_group->free_space_lock);
845 ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
847 mutex_unlock(&block_group->free_space_lock);
849 btrfs_put_block_group(block_group);
851 btrfs_free_path(path);
853 btrfs_abort_transaction(trans, ret);
857 static int add_free_space_extent(struct btrfs_trans_handle *trans,
858 struct btrfs_fs_info *fs_info,
859 struct btrfs_block_group_cache *block_group,
860 struct btrfs_path *path,
863 struct btrfs_root *root = fs_info->free_space_root;
864 struct btrfs_key key, new_key;
865 u64 found_start, found_end;
866 u64 end = start + size;
871 * We are adding a new extent of free space, but we need to merge
872 * extents. There are four cases here:
874 * 1. The new extent does not have any immediate neighbors to merge
875 * with: add the new key and increment the free space extent count. We
876 * may need to convert the block group to bitmaps as a result.
877 * 2. The new extent has an immediate neighbor before it: remove the
878 * previous key and insert a new key combining both of them. There is no
879 * net change in the number of extents.
880 * 3. The new extent has an immediate neighbor after it: remove the next
881 * key and insert a new key combining both of them. There is no net
882 * change in the number of extents.
883 * 4. The new extent has immediate neighbors on both sides: remove both
884 * of the keys and insert a new key combining all of them. Where we used
885 * to have two extents, we now have one, so decrement the extent count.
888 new_key.objectid = start;
889 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
890 new_key.offset = size;
892 /* Search for a neighbor on the left. */
893 if (start == block_group->key.objectid)
895 key.objectid = start - 1;
897 key.offset = (u64)-1;
899 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
903 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
905 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
906 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
907 btrfs_release_path(path);
911 found_start = key.objectid;
912 found_end = key.objectid + key.offset;
913 ASSERT(found_start >= block_group->key.objectid &&
914 found_end > block_group->key.objectid);
915 ASSERT(found_start < start && found_end <= start);
918 * Delete the neighbor on the left and absorb it into the new key (cases
921 if (found_end == start) {
922 ret = btrfs_del_item(trans, root, path);
925 new_key.objectid = found_start;
926 new_key.offset += key.offset;
929 btrfs_release_path(path);
932 /* Search for a neighbor on the right. */
933 if (end == block_group->key.objectid + block_group->key.offset)
937 key.offset = (u64)-1;
939 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
943 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
945 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
946 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
947 btrfs_release_path(path);
951 found_start = key.objectid;
952 found_end = key.objectid + key.offset;
953 ASSERT(found_start >= block_group->key.objectid &&
954 found_end > block_group->key.objectid);
955 ASSERT((found_start < start && found_end <= start) ||
956 (found_start >= end && found_end > end));
959 * Delete the neighbor on the right and absorb it into the new key
962 if (found_start == end) {
963 ret = btrfs_del_item(trans, root, path);
966 new_key.offset += key.offset;
969 btrfs_release_path(path);
972 /* Insert the new key (cases 1-4). */
973 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
977 btrfs_release_path(path);
978 ret = update_free_space_extent_count(trans, fs_info, block_group, path,
985 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
986 struct btrfs_fs_info *fs_info,
987 struct btrfs_block_group_cache *block_group,
988 struct btrfs_path *path, u64 start, u64 size)
990 struct btrfs_free_space_info *info;
994 if (block_group->needs_free_space) {
995 ret = __add_block_group_free_space(trans, fs_info, block_group,
1001 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1003 return PTR_ERR(info);
1004 flags = btrfs_free_space_flags(path->nodes[0], info);
1005 btrfs_release_path(path);
1007 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1008 return modify_free_space_bitmap(trans, fs_info, block_group,
1009 path, start, size, 0);
1011 return add_free_space_extent(trans, fs_info, block_group, path,
1016 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1017 struct btrfs_fs_info *fs_info,
1018 u64 start, u64 size)
1020 struct btrfs_block_group_cache *block_group;
1021 struct btrfs_path *path;
1024 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1027 path = btrfs_alloc_path();
1033 block_group = btrfs_lookup_block_group(fs_info, start);
1040 mutex_lock(&block_group->free_space_lock);
1041 ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
1043 mutex_unlock(&block_group->free_space_lock);
1045 btrfs_put_block_group(block_group);
1047 btrfs_free_path(path);
1049 btrfs_abort_transaction(trans, ret);
1054 * Populate the free space tree by walking the extent tree. Operations on the
1055 * extent tree that happen as a result of writes to the free space tree will go
1056 * through the normal add/remove hooks.
