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>
11 #include "free-space-tree.h"
12 #include "transaction.h"
13 #include "block-group.h"
15 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
16 struct btrfs_block_group *block_group,
17 struct btrfs_path *path);
19 void set_free_space_tree_thresholds(struct btrfs_block_group *cache)
23 u64 num_bitmaps, total_bitmap_size;
25 if (WARN_ON(cache->length == 0))
26 btrfs_warn(cache->fs_info, "block group %llu length is zero",
30 * We convert to bitmaps when the disk space required for using extents
31 * exceeds that required for using bitmaps.
33 bitmap_range = cache->fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
34 num_bitmaps = div_u64(cache->length + bitmap_range - 1, bitmap_range);
35 bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
36 total_bitmap_size = num_bitmaps * bitmap_size;
37 cache->bitmap_high_thresh = div_u64(total_bitmap_size,
38 sizeof(struct btrfs_item));
41 * We allow for a small buffer between the high threshold and low
42 * threshold to avoid thrashing back and forth between the two formats.
44 if (cache->bitmap_high_thresh > 100)
45 cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
47 cache->bitmap_low_thresh = 0;
50 static int add_new_free_space_info(struct btrfs_trans_handle *trans,
51 struct btrfs_block_group *block_group,
52 struct btrfs_path *path)
54 struct btrfs_root *root = trans->fs_info->free_space_root;
55 struct btrfs_free_space_info *info;
57 struct extent_buffer *leaf;
60 key.objectid = block_group->start;
61 key.type = BTRFS_FREE_SPACE_INFO_KEY;
62 key.offset = block_group->length;
64 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
68 leaf = path->nodes[0];
69 info = btrfs_item_ptr(leaf, path->slots[0],
70 struct btrfs_free_space_info);
71 btrfs_set_free_space_extent_count(leaf, info, 0);
72 btrfs_set_free_space_flags(leaf, info, 0);
73 btrfs_mark_buffer_dirty(leaf);
77 btrfs_release_path(path);
82 struct btrfs_free_space_info *search_free_space_info(
83 struct btrfs_trans_handle *trans,
84 struct btrfs_block_group *block_group,
85 struct btrfs_path *path, int cow)
87 struct btrfs_fs_info *fs_info = block_group->fs_info;
88 struct btrfs_root *root = fs_info->free_space_root;
92 key.objectid = block_group->start;
93 key.type = BTRFS_FREE_SPACE_INFO_KEY;
94 key.offset = block_group->length;
96 ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
100 btrfs_warn(fs_info, "missing free space info for %llu",
103 return ERR_PTR(-ENOENT);
106 return btrfs_item_ptr(path->nodes[0], path->slots[0],
107 struct btrfs_free_space_info);
111 * btrfs_search_slot() but we're looking for the greatest key less than the
114 static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
115 struct btrfs_root *root,
116 struct btrfs_key *key, struct btrfs_path *p,
117 int ins_len, int cow)
121 ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
130 if (p->slots[0] == 0) {
139 static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
141 return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
144 static unsigned long *alloc_bitmap(u32 bitmap_size)
147 unsigned int nofs_flag;
148 u32 bitmap_rounded_size = round_up(bitmap_size, sizeof(unsigned long));
151 * GFP_NOFS doesn't work with kvmalloc(), but we really can't recurse
152 * into the filesystem as the free space bitmap can be modified in the
153 * critical section of a transaction commit.
155 * TODO: push the memalloc_nofs_{save,restore}() to the caller where we
156 * know that recursion is unsafe.
