Loading...
1/*
2 * Copyright (C) 2011 Red Hat, Inc.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-space-map-common.h"
8#include "dm-transaction-manager.h"
9
10#include <linux/bitops.h>
11#include <linux/device-mapper.h>
12
13#define DM_MSG_PREFIX "space map common"
14
15/*----------------------------------------------------------------*/
16
17/*
18 * Index validator.
19 */
20#define INDEX_CSUM_XOR 160478
21
22static void index_prepare_for_write(struct dm_block_validator *v,
23 struct dm_block *b,
24 size_t block_size)
25{
26 struct disk_metadata_index *mi_le = dm_block_data(b);
27
28 mi_le->blocknr = cpu_to_le64(dm_block_location(b));
29 mi_le->csum = cpu_to_le32(dm_bm_checksum(&mi_le->padding,
30 block_size - sizeof(__le32),
31 INDEX_CSUM_XOR));
32}
33
34static int index_check(struct dm_block_validator *v,
35 struct dm_block *b,
36 size_t block_size)
37{
38 struct disk_metadata_index *mi_le = dm_block_data(b);
39 __le32 csum_disk;
40
41 if (dm_block_location(b) != le64_to_cpu(mi_le->blocknr)) {
42 DMERR_LIMIT("index_check failed: blocknr %llu != wanted %llu",
43 le64_to_cpu(mi_le->blocknr), dm_block_location(b));
44 return -ENOTBLK;
45 }
46
47 csum_disk = cpu_to_le32(dm_bm_checksum(&mi_le->padding,
48 block_size - sizeof(__le32),
49 INDEX_CSUM_XOR));
50 if (csum_disk != mi_le->csum) {
51 DMERR_LIMIT("index_check failed: csum %u != wanted %u",
52 le32_to_cpu(csum_disk), le32_to_cpu(mi_le->csum));
53 return -EILSEQ;
54 }
55
56 return 0;
57}
58
59static struct dm_block_validator index_validator = {
60 .name = "index",
61 .prepare_for_write = index_prepare_for_write,
62 .check = index_check
63};
64
65/*----------------------------------------------------------------*/
66
67/*
68 * Bitmap validator
69 */
70#define BITMAP_CSUM_XOR 240779
71
72static void dm_bitmap_prepare_for_write(struct dm_block_validator *v,
73 struct dm_block *b,
74 size_t block_size)
75{
76 struct disk_bitmap_header *disk_header = dm_block_data(b);
77
78 disk_header->blocknr = cpu_to_le64(dm_block_location(b));
79 disk_header->csum = cpu_to_le32(dm_bm_checksum(&disk_header->not_used,
80 block_size - sizeof(__le32),
81 BITMAP_CSUM_XOR));
82}
83
84static int dm_bitmap_check(struct dm_block_validator *v,
85 struct dm_block *b,
86 size_t block_size)
87{
88 struct disk_bitmap_header *disk_header = dm_block_data(b);
89 __le32 csum_disk;
90
91 if (dm_block_location(b) != le64_to_cpu(disk_header->blocknr)) {
92 DMERR_LIMIT("bitmap check failed: blocknr %llu != wanted %llu",
93 le64_to_cpu(disk_header->blocknr), dm_block_location(b));
94 return -ENOTBLK;
95 }
96
97 csum_disk = cpu_to_le32(dm_bm_checksum(&disk_header->not_used,
98 block_size - sizeof(__le32),
99 BITMAP_CSUM_XOR));
100 if (csum_disk != disk_header->csum) {
101 DMERR_LIMIT("bitmap check failed: csum %u != wanted %u",
102 le32_to_cpu(csum_disk), le32_to_cpu(disk_header->csum));
103 return -EILSEQ;
104 }
105
106 return 0;
107}
108
109static struct dm_block_validator dm_sm_bitmap_validator = {
110 .name = "sm_bitmap",
111 .prepare_for_write = dm_bitmap_prepare_for_write,
112 .check = dm_bitmap_check,
113};
114
115/*----------------------------------------------------------------*/
116
117#define ENTRIES_PER_WORD 32
118#define ENTRIES_SHIFT 5
119
120static void *dm_bitmap_data(struct dm_block *b)
121{
122 return dm_block_data(b) + sizeof(struct disk_bitmap_header);
123}
124
125#define WORD_MASK_HIGH 0xAAAAAAAAAAAAAAAAULL
126
127static unsigned dm_bitmap_word_used(void *addr, unsigned b)
128{
129 __le64 *words_le = addr;
130 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
131
132 uint64_t bits = le64_to_cpu(*w_le);
133 uint64_t mask = (bits + WORD_MASK_HIGH + 1) & WORD_MASK_HIGH;
134
135 return !(~bits & mask);
136}
137
138static unsigned sm_lookup_bitmap(void *addr, unsigned b)
139{
140 __le64 *words_le = addr;
141 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
142 unsigned hi, lo;
143
144 b = (b & (ENTRIES_PER_WORD - 1)) << 1;
145 hi = !!test_bit_le(b, (void *) w_le);
146 lo = !!test_bit_le(b + 1, (void *) w_le);
147 return (hi << 1) | lo;
148}
149
150static void sm_set_bitmap(void *addr, unsigned b, unsigned val)
151{
152 __le64 *words_le = addr;
153 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
154
155 b = (b & (ENTRIES_PER_WORD - 1)) << 1;
156
157 if (val & 2)
158 __set_bit_le(b, (void *) w_le);
159 else
160 __clear_bit_le(b, (void *) w_le);
161
162 if (val & 1)
163 __set_bit_le(b + 1, (void *) w_le);
164 else
165 __clear_bit_le(b + 1, (void *) w_le);
166}
167
168static int sm_find_free(void *addr, unsigned begin, unsigned end,
169 unsigned *result)
170{
171 while (begin < end) {
172 if (!(begin & (ENTRIES_PER_WORD - 1)) &&
173 dm_bitmap_word_used(addr, begin)) {
174 begin += ENTRIES_PER_WORD;
175 continue;
176 }
177
178 if (!sm_lookup_bitmap(addr, begin)) {
179 *result = begin;
180 return 0;
181 }
182
183 begin++;
184 }
185
186 return -ENOSPC;
187}
188
189/*----------------------------------------------------------------*/
190
191static int sm_ll_init(struct ll_disk *ll, struct dm_transaction_manager *tm)
192{
193 memset(ll, 0, sizeof(struct ll_disk));
194
195 ll->tm = tm;
196
197 ll->bitmap_info.tm = tm;
198 ll->bitmap_info.levels = 1;
199
200 /*
201 * Because the new bitmap blocks are created via a shadow
202 * operation, the old entry has already had its reference count
203 * decremented and we don't need the btree to do any bookkeeping.
