Merge tag 'firewire-updates' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee139...
[firefly-linux-kernel-4.4.55.git] / drivers / md / dm-cache-metadata.c
1 /*
2  * Copyright (C) 2012 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6
7 #include "dm-cache-metadata.h"
8
9 #include "persistent-data/dm-array.h"
10 #include "persistent-data/dm-bitset.h"
11 #include "persistent-data/dm-space-map.h"
12 #include "persistent-data/dm-space-map-disk.h"
13 #include "persistent-data/dm-transaction-manager.h"
14
15 #include <linux/device-mapper.h>
16
17 /*----------------------------------------------------------------*/
18
19 #define DM_MSG_PREFIX   "cache metadata"
20
21 #define CACHE_SUPERBLOCK_MAGIC 06142003
22 #define CACHE_SUPERBLOCK_LOCATION 0
23
24 /*
25  * defines a range of metadata versions that this module can handle.
26  */
27 #define MIN_CACHE_VERSION 1
28 #define MAX_CACHE_VERSION 1
29
30 #define CACHE_METADATA_CACHE_SIZE 64
31
32 /*
33  *  3 for btree insert +
34  *  2 for btree lookup used within space map
35  */
36 #define CACHE_MAX_CONCURRENT_LOCKS 5
37 #define SPACE_MAP_ROOT_SIZE 128
38
39 enum superblock_flag_bits {
40         /* for spotting crashes that would invalidate the dirty bitset */
41         CLEAN_SHUTDOWN,
42 };
43
44 /*
45  * Each mapping from cache block -> origin block carries a set of flags.
46  */
47 enum mapping_bits {
48         /*
49          * A valid mapping.  Because we're using an array we clear this
50          * flag for an non existant mapping.
51          */
52         M_VALID = 1,
53
54         /*
55          * The data on the cache is different from that on the origin.
56          */
57         M_DIRTY = 2
58 };
59
60 struct cache_disk_superblock {
61         __le32 csum;
62         __le32 flags;
63         __le64 blocknr;
64
65         __u8 uuid[16];
66         __le64 magic;
67         __le32 version;
68
69         __u8 policy_name[CACHE_POLICY_NAME_SIZE];
70         __le32 policy_hint_size;
71
72         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
73         __le64 mapping_root;
74         __le64 hint_root;
75
76         __le64 discard_root;
77         __le64 discard_block_size;
78         __le64 discard_nr_blocks;
79
80         __le32 data_block_size;
81         __le32 metadata_block_size;
82         __le32 cache_blocks;
83
84         __le32 compat_flags;
85         __le32 compat_ro_flags;
86         __le32 incompat_flags;
87
88         __le32 read_hits;
89         __le32 read_misses;
90         __le32 write_hits;
91         __le32 write_misses;
92
93         __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
94 } __packed;
95
96 struct dm_cache_metadata {
97         atomic_t ref_count;
98         struct list_head list;
99
100         struct block_device *bdev;
101         struct dm_block_manager *bm;
102         struct dm_space_map *metadata_sm;
103         struct dm_transaction_manager *tm;
104
105         struct dm_array_info info;
106         struct dm_array_info hint_info;
107         struct dm_disk_bitset discard_info;
108
109         struct rw_semaphore root_lock;
110         dm_block_t root;
111         dm_block_t hint_root;
112         dm_block_t discard_root;
113
114         sector_t discard_block_size;
115         dm_dblock_t discard_nr_blocks;
116
117         sector_t data_block_size;
118         dm_cblock_t cache_blocks;
119         bool changed:1;
120         bool clean_when_opened:1;
121
122         char policy_name[CACHE_POLICY_NAME_SIZE];
123         unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
124         size_t policy_hint_size;
125         struct dm_cache_statistics stats;
126
127         /*
128          * Reading the space map root can fail, so we read it into this
129          * buffer before the superblock is locked and updated.
