1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2015 Facebook. All rights reserved.
6 #include <linux/kernel.h>
7 #include <linux/sched/mm.h>
12 #include "free-space-tree.h"
13 #include "transaction.h"
14 #include "block-group.h"
16 #include "accessors.h"
17 #include "extent-tree.h"
18 #include "root-tree.h"
20 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
21 struct btrfs_block_group *block_group,
22 struct btrfs_path *path);
24 static struct btrfs_root *btrfs_free_space_root(
25 struct btrfs_block_group *block_group)
27 struct btrfs_key key = {
28 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
29 .type = BTRFS_ROOT_ITEM_KEY,
33 if (btrfs_fs_incompat(block_group->fs_info, EXTENT_TREE_V2))
34 key.offset = block_group->global_root_id;
35 return btrfs_global_root(block_group->fs_info, &key);
38 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
42 u64 num_bitmaps, total_bitmap_size;
44 if (WARN_ON(cache->length == 0))
45 btrfs_warn(cache->fs_info, "block group %llu length is zero",
49 * We convert to bitmaps when the disk space required for using extents
50 * exceeds that required for using bitmaps.
52 bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
53 num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
54 bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
55 total_bitmap_size = num_bitmaps * bitmap_size;
56 cache->bitmap_high_thresh = div_u64(total_bitmap_size,
57 sizeof(struct btrfs_item));
60 * We allow for a small buffer between the high threshold and low
61 * threshold to avoid thrashing back and forth between the two formats.
63 if (cache->bitmap_high_thresh > 100)
64 cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
66 cache->bitmap_low_thresh = 0;
69 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
70 struct btrfs_block_group *block_group,
71 struct btrfs_path *path)
73 struct btrfs_root *root = btrfs_free_space_root(block_group);
74 struct btrfs_free_space_info *info;
76 struct extent_buffer *leaf;
79 key.objectid = block_group->start;
80 key.type = BTRFS_FREE_SPACE_INFO_KEY;
81 key.offset = block_group->length;
83 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
87 leaf = path->nodes[0];
88 info = btrfs_item_ptr(leaf, path->slots[0],
89 struct btrfs_free_space_info);
90 btrfs_set_free_space_extent_count(leaf, info, 0);
91 btrfs_set_free_space_flags(leaf, info, 0);
92 btrfs_mark_buffer_dirty(leaf);
96 btrfs_release_path(path);
101 struct btrfs_free_space_info *search_free_space_info(
102 struct btrfs_trans_handle *trans,
103 struct btrfs_block_group *block_group,
104 struct btrfs_path *path, int cow)
106 struct btrfs_fs_info *fs_info = block_group->fs_info;
107 struct btrfs_root *root = btrfs_free_space_root(block_group);
108 struct btrfs_key key;
111 key.objectid = block_group->start;
112 key.type = BTRFS_FREE_SPACE_INFO_KEY;
113 key.offset = block_group->length;
115 ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
119 btrfs_warn(fs_info, "missing free space info for %llu",
122 return ERR_PTR(-ENOENT);
125 return btrfs_item_ptr(path->nodes[0], path->slots[0],
126 struct btrfs_free_space_info);
130 * btrfs_search_slot() but we're looking for the greatest key less than the
133 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root,
135 struct btrfs_key *key, struct btrfs_path *p,
136 int ins_len, int cow)
140 ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
149 if (p->slots[0] == 0) {
158 static inline u32 free_space_bitmap_size(const struct btrfs_fs_info *fs_info,
161 return DIV_ROUND_UP(size >> fs_info->sectorsize_bits, BITS_PER_BYTE);
164 static unsigned long *alloc_bitmap(u32 bitmap_size)
167 unsigned int nofs_flag;
168 u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
171 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
172 * into the filesystem as the free space bitmap can be modified in the
173 * critical section of a transaction commit.
175 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
176 * know that recursion is unsafe.