204 */
205 ll->bitmap_info.value_type.size = sizeof(struct disk_index_entry);
206 ll->bitmap_info.value_type.inc = NULL;
207 ll->bitmap_info.value_type.dec = NULL;
208 ll->bitmap_info.value_type.equal = NULL;
209
210 ll->ref_count_info.tm = tm;
211 ll->ref_count_info.levels = 1;
212 ll->ref_count_info.value_type.size = sizeof(uint32_t);
213 ll->ref_count_info.value_type.inc = NULL;
214 ll->ref_count_info.value_type.dec = NULL;
215 ll->ref_count_info.value_type.equal = NULL;
216
217 ll->block_size = dm_bm_block_size(dm_tm_get_bm(tm));
218
219 if (ll->block_size > (1 << 30)) {
220 DMERR("block size too big to hold bitmaps");
221 return -EINVAL;
222 }
223
224 ll->entries_per_block = (ll->block_size - sizeof(struct disk_bitmap_header)) *
225 ENTRIES_PER_BYTE;
226 ll->nr_blocks = 0;
227 ll->bitmap_root = 0;
228 ll->ref_count_root = 0;
229 ll->bitmap_index_changed = false;
230
231 return 0;
232}
233
234int sm_ll_extend(struct ll_disk *ll, dm_block_t extra_blocks)
235{
236 int r;
237 dm_block_t i, nr_blocks, nr_indexes;
238 unsigned old_blocks, blocks;
239
240 nr_blocks = ll->nr_blocks + extra_blocks;
241 old_blocks = dm_sector_div_up(ll->nr_blocks, ll->entries_per_block);
242 blocks = dm_sector_div_up(nr_blocks, ll->entries_per_block);
243
244 nr_indexes = dm_sector_div_up(nr_blocks, ll->entries_per_block);
245 if (nr_indexes > ll->max_entries(ll)) {
246 DMERR("space map too large");
247 return -EINVAL;
248 }
249
250 /*
251 * We need to set this before the dm_tm_new_block() call below.
252 */
253 ll->nr_blocks = nr_blocks;
254 for (i = old_blocks; i < blocks; i++) {
255 struct dm_block *b;
256 struct disk_index_entry idx;
257
258 r = dm_tm_new_block(ll->tm, &dm_sm_bitmap_validator, &b);
259 if (r < 0)
260 return r;
261
262 idx.blocknr = cpu_to_le64(dm_block_location(b));
263
264 dm_tm_unlock(ll->tm, b);
265
266 idx.nr_free = cpu_to_le32(ll->entries_per_block);
267 idx.none_free_before = 0;
268
269 r = ll->save_ie(ll, i, &idx);
270 if (r < 0)
271 return r;
272 }
273
274 return 0;
275}
276
277int sm_ll_lookup_bitmap(struct ll_disk *ll, dm_block_t b, uint32_t *result)
278{
279 int r;
280 dm_block_t index = b;
281 struct disk_index_entry ie_disk;
282 struct dm_block *blk;
283
284 b = do_div(index, ll->entries_per_block);
285 r = ll->load_ie(ll, index, &ie_disk);
286 if (r < 0)
287 return r;
288
289 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr),
290 &dm_sm_bitmap_validator, &blk);
291 if (r < 0)
292 return r;
293
294 *result = sm_lookup_bitmap(dm_bitmap_data(blk), b);
295
296 dm_tm_unlock(ll->tm, blk);
297
298 return 0;
299}
300
301static int sm_ll_lookup_big_ref_count(struct ll_disk *ll, dm_block_t b,
302 uint32_t *result)
303{
304 __le32 le_rc;
305 int r;
306
307 r = dm_btree_lookup(&ll->ref_count_info, ll->ref_count_root, &b, &le_rc);
308 if (r < 0)
309 return r;
310
311 *result = le32_to_cpu(le_rc);
312
313 return r;
314}
315
316int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result)
317{
318 int r = sm_ll_lookup_bitmap(ll, b, result);
319
320 if (r)
321 return r;
322
323 if (*result != 3)
324 return r;
325
326 return sm_ll_lookup_big_ref_count(ll, b, result);
327}
328
329int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin,
330 dm_block_t end, dm_block_t *result)
331{
332 int r;
333 struct disk_index_entry ie_disk;
334 dm_block_t i, index_begin = begin;
335 dm_block_t index_end = dm_sector_div_up(end, ll->entries_per_block);
336
337 /*
338 * FIXME: Use shifts
339 */
340 begin = do_div(index_begin, ll->entries_per_block);
341 end = do_div(end, ll->entries_per_block);
342
343 for (i = index_begin; i < index_end; i++, begin = 0) {
344 struct dm_block *blk;
345 unsigned position;
346 uint32_t bit_end;
347
348 r = ll->load_ie(ll, i, &ie_disk);
349 if (r < 0)
350 return r;
351
352 if (le32_to_cpu(ie_disk.nr_free) == 0)
353 continue;
354
355 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr),
356 &dm_sm_bitmap_validator, &blk);
357 if (r < 0)
358 return r;
359
360 bit_end = (i == index_end - 1) ? end : ll->entries_per_block;
361
362 r = sm_find_free(dm_bitmap_data(blk),
363 max_t(unsigned, begin, le32_to_cpu(ie_disk.none_free_before)),
364 bit_end, &position);
365 if (r == -ENOSPC) {
366 /*
367 * This might happen because we started searching
368 * part way through the bitmap.
