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 unsigned long *alloc_bitmap(u32 bitmap_size
)
159 * The allocation size varies, observed numbers were < 4K up to 16K.
160 * Using vmalloc unconditionally would be too heavy, we'll try
161 * contiguous allocations first.
163 if (bitmap_size
<= PAGE_SIZE
)
164 return kzalloc(bitmap_size
, GFP_NOFS
);
166 mem
= kzalloc(bitmap_size
, GFP_NOFS
| __GFP_NOWARN
);
170 return __vmalloc(bitmap_size
, GFP_NOFS
| __GFP_HIGHMEM
| __GFP_ZERO
,
174 int convert_free_space_to_bitmaps(struct btrfs_trans_handle
*trans
,
175 struct btrfs_fs_info
*fs_info
,
176 struct btrfs_block_group_cache
*block_group
,
177 struct btrfs_path
*path
)
179 struct btrfs_root
*root
= fs_info
->free_space_root
;
180 struct btrfs_free_space_info
*info
;
181 struct btrfs_key key
, found_key
;
182 struct extent_buffer
*leaf
;
183 unsigned long *bitmap
;
187 u32 bitmap_size
, flags
, expected_extent_count
;
188 u32 extent_count
= 0;
192 bitmap_size
= free_space_bitmap_size(block_group
->key
.offset
,
193 block_group
->sectorsize
);
194 bitmap
= alloc_bitmap(bitmap_size
);
200 start
= block_group
->key
.objectid
;
201 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
203 key
.objectid
= end
- 1;
205 key
.offset
= (u64
)-1;
208 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
212 leaf
= path
->nodes
[0];
215 while (path
->slots
[0] > 0) {
216 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0] - 1);
218 if (found_key
.type
== BTRFS_FREE_SPACE_INFO_KEY
) {
219 ASSERT(found_key
.objectid
== block_group
->key
.objectid
);
220 ASSERT(found_key
.offset
== block_group
->key
.offset
);
223 } else if (found_key
.type
== BTRFS_FREE_SPACE_EXTENT_KEY
) {
226 ASSERT(found_key
.objectid
>= start
);
227 ASSERT(found_key
.objectid
< end
);
228 ASSERT(found_key
.objectid
+ found_key
.offset
<= end
);
230 first
= div_u64(found_key
.objectid
- start
,
231 block_group
->sectorsize
);
232 last
= div_u64(found_key
.objectid
+ found_key
.offset
- start
,
233 block_group
->sectorsize
);
234 bitmap_set(bitmap
, first
, last
- first
);
244 ret
= btrfs_del_items(trans
, root
, path
, path
->slots
[0], nr
);
247 btrfs_release_path(path
);
250 info
= search_free_space_info(trans
, fs_info
, block_group
, path
, 1);
255 leaf
= path
->nodes
[0];
256 flags
= btrfs_free_space_flags(leaf
, info
);
257 flags
|= BTRFS_FREE_SPACE_USING_BITMAPS
;
258 btrfs_set_free_space_flags(leaf
, info
, flags
);
259 expected_extent_count
= btrfs_free_space_extent_count(leaf
, info
);
260 btrfs_mark_buffer_dirty(leaf
);
261 btrfs_release_path(path
);
263 if (extent_count
!= expected_extent_count
) {
264 btrfs_err(fs_info
, "incorrect extent count for %llu; counted %u, expected %u",
265 block_group
->key
.objectid
, extent_count
,
266 expected_extent_count
);
272 bitmap_cursor
= (char *)bitmap
;
273 bitmap_range
= block_group
->sectorsize
* BTRFS_FREE_SPACE_BITMAP_BITS
;
280 extent_size
= min(end
- i
, bitmap_range
);
281 data_size
= free_space_bitmap_size(extent_size
,
282 block_group
->sectorsize
);
285 key
.type
= BTRFS_FREE_SPACE_BITMAP_KEY
;
286 key
.offset
= extent_size
;
288 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
293 leaf
= path
->nodes
[0];
294 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
295 write_extent_buffer(leaf
, bitmap_cursor
, ptr
,
297 btrfs_mark_buffer_dirty(leaf
);
298 btrfs_release_path(path
);
301 bitmap_cursor
+= data_size
;
308 btrfs_abort_transaction(trans
, root
, ret
);
312 int convert_free_space_to_extents(struct btrfs_trans_handle
*trans
,
313 struct btrfs_fs_info
*fs_info
,
314 struct btrfs_block_group_cache
*block_group
,
315 struct btrfs_path
*path
)
317 struct btrfs_root
*root
= fs_info
->free_space_root
;
318 struct btrfs_free_space_info
*info
;
319 struct btrfs_key key
, found_key
;
320 struct extent_buffer
*leaf
;
321 unsigned long *bitmap
;
323 /* Initialize to silence GCC. */
324 u64 extent_start
= 0;
326 u32 bitmap_size