158 nofs_flag = memalloc_nofs_save();
159 ret = kvzalloc(bitmap_rounded_size, GFP_KERNEL);
160 memalloc_nofs_restore(nofs_flag);
164 static void le_bitmap_set(unsigned long *map, unsigned int start, int len)
166 u8 *p = ((u8 *)map) + BIT_BYTE(start);
167 const unsigned int size = start + len;
168 int bits_to_set = BITS_PER_BYTE - (start % BITS_PER_BYTE);
169 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(start);
171 while (len - bits_to_set >= 0) {
174 bits_to_set = BITS_PER_BYTE;
179 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
185 int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
186 struct btrfs_block_group *block_group,
187 struct btrfs_path *path)
189 struct btrfs_fs_info *fs_info = trans->fs_info;
190 struct btrfs_root *root = fs_info->free_space_root;
191 struct btrfs_free_space_info *info;
192 struct btrfs_key key, found_key;
193 struct extent_buffer *leaf;
194 unsigned long *bitmap;
198 u32 bitmap_size, flags, expected_extent_count;
199 u32 extent_count = 0;
203 bitmap_size = free_space_bitmap_size(block_group->length,
204 fs_info->sectorsize);
205 bitmap = alloc_bitmap(bitmap_size);
211 start = block_group->start;
212 end = block_group->start + block_group->length;
214 key.objectid = end - 1;
216 key.offset = (u64)-1;
219 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
223 leaf = path->nodes[0];
226 while (path->slots[0] > 0) {
227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
229 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
230 ASSERT(found_key.objectid == block_group->start);
231 ASSERT(found_key.offset == block_group->length);
234 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
237 ASSERT(found_key.objectid >= start);
238 ASSERT(found_key.objectid < end);
239 ASSERT(found_key.objectid + found_key.offset <= end);
241 first = div_u64(found_key.objectid - start,
242 fs_info->sectorsize);
243 last = div_u64(found_key.objectid + found_key.offset - start,
244 fs_info->sectorsize);
245 le_bitmap_set(bitmap, first, last - first);
255 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
258 btrfs_release_path(path);
261 info = search_free_space_info(trans, block_group, path, 1);
266 leaf = path->nodes[0];
267 flags = btrfs_free_space_flags(leaf, info);
268 flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
269 btrfs_set_free_space_flags(leaf, info, flags);
270 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
271 btrfs_mark_buffer_dirty(leaf);
272 btrfs_release_path(path);
274 if (extent_count != expected_extent_count) {
276 "incorrect extent count for %llu; counted %u, expected %u",
277 block_group->start, extent_count,
278 expected_extent_count);
284 bitmap_cursor = (char *)bitmap;
285 bitmap_range = fs_info->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
292 extent_size = min(end - i, bitmap_range);
293 data_size = free_space_bitmap_size(extent_size,
294 fs_info->sectorsize);
297 key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
298 key.offset = extent_size;
300 ret = btrfs_insert_empty_item(trans, root, path, &key,
305 leaf = path->nodes[0];
306 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
307 write_extent_buffer(leaf, bitmap_cursor, ptr,
309 btrfs_mark_buffer_dirty(leaf);
310 btrfs_release_path(path);
313 bitmap_cursor += data_size;
320 btrfs_abort_transaction(trans, ret);
325 int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
326 struct btrfs_block_group *block_group,
327 struct btrfs_path *path)
329 struct btrfs_fs_info *fs_info = trans->fs_info;
330 struct btrfs_root *root = fs_info->free_space_root;
331 struct btrfs_free_space_info *info;
332 struct btrfs_key key, found_key;
333 struct extent_buffer *leaf;
334 unsigned long *bitmap;
336 u32 bitmap_size, flags, expected_extent_count;
337 unsigned long nrbits, start_bit, end_bit;
338 u32 extent_count = 0;
342 bitmap_size = free_space_bitmap_size(block_group->length,
343 fs_info->sectorsize);
344 bitmap = alloc_bitmap(bitmap_size);
350 start = block_group->start;
351 end = block_group->start + block_group->length;
353 key.objectid = end - 1;
355 key.offset = (u64)-1;
358 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
362 leaf = path->nodes[0];
365 while (path->slots[0] > 0) {
366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
368 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
369 ASSERT(found_key.objectid == block_group->start);
370 ASSERT(found_key.offset == block_group->length);
373 } else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
376 u32 bitmap_pos, data_size;
378 ASSERT(found_key.objectid >= start);
379 ASSERT(found_key.objectid < end);
380 ASSERT(found_key.objectid + found_key.offset <= end);
382 bitmap_pos = div_u64(found_key.objectid - start,
383 fs_info->sectorsize *
385 bitmap_cursor = ((char *)bitmap) + bitmap_pos;
386 data_size = free_space_bitmap_size(found_key.offset,
387 fs_info->sectorsize);
389 ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
390 read_extent_buffer(leaf, bitmap_cursor, ptr,
400 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
