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>
16 #include <linux/refcount.h>
18 /*----------------------------------------------------------------*/
20 #define DM_MSG_PREFIX "cache metadata"
22 #define CACHE_SUPERBLOCK_MAGIC 06142003
23 #define CACHE_SUPERBLOCK_LOCATION 0
26 * defines a range of metadata versions that this module can handle.
28 #define MIN_CACHE_VERSION 1
29 #define MAX_CACHE_VERSION 2
32 * 3 for btree insert +
33 * 2 for btree lookup used within space map
35 #define CACHE_MAX_CONCURRENT_LOCKS 5
36 #define SPACE_MAP_ROOT_SIZE 128
38 enum superblock_flag_bits {
39 /* for spotting crashes that would invalidate the dirty bitset */
41 /* metadata must be checked using the tools */
46 * Each mapping from cache block -> origin block carries a set of flags.
50 * A valid mapping. Because we're using an array we clear this
51 * flag for an non existant mapping.
56 * The data on the cache is different from that on the origin.
57 * This flag is only used by metadata format 1.
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 * Metadata format 2 fields.
103 struct dm_cache_metadata {
104 refcount_t ref_count;
105 struct list_head list;
107 unsigned int version;
108 struct block_device *bdev;
109 struct dm_block_manager *bm;
110 struct dm_space_map *metadata_sm;
111 struct dm_transaction_manager *tm;
113 struct dm_array_info info;
114 struct dm_array_info hint_info;
115 struct dm_disk_bitset discard_info;
117 struct rw_semaphore root_lock;
120 dm_block_t hint_root;
121 dm_block_t discard_root;
123 sector_t discard_block_size;
124 dm_dblock_t discard_nr_blocks;
126 sector_t data_block_size;
127 dm_cblock_t cache_blocks;
129 bool clean_when_opened:1;
131 char policy_name[CACHE_POLICY_NAME_SIZE];
132 unsigned int policy_version[CACHE_POLICY_VERSION_SIZE];
133 size_t policy_hint_size;
134 struct dm_cache_statistics stats;
137 * Reading the space map root can fail, so we read it into this
138 * buffer before the superblock is locked and updated.
140 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
143 * Set if a transaction has to be aborted but the attempt to roll
144 * back to the previous (good) transaction failed. The only
145 * metadata operation permissible in this state is the closing of
151 * Metadata format 2 fields.
153 dm_block_t dirty_root;
154 struct dm_disk_bitset dirty_info;
157 * These structures are used when loading metadata. They're too
158 * big to put on the stack.
160 struct dm_array_cursor mapping_cursor;
161 struct dm_array_cursor hint_cursor;
162 struct dm_bitset_cursor dirty_cursor;
165 /*-------------------------------------------------------------------
166 * superblock validator
167 *-----------------------------------------------------------------*/
169 #define SUPERBLOCK_CSUM_XOR 9031977
171 static void sb_prepare_for_write(struct dm_block_validator *v,
173 size_t sb_block_size)
175 struct cache_disk_superblock *disk_super = dm_block_data(b);
177 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
178 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
179 sb_block_size - sizeof(__le32),
180 SUPERBLOCK_CSUM_XOR));
183 static int check_metadata_version(struct cache_disk_superblock *disk_super)
185 uint32_t metadata_version = le32_to_cpu(disk_super->version);
187 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
188 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
189 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
196 static int sb_check(struct dm_block_validator *v,
198 size_t sb_block_size)
200 struct cache_disk_superblock *disk_super = dm_block_data(b);
203 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
204 DMERR("sb_check failed: blocknr %llu: wanted %llu",
205 le64_to_cpu(disk_super->blocknr),
206 (unsigned long long)dm_block_location(b));
210 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
211 DMERR("sb_check failed: magic %llu: wanted %llu",
212 le64_to_cpu(disk_super->magic),
213 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
217 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
218 sb_block_size - sizeof(__le32),
219 SUPERBLOCK_CSUM_XOR));
220 if (csum_le != disk_super->csum) {
221 DMERR("sb_check failed: csum %u: wanted %u",
222 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
226 return check_metadata_version(disk_super);
229 static struct dm_block_validator sb_validator = {
230 .name = "superblock",
231 .prepare_for_write = sb_prepare_for_write,
235 /*----------------------------------------------------------------*/
237 static int superblock_read_lock(struct dm_cache_metadata *cmd,
238 struct dm_block **sblock)
240 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
241 &sb_validator, sblock);
244 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
245 struct dm_block **sblock)
247 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
248 &sb_validator, sblock);
251 static int superblock_lock(struct dm_cache_metadata *cmd,
252 struct dm_block **sblock)
254 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
255 &sb_validator, sblock);
258 /*----------------------------------------------------------------*/
260 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
265 __le64 *data_le, zero = cpu_to_le64(0);
266 unsigned int sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
269 * We can't use a validator here - it may be all zeroes.
