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 /*-------------------------------------------------------------------
146 * superblock validator
147 *-----------------------------------------------------------------*/
149 #define SUPERBLOCK_CSUM_XOR 9031977
151 static void sb_prepare_for_write(struct dm_block_validator *v,
153 size_t sb_block_size)
155 struct cache_disk_superblock *disk_super = dm_block_data(b);
157 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
158 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
159 sb_block_size - sizeof(__le32),
160 SUPERBLOCK_CSUM_XOR));
163 static int check_metadata_version(struct cache_disk_superblock *disk_super)
165 uint32_t metadata_version = le32_to_cpu(disk_super->version);
166 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
167 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
168 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
175 static int sb_check(struct dm_block_validator *v,
177 size_t sb_block_size)
179 struct cache_disk_superblock *disk_super = dm_block_data(b);
182 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
183 DMERR("sb_check failed: blocknr %llu: wanted %llu",
184 le64_to_cpu(disk_super->blocknr),
185 (unsigned long long)dm_block_location(b));
189 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
190 DMERR("sb_check failed: magic %llu: wanted %llu",
191 le64_to_cpu(disk_super->magic),
192 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
196 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
197 sb_block_size - sizeof(__le32),
198 SUPERBLOCK_CSUM_XOR));
199 if (csum_le != disk_super->csum) {
200 DMERR("sb_check failed: csum %u: wanted %u",
201 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
205 return check_metadata_version(disk_super);
208 static struct dm_block_validator sb_validator = {
209 .name = "superblock",
210 .prepare_for_write = sb_prepare_for_write,
214 /*----------------------------------------------------------------*/
216 static int superblock_read_lock(struct dm_cache_metadata *cmd,
217 struct dm_block **sblock)
219 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
220 &sb_validator, sblock);
223 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
224 struct dm_block **sblock)
226 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
227 &sb_validator, sblock);
230 static int superblock_lock(struct dm_cache_metadata *cmd,
231 struct dm_block **sblock)
233 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
234 &sb_validator, sblock);
237 /*----------------------------------------------------------------*/
239 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
244 __le64 *data_le, zero = cpu_to_le64(0);
245 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
248 * We can't use a validator here - it may be all zeroes.
250 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
254 data_le = dm_block_data(b);
256 for (i = 0; i < sb_block_size; i++) {
257 if (data_le[i] != zero) {
268 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
270 struct dm_btree_value_type vt;
273 vt.size = sizeof(__le64);
277 dm_array_info_init(&cmd->info, cmd->tm, &vt);
279 if (cmd->policy_hint_size) {
280 vt.size = sizeof(__le32);
281 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
285 static int __save_sm_root(struct dm_cache_metadata *cmd)
290 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
294 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
298 static void __copy_sm_root(struct dm_cache_metadata *cmd,
299 struct cache_disk_superblock *disk_super)
301 memcpy(&disk_super->metadata_space_map_root,
302 &cmd->metadata_space_map_root,
303 sizeof(cmd->metadata_space_map_root));
306 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
309 struct dm_block *sblock;
310 struct cache_disk_superblock *disk_super;
311 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
313 /* FIXME: see if we can lose the max sectors limit */
314 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
315 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
317 r = dm_tm_pre_commit(cmd->tm);
322 * dm_sm_copy_root() can fail. So we need to do it before we start
323 * updating the superblock.
325 r = __save_sm_root(cmd);
329 r = superblock_lock_zero(cmd, &sblock);
333 disk_super = dm_block_data(sblock);
334 disk_super->flags = 0;
335 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
336 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
337 disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
338 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
339 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
340 disk_super->policy_hint_size = 0;
342 __copy_sm_root(cmd, disk_super);
344 disk_super->mapping_root = cpu_to_le64(cmd->root);
345 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
346 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
347 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
348 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
349 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
350 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
351 disk_super->cache_blocks = cpu_to_le32(0);
353 disk_super->read_hits = cpu_to_le32(0);
354 disk_super->read_misses = cpu_to_le32(0);
355 disk_super->write_hits = cpu_to_le32(0);
356 disk_super->write_misses = cpu_to_le32(0);
358 return dm_tm_commit(cmd->tm, sblock);
361 static int __format_metadata(struct dm_cache_metadata *cmd)
365 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
366 &cmd->tm, &cmd->metadata_sm);
368 DMERR("tm_create_with_sm failed");
372 __setup_mapping_info(cmd);
374 r = dm_array_empty(&cmd->info, &cmd->root);
378 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
380 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
384 cmd->discard_block_size = 0;
385 cmd->discard_nr_blocks = 0;
387 r = __write_initial_superblock(cmd);
391 cmd->clean_when_opened = true;
395 dm_tm_destroy(cmd->tm);
396 dm_sm_destroy(cmd->metadata_sm);
401 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
402 struct dm_cache_metadata *cmd)
406 features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
408 DMERR("could not access metadata due to unsupported optional features (%lx).",
409 (unsigned long)features);
414 * Check for read-only metadata to skip the following RDWR checks.