403 btrfs_release_path(path);
406 info = search_free_space_info(trans, block_group, path, 1);
411 leaf = path->nodes[0];
412 flags = btrfs_free_space_flags(leaf, info);
413 flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
414 btrfs_set_free_space_flags(leaf, info, flags);
415 expected_extent_count = btrfs_free_space_extent_count(leaf, info);
416 btrfs_mark_buffer_dirty(leaf);
417 btrfs_release_path(path);
419 nrbits = div_u64(block_group->length, block_group->fs_info->sectorsize);
420 start_bit = find_next_bit_le(bitmap, nrbits, 0);
422 while (start_bit < nrbits) {
423 end_bit = find_next_zero_bit_le(bitmap, nrbits, start_bit);
424 ASSERT(start_bit < end_bit);
426 key.objectid = start + start_bit * block_group->fs_info->sectorsize;
427 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
428 key.offset = (end_bit - start_bit) * block_group->fs_info->sectorsize;
430 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
433 btrfs_release_path(path);
437 start_bit = find_next_bit_le(bitmap, nrbits, end_bit);
440 if (extent_count != expected_extent_count) {
442 "incorrect extent count for %llu; counted %u, expected %u",
443 block_group->start, extent_count,
444 expected_extent_count);
454 btrfs_abort_transaction(trans, ret);
458 static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
459 struct btrfs_block_group *block_group,
460 struct btrfs_path *path,
463 struct btrfs_free_space_info *info;
468 if (new_extents == 0)
471 info = search_free_space_info(trans, block_group, path, 1);
476 flags = btrfs_free_space_flags(path->nodes[0], info);
477 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
479 extent_count += new_extents;
480 btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
481 btrfs_mark_buffer_dirty(path->nodes[0]);
482 btrfs_release_path(path);
484 if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
485 extent_count > block_group->bitmap_high_thresh) {
486 ret = convert_free_space_to_bitmaps(trans, block_group, path);
487 } else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
488 extent_count < block_group->bitmap_low_thresh) {
489 ret = convert_free_space_to_extents(trans, block_group, path);
497 int free_space_test_bit(struct btrfs_block_group *block_group,
498 struct btrfs_path *path, u64 offset)
500 struct extent_buffer *leaf;
501 struct btrfs_key key;
502 u64 found_start, found_end;
503 unsigned long ptr, i;
505 leaf = path->nodes[0];
506 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
507 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
509 found_start = key.objectid;
510 found_end = key.objectid + key.offset;
511 ASSERT(offset >= found_start && offset < found_end);
513 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
514 i = div_u64(offset - found_start,
515 block_group->fs_info->sectorsize);
516 return !!extent_buffer_test_bit(leaf, ptr, i);
519 static void free_space_set_bits(struct btrfs_block_group *block_group,
520 struct btrfs_path *path, u64 *start, u64 *size,
523 struct btrfs_fs_info *fs_info = block_group->fs_info;
524 struct extent_buffer *leaf;
525 struct btrfs_key key;
526 u64 end = *start + *size;
527 u64 found_start, found_end;
528 unsigned long ptr, first, last;
530 leaf = path->nodes[0];
531 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
532 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
534 found_start = key.objectid;
535 found_end = key.objectid + key.offset;
536 ASSERT(*start >= found_start && *start < found_end);
537 ASSERT(end > found_start);
542 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
543 first = div_u64(*start - found_start, fs_info->sectorsize);
544 last = div_u64(end - found_start, fs_info->sectorsize);
546 extent_buffer_bitmap_set(leaf, ptr, first, last - first);
548 extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
549 btrfs_mark_buffer_dirty(leaf);
551 *size -= end - *start;
556 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
557 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
558 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
561 static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
562 struct btrfs_root *root, struct btrfs_path *p)
564 struct btrfs_key key;
566 if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
571 btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
572 btrfs_release_path(p);
574 key.objectid += key.offset;
576 key.offset = (u64)-1;
578 return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
582 * If remove is 1, then we are removing free space, thus clearing bits in the
583 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
586 static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
587 struct btrfs_block_group *block_group,
588 struct btrfs_path *path,
589 u64 start, u64 size, int remove)
591 struct btrfs_root *root = block_group->fs_info->free_space_root;
592 struct btrfs_key key;
593 u64 end = start + size;
594 u64 cur_start, cur_size;
595 int prev_bit, next_bit;
600 * Read the bit for the block immediately before the extent of space if
601 * that block is within the block group.