271 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
275 data_le = dm_block_data(b);
277 for (i = 0; i < sb_block_size; i++) {
278 if (data_le[i] != zero) {
289 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
291 struct dm_btree_value_type vt;
294 vt.size = sizeof(__le64);
298 dm_array_info_init(&cmd->info, cmd->tm, &vt);
300 if (cmd->policy_hint_size) {
301 vt.size = sizeof(__le32);
302 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
306 static int __save_sm_root(struct dm_cache_metadata *cmd)
311 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
315 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
319 static void __copy_sm_root(struct dm_cache_metadata *cmd,
320 struct cache_disk_superblock *disk_super)
322 memcpy(&disk_super->metadata_space_map_root,
323 &cmd->metadata_space_map_root,
324 sizeof(cmd->metadata_space_map_root));
327 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
329 return cmd->version >= 2;
332 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
335 struct dm_block *sblock;
336 struct cache_disk_superblock *disk_super;
337 sector_t bdev_size = bdev_nr_sectors(cmd->bdev);
339 /* FIXME: see if we can lose the max sectors limit */
340 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
341 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
343 r = dm_tm_pre_commit(cmd->tm);
348 * dm_sm_copy_root() can fail. So we need to do it before we start
349 * updating the superblock.
351 r = __save_sm_root(cmd);
355 r = superblock_lock_zero(cmd, &sblock);
359 disk_super = dm_block_data(sblock);
360 disk_super->flags = 0;
361 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
362 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
363 disk_super->version = cpu_to_le32(cmd->version);
364 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
365 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
366 disk_super->policy_hint_size = cpu_to_le32(0);
368 __copy_sm_root(cmd, disk_super);
370 disk_super->mapping_root = cpu_to_le64(cmd->root);
371 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
372 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
373 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
374 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
375 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
376 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
377 disk_super->cache_blocks = cpu_to_le32(0);
379 disk_super->read_hits = cpu_to_le32(0);
380 disk_super->read_misses = cpu_to_le32(0);
381 disk_super->write_hits = cpu_to_le32(0);
382 disk_super->write_misses = cpu_to_le32(0);
384 if (separate_dirty_bits(cmd))
385 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
387 return dm_tm_commit(cmd->tm, sblock);
390 static int __format_metadata(struct dm_cache_metadata *cmd)
394 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
395 &cmd->tm, &cmd->metadata_sm);
397 DMERR("tm_create_with_sm failed");
401 __setup_mapping_info(cmd);
403 r = dm_array_empty(&cmd->info, &cmd->root);
407 if (separate_dirty_bits(cmd)) {
408 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
409 r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
414 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
415 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
419 cmd->discard_block_size = 0;
420 cmd->discard_nr_blocks = 0;
422 r = __write_initial_superblock(cmd);
426 cmd->clean_when_opened = true;
430 dm_tm_destroy(cmd->tm);
431 dm_sm_destroy(cmd->metadata_sm);
436 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
437 struct dm_cache_metadata *cmd)
439 uint32_t incompat_flags, features;
441 incompat_flags = le32_to_cpu(disk_super->incompat_flags);
442 features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
444 DMERR("could not access metadata due to unsupported optional features (%lx).",
445 (unsigned long)features);
450 * Check for read-only metadata to skip the following RDWR checks.