369 */
370 dm_tm_unlock(ll->tm, blk);
371 continue;
372 }
373
374 dm_tm_unlock(ll->tm, blk);
375
376 *result = i * ll->entries_per_block + (dm_block_t) position;
377 return 0;
378 }
379
380 return -ENOSPC;
381}
382
383static int sm_ll_mutate(struct ll_disk *ll, dm_block_t b,
384 int (*mutator)(void *context, uint32_t old, uint32_t *new),
385 void *context, enum allocation_event *ev)
386{
387 int r;
388 uint32_t bit, old, ref_count;
389 struct dm_block *nb;
390 dm_block_t index = b;
391 struct disk_index_entry ie_disk;
392 void *bm_le;
393 int inc;
394
395 bit = do_div(index, ll->entries_per_block);
396 r = ll->load_ie(ll, index, &ie_disk);
397 if (r < 0)
398 return r;
399
400 r = dm_tm_shadow_block(ll->tm, le64_to_cpu(ie_disk.blocknr),
401 &dm_sm_bitmap_validator, &nb, &inc);
402 if (r < 0) {
403 DMERR("dm_tm_shadow_block() failed");
404 return r;
405 }
406 ie_disk.blocknr = cpu_to_le64(dm_block_location(nb));
407
408 bm_le = dm_bitmap_data(nb);
409 old = sm_lookup_bitmap(bm_le, bit);
410
411 if (old > 2) {
412 r = sm_ll_lookup_big_ref_count(ll, b, &old);
413 if (r < 0) {
414 dm_tm_unlock(ll->tm, nb);
415 return r;
416 }
417 }
418
419 r = mutator(context, old, &ref_count);
420 if (r) {
421 dm_tm_unlock(ll->tm, nb);
422 return r;
423 }
424
425 if (ref_count <= 2) {
426 sm_set_bitmap(bm_le, bit, ref_count);
427
428 dm_tm_unlock(ll->tm, nb);
429
430 if (old > 2) {
431 r = dm_btree_remove(&ll->ref_count_info,
432 ll->ref_count_root,
433 &b, &ll->ref_count_root);
434 if (r)
435 return r;
436 }
437
438 } else {
439 __le32 le_rc = cpu_to_le32(ref_count);
440
441 sm_set_bitmap(bm_le, bit, 3);
442 dm_tm_unlock(ll->tm, nb);
443
444 __dm_bless_for_disk(&le_rc);
445 r = dm_btree_insert(&ll->ref_count_info, ll->ref_count_root,
446 &b, &le_rc, &ll->ref_count_root);
447 if (r < 0) {
448 DMERR("ref count insert failed");
449 return r;
450 }
451 }
452
453 if (ref_count && !old) {
454 *ev = SM_ALLOC;
455 ll->nr_allocated++;
456 le32_add_cpu(&ie_disk.nr_free, -1);
457 if (le32_to_cpu(ie_disk.none_free_before) == bit)
458 ie_disk.none_free_before = cpu_to_le32(bit + 1);
459
460 } else if (old && !ref_count) {
461 *ev = SM_FREE;
462 ll->nr_allocated--;
463 le32_add_cpu(&ie_disk.nr_free, 1);
464 ie_disk.none_free_before = cpu_to_le32(min(le32_to_cpu(ie_disk.none_free_before), bit));
465 } else
466 *ev = SM_NONE;
467
468 return ll->save_ie(ll, index, &ie_disk);
469}
470
471static int set_ref_count(void *context, uint32_t old, uint32_t *new)
472{
473 *new = *((uint32_t *) context);
474 return 0;
475}
476
477int sm_ll_insert(struct ll_disk *ll, dm_block_t b,
478 uint32_t ref_count, enum allocation_event *ev)
479{
480 return sm_ll_mutate(ll, b, set_ref_count, &ref_count, ev);
481}
482
483static int inc_ref_count(void *context, uint32_t old, uint32_t *new)
484{
485 *new = old + 1;
486 return 0;
487}
488
489int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
490{
491 return sm_ll_mutate(ll, b, inc_ref_count, NULL, ev);
492}
493
494static int dec_ref_count(void *context, uint32_t old, uint32_t *new)
495{
496 if (!old) {
497 DMERR_LIMIT("unable to decrement a reference count below 0");
498 return -EINVAL;
499 }
500
501 *new = old - 1;
502 return 0;
503}
504
505int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
506{
507 return sm_ll_mutate(ll, b, dec_ref_count, NULL, ev);
508}
509
510int sm_ll_commit(struct ll_disk *ll)
511{
512 int r = 0;
513
514 if (ll->bitmap_index_changed) {
515 r = ll->commit(ll);
516 if (!r)
517 ll->bitmap_index_changed = false;
518 }
519
520 return r;
521}
522
523/*----------------------------------------------------------------*/
524
525static int metadata_ll_load_ie(struct ll_disk *ll, dm_block_t index,
526 struct disk_index_entry *ie)
527{
528 memcpy(ie, ll->mi_le.index + index, sizeof(*ie));
529 return 0;
530}
531
532static int metadata_ll_save_ie(struct ll_disk *ll, dm_block_t index,
533 struct disk_index_entry *ie)
534{
535 ll->bitmap_index_changed = true;
536 memcpy(ll->mi_le.index + index, ie, sizeof(*ie));
537 return 0;
538}
539
540static int metadata_ll_init_index(struct ll_disk *ll)
541{
542 int r;
543 struct dm_block *b;
544
545 r = dm_tm_new_block(ll->tm, &index_validator, &b);
546 if (r < 0)
547 return r;
548
549 ll->bitmap_root = dm_block_location(b);
550
551 dm_tm_unlock(ll->tm, b);
552
553 return 0;
554}
555
556static int metadata_ll_open(struct ll_disk *ll)
557{
558 int r;
559 struct dm_block *block;
560
561 r = dm_tm_read_lock(ll->tm, ll->bitmap_root,
562 &index_validator, &block);
563 if (r)
564 return r;
565
566 memcpy(&ll->mi_le, dm_block_data(block), sizeof(ll->mi_le));
567 dm_tm_unlock(ll->tm, block);
