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;
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 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 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 = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
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 = 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 (get_disk_ro(cmd->bdev->bd_disk))
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 return PTR_ERR(cmd->bm);
543 r = __open_or_format_metadata(cmd, may_format_device);
545 dm_block_manager_destroy(cmd->bm);
550 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
552 dm_sm_destroy(cmd->metadata_sm);
553 dm_tm_destroy(cmd->tm);
554 dm_block_manager_destroy(cmd->bm);
557 typedef unsigned long (*flags_mutator)(unsigned long);
559 static void update_flags(struct cache_disk_superblock *disk_super,
560 flags_mutator mutator)
562 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
563 disk_super->flags = cpu_to_le32(sb_flags);
566 static unsigned long set_clean_shutdown(unsigned long flags)
568 set_bit(CLEAN_SHUTDOWN, &flags);
572 static unsigned long clear_clean_shutdown(unsigned long flags)
574 clear_bit(CLEAN_SHUTDOWN, &flags);
578 static void read_superblock_fields(struct dm_cache_metadata *cmd,
579 struct cache_disk_superblock *disk_super)
581 cmd->version = le32_to_cpu(disk_super->version);
582 cmd->flags = le32_to_cpu(disk_super->flags);
583 cmd->root = le64_to_cpu(disk_super->mapping_root);
584 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
585 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
586 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
587 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
588 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
589 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
590 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
591 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
592 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
593 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
594 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
596 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
597 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
598 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
599 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
601 if (separate_dirty_bits(cmd))
602 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
604 cmd->changed = false;
608 * The mutator updates the superblock flags.
610 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
611 flags_mutator mutator)
614 struct cache_disk_superblock *disk_super;
615 struct dm_block *sblock;
617 r = superblock_lock(cmd, &sblock);
621 disk_super = dm_block_data(sblock);
622 update_flags(disk_super, mutator);
623 read_superblock_fields(cmd, disk_super);
624 dm_bm_unlock(sblock);
626 return dm_bm_flush(cmd->bm);
629 static int __begin_transaction(struct dm_cache_metadata *cmd)
632 struct cache_disk_superblock *disk_super;
633 struct dm_block *sblock;
636 * We re-read the superblock every time. Shouldn't need to do this
639 r = superblock_read_lock(cmd, &sblock);
643 disk_super = dm_block_data(sblock);
644 read_superblock_fields(cmd, disk_super);
645 dm_bm_unlock(sblock);
650 static int __commit_transaction(struct dm_cache_metadata *cmd,
651 flags_mutator mutator)
654 struct cache_disk_superblock *disk_super;
655 struct dm_block *sblock;
658 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
660 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
662 if (separate_dirty_bits(cmd)) {
663 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
669 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
674 r = dm_tm_pre_commit(cmd->tm);
678 r = __save_sm_root(cmd);
682 r = superblock_lock(cmd, &sblock);
686 disk_super = dm_block_data(sblock);
688 disk_super->flags = cpu_to_le32(cmd->flags);
690 update_flags(disk_super, mutator);
692 disk_super->mapping_root = cpu_to_le64(cmd->root);
693 if (separate_dirty_bits(cmd))
694 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
695 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
696 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
697 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
698 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
699 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
700 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
701 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
702 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
703 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
705 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
706 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
707 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
708 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
709 __copy_sm_root(cmd, disk_super);
711 return dm_tm_commit(cmd->tm, sblock);
714 /*----------------------------------------------------------------*/
717 * The mappings are held in a dm-array that has 64-bit values stored in
718 * little-endian format. The index is the cblock, the high 48bits of the
719 * value are the oblock and the low 16 bit the flags.