452 if (bdev_read_only(cmd->bdev))
455 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
457 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
458 (unsigned long)features);
465 static int __open_metadata(struct dm_cache_metadata *cmd)
468 struct dm_block *sblock;
469 struct cache_disk_superblock *disk_super;
470 unsigned long sb_flags;
472 r = superblock_read_lock(cmd, &sblock);
474 DMERR("couldn't read lock superblock");
478 disk_super = dm_block_data(sblock);
480 /* Verify the data block size hasn't changed */
481 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
482 DMERR("changing the data block size (from %u to %llu) is not supported",
483 le32_to_cpu(disk_super->data_block_size),
484 (unsigned long long)cmd->data_block_size);
489 r = __check_incompat_features(disk_super, cmd);
493 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
494 disk_super->metadata_space_map_root,
495 sizeof(disk_super->metadata_space_map_root),
496 &cmd->tm, &cmd->metadata_sm);
498 DMERR("tm_open_with_sm failed");
502 __setup_mapping_info(cmd);
503 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
504 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
505 sb_flags = le32_to_cpu(disk_super->flags);
506 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
507 dm_bm_unlock(sblock);
512 dm_bm_unlock(sblock);
516 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
520 bool unformatted = false;
522 r = __superblock_all_zeroes(cmd->bm, &unformatted);
527 return format_device ? __format_metadata(cmd) : -EPERM;
529 return __open_metadata(cmd);
532 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
533 bool may_format_device)
536 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
537 CACHE_MAX_CONCURRENT_LOCKS);
538 if (IS_ERR(cmd->bm)) {
539 DMERR("could not create block manager");
540 r = PTR_ERR(cmd->bm);
545 r = __open_or_format_metadata(cmd, may_format_device);
547 dm_block_manager_destroy(cmd->bm);
554 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd,
557 dm_sm_destroy(cmd->metadata_sm);
558 dm_tm_destroy(cmd->tm);
560 dm_block_manager_destroy(cmd->bm);
563 typedef unsigned long (*flags_mutator)(unsigned long);
565 static void update_flags(struct cache_disk_superblock *disk_super,
566 flags_mutator mutator)
568 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
569 disk_super->flags = cpu_to_le32(sb_flags);
572 static unsigned long set_clean_shutdown(unsigned long flags)
574 set_bit(CLEAN_SHUTDOWN, &flags);
578 static unsigned long clear_clean_shutdown(unsigned long flags)
580 clear_bit(CLEAN_SHUTDOWN, &flags);
584 static void read_superblock_fields(struct dm_cache_metadata *cmd,
585 struct cache_disk_superblock *disk_super)
587 cmd->version = le32_to_cpu(disk_super->version);
588 cmd->flags = le32_to_cpu(disk_super->flags);
589 cmd->root = le64_to_cpu(disk_super->mapping_root);
590 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
591 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
592 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
593 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
594 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
595 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
596 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
597 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
598 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
599 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
600 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
602 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
603 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
604 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
605 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
607 if (separate_dirty_bits(cmd))
608 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
610 cmd->changed = false;
614 * The mutator updates the superblock flags.