, flags
, expected_extent_count
;
327 int prev_bit
= 0, bit
, bitnr
;
328 u32 extent_count
= 0;
332 bitmap_size
= free_space_bitmap_size(block_group
->key
.offset
,
333 block_group
->sectorsize
);
334 bitmap
= alloc_bitmap(bitmap_size
);
340 start
= block_group
->key
.objectid
;
341 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
343 key
.objectid
= end
- 1;
345 key
.offset
= (u64
)-1;
348 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
352 leaf
= path
->nodes
[0];
355 while (path
->slots
[0] > 0) {
356 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0] - 1);
358 if (found_key
.type
== BTRFS_FREE_SPACE_INFO_KEY
) {
359 ASSERT(found_key
.objectid
== block_group
->key
.objectid
);
360 ASSERT(found_key
.offset
== block_group
->key
.offset
);
363 } else if (found_key
.type
== BTRFS_FREE_SPACE_BITMAP_KEY
) {
366 u32 bitmap_pos
, data_size
;
368 ASSERT(found_key
.objectid
>= start
);
369 ASSERT(found_key
.objectid
< end
);
370 ASSERT(found_key
.objectid
+ found_key
.offset
<= end
);
372 bitmap_pos
= div_u64(found_key
.objectid
- start
,
373 block_group
->sectorsize
*
375 bitmap_cursor
= ((char *)bitmap
) + bitmap_pos
;
376 data_size
= free_space_bitmap_size(found_key
.offset
,
377 block_group
->sectorsize
);
379 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1);
380 read_extent_buffer(leaf
, bitmap_cursor
, ptr
,
390 ret
= btrfs_del_items(trans
, root
, path
, path
->slots
[0], nr
);
393 btrfs_release_path(path
);
396 info
= search_free_space_info(trans
, fs_info
, block_group
, path
, 1);
401 leaf
= path
->nodes
[0];
402 flags
= btrfs_free_space_flags(leaf
, info
);
403 flags
&= ~BTRFS_FREE_SPACE_USING_BITMAPS
;
404 btrfs_set_free_space_flags(leaf
, info
, flags
);
405 expected_extent_count
= btrfs_free_space_extent_count(leaf
, info
);
406 btrfs_mark_buffer_dirty(leaf
);
407 btrfs_release_path(path
);
411 while (offset
< end
) {
412 bit
= !!test_bit(bitnr
, bitmap
);
413 if (prev_bit
== 0 && bit
== 1) {
414 extent_start
= offset
;
415 } else if (prev_bit
== 1 && bit
== 0) {
416 key
.objectid
= extent_start
;
417 key
.type
= BTRFS_FREE_SPACE_EXTENT_KEY
;
418 key
.offset
= offset
- extent_start
;
420 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
423 btrfs_release_path(path
);
428 offset
+= block_group
->sectorsize
;
432 key
.objectid
= extent_start
;
433 key
.type
= BTRFS_FREE_SPACE_EXTENT_KEY
;
434 key
.offset
= end
- extent_start
;
436 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
439 btrfs_release_path(path
);
444 if (extent_count
!= expected_extent_count
) {
445 btrfs_err(fs_info
, "incorrect extent count for %llu; counted %u, expected %u",
446 block_group
->key
.objectid
, extent_count
,
447 expected_extent_count
);
457 btrfs_abort_transaction(trans
, root
, ret
);
461 static int update_free_space_extent_count(struct btrfs_trans_handle
*trans
,
462 struct btrfs_fs_info
*fs_info
,
463 struct btrfs_block_group_cache
*block_group
,
464 struct btrfs_path
*path
,
467 struct btrfs_free_space_info
*info
;
472 if (new_extents
== 0)
475 info
= search_free_space_info(trans
, fs_info
, block_group
, path
, 1);
480 flags
= btrfs_free_space_flags(path
->nodes
[0], info
);
481 extent_count
= btrfs_free_space_extent_count(path
->nodes
[0], info
);
483 extent_count
+= new_extents
;
484 btrfs_set_free_space_extent_count(path
->nodes
[0], info
, extent_count
);
485 btrfs_mark_buffer_dirty(path
->nodes
[0]);
486 btrfs_release_path(path
);
488 if (!(flags
& BTRFS_FREE_SPACE_USING_BITMAPS
) &&
489 extent_count
> block_group
->bitmap_high_thresh
) {
490 ret
= convert_free_space_to_bitmaps(trans
, fs_info
, block_group
,
492 } else if ((flags
& BTRFS_FREE_SPACE_USING_BITMAPS
) &&
493 extent_count
< block_group
->bitmap_low_thresh
) {
494 ret
= convert_free_space_to_extents(trans
, fs_info
, block_group
,
502 int free_space_test_bit(struct btrfs_block_group_cache
*block_group
,
503 struct btrfs_path
*path
, u64 offset
)