603 if (start > block_group->start) {
604 u64 prev_block = start - block_group->fs_info->sectorsize;
606 key.objectid = prev_block;
608 key.offset = (u64)-1;
610 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
614 prev_bit = free_space_test_bit(block_group, path, prev_block);
616 /* The previous block may have been in the previous bitmap. */
617 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
618 if (start >= key.objectid + key.offset) {
619 ret = free_space_next_bitmap(trans, root, path);
624 key.objectid = start;
626 key.offset = (u64)-1;
628 ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
636 * Iterate over all of the bitmaps overlapped by the extent of space,
637 * clearing/setting bits as required.
642 free_space_set_bits(block_group, path, &cur_start, &cur_size,
646 ret = free_space_next_bitmap(trans, root, path);
652 * Read the bit for the block immediately after the extent of space if
653 * that block is within the block group.
655 if (end < block_group->start + block_group->length) {
656 /* The next block may be in the next bitmap. */
657 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
658 if (end >= key.objectid + key.offset) {
659 ret = free_space_next_bitmap(trans, root, path);
664 next_bit = free_space_test_bit(block_group, path, end);
672 /* Leftover on the left. */
676 /* Leftover on the right. */
682 /* Merging with neighbor on the left. */
686 /* Merging with neighbor on the right. */
691 btrfs_release_path(path);
692 ret = update_free_space_extent_count(trans, block_group, path,
699 static int remove_free_space_extent(struct btrfs_trans_handle *trans,
700 struct btrfs_block_group *block_group,
701 struct btrfs_path *path,
704 struct btrfs_root *root = trans->fs_info->free_space_root;
705 struct btrfs_key key;
706 u64 found_start, found_end;
707 u64 end = start + size;
708 int new_extents = -1;
711 key.objectid = start;
713 key.offset = (u64)-1;
715 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
719 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
721 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
723 found_start = key.objectid;
724 found_end = key.objectid + key.offset;
725 ASSERT(start >= found_start && end <= found_end);
728 * Okay, now that we've found the free space extent which contains the
729 * free space that we are removing, there are four cases:
731 * 1. We're using the whole extent: delete the key we found and
732 * decrement the free space extent count.
733 * 2. We are using part of the extent starting at the beginning: delete
734 * the key we found and insert a new key representing the leftover at
735 * the end. There is no net change in the number of extents.
736 * 3. We are using part of the extent ending at the end: delete the key
737 * we found and insert a new key representing the leftover at the
738 * beginning. There is no net change in the number of extents.