616 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
617 flags_mutator mutator)
620 struct cache_disk_superblock *disk_super;
621 struct dm_block *sblock;
623 r = superblock_lock(cmd, &sblock);
627 disk_super = dm_block_data(sblock);
628 update_flags(disk_super, mutator);
629 read_superblock_fields(cmd, disk_super);
630 dm_bm_unlock(sblock);
632 return dm_bm_flush(cmd->bm);
635 static int __begin_transaction(struct dm_cache_metadata *cmd)
638 struct cache_disk_superblock *disk_super;
639 struct dm_block *sblock;
642 * We re-read the superblock every time. Shouldn't need to do this
645 r = superblock_read_lock(cmd, &sblock);
649 disk_super = dm_block_data(sblock);
650 read_superblock_fields(cmd, disk_super);
651 dm_bm_unlock(sblock);
656 static int __commit_transaction(struct dm_cache_metadata *cmd,
657 flags_mutator mutator)
660 struct cache_disk_superblock *disk_super;
661 struct dm_block *sblock;
664 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
666 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
668 if (separate_dirty_bits(cmd)) {
669 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
675 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
680 r = dm_tm_pre_commit(cmd->tm);
684 r = __save_sm_root(cmd);
688 r = superblock_lock(cmd, &sblock);
692 disk_super = dm_block_data(sblock);
694 disk_super->flags = cpu_to_le32(cmd->flags);
696 update_flags(disk_super, mutator);
698 disk_super->mapping_root = cpu_to_le64(cmd->root);
699 if (separate_dirty_bits(cmd))
700 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
701 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
702 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
703 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
704 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
705 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
706 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
707 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
708 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
709 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
710 disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
712 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
713 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
714 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
715 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
716 __copy_sm_root(cmd, disk_super);
718 return dm_tm_commit(cmd->tm, sblock);
721 /*----------------------------------------------------------------*/
724 * The mappings are held in a dm-array that has 64-bit values stored in
725 * little-endian format. The index is the cblock, the high 48bits of the
726 * value are the oblock and the low 16 bit the flags.
728 #define FLAGS_MASK ((1 << 16) - 1)
730 static __le64 pack_value(dm_oblock_t block, unsigned int flags)
732 uint64_t value = from_oblock(block);
734 value = value | (flags & FLAGS_MASK);
735 return cpu_to_le64(value);
738 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned int *flags)
740 uint64_t value = le64_to_cpu(value_le);
741 uint64_t b = value >> 16;
742 *block = to_oblock(b);
743 *flags = value & FLAGS_MASK;
746 /*----------------------------------------------------------------*/
748 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
749 sector_t data_block_size,
750 bool may_format_device,
751 size_t policy_hint_size,
752 unsigned int metadata_version)
755 struct dm_cache_metadata *cmd;
757 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
759 DMERR("could not allocate metadata struct");
760 return ERR_PTR(-ENOMEM);
763 cmd->version = metadata_version;
764 refcount_set(&cmd->ref_count, 1);
765 init_rwsem(&cmd->root_lock);
767 cmd->data_block_size = data_block_size;
768 cmd->cache_blocks = 0;
769 cmd->policy_hint_size = policy_hint_size;
771 cmd->fail_io = false;
773 r = __create_persistent_data_objects(cmd, may_format_device);
779 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
781 dm_cache_metadata_close(cmd);
789 * We keep a little list of ref counted metadata objects to prevent two
790 * different target instances creating separate bufio instances. This is
791 * an issue if a table is reloaded before the suspend.