505 struct extent_buffer
*leaf
;
506 struct btrfs_key key
;
507 u64 found_start
, found_end
;
508 unsigned long ptr
, i
;
510 leaf
= path
->nodes
[0];
511 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
512 ASSERT(key
.type
== BTRFS_FREE_SPACE_BITMAP_KEY
);
514 found_start
= key
.objectid
;
515 found_end
= key
.objectid
+ key
.offset
;
516 ASSERT(offset
>= found_start
&& offset
< found_end
);
518 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
519 i
= div_u64(offset
- found_start
, block_group
->sectorsize
);
520 return !!extent_buffer_test_bit(leaf
, ptr
, i
);
523 static void free_space_set_bits(struct btrfs_block_group_cache
*block_group
,
524 struct btrfs_path
*path
, u64
*start
, u64
*size
,
527 struct extent_buffer
*leaf
;
528 struct btrfs_key key
;
529 u64 end
= *start
+ *size
;
530 u64 found_start
, found_end
;
531 unsigned long ptr
, first
, last
;
533 leaf
= path
->nodes
[0];
534 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
535 ASSERT(key
.type
== BTRFS_FREE_SPACE_BITMAP_KEY
);
537 found_start
= key
.objectid
;
538 found_end
= key
.objectid
+ key
.offset
;
539 ASSERT(*start
>= found_start
&& *start
< found_end
);
540 ASSERT(end
> found_start
);
545 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
546 first
= div_u64(*start
- found_start
, block_group
->sectorsize
);
547 last
= div_u64(end
- found_start
, block_group
->sectorsize
);
549 extent_buffer_bitmap_set(leaf
, ptr
, first
, last
- first
);
551 extent_buffer_bitmap_clear(leaf
, ptr
, first
, last
- first
);
552 btrfs_mark_buffer_dirty(leaf
);
554 *size
-= end
- *start
;
559 * We can't use btrfs_next_item() in modify_free_space_bitmap() because
560 * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
561 * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
564 static int free_space_next_bitmap(struct btrfs_trans_handle
*trans
,
565 struct btrfs_root
*root
, struct btrfs_path
*p
)
567 struct btrfs_key key
;
569 if (p
->slots
[0] + 1 < btrfs_header_nritems(p
->nodes
[0])) {
574 btrfs_item_key_to_cpu(p
->nodes
[0], &key
, p
->slots
[0]);
575 btrfs_release_path(p
);
577 key
.objectid
+= key
.offset
;
579 key
.offset
= (u64
)-1;
581 return btrfs_search_prev_slot(trans
, root
, &key
, p
, 0, 1);
585 * If remove is 1, then we are removing free space, thus clearing bits in the
586 * bitmap. If remove is 0, then we are adding free space, thus setting bits in
589 static int modify_free_space_bitmap(struct btrfs_trans_handle
*trans
,
590 struct btrfs_fs_info
*fs_info
,
591 struct btrfs_block_group_cache
*block_group
,
592 struct btrfs_path
*path
,
593 u64 start
, u64 size
, int remove
)
595 struct btrfs_root
*root
= fs_info
->free_space_root
;
596 struct btrfs_key key
;
597 u64 end
= start
+ size
;
598 u64 cur_start
, cur_size
;
599 int prev_bit
, next_bit
;
604 * Read the bit for the block immediately before the extent of space if
605 * that block is within the block group.
607 if (start
> block_group
->key
.objectid
) {
608 u64 prev_block
= start
- block_group
->sectorsize
;
610 key
.objectid
= prev_block
;
612 key
.offset
= (u64
)-1;
614 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, 0, 1);
618 prev_bit
= free_space_test_bit(block_group
, path
, prev_block
);
620 /* The previous block may have been in the previous bitmap. */
621 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
622 if (start
>= key
.objectid
+ key
.offset
) {
623 ret
= free_space_next_bitmap(trans
, root
, path
);
628 key
.objectid
= start
;
630 key
.offset
= (u64
)-1;
632 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, 0, 1);
640 * Iterate over all of the bitmaps overlapped by the extent of space,
641 * clearing/setting bits as required.
646 free_space_set_bits(block_group
, path
, &cur_start
, &cur_size
,
650 ret
= free_space_next_bitmap(trans
, root
, path
);
656 * Read the bit for the block immediately after the extent of space if
657 * that block is within the block group.