130          */
131         __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
132 };
133
134 /*-------------------------------------------------------------------
135  * superblock validator
136  *-----------------------------------------------------------------*/
137
138 #define SUPERBLOCK_CSUM_XOR 9031977
139
140 static void sb_prepare_for_write(struct dm_block_validator *v,
141                                  struct dm_block *b,
142                                  size_t sb_block_size)
143 {
144         struct cache_disk_superblock *disk_super = dm_block_data(b);
145
146         disk_super->blocknr = cpu_to_le64(dm_block_location(b));
147         disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
148                                                       sb_block_size - sizeof(__le32),
149                                                       SUPERBLOCK_CSUM_XOR));
150 }
151
152 static int check_metadata_version(struct cache_disk_superblock *disk_super)
153 {
154         uint32_t metadata_version = le32_to_cpu(disk_super->version);
155         if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
156                 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
157                       metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
158                 return -EINVAL;
159         }
160
161         return 0;
162 }
163
164 static int sb_check(struct dm_block_validator *v,
165                     struct dm_block *b,
166                     size_t sb_block_size)
167 {
168         struct cache_disk_superblock *disk_super = dm_block_data(b);
169         __le32 csum_le;
170
171         if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
172                 DMERR("sb_check failed: blocknr %llu: wanted %llu",
173                       le64_to_cpu(disk_super->blocknr),
174                       (unsigned long long)dm_block_location(b));
175                 return -ENOTBLK;
176         }
177
178         if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
179                 DMERR("sb_check failed: magic %llu: wanted %llu",
180                       le64_to_cpu(disk_super->magic),
181                       (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
182                 return -EILSEQ;
183         }
184
185         csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
186                                              sb_block_size - sizeof(__le32),
187                                              SUPERBLOCK_CSUM_XOR));
188         if (csum_le != disk_super->csum) {
189                 DMERR("sb_check failed: csum %u: wanted %u",
190                       le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
191                 return -EILSEQ;
192         }
193
194         return check_metadata_version(disk_super);
195 }
196
197 static struct dm_block_validator sb_validator = {
198         .name = "superblock",
199         .prepare_for_write = sb_prepare_for_write,
200         .check = sb_check
201 };
202
203 /*----------------------------------------------------------------*/
204
205 static int superblock_read_lock(struct dm_cache_metadata *cmd,
206                                 struct dm_block **sblock)
207 {
208         return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
209                                &sb_validator, sblock);
210 }
211
212 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
213                                 struct dm_block **sblock)
214 {
215         return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
216                                      &sb_validator, sblock);
217 }
218
219 static int superblock_lock(struct dm_cache_metadata *cmd,
220                            struct dm_block **sblock)
221 {
222         return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
223                                 &sb_validator, sblock);
224 }
225
226 /*----------------------------------------------------------------*/
227
228 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
229 {
230         int r;
231         unsigned i;
232         struct dm_block *b;
233         __le64 *data_le, zero = cpu_to_le64(0);
234         unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
235
236         /*
237          * We can't use a validator here - it may be all zeroes.
238          */
239         r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
240         if (r)
241                 return r;
242
243         data_le = dm_block_data(b);
244         *result = true;
245         for (i = 0; i < sb_block_size; i++) {
246                 if (data_le[i] != zero) {
247                         *result = false;
248                         break;
249                 }
250         }
251
252         return dm_bm_unlock(b);
253 }
254
255 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
256 {
257         struct dm_btree_value_type vt;
258
259         vt.context = NULL;
260         vt.size = sizeof(__le64);
261         vt.inc = NULL;
262         vt.dec = NULL;
263         vt.equal = NULL;
264         dm_array_info_init(&cmd->info, cmd->tm, &vt);
265
266         if (cmd->policy_hint_size) {
267                 vt.size = sizeof(__le32);
268                 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
269         }
270 }
271
272 static int __save_sm_root(struct dm_cache_metadata *cmd)
273 {
274         int r;
275         size_t metadata_len;
276
277         r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
278         if (r < 0)
279                 return r;
280
281         return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
282                                metadata_len);
283 }
284
285 static void __copy_sm_root(struct dm_cache_metadata *cmd,
286                            struct cache_disk_superblock *disk_super)
287 {
288         memcpy(&disk_super->metadata_space_map_root,
289                &cmd->metadata_space_map_root,
290                sizeof(cmd->metadata_space_map_root));
291 }
292
293 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
294 {
295         int r;
296         struct dm_block *sblock;
297         struct cache_disk_superblock *disk_super;
298         sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
299
300         /* FIXME: see if we can lose the max sectors limit */
301         if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
302                 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
303
304         r = dm_tm_pre_commit(cmd->tm);
305         if (r < 0)
306                 return r;
307
308         /*
309          * dm_sm_copy_root() can fail.  So we need to do it before we start
310          * updating the superblock.