568
569 return 0;
570}
571
572static dm_block_t metadata_ll_max_entries(struct ll_disk *ll)
573{
574 return MAX_METADATA_BITMAPS;
575}
576
577static int metadata_ll_commit(struct ll_disk *ll)
578{
579 int r, inc;
580 struct dm_block *b;
581
582 r = dm_tm_shadow_block(ll->tm, ll->bitmap_root, &index_validator, &b, &inc);
583 if (r)
584 return r;
585
586 memcpy(dm_block_data(b), &ll->mi_le, sizeof(ll->mi_le));
587 ll->bitmap_root = dm_block_location(b);
588
589 dm_tm_unlock(ll->tm, b);
590
591 return 0;
592}
593
594int sm_ll_new_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm)
595{
596 int r;
597
598 r = sm_ll_init(ll, tm);
599 if (r < 0)
600 return r;
601
602 ll->load_ie = metadata_ll_load_ie;
603 ll->save_ie = metadata_ll_save_ie;
604 ll->init_index = metadata_ll_init_index;
605 ll->open_index = metadata_ll_open;
606 ll->max_entries = metadata_ll_max_entries;
607 ll->commit = metadata_ll_commit;
608
609 ll->nr_blocks = 0;
610 ll->nr_allocated = 0;
611
612 r = ll->init_index(ll);
613 if (r < 0)
614 return r;
615
616 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root);
617 if (r < 0)
618 return r;
619
620 return 0;
621}
622
623int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm,
624 void *root_le, size_t len)
625{
626 int r;
627 struct disk_sm_root smr;
628
629 if (len < sizeof(struct disk_sm_root)) {
630 DMERR("sm_metadata root too small");
631 return -ENOMEM;
632 }
633
634 /*
635 * We don't know the alignment of the root_le buffer, so need to
636 * copy into a new structure.
637 */
638 memcpy(&smr, root_le, sizeof(smr));
639
640 r = sm_ll_init(ll, tm);
641 if (r < 0)
642 return r;
643
644 ll->load_ie = metadata_ll_load_ie;
645 ll->save_ie = metadata_ll_save_ie;
646 ll->init_index = metadata_ll_init_index;
647 ll->open_index = metadata_ll_open;
648 ll->max_entries = metadata_ll_max_entries;
649 ll->commit = metadata_ll_commit;
650
651 ll->nr_blocks = le64_to_cpu(smr.nr_blocks);
652 ll->nr_allocated = le64_to_cpu(smr.nr_allocated);
653 ll->bitmap_root = le64_to_cpu(smr.bitmap_root);
654 ll->ref_count_root = le64_to_cpu(smr.ref_count_root);
655
656 return ll->open_index(ll);
657}
658
659/*----------------------------------------------------------------*/
660
661static int disk_ll_load_ie(struct ll_disk *ll, dm_block_t index,
662 struct disk_index_entry *ie)
663{
664 return dm_btree_lookup(&ll->bitmap_info, ll->bitmap_root, &index, ie);
665}
666
667static int disk_ll_save_ie(struct ll_disk *ll, dm_block_t index,
668 struct disk_index_entry *ie)
669{
670 __dm_bless_for_disk(ie);
671 return dm_btree_insert(&ll->bitmap_info, ll->bitmap_root,
672 &index, ie, &ll->bitmap_root);
673}
674
675static int disk_ll_init_index(struct ll_disk *ll)
676{
677 return dm_btree_empty(&ll->bitmap_info, &ll->bitmap_root);
678}
679
680static int disk_ll_open(struct ll_disk *ll)
681{
682 /* nothing to do */
683 return 0;
684}
685
686static dm_block_t disk_ll_max_entries(struct ll_disk *ll)
687{
688 return -1ULL;
689}
690
691static int disk_ll_commit(struct ll_disk *ll)
692{
693 return 0;
694}
695
696int sm_ll_new_disk(struct ll_disk *ll, struct dm_transaction_manager *tm)
697{
698 int r;
699
700 r = sm_ll_init(ll, tm);
701 if (r < 0)
702 return r;
703
704 ll->load_ie = disk_ll_load_ie;
705 ll->save_ie = disk_ll_save_ie;
706 ll->init_index = disk_ll_init_index;
707 ll->open_index = disk_ll_open;
708 ll->max_entries = disk_ll_max_entries;
709 ll->commit = disk_ll_commit;
710
711 ll->nr_blocks = 0;
712 ll->nr_allocated = 0;
713
714 r = ll->init_index(ll);
715 if (r < 0)
716 return r;
717
718 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root);
719 if (r < 0)
720 return r;
721
722 return 0;
723}
724
725int sm_ll_open_disk(struct ll_disk *ll, struct dm_transaction_manager *tm,
726 void *root_le, size_t len)
727{
728 int r;
729 struct disk_sm_root *smr = root_le;
730
731 if (len < sizeof(struct disk_sm_root)) {
732 DMERR("sm_metadata root too small");
733 return -ENOMEM;
734 }
735
736 r = sm_ll_init(ll, tm);
737 if (r < 0)
738 return r;
739
740 ll->load_ie = disk_ll_load_ie;
741 ll->save_ie = disk_ll_save_ie;
742 ll->init_index = disk_ll_init_index;
743 ll->open_index = disk_ll_open;
744 ll->max_entries = disk_ll_max_entries;
745 ll->commit = disk_ll_commit;
746
747 ll->nr_blocks = le64_to_cpu(smr->nr_blocks);
748 ll->nr_allocated = le64_to_cpu(smr->nr_allocated);
749 ll->bitmap_root = le64_to_cpu(smr->bitmap_root);
750 ll->ref_count_root = le64_to_cpu(smr->ref_count_root);
751
752 return ll->open_index(ll);
753}
754
755/*----------------------------------------------------------------*/
1/*
2 * Copyright (C) 2011 Red Hat, Inc.