793 static DEFINE_MUTEX(table_lock);
794 static LIST_HEAD(table);
796 static struct dm_cache_metadata *lookup(struct block_device *bdev)
798 struct dm_cache_metadata *cmd;
800 list_for_each_entry(cmd, &table, list)
801 if (cmd->bdev == bdev) {
802 refcount_inc(&cmd->ref_count);
809 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
810 sector_t data_block_size,
811 bool may_format_device,
812 size_t policy_hint_size,
813 unsigned int metadata_version)
815 struct dm_cache_metadata *cmd, *cmd2;
817 mutex_lock(&table_lock);
819 mutex_unlock(&table_lock);
824 cmd = metadata_open(bdev, data_block_size, may_format_device,
825 policy_hint_size, metadata_version);
827 mutex_lock(&table_lock);
830 mutex_unlock(&table_lock);
831 __destroy_persistent_data_objects(cmd, true);
835 list_add(&cmd->list, &table);
836 mutex_unlock(&table_lock);
842 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
844 if (cmd->data_block_size != data_block_size) {
845 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
846 (unsigned long long) data_block_size,
847 (unsigned long long) cmd->data_block_size);
854 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
855 sector_t data_block_size,
856 bool may_format_device,
857 size_t policy_hint_size,
858 unsigned int metadata_version)
860 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
861 policy_hint_size, metadata_version);
863 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
864 dm_cache_metadata_close(cmd);
865 return ERR_PTR(-EINVAL);
871 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
873 if (refcount_dec_and_test(&cmd->ref_count)) {
874 mutex_lock(&table_lock);
875 list_del(&cmd->list);
876 mutex_unlock(&table_lock);
879 __destroy_persistent_data_objects(cmd, true);
885 * Checks that the given cache block is either unmapped or clean.
887 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
895 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
899 unpack_value(value, &ob, &flags);
900 *result = !((flags & M_VALID) && (flags & M_DIRTY));
905 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
906 dm_cblock_t begin, dm_cblock_t end,
912 while (begin != end) {
913 r = block_clean_combined_dirty(cmd, begin, result);
915 DMERR("block_clean_combined_dirty failed");
920 DMERR("cache block %llu is dirty",
921 (unsigned long long) from_cblock(begin));
925 begin = to_cblock(from_cblock(begin) + 1);
931 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
932 dm_cblock_t begin, dm_cblock_t end,
939 if (from_cblock(cmd->cache_blocks) == 0)
943 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
944 from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
946 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
950 r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
952 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
953 dm_bitset_cursor_end(&cmd->dirty_cursor);
957 while (begin != end) {
959 * We assume that unmapped blocks have their dirty bit
962 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
964 DMERR("%s: cache block %llu is dirty", __func__,
965 (unsigned long long) from_cblock(begin));
966 dm_bitset_cursor_end(&cmd->dirty_cursor);
971 begin = to_cblock(from_cblock(begin) + 1);
975 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
977 DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
978 dm_bitset_cursor_end(&cmd->dirty_cursor);
983 dm_bitset_cursor_end(&cmd->dirty_cursor);
988 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
989 dm_cblock_t begin, dm_cblock_t end,
992 if (separate_dirty_bits(cmd))
993 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
995 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
998 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
1000 down_write(&cmd->root_lock);
1001 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
1002 up_write(&cmd->root_lock);
1008 #define WRITE_LOCK(cmd) \
1010 if (!cmd_write_lock((cmd))) \
1014 #define WRITE_LOCK_VOID(cmd) \
1016 if (!cmd_write_lock((cmd))) \
1020 #define WRITE_UNLOCK(cmd) \
1021 up_write(&(cmd)->root_lock)
1023 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1025 down_read(&cmd->root_lock);
1027 up_read(&cmd->root_lock);
1033 #define READ_LOCK(cmd) \
1035 if (!cmd_read_lock((cmd))) \
1039 #define READ_LOCK_VOID(cmd) \