721 #define FLAGS_MASK ((1 << 16) - 1)
723 static __le64 pack_value(dm_oblock_t block, unsigned flags)
725 uint64_t value = from_oblock(block);
727 value = value | (flags & FLAGS_MASK);
728 return cpu_to_le64(value);
731 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
733 uint64_t value = le64_to_cpu(value_le);
734 uint64_t b = value >> 16;
735 *block = to_oblock(b);
736 *flags = value & FLAGS_MASK;
739 /*----------------------------------------------------------------*/
741 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
742 sector_t data_block_size,
743 bool may_format_device,
744 size_t policy_hint_size,
745 unsigned metadata_version)
748 struct dm_cache_metadata *cmd;
750 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
752 DMERR("could not allocate metadata struct");
753 return ERR_PTR(-ENOMEM);
756 cmd->version = metadata_version;
757 refcount_set(&cmd->ref_count, 1);
758 init_rwsem(&cmd->root_lock);
760 cmd->data_block_size = data_block_size;
761 cmd->cache_blocks = 0;
762 cmd->policy_hint_size = policy_hint_size;
764 cmd->fail_io = false;
766 r = __create_persistent_data_objects(cmd, may_format_device);
772 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
774 dm_cache_metadata_close(cmd);
782 * We keep a little list of ref counted metadata objects to prevent two
783 * different target instances creating separate bufio instances. This is
784 * an issue if a table is reloaded before the suspend.
786 static DEFINE_MUTEX(table_lock);
787 static LIST_HEAD(table);
789 static struct dm_cache_metadata *lookup(struct block_device *bdev)
791 struct dm_cache_metadata *cmd;
793 list_for_each_entry(cmd, &table, list)
794 if (cmd->bdev == bdev) {
795 refcount_inc(&cmd->ref_count);
802 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
803 sector_t data_block_size,
804 bool may_format_device,
805 size_t policy_hint_size,
806 unsigned metadata_version)
808 struct dm_cache_metadata *cmd, *cmd2;
810 mutex_lock(&table_lock);
812 mutex_unlock(&table_lock);
817 cmd = metadata_open(bdev, data_block_size, may_format_device,
818 policy_hint_size, metadata_version);
820 mutex_lock(&table_lock);
823 mutex_unlock(&table_lock);
824 __destroy_persistent_data_objects(cmd);
828 list_add(&cmd->list, &table);
829 mutex_unlock(&table_lock);
835 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
837 if (cmd->data_block_size != data_block_size) {
838 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
839 (unsigned long long) data_block_size,
840 (unsigned long long) cmd->data_block_size);
847 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
848 sector_t data_block_size,
849 bool may_format_device,
850 size_t policy_hint_size,
851 unsigned metadata_version)
853 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
854 policy_hint_size, metadata_version);
856 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
857 dm_cache_metadata_close(cmd);
858 return ERR_PTR(-EINVAL);
864 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
866 if (refcount_dec_and_test(&cmd->ref_count)) {
867 mutex_lock(&table_lock);
868 list_del(&cmd->list);
869 mutex_unlock(&table_lock);
872 __destroy_persistent_data_objects(cmd);
878 * Checks that the given cache block is either unmapped or clean.