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-space-map-common.h"
8#include "dm-transaction-manager.h"
9
10#include <linux/bitops.h>
11#include <linux/device-mapper.h>
12
13#define DM_MSG_PREFIX "space map common"
14
15/*----------------------------------------------------------------*/
16
17/*
18 * Index validator.
19 */
20#define INDEX_CSUM_XOR 160478
21
22static void index_prepare_for_write(struct dm_block_validator *v,
23 struct dm_block *b,
24 size_t block_size)
25{
26 struct disk_metadata_index *mi_le = dm_block_data(b);
27
28 mi_le->blocknr = cpu_to_le64(dm_block_location(b));
29 mi_le->csum = cpu_to_le32(dm_bm_checksum(&mi_le->padding,
30 block_size - sizeof(__le32),
31 INDEX_CSUM_XOR));
32}
33
34static int index_check(struct dm_block_validator *v,
35 struct dm_block *b,
36 size_t block_size)
37{
38 struct disk_metadata_index *mi_le = dm_block_data(b);
39 __le32 csum_disk;
40
41 if (dm_block_location(b) != le64_to_cpu(mi_le->blocknr)) {
42 DMERR_LIMIT("index_check failed: blocknr %llu != wanted %llu",
43 le64_to_cpu(mi_le->blocknr), dm_block_location(b));
44 return -ENOTBLK;
45 }
46
47 csum_disk = cpu_to_le32(dm_bm_checksum(&mi_le->padding,
48 block_size - sizeof(__le32),
49 INDEX_CSUM_XOR));
50 if (csum_disk != mi_le->csum) {
51 DMERR_LIMIT("index_check failed: csum %u != wanted %u",
52 le32_to_cpu(csum_disk), le32_to_cpu(mi_le->csum));
53 return -EILSEQ;
54 }
55
56 return 0;
57}
58
59static struct dm_block_validator index_validator = {
60 .name = "index",
61 .prepare_for_write = index_prepare_for_write,
62 .check = index_check
63};
64
65/*----------------------------------------------------------------*/
66
67/*
68 * Bitmap validator
69 */
70#define BITMAP_CSUM_XOR 240779
71
72static void bitmap_prepare_for_write(struct dm_block_validator *v,
73 struct dm_block *b,
74 size_t block_size)
75{
76 struct disk_bitmap_header *disk_header = dm_block_data(b);
77
78 disk_header->blocknr = cpu_to_le64(dm_block_location(b));
79 disk_header->csum = cpu_to_le32(dm_bm_checksum(&disk_header->not_used,
80 block_size - sizeof(__le32),
81 BITMAP_CSUM_XOR));
82}
83
84static int bitmap_check(struct dm_block_validator *v,
85 struct dm_block *b,
86 size_t block_size)
87{
88 struct disk_bitmap_header *disk_header = dm_block_data(b);
89 __le32 csum_disk;
90
91 if (dm_block_location(b) != le64_to_cpu(disk_header->blocknr)) {
92 DMERR_LIMIT("bitmap check failed: blocknr %llu != wanted %llu",
93 le64_to_cpu(disk_header->blocknr), dm_block_location(b));
94 return -ENOTBLK;
95 }
96
97 csum_disk = cpu_to_le32(dm_bm_checksum(&disk_header->not_used,
98 block_size - sizeof(__le32),
99 BITMAP_CSUM_XOR));
100 if (csum_disk != disk_header->csum) {
101 DMERR_LIMIT("bitmap check failed: csum %u != wanted %u",
102 le32_to_cpu(csum_disk), le32_to_cpu(disk_header->csum));
103 return -EILSEQ;
104 }
105
106 return 0;
107}
108
109static struct dm_block_validator dm_sm_bitmap_validator = {
110 .name = "sm_bitmap",
111 .prepare_for_write = bitmap_prepare_for_write,
112 .check = bitmap_check
113};
114
115/*----------------------------------------------------------------*/
116
117#define ENTRIES_PER_WORD 32
118#define ENTRIES_SHIFT 5
119
120static void *dm_bitmap_data(struct dm_block *b)
121{
122 return dm_block_data(b) + sizeof(struct disk_bitmap_header);
123}
124
125#define WORD_MASK_HIGH 0xAAAAAAAAAAAAAAAAULL
126
127static unsigned bitmap_word_used(void *addr, unsigned b)
128{
129 __le64 *words_le = addr;
130 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
131
132 uint64_t bits = le64_to_cpu(*w_le);
133 uint64_t mask = (bits + WORD_MASK_HIGH + 1) & WORD_MASK_HIGH;
134
135 return !(~bits & mask);
136}
137
138static unsigned sm_lookup_bitmap(void *addr, unsigned b)
139{
140 __le64 *words_le = addr;
141 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
142 unsigned hi, lo;
143
144 b = (b & (ENTRIES_PER_WORD - 1)) << 1;
145 hi = !!test_bit_le(b, (void *) w_le);
146 lo = !!test_bit_le(b + 1, (void *) w_le);
147 return (hi << 1) | lo;
148}
149
150static void sm_set_bitmap(void *addr, unsigned b, unsigned val)
151{
152 __le64 *words_le = addr;
153 __le64 *w_le = words_le + (b >> ENTRIES_SHIFT);
154
155 b = (b & (ENTRIES_PER_WORD - 1)) << 1;
156
157 if (val & 2)
158 __set_bit_le(b, (void *) w_le);
159 else
160 __clear_bit_le(b, (void *) w_le);
161
162 if (val & 1)
163 __set_bit_le(b + 1, (void *) w_le);
164 else
165 __clear_bit_le(b + 1, (void *) w_le);
166}
167
168static int sm_find_free(void *addr, unsigned begin, unsigned end,
169 unsigned *result)
170{
171 while (begin < end) {
172 if (!(begin & (ENTRIES_PER_WORD - 1)) &&
173 bitmap_word_used(addr, begin)) {
174 begin += ENTRIES_PER_WORD;
175 continue;
176 }
177
178 if (!sm_lookup_bitmap(addr, begin)) {
179 *result = begin;
180 return 0;
181 }
182
183 begin++;
184 }
185
186 return -ENOSPC;
187}
188
189/*----------------------------------------------------------------*/
190
191static int sm_ll_init(struct ll_disk *ll, struct dm_transaction_manager *tm)
192{
193 ll->tm = tm;
194
195 ll->bitmap_info.tm = tm;
196 ll->bitmap_info.levels = 1;
197
198 /*
199 * Because the new bitmap blocks are created via a shadow
200 * operation, the old entry has already had its reference count
201 * decremented and we don't need the btree to do any bookkeeping.