1058 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1059 struct btrfs_fs_info *fs_info,
1060 struct btrfs_block_group_cache *block_group)
1062 struct btrfs_root *extent_root = fs_info->extent_root;
1063 struct btrfs_path *path, *path2;
1064 struct btrfs_key key;
1068 path = btrfs_alloc_path();
1073 path2 = btrfs_alloc_path();
1075 btrfs_free_path(path);
1079 ret = add_new_free_space_info(trans, fs_info, block_group, path2);
1083 mutex_lock(&block_group->free_space_lock);
1086 * Iterate through all of the extent and metadata items in this block
1087 * group, adding the free space between them and the free space at the
1088 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1089 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1092 key.objectid = block_group->key.objectid;
1093 key.type = BTRFS_EXTENT_ITEM_KEY;
1096 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1101 start = block_group->key.objectid;
1102 end = block_group->key.objectid + block_group->key.offset;
1104 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1106 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1107 key.type == BTRFS_METADATA_ITEM_KEY) {
1108 if (key.objectid >= end)
1111 if (start < key.objectid) {
1112 ret = __add_to_free_space_tree(trans, fs_info,
1120 start = key.objectid;
1121 if (key.type == BTRFS_METADATA_ITEM_KEY)
1122 start += fs_info->tree_root->nodesize;
1124 start += key.offset;
1125 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1126 if (key.objectid != block_group->key.objectid)
1130 ret = btrfs_next_item(extent_root, path);
1137 ret = __add_to_free_space_tree(trans, fs_info, block_group,
1138 path2, start, end - start);
1145 mutex_unlock(&block_group->free_space_lock);
1147 btrfs_free_path(path2);
1148 btrfs_free_path(path);
1152 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1154 struct btrfs_trans_handle *trans;
1155 struct btrfs_root *tree_root = fs_info->tree_root;
1156 struct btrfs_root *free_space_root;
1157 struct btrfs_block_group_cache *block_group;
1158 struct rb_node *node;
1161 trans = btrfs_start_transaction(tree_root, 0);
1163 return PTR_ERR(trans);
1165 fs_info->creating_free_space_tree = 1;
1166 free_space_root = btrfs_create_tree(trans, fs_info,
1167 BTRFS_FREE_SPACE_TREE_OBJECTID);
1168 if (IS_ERR(free_space_root)) {
1169 ret = PTR_ERR(free_space_root);
1172 fs_info->free_space_root = free_space_root;
1174 node = rb_first(&fs_info->block_group_cache_tree);
1176 block_group = rb_entry(node, struct btrfs_block_group_cache,
1178 ret = populate_free_space_tree(trans, fs_info, block_group);
1181 node = rb_next(node);
1184 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1185 fs_info->creating_free_space_tree = 0;
1187 ret = btrfs_commit_transaction(trans, tree_root);
1194 fs_info->creating_free_space_tree = 0;
1195 btrfs_abort_transaction(trans, ret);
1196 btrfs_end_transaction(trans, tree_root);
1200 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1201 struct btrfs_root *root)
1203 struct btrfs_path *path;
1204 struct btrfs_key key;
1208 path = btrfs_alloc_path();
1212 path->leave_spinning = 1;
1219 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1223 nr = btrfs_header_nritems(path->nodes[0]);
1228 ret = btrfs_del_items(trans, root, path, 0, nr);
1232 btrfs_release_path(path);
1237 btrfs_free_path(path);
1241 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1243 struct btrfs_trans_handle *trans;
1244 struct btrfs_root *tree_root = fs_info->tree_root;
1245 struct btrfs_root *free_space_root = fs_info->free_space_root;
1248 trans = btrfs_start_transaction(tree_root, 0);
1250 return PTR_ERR(trans);
1252 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1253 fs_info->free_space_root = NULL;
1255 ret = clear_free_space_tree(trans, free_space_root);
1259 ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
1263 list_del(&free_space_root->dirty_list);
1265 btrfs_tree_lock(free_space_root->node);
1266 clean_tree_block(trans, tree_root->fs_info, free_space_root->node);
1267 btrfs_tree_unlock(free_space_root->node);
1268 btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1271 free_extent_buffer(free_space_root->node);
1272 free_extent_buffer(free_space_root->commit_root);
1273 kfree(free_space_root);
1275 ret = btrfs_commit_transaction(trans, tree_root);
1282 btrfs_abort_transaction(trans, ret);
1283 btrfs_end_transaction(trans, tree_root);
1287 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1288 struct btrfs_fs_info *fs_info,
1289 struct btrfs_block_group_cache *block_group,
1290 struct btrfs_path *path)
1295 start = block_group->key.objectid;
1296 end = block_group->key.objectid + block_group->key.offset;
1298 block_group->needs_free_space = 0;
1300 ret = add_new_free_space_info(trans, fs_info, block_group, path);
1304 return __add_to_free_space_tree(trans, fs_info, block_group, path,
1305 block_group->key.objectid,
1306 block_group->key.offset);
1309 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1310 struct btrfs_fs_info *fs_info,
1311 struct btrfs_block_group_cache *block_group)
1313 struct btrfs_path *path = NULL;
1316 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1319 mutex_lock(&block_group->free_space_lock);
1320 if (!block_group->needs_free_space)
1323 path = btrfs_alloc_path();
1329 ret = __add_block_group_free_space(trans, fs_info, block_group, path);
1332 btrfs_free_path(path);
1333 mutex_unlock(&block_group->free_space_lock);