178 nofs_flag = memalloc_nofs_save();
179 ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
180 memalloc_nofs_restore(nofs_flag);
184 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
186 u8 *p = ((u8 *)map) + BIT_BYTE(start);
187 const unsigned int size = start + len;
188 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
189 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
191 while (len - bits_to_set >= 0) {
194 bits_to_set = BITS_PER_BYTE;
199 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
205 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
206 struct btrfs_block_group *block_group,
207 struct btrfs_path *path)
209 struct btrfs_fs_info *fs_info = trans->fs_info;
210 struct btrfs_root *root = btrfs_free_space_root(block_group);
211 struct btrfs_free_space_info *info;
212 struct btrfs_key key, found_key;
213 struct extent_buffer *leaf;
214 unsigned long *bitmap;
218 u32 bitmap_size, flags, expected_extent_count;
219 u32 extent_count = 0;
223 bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
224 bitmap = alloc_bitmap(bitmap_size);
230 start = block_group->start;
231 end = block_group->start + block_group->length;
233 key.objectid = end - 1;
235 key.offset = (u64)-1;
238 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
242 leaf = path->nodes[0];
245 while (path->slots[0] > 0) {
246 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
248 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
249 ASSERT(found_key.objectid == block_group->start);
250 ASSERT(found_key.offset == block_group->length);
253 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
256 ASSERT(found_key.objectid >= start);
257 ASSERT(found_key.objectid < end);
258 ASSERT(found_key.objectid + found_key.offset <= end);
260 first = div_u64(found_key.objectid - start,
261 fs_info->sectorsize);
262 last = div_u64(found_key.objectid + found_key.offset - start,
263 fs_info->sectorsize);
264 le_bitmap_set(bitmap, first, last - first);
274 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
277 btrfs_release_path(path);
280 info = search_free_space_info(trans, block_group, path, 1);
285 leaf = path->nodes[0];
286 flags = btrfs_free_space_flags(leaf, info);
287 flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
288 btrfs_set_free_space_flags(leaf, info, flags);
289 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
290 btrfs_mark_buffer_dirty(leaf);
291 btrfs_release_path(path);
293 if (extent_count != expected_extent_count) {
295 "incorrect extent count for %llu; counted %u, expected %u",
296 block_group->start, extent_count,
297 expected_extent_count);
303 bitmap_cursor = (char *)bitmap;
304 bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
311 extent_size = min(end - i, bitmap_range);
312 data_size = free_space_bitmap_size(fs_info, extent_size);
315 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
316 key.offset = extent_size;
318 ret = btrfs_insert_empty_item(trans, root, path, &key,
323 leaf = path->nodes[0];
324 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
325 write_extent_buffer(leaf, bitmap_cursor, ptr,
327 btrfs_mark_buffer_dirty(leaf);
328 btrfs_release_path(path);
331 bitmap_cursor += data_size;
338 btrfs_abort_transaction(trans, ret);
343 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
344 struct btrfs_block_group *block_group,
345 struct btrfs_path *path)
347 struct btrfs_fs_info *fs_info = trans->fs_info;
348 struct btrfs_root *root = btrfs_free_space_root(block_group);
349 struct btrfs_free_space_info *info;
350 struct btrfs_key key, found_key;
351 struct extent_buffer *leaf;
352 unsigned long *bitmap;
354 u32 bitmap_size, flags, expected_extent_count;
355 unsigned long nrbits, start_bit, end_bit;
356 u32 extent_count = 0;
360 bitmap_size = free_space_bitmap_size(fs_info, block_group->length);
361 bitmap = alloc_bitmap(bitmap_size);
367 start = block_group->start;
368 end = block_group->start + block_group->length;
370 key.objectid = end - 1;
372 key.offset = (u64)-1;
375 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
379 leaf = path->nodes[0];
382 while (path->slots[0] > 0) {
383 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
385 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
386 ASSERT(found_key.objectid == block_group->start);
387 ASSERT(found_key.offset == block_group->length);
390 } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
393 u32 bitmap_pos, data_size;
395 ASSERT(found_key.objectid >= start);
396 ASSERT(found_key.objectid < end);
397 ASSERT(found_key.objectid + found_key.offset <= end);
399 bitmap_pos = div_u64(found_key.objectid - start,
400 fs_info->sectorsize *
402 bitmap_cursor = ((char *)bitmap) + bitmap_pos;
403 data_size = free_space_bitmap_size(fs_info,
406 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
407 read_extent_buffer(leaf, bitmap_cursor, ptr,
417 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
420 btrfs_release_path(path);