659 if (end
< block_group
->key
.objectid
+ block_group
->key
.offset
) {
660 /* The next block may be in the next bitmap. */
661 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
662 if (end
>= key
.objectid
+ key
.offset
) {
663 ret
= free_space_next_bitmap(trans
, root
, path
);
668 next_bit
= free_space_test_bit(block_group
, path
, end
);
676 /* Leftover on the left. */
680 /* Leftover on the right. */
686 /* Merging with neighbor on the left. */
690 /* Merging with neighbor on the right. */
695 btrfs_release_path(path
);
696 ret
= update_free_space_extent_count(trans
, fs_info
, block_group
, path
,
703 static int remove_free_space_extent(struct btrfs_trans_handle
*trans
,
704 struct btrfs_fs_info
*fs_info
,
705 struct btrfs_block_group_cache
*block_group
,
706 struct btrfs_path
*path
,
709 struct btrfs_root
*root
= fs_info
->free_space_root
;
710 struct btrfs_key key
;
711 u64 found_start
, found_end
;
712 u64 end
= start
+ size
;
713 int new_extents
= -1;
716 key
.objectid
= start
;
718 key
.offset
= (u64
)-1;
720 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
724 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
726 ASSERT(key
.type
== BTRFS_FREE_SPACE_EXTENT_KEY
);
728 found_start
= key
.objectid
;
729 found_end
= key
.objectid
+ key
.offset
;
730 ASSERT(start
>= found_start
&& end
<= found_end
);
733 * Okay, now that we've found the free space extent which contains the
734 * free space that we are removing, there are four cases:
736 * 1. We're using the whole extent: delete the key we found and
737 * decrement the free space extent count.
738 * 2. We are using part of the extent starting at the beginning: delete
739 * the key we found and insert a new key representing the leftover at
740 * the end. There is no net change in the number of extents.
741 * 3. We are using part of the extent ending at the end: delete the key
742 * we found and insert a new key representing the leftover at the
743 * beginning. There is no net change in the number of extents.
744 * 4. We are using part of the extent in the middle: delete the key we
745 * found and insert two new keys representing the leftovers on each
746 * side. Where we used to have one extent, we now have two, so increment
747 * the extent count. We may need to convert the block group to bitmaps
751 /* Delete the existing key (cases 1-4). */
752 ret
= btrfs_del_item(trans
, root
, path
);
756 /* Add a key for leftovers at the beginning (cases 3 and 4). */
757 if (start
> found_start
) {
758 key
.objectid
= found_start
;
759 key
.type
= BTRFS_FREE_SPACE_EXTENT_KEY
;
760 key
.offset
= start
- found_start
;
762 btrfs_release_path(path
);
763 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
769 /* Add a key for leftovers at the end (cases 2 and 4). */
770 if (end
< found_end
) {
772 key
.type
= BTRFS_FREE_SPACE_EXTENT_KEY
;
773 key
.offset
= found_end
- end
;
775 btrfs_release_path(path
);
776 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
782 btrfs_release_path(path
);
783 ret
= update_free_space_extent_count(trans
, fs_info
, block_group
, path
,
790 int __remove_from_free_space_tree(struct btrfs_trans_handle
*trans
,
791 struct btrfs_fs_info
*fs_info
,
792 struct btrfs_block_group_cache
*block_group
,
793 struct btrfs_path
*path
, u64 start
, u64 size
)
795 struct btrfs_free_space_info
*info
;
799 if (block_group
->needs_free_space
) {
800 ret
= __add_block_group_free_space(trans
, fs_info
, block_group
,
806 info
= search_free_space_info(NULL
, fs_info
, block_group
, path
, 0);
808 return PTR_ERR(info
);
809 flags
= btrfs_free_space_flags(path
->nodes
[0], info
);
810 btrfs_release_path(path
);
812 if (flags
& BTRFS_FREE_SPACE_USING_BITMAPS
) {
813 return modify_free_space_bitmap(trans
, fs_info
, block_group
,
814 path
, start
, size
, 1);
816 return remove_free_space_extent(trans
, fs_info
, block_group
,
821 int remove_from_free_space_tree(struct btrfs_trans_handle
*trans
,
822 struct btrfs_fs_info
*fs_info
,
825 struct btrfs_block_group_cache
*block_group
;
826 struct btrfs_path
*path
;
829 if (!btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
832 path
= btrfs_alloc_path();
838 block_group
= btrfs_lookup_block_group(fs_info
, start
);
845 mutex_lock(&block_group
->free_space_lock
);
846 ret
= __remove_from_free_space_tree(trans
, fs_info
, block_group
, path
,
848 mutex_unlock(&block_group
->free_space_lock
);
850 btrfs_put_block_group(block_group
);
852 btrfs_free_path(path
);
854 btrfs_abort_transaction(trans
, fs_info
->free_space_root
, ret
);
858 static int add_free_space_extent(struct btrfs_trans_handle
*trans
,
859 struct btrfs_fs_info
*fs_info
,
860 struct btrfs_block_group_cache
*block_group
,
861 struct btrfs_path
*path
,
864 struct btrfs_root
*root
= fs_info
->free_space_root
;
865 struct btrfs_key key
, new_key
;
866 u64 found_start
, found_end
;
867 u64 end
= start
+ size
;
872 * We are adding a new extent of free space, but we need to merge
873 * extents. There are four cases here:
875 * 1. The new extent does not have any immediate neighbors to merge
876 * with: add the new key and increment the free space extent count. We
877 * may need to convert the block group to bitmaps as a result.