311          */
312         r = __save_sm_root(cmd);
313         if (r)
314                 return r;
315
316         r = superblock_lock_zero(cmd, &sblock);
317         if (r)
318                 return r;
319
320         disk_super = dm_block_data(sblock);
321         disk_super->flags = 0;
322         memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
323         disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
324         disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
325         memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
326         memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
327         disk_super->policy_hint_size = 0;
328
329         __copy_sm_root(cmd, disk_super);
330
331         disk_super->mapping_root = cpu_to_le64(cmd->root);
332         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
333         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
334         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
335         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
336         disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
337         disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
338         disk_super->cache_blocks = cpu_to_le32(0);
339
340         disk_super->read_hits = cpu_to_le32(0);
341         disk_super->read_misses = cpu_to_le32(0);
342         disk_super->write_hits = cpu_to_le32(0);
343         disk_super->write_misses = cpu_to_le32(0);
344
345         return dm_tm_commit(cmd->tm, sblock);
346 }
347
348 static int __format_metadata(struct dm_cache_metadata *cmd)
349 {
350         int r;
351
352         r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
353                                  &cmd->tm, &cmd->metadata_sm);
354         if (r < 0) {
355                 DMERR("tm_create_with_sm failed");
356                 return r;
357         }
358
359         __setup_mapping_info(cmd);
360
361         r = dm_array_empty(&cmd->info, &cmd->root);
362         if (r < 0)
363                 goto bad;
364
365         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
366
367         r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
368         if (r < 0)
369                 goto bad;
370
371         cmd->discard_block_size = 0;
372         cmd->discard_nr_blocks = 0;
373
374         r = __write_initial_superblock(cmd);
375         if (r)
376                 goto bad;
377
378         cmd->clean_when_opened = true;
379         return 0;
380
381 bad:
382         dm_tm_destroy(cmd->tm);
383         dm_sm_destroy(cmd->metadata_sm);
384
385         return r;
386 }
387
388 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
389                                      struct dm_cache_metadata *cmd)
390 {
391         uint32_t features;
392
393         features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
394         if (features) {
395                 DMERR("could not access metadata due to unsupported optional features (%lx).",
396                       (unsigned long)features);
397                 return -EINVAL;
398         }
399
400         /*
401          * Check for read-only metadata to skip the following RDWR checks.
402          */
403         if (get_disk_ro(cmd->bdev->bd_disk))
404                 return 0;
405
406         features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
407         if (features) {
408                 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
409                       (unsigned long)features);
410                 return -EINVAL;
411         }
412
413         return 0;
414 }
415
416 static int __open_metadata(struct dm_cache_metadata *cmd)
417 {
418         int r;
419         struct dm_block *sblock;
420         struct cache_disk_superblock *disk_super;
421         unsigned long sb_flags;
422
423         r = superblock_read_lock(cmd, &sblock);
424         if (r < 0) {
425                 DMERR("couldn't read lock superblock");
426                 return r;
427         }
428
429         disk_super = dm_block_data(sblock);
430
431         /* Verify the data block size hasn't changed */
432         if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
433                 DMERR("changing the data block size (from %u to %llu) is not supported",
434                       le32_to_cpu(disk_super->data_block_size),
435                       (unsigned long long)cmd->data_block_size);
436                 r = -EINVAL;
437                 goto bad;
438         }
439
440         r = __check_incompat_features(disk_super, cmd);
441         if (r < 0)
442                 goto bad;
443
444         r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
445                                disk_super->metadata_space_map_root,
446                                sizeof(disk_super->metadata_space_map_root),
447                                &cmd->tm, &cmd->metadata_sm);
448         if (r < 0) {
449                 DMERR("tm_open_with_sm failed");
450                 goto bad;
451         }
452
453         __setup_mapping_info(cmd);
454         dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
455         sb_flags = le32_to_cpu(disk_super->flags);
456         cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
457         return dm_bm_unlock(sblock);
458
459 bad:
460         dm_bm_unlock(sblock);
461         return r;
462 }
463
464 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
465                                      bool format_device)
466 {
467         int r;
468         bool unformatted = false;
469
470         r = __superblock_all_zeroes(cmd->bm, &unformatted);
471         if (r)
472                 return r;
473
474         if (unformatted)
475                 return format_device ? __format_metadata(cmd) : -EPERM;
476
477         return __open_metadata(cmd);
478 }
479
480 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
481                                             bool may_format_device)
482 {
483         int r;
484         cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
485                                           CACHE_METADATA_CACHE_SIZE,