416 if (get_disk_ro(cmd->bdev->bd_disk))
419 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
421 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
422 (unsigned long)features);
429 static int __open_metadata(struct dm_cache_metadata *cmd)
432 struct dm_block *sblock;
433 struct cache_disk_superblock *disk_super;
434 unsigned long sb_flags;
436 r = superblock_read_lock(cmd, &sblock);
438 DMERR("couldn't read lock superblock");
442 disk_super = dm_block_data(sblock);
444 /* Verify the data block size hasn't changed */
445 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
446 DMERR("changing the data block size (from %u to %llu) is not supported",
447 le32_to_cpu(disk_super->data_block_size),
448 (unsigned long long)cmd->data_block_size);
453 r = __check_incompat_features(disk_super, cmd);
457 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
458 disk_super->metadata_space_map_root,
459 sizeof(disk_super->metadata_space_map_root),
460 &cmd->tm, &cmd->metadata_sm);
462 DMERR("tm_open_with_sm failed");
466 __setup_mapping_info(cmd);
467 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
468 sb_flags = le32_to_cpu(disk_super->flags);
469 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
470 dm_bm_unlock(sblock);
475 dm_bm_unlock(sblock);
479 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
483 bool unformatted = false;
485 r = __superblock_all_zeroes(cmd->bm, &unformatted);
490 return format_device ? __format_metadata(cmd) : -EPERM;
492 return __open_metadata(cmd);
495 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
496 bool may_format_device)
499 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
500 CACHE_METADATA_CACHE_SIZE,
501 CACHE_MAX_CONCURRENT_LOCKS);
502 if (IS_ERR(cmd->bm)) {
503 DMERR("could not create block manager");
504 return PTR_ERR(cmd->bm);
507 r = __open_or_format_metadata(cmd, may_format_device);
509 dm_block_manager_destroy(cmd->bm);
514 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
516 dm_sm_destroy(cmd->metadata_sm);
517 dm_tm_destroy(cmd->tm);
518 dm_block_manager_destroy(cmd->bm);
521 typedef unsigned long (*flags_mutator)(unsigned long);
523 static void update_flags(struct cache_disk_superblock *disk_super,
524 flags_mutator mutator)
526 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
527 disk_super->flags = cpu_to_le32(sb_flags);
530 static unsigned long set_clean_shutdown(unsigned long flags)
532 set_bit(CLEAN_SHUTDOWN, &flags);
536 static unsigned long clear_clean_shutdown(unsigned long flags)
538 clear_bit(CLEAN_SHUTDOWN, &flags);
542 static void read_superblock_fields(struct dm_cache_metadata *cmd,
543 struct cache_disk_superblock *disk_super)
545 cmd->flags = le32_to_cpu(disk_super->flags);
546 cmd->root = le64_to_cpu(disk_super->mapping_root);
547 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
548 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
549 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
550 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
551 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
552 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
553 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
554 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
555 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
556 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
557 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
559 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
560 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
561 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
562 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
564 cmd->changed = false;
568 * The mutator updates the superblock flags.