423 info = search_free_space_info(trans, block_group, path, 1);
428 leaf = path->nodes[0];
429 flags = btrfs_free_space_flags(leaf, info);
430 flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
431 btrfs_set_free_space_flags(leaf, info, flags);
432 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
433 btrfs_mark_buffer_dirty(leaf);
434 btrfs_release_path(path);
436 nrbits = block_group->length >> block_group->fs_info->sectorsize_bits;
437 start_bit = find_next_bit_le(bitmap, nrbits, 0);
439 while (start_bit < nrbits) {
440 end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
441 ASSERT(start_bit < end_bit);
443 key.objectid = start + start_bit * block_group->fs_info->sectorsize;
444 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
445 key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
447 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
450 btrfs_release_path(path);
454 start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
457 if (extent_count != expected_extent_count) {
459 "incorrect extent count for %llu; counted %u, expected %u",
460 block_group->start, extent_count,
461 expected_extent_count);
471 btrfs_abort_transaction(trans, ret);
475 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
476 struct btrfs_block_group *block_group,
477 struct btrfs_path *path,
480 struct btrfs_free_space_info *info;
485 if (new_extents == 0)
488 info = search_free_space_info(trans, block_group, path, 1);
493 flags = btrfs_free_space_flags(path->nodes[0], info);
494 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
496 extent_count += new_extents;
497 btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
498 btrfs_mark_buffer_dirty(path->nodes[0]);
499 btrfs_release_path(path);
501 if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
502 extent_count > block_group->bitmap_high_thresh) {
503 ret = convert_free_space_to_bitmaps(trans, block_group, path);
504 } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
505 extent_count < block_group->bitmap_low_thresh) {
506 ret = convert_free_space_to_extents(trans, block_group, path);
514 int free_space_test_bit(struct btrfs_block_group *block_group,
515 struct btrfs_path *path, u64 offset)
517 struct extent_buffer *leaf;
518 struct btrfs_key key;
519 u64 found_start, found_end;
520 unsigned long ptr, i;
522 leaf = path->nodes[0];
523 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
524 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
526 found_start = key.objectid;
527 found_end = key.objectid + key.offset;
528 ASSERT(offset >= found_start && offset < found_end);
530 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
531 i = div_u64(offset - found_start,
532 block_group->fs_info->sectorsize);
533 return !!extent_buffer_test_bit(leaf, ptr, i);
536 static void free_space_set_bits(struct btrfs_block_group *block_group,
537 struct btrfs_path *path, u64 *start, u64 *size,
540 struct btrfs_fs_info *fs_info = block_group->fs_info;
541 struct extent_buffer *leaf;
542 struct btrfs_key key;
543 u64 end = *start + *size;
544 u64 found_start, found_end;
545 unsigned long ptr, first, last;
547 leaf = path->nodes[0];
548 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
549 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
551 found_start = key.objectid;
552 found_end = key.objectid + key.offset;
553 ASSERT(*start >= found_start && *start < found_end);
554 ASSERT(end > found_start);
559 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
560 first = (*start - found_start) >> fs_info->sectorsize_bits;
561 last = (end - found_start) >> fs_info->sectorsize_bits;
563 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
565 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
566 btrfs_mark_buffer_dirty(leaf);
568 *size -= end - *start;
573 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
574 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
575 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
578 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
579 struct btrfs_root *root, struct btrfs_path *p)
581 struct btrfs_key key;
583 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
588 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
589 btrfs_release_path(p);
591 key.objectid += key.offset;
593 key.offset = (u64)-1;
595 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
599 * If remove is 1, then we are removing free space, thus clearing bits in the
600 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
603 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
604 struct btrfs_block_group *block_group,
605 struct btrfs_path *path,
606 u64 start, u64 size, int remove)
608 struct btrfs_root *root = btrfs_free_space_root(block_group);
609 struct btrfs_key key;
610 u64 end = start + size;
611 u64 cur_start, cur_size;
612 int prev_bit, next_bit;
617 * Read the bit for the block immediately before the extent of space if
618 * that block is within the block group.