880 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
888 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
892 unpack_value(value, &ob, &flags);
893 *result = !((flags & M_VALID) && (flags & M_DIRTY));
898 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
899 dm_cblock_t begin, dm_cblock_t end,
905 while (begin != end) {
906 r = block_clean_combined_dirty(cmd, begin, result);
908 DMERR("block_clean_combined_dirty failed");
913 DMERR("cache block %llu is dirty",
914 (unsigned long long) from_cblock(begin));
918 begin = to_cblock(from_cblock(begin) + 1);
924 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
925 dm_cblock_t begin, dm_cblock_t end,
932 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
933 from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
935 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
939 r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
941 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
942 dm_bitset_cursor_end(&cmd->dirty_cursor);
946 while (begin != end) {
948 * We assume that unmapped blocks have their dirty bit
951 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
953 DMERR("%s: cache block %llu is dirty", __func__,
954 (unsigned long long) from_cblock(begin));
955 dm_bitset_cursor_end(&cmd->dirty_cursor);
960 begin = to_cblock(from_cblock(begin) + 1);
964 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
966 DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
967 dm_bitset_cursor_end(&cmd->dirty_cursor);
972 dm_bitset_cursor_end(&cmd->dirty_cursor);
977 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
978 dm_cblock_t begin, dm_cblock_t end,
981 if (separate_dirty_bits(cmd))
982 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
984 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
987 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
989 down_write(&cmd->root_lock);
990 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
991 up_write(&cmd->root_lock);
997 #define WRITE_LOCK(cmd) \
999 if (!cmd_write_lock((cmd))) \
1003 #define WRITE_LOCK_VOID(cmd) \
1005 if (!cmd_write_lock((cmd))) \
1009 #define WRITE_UNLOCK(cmd) \
1010 up_write(&(cmd)->root_lock)
1012 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1014 down_read(&cmd->root_lock);
1016 up_read(&cmd->root_lock);
1022 #define READ_LOCK(cmd) \
1024 if (!cmd_read_lock((cmd))) \
1028 #define READ_LOCK_VOID(cmd) \
1030 if (!cmd_read_lock((cmd))) \
1034 #define READ_UNLOCK(cmd) \
1035 up_read(&(cmd)->root_lock)
1037 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1041 __le64 null_mapping = pack_value(0, 0);
1044 __dm_bless_for_disk(&null_mapping);
1046 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1047 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1049 __dm_unbless_for_disk(&null_mapping);
1054 DMERR("unable to shrink cache due to dirty blocks");
1056 __dm_unbless_for_disk(&null_mapping);
1061 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1062 from_cblock(new_cache_size),
1063 &null_mapping, &cmd->root);
1067 if (separate_dirty_bits(cmd)) {
1068 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1069 from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1070 false, &cmd->dirty_root);
1075 cmd->cache_blocks = new_cache_size;
1076 cmd->changed = true;
1084 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1085 sector_t discard_block_size,
1086 dm_dblock_t new_nr_entries)
1091 r = dm_bitset_resize(&cmd->discard_info,
1093 from_dblock(cmd->discard_nr_blocks),
1094 from_dblock(new_nr_entries),
1095 false, &cmd->discard_root);
1097 cmd->discard_block_size = discard_block_size;
1098 cmd->discard_nr_blocks = new_nr_entries;
1101 cmd->changed = true;
1107 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1109 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1110 from_dblock(b), &cmd->discard_root);
1113 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1115 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1116 from_dblock(b), &cmd->discard_root);
1119 static int __discard(struct dm_cache_metadata *cmd,
1120 dm_dblock_t dblock, bool discard)
1124 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1128 cmd->changed = true;
1132 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1133 dm_dblock_t dblock, bool discard)
1138 r = __discard(cmd, dblock, discard);
1144 static int __load_discards(struct dm_cache_metadata *cmd,
1145 load_discard_fn fn, void *context)
1149 struct dm_bitset_cursor c;
1151 if (from_dblock(cmd->discard_nr_blocks) == 0)
1155 if (cmd->clean_when_opened) {
1156 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1160 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1161 from_dblock(cmd->discard_nr_blocks), &c);
1165 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1166 r = fn(context, cmd->discard_block_size, to_dblock(b),
1167 dm_bitset_cursor_get_value(&c));
1172 dm_bitset_cursor_end(&c);
1175 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1176 r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1185 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1186 load_discard_fn fn, void *context)
1191 r = __load_discards(cmd, fn, context);
1197 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1200 *result = cmd->cache_blocks;
1206 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1209 __le64 value = pack_value(0, 0);
1211 __dm_bless_for_disk(&value);
1212 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1213 &value, &cmd->root);
1217 cmd->changed = true;
1221 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1226 r = __remove(cmd, cblock);
1232 static int __insert(struct dm_cache_metadata *cmd,
1233 dm_cblock_t cblock, dm_oblock_t oblock)
1236 __le64 value = pack_value(oblock, M_VALID);
1237 __dm_bless_for_disk(&value);
1239 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1240 &value, &cmd->root);
1244 cmd->changed = true;
1248 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1249 dm_cblock_t cblock, dm_oblock_t oblock)
1254 r = __insert(cmd, cblock, oblock);
1264 struct dm_cache_metadata *cmd;
1265 bool respect_dirty_flags;
1269 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1270 struct dm_cache_policy *policy)
1272 const char *policy_name = dm_cache_policy_get_name(policy);
1273 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1274 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1277 * Ensure policy names match.