486                                           CACHE_MAX_CONCURRENT_LOCKS);
487         if (IS_ERR(cmd->bm)) {
488                 DMERR("could not create block manager");
489                 return PTR_ERR(cmd->bm);
490         }
491
492         r = __open_or_format_metadata(cmd, may_format_device);
493         if (r)
494                 dm_block_manager_destroy(cmd->bm);
495
496         return r;
497 }
498
499 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
500 {
501         dm_sm_destroy(cmd->metadata_sm);
502         dm_tm_destroy(cmd->tm);
503         dm_block_manager_destroy(cmd->bm);
504 }
505
506 typedef unsigned long (*flags_mutator)(unsigned long);
507
508 static void update_flags(struct cache_disk_superblock *disk_super,
509                          flags_mutator mutator)
510 {
511         uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
512         disk_super->flags = cpu_to_le32(sb_flags);
513 }
514
515 static unsigned long set_clean_shutdown(unsigned long flags)
516 {
517         set_bit(CLEAN_SHUTDOWN, &flags);
518         return flags;
519 }
520
521 static unsigned long clear_clean_shutdown(unsigned long flags)
522 {
523         clear_bit(CLEAN_SHUTDOWN, &flags);
524         return flags;
525 }
526
527 static void read_superblock_fields(struct dm_cache_metadata *cmd,
528                                    struct cache_disk_superblock *disk_super)
529 {
530         cmd->root = le64_to_cpu(disk_super->mapping_root);
531         cmd->hint_root = le64_to_cpu(disk_super->hint_root);
532         cmd->discard_root = le64_to_cpu(disk_super->discard_root);
533         cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
534         cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
535         cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
536         cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
537         strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
538         cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
539         cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
540         cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
541         cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
542
543         cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
544         cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
545         cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
546         cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
547
548         cmd->changed = false;
549 }
550
551 /*
552  * The mutator updates the superblock flags.
553  */
554 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
555                                      flags_mutator mutator)
556 {
557         int r;
558         struct cache_disk_superblock *disk_super;
559         struct dm_block *sblock;
560
561         r = superblock_lock(cmd, &sblock);
562         if (r)
563                 return r;
564
565         disk_super = dm_block_data(sblock);
566         update_flags(disk_super, mutator);
567         read_superblock_fields(cmd, disk_super);
568         dm_bm_unlock(sblock);
569
570         return dm_bm_flush(cmd->bm);
571 }
572
573 static int __begin_transaction(struct dm_cache_metadata *cmd)
574 {
575         int r;
576         struct cache_disk_superblock *disk_super;
577         struct dm_block *sblock;
578
579         /*
580          * We re-read the superblock every time.  Shouldn't need to do this
581          * really.
582          */
583         r = superblock_read_lock(cmd, &sblock);
584         if (r)
585                 return r;
586
587         disk_super = dm_block_data(sblock);
588         read_superblock_fields(cmd, disk_super);
589         dm_bm_unlock(sblock);
590
591         return 0;
592 }
593
594 static int __commit_transaction(struct dm_cache_metadata *cmd,
595                                 flags_mutator mutator)
596 {
597         int r;
598         struct cache_disk_superblock *disk_super;
599         struct dm_block *sblock;
600
601         /*
602          * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
603          */
604         BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
605
606         r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
607                             &cmd->discard_root);
608         if (r)
609                 return r;
610
611         r = dm_tm_pre_commit(cmd->tm);
612         if (r < 0)
613                 return r;
614
615         r = __save_sm_root(cmd);
616         if (r)
617                 return r;
618
619         r = superblock_lock(cmd, &sblock);
620         if (r)
621                 return r;
622
623         disk_super = dm_block_data(sblock);
624
625         if (mutator)
626                 update_flags(disk_super, mutator);
627
628         disk_super->mapping_root = cpu_to_le64(cmd->root);
629         disk_super->hint_root = cpu_to_le64(cmd->hint_root);
630         disk_super->discard_root = cpu_to_le64(cmd->discard_root);
631         disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
632         disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
633         disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
634         strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
635         disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
636         disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
637         disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
638
639         disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
640         disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
641         disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
642         disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
643         __copy_sm_root(cmd, disk_super);
644
645         return dm_tm_commit(cmd->tm, sblock);
646 }
647
648 /*----------------------------------------------------------------*/
649
650 /*
651  * The mappings are held in a dm-array that has 64-bit values stored in
652  * little-endian format.  The index is the cblock, the high 48bits of the
653  * value are the oblock and the low 16 bit the flags.
