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);
387 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
391 cmd->discard_block_size = 0;
392 cmd->discard_nr_blocks = 0;
394 r = __write_initial_superblock(cmd);
398 cmd->clean_when_opened = true;
402 dm_tm_destroy(cmd->tm);
403 dm_sm_destroy(cmd->metadata_sm);
408 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
409 struct dm_cache_metadata *cmd)
413 features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
415 DMERR("could not access metadata due to unsupported optional features (%lx).",
416 (unsigned long)features);
421 * Check for read-only metadata to skip the following RDWR checks.
423 if (get_disk_ro(cmd->bdev->bd_disk))
426 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
428 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
429 (unsigned long)features);
436 static int __open_metadata(struct dm_cache_metadata *cmd)
439 struct dm_block *sblock;
440 struct cache_disk_superblock *disk_super;
441 unsigned long sb_flags;
443 r = superblock_read_lock(cmd, &sblock);
445 DMERR("couldn't read lock superblock");
449 disk_super = dm_block_data(sblock);
451 /* Verify the data block size hasn't changed */
452 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
453 DMERR("changing the data block size (from %u to %llu) is not supported",
454 le32_to_cpu(disk_super->data_block_size),
455 (unsigned long long)cmd->data_block_size);
460 r = __check_incompat_features(disk_super, cmd);
464 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
465 disk_super->metadata_space_map_root,
466 sizeof(disk_super->metadata_space_map_root),
467 &cmd->tm, &cmd->metadata_sm);
469 DMERR("tm_open_with_sm failed");
473 __setup_mapping_info(cmd);
474 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
475 sb_flags = le32_to_cpu(disk_super->flags);
476 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
477 dm_bm_unlock(sblock);
482 dm_bm_unlock(sblock);
486 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
490 bool unformatted = false;
492 r = __superblock_all_zeroes(cmd->bm, &unformatted);
497 return format_device ? __format_metadata(cmd) : -EPERM;
499 return __open_metadata(cmd);
502 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
503 bool may_format_device)
506 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
507 CACHE_METADATA_CACHE_SIZE,
508 CACHE_MAX_CONCURRENT_LOCKS);
509 if (IS_ERR(cmd->bm)) {
510 DMERR("could not create block manager");
511 return PTR_ERR(cmd->bm);
514 r = __open_or_format_metadata(cmd, may_format_device);
516 dm_block_manager_destroy(cmd->bm);
521 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
523 dm_sm_destroy(cmd->metadata_sm);
524 dm_tm_destroy(cmd->tm);
525 dm_block_manager_destroy(cmd->bm);
528 typedef unsigned long (*flags_mutator)(unsigned long);
530 static void update_flags(struct cache_disk_superblock *disk_super,
531 flags_mutator mutator)
533 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
534 disk_super->flags = cpu_to_le32(sb_flags);
537 static unsigned long set_clean_shutdown(unsigned long flags)
539 set_bit(CLEAN_SHUTDOWN, &flags);
543 static unsigned long clear_clean_shutdown(unsigned long flags)
545 clear_bit(CLEAN_SHUTDOWN, &flags);
549 static void read_superblock_fields(struct dm_cache_metadata *cmd,
550 struct cache_disk_superblock *disk_super)
552 cmd->flags = le32_to_cpu(disk_super->flags);
553 cmd->root = le64_to_cpu(disk_super->mapping_root);
554 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
555 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
556 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
557 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
558 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
559 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
560 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
561 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
562 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
563 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
564 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
566 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
567 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
568 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
569 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
571 cmd->changed = false;
575 * The mutator updates the superblock flags.