739 * 4. We are using part of the extent in the middle: delete the key we
740 * found and insert two new keys representing the leftovers on each
741 * side. Where we used to have one extent, we now have two, so increment
742 * the extent count. We may need to convert the block group to bitmaps
746 /* Delete the existing key (cases 1-4). */
747 ret = btrfs_del_item(trans, root, path);
751 /* Add a key for leftovers at the beginning (cases 3 and 4). */
752 if (start > found_start) {
753 key.objectid = found_start;
754 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
755 key.offset = start - found_start;
757 btrfs_release_path(path);
758 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
764 /* Add a key for leftovers at the end (cases 2 and 4). */
765 if (end < found_end) {
767 key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
768 key.offset = found_end - end;
770 btrfs_release_path(path);
771 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
777 btrfs_release_path(path);
778 ret = update_free_space_extent_count(trans, block_group, path,
786 int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
787 struct btrfs_block_group *block_group,
788 struct btrfs_path *path, u64 start, u64 size)
790 struct btrfs_free_space_info *info;
794 if (block_group->needs_free_space) {
795 ret = __add_block_group_free_space(trans, block_group, path);
800 info = search_free_space_info(NULL, block_group, path, 0);
802 return PTR_ERR(info);
803 flags = btrfs_free_space_flags(path->nodes[0], info);
804 btrfs_release_path(path);
806 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
807 return modify_free_space_bitmap(trans, block_group, path,
810 return remove_free_space_extent(trans, block_group, path,
815 int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
818 struct btrfs_block_group *block_group;
819 struct btrfs_path *path;
822 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
825 path = btrfs_alloc_path();
831 block_group = btrfs_lookup_block_group(trans->fs_info, start);
838 mutex_lock(&block_group->free_space_lock);
839 ret = __remove_from_free_space_tree(trans, block_group, path, start,
841 mutex_unlock(&block_group->free_space_lock);
843 btrfs_put_block_group(block_group);
845 btrfs_free_path(path);
847 btrfs_abort_transaction(trans, ret);
851 static int add_free_space_extent(struct btrfs_trans_handle *trans,
852 struct btrfs_block_group *block_group,
853 struct btrfs_path *path,
856 struct btrfs_root *root = trans->fs_info->free_space_root;
857 struct btrfs_key key, new_key;
858 u64 found_start, found_end;
859 u64 end = start + size;
864 * We are adding a new extent of free space, but we need to merge
865 * extents. There are four cases here:
867 * 1. The new extent does not have any immediate neighbors to merge
868 * with: add the new key and increment the free space extent count. We
869 * may need to convert the block group to bitmaps as a result.
870 * 2. The new extent has an immediate neighbor before it: remove the
871 * previous key and insert a new key combining both of them. There is no
872 * net change in the number of extents.
873 * 3. The new extent has an immediate neighbor after it: remove the next
874 * key and insert a new key combining both of them. There is no net
875 * change in the number of extents.
876 * 4. The new extent has immediate neighbors on both sides: remove both
877 * of the keys and insert a new key combining all of them. Where we used
878 * to have two extents, we now have one, so decrement the extent count.
881 new_key.objectid = start;
882 new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
883 new_key.offset = size;
885 /* Search for a neighbor on the left. */
886 if (start == block_group->start)
888 key.objectid = start - 1;
890 key.offset = (u64)-1;
892 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
896 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
898 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
899 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
900 btrfs_release_path(path);
904 found_start = key.objectid;
905 found_end = key.objectid + key.offset;
906 ASSERT(found_start >= block_group->start &&
907 found_end > block_group->start);
908 ASSERT(found_start < start && found_end <= start);
911 * Delete the neighbor on the left and absorb it into the new key (cases
914 if (found_end == start) {
915 ret = btrfs_del_item(trans, root, path);
918 new_key.objectid = found_start;
919 new_key.offset += key.offset;
922 btrfs_release_path(path);
925 /* Search for a neighbor on the right. */
926 if (end == block_group->start + block_group->length)
930 key.offset = (u64)-1;
932 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
936 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
938 if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
939 ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
940 btrfs_release_path(path);
944 found_start = key.objectid;
945 found_end = key.objectid + key.offset;
946 ASSERT(found_start >= block_group->start &&
