2 * Copyright (C) 2012 Red Hat, Inc.
4 * This file is released under the GPL.
7 #include "dm-cache-metadata.h"
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"
15 #include <linux/device-mapper.h>
17 /*----------------------------------------------------------------*/
19 #define DM_MSG_PREFIX "cache metadata"
21 #define CACHE_SUPERBLOCK_MAGIC 06142003
22 #define CACHE_SUPERBLOCK_LOCATION 0
25 * defines a range of metadata versions that this module can handle.
27 #define MIN_CACHE_VERSION 1
28 #define MAX_CACHE_VERSION 1
30 #define CACHE_METADATA_CACHE_SIZE 64
33 * 3 for btree insert +
34 * 2 for btree lookup used within space map
36 #define CACHE_MAX_CONCURRENT_LOCKS 5
37 #define SPACE_MAP_ROOT_SIZE 128
39 enum superblock_flag_bits {
40 /* for spotting crashes that would invalidate the dirty bitset */
42 /* metadata must be checked using the tools */
47 * Each mapping from cache block -> origin block carries a set of flags.
51 * A valid mapping. Because we're using an array we clear this
52 * flag for an non existant mapping.
57 * The data on the cache is different from that on the origin.
62 struct cache_disk_superblock {
71 __u8 policy_name[CACHE_POLICY_NAME_SIZE];
72 __le32 policy_hint_size;
74 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
79 __le64 discard_block_size;
80 __le64 discard_nr_blocks;
82 __le32 data_block_size;
83 __le32 metadata_block_size;
87 __le32 compat_ro_flags;
88 __le32 incompat_flags;
95 __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
98 struct dm_cache_metadata {
100 struct list_head list;
102 struct block_device *bdev;
103 struct dm_block_manager *bm;
104 struct dm_space_map *metadata_sm;
105 struct dm_transaction_manager *tm;
107 struct dm_array_info info;
108 struct dm_array_info hint_info;
109 struct dm_disk_bitset discard_info;
111 struct rw_semaphore root_lock;
114 dm_block_t hint_root;
115 dm_block_t discard_root;
117 sector_t discard_block_size;
118 dm_dblock_t discard_nr_blocks;
120 sector_t data_block_size;
121 dm_cblock_t cache_blocks;
123 bool clean_when_opened:1;
125 char policy_name[CACHE_POLICY_NAME_SIZE];
126 unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
127 size_t policy_hint_size;
128 struct dm_cache_statistics stats;
131 * Reading the space map root can fail, so we read it into this
132 * buffer before the superblock is locked and updated.
134 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
137 * Set if a transaction has to be aborted but the attempt to roll
138 * back to the previous (good) transaction failed. The only
139 * metadata operation permissible in this state is the closing of
145 * These structures are used when loading metadata. They're too
146 * big to put on the stack.
148 struct dm_array_cursor mapping_cursor;
149 struct dm_array_cursor hint_cursor;
152 /*-------------------------------------------------------------------
153 * superblock validator
154 *-----------------------------------------------------------------*/
156 #define SUPERBLOCK_CSUM_XOR 9031977
158 static void sb_prepare_for_write(struct dm_block_validator *v,
160 size_t sb_block_size)
162 struct cache_disk_superblock *disk_super = dm_block_data(b);
164 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
165 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
166 sb_block_size - sizeof(__le32),
167 SUPERBLOCK_CSUM_XOR));
170 static int check_metadata_version(struct cache_disk_superblock *disk_super)
172 uint32_t metadata_version = le32_to_cpu(disk_super->version);
173 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
174 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
175 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
182 static int sb_check(struct dm_block_validator *v,
184 size_t sb_block_size)
186 struct cache_disk_superblock *disk_super = dm_block_data(b);
189 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
190 DMERR("sb_check failed: blocknr %llu: wanted %llu",
191 le64_to_cpu(disk_super->blocknr),
192 (unsigned long long)dm_block_location(b));
196 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
197 DMERR("sb_check failed: magic %llu: wanted %llu",
198 le64_to_cpu(disk_super->magic),
199 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
203 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
204 sb_block_size - sizeof(__le32),
205 SUPERBLOCK_CSUM_XOR));
206 if (csum_le != disk_super->csum) {
207 DMERR("sb_check failed: csum %u: wanted %u",
208 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
212 return check_metadata_version(disk_super);
215 static struct dm_block_validator sb_validator = {
216 .name = "superblock",
217 .prepare_for_write = sb_prepare_for_write,
221 /*----------------------------------------------------------------*/
223 static int superblock_read_lock(struct dm_cache_metadata *cmd,
224 struct dm_block **sblock)
226 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
227 &sb_validator, sblock);
230 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
231 struct dm_block **sblock)
233 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
234 &sb_validator, sblock);
237 static int superblock_lock(struct dm_cache_metadata *cmd,
238 struct dm_block **sblock)
240 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
241 &sb_validator, sblock);
244 /*----------------------------------------------------------------*/
246 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
251 __le64 *data_le, zero = cpu_to_le64(0);
252 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
255 * We can't use a validator here - it may be all zeroes.