570 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
571 flags_mutator mutator)
574 struct cache_disk_superblock *disk_super;
575 struct dm_block *sblock;
577 r = superblock_lock(cmd, &sblock);
581 disk_super = dm_block_data(sblock);
582 update_flags(disk_super, mutator);
583 read_superblock_fields(cmd, disk_super);
584 dm_bm_unlock(sblock);
586 return dm_bm_flush(cmd->bm);
589 static int __begin_transaction(struct dm_cache_metadata *cmd)
592 struct cache_disk_superblock *disk_super;
593 struct dm_block *sblock;
596 * We re-read the superblock every time. Shouldn't need to do this
599 r = superblock_read_lock(cmd, &sblock);
603 disk_super = dm_block_data(sblock);
604 read_superblock_fields(cmd, disk_super);
605 dm_bm_unlock(sblock);
610 static int __commit_transaction(struct dm_cache_metadata *cmd,
611 flags_mutator mutator)
614 struct cache_disk_superblock *disk_super;
615 struct dm_block *sblock;
618 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
620 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
622 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
627 r = dm_tm_pre_commit(cmd->tm);
631 r = __save_sm_root(cmd);
635 r = superblock_lock(cmd, &sblock);
639 disk_super = dm_block_data(sblock);
641 disk_super->flags = cpu_to_le32(cmd->flags);
643 update_flags(disk_super, mutator);
645 disk_super->mapping_root = cpu_to_le64(cmd->root);
646 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
647 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
648 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
649 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
650 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
651 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
652 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
653 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
654 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
656 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
657 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
658 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
659 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
660 __copy_sm_root(cmd, disk_super);
662 return dm_tm_commit(cmd->tm, sblock);
665 /*----------------------------------------------------------------*/
668 * The mappings are held in a dm-array that has 64-bit values stored in
669 * little-endian format. The index is the cblock, the high 48bits of the
670 * value are the oblock and the low 16 bit the flags.
672 #define FLAGS_MASK ((1 << 16) - 1)
674 static __le64 pack_value(dm_oblock_t block, unsigned flags)
676 uint64_t value = from_oblock(block);
678 value = value | (flags & FLAGS_MASK);
679 return cpu_to_le64(value);
682 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
684 uint64_t value = le64_to_cpu(value_le);
685 uint64_t b = value >> 16;
686 *block = to_oblock(b);
687 *flags = value & FLAGS_MASK;
690 /*----------------------------------------------------------------*/
692 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
693 sector_t data_block_size,
694 bool may_format_device,
695 size_t policy_hint_size)
698 struct dm_cache_metadata *cmd;
700 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
702 DMERR("could not allocate metadata struct");
703 return ERR_PTR(-ENOMEM);
706 atomic_set(&cmd->ref_count, 1);
707 init_rwsem(&cmd->root_lock);
709 cmd->data_block_size = data_block_size;
710 cmd->cache_blocks = 0;
711 cmd->policy_hint_size = policy_hint_size;
713 cmd->fail_io = false;
715 r = __create_persistent_data_objects(cmd, may_format_device);
721 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
723 dm_cache_metadata_close(cmd);
731 * We keep a little list of ref counted metadata objects to prevent two
732 * different target instances creating separate bufio instances. This is
733 * an issue if a table is reloaded before the suspend.
735 static DEFINE_MUTEX(table_lock);
736 static LIST_HEAD(table);
738 static struct dm_cache_metadata *lookup(struct block_device *bdev)
740 struct dm_cache_metadata *cmd;
742 list_for_each_entry(cmd, &table, list)
743 if (cmd->bdev == bdev) {
744 atomic_inc(&cmd->ref_count);
751 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
752 sector_t data_block_size,
753 bool may_format_device,
754 size_t policy_hint_size)
756 struct dm_cache_metadata *cmd, *cmd2;
758 mutex_lock(&table_lock);
760 mutex_unlock(&table_lock);
765 cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
767 mutex_lock(&table_lock);
770 mutex_unlock(&table_lock);
771 __destroy_persistent_data_objects(cmd);
775 list_add(&cmd->list, &table);
776 mutex_unlock(&table_lock);
782 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
784 if (cmd->data_block_size != data_block_size) {
785 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
786 (unsigned long long) data_block_size,
787 (unsigned long long) cmd->data_block_size);
794 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
795 sector_t data_block_size,
796 bool may_format_device,
797 size_t policy_hint_size)
799 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
800 may_format_device, policy_hint_size);
802 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
803 dm_cache_metadata_close(cmd);
804 return ERR_PTR(-EINVAL);
810 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
812 if (atomic_dec_and_test(&cmd->ref_count)) {
813 mutex_lock(&table_lock);
814 list_del(&cmd->list);
815 mutex_unlock(&table_lock);
818 __destroy_persistent_data_objects(cmd);
824 * Checks that the given cache block is either unmapped or clean.