654  */
655 #define FLAGS_MASK ((1 << 16) - 1)
656
657 static __le64 pack_value(dm_oblock_t block, unsigned flags)
658 {
659         uint64_t value = from_oblock(block);
660         value <<= 16;
661         value = value | (flags & FLAGS_MASK);
662         return cpu_to_le64(value);
663 }
664
665 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
666 {
667         uint64_t value = le64_to_cpu(value_le);
668         uint64_t b = value >> 16;
669         *block = to_oblock(b);
670         *flags = value & FLAGS_MASK;
671 }
672
673 /*----------------------------------------------------------------*/
674
675 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
676                                                sector_t data_block_size,
677                                                bool may_format_device,
678                                                size_t policy_hint_size)
679 {
680         int r;
681         struct dm_cache_metadata *cmd;
682
683         cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
684         if (!cmd) {
685                 DMERR("could not allocate metadata struct");
686                 return ERR_PTR(-ENOMEM);
687         }
688
689         atomic_set(&cmd->ref_count, 1);
690         init_rwsem(&cmd->root_lock);
691         cmd->bdev = bdev;
692         cmd->data_block_size = data_block_size;
693         cmd->cache_blocks = 0;
694         cmd->policy_hint_size = policy_hint_size;
695         cmd->changed = true;
696
697         r = __create_persistent_data_objects(cmd, may_format_device);
698         if (r) {
699                 kfree(cmd);
700                 return ERR_PTR(r);
701         }
702
703         r = __begin_transaction_flags(cmd, clear_clean_shutdown);
704         if (r < 0) {
705                 dm_cache_metadata_close(cmd);
706                 return ERR_PTR(r);
707         }
708
709         return cmd;
710 }
711
712 /*
713  * We keep a little list of ref counted metadata objects to prevent two
714  * different target instances creating separate bufio instances.  This is
715  * an issue if a table is reloaded before the suspend.
716  */
717 static DEFINE_MUTEX(table_lock);
718 static LIST_HEAD(table);
719
720 static struct dm_cache_metadata *lookup(struct block_device *bdev)
721 {
722         struct dm_cache_metadata *cmd;
723
724         list_for_each_entry(cmd, &table, list)
725                 if (cmd->bdev == bdev) {
726                         atomic_inc(&cmd->ref_count);
727                         return cmd;
728                 }
729
730         return NULL;
731 }
732
733 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
734                                                 sector_t data_block_size,
735                                                 bool may_format_device,
736                                                 size_t policy_hint_size)
737 {
738         struct dm_cache_metadata *cmd, *cmd2;
739
740         mutex_lock(&table_lock);
741         cmd = lookup(bdev);
742         mutex_unlock(&table_lock);
743
744         if (cmd)
745                 return cmd;
746
747         cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
748         if (!IS_ERR(cmd)) {
749                 mutex_lock(&table_lock);
750                 cmd2 = lookup(bdev);
751                 if (cmd2) {
752                         mutex_unlock(&table_lock);
753                         __destroy_persistent_data_objects(cmd);
754                         kfree(cmd);
755                         return cmd2;
756                 }
757                 list_add(&cmd->list, &table);
758                 mutex_unlock(&table_lock);
759         }
760
761         return cmd;
762 }
763
764 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
765 {
766         if (cmd->data_block_size != data_block_size) {
767                 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
768                       (unsigned long long) data_block_size,
769                       (unsigned long long) cmd->data_block_size);
770                 return false;
771         }
772
773         return true;
774 }
775
776 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
777                                                  sector_t data_block_size,
778                                                  bool may_format_device,
779                                                  size_t policy_hint_size)
780 {
781         struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
782                                                        may_format_device, policy_hint_size);
783
784         if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
785                 dm_cache_metadata_close(cmd);
786                 return ERR_PTR(-EINVAL);
787         }
788
789         return cmd;
790 }
791
792 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
793 {
794         if (atomic_dec_and_test(&cmd->ref_count)) {
795                 mutex_lock(&table_lock);
796                 list_del(&cmd->list);
797                 mutex_unlock(&table_lock);
798
799                 __destroy_persistent_data_objects(cmd);
800                 kfree(cmd);
801         }
802 }
803
804 /*
805  * Checks that the given cache block is either unmapped or clean.