257 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
261 data_le = dm_block_data(b);
263 for (i = 0; i < sb_block_size; i++) {
264 if (data_le[i] != zero) {
275 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
277 struct dm_btree_value_type vt;
280 vt.size = sizeof(__le64);
284 dm_array_info_init(&cmd->info, cmd->tm, &vt);
286 if (cmd->policy_hint_size) {
287 vt.size = sizeof(__le32);
288 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
292 static int __save_sm_root(struct dm_cache_metadata *cmd)
297 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
301 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
305 static void __copy_sm_root(struct dm_cache_metadata *cmd,
306 struct cache_disk_superblock *disk_super)
308 memcpy(&disk_super->metadata_space_map_root,
309 &cmd->metadata_space_map_root,
310 sizeof(cmd->metadata_space_map_root));
313 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
316 struct dm_block *sblock;
317 struct cache_disk_superblock *disk_super;
318 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
320 /* FIXME: see if we can lose the max sectors limit */
321 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
322 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
324 r = dm_tm_pre_commit(cmd->tm);
329 * dm_sm_copy_root() can fail. So we need to do it before we start
330 * updating the superblock.
332 r = __save_sm_root(cmd);
336 r = superblock_lock_zero(cmd, &sblock);
340 disk_super = dm_block_data(sblock);
341 disk_super->flags = 0;
342 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
343 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
344 disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
345 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
346 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
347 disk_super->policy_hint_size = 0;
349 __copy_sm_root(cmd, disk_super);
351 disk_super->mapping_root = cpu_to_le64(cmd->root);
352 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
353 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
354 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
355 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
356 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
357 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
358 disk_super->cache_blocks = cpu_to_le32(0);
360 disk_super->read_hits = cpu_to_le32(0);
361 disk_super->read_misses = cpu_to_le32(0);
362 disk_super->write_hits = cpu_to_le32(0);
363 disk_super->write_misses = cpu_to_le32(0);
365 return dm_tm_commit(cmd->tm, sblock);
368 static int __format_metadata(struct dm_cache_metadata *cmd)
372 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
373 &cmd->tm, &cmd->metadata_sm);
375 DMERR("tm_create_with_sm failed");
379 __setup_mapping_info(cmd);
381 r = dm_array_empty(&cmd->info, &cmd->root);
385 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
386 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
390 cmd->discard_block_size = 0;
391 cmd->discard_nr_blocks = 0;
393 r = __write_initial_superblock(cmd);
397 cmd->clean_when_opened = true;
401 dm_tm_destroy(cmd->tm);
402 dm_sm_destroy(cmd->metadata_sm);
407 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
408 struct dm_cache_metadata *cmd)
412 features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
414 DMERR("could not access metadata due to unsupported optional features (%lx).",
415 (unsigned long)features);
420 * Check for read-only metadata to skip the following RDWR checks.