1041 if (!cmd_read_lock((cmd))) \
1045 #define READ_UNLOCK(cmd) \
1046 up_read(&(cmd)->root_lock)
1048 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1052 __le64 null_mapping = pack_value(0, 0);
1055 __dm_bless_for_disk(&null_mapping);
1057 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1058 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1060 __dm_unbless_for_disk(&null_mapping);
1065 DMERR("unable to shrink cache due to dirty blocks");
1067 __dm_unbless_for_disk(&null_mapping);
1072 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1073 from_cblock(new_cache_size),
1074 &null_mapping, &cmd->root);
1078 if (separate_dirty_bits(cmd)) {
1079 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1080 from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1081 false, &cmd->dirty_root);
1086 cmd->cache_blocks = new_cache_size;
1087 cmd->changed = true;
1095 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1096 sector_t discard_block_size,
1097 dm_dblock_t new_nr_entries)
1102 r = dm_bitset_resize(&cmd->discard_info,
1104 from_dblock(cmd->discard_nr_blocks),
1105 from_dblock(new_nr_entries),
1106 false, &cmd->discard_root);
1108 cmd->discard_block_size = discard_block_size;
1109 cmd->discard_nr_blocks = new_nr_entries;
1112 cmd->changed = true;
1118 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1120 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1121 from_dblock(b), &cmd->discard_root);
1124 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1126 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1127 from_dblock(b), &cmd->discard_root);
1130 static int __discard(struct dm_cache_metadata *cmd,
1131 dm_dblock_t dblock, bool discard)
1135 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1139 cmd->changed = true;
1143 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1144 dm_dblock_t dblock, bool discard)
1149 r = __discard(cmd, dblock, discard);
1155 static int __load_discards(struct dm_cache_metadata *cmd,
1156 load_discard_fn fn, void *context)
1160 struct dm_bitset_cursor c;
1162 if (from_dblock(cmd->discard_nr_blocks) == 0)
1166 if (cmd->clean_when_opened) {
1167 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1171 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1172 from_dblock(cmd->discard_nr_blocks), &c);
1176 for (b = 0; ; b++) {
1177 r = fn(context, cmd->discard_block_size, to_dblock(b),
1178 dm_bitset_cursor_get_value(&c));
1182 if (b >= (from_dblock(cmd->discard_nr_blocks) - 1))
1185 r = dm_bitset_cursor_next(&c);
1190 dm_bitset_cursor_end(&c);
1193 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1194 r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1203 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1204 load_discard_fn fn, void *context)
1209 r = __load_discards(cmd, fn, context);
1215 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1218 *result = cmd->cache_blocks;
1224 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1227 __le64 value = pack_value(0, 0);
1229 __dm_bless_for_disk(&value);
1230 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1231 &value, &cmd->root);
1235 cmd->changed = true;
1239 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1244 r = __remove(cmd, cblock);
1250 static int __insert(struct dm_cache_metadata *cmd,
1251 dm_cblock_t cblock, dm_oblock_t oblock)
1254 __le64 value = pack_value(oblock, M_VALID);
1255 __dm_bless_for_disk(&value);
1257 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1258 &value, &cmd->root);
1262 cmd->changed = true;
1266 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1267 dm_cblock_t cblock, dm_oblock_t oblock)
1272 r = __insert(cmd, cblock, oblock);
1282 struct dm_cache_metadata *cmd;
1283 bool respect_dirty_flags;
1287 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1288 struct dm_cache_policy *policy)
1290 const char *policy_name = dm_cache_policy_get_name(policy);
1291 const unsigned int *policy_version = dm_cache_policy_get_version(policy);
1292 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1295 * Ensure policy names match.
1297 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1301 * Ensure policy major versions match.
1303 if (cmd->policy_version[0] != policy_version[0])
1307 * Ensure policy hint sizes match.