1279 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1283 * Ensure policy major versions match.
1285 if (cmd->policy_version[0] != policy_version[0])
1289 * Ensure policy hint sizes match.
1291 if (cmd->policy_hint_size != policy_hint_size)
1297 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1299 return cmd->hint_root && cmd->policy_hint_size;
1302 static bool hints_array_available(struct dm_cache_metadata *cmd,
1303 struct dm_cache_policy *policy)
1305 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1306 hints_array_initialized(cmd);
1309 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1310 uint64_t cb, bool hints_valid,
1311 struct dm_array_cursor *mapping_cursor,
1312 struct dm_array_cursor *hint_cursor,
1313 load_mapping_fn fn, void *context)
1320 __le64 *mapping_value_le;
1321 __le32 *hint_value_le;
1326 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1327 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1328 unpack_value(mapping, &oblock, &flags);
1330 if (flags & M_VALID) {
1332 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1333 memcpy(&hint, hint_value_le, sizeof(hint));
1336 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1337 le32_to_cpu(hint), hints_valid);
1339 DMERR("policy couldn't load cache block %llu",
1340 (unsigned long long) from_cblock(to_cblock(cb)));
1347 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1348 uint64_t cb, bool hints_valid,
1349 struct dm_array_cursor *mapping_cursor,
1350 struct dm_array_cursor *hint_cursor,
1351 struct dm_bitset_cursor *dirty_cursor,
1352 load_mapping_fn fn, void *context)
1359 __le64 *mapping_value_le;
1360 __le32 *hint_value_le;
1366 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1367 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1368 unpack_value(mapping, &oblock, &flags);
1370 if (flags & M_VALID) {
1372 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1373 memcpy(&hint, hint_value_le, sizeof(hint));
1376 dirty = dm_bitset_cursor_get_value(dirty_cursor);
1377 r = fn(context, oblock, to_cblock(cb), dirty,
1378 le32_to_cpu(hint), hints_valid);
1380 DMERR("policy couldn't load cache block %llu",
1381 (unsigned long long) from_cblock(to_cblock(cb)));
1388 static int __load_mappings(struct dm_cache_metadata *cmd,
1389 struct dm_cache_policy *policy,
1390 load_mapping_fn fn, void *context)
1395 bool hints_valid = hints_array_available(cmd, policy);
1397 if (from_cblock(cmd->cache_blocks) == 0)
1401 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1406 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1408 dm_array_cursor_end(&cmd->mapping_cursor);
1413 if (separate_dirty_bits(cmd)) {
1414 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1415 from_cblock(cmd->cache_blocks),
1416 &cmd->dirty_cursor);
1418 dm_array_cursor_end(&cmd->hint_cursor);
1419 dm_array_cursor_end(&cmd->mapping_cursor);
1424 for (cb = 0; ; cb++) {
1425 if (separate_dirty_bits(cmd))
1426 r = __load_mapping_v2(cmd, cb, hints_valid,
1427 &cmd->mapping_cursor,
1432 r = __load_mapping_v1(cmd, cb, hints_valid,
1433 &cmd->mapping_cursor, &cmd->hint_cursor,
1439 * We need to break out before we move the cursors.