422 if (get_disk_ro(cmd->bdev->bd_disk))
425 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
427 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
428 (unsigned long)features);
435 static int __open_metadata(struct dm_cache_metadata *cmd)
438 struct dm_block *sblock;
439 struct cache_disk_superblock *disk_super;
440 unsigned long sb_flags;
442 r = superblock_read_lock(cmd, &sblock);
444 DMERR("couldn't read lock superblock");
448 disk_super = dm_block_data(sblock);
450 /* Verify the data block size hasn't changed */
451 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
452 DMERR("changing the data block size (from %u to %llu) is not supported",
453 le32_to_cpu(disk_super->data_block_size),
454 (unsigned long long)cmd->data_block_size);
459 r = __check_incompat_features(disk_super, cmd);
463 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
464 disk_super->metadata_space_map_root,
465 sizeof(disk_super->metadata_space_map_root),
466 &cmd->tm, &cmd->metadata_sm);
468 DMERR("tm_open_with_sm failed");
472 __setup_mapping_info(cmd);
473 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
474 sb_flags = le32_to_cpu(disk_super->flags);
475 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
476 dm_bm_unlock(sblock);
481 dm_bm_unlock(sblock);
485 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
489 bool unformatted = false;
491 r = __superblock_all_zeroes(cmd->bm, &unformatted);
496 return format_device ? __format_metadata(cmd) : -EPERM;
498 return __open_metadata(cmd);
501 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
502 bool may_format_device)
505 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
506 CACHE_METADATA_CACHE_SIZE,
507 CACHE_MAX_CONCURRENT_LOCKS);
508 if (IS_ERR(cmd->bm)) {
509 DMERR("could not create block manager");
510 return PTR_ERR(cmd->bm);
513 r = __open_or_format_metadata(cmd, may_format_device);
515 dm_block_manager_destroy(cmd->bm);
520 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
522 dm_sm_destroy(cmd->metadata_sm);
523 dm_tm_destroy(cmd->tm);
524 dm_block_manager_destroy(cmd->bm);
527 typedef unsigned long (*flags_mutator)(unsigned long);
529 static void update_flags(struct cache_disk_superblock *disk_super,
530 flags_mutator mutator)
532 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
533 disk_super->flags = cpu_to_le32(sb_flags);
536 static unsigned long set_clean_shutdown(unsigned long flags)
538 set_bit(CLEAN_SHUTDOWN, &flags);
542 static unsigned long clear_clean_shutdown(unsigned long flags)
544 clear_bit(CLEAN_SHUTDOWN, &flags);
548 static void read_superblock_fields(struct dm_cache_metadata *cmd,
549 struct cache_disk_superblock *disk_super)
551 cmd->flags = le32_to_cpu(disk_super->flags);
552 cmd->root = le64_to_cpu(disk_super->mapping_root);
553 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
554 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
555 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
556 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
557 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
558 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
559 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
560 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
561 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
562 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
563 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
565 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
566 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
567 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
568 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
570 cmd->changed = false;
574 * The mutator updates the superblock flags.
576 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
577 flags_mutator mutator)
580 struct cache_disk_superblock *disk_super;
581 struct dm_block *sblock;
583 r = superblock_lock(cmd, &sblock);
587 disk_super = dm_block_data(sblock);
588 update_flags(disk_super, mutator);
589 read_superblock_fields(cmd, disk_super);
590 dm_bm_unlock(sblock);
592 return dm_bm_flush(cmd->bm);
595 static int __begin_transaction(struct dm_cache_metadata *cmd)
598 struct cache_disk_superblock *disk_super;
599 struct dm_block *sblock;
602 * We re-read the superblock every time. Shouldn't need to do this
605 r = superblock_read_lock(cmd, &sblock);
609 disk_super = dm_block_data(sblock);
610 read_superblock_fields(cmd, disk_super);
611 dm_bm_unlock(sblock);
616 static int __commit_transaction(struct dm_cache_metadata *cmd,
617 flags_mutator mutator)
620 struct cache_disk_superblock *disk_super;
621 struct dm_block *sblock;
624 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
626 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
628 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
633 r = dm_tm_pre_commit(cmd->tm);
637 r = __save_sm_root(cmd);
641 r = superblock_lock(cmd, &sblock);
645 disk_super = dm_block_data(sblock);
647 disk_super->flags = cpu_to_le32(cmd->flags);
649 update_flags(disk_super, mutator);
651 disk_super->mapping_root = cpu_to_le64(cmd->root);
652 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
653 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
654 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
655 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
656 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
657 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
658 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
659 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
660 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
662 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
663 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
664 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
665 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
666 __copy_sm_root(cmd, disk_super);
668 return dm_tm_commit(cmd->tm, sblock);
671 /*----------------------------------------------------------------*/
674 * The mappings are held in a dm-array that has 64-bit values stored in
675 * little-endian format. The index is the cblock, the high 48bits of the
676 * value are the oblock and the low 16 bit the flags.