947 found_end > block_group->start);
948 ASSERT((found_start < start && found_end <= start) ||
949 (found_start >= end && found_end > end));
952 * Delete the neighbor on the right and absorb it into the new key
955 if (found_start == end) {
956 ret = btrfs_del_item(trans, root, path);
959 new_key.offset += key.offset;
962 btrfs_release_path(path);
965 /* Insert the new key (cases 1-4). */
966 ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
970 btrfs_release_path(path);
971 ret = update_free_space_extent_count(trans, block_group, path,
979 int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
980 struct btrfs_block_group *block_group,
981 struct btrfs_path *path, u64 start, u64 size)
983 struct btrfs_free_space_info *info;
987 if (block_group->needs_free_space) {
988 ret = __add_block_group_free_space(trans, block_group, path);
993 info = search_free_space_info(NULL, block_group, path, 0);
995 return PTR_ERR(info);
996 flags = btrfs_free_space_flags(path->nodes[0], info);
997 btrfs_release_path(path);
999 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
1000 return modify_free_space_bitmap(trans, block_group, path,
1003 return add_free_space_extent(trans, block_group, path, start,
1008 int add_to_free_space_tree(struct btrfs_trans_handle *trans,
1009 u64 start, u64 size)
1011 struct btrfs_block_group *block_group;
1012 struct btrfs_path *path;
1015 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1018 path = btrfs_alloc_path();
1024 block_group = btrfs_lookup_block_group(trans->fs_info, start);
1031 mutex_lock(&block_group->free_space_lock);
1032 ret = __add_to_free_space_tree(trans, block_group, path, start, size);
1033 mutex_unlock(&block_group->free_space_lock);
1035 btrfs_put_block_group(block_group);
1037 btrfs_free_path(path);
1039 btrfs_abort_transaction(trans, ret);
1044 * Populate the free space tree by walking the extent tree. Operations on the
1045 * extent tree that happen as a result of writes to the free space tree will go
1046 * through the normal add/remove hooks.
1048 static int populate_free_space_tree(struct btrfs_trans_handle *trans,
1049 struct btrfs_block_group *block_group)
1051 struct btrfs_root *extent_root = trans->fs_info->extent_root;
1052 struct btrfs_path *path, *path2;
1053 struct btrfs_key key;
1057 path = btrfs_alloc_path();
1060 path->reada = READA_FORWARD;
1062 path2 = btrfs_alloc_path();
1064 btrfs_free_path(path);
1068 ret = add_new_free_space_info(trans, block_group, path2);
1072 mutex_lock(&block_group->free_space_lock);
1075 * Iterate through all of the extent and metadata items in this block
1076 * group, adding the free space between them and the free space at the
1077 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1078 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1081 key.objectid = block_group->start;
1082 key.type = BTRFS_EXTENT_ITEM_KEY;
1085 ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
1090 start = block_group->start;
1091 end = block_group->start + block_group->length;
1093 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1095 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1096 key.type == BTRFS_METADATA_ITEM_KEY) {
1097 if (key.objectid >= end)
1100 if (start < key.objectid) {
1101 ret = __add_to_free_space_tree(trans,
1109 start = key.objectid;
1110 if (key.type == BTRFS_METADATA_ITEM_KEY)
1111 start += trans->fs_info->nodesize;
1113 start += key.offset;
1114 } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
1115 if (key.objectid != block_group->start)
1119 ret = btrfs_next_item(extent_root, path);
1126 ret = __add_to_free_space_tree(trans, block_group, path2,
1127 start, end - start);
1134 mutex_unlock(&block_group->free_space_lock);
1136 btrfs_free_path(path2);
1137 btrfs_free_path(path);
1141 int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
1143 struct btrfs_trans_handle *trans;
1144 struct btrfs_root *tree_root = fs_info->tree_root;
1145 struct btrfs_root *free_space_root;
1146 struct btrfs_block_group *block_group;
1147 struct rb_node *node;
1150 trans = btrfs_start_transaction(tree_root, 0);
1152 return PTR_ERR(trans);
1154 set_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1155 set_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1156 free_space_root = btrfs_create_tree(trans,
1157 BTRFS_FREE_SPACE_TREE_OBJECTID);
1158 if (IS_ERR(free_space_root)) {
1159 ret = PTR_ERR(free_space_root);
1162 fs_info->free_space_root = free_space_root;
1164 node = rb_first(&fs_info->block_group_cache_tree);
1166 block_group = rb_entry(node, struct btrfs_block_group,
1168 ret = populate_free_space_tree(trans, block_group);
1171 node = rb_next(node);
1174 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1175 btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1176 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1177 ret = btrfs_commit_transaction(trans);
1180 * Now that we've committed the transaction any reading of our commit
1181 * root will be safe, so we can cache from the free space tree now.