878 * 2. The new extent has an immediate neighbor before it: remove the
879 * previous key and insert a new key combining both of them. There is no
880 * net change in the number of extents.
881 * 3. The new extent has an immediate neighbor after it: remove the next
882 * key and insert a new key combining both of them. There is no net
883 * change in the number of extents.
884 * 4. The new extent has immediate neighbors on both sides: remove both
885 * of the keys and insert a new key combining all of them. Where we used
886 * to have two extents, we now have one, so decrement the extent count.
889 new_key
.objectid
= start
;
890 new_key
.type
= BTRFS_FREE_SPACE_EXTENT_KEY
;
891 new_key
.offset
= size
;
893 /* Search for a neighbor on the left. */
894 if (start
== block_group
->key
.objectid
)
896 key
.objectid
= start
- 1;
898 key
.offset
= (u64
)-1;
900 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
904 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
906 if (key
.type
!= BTRFS_FREE_SPACE_EXTENT_KEY
) {
907 ASSERT(key
.type
== BTRFS_FREE_SPACE_INFO_KEY
);
908 btrfs_release_path(path
);
912 found_start
= key
.objectid
;
913 found_end
= key
.objectid
+ key
.offset
;
914 ASSERT(found_start
>= block_group
->key
.objectid
&&
915 found_end
> block_group
->key
.objectid
);
916 ASSERT(found_start
< start
&& found_end
<= start
);
919 * Delete the neighbor on the left and absorb it into the new key (cases
922 if (found_end
== start
) {
923 ret
= btrfs_del_item(trans
, root
, path
);
926 new_key
.objectid
= found_start
;
927 new_key
.offset
+= key
.offset
;
930 btrfs_release_path(path
);
933 /* Search for a neighbor on the right. */
934 if (end
== block_group
->key
.objectid
+ block_group
->key
.offset
)
938 key
.offset
= (u64
)-1;
940 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
944 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
946 if (key
.type
!= BTRFS_FREE_SPACE_EXTENT_KEY
) {
947 ASSERT(key
.type
== BTRFS_FREE_SPACE_INFO_KEY
);
948 btrfs_release_path(path
);
952 found_start
= key
.objectid
;
953 found_end
= key
.objectid
+ key
.offset
;
954 ASSERT(found_start
>= block_group
->key
.objectid
&&
955 found_end
> block_group
->key
.objectid
);
956 ASSERT((found_start
< start
&& found_end
<= start
) ||
957 (found_start
>= end
&& found_end
> end
));
960 * Delete the neighbor on the right and absorb it into the new key
963 if (found_start
== end
) {
964 ret
= btrfs_del_item(trans
, root
, path
);
967 new_key
.offset
+= key
.offset
;
970 btrfs_release_path(path
);
973 /* Insert the new key (cases 1-4). */
974 ret
= btrfs_insert_empty_item(trans
, root
, path
, &new_key
, 0);
978 btrfs_release_path(path
);
979 ret
= update_free_space_extent_count(trans
, fs_info
, block_group
, path
,
986 int __add_to_free_space_tree(struct btrfs_trans_handle
*trans
,
987 struct btrfs_fs_info
*fs_info
,
988 struct btrfs_block_group_cache
*block_group
,
989 struct btrfs_path
*path
, u64 start
, u64 size
)
991 struct btrfs_free_space_info
*info
;
995 if (block_group
->needs_free_space
) {
996 ret
= __add_block_group_free_space(trans
, fs_info
, block_group
,
1002 info
= search_free_space_info(NULL
, fs_info
, block_group
, path
, 0);
1004 return PTR_ERR(info
);
1005 flags
= btrfs_free_space_flags(path
->nodes
[0], info
);
1006 btrfs_release_path(path
);
1008 if (flags
& BTRFS_FREE_SPACE_USING_BITMAPS
) {
1009 return modify_free_space_bitmap(trans
, fs_info
, block_group
,
1010 path
, start
, size
, 0);
1012 return add_free_space_extent(trans
, fs_info
, block_group
, path
,
1017 int add_to_free_space_tree(struct btrfs_trans_handle
*trans
,
1018 struct btrfs_fs_info
*fs_info
,
1019 u64 start
, u64 size
)
1021 struct btrfs_block_group_cache
*block_group
;
1022 struct btrfs_path
*path
;
1025 if (!btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
1028 path
= btrfs_alloc_path();
1034 block_group
= btrfs_lookup_block_group(fs_info
, start
);
1041 mutex_lock(&block_group
->free_space_lock
);
1042 ret
= __add_to_free_space_tree(trans
, fs_info
, block_group
, path
, start
,
1044 mutex_unlock(&block_group
->free_space_lock
);
1046 btrfs_put_block_group(block_group
);
1048 btrfs_free_path(path
);
1050 btrfs_abort_transaction(trans
, fs_info
->free_space_root
, ret
);
1055 * Populate the free space tree by walking the extent tree. Operations on the
1056 * extent tree that happen as a result of writes to the free space tree will go
1057 * through the normal add/remove hooks.