577 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
578 flags_mutator mutator)
581 struct cache_disk_superblock *disk_super;
582 struct dm_block *sblock;
584 r = superblock_lock(cmd, &sblock);
588 disk_super = dm_block_data(sblock);
589 update_flags(disk_super, mutator);
590 read_superblock_fields(cmd, disk_super);
591 dm_bm_unlock(sblock);
593 return dm_bm_flush(cmd->bm);
596 static int __begin_transaction(struct dm_cache_metadata *cmd)
599 struct cache_disk_superblock *disk_super;
600 struct dm_block *sblock;
603 * We re-read the superblock every time. Shouldn't need to do this
606 r = superblock_read_lock(cmd, &sblock);
610 disk_super = dm_block_data(sblock);
611 read_superblock_fields(cmd, disk_super);
612 dm_bm_unlock(sblock);
617 static int __commit_transaction(struct dm_cache_metadata *cmd,
618 flags_mutator mutator)
621 struct cache_disk_superblock *disk_super;
622 struct dm_block *sblock;
625 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
627 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
629 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
634 r = dm_tm_pre_commit(cmd->tm);
638 r = __save_sm_root(cmd);
642 r = superblock_lock(cmd, &sblock);
646 disk_super = dm_block_data(sblock);
648 disk_super->flags = cpu_to_le32(cmd->flags);
650 update_flags(disk_super, mutator);
652 disk_super->mapping_root = cpu_to_le64(cmd->root);
653 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
654 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
655 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
656 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
657 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
658 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
659 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
660 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
661 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
663 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
664 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
665 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
666 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
667 __copy_sm_root(cmd, disk_super);
669 return dm_tm_commit(cmd->tm, sblock);
672 /*----------------------------------------------------------------*/
675 * The mappings are held in a dm-array that has 64-bit values stored in
676 * little-endian format. The index is the cblock, the high 48bits of the
677 * value are the oblock and the low 16 bit the flags.
679 #define FLAGS_MASK ((1 << 16) - 1)
681 static __le64 pack_value(dm_oblock_t block, unsigned flags)
683 uint64_t value = from_oblock(block);
685 value = value | (flags & FLAGS_MASK);
686 return cpu_to_le64(value);
689 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
691 uint64_t value = le64_to_cpu(value_le);
692 uint64_t b = value >> 16;
693 *block = to_oblock(b);
694 *flags = value & FLAGS_MASK;
697 /*----------------------------------------------------------------*/
699 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
700 sector_t data_block_size,
701 bool may_format_device,
702 size_t policy_hint_size)
705 struct dm_cache_metadata *cmd;
707 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
709 DMERR("could not allocate metadata struct");
710 return ERR_PTR(-ENOMEM);
713 atomic_set(&cmd->ref_count, 1);
714 init_rwsem(&cmd->root_lock);
716 cmd->data_block_size = data_block_size;
717 cmd->cache_blocks = 0;
718 cmd->policy_hint_size = policy_hint_size;
720 cmd->fail_io = false;
722 r = __create_persistent_data_objects(cmd, may_format_device);
728 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
730 dm_cache_metadata_close(cmd);
738 * We keep a little list of ref counted metadata objects to prevent two
739 * different target instances creating separate bufio instances. This is
740 * an issue if a table is reloaded before the suspend.
742 static DEFINE_MUTEX(table_lock);
743 static LIST_HEAD(table);
745 static struct dm_cache_metadata *lookup(struct block_device *bdev)
747 struct dm_cache_metadata *cmd;
749 list_for_each_entry(cmd, &table, list)
750 if (cmd->bdev == bdev) {
751 atomic_inc(&cmd->ref_count);
758 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
759 sector_t data_block_size,
760 bool may_format_device,
761 size_t policy_hint_size)
763 struct dm_cache_metadata *cmd, *cmd2;
765 mutex_lock(&table_lock);
767 mutex_unlock(&table_lock);
772 cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
774 mutex_lock(&table_lock);
777 mutex_unlock(&table_lock);
778 __destroy_persistent_data_objects(cmd);
782 list_add(&cmd->list, &table);
783 mutex_unlock(&table_lock);
789 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
791 if (cmd->data_block_size != data_block_size) {
792 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
793 (unsigned long long) data_block_size,
794 (unsigned long long) cmd->data_block_size);
801 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
802 sector_t data_block_size,
803 bool may_format_device,
804 size_t policy_hint_size)
806 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
807 may_format_device, policy_hint_size);
809 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
810 dm_cache_metadata_close(cmd);
811 return ERR_PTR(-EINVAL);
817 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
819 if (atomic_dec_and_test(&cmd->ref_count)) {
820 mutex_lock(&table_lock);
821 list_del(&cmd->list);
822 mutex_unlock(&table_lock);
825 __destroy_persistent_data_objects(cmd);
831 * Checks that the given cache block is either unmapped or clean.