1183 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1187 clear_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags);
1188 clear_bit(BTRFS_FS_FREE_SPACE_TREE_UNTRUSTED, &fs_info->flags);
1189 btrfs_abort_transaction(trans, ret);
1190 btrfs_end_transaction(trans);
1194 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root)
1197 struct btrfs_path *path;
1198 struct btrfs_key key;
1202 path = btrfs_alloc_path();
1206 path->leave_spinning = 1;
1213 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1217 nr = btrfs_header_nritems(path->nodes[0]);
1222 ret = btrfs_del_items(trans, root, path, 0, nr);
1226 btrfs_release_path(path);
1231 btrfs_free_path(path);
1235 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
1237 struct btrfs_trans_handle *trans;
1238 struct btrfs_root *tree_root = fs_info->tree_root;
1239 struct btrfs_root *free_space_root = fs_info->free_space_root;
1242 trans = btrfs_start_transaction(tree_root, 0);
1244 return PTR_ERR(trans);
1246 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
1247 btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
1248 fs_info->free_space_root = NULL;
1250 ret = clear_free_space_tree(trans, free_space_root);
1254 ret = btrfs_del_root(trans, &free_space_root->root_key);
1258 list_del(&free_space_root->dirty_list);
1260 btrfs_tree_lock(free_space_root->node);
1261 btrfs_clean_tree_block(free_space_root->node);
1262 btrfs_tree_unlock(free_space_root->node);
1263 btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
1266 btrfs_put_root(free_space_root);
1268 return btrfs_commit_transaction(trans);
1271 btrfs_abort_transaction(trans, ret);
1272 btrfs_end_transaction(trans);
1276 static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
1277 struct btrfs_block_group *block_group,
1278 struct btrfs_path *path)
1282 block_group->needs_free_space = 0;
1284 ret = add_new_free_space_info(trans, block_group, path);
1288 return __add_to_free_space_tree(trans, block_group, path,
1290 block_group->length);
1293 int add_block_group_free_space(struct btrfs_trans_handle *trans,
1294 struct btrfs_block_group *block_group)
1296 struct btrfs_fs_info *fs_info = trans->fs_info;
1297 struct btrfs_path *path = NULL;
1300 if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
1303 mutex_lock(&block_group->free_space_lock);
1304 if (!block_group->needs_free_space)
1307 path = btrfs_alloc_path();
1313 ret = __add_block_group_free_space(trans, block_group, path);
1316 btrfs_free_path(path);
1317 mutex_unlock(&block_group->free_space_lock);
1319 btrfs_abort_transaction(trans, ret);
1323 int remove_block_group_free_space(struct btrfs_trans_handle *trans,
1324 struct btrfs_block_group *block_group)
1326 struct btrfs_root *root = trans->fs_info->free_space_root;
1327 struct btrfs_path *path;
1328 struct btrfs_key key, found_key;
1329 struct extent_buffer *leaf;
1334 if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
1337 if (block_group->needs_free_space) {
1338 /* We never added this block group to the free space tree. */
1342 path = btrfs_alloc_path();
1348 start = block_group->start;
1349 end = block_group->start + block_group->length;
1351 key.objectid = end - 1;
1353 key.offset = (u64)-1;
1356 ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
1360 leaf = path->nodes[0];
1363 while (path->slots[0] > 0) {
1364 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
1366 if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
1367 ASSERT(found_key.objectid == block_group->start);
1368 ASSERT(found_key.offset == block_group->length);
1373 } else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
1374 found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
1375 ASSERT(found_key.objectid >= start);
1376 ASSERT(found_key.objectid < end);
1377 ASSERT(found_key.objectid + found_key.offset <= end);
1385 ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
1388 btrfs_release_path(path);
1393 btrfs_free_path(path);
1395 btrfs_abort_transaction(trans, ret);