1059 static int populate_free_space_tree(struct btrfs_trans_handle
*trans
,
1060 struct btrfs_fs_info
*fs_info
,
1061 struct btrfs_block_group_cache
*block_group
)
1063 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
1064 struct btrfs_path
*path
, *path2
;
1065 struct btrfs_key key
;
1069 path
= btrfs_alloc_path();
1074 path2
= btrfs_alloc_path();
1076 btrfs_free_path(path
);
1080 ret
= add_new_free_space_info(trans
, fs_info
, block_group
, path2
);
1084 mutex_lock(&block_group
->free_space_lock
);
1087 * Iterate through all of the extent and metadata items in this block
1088 * group, adding the free space between them and the free space at the
1089 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
1090 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
1093 key
.objectid
= block_group
->key
.objectid
;
1094 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1097 ret
= btrfs_search_slot_for_read(extent_root
, &key
, path
, 1, 0);
1102 start
= block_group
->key
.objectid
;
1103 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
1105 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1107 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
1108 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
1109 if (key
.objectid
>= end
)
1112 if (start
< key
.objectid
) {
1113 ret
= __add_to_free_space_tree(trans
, fs_info
,
1121 start
= key
.objectid
;
1122 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
1123 start
+= fs_info
->tree_root
->nodesize
;
1125 start
+= key
.offset
;
1126 } else if (key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1127 if (key
.objectid
!= block_group
->key
.objectid
)
1131 ret
= btrfs_next_item(extent_root
, path
);
1138 ret
= __add_to_free_space_tree(trans
, fs_info
, block_group
,
1139 path2
, start
, end
- start
);
1146 mutex_unlock(&block_group
->free_space_lock
);
1148 btrfs_free_path(path2
);
1149 btrfs_free_path(path
);
1153 int btrfs_create_free_space_tree(struct btrfs_fs_info
*fs_info
)
1155 struct btrfs_trans_handle
*trans
;
1156 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1157 struct btrfs_root
*free_space_root
;
1158 struct btrfs_block_group_cache
*block_group
;
1159 struct rb_node
*node
;
1162 trans
= btrfs_start_transaction(tree_root
, 0);
1164 return PTR_ERR(trans
);
1166 fs_info
->creating_free_space_tree
= 1;
1167 free_space_root
= btrfs_create_tree(trans
, fs_info
,
1168 BTRFS_FREE_SPACE_TREE_OBJECTID
);
1169 if (IS_ERR(free_space_root
)) {
1170 ret
= PTR_ERR(free_space_root
);
1173 fs_info
->free_space_root
= free_space_root
;
1175 node
= rb_first(&fs_info
->block_group_cache_tree
);
1177 block_group
= rb_entry(node
, struct btrfs_block_group_cache
,
1179 ret
= populate_free_space_tree(trans
, fs_info
, block_group
);
1182 node
= rb_next(node
);
1185 btrfs_set_fs_compat_ro(fs_info
, FREE_SPACE_TREE
);
1186 fs_info
->creating_free_space_tree
= 0;
1188 ret
= btrfs_commit_transaction(trans
, tree_root
);
1195 fs_info
->creating_free_space_tree
= 0;
1196 btrfs_abort_transaction(trans
, tree_root
, ret
);
1197 btrfs_end_transaction(trans
, tree_root
);
1201 static int clear_free_space_tree(struct btrfs_trans_handle
*trans
,
1202 struct btrfs_root
*root
)
1204 struct btrfs_path
*path
;
1205 struct btrfs_key key
;
1209 path
= btrfs_alloc_path();
1213 path
->leave_spinning
= 1;
1220 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1224 nr
= btrfs_header_nritems(path
->nodes
[0]);
1229 ret
= btrfs_del_items(trans
, root
, path
, 0, nr
);
1233 btrfs_release_path(path
);
1238 btrfs_free_path(path
);
1242 int btrfs_clear_free_space_tree(struct btrfs_fs_info
*fs_info
)
1244 struct btrfs_trans_handle
*trans
;
1245 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
1246 struct btrfs_root
*free_space_root
= fs_info
->free_space_root
;
1249 trans
= btrfs_start_transaction(tree_root
, 0);
1251 return PTR_ERR(trans
);
1253 btrfs_clear_fs_compat_ro(fs_info