806  */
807 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
808                                    bool *result)
809 {
810         int r;
811         __le64 value;
812         dm_oblock_t ob;
813         unsigned flags;
814
815         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
816         if (r) {
817                 DMERR("block_unmapped_or_clean failed");
818                 return r;
819         }
820
821         unpack_value(value, &ob, &flags);
822         *result = !((flags & M_VALID) && (flags & M_DIRTY));
823
824         return 0;
825 }
826
827 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
828                                         dm_cblock_t begin, dm_cblock_t end,
829                                         bool *result)
830 {
831         int r;
832         *result = true;
833
834         while (begin != end) {
835                 r = block_unmapped_or_clean(cmd, begin, result);
836                 if (r)
837                         return r;
838
839                 if (!*result) {
840                         DMERR("cache block %llu is dirty",
841                               (unsigned long long) from_cblock(begin));
842                         return 0;
843                 }
844
845                 begin = to_cblock(from_cblock(begin) + 1);
846         }
847
848         return 0;
849 }
850
851 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
852 {
853         int r;
854         bool clean;
855         __le64 null_mapping = pack_value(0, 0);
856
857         down_write(&cmd->root_lock);
858         __dm_bless_for_disk(&null_mapping);
859
860         if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
861                 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
862                 if (r) {
863                         __dm_unbless_for_disk(&null_mapping);
864                         goto out;
865                 }
866
867                 if (!clean) {
868                         DMERR("unable to shrink cache due to dirty blocks");
869                         r = -EINVAL;
870                         __dm_unbless_for_disk(&null_mapping);
871                         goto out;
872                 }
873         }
874
875         r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
876                             from_cblock(new_cache_size),
877                             &null_mapping, &cmd->root);
878         if (!r)
879                 cmd->cache_blocks = new_cache_size;
880         cmd->changed = true;
881
882 out:
883         up_write(&cmd->root_lock);
884
885         return r;
886 }
887
888 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
889                                    sector_t discard_block_size,
890                                    dm_dblock_t new_nr_entries)
891 {
892         int r;
893
894         down_write(&cmd->root_lock);
895         r = dm_bitset_resize(&cmd->discard_info,
896                              cmd->discard_root,
897                              from_dblock(cmd->discard_nr_blocks),
898                              from_dblock(new_nr_entries),
899                              false, &cmd->discard_root);
900         if (!r) {
901                 cmd->discard_block_size = discard_block_size;
902                 cmd->discard_nr_blocks = new_nr_entries;
903         }
904
905         cmd->changed = true;
906         up_write(&cmd->root_lock);
907
908         return r;
909 }
910
911 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
912 {
913         return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
914                                  from_dblock(b), &cmd->discard_root);
915 }
916
917 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
918 {
919         return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
920                                    from_dblock(b), &cmd->discard_root);
921 }
922
923 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
924                           bool *is_discarded)
925 {
926         return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
927                                   from_dblock(b), &cmd->discard_root,
928                                   is_discarded);
929 }
930
931 static int __discard(struct dm_cache_metadata *cmd,
932                      dm_dblock_t dblock, bool discard)
933 {
934         int r;
935
936         r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
937         if (r)
938                 return r;
939
940         cmd->changed = true;
941         return 0;
942 }
943
944 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
945                          dm_dblock_t dblock, bool discard)
946 {
947         int r;
948
949         down_write(&cmd->root_lock);
950         r = __discard(cmd, dblock, discard);
951         up_write(&cmd->root_lock);
952
953         return r;
954 }
955
956 static int __load_discards(struct dm_cache_metadata *cmd,
957                            load_discard_fn fn, void *context)
958 {
959         int r = 0;
960         dm_block_t b;
961         bool discard;
962
963         for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
964                 dm_dblock_t dblock = to_dblock(b);
965
966                 if (cmd->clean_when_opened) {
967                         r = __is_discarded(cmd, dblock, &discard);
968                         if (r)
969                                 return r;
970                 } else
971                         discard = false;
972
973                 r = fn(context, cmd->discard_block_size, dblock, discard);
974                 if (r)
975                         break;
976         }
977
978         return r;
979 }
980
981 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
982                            load_discard_fn fn, void *context)
983 {
984         int r;
985
986         down_read(&cmd->root_lock);
987         r = __load_discards(cmd, fn, context);
988         up_read(&cmd->root_lock);
989
990         return r;
991 }
992
993 dm_cblock_t dm_cache_size(struct dm_cache_metadata *cmd)
994 {
995         dm_cblock_t r;
996