620 if (start > block_group->start) {
621 u64 prev_block = start - block_group->fs_info->sectorsize;
623 key.objectid = prev_block;
625 key.offset = (u64)-1;
627 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
631 prev_bit = free_space_test_bit(block_group, path, prev_block);
633 /* The previous block may have been in the previous bitmap. */
634 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
635 if (start >= key.objectid + key.offset) {
636 ret = free_space_next_bitmap(trans, root, path);
641 key.objectid = start;
643 key.offset = (u64)-1;
645 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
653 * Iterate over all of the bitmaps overlapped by the extent of space,
654 * clearing/setting bits as required.
659 free_space_set_bits(block_group, path, &cur_start, &cur_size,
663 ret = free_space_next_bitmap(trans, root, path);
669 * Read the bit for the block immediately after the extent of space if
670 * that block is within the block group.
672 if (end < block_group->start + block_group->length) {
673 /* The next block may be in the next bitmap. */
674 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
675 if (end >= key.objectid + key.offset) {
676 ret = free_space_next_bitmap(trans, root, path);
681 next_bit = free_space_test_bit(block_group, path, end);
689 /* Leftover on the left. */
693 /* Leftover on the right. */
699 /* Merging with neighbor on the left. */
703 /* Merging with neighbor on the right. */
708 btrfs_release_path(path);
709 ret = update_free_space_extent_count(trans, block_group, path,
716 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
717 struct btrfs_block_group *block_group,
718 struct btrfs_path *path,
721 struct btrfs_root *root = btrfs_free_space_root(block_group);
722 struct btrfs_key key;
723 u64 found_start, found_end;
724 u64 end = start + size;
725 int new_extents = -1;
728 key.objectid = start;
730 key.offset = (u64)-1;
732 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
736 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
738 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
740 found_start = key.objectid;
741 found_end = key.objectid + key.offset;
742 ASSERT(start >= found_start && end <= found_end);
745 * Okay, now that we've found the free space extent which contains the
746 * free space that we are removing, there are four cases:
748 * 1. We're using the whole extent: delete the key we found and
749 * decrement the free space extent count.
750 * 2. We are using part of the extent starting at the beginning: delete
751 * the key we found and insert a new key representing the leftover at
752 * the end. There is no net change in the number of extents.
753 * 3. We are using part of the extent ending at the end: delete the key
754 * we found and insert a new key representing the leftover at the
755 * beginning. There is no net change in the number of extents.
756 * 4. We are using part of the extent in the middle: delete the key we
757 * found and insert two new keys representing the leftovers on each
758 * side. Where we used to have one extent, we now have two, so increment
759 * the extent count. We may need to convert the block group to bitmaps
763 /* Delete the existing key (cases 1-4). */
764 ret = btrfs_del_item(trans, root, path);
768 /* Add a key for leftovers at the beginning (cases 3 and 4). */
769 if (start > found_start) {
770 key.objectid = found_start;
771 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
772 key.offset = start - found_start;
774 btrfs_release_path(path);
775 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
781 /* Add a key for leftovers at the end (cases 2 and 4). */
782 if (end < found_end) {
784 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
785 key.offset = found_end - end;
787 btrfs_release_path(path);
788 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
794 btrfs_release_path(path);
795 ret = update_free_space_extent_count(trans, block_group, path,
803 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
804 struct btrfs_block_group *block_group,
805 struct btrfs_path *path, u64 start, u64 size)
807 struct btrfs_free_space_info *info;
811 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
812 ret = __add_block_group_free_space(trans, block_group, path);
817 info = search_free_space_info(NULL, block_group, path, 0);
819 return PTR_ERR(info);
820 flags = btrfs_free_space_flags(path->nodes[0], info);
821 btrfs_release_path(path);
823 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
824 return modify_free_space_bitmap(trans, block_group, path,
827 return remove_free_space_extent(trans, block_group, path,
832 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_block_group *block_group;
836 struct btrfs_path *path;
839 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
842 path = btrfs_alloc_path();
848 block_group = btrfs_lookup_block_group(trans->fs_info, start);
855 mutex_lock(&block_group->free_space_lock);