, FREE_SPACE_TREE
);
1254 fs_info
->free_space_root
= NULL
;
1256 ret
= clear_free_space_tree(trans
, free_space_root
);
1260 ret
= btrfs_del_root(trans
, tree_root
, &free_space_root
->root_key
);
1264 list_del(&free_space_root
->dirty_list
);
1266 btrfs_tree_lock(free_space_root
->node
);
1267 clean_tree_block(trans
, tree_root
->fs_info
, free_space_root
->node
);
1268 btrfs_tree_unlock(free_space_root
->node
);
1269 btrfs_free_tree_block(trans
, free_space_root
, free_space_root
->node
,
1272 free_extent_buffer(free_space_root
->node
);
1273 free_extent_buffer(free_space_root
->commit_root
);
1274 kfree(free_space_root
);
1276 ret
= btrfs_commit_transaction(trans
, tree_root
);
1283 btrfs_abort_transaction(trans
, tree_root
, ret
);
1284 btrfs_end_transaction(trans
, tree_root
);
1288 static int __add_block_group_free_space(struct btrfs_trans_handle
*trans
,
1289 struct btrfs_fs_info
*fs_info
,
1290 struct btrfs_block_group_cache
*block_group
,
1291 struct btrfs_path
*path
)
1296 start
= block_group
->key
.objectid
;
1297 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
1299 block_group
->needs_free_space
= 0;
1301 ret
= add_new_free_space_info(trans
, fs_info
, block_group
, path
);
1305 return __add_to_free_space_tree(trans
, fs_info
, block_group
, path
,
1306 block_group
->key
.objectid
,
1307 block_group
->key
.offset
);
1310 int add_block_group_free_space(struct btrfs_trans_handle
*trans
,
1311 struct btrfs_fs_info
*fs_info
,
1312 struct btrfs_block_group_cache
*block_group
)
1314 struct btrfs_path
*path
= NULL
;
1317 if (!btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
1320 mutex_lock(&block_group
->free_space_lock
);
1321 if (!block_group
->needs_free_space
)
1324 path
= btrfs_alloc_path();
1330 ret
= __add_block_group_free_space(trans
, fs_info
, block_group
, path
);
1333 btrfs_free_path(path
);
1334 mutex_unlock(&block_group
->free_space_lock
);
1336 btrfs_abort_transaction(trans
, fs_info
->free_space_root
, ret
);
1340 int remove_block_group_free_space(struct btrfs_trans_handle
*trans
,
1341 struct btrfs_fs_info
*fs_info
,
1342 struct btrfs_block_group_cache
*block_group
)
1344 struct btrfs_root
*root
= fs_info
->free_space_root
;
1345 struct btrfs_path
*path
;
1346 struct btrfs_key key
, found_key
;
1347 struct extent_buffer
*leaf
;
1352 if (!btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
1355 if (block_group
->needs_free_space
) {
1356 /* We never added this block group to the free space tree. */
1360 path
= btrfs_alloc_path();
1366 start
= block_group
->key
.objectid
;
1367 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
1369 key
.objectid
= end
- 1;
1371 key
.offset
= (u64
)-1;
1374 ret
= btrfs_search_prev_slot(trans
, root
, &key
, path
, -1, 1);
1378 leaf
= path
->nodes
[0];
1381 while (path
->slots
[0] > 0) {
1382 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0] - 1);
1384 if (found_key
.type
== BTRFS_FREE_SPACE_INFO_KEY
) {
1385 ASSERT(found_key
.objectid
== block_group
->key
.objectid
);
1386 ASSERT(found_key
.offset
== block_group
->key
.offset
);
1391 } else if (found_key
.type
== BTRFS_FREE_SPACE_EXTENT_KEY
||
1392 found_key
.type
== BTRFS_FREE_SPACE_BITMAP_KEY
) {
1393 ASSERT(found_key
.objectid
>= start
);
1394 ASSERT(found_key
.objectid
< end
);
1395 ASSERT(found_key
.objectid
+ found_key
.offset
<= end
);
1403 ret
= btrfs_del_items(trans
, root
, path
, path
->slots
[0], nr
);
1406 btrfs_release_path(path
);
1411 btrfs_free_path(path
);
1413 btrfs_abort_transaction(trans
, root
, ret
);
1417 static int load_free_space_bitmaps(struct btrfs_caching_control
*caching_ctl
,
1418 struct btrfs_path
*path
,
1419 u32 expected_extent_count
)
1421 struct btrfs_block_group_cache
*block_group
;
1422 struct btrfs_fs_info
*fs_info
;
1423 struct btrfs_root
*root
;
1424 struct btrfs_key key
;
1425 int prev_bit
= 0, bit
;
1426 /* Initialize to silence GCC. */