678 #define FLAGS_MASK ((1 << 16) - 1)
680 static __le64 pack_value(dm_oblock_t block, unsigned flags)
682 uint64_t value = from_oblock(block);
684 value = value | (flags & FLAGS_MASK);
685 return cpu_to_le64(value);
688 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
690 uint64_t value = le64_to_cpu(value_le);
691 uint64_t b = value >> 16;
692 *block = to_oblock(b);
693 *flags = value & FLAGS_MASK;
696 /*----------------------------------------------------------------*/
698 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
699 sector_t data_block_size,
700 bool may_format_device,
701 size_t policy_hint_size)
704 struct dm_cache_metadata *cmd;
706 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
708 DMERR("could not allocate metadata struct");
709 return ERR_PTR(-ENOMEM);
712 atomic_set(&cmd->ref_count, 1);
713 init_rwsem(&cmd->root_lock);
715 cmd->data_block_size = data_block_size;
716 cmd->cache_blocks = 0;
717 cmd->policy_hint_size = policy_hint_size;
719 cmd->fail_io = false;
721 r = __create_persistent_data_objects(cmd, may_format_device);
727 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
729 dm_cache_metadata_close(cmd);
737 * We keep a little list of ref counted metadata objects to prevent two
738 * different target instances creating separate bufio instances. This is
739 * an issue if a table is reloaded before the suspend.
741 static DEFINE_MUTEX(table_lock);
742 static LIST_HEAD(table);
744 static struct dm_cache_metadata *lookup(struct block_device *bdev)
746 struct dm_cache_metadata *cmd;
748 list_for_each_entry(cmd, &table, list)
749 if (cmd->bdev == bdev) {
750 atomic_inc(&cmd->ref_count);
757 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
758 sector_t data_block_size,
759 bool may_format_device,
760 size_t policy_hint_size)
762 struct dm_cache_metadata *cmd, *cmd2;
764 mutex_lock(&table_lock);
766 mutex_unlock(&table_lock);
771 cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
773 mutex_lock(&table_lock);
776 mutex_unlock(&table_lock);
777 __destroy_persistent_data_objects(cmd);
781 list_add(&cmd->list, &table);
782 mutex_unlock(&table_lock);
788 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
790 if (cmd->data_block_size != data_block_size) {
791 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
792 (unsigned long long) data_block_size,
793 (unsigned long long) cmd->data_block_size);
800 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
801 sector_t data_block_size,
802 bool may_format_device,
803 size_t policy_hint_size)
805 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
806 may_format_device, policy_hint_size);
808 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
809 dm_cache_metadata_close(cmd);
810 return ERR_PTR(-EINVAL);
816 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
818 if (atomic_dec_and_test(&cmd->ref_count)) {
819 mutex_lock(&table_lock);
820 list_del(&cmd->list);
821 mutex_unlock(&table_lock);
824 __destroy_persistent_data_objects(cmd);
830 * Checks that the given cache block is either unmapped or clean.