1441 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1444 r = dm_array_cursor_next(&cmd->mapping_cursor);
1446 DMERR("dm_array_cursor_next for mapping failed");
1451 r = dm_array_cursor_next(&cmd->hint_cursor);
1453 DMERR("dm_array_cursor_next for hint failed");
1458 if (separate_dirty_bits(cmd)) {
1459 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1461 DMERR("dm_bitset_cursor_next for dirty failed");
1467 dm_array_cursor_end(&cmd->mapping_cursor);
1469 dm_array_cursor_end(&cmd->hint_cursor);
1471 if (separate_dirty_bits(cmd))
1472 dm_bitset_cursor_end(&cmd->dirty_cursor);
1477 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1478 struct dm_cache_policy *policy,
1479 load_mapping_fn fn, void *context)
1484 r = __load_mappings(cmd, policy, fn, context);
1490 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1497 memcpy(&value, leaf, sizeof(value));
1498 unpack_value(value, &oblock, &flags);
1503 static int __dump_mappings(struct dm_cache_metadata *cmd)
1505 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1508 void dm_cache_dump(struct dm_cache_metadata *cmd)
1510 READ_LOCK_VOID(cmd);
1511 __dump_mappings(cmd);
1515 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1526 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1533 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1537 unpack_value(value, &oblock, &flags);
1539 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1540 /* nothing to be done */
1543 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1544 __dm_bless_for_disk(&value);
1546 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1547 &value, &cmd->root);
1551 cmd->changed = true;
1556 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1560 for (i = 0; i < nr_bits; i++) {
1561 r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1569 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1571 unsigned long *bits = context;
1572 *value = test_bit(index, bits);
1576 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1580 /* nr_bits is really just a sanity check */
1581 if (nr_bits != from_cblock(cmd->cache_blocks)) {
1582 DMERR("dirty bitset is wrong size");
1586 r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1590 cmd->changed = true;
1591 return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1594 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1596 unsigned long *bits)
1601 if (separate_dirty_bits(cmd))
1602 r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1604 r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1610 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1611 struct dm_cache_statistics *stats)
1613 READ_LOCK_VOID(cmd);
1614 *stats = cmd->stats;
1618 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1619 struct dm_cache_statistics *stats)
1621 WRITE_LOCK_VOID(cmd);
1622 cmd->stats = *stats;
1626 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1629 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1630 clear_clean_shutdown);
1636 r = __commit_transaction(cmd, mutator);
1640 r = __begin_transaction(cmd);
1646 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1653 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1659 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1666 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1672 /*----------------------------------------------------------------*/
1674 static int get_hint(uint32_t index, void *value_le, void *context)
1677 struct dm_cache_policy *policy = context;
1679 value = policy_get_hint(policy, to_cblock(index));
1680 *((__le32 *) value_le) = cpu_to_le32(value);
1686 * It's quicker to always delete the hint array, and recreate with
1689 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1693 const char *policy_name = dm_cache_policy_get_name(policy);
1694 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1696 if (!policy_name[0] ||
1697 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1700 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1701 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1703 hint_size = dm_cache_policy_get_hint_size(policy);
1705 return 0; /* short-circuit hints initialization */
1706 cmd->policy_hint_size = hint_size;
1708 if (cmd->hint_root) {
1709 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1714 return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1715 from_cblock(cmd->cache_blocks),
1719 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1724 r = write_hints(cmd, policy);
1730 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1735 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1741 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1743 WRITE_LOCK_VOID(cmd);
1744 dm_bm_set_read_only(cmd->bm);
1748 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1750 WRITE_LOCK_VOID(cmd);
1751 dm_bm_set_read_write(cmd->bm);
1755 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1758 struct dm_block *sblock;
1759 struct cache_disk_superblock *disk_super;
1762 set_bit(NEEDS_CHECK, &cmd->flags);
1764 r = superblock_lock(cmd, &sblock);
1766 DMERR("couldn't read superblock");
1770 disk_super = dm_block_data(sblock);
1771 disk_super->flags = cpu_to_le32(cmd->flags);
1773 dm_bm_unlock(sblock);
1780 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1783 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1789 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1794 __destroy_persistent_data_objects(cmd);
1795 r = __create_persistent_data_objects(cmd, false);
1797 cmd->fail_io = true;