826 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
834 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
836 DMERR("block_unmapped_or_clean failed");
840 unpack_value(value, &ob, &flags);
841 *result = !((flags & M_VALID) && (flags & M_DIRTY));
846 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
847 dm_cblock_t begin, dm_cblock_t end,
853 while (begin != end) {
854 r = block_unmapped_or_clean(cmd, begin, result);
859 DMERR("cache block %llu is dirty",
860 (unsigned long long) from_cblock(begin));
864 begin = to_cblock(from_cblock(begin) + 1);
870 #define WRITE_LOCK(cmd) \
871 down_write(&cmd->root_lock); \
872 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) { \
873 up_write(&cmd->root_lock); \
877 #define WRITE_LOCK_VOID(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); \
884 #define WRITE_UNLOCK(cmd) \
885 up_write(&cmd->root_lock)
887 #define READ_LOCK(cmd) \
888 down_read(&cmd->root_lock); \
889 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) { \
890 up_read(&cmd->root_lock); \
894 #define READ_LOCK_VOID(cmd) \
895 down_read(&cmd->root_lock); \
896 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) { \
897 up_read(&cmd->root_lock); \
901 #define READ_UNLOCK(cmd) \
902 up_read(&cmd->root_lock)
904 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
908 __le64 null_mapping = pack_value(0, 0);
911 __dm_bless_for_disk(&null_mapping);
913 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
914 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
916 __dm_unbless_for_disk(&null_mapping);
921 DMERR("unable to shrink cache due to dirty blocks");
923 __dm_unbless_for_disk(&null_mapping);
928 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
929 from_cblock(new_cache_size),
930 &null_mapping, &cmd->root);
932 cmd->cache_blocks = new_cache_size;
941 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
942 sector_t discard_block_size,
943 dm_dblock_t new_nr_entries)
948 r = dm_bitset_resize(&cmd->discard_info,
950 from_dblock(cmd->discard_nr_blocks),
951 from_dblock(new_nr_entries),
952 false, &cmd->discard_root);
954 cmd->discard_block_size = discard_block_size;
955 cmd->discard_nr_blocks = new_nr_entries;
964 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
966 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
967 from_dblock(b), &cmd->discard_root);
970 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
972 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
973 from_dblock(b), &cmd->discard_root);
976 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
979 return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
980 from_dblock(b), &cmd->discard_root,
984 static int __discard(struct dm_cache_metadata *cmd,
985 dm_dblock_t dblock, bool discard)
989 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
997 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
998 dm_dblock_t dblock, bool discard)
1003 r = __discard(cmd, dblock, discard);
1009 static int __load_discards(struct dm_cache_metadata *cmd,
1010 load_discard_fn fn, void *context)
1016 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1017 dm_dblock_t dblock = to_dblock(b);
1019 if (cmd->clean_when_opened) {
1020 r = __is_discarded(cmd, dblock, &discard);
1026 r = fn(context, cmd->discard_block_size, dblock, discard);
1034 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1035 load_discard_fn fn, void *context)
1040 r = __load_discards(cmd, fn, context);
1046 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1049 *result = cmd->cache_blocks;
1055 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1058 __le64 value = pack_value(0, 0);
1060 __dm_bless_for_disk(&value);
1061 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1062 &value, &cmd->root);
1066 cmd->changed = true;
1070 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1075 r = __remove(cmd, cblock);
1081 static int __insert(struct dm_cache_metadata *cmd,
1082 dm_cblock_t cblock, dm_oblock_t oblock)
1085 __le64 value = pack_value(oblock, M_VALID);
1086 __dm_bless_for_disk(&value);
1088 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1089 &value, &cmd->root);
1093 cmd->changed = true;
1097 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1098 dm_cblock_t cblock, dm_oblock_t oblock)
1103 r = __insert(cmd, cblock, oblock);
1113 struct dm_cache_metadata *cmd;
1114 bool respect_dirty_flags;
1118 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1119 struct dm_cache_policy *policy)
1121 const char *policy_name = dm_cache_policy_get_name(policy);
1122 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1123 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1126 * Ensure policy names match.
1128 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1132 * Ensure policy major versions match.
1134 if (cmd->policy_version[0] != policy_version[0])
1138 * Ensure policy hint sizes match.