832 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
840 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
842 DMERR("block_unmapped_or_clean failed");
846 unpack_value(value, &ob, &flags);
847 *result = !((flags & M_VALID) && (flags & M_DIRTY));
852 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
853 dm_cblock_t begin, dm_cblock_t end,
859 while (begin != end) {
860 r = block_unmapped_or_clean(cmd, begin, result);
865 DMERR("cache block %llu is dirty",
866 (unsigned long long) from_cblock(begin));
870 begin = to_cblock(from_cblock(begin) + 1);
876 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
878 down_write(&cmd->root_lock);
879 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
880 up_write(&cmd->root_lock);
886 #define WRITE_LOCK(cmd) \
888 if (!cmd_write_lock((cmd))) \
892 #define WRITE_LOCK_VOID(cmd) \
894 if (!cmd_write_lock((cmd))) \
898 #define WRITE_UNLOCK(cmd) \
899 up_write(&(cmd)->root_lock)
901 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
903 down_read(&cmd->root_lock);
905 up_read(&cmd->root_lock);
911 #define READ_LOCK(cmd) \
913 if (!cmd_read_lock((cmd))) \
917 #define READ_LOCK_VOID(cmd) \
919 if (!cmd_read_lock((cmd))) \
923 #define READ_UNLOCK(cmd) \
924 up_read(&(cmd)->root_lock)
926 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
930 __le64 null_mapping = pack_value(0, 0);
933 __dm_bless_for_disk(&null_mapping);
935 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
936 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
938 __dm_unbless_for_disk(&null_mapping);
943 DMERR("unable to shrink cache due to dirty blocks");
945 __dm_unbless_for_disk(&null_mapping);
950 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
951 from_cblock(new_cache_size),
952 &null_mapping, &cmd->root);
954 cmd->cache_blocks = new_cache_size;
963 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
964 sector_t discard_block_size,
965 dm_dblock_t new_nr_entries)
970 r = dm_bitset_resize(&cmd->discard_info,
972 from_dblock(cmd->discard_nr_blocks),
973 from_dblock(new_nr_entries),
974 false, &cmd->discard_root);
976 cmd->discard_block_size = discard_block_size;
977 cmd->discard_nr_blocks = new_nr_entries;
986 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
988 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
989 from_dblock(b), &cmd->discard_root);
992 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
994 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
995 from_dblock(b), &cmd->discard_root);
998 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
1001 return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1002 from_dblock(b), &cmd->discard_root,
1006 static int __discard(struct dm_cache_metadata *cmd,
1007 dm_dblock_t dblock, bool discard)
1011 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1015 cmd->changed = true;
1019 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1020 dm_dblock_t dblock, bool discard)
1025 r = __discard(cmd, dblock, discard);
1031 static int __load_discards(struct dm_cache_metadata *cmd,
1032 load_discard_fn fn, void *context)
1038 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1039 dm_dblock_t dblock = to_dblock(b);
1041 if (cmd->clean_when_opened) {
1042 r = __is_discarded(cmd, dblock, &discard);
1048 r = fn(context, cmd->discard_block_size, dblock, discard);
1056 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1057 load_discard_fn fn, void *context)
1062 r = __load_discards(cmd, fn, context);
1068 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1071 *result = cmd->cache_blocks;
1077 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1080 __le64 value = pack_value(0, 0);
1082 __dm_bless_for_disk(&value);
1083 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1084 &value, &cmd->root);
1088 cmd->changed = true;
1092 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1097 r = __remove(cmd, cblock);
1103 static int __insert(struct dm_cache_metadata *cmd,
1104 dm_cblock_t cblock, dm_oblock_t oblock)
1107 __le64 value = pack_value(oblock, M_VALID);
1108 __dm_bless_for_disk(&value);
1110 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1111 &value, &cmd->root);
1115 cmd->changed = true;
1119 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1120 dm_cblock_t cblock, dm_oblock_t oblock)
1125 r = __insert(cmd, cblock, oblock);
1135 struct dm_cache_metadata *cmd;
1136 bool respect_dirty_flags;
1140 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1141 struct dm_cache_policy *policy)
1143 const char *policy_name = dm_cache_policy_get_name(policy);
1144 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1145 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1148 * Ensure policy names match.
1150 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1154 * Ensure policy major versions match.
1156 if (cmd->policy_version[0] != policy_version[0])
1160 * Ensure policy hint sizes match.
1162 if (cmd->policy_hint_size != policy_hint_size)
1168 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1170 return cmd->hint_root && cmd->policy_hint_size;
1173 static bool hints_array_available(struct dm_cache_metadata *cmd,
1174 struct dm_cache_policy *policy)
1176 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1177 hints_array_initialized(cmd);
1180 static int __load_mapping(struct dm_cache_metadata *cmd,
1181 uint64_t cb, bool hints_valid,
1182 struct dm_array_cursor *mapping_cursor,
1183 struct dm_array_cursor *hint_cursor,
1184 load_mapping_fn fn, void *context)
1191 __le64 *mapping_value_le;
1192 __le32 *hint_value_le;
1197 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1198 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1199 unpack_value(mapping, &oblock, &flags);
1201 if (flags & M_VALID) {
1203 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1204 memcpy(&hint, hint_value_le, sizeof(hint));
1207 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1208 le32_to_cpu(hint), hints_valid);
1210 DMERR("policy couldn't load cblock");
1216 static int __load_mappings(struct dm_cache_metadata *cmd,
1217 struct dm_cache_policy *policy,
1218 load_mapping_fn fn, void *context)
1223 bool hints_valid = hints_array_available(cmd, policy);
1225 if (from_cblock(cmd->cache_blocks) == 0)
1229 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1234 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1236 dm_array_cursor_end(&cmd->mapping_cursor);
1241 for (cb = 0; ; cb++) {
1242 r = __load_mapping(cmd, cb, hints_valid,
1243 &cmd->mapping_cursor, &cmd->hint_cursor,
1249 * We need to break out before we move the cursors.