1309 if (cmd->policy_hint_size != policy_hint_size)
1315 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1317 return cmd->hint_root && cmd->policy_hint_size;
1320 static bool hints_array_available(struct dm_cache_metadata *cmd,
1321 struct dm_cache_policy *policy)
1323 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1324 hints_array_initialized(cmd);
1327 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1328 uint64_t cb, bool hints_valid,
1329 struct dm_array_cursor *mapping_cursor,
1330 struct dm_array_cursor *hint_cursor,
1331 load_mapping_fn fn, void *context)
1338 __le64 *mapping_value_le;
1339 __le32 *hint_value_le;
1345 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1346 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1347 unpack_value(mapping, &oblock, &flags);
1349 if (flags & M_VALID) {
1351 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1352 memcpy(&hint, hint_value_le, sizeof(hint));
1354 if (cmd->clean_when_opened)
1355 dirty = flags & M_DIRTY;
1357 r = fn(context, oblock, to_cblock(cb), dirty,
1358 le32_to_cpu(hint), hints_valid);
1360 DMERR("policy couldn't load cache block %llu",
1361 (unsigned long long) from_cblock(to_cblock(cb)));
1368 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1369 uint64_t cb, bool hints_valid,
1370 struct dm_array_cursor *mapping_cursor,
1371 struct dm_array_cursor *hint_cursor,
1372 struct dm_bitset_cursor *dirty_cursor,
1373 load_mapping_fn fn, void *context)
1380 __le64 *mapping_value_le;
1381 __le32 *hint_value_le;
1387 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1388 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1389 unpack_value(mapping, &oblock, &flags);
1391 if (flags & M_VALID) {
1393 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1394 memcpy(&hint, hint_value_le, sizeof(hint));
1396 if (cmd->clean_when_opened)
1397 dirty = dm_bitset_cursor_get_value(dirty_cursor);
1399 r = fn(context, oblock, to_cblock(cb), dirty,
1400 le32_to_cpu(hint), hints_valid);
1402 DMERR("policy couldn't load cache block %llu",
1403 (unsigned long long) from_cblock(to_cblock(cb)));
1410 static int __load_mappings(struct dm_cache_metadata *cmd,
1411 struct dm_cache_policy *policy,
1412 load_mapping_fn fn, void *context)
1417 bool hints_valid = hints_array_available(cmd, policy);
1419 if (from_cblock(cmd->cache_blocks) == 0)
1423 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1428 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1430 dm_array_cursor_end(&cmd->mapping_cursor);
1435 if (separate_dirty_bits(cmd)) {
1436 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1437 from_cblock(cmd->cache_blocks),
1438 &cmd->dirty_cursor);
1440 dm_array_cursor_end(&cmd->hint_cursor);
1441 dm_array_cursor_end(&cmd->mapping_cursor);
1446 for (cb = 0; ; cb++) {
1447 if (separate_dirty_bits(cmd))
1448 r = __load_mapping_v2(cmd, cb, hints_valid,
1449 &cmd->mapping_cursor,
1454 r = __load_mapping_v1(cmd, cb, hints_valid,
1455 &cmd->mapping_cursor, &cmd->hint_cursor,
1461 * We need to break out before we move the cursors.
1463 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1466 r = dm_array_cursor_next(&cmd->mapping_cursor);
1468 DMERR("dm_array_cursor_next for mapping failed");
1473 r = dm_array_cursor_next(&cmd->hint_cursor);
1475 dm_array_cursor_end(&cmd->hint_cursor);
1476 hints_valid = false;
1480 if (separate_dirty_bits(cmd)) {
1481 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1483 DMERR("dm_bitset_cursor_next for dirty failed");
1489 dm_array_cursor_end(&cmd->mapping_cursor);
1491 dm_array_cursor_end(&cmd->hint_cursor);
1493 if (separate_dirty_bits(cmd))
1494 dm_bitset_cursor_end(&cmd->dirty_cursor);
1499 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1500 struct dm_cache_policy *policy,
1501 load_mapping_fn fn, void *context)
1506 r = __load_mappings(cmd, policy, fn, context);
1512 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1518 memcpy(&value, leaf, sizeof(value));
1519 unpack_value(value, &oblock, &flags);
1524 static int __dump_mappings(struct dm_cache_metadata *cmd)
1526 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1529 void dm_cache_dump(struct dm_cache_metadata *cmd)
1531 READ_LOCK_VOID(cmd);
1532 __dump_mappings(cmd);
1536 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1547 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1554 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1558 unpack_value(value, &oblock, &flags);
1560 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1561 /* nothing to be done */
1564 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1565 __dm_bless_for_disk(&value);
1567 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1568 &value, &cmd->root);
1572 cmd->changed = true;