856 ret = __remove_from_free_space_tree(trans, block_group, path, start,
858 mutex_unlock(&block_group->free_space_lock);
860 btrfs_put_block_group(block_group);
862 btrfs_free_path(path);
864 btrfs_abort_transaction(trans, ret);
868 static int add_free_space_extent(struct btrfs_trans_handle *trans,
869 struct btrfs_block_group *block_group,
870 struct btrfs_path *path,
873 struct btrfs_root *root = btrfs_free_space_root(block_group);
874 struct btrfs_key key, new_key;
875 u64 found_start, found_end;
876 u64 end = start + size;
881 * We are adding a new extent of free space, but we need to merge
882 * extents. There are four cases here:
884 * 1. The new extent does not have any immediate neighbors to merge
885 * with: add the new key and increment the free space extent count. We
886 * may need to convert the block group to bitmaps as a result.
887 * 2. The new extent has an immediate neighbor before it: remove the
888 * previous key and insert a new key combining both of them. There is no
889 * net change in the number of extents.
890 * 3. The new extent has an immediate neighbor after it: remove the next
891 * key and insert a new key combining both of them. There is no net
892 * change in the number of extents.
893 * 4. The new extent has immediate neighbors on both sides: remove both
894 * of the keys and insert a new key combining all of them. Where we used
895 * to have two extents, we now have one, so decrement the extent count.
898 new_key.objectid = start;
899 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
900 new_key.offset = size;
902 /* Search for a neighbor on the left. */
903 if (start == block_group->start)
905 key.objectid = start - 1;
907 key.offset = (u64)-1;
909 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
913 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
915 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
916 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
917 btrfs_release_path(path);
921 found_start = key.objectid;
922 found_end = key.objectid + key.offset;
923 ASSERT(found_start >= block_group->start &&
924 found_end > block_group->start);
925 ASSERT(found_start < start && found_end <= start);
928 * Delete the neighbor on the left and absorb it into the new key (cases
931 if (found_end == start) {
932 ret = btrfs_del_item(trans, root, path);
935 new_key.objectid = found_start;
936 new_key.offset += key.offset;
939 btrfs_release_path(path);
942 /* Search for a neighbor on the right. */
943 if (end == block_group->start + block_group->length)
947 key.offset = (u64)-1;
949 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
953 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
955 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
956 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
957 btrfs_release_path(path);
961 found_start = key.objectid;
962 found_end = key.objectid + key.offset;
963 ASSERT(found_start >= block_group->start &&
964 found_end > block_group->start);
965 ASSERT((found_start < start && found_end <= start) ||
966 (found_start >= end && found_end > end));
969 * Delete the neighbor on the right and absorb it into the new key
972 if (found_start == end) {
973 ret = btrfs_del_item(trans, root, path);
976 new_key.offset += key.offset;
979 btrfs_release_path(path);
982 /* Insert the new key (cases 1-4). */
983 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
987 btrfs_release_path(path);
988 ret = update_free_space_extent_count(trans, block_group, path,
996 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
997 struct btrfs_block_group *block_group,
998 struct btrfs_path *path, u64 start, u64 size)
1000 struct btrfs_free_space_info *info;
1004 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1005 ret = __add_block_group_free_space(trans, block_group, path);
1010 info = search_free_space_info(NULL, block_group, path, 0);
1012 return PTR_ERR(info);
1013 flags = btrfs_free_space_flags(path->nodes[0], info);
1014 btrfs_release_path(path);
1016 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1017 return modify_free_space_bitmap(trans, block_group, path,
1020 return add_free_space_extent(trans, block_group, path, start,
1025 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1026 u64 start, u64 size)
1028 struct btrfs_block_group *block_group;
1029 struct btrfs_path *path;
1032 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1035 path = btrfs_alloc_path();
1041 block_group = btrfs_lookup_block_group(trans->fs_info, start);
1048 mutex_lock(&block_group->free_space_lock);
1049 ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1050 mutex_unlock(&block_group->free_space_lock);
1052 btrfs_put_block_group(block_group);
1054 btrfs_free_path(path);
1056 btrfs_abort_transaction(trans, ret);
1061 * Populate the free space tree by walking the extent tree. Operations on the
1062 * extent tree that happen as a result of writes to the free space tree will go
1063 * through the normal add/remove hooks.