997         down_read(&cmd->root_lock);
998         r = cmd->cache_blocks;
999         up_read(&cmd->root_lock);
1000
1001         return r;
1002 }
1003
1004 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1005 {
1006         int r;
1007         __le64 value = pack_value(0, 0);
1008
1009         __dm_bless_for_disk(&value);
1010         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1011                                &value, &cmd->root);
1012         if (r)
1013                 return r;
1014
1015         cmd->changed = true;
1016         return 0;
1017 }
1018
1019 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1020 {
1021         int r;
1022
1023         down_write(&cmd->root_lock);
1024         r = __remove(cmd, cblock);
1025         up_write(&cmd->root_lock);
1026
1027         return r;
1028 }
1029
1030 static int __insert(struct dm_cache_metadata *cmd,
1031                     dm_cblock_t cblock, dm_oblock_t oblock)
1032 {
1033         int r;
1034         __le64 value = pack_value(oblock, M_VALID);
1035         __dm_bless_for_disk(&value);
1036
1037         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1038                                &value, &cmd->root);
1039         if (r)
1040                 return r;
1041
1042         cmd->changed = true;
1043         return 0;
1044 }
1045
1046 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1047                             dm_cblock_t cblock, dm_oblock_t oblock)
1048 {
1049         int r;
1050
1051         down_write(&cmd->root_lock);
1052         r = __insert(cmd, cblock, oblock);
1053         up_write(&cmd->root_lock);
1054
1055         return r;
1056 }
1057
1058 struct thunk {
1059         load_mapping_fn fn;
1060         void *context;
1061
1062         struct dm_cache_metadata *cmd;
1063         bool respect_dirty_flags;
1064         bool hints_valid;
1065 };
1066
1067 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1068                              struct dm_cache_policy *policy)
1069 {
1070         const char *policy_name = dm_cache_policy_get_name(policy);
1071         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1072         size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1073
1074         /*
1075          * Ensure policy names match.
1076          */
1077         if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1078                 return false;
1079
1080         /*
1081          * Ensure policy major versions match.
1082          */
1083         if (cmd->policy_version[0] != policy_version[0])
1084                 return false;
1085
1086         /*
1087          * Ensure policy hint sizes match.
1088          */
1089         if (cmd->policy_hint_size != policy_hint_size)
1090                 return false;
1091
1092         return true;
1093 }
1094
1095 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1096 {
1097         return cmd->hint_root && cmd->policy_hint_size;
1098 }
1099
1100 static bool hints_array_available(struct dm_cache_metadata *cmd,
1101                                   struct dm_cache_policy *policy)
1102 {
1103         return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1104                 hints_array_initialized(cmd);
1105 }
1106
1107 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1108 {
1109         int r = 0;
1110         bool dirty;
1111         __le64 value;
1112         __le32 hint_value = 0;
1113         dm_oblock_t oblock;
1114         unsigned flags;
1115         struct thunk *thunk = context;
1116         struct dm_cache_metadata *cmd = thunk->cmd;
1117
1118         memcpy(&value, leaf, sizeof(value));
1119         unpack_value(value, &oblock, &flags);
1120
1121         if (flags & M_VALID) {
1122                 if (thunk->hints_valid) {
1123                         r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1124                                                cblock, &hint_value);
1125                         if (r && r != -ENODATA)
1126                                 return r;
1127                 }
1128
1129                 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1130                 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1131                               dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1132         }
1133
1134         return r;
1135 }
1136
1137 static int __load_mappings(struct dm_cache_metadata *cmd,
1138                            struct dm_cache_policy *policy,
1139                            load_mapping_fn fn, void *context)
1140 {
1141         struct thunk thunk;
1142
1143         thunk.fn = fn;
1144         thunk.context = context;
1145
1146         thunk.cmd = cmd;
1147         thunk.respect_dirty_flags = cmd->clean_when_opened;
1148         thunk.hints_valid = hints_array_available(cmd, policy);
1149
1150         return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1151 }
1152
1153 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1154                            struct dm_cache_policy *policy,
1155                            load_mapping_fn fn, void *context)
1156 {
1157         int r;
1158
1159         down_read(&cmd->root_lock);
1160         r = __load_mappings(cmd, policy, fn, context);
1161         up_read(&cmd->root_lock);
1162
1163         return r;
1164 }
1165
1166 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1167 {
1168         int r = 0;
1169         __le64 value;
1170         dm_oblock_t oblock;
1171         unsigned flags;
1172
1173         memcpy(&value, leaf, sizeof(value));
1174         unpack_value(value, &oblock, &flags);
1175
1176         return r;
1177 }
1178
1179 static int __dump_mappings(struct dm_cache_metadata *cmd)
1180 {
1181         return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1182 }
1183
1184 void dm_cache_dump(struct dm_cache_metadata *cmd)
1185 {
1186         down_read(&cmd->root_lock);
1187         __dump_mappings(cmd);
1188         up_read(&cmd->root_lock);
1189 }
1190
1191 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1192 {
1193         int r;
1194