1427 u64 extent_start
= 0;
1429 u64 total_found
= 0;
1430 u32 extent_count
= 0;
1433 block_group
= caching_ctl
->block_group
;
1434 fs_info
= block_group
->fs_info
;
1435 root
= fs_info
->free_space_root
;
1437 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
1440 ret
= btrfs_next_item(root
, path
);
1446 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1448 if (key
.type
== BTRFS_FREE_SPACE_INFO_KEY
)
1451 ASSERT(key
.type
== BTRFS_FREE_SPACE_BITMAP_KEY
);
1452 ASSERT(key
.objectid
< end
&& key
.objectid
+ key
.offset
<= end
);
1454 caching_ctl
->progress
= key
.objectid
;
1456 offset
= key
.objectid
;
1457 while (offset
< key
.objectid
+ key
.offset
) {
1458 bit
= free_space_test_bit(block_group
, path
, offset
);
1459 if (prev_bit
== 0 && bit
== 1) {
1460 extent_start
= offset
;
1461 } else if (prev_bit
== 1 && bit
== 0) {
1462 total_found
+= add_new_free_space(block_group
,
1466 if (total_found
> CACHING_CTL_WAKE_UP
) {
1468 wake_up(&caching_ctl
->wait
);
1473 offset
+= block_group
->sectorsize
;
1476 if (prev_bit
== 1) {
1477 total_found
+= add_new_free_space(block_group
, fs_info
,
1482 if (extent_count
!= expected_extent_count
) {
1483 btrfs_err(fs_info
, "incorrect extent count for %llu; counted %u, expected %u",
1484 block_group
->key
.objectid
, extent_count
,
1485 expected_extent_count
);
1491 caching_ctl
->progress
= (u64
)-1;
1498 static int load_free_space_extents(struct btrfs_caching_control
*caching_ctl
,
1499 struct btrfs_path
*path
,
1500 u32 expected_extent_count
)
1502 struct btrfs_block_group_cache
*block_group
;
1503 struct btrfs_fs_info
*fs_info
;
1504 struct btrfs_root
*root
;
1505 struct btrfs_key key
;
1507 u64 total_found
= 0;
1508 u32 extent_count
= 0;
1511 block_group
= caching_ctl
->block_group
;
1512 fs_info
= block_group
->fs_info
;
1513 root
= fs_info
->free_space_root
;
1515 end
= block_group
->key
.objectid
+ block_group
->key
.offset
;
1518 ret
= btrfs_next_item(root
, path
);
1524 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
1526 if (key
.type
== BTRFS_FREE_SPACE_INFO_KEY
)
1529 ASSERT(key
.type
== BTRFS_FREE_SPACE_EXTENT_KEY
);
1530 ASSERT(key
.objectid
< end
&& key
.objectid
+ key
.offset
<= end
);
1532 caching_ctl
->progress
= key
.objectid
;
1534 total_found
+= add_new_free_space(block_group
, fs_info
,
1536 key
.objectid
+ key
.offset
);
1537 if (total_found
> CACHING_CTL_WAKE_UP
) {
1539 wake_up(&caching_ctl
->wait
);
1544 if (extent_count
!= expected_extent_count
) {
1545 btrfs_err(fs_info
, "incorrect extent count for %llu; counted %u, expected %u",
1546 block_group
->key
.objectid
, extent_count
,
1547 expected_extent_count
);
1553 caching_ctl
->progress
= (u64
)-1;
1560 int load_free_space_tree(struct btrfs_caching_control
*caching_ctl
)
1562 struct btrfs_block_group_cache
*block_group
;
1563 struct btrfs_fs_info
*fs_info
;
1564 struct btrfs_free_space_info
*info
;
1565 struct btrfs_path
*path
;
1566 u32 extent_count
, flags
;
1569 block_group
= caching_ctl
->block_group
;
1570 fs_info
= block_group
->fs_info
;
1572 path
= btrfs_alloc_path();
1577 * Just like caching_thread() doesn't want to deadlock on the extent
1578 * tree, we don't want to deadlock on the free space tree.
1580 path
->skip_locking
= 1;
1581 path
->search_commit_root
= 1;
1584 info
= search_free_space_info(NULL
, fs_info
, block_group
, path
, 0);
1586 ret
= PTR_ERR(info
);
1589 extent_count
= btrfs_free_space_extent_count(path
->nodes
[0], info
);
1590 flags
= btrfs_free_space_flags(path
->nodes
[0], info
);
1593 * We left path pointing to the free space info item, so now
1594 * load_free_space_foo can just iterate through the free space tree from
1597 if (flags
& BTRFS_FREE_SPACE_USING_BITMAPS
)
1598 ret
= load_free_space_bitmaps(caching_ctl
, path
, extent_count
);
1600 ret
= load_free_space_extents(caching_ctl
, path
, extent_count
);
1603 btrfs_free_path(path
);