833 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
841 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
843 DMERR("block_unmapped_or_clean failed");
847 unpack_value(value, &ob, &flags);
848 *result = !((flags & M_VALID) && (flags & M_DIRTY));
853 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
854 dm_cblock_t begin, dm_cblock_t end,
860 while (begin != end) {
861 r = block_unmapped_or_clean(cmd, begin, result);
866 DMERR("cache block %llu is dirty",
867 (unsigned long long) from_cblock(begin));
871 begin = to_cblock(from_cblock(begin) + 1);
877 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
879 down_write(&cmd->root_lock);
880 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
881 up_write(&cmd->root_lock);
887 #define WRITE_LOCK(cmd) \
889 if (!cmd_write_lock((cmd))) \
893 #define WRITE_LOCK_VOID(cmd) \
895 if (!cmd_write_lock((cmd))) \
899 #define WRITE_UNLOCK(cmd) \
900 up_write(&(cmd)->root_lock)
902 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
904 down_read(&cmd->root_lock);
906 up_read(&cmd->root_lock);
912 #define READ_LOCK(cmd) \
914 if (!cmd_read_lock((cmd))) \
918 #define READ_LOCK_VOID(cmd) \
920 if (!cmd_read_lock((cmd))) \
924 #define READ_UNLOCK(cmd) \
925 up_read(&(cmd)->root_lock)
927 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
931 __le64 null_mapping = pack_value(0, 0);
934 __dm_bless_for_disk(&null_mapping);
936 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
937 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
939 __dm_unbless_for_disk(&null_mapping);
944 DMERR("unable to shrink cache due to dirty blocks");
946 __dm_unbless_for_disk(&null_mapping);
951 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
952 from_cblock(new_cache_size),
953 &null_mapping, &cmd->root);
955 cmd->cache_blocks = new_cache_size;
964 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
965 sector_t discard_block_size,
966 dm_dblock_t new_nr_entries)
971 r = dm_bitset_resize(&cmd->discard_info,
973 from_dblock(cmd->discard_nr_blocks),
974 from_dblock(new_nr_entries),
975 false, &cmd->discard_root);
977 cmd->discard_block_size = discard_block_size;
978 cmd->discard_nr_blocks = new_nr_entries;
987 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
989 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
990 from_dblock(b), &cmd->discard_root);
993 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
995 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
996 from_dblock(b), &cmd->discard_root);
999 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
1002 return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1003 from_dblock(b), &cmd->discard_root,
1007 static int __discard(struct dm_cache_metadata *cmd,
1008 dm_dblock_t dblock, bool discard)
1012 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1016 cmd->changed = true;
1020 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1021 dm_dblock_t dblock, bool discard)
1026 r = __discard(cmd, dblock, discard);
1032 static int __load_discards(struct dm_cache_metadata *cmd,
1033 load_discard_fn fn, void *context)
1039 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1040 dm_dblock_t dblock = to_dblock(b);
1042 if (cmd->clean_when_opened) {
1043 r = __is_discarded(cmd, dblock, &discard);
1049 r = fn(context, cmd->discard_block_size, dblock, discard);
1057 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1058 load_discard_fn fn, void *context)
1063 r = __load_discards(cmd, fn, context);
1069 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1072 *result = cmd->cache_blocks;
1078 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1081 __le64 value = pack_value(0, 0);
1083 __dm_bless_for_disk(&value);
1084 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1085 &value, &cmd->root);
1089 cmd->changed = true;
1093 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1098 r = __remove(cmd, cblock);
1104 static int __insert(struct dm_cache_metadata *cmd,
1105 dm_cblock_t cblock, dm_oblock_t oblock)
1108 __le64 value = pack_value(oblock, M_VALID);
1109 __dm_bless_for_disk(&value);
1111 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1112 &value, &cmd->root);
1116 cmd->changed = true;
1120 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1121 dm_cblock_t cblock, dm_oblock_t oblock)
1126 r = __insert(cmd, cblock, oblock);
1136 struct dm_cache_metadata *cmd;
1137 bool respect_dirty_flags;
1141 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1142 struct dm_cache_policy *policy)
1144 const char *policy_name = dm_cache_policy_get_name(policy);
1145 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1146 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1149 * Ensure policy names match.
1151 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1155 * Ensure policy major versions match.
1157 if (cmd->policy_version[0] != policy_version[0])
1161 * Ensure policy hint sizes match.