202 */
203 ll->bitmap_info.value_type.size = sizeof(struct disk_index_entry);
204 ll->bitmap_info.value_type.inc = NULL;
205 ll->bitmap_info.value_type.dec = NULL;
206 ll->bitmap_info.value_type.equal = NULL;
207
208 ll->ref_count_info.tm = tm;
209 ll->ref_count_info.levels = 1;
210 ll->ref_count_info.value_type.size = sizeof(uint32_t);
211 ll->ref_count_info.value_type.inc = NULL;
212 ll->ref_count_info.value_type.dec = NULL;
213 ll->ref_count_info.value_type.equal = NULL;
214
215 ll->block_size = dm_bm_block_size(dm_tm_get_bm(tm));
216
217 if (ll->block_size > (1 << 30)) {
218 DMERR("block size too big to hold bitmaps");
219 return -EINVAL;
220 }
221
222 ll->entries_per_block = (ll->block_size - sizeof(struct disk_bitmap_header)) *
223 ENTRIES_PER_BYTE;
224 ll->nr_blocks = 0;
225 ll->bitmap_root = 0;
226 ll->ref_count_root = 0;
227 ll->bitmap_index_changed = false;
228
229 return 0;
230}
231
232int sm_ll_extend(struct ll_disk *ll, dm_block_t extra_blocks)
233{
234 int r;
235 dm_block_t i, nr_blocks, nr_indexes;
236 unsigned old_blocks, blocks;
237
238 nr_blocks = ll->nr_blocks + extra_blocks;
239 old_blocks = dm_sector_div_up(ll->nr_blocks, ll->entries_per_block);
240 blocks = dm_sector_div_up(nr_blocks, ll->entries_per_block);
241
242 nr_indexes = dm_sector_div_up(nr_blocks, ll->entries_per_block);
243 if (nr_indexes > ll->max_entries(ll)) {
244 DMERR("space map too large");
245 return -EINVAL;
246 }
247
248 /*
249 * We need to set this before the dm_tm_new_block() call below.
250 */
251 ll->nr_blocks = nr_blocks;
252 for (i = old_blocks; i < blocks; i++) {
253 struct dm_block *b;
254 struct disk_index_entry idx;
255
256 r = dm_tm_new_block(ll->tm, &dm_sm_bitmap_validator, &b);
257 if (r < 0)
258 return r;
259
260 idx.blocknr = cpu_to_le64(dm_block_location(b));
261
262 dm_tm_unlock(ll->tm, b);
263
264 idx.nr_free = cpu_to_le32(ll->entries_per_block);
265 idx.none_free_before = 0;
266
267 r = ll->save_ie(ll, i, &idx);
268 if (r < 0)
269 return r;
270 }
271
272 return 0;
273}
274
275int sm_ll_lookup_bitmap(struct ll_disk *ll, dm_block_t b, uint32_t *result)
276{
277 int r;
278 dm_block_t index = b;
279 struct disk_index_entry ie_disk;
280 struct dm_block *blk;
281
282 b = do_div(index, ll->entries_per_block);
283 r = ll->load_ie(ll, index, &ie_disk);
284 if (r < 0)
285 return r;
286
287 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr),
288 &dm_sm_bitmap_validator, &blk);
289 if (r < 0)
290 return r;
291
292 *result = sm_lookup_bitmap(dm_bitmap_data(blk), b);
293
294 dm_tm_unlock(ll->tm, blk);
295
296 return 0;
297}
298
299static int sm_ll_lookup_big_ref_count(struct ll_disk *ll, dm_block_t b,
300 uint32_t *result)
301{
302 __le32 le_rc;
303 int r;
304
305 r = dm_btree_lookup(&ll->ref_count_info, ll->ref_count_root, &b, &le_rc);
306 if (r < 0)
307 return r;
308
309 *result = le32_to_cpu(le_rc);
310
311 return r;
312}
313
314int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result)
315{
316 int r = sm_ll_lookup_bitmap(ll, b, result);
317
318 if (r)
319 return r;
320
321 if (*result != 3)
322 return r;
323
324 return sm_ll_lookup_big_ref_count(ll, b, result);
325}
326
327int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin,
328 dm_block_t end, dm_block_t *result)
329{
330 int r;
331 struct disk_index_entry ie_disk;
332 dm_block_t i, index_begin = begin;
333 dm_block_t index_end = dm_sector_div_up(end, ll->entries_per_block);
334
335 /*
336 * FIXME: Use shifts
337 */
338 begin = do_div(index_begin, ll->entries_per_block);
339 end = do_div(end, ll->entries_per_block);
340
341 for (i = index_begin; i < index_end; i++, begin = 0) {
342 struct dm_block *blk;
343 unsigned position;
344 uint32_t bit_end;
345
346 r = ll->load_ie(ll, i, &ie_disk);
347 if (r < 0)
348 return r;
349
350 if (le32_to_cpu(ie_disk.nr_free) == 0)
351 continue;
352
353 r = dm_tm_read_lock(ll->tm, le64_to_cpu(ie_disk.blocknr),
354 &dm_sm_bitmap_validator, &blk);
355 if (r < 0)
356 return r;
357
358 bit_end = (i == index_end - 1) ? end : ll->entries_per_block;
359
360 r = sm_find_free(dm_bitmap_data(blk),
361 max_t(unsigned, begin, le32_to_cpu(ie_disk.none_free_before)),
362 bit_end, &position);
363 if (r == -ENOSPC) {
364 /*
365 * This might happen because we started searching
366 * part way through the bitmap.