1065 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1066 struct btrfs_block_group *block_group)
1068 struct btrfs_root *extent_root;
1069 struct btrfs_path *path, *path2;
1070 struct btrfs_key key;
1074 path = btrfs_alloc_path();
1077 path->reada = READA_FORWARD;
1079 path2 = btrfs_alloc_path();
1081 btrfs_free_path(path);
1085 ret = add_new_free_space_info(trans, block_group, path2);
1089 mutex_lock(&block_group->free_space_lock);
1092 * Iterate through all of the extent and metadata items in this block
1093 * group, adding the free space between them and the free space at the
1094 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1095 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1098 key.objectid = block_group->start;
1099 key.type = BTRFS_EXTENT_ITEM_KEY;
1102 extent_root = btrfs_extent_root(trans->fs_info, key.objectid);
1103 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1108 start = block_group->start;
1109 end = block_group->start + block_group->length;
1111 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1113 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1114 key.type == BTRFS_METADATA_ITEM_KEY) {
1115 if (key.objectid >= end)
1118 if (start < key.objectid) {
1119 ret = __add_to_free_space_tree(trans,
1127 start = key.objectid;
1128 if (key.type == BTRFS_METADATA_ITEM_KEY)
1129 start += trans->fs_info->nodesize;
1131 start += key.offset;
1132 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1133 if (key.objectid != block_group->start)
1137 ret = btrfs_next_item(extent_root, path);
1144 ret = __add_to_free_space_tree(trans, block_group, path2,
1145 start, end - start);
1152 mutex_unlock(&block_group->free_space_lock);
1154 btrfs_free_path(path2);
1155 btrfs_free_path(path);
1159 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1161 struct btrfs_trans_handle *trans;
1162 struct btrfs_root *tree_root = fs_info->tree_root;
1163 struct btrfs_root *free_space_root;
1164 struct btrfs_block_group *block_group;
1165 struct rb_node *node;
1168 trans = btrfs_start_transaction(tree_root, 0);
1170 return PTR_ERR(trans);
1172 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1173 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1174 free_space_root = btrfs_create_tree(trans,
1175 BTRFS_FREE_SPACE_TREE_OBJECTID);
1176 if (IS_ERR(free_space_root)) {
1177 ret = PTR_ERR(free_space_root);
1180 ret = btrfs_global_root_insert(free_space_root);
1182 btrfs_put_root(free_space_root);
1186 node = rb_first_cached(&fs_info->block_group_cache_tree);
1188 block_group = rb_entry(node, struct btrfs_block_group,
1190 ret = populate_free_space_tree(trans, block_group);
1193 node = rb_next(node);
1196 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1197 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1198 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1199 ret = btrfs_commit_transaction(trans);
1202 * Now that we've committed the transaction any reading of our commit
1203 * root will be safe, so we can cache from the free space tree now.