1251 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1254 r = dm_array_cursor_next(&cmd->mapping_cursor);
1256 DMERR("dm_array_cursor_next for mapping failed");
1261 r = dm_array_cursor_next(&cmd->hint_cursor);
1263 DMERR("dm_array_cursor_next for hint failed");
1269 dm_array_cursor_end(&cmd->mapping_cursor);
1271 dm_array_cursor_end(&cmd->hint_cursor);
1276 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1277 struct dm_cache_policy *policy,
1278 load_mapping_fn fn, void *context)
1283 r = __load_mappings(cmd, policy, fn, context);
1289 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1296 memcpy(&value, leaf, sizeof(value));
1297 unpack_value(value, &oblock, &flags);
1302 static int __dump_mappings(struct dm_cache_metadata *cmd)
1304 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1307 void dm_cache_dump(struct dm_cache_metadata *cmd)
1309 READ_LOCK_VOID(cmd);
1310 __dump_mappings(cmd);
1314 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1325 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1332 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1336 unpack_value(value, &oblock, &flags);
1338 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1339 /* nothing to be done */
1342 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1343 __dm_bless_for_disk(&value);
1345 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1346 &value, &cmd->root);
1350 cmd->changed = true;
1355 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1356 dm_cblock_t cblock, bool dirty)
1361 r = __dirty(cmd, cblock, dirty);
1367 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1368 struct dm_cache_statistics *stats)
1370 READ_LOCK_VOID(cmd);
1371 *stats = cmd->stats;
1375 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1376 struct dm_cache_statistics *stats)
1378 WRITE_LOCK_VOID(cmd);
1379 cmd->stats = *stats;
1383 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1386 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1387 clear_clean_shutdown);
1390 r = __commit_transaction(cmd, mutator);
1394 r = __begin_transaction(cmd);
1401 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1407 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1413 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1419 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1425 /*----------------------------------------------------------------*/
1427 static int get_hint(uint32_t index, void *value_le, void *context)
1430 struct dm_cache_policy *policy = context;
1432 value = policy_get_hint(policy, to_cblock(index));
1433 *((__le32 *) value_le) = cpu_to_le32(value);
1439 * It's quicker to always delete the hint array, and recreate with
1442 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1446 const char *policy_name = dm_cache_policy_get_name(policy);
1447 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1449 if (!policy_name[0] ||
1450 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1453 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1454 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1456 hint_size = dm_cache_policy_get_hint_size(policy);
1458 return 0; /* short-circuit hints initialization */
1459 cmd->policy_hint_size = hint_size;
1461 if (cmd->hint_root) {
1462 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1467 return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1468 from_cblock(cmd->cache_blocks),
1472 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1477 r = write_hints(cmd, policy);
1483 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1488 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1494 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1496 WRITE_LOCK_VOID(cmd);
1497 dm_bm_set_read_only(cmd->bm);
1501 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1503 WRITE_LOCK_VOID(cmd);
1504 dm_bm_set_read_write(cmd->bm);
1508 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1511 struct dm_block *sblock;
1512 struct cache_disk_superblock *disk_super;
1515 set_bit(NEEDS_CHECK, &cmd->flags);
1517 r = superblock_lock(cmd, &sblock);
1519 DMERR("couldn't read superblock");
1523 disk_super = dm_block_data(sblock);
1524 disk_super->flags = cpu_to_le32(cmd->flags);
1526 dm_bm_unlock(sblock);
1533 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1536 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1542 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1547 __destroy_persistent_data_objects(cmd);
1548 r = __create_persistent_data_objects(cmd, false);
1550 cmd->fail_io = true;