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 2
31 * 3 for btree insert +
32 * 2 for btree lookup used within space map
34 #define CACHE_MAX_CONCURRENT_LOCKS 5
35 #define SPACE_MAP_ROOT_SIZE 128
37 enum superblock_flag_bits {
38 /* for spotting crashes that would invalidate the dirty bitset */
40 /* metadata must be checked using the tools */
45 * Each mapping from cache block -> origin block carries a set of flags.
49 * A valid mapping. Because we're using an array we clear this
50 * flag for an non existant mapping.
55 * The data on the cache is different from that on the origin.
56 * This flag is only used by metadata format 1.
61 struct cache_disk_superblock {
70 __u8 policy_name[CACHE_POLICY_NAME_SIZE];
71 __le32 policy_hint_size;
73 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
78 __le64 discard_block_size;
79 __le64 discard_nr_blocks;
81 __le32 data_block_size;
82 __le32 metadata_block_size;
86 __le32 compat_ro_flags;
87 __le32 incompat_flags;
94 __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
97 * Metadata format 2 fields.
102 struct dm_cache_metadata {
104 struct list_head list;
107 struct block_device *bdev;
108 struct dm_block_manager *bm;
109 struct dm_space_map *metadata_sm;
110 struct dm_transaction_manager *tm;
112 struct dm_array_info info;
113 struct dm_array_info hint_info;
114 struct dm_disk_bitset discard_info;
116 struct rw_semaphore root_lock;
119 dm_block_t hint_root;
120 dm_block_t discard_root;
122 sector_t discard_block_size;
123 dm_dblock_t discard_nr_blocks;
125 sector_t data_block_size;
126 dm_cblock_t cache_blocks;
128 bool clean_when_opened:1;
130 char policy_name[CACHE_POLICY_NAME_SIZE];
131 unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
132 size_t policy_hint_size;
133 struct dm_cache_statistics stats;
136 * Reading the space map root can fail, so we read it into this
137 * buffer before the superblock is locked and updated.
139 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
142 * Set if a transaction has to be aborted but the attempt to roll
143 * back to the previous (good) transaction failed. The only
144 * metadata operation permissible in this state is the closing of
150 * Metadata format 2 fields.
152 dm_block_t dirty_root;
153 struct dm_disk_bitset dirty_info;
156 * These structures are used when loading metadata. They're too
157 * big to put on the stack.
159 struct dm_array_cursor mapping_cursor;
160 struct dm_array_cursor hint_cursor;
161 struct dm_bitset_cursor dirty_cursor;
164 /*-------------------------------------------------------------------
165 * superblock validator
166 *-----------------------------------------------------------------*/
168 #define SUPERBLOCK_CSUM_XOR 9031977
170 static void sb_prepare_for_write(struct dm_block_validator *v,
172 size_t sb_block_size)
174 struct cache_disk_superblock *disk_super = dm_block_data(b);
176 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
177 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
178 sb_block_size - sizeof(__le32),
179 SUPERBLOCK_CSUM_XOR));
182 static int check_metadata_version(struct cache_disk_superblock *disk_super)
184 uint32_t metadata_version = le32_to_cpu(disk_super->version);
186 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
187 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
188 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
195 static int sb_check(struct dm_block_validator *v,
197 size_t sb_block_size)
199 struct cache_disk_superblock *disk_super = dm_block_data(b);
202 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
203 DMERR("sb_check failed: blocknr %llu: wanted %llu",
204 le64_to_cpu(disk_super->blocknr),
205 (unsigned long long)dm_block_location(b));
209 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
210 DMERR("sb_check failed: magic %llu: wanted %llu",
211 le64_to_cpu(disk_super->magic),
212 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
216 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
217 sb_block_size - sizeof(__le32),
218 SUPERBLOCK_CSUM_XOR));
219 if (csum_le != disk_super->csum) {
220 DMERR("sb_check failed: csum %u: wanted %u",
221 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
225 return check_metadata_version(disk_super);
228 static struct dm_block_validator sb_validator = {
229 .name = "superblock",
230 .prepare_for_write = sb_prepare_for_write,
234 /*----------------------------------------------------------------*/
236 static int superblock_read_lock(struct dm_cache_metadata *cmd,
237 struct dm_block **sblock)
239 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
240 &sb_validator, sblock);
243 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
244 struct dm_block **sblock)
246 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
247 &sb_validator, sblock);
250 static int superblock_lock(struct dm_cache_metadata *cmd,
251 struct dm_block **sblock)
253 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
254 &sb_validator, sblock);
257 /*----------------------------------------------------------------*/
259 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
264 __le64 *data_le, zero = cpu_to_le64(0);
265 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
268 * We can't use a validator here - it may be all zeroes.