367 */
368 dm_tm_unlock(ll->tm, blk);
369 continue;
370
371 } else if (r < 0) {
372 dm_tm_unlock(ll->tm, blk);
373 return r;
374 }
375
376 dm_tm_unlock(ll->tm, blk);
377
378 *result = i * ll->entries_per_block + (dm_block_t) position;
379 return 0;
380 }
381
382 return -ENOSPC;
383}
384
385static int sm_ll_mutate(struct ll_disk *ll, dm_block_t b,
386 int (*mutator)(void *context, uint32_t old, uint32_t *new),
387 void *context, enum allocation_event *ev)
388{
389 int r;
390 uint32_t bit, old, ref_count;
391 struct dm_block *nb;
392 dm_block_t index = b;
393 struct disk_index_entry ie_disk;
394 void *bm_le;
395 int inc;
396
397 bit = do_div(index, ll->entries_per_block);
398 r = ll->load_ie(ll, index, &ie_disk);
399 if (r < 0)
400 return r;
401
402 r = dm_tm_shadow_block(ll->tm, le64_to_cpu(ie_disk.blocknr),
403 &dm_sm_bitmap_validator, &nb, &inc);
404 if (r < 0) {
405 DMERR("dm_tm_shadow_block() failed");
406 return r;
407 }
408 ie_disk.blocknr = cpu_to_le64(dm_block_location(nb));
409
410 bm_le = dm_bitmap_data(nb);
411 old = sm_lookup_bitmap(bm_le, bit);
412
413 if (old > 2) {
414 r = sm_ll_lookup_big_ref_count(ll, b, &old);
415 if (r < 0) {
416 dm_tm_unlock(ll->tm, nb);
417 return r;
418 }
419 }
420
421 r = mutator(context, old, &ref_count);
422 if (r) {
423 dm_tm_unlock(ll->tm, nb);
424 return r;
425 }
426
427 if (ref_count <= 2) {
428 sm_set_bitmap(bm_le, bit, ref_count);
429
430 dm_tm_unlock(ll->tm, nb);
431
432 if (old > 2) {
433 r = dm_btree_remove(&ll->ref_count_info,
434 ll->ref_count_root,
435 &b, &ll->ref_count_root);
436 if (r)
437 return r;
438 }
439
440 } else {
441 __le32 le_rc = cpu_to_le32(ref_count);
442
443 sm_set_bitmap(bm_le, bit, 3);
444 dm_tm_unlock(ll->tm, nb);
445
446 __dm_bless_for_disk(&le_rc);
447 r = dm_btree_insert(&ll->ref_count_info, ll->ref_count_root,
448 &b, &le_rc, &ll->ref_count_root);
449 if (r < 0) {
450 DMERR("ref count insert failed");
451 return r;
452 }
453 }
454
455 if (ref_count && !old) {
456 *ev = SM_ALLOC;
457 ll->nr_allocated++;
458 le32_add_cpu(&ie_disk.nr_free, -1);
459 if (le32_to_cpu(ie_disk.none_free_before) == bit)
460 ie_disk.none_free_before = cpu_to_le32(bit + 1);
461
462 } else if (old && !ref_count) {
463 *ev = SM_FREE;
464 ll->nr_allocated--;
465 le32_add_cpu(&ie_disk.nr_free, 1);
466 ie_disk.none_free_before = cpu_to_le32(min(le32_to_cpu(ie_disk.none_free_before), bit));
467 }
468
469 return ll->save_ie(ll, index, &ie_disk);
470}
471
472static int set_ref_count(void *context, uint32_t old, uint32_t *new)
473{
474 *new = *((uint32_t *) context);
475 return 0;
476}
477
478int sm_ll_insert(struct ll_disk *ll, dm_block_t b,
479 uint32_t ref_count, enum allocation_event *ev)
480{
481 return sm_ll_mutate(ll, b, set_ref_count, &ref_count, ev);
482}
483
484static int inc_ref_count(void *context, uint32_t old, uint32_t *new)
485{
486 *new = old + 1;
487 return 0;
488}
489
490int sm_ll_inc(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
491{
492 return sm_ll_mutate(ll, b, inc_ref_count, NULL, ev);
493}
494
495static int dec_ref_count(void *context, uint32_t old, uint32_t *new)
496{
497 if (!old) {
498 DMERR_LIMIT("unable to decrement a reference count below 0");
499 return -EINVAL;
500 }
501
502 *new = old - 1;
503 return 0;
504}
505
506int sm_ll_dec(struct ll_disk *ll, dm_block_t b, enum allocation_event *ev)
507{
508 return sm_ll_mutate(ll, b, dec_ref_count, NULL, ev);
509}
510
511int sm_ll_commit(struct ll_disk *ll)
512{
513 int r = 0;
514
515 if (ll->bitmap_index_changed) {
516 r = ll->commit(ll);
517 if (!r)
518 ll->bitmap_index_changed = false;
519 }
520
521 return r;
522}
523
524/*----------------------------------------------------------------*/
525
526static int metadata_ll_load_ie(struct ll_disk *ll, dm_block_t index,
527 struct disk_index_entry *ie)
528{
529 memcpy(ie, ll->mi_le.index + index, sizeof(*ie));
530 return 0;
531}
532
533static int metadata_ll_save_ie(struct ll_disk *ll, dm_block_t index,
534 struct disk_index_entry *ie)
535{
536 ll->bitmap_index_changed = true;
537 memcpy(ll->mi_le.index + index, ie, sizeof(*ie));
538 return 0;
539}
540
541static int metadata_ll_init_index(struct ll_disk *ll)
542{
543 int r;
544 struct dm_block *b;
545
546 r = dm_tm_new_block(ll->tm, &index_validator, &b);
547 if (r < 0)
548 return r;
549
550 memcpy(dm_block_data(b), &ll->mi_le, sizeof(ll->mi_le));
551 ll->bitmap_root = dm_block_location(b);
552
553 dm_tm_unlock(ll->tm, b);
554
555 return 0;
556}
557
558static int metadata_ll_open(struct ll_disk *ll)
559{
560 int r;
561 struct dm_block *block;
562
563 r = dm_tm_read_lock(ll->tm, ll->bitmap_root,