1205 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1209 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1210 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1211 btrfs_abort_transaction(trans, ret);
1212 btrfs_end_transaction(trans);
1216 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1217 struct btrfs_root *root)
1219 struct btrfs_path *path;
1220 struct btrfs_key key;
1224 path = btrfs_alloc_path();
1233 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1237 nr = btrfs_header_nritems(path->nodes[0]);
1242 ret = btrfs_del_items(trans, root, path, 0, nr);
1246 btrfs_release_path(path);
1251 btrfs_free_path(path);
1255 int btrfs_delete_free_space_tree(struct btrfs_fs_info *fs_info)
1257 struct btrfs_trans_handle *trans;
1258 struct btrfs_root *tree_root = fs_info->tree_root;
1259 struct btrfs_key key = {
1260 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1261 .type = BTRFS_ROOT_ITEM_KEY,
1264 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1267 trans = btrfs_start_transaction(tree_root, 0);
1269 return PTR_ERR(trans);
1271 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1272 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1274 ret = clear_free_space_tree(trans, free_space_root);
1278 ret = btrfs_del_root(trans, &free_space_root->root_key);
1282 btrfs_global_root_delete(free_space_root);
1284 spin_lock(&fs_info->trans_lock);
1285 list_del(&free_space_root->dirty_list);
1286 spin_unlock(&fs_info->trans_lock);
1288 btrfs_tree_lock(free_space_root->node);
1289 btrfs_clear_buffer_dirty(trans, free_space_root->node);
1290 btrfs_tree_unlock(free_space_root->node);
1291 btrfs_free_tree_block(trans, btrfs_root_id(free_space_root),
1292 free_space_root->node, 0, 1);
1294 btrfs_put_root(free_space_root);
1296 return btrfs_commit_transaction(trans);
1299 btrfs_abort_transaction(trans, ret);
1300 btrfs_end_transaction(trans);
1304 int btrfs_rebuild_free_space_tree(struct btrfs_fs_info *fs_info)
1306 struct btrfs_trans_handle *trans;
1307 struct btrfs_key key = {
1308 .objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
1309 .type = BTRFS_ROOT_ITEM_KEY,
1312 struct btrfs_root *free_space_root = btrfs_global_root(fs_info, &key);
1313 struct rb_node *node;
1316 trans = btrfs_start_transaction(free_space_root, 1);
1318 return PTR_ERR(trans);
1320 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1321 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1323 ret = clear_free_space_tree(trans, free_space_root);
1327 node = rb_first_cached(&fs_info->block_group_cache_tree);
1329 struct btrfs_block_group *block_group;
1331 block_group = rb_entry(node, struct btrfs_block_group,
1333 ret = populate_free_space_tree(trans, block_group);
1336 node = rb_next(node);
1339 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1340 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1341 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1343 ret = btrfs_commit_transaction(trans);
1344 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1347 btrfs_abort_transaction(trans, ret);
1348 btrfs_end_transaction(trans);
1352 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1353 struct btrfs_block_group *block_group,
1354 struct btrfs_path *path)
1358 clear_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags);
1360 ret = add_new_free_space_info(trans, block_group, path);
1364 return __add_to_free_space_tree(trans, block_group, path,
1366 block_group->length);
1369 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1370 struct btrfs_block_group *block_group)
1372 struct btrfs_fs_info *fs_info = trans->fs_info;
1373 struct btrfs_path *path = NULL;
1376 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1379 mutex_lock(&block_group->free_space_lock);
1380 if (!test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags))
1383 path = btrfs_alloc_path();
1389 ret = __add_block_group_free_space(trans, block_group, path);
1392 btrfs_free_path(path);
1393 mutex_unlock(&block_group->free_space_lock);
1395 btrfs_abort_transaction(trans, ret);
1399 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1400 struct btrfs_block_group *block_group)
1402 struct btrfs_root *root = btrfs_free_space_root(block_group);
1403 struct btrfs_path *path;
1404 struct btrfs_key key, found_key;
1405 struct extent_buffer *leaf;
1410 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1413 if (test_bit(BLOCK_GROUP_FLAG_NEEDS_FREE_SPACE, &block_group->runtime_flags)) {
1414 /* We never added this block group to the free space tree. */
1418 path = btrfs_alloc_path();
1424 start = block_group->start;
1425 end = block_group->start + block_group->length;