1140 if (cmd->policy_hint_size != policy_hint_size)
1146 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1148 return cmd->hint_root && cmd->policy_hint_size;
1151 static bool hints_array_available(struct dm_cache_metadata *cmd,
1152 struct dm_cache_policy *policy)
1154 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1155 hints_array_initialized(cmd);
1158 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1163 __le32 hint_value = 0;
1166 struct thunk *thunk = context;
1167 struct dm_cache_metadata *cmd = thunk->cmd;
1169 memcpy(&value, leaf, sizeof(value));
1170 unpack_value(value, &oblock, &flags);
1172 if (flags & M_VALID) {
1173 if (thunk->hints_valid) {
1174 r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1175 cblock, &hint_value);
1176 if (r && r != -ENODATA)
1180 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1181 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1182 dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1188 static int __load_mappings(struct dm_cache_metadata *cmd,
1189 struct dm_cache_policy *policy,
1190 load_mapping_fn fn, void *context)
1195 thunk.context = context;
1198 thunk.respect_dirty_flags = cmd->clean_when_opened;
1199 thunk.hints_valid = hints_array_available(cmd, policy);
1201 return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1204 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1205 struct dm_cache_policy *policy,
1206 load_mapping_fn fn, void *context)
1211 r = __load_mappings(cmd, policy, fn, context);
1217 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1224 memcpy(&value, leaf, sizeof(value));
1225 unpack_value(value, &oblock, &flags);
1230 static int __dump_mappings(struct dm_cache_metadata *cmd)
1232 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1235 void dm_cache_dump(struct dm_cache_metadata *cmd)
1237 READ_LOCK_VOID(cmd);
1238 __dump_mappings(cmd);
1242 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1253 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1260 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1264 unpack_value(value, &oblock, &flags);
1266 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1267 /* nothing to be done */
1270 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1271 __dm_bless_for_disk(&value);
1273 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1274 &value, &cmd->root);
1278 cmd->changed = true;
1283 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1284 dm_cblock_t cblock, bool dirty)
1289 r = __dirty(cmd, cblock, dirty);
1295 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1296 struct dm_cache_statistics *stats)
1298 READ_LOCK_VOID(cmd);
1299 *stats = cmd->stats;
1303 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1304 struct dm_cache_statistics *stats)
1306 WRITE_LOCK_VOID(cmd);
1307 cmd->stats = *stats;
1311 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1314 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1315 clear_clean_shutdown);
1318 r = __commit_transaction(cmd, mutator);
1322 r = __begin_transaction(cmd);
1329 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1335 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1341 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1347 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1353 /*----------------------------------------------------------------*/
1355 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1360 const char *policy_name = dm_cache_policy_get_name(policy);
1361 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1363 if (!policy_name[0] ||
1364 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1367 if (!policy_unchanged(cmd, policy)) {
1368 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1369 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1371 hint_size = dm_cache_policy_get_hint_size(policy);
1373 return 0; /* short-circuit hints initialization */
1374 cmd->policy_hint_size = hint_size;
1376 if (cmd->hint_root) {
1377 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1382 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1386 value = cpu_to_le32(0);
1387 __dm_bless_for_disk(&value);
1388 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1389 from_cblock(cmd->cache_blocks),
1390 &value, &cmd->hint_root);
1398 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1400 struct dm_cache_metadata *cmd = context;
1401 __le32 value = cpu_to_le32(hint);
1404 __dm_bless_for_disk(&value);
1406 r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1407 from_cblock(cblock), &value, &cmd->hint_root);
1408 cmd->changed = true;
1413 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1417 r = begin_hints(cmd, policy);
1419 DMERR("begin_hints failed");
1423 return policy_walk_mappings(policy, save_hint, cmd);
1426 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1431 r = write_hints(cmd, policy);
1437 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1442 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1448 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1450 WRITE_LOCK_VOID(cmd);
1451 dm_bm_set_read_only(cmd->bm);
1455 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1457 WRITE_LOCK_VOID(cmd);
1458 dm_bm_set_read_write(cmd->bm);
1462 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1465 struct dm_block *sblock;
1466 struct cache_disk_superblock *disk_super;
1469 set_bit(NEEDS_CHECK, &cmd->flags);
1471 r = superblock_lock(cmd, &sblock);
1473 DMERR("couldn't read superblock");
1477 disk_super = dm_block_data(sblock);
1478 disk_super->flags = cpu_to_le32(cmd->flags);
1480 dm_bm_unlock(sblock);
1487 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1490 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1496 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1501 __destroy_persistent_data_objects(cmd);
1502 r = __create_persistent_data_objects(cmd, false);
1504 cmd->fail_io = true;