564 &index_validator, &block);
565 if (r)
566 return r;
567
568 memcpy(&ll->mi_le, dm_block_data(block), sizeof(ll->mi_le));
569 dm_tm_unlock(ll->tm, block);
570
571 return 0;
572}
573
574static dm_block_t metadata_ll_max_entries(struct ll_disk *ll)
575{
576 return MAX_METADATA_BITMAPS;
577}
578
579static int metadata_ll_commit(struct ll_disk *ll)
580{
581 int r, inc;
582 struct dm_block *b;
583
584 r = dm_tm_shadow_block(ll->tm, ll->bitmap_root, &index_validator, &b, &inc);
585 if (r)
586 return r;
587
588 memcpy(dm_block_data(b), &ll->mi_le, sizeof(ll->mi_le));
589 ll->bitmap_root = dm_block_location(b);
590
591 dm_tm_unlock(ll->tm, b);
592
593 return 0;
594}
595
596int sm_ll_new_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm)
597{
598 int r;
599
600 r = sm_ll_init(ll, tm);
601 if (r < 0)
602 return r;
603
604 ll->load_ie = metadata_ll_load_ie;
605 ll->save_ie = metadata_ll_save_ie;
606 ll->init_index = metadata_ll_init_index;
607 ll->open_index = metadata_ll_open;
608 ll->max_entries = metadata_ll_max_entries;
609 ll->commit = metadata_ll_commit;
610
611 ll->nr_blocks = 0;
612 ll->nr_allocated = 0;
613
614 r = ll->init_index(ll);
615 if (r < 0)
616 return r;
617
618 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root);
619 if (r < 0)
620 return r;
621
622 return 0;
623}
624
625int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm,
626 void *root_le, size_t len)
627{
628 int r;
629 struct disk_sm_root *smr = root_le;
630
631 if (len < sizeof(struct disk_sm_root)) {
632 DMERR("sm_metadata root too small");
633 return -ENOMEM;
634 }
635
636 r = sm_ll_init(ll, tm);
637 if (r < 0)
638 return r;
639
640 ll->load_ie = metadata_ll_load_ie;
641 ll->save_ie = metadata_ll_save_ie;
642 ll->init_index = metadata_ll_init_index;
643 ll->open_index = metadata_ll_open;
644 ll->max_entries = metadata_ll_max_entries;
645 ll->commit = metadata_ll_commit;
646
647 ll->nr_blocks = le64_to_cpu(smr->nr_blocks);
648 ll->nr_allocated = le64_to_cpu(smr->nr_allocated);
649 ll->bitmap_root = le64_to_cpu(smr->bitmap_root);
650 ll->ref_count_root = le64_to_cpu(smr->ref_count_root);
651
652 return ll->open_index(ll);
653}
654
655/*----------------------------------------------------------------*/
656
657static int disk_ll_load_ie(struct ll_disk *ll, dm_block_t index,
658 struct disk_index_entry *ie)
659{
660 return dm_btree_lookup(&ll->bitmap_info, ll->bitmap_root, &index, ie);
661}
662
663static int disk_ll_save_ie(struct ll_disk *ll, dm_block_t index,
664 struct disk_index_entry *ie)
665{
666 __dm_bless_for_disk(ie);
667 return dm_btree_insert(&ll->bitmap_info, ll->bitmap_root,
668 &index, ie, &ll->bitmap_root);
669}
670
671static int disk_ll_init_index(struct ll_disk *ll)
672{
673 return dm_btree_empty(&ll->bitmap_info, &ll->bitmap_root);
674}
675
676static int disk_ll_open(struct ll_disk *ll)
677{
678 /* nothing to do */
679 return 0;
680}
681
682static dm_block_t disk_ll_max_entries(struct ll_disk *ll)
683{
684 return -1ULL;
685}
686
687static int disk_ll_commit(struct ll_disk *ll)
688{
689 return 0;
690}
691
692int sm_ll_new_disk(struct ll_disk *ll, struct dm_transaction_manager *tm)
693{
694 int r;
695
696 r = sm_ll_init(ll, tm);
697 if (r < 0)
698 return r;
699
700 ll->load_ie = disk_ll_load_ie;
701 ll->save_ie = disk_ll_save_ie;
702 ll->init_index = disk_ll_init_index;
703 ll->open_index = disk_ll_open;
704 ll->max_entries = disk_ll_max_entries;
705 ll->commit = disk_ll_commit;
706
707 ll->nr_blocks = 0;
708 ll->nr_allocated = 0;
709
710 r = ll->init_index(ll);
711 if (r < 0)
712 return r;
713
714 r = dm_btree_empty(&ll->ref_count_info, &ll->ref_count_root);
715 if (r < 0)
716 return r;
717
718 return 0;
719}
720
721int sm_ll_open_disk(struct ll_disk *ll, struct dm_transaction_manager *tm,
722 void *root_le, size_t len)
723{
724 int r;
725 struct disk_sm_root *smr = root_le;
726
727 if (len < sizeof(struct disk_sm_root)) {
728 DMERR("sm_metadata root too small");
729 return -ENOMEM;
730 }
731
732 r = sm_ll_init(ll, tm);
733 if (r < 0)
734 return r;
735
736 ll->load_ie = disk_ll_load_ie;
737 ll->save_ie = disk_ll_save_ie;
738 ll->init_index = disk_ll_init_index;
739 ll->open_index = disk_ll_open;
740 ll->max_entries = disk_ll_max_entries;
741 ll->commit = disk_ll_commit;
742
743 ll->nr_blocks = le64_to_cpu(smr->nr_blocks);
744 ll->nr_allocated = le64_to_cpu(smr->nr_allocated);
745 ll->bitmap_root = le64_to_cpu(smr->bitmap_root);
746 ll->ref_count_root = le64_to_cpu(smr->ref_count_root);
747
748 return ll->open_index(ll);
749}
750
751/*----------------------------------------------------------------*/