1163 if (cmd->policy_hint_size != policy_hint_size)
1169 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1171 return cmd->hint_root && cmd->policy_hint_size;
1174 static bool hints_array_available(struct dm_cache_metadata *cmd,
1175 struct dm_cache_policy *policy)
1177 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1178 hints_array_initialized(cmd);
1181 static int __load_mapping(struct dm_cache_metadata *cmd,
1182 uint64_t cb, bool hints_valid,
1183 struct dm_array_cursor *mapping_cursor,
1184 struct dm_array_cursor *hint_cursor,
1185 load_mapping_fn fn, void *context)
1192 __le64 *mapping_value_le;
1193 __le32 *hint_value_le;
1198 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1199 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1200 unpack_value(mapping, &oblock, &flags);
1202 if (flags & M_VALID) {
1204 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1205 memcpy(&hint, hint_value_le, sizeof(hint));
1208 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1209 le32_to_cpu(hint), hints_valid);
1211 DMERR("policy couldn't load cblock");
1217 static int __load_mappings(struct dm_cache_metadata *cmd,
1218 struct dm_cache_policy *policy,
1219 load_mapping_fn fn, void *context)
1224 bool hints_valid = hints_array_available(cmd, policy);
1226 if (from_cblock(cmd->cache_blocks) == 0)
1230 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1235 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1237 dm_array_cursor_end(&cmd->mapping_cursor);
1242 for (cb = 0; ; cb++) {
1243 r = __load_mapping(cmd, cb, hints_valid,
1244 &cmd->mapping_cursor, &cmd->hint_cursor,
1250 * We need to break out before we move the cursors.
1252 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1255 r = dm_array_cursor_next(&cmd->mapping_cursor);
1257 DMERR("dm_array_cursor_next for mapping failed");
1262 r = dm_array_cursor_next(&cmd->hint_cursor);
1264 DMERR("dm_array_cursor_next for hint failed");
1270 dm_array_cursor_end(&cmd->mapping_cursor);
1272 dm_array_cursor_end(&cmd->hint_cursor);
1277 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1278 struct dm_cache_policy *policy,
1279 load_mapping_fn fn, void *context)
1284 r = __load_mappings(cmd, policy, fn, context);
1290 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1297 memcpy(&value, leaf, sizeof(value));
1298 unpack_value(value, &oblock, &flags);
1303 static int __dump_mappings(struct dm_cache_metadata *cmd)
1305 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1308 void dm_cache_dump(struct dm_cache_metadata *cmd)
1310 READ_LOCK_VOID(cmd);
1311 __dump_mappings(cmd);
1315 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1326 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1333 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1337 unpack_value(value, &oblock, &flags);
1339 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1340 /* nothing to be done */
1343 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1344 __dm_bless_for_disk(&value);
1346 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1347 &value, &cmd->root);
1351 cmd->changed = true;
1356 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1357 dm_cblock_t cblock, bool dirty)
1362 r = __dirty(cmd, cblock, dirty);
1368 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1369 struct dm_cache_statistics *stats)
1371 READ_LOCK_VOID(cmd);
1372 *stats = cmd->stats;
1376 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1377 struct dm_cache_statistics *stats)
1379 WRITE_LOCK_VOID(cmd);
1380 cmd->stats = *stats;
1384 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1387 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1388 clear_clean_shutdown);
1391 r = __commit_transaction(cmd, mutator);
1395 r = __begin_transaction(cmd);
1402 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1408 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1414 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1420 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1426 /*----------------------------------------------------------------*/
1428 static int get_hint(uint32_t index, void *value_le, void *context)
1431 struct dm_cache_policy *policy = context;
1433 value = policy_get_hint(policy, to_cblock(index));
1434 *((__le32 *) value_le) = cpu_to_le32(value);
1440 * It's quicker to always delete the hint array, and recreate with
1443 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1447 const char *policy_name = dm_cache_policy_get_name(policy);
1448 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1450 if (!policy_name[0] ||
1451 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1454 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1455 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1457 hint_size = dm_cache_policy_get_hint_size(policy);
1459 return 0; /* short-circuit hints initialization */
1460 cmd->policy_hint_size = hint_size;
1462 if (cmd->hint_root) {
1463 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1468 return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1469 from_cblock(cmd->cache_blocks),
1473 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1478 r = write_hints(cmd, policy);
1484 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1489 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1495 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1497 WRITE_LOCK_VOID(cmd);
1498 dm_bm_set_read_only(cmd->bm);
1502 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1504 WRITE_LOCK_VOID(cmd);
1505 dm_bm_set_read_write(cmd->bm);
1509 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1512 struct dm_block *sblock;
1513 struct cache_disk_superblock *disk_super;
1516 set_bit(NEEDS_CHECK, &cmd->flags);
1518 r = superblock_lock(cmd, &sblock);
1520 DMERR("couldn't read superblock");
1524 disk_super = dm_block_data(sblock);
1525 disk_super->flags = cpu_to_le32(cmd->flags);
1527 dm_bm_unlock(sblock);
1534 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1537 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1543 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1548 __destroy_persistent_data_objects(cmd);
1549 r = __create_persistent_data_objects(cmd, false);
1551 cmd->fail_io = true;