270 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
274 data_le = dm_block_data(b);
276 for (i = 0; i < sb_block_size; i++) {
277 if (data_le[i] != zero) {
288 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
290 struct dm_btree_value_type vt;
293 vt.size = sizeof(__le64);
297 dm_array_info_init(&cmd->info, cmd->tm, &vt);
299 if (cmd->policy_hint_size) {
300 vt.size = sizeof(__le32);
301 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
305 static int __save_sm_root(struct dm_cache_metadata *cmd)
310 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
314 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
318 static void __copy_sm_root(struct dm_cache_metadata *cmd,
319 struct cache_disk_superblock *disk_super)
321 memcpy(&disk_super->metadata_space_map_root,
322 &cmd->metadata_space_map_root,
323 sizeof(cmd->metadata_space_map_root));
326 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
328 return cmd->version >= 2;
331 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
334 struct dm_block *sblock;
335 struct cache_disk_superblock *disk_super;
336 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
338 /* FIXME: see if we can lose the max sectors limit */
339 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
340 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
342 r = dm_tm_pre_commit(cmd->tm);
347 * dm_sm_copy_root() can fail. So we need to do it before we start
348 * updating the superblock.
350 r = __save_sm_root(cmd);
354 r = superblock_lock_zero(cmd, &sblock);
358 disk_super = dm_block_data(sblock);
359 disk_super->flags = 0;
360 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
361 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
362 disk_super->version = cpu_to_le32(cmd->version);
363 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
364 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
365 disk_super->policy_hint_size = cpu_to_le32(0);
367 __copy_sm_root(cmd, disk_super);
369 disk_super->mapping_root = cpu_to_le64(cmd->root);
370 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
371 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
372 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
373 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
374 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
375 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
376 disk_super->cache_blocks = cpu_to_le32(0);
378 disk_super->read_hits = cpu_to_le32(0);
379 disk_super->read_misses = cpu_to_le32(0);
380 disk_super->write_hits = cpu_to_le32(0);
381 disk_super->write_misses = cpu_to_le32(0);
383 if (separate_dirty_bits(cmd))
384 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
386 return dm_tm_commit(cmd->tm, sblock);
389 static int __format_metadata(struct dm_cache_metadata *cmd)
393 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
394 &cmd->tm, &cmd->metadata_sm);
396 DMERR("tm_create_with_sm failed");
400 __setup_mapping_info(cmd);
402 r = dm_array_empty(&cmd->info, &cmd->root);
406 if (separate_dirty_bits(cmd)) {
407 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
408 r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
413 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
414 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
418 cmd->discard_block_size = 0;
419 cmd->discard_nr_blocks = 0;
421 r = __write_initial_superblock(cmd);
425 cmd->clean_when_opened = true;
429 dm_tm_destroy(cmd->tm);
430 dm_sm_destroy(cmd->metadata_sm);
435 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
436 struct dm_cache_metadata *cmd)
438 uint32_t incompat_flags, features;
440 incompat_flags = le32_to_cpu(disk_super->incompat_flags);
441 features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
443 DMERR("could not access metadata due to unsupported optional features (%lx).",
444 (unsigned long)features);
449 * Check for read-only metadata to skip the following RDWR checks.