1195         down_read(&cmd->root_lock);
1196         r = cmd->changed;
1197         up_read(&cmd->root_lock);
1198
1199         return r;
1200 }
1201
1202 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1203 {
1204         int r;
1205         unsigned flags;
1206         dm_oblock_t oblock;
1207         __le64 value;
1208
1209         r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1210         if (r)
1211                 return r;
1212
1213         unpack_value(value, &oblock, &flags);
1214
1215         if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1216                 /* nothing to be done */
1217                 return 0;
1218
1219         value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1220         __dm_bless_for_disk(&value);
1221
1222         r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1223                                &value, &cmd->root);
1224         if (r)
1225                 return r;
1226
1227         cmd->changed = true;
1228         return 0;
1229
1230 }
1231
1232 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1233                        dm_cblock_t cblock, bool dirty)
1234 {
1235         int r;
1236
1237         down_write(&cmd->root_lock);
1238         r = __dirty(cmd, cblock, dirty);
1239         up_write(&cmd->root_lock);
1240
1241         return r;
1242 }
1243
1244 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1245                                  struct dm_cache_statistics *stats)
1246 {
1247         down_read(&cmd->root_lock);
1248         *stats = cmd->stats;
1249         up_read(&cmd->root_lock);
1250 }
1251
1252 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1253                                  struct dm_cache_statistics *stats)
1254 {
1255         down_write(&cmd->root_lock);
1256         cmd->stats = *stats;
1257         up_write(&cmd->root_lock);
1258 }
1259
1260 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1261 {
1262         int r;
1263         flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1264                                  clear_clean_shutdown);
1265
1266         down_write(&cmd->root_lock);
1267         r = __commit_transaction(cmd, mutator);
1268         if (r)
1269                 goto out;
1270
1271         r = __begin_transaction(cmd);
1272
1273 out:
1274         up_write(&cmd->root_lock);
1275         return r;
1276 }
1277
1278 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1279                                            dm_block_t *result)
1280 {
1281         int r = -EINVAL;
1282
1283         down_read(&cmd->root_lock);
1284         r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1285         up_read(&cmd->root_lock);
1286
1287         return r;
1288 }
1289
1290 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1291                                    dm_block_t *result)
1292 {
1293         int r = -EINVAL;
1294
1295         down_read(&cmd->root_lock);
1296         r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1297         up_read(&cmd->root_lock);
1298
1299         return r;
1300 }
1301
1302 /*----------------------------------------------------------------*/
1303
1304 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1305 {
1306         int r;
1307         __le32 value;
1308         size_t hint_size;
1309         const char *policy_name = dm_cache_policy_get_name(policy);
1310         const unsigned *policy_version = dm_cache_policy_get_version(policy);
1311
1312         if (!policy_name[0] ||
1313             (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1314                 return -EINVAL;
1315
1316         if (!policy_unchanged(cmd, policy)) {
1317                 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1318                 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1319
1320                 hint_size = dm_cache_policy_get_hint_size(policy);
1321                 if (!hint_size)
1322                         return 0; /* short-circuit hints initialization */
1323                 cmd->policy_hint_size = hint_size;
1324
1325                 if (cmd->hint_root) {
1326                         r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1327                         if (r)
1328                                 return r;
1329                 }
1330
1331                 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1332                 if (r)
1333                         return r;
1334
1335                 value = cpu_to_le32(0);
1336                 __dm_bless_for_disk(&value);
1337                 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1338                                     from_cblock(cmd->cache_blocks),
1339                                     &value, &cmd->hint_root);
1340                 if (r)
1341                         return r;
1342         }
1343
1344         return 0;
1345 }
1346
1347 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1348 {
1349         struct dm_cache_metadata *cmd = context;
1350         __le32 value = cpu_to_le32(hint);
1351         int r;
1352
1353         __dm_bless_for_disk(&value);
1354
1355         r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1356                                from_cblock(cblock), &value, &cmd->hint_root);
1357         cmd->changed = true;
1358
1359         return r;
1360 }
1361
1362 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1363 {
1364         int r;
1365
1366         r = begin_hints(cmd, policy);
1367         if (r) {
1368                 DMERR("begin_hints failed");
1369                 return r;
1370         }
1371
1372         return policy_walk_mappings(policy, save_hint, cmd);
1373 }
1374
1375 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1376 {
1377         int r;
1378
1379         down_write(&cmd->root_lock);
1380         r = write_hints(cmd, policy);
1381         up_write(&cmd->root_lock);
1382
1383         return r;
1384 }
1385
1386 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1387 {
1388         return blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1389 }