1577 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned int nr_bits, unsigned long *bits)
1581 for (i = 0; i < nr_bits; i++) {
1582 r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1590 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1592 unsigned long *bits = context;
1593 *value = test_bit(index, bits);
1597 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned int nr_bits, unsigned long *bits)
1601 /* nr_bits is really just a sanity check */
1602 if (nr_bits != from_cblock(cmd->cache_blocks)) {
1603 DMERR("dirty bitset is wrong size");
1607 r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1611 cmd->changed = true;
1612 return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1615 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1616 unsigned int nr_bits,
1617 unsigned long *bits)
1622 if (separate_dirty_bits(cmd))
1623 r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1625 r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1631 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1632 struct dm_cache_statistics *stats)
1634 READ_LOCK_VOID(cmd);
1635 *stats = cmd->stats;
1639 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1640 struct dm_cache_statistics *stats)
1642 WRITE_LOCK_VOID(cmd);
1643 cmd->stats = *stats;
1647 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1650 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1651 clear_clean_shutdown);
1657 r = __commit_transaction(cmd, mutator);
1661 r = __begin_transaction(cmd);
1667 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1674 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1680 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1687 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1693 /*----------------------------------------------------------------*/
1695 static int get_hint(uint32_t index, void *value_le, void *context)
1698 struct dm_cache_policy *policy = context;
1700 value = policy_get_hint(policy, to_cblock(index));
1701 *((__le32 *) value_le) = cpu_to_le32(value);
1707 * It's quicker to always delete the hint array, and recreate with
1710 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1714 const char *policy_name = dm_cache_policy_get_name(policy);
1715 const unsigned int *policy_version = dm_cache_policy_get_version(policy);
1717 if (!policy_name[0] ||
1718 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1721 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1722 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1724 hint_size = dm_cache_policy_get_hint_size(policy);
1726 return 0; /* short-circuit hints initialization */
1727 cmd->policy_hint_size = hint_size;
1729 if (cmd->hint_root) {
1730 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1735 return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1736 from_cblock(cmd->cache_blocks),
1740 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1745 r = write_hints(cmd, policy);
1751 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1756 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1762 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1764 WRITE_LOCK_VOID(cmd);
1765 dm_bm_set_read_only(cmd->bm);
1769 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1771 WRITE_LOCK_VOID(cmd);
1772 dm_bm_set_read_write(cmd->bm);
1776 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1779 struct dm_block *sblock;
1780 struct cache_disk_superblock *disk_super;
1783 set_bit(NEEDS_CHECK, &cmd->flags);
1785 r = superblock_lock(cmd, &sblock);
1787 DMERR("couldn't read superblock");
1791 disk_super = dm_block_data(sblock);
1792 disk_super->flags = cpu_to_le32(cmd->flags);
1794 dm_bm_unlock(sblock);
1801 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1804 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1810 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1813 struct dm_block_manager *old_bm = NULL, *new_bm = NULL;
1815 /* fail_io is double-checked with cmd->root_lock held below */
1816 if (unlikely(cmd->fail_io))
1820 * Replacement block manager (new_bm) is created and old_bm destroyed outside of
1821 * cmd root_lock to avoid ABBA deadlock that would result (due to life-cycle of
1822 * shrinker associated with the block manager's bufio client vs cmd root_lock).
1823 * - must take shrinker_rwsem without holding cmd->root_lock
1825 new_bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
1826 CACHE_MAX_CONCURRENT_LOCKS);
1834 __destroy_persistent_data_objects(cmd, false);
1836 if (IS_ERR(new_bm)) {
1837 DMERR("could not create block manager during abort");
1839 r = PTR_ERR(new_bm);
1844 r = __open_or_format_metadata(cmd, false);
1852 cmd->fail_io = true;
1854 dm_block_manager_destroy(old_bm);
1856 if (new_bm && !IS_ERR(new_bm))
1857 dm_block_manager_destroy(new_bm);