1335 btrfs_abort_transaction(trans, ret);
1339 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1340 struct btrfs_fs_info *fs_info,
1341 struct btrfs_block_group_cache *block_group)
1343 struct btrfs_root *root = fs_info->free_space_root;
1344 struct btrfs_path *path;
1345 struct btrfs_key key, found_key;
1346 struct extent_buffer *leaf;
1351 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1354 if (block_group->needs_free_space) {
1355 /* We never added this block group to the free space tree. */
1359 path = btrfs_alloc_path();
1365 start = block_group->key.objectid;
1366 end = block_group->key.objectid + block_group->key.offset;
1368 key.objectid = end - 1;
1370 key.offset = (u64)-1;
1373 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1377 leaf = path->nodes[0];
1380 while (path->slots[0] > 0) {
1381 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1383 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1384 ASSERT(found_key.objectid == block_group->key.objectid);
1385 ASSERT(found_key.offset == block_group->key.offset);
1390 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1391 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1392 ASSERT(found_key.objectid >= start);
1393 ASSERT(found_key.objectid < end);
1394 ASSERT(found_key.objectid + found_key.offset <= end);
1402 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1405 btrfs_release_path(path);
1410 btrfs_free_path(path);
1412 btrfs_abort_transaction(trans, ret);
1416 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1417 struct btrfs_path *path,
1418 u32 expected_extent_count)
1420 struct btrfs_block_group_cache *block_group;
1421 struct btrfs_fs_info *fs_info;
1422 struct btrfs_root *root;
1423 struct btrfs_key key;
1424 int prev_bit = 0, bit;
1425 /* Initialize to silence GCC. */
1426 u64 extent_start = 0;
1428 u64 total_found = 0;
1429 u32 extent_count = 0;
1432 block_group = caching_ctl->block_group;
1433 fs_info = block_group->fs_info;
1434 root = fs_info->free_space_root;
1436 end = block_group->key.objectid + block_group->key.offset;
1439 ret = btrfs_next_item(root, path);
1445 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1447 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1450 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1451 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1453 caching_ctl->progress = key.objectid;
1455 offset = key.objectid;
1456 while (offset < key.objectid + key.offset) {
1457 bit = free_space_test_bit(block_group, path, offset);
1458 if (prev_bit == 0 && bit == 1) {
1459 extent_start = offset;
1460 } else if (prev_bit == 1 && bit == 0) {
1461 total_found += add_new_free_space(block_group,
1465 if (total_found > CACHING_CTL_WAKE_UP) {
1467 wake_up(&caching_ctl->wait);
1472 offset += block_group->sectorsize;
1475 if (prev_bit == 1) {
1476 total_found += add_new_free_space(block_group, fs_info,
1481 if (extent_count != expected_extent_count) {
1482 btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
1483 block_group->key.objectid, extent_count,
1484 expected_extent_count);
1490 caching_ctl->progress = (u64)-1;
1497 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1498 struct btrfs_path *path,
1499 u32 expected_extent_count)
1501 struct btrfs_block_group_cache *block_group;
1502 struct btrfs_fs_info *fs_info;
1503 struct btrfs_root *root;
1504 struct btrfs_key key;
1506 u64 total_found = 0;
1507 u32 extent_count = 0;
1510 block_group = caching_ctl->block_group;
1511 fs_info = block_group->fs_info;
1512 root = fs_info->free_space_root;
1514 end = block_group->key.objectid + block_group->key.offset;
1517 ret = btrfs_next_item(root, path);
1523 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1525 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1528 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1529 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1531 caching_ctl->progress = key.objectid;
1533 total_found += add_new_free_space(block_group, fs_info,
1535 key.objectid + key.offset);
1536 if (total_found > CACHING_CTL_WAKE_UP) {
1538 wake_up(&caching_ctl->wait);
1543 if (extent_count != expected_extent_count) {
1544 btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
1545 block_group->key.objectid, extent_count,
1546 expected_extent_count);
1552 caching_ctl->progress = (u64)-1;
1559 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1561 struct btrfs_block_group_cache *block_group;
1562 struct btrfs_fs_info *fs_info;
1563 struct btrfs_free_space_info *info;
1564 struct btrfs_path *path;
1565 u32 extent_count, flags;
1568 block_group = caching_ctl->block_group;
1569 fs_info = block_group->fs_info;
1571 path = btrfs_alloc_path();
1576 * Just like caching_thread() doesn't want to deadlock on the extent
1577 * tree, we don't want to deadlock on the free space tree.
1579 path->skip_locking = 1;
1580 path->search_commit_root = 1;
1583 info = search_free_space_info(NULL, fs_info, block_group, path, 0);
1585 ret = PTR_ERR(info);
1588 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1589 flags = btrfs_free_space_flags(path->nodes[0], info);
1592 * We left path pointing to the free space info item, so now
1593 * load_free_space_foo can just iterate through the free space tree from
1596 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1597 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1599 ret = load_free_space_extents(caching_ctl, path, extent_count);
1602 btrfs_free_path(path);