1399 static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
1400 struct btrfs_path *path,
1401 u32 expected_extent_count)
1403 struct btrfs_block_group *block_group;
1404 struct btrfs_fs_info *fs_info;
1405 struct btrfs_root *root;
1406 struct btrfs_key key;
1407 int prev_bit = 0, bit;
1408 /* Initialize to silence GCC. */
1409 u64 extent_start = 0;
1411 u64 total_found = 0;
1412 u32 extent_count = 0;
1415 block_group = caching_ctl->block_group;
1416 fs_info = block_group->fs_info;
1417 root = fs_info->free_space_root;
1419 end = block_group->start + block_group->length;
1422 ret = btrfs_next_item(root, path);
1428 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1430 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1433 ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
1434 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1436 caching_ctl->progress = key.objectid;
1438 offset = key.objectid;
1439 while (offset < key.objectid + key.offset) {
1440 bit = free_space_test_bit(block_group, path, offset);
1441 if (prev_bit == 0 && bit == 1) {
1442 extent_start = offset;
1443 } else if (prev_bit == 1 && bit == 0) {
1444 total_found += add_new_free_space(block_group,
1447 if (total_found > CACHING_CTL_WAKE_UP) {
1449 wake_up(&caching_ctl->wait);
1454 offset += fs_info->sectorsize;
1457 if (prev_bit == 1) {
1458 total_found += add_new_free_space(block_group, extent_start,
1463 if (extent_count != expected_extent_count) {
1465 "incorrect extent count for %llu; counted %u, expected %u",
1466 block_group->start, extent_count,
1467 expected_extent_count);
1473 caching_ctl->progress = (u64)-1;
1480 static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
1481 struct btrfs_path *path,
1482 u32 expected_extent_count)
1484 struct btrfs_block_group *block_group;
1485 struct btrfs_fs_info *fs_info;
1486 struct btrfs_root *root;
1487 struct btrfs_key key;
1489 u64 total_found = 0;
1490 u32 extent_count = 0;
1493 block_group = caching_ctl->block_group;
1494 fs_info = block_group->fs_info;
1495 root = fs_info->free_space_root;
1497 end = block_group->start + block_group->length;
1500 ret = btrfs_next_item(root, path);
1506 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1508 if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
1511 ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
1512 ASSERT(key.objectid < end && key.objectid + key.offset <= end);
1514 caching_ctl->progress = key.objectid;
1516 total_found += add_new_free_space(block_group, key.objectid,
1517 key.objectid + key.offset);
1518 if (total_found > CACHING_CTL_WAKE_UP) {
1520 wake_up(&caching_ctl->wait);
1525 if (extent_count != expected_extent_count) {
1527 "incorrect extent count for %llu; counted %u, expected %u",
1528 block_group->start, extent_count,
1529 expected_extent_count);
1535 caching_ctl->progress = (u64)-1;
1542 int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
1544 struct btrfs_block_group *block_group;
1545 struct btrfs_free_space_info *info;
1546 struct btrfs_path *path;
1547 u32 extent_count, flags;
1550 block_group = caching_ctl->block_group;
1552 path = btrfs_alloc_path();
1557 * Just like caching_thread() doesn't want to deadlock on the extent
1558 * tree, we don't want to deadlock on the free space tree.
1560 path->skip_locking = 1;
1561 path->search_commit_root = 1;
1562 path->reada = READA_FORWARD;
1564 info = search_free_space_info(NULL, block_group, path, 0);
1566 ret = PTR_ERR(info);
1569 extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
1570 flags = btrfs_free_space_flags(path->nodes[0], info);
1573 * We left path pointing to the free space info item, so now
1574 * load_free_space_foo can just iterate through the free space tree from
1577 if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
1578 ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
1580 ret = load_free_space_extents(caching_ctl, path, extent_count);
1583 btrfs_free_path(path);