1427 key.objectid = end - 1;
1429 key.offset = (u64)-1;
1432 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1436 leaf = path->nodes[0];
1439 while (path->slots[0] > 0) {
1440 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1442 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1443 ASSERT(found_key.objectid == block_group->start);
1444 ASSERT(found_key.offset == block_group->length);
1449 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1450 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1451 ASSERT(found_key.objectid >= start);
1452 ASSERT(found_key.objectid < end);
1453 ASSERT(found_key.objectid + found_key.offset <= end);
1461 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1464 btrfs_release_path(path);
1469 btrfs_free_path(path);
1471 btrfs_abort_transaction(trans, ret);
1475 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1476 struct btrfs_path *path,
1477 u32 expected_extent_count)
1479 struct btrfs_block_group *block_group;
1480 struct btrfs_fs_info *fs_info;
1481 struct btrfs_root *root;
1482 struct btrfs_key key;
1483 int prev_bit = 0, bit;
1484 /* Initialize to silence GCC. */
1485 u64 extent_start = 0;
1487 u64 total_found = 0;
1488 u32 extent_count = 0;
1491 block_group = caching_ctl->block_group;
1492 fs_info = block_group->fs_info;
1493 root = btrfs_free_space_root(block_group);
1495 end = block_group->start + block_group->length;
1498 ret = btrfs_next_item(root, path);
1504 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1506 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1509 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1510 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1512 offset = key.objectid;
1513 while (offset < key.objectid + key.offset) {
1514 bit = free_space_test_bit(block_group, path, offset);
1515 if (prev_bit == 0 && bit == 1) {
1516 extent_start = offset;
1517 } else if (prev_bit == 1 && bit == 0) {
1520 ret = btrfs_add_new_free_space(block_group,
1526 total_found += space_added;
1527 if (total_found > CACHING_CTL_WAKE_UP) {
1529 wake_up(&caching_ctl->wait);
1534 offset += fs_info->sectorsize;
1537 if (prev_bit == 1) {
1538 ret = btrfs_add_new_free_space(block_group, extent_start, end, NULL);
1544 if (extent_count != expected_extent_count) {
1546 "incorrect extent count for %llu; counted %u, expected %u",
1547 block_group->start, extent_count,
1548 expected_extent_count);
1559 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1560 struct btrfs_path *path,
1561 u32 expected_extent_count)
1563 struct btrfs_block_group *block_group;
1564 struct btrfs_fs_info *fs_info;
1565 struct btrfs_root *root;
1566 struct btrfs_key key;
1568 u64 total_found = 0;
1569 u32 extent_count = 0;
1572 block_group = caching_ctl->block_group;
1573 fs_info = block_group->fs_info;
1574 root = btrfs_free_space_root(block_group);
1576 end = block_group->start + block_group->length;
1581 ret = btrfs_next_item(root, path);
1587 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1589 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1592 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1593 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1595 ret = btrfs_add_new_free_space(block_group, key.objectid,
1596 key.objectid + key.offset,
1600 total_found += space_added;
1601 if (total_found > CACHING_CTL_WAKE_UP) {
1603 wake_up(&caching_ctl->wait);
1608 if (extent_count != expected_extent_count) {
1610 "incorrect extent count for %llu; counted %u, expected %u",
1611 block_group->start, extent_count,
1612 expected_extent_count);
1623 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1625 struct btrfs_block_group *block_group;
1626 struct btrfs_free_space_info *info;
1627 struct btrfs_path *path;
1628 u32 extent_count, flags;
1631 block_group = caching_ctl->block_group;
1633 path = btrfs_alloc_path();
1638 * Just like caching_thread() doesn't want to deadlock on the extent
1639 * tree, we don't want to deadlock on the free space tree.
1641 path->skip_locking = 1;
1642 path->search_commit_root = 1;
1643 path->reada = READA_FORWARD;
1645 info = search_free_space_info(NULL, block_group, path, 0);
1647 ret = PTR_ERR(info);
1650 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1651 flags = btrfs_free_space_flags(path->nodes[0], info);
1654 * We left path pointing to the free space info item, so now
1655 * load_free_space_foo can just iterate through the free space tree from
1658 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1659 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1661 ret = load_free_space_extents(caching_ctl, path, extent_count);
1664 btrfs_free_path(path);