451 if (get_disk_ro(cmd->bdev->bd_disk))
454 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
456 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
457 (unsigned long)features);
464 static int __open_metadata(struct dm_cache_metadata *cmd)
467 struct dm_block *sblock;
468 struct cache_disk_superblock *disk_super;
469 unsigned long sb_flags;
471 r = superblock_read_lock(cmd, &sblock);
473 DMERR("couldn't read lock superblock");
477 disk_super = dm_block_data(sblock);
479 /* Verify the data block size hasn't changed */
480 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
481 DMERR("changing the data block size (from %u to %llu) is not supported",
482 le32_to_cpu(disk_super->data_block_size),
483 (unsigned long long)cmd->data_block_size);
488 r = __check_incompat_features(disk_super, cmd);
492 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
493 disk_super->metadata_space_map_root,
494 sizeof(disk_super->metadata_space_map_root),
495 &cmd->tm, &cmd->metadata_sm);
497 DMERR("tm_open_with_sm failed");
501 __setup_mapping_info(cmd);
502 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
503 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
504 sb_flags = le32_to_cpu(disk_super->flags);
505 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
506 dm_bm_unlock(sblock);
511 dm_bm_unlock(sblock);
515 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
519 bool unformatted = false;
521 r = __superblock_all_zeroes(cmd->bm, &unformatted);
526 return format_device ? __format_metadata(cmd) : -EPERM;
528 return __open_metadata(cmd);
531 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
532 bool may_format_device)
535 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
536 CACHE_MAX_CONCURRENT_LOCKS);
537 if (IS_ERR(cmd->bm)) {
538 DMERR("could not create block manager");
539 r = PTR_ERR(cmd->bm);
544 r = __open_or_format_metadata(cmd, may_format_device);
546 dm_block_manager_destroy(cmd->bm);
553 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd,
556 dm_sm_destroy(cmd->metadata_sm);
557 dm_tm_destroy(cmd->tm);
559 dm_block_manager_destroy(cmd->bm);
562 typedef unsigned long (*flags_mutator)(unsigned long);
564 static void update_flags(struct cache_disk_superblock *disk_super,
565 flags_mutator mutator)
567 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
568 disk_super->flags = cpu_to_le32(sb_flags);
571 static unsigned long set_clean_shutdown(unsigned long flags)
573 set_bit(CLEAN_SHUTDOWN, &flags);
577 static unsigned long clear_clean_shutdown(unsigned long flags)
579 clear_bit(CLEAN_SHUTDOWN, &flags);
583 static void read_superblock_fields(struct dm_cache_metadata *cmd,
584 struct cache_disk_superblock *disk_super)
586 cmd->version = le32_to_cpu(disk_super->version);
587 cmd->flags = le32_to_cpu(disk_super->flags);
588 cmd->root = le64_to_cpu(disk_super->mapping_root);
589 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
590 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
591 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
592 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
593 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
594 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
595 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
596 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
597 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
598 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
599 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
601 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
602 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
603 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
604 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
606 if (separate_dirty_bits(cmd))
607 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
609 cmd->changed = false;
613 * The mutator updates the superblock flags.
615 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
616 flags_mutator mutator)
619 struct cache_disk_superblock *disk_super;
620 struct dm_block *sblock;
622 r = superblock_lock(cmd, &sblock);
626 disk_super = dm_block_data(sblock);
627 update_flags(disk_super, mutator);
628 read_superblock_fields(cmd, disk_super);
629 dm_bm_unlock(sblock);
631 return dm_bm_flush(cmd->bm);
634 static int __begin_transaction(struct dm_cache_metadata *cmd)
637 struct cache_disk_superblock *disk_super;
638 struct dm_block *sblock;
641 * We re-read the superblock every time. Shouldn't need to do this
644 r = superblock_read_lock(cmd, &sblock);
648 disk_super = dm_block_data(sblock);
649 read_superblock_fields(cmd, disk_super);
650 dm_bm_unlock(sblock);
655 static int __commit_transaction(struct dm_cache_metadata *cmd,
656 flags_mutator mutator)
659 struct cache_disk_superblock *disk_super;
660 struct dm_block *sblock;
663 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
665 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
667 if (separate_dirty_bits(cmd)) {
668 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
674 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
679 r = dm_tm_pre_commit(cmd->tm);
683 r = __save_sm_root(cmd);
687 r = superblock_lock(cmd, &sblock);
691 disk_super = dm_block_data(sblock);
693 disk_super->flags = cpu_to_le32(cmd->flags);
695 update_flags(disk_super, mutator);
697 disk_super->mapping_root = cpu_to_le64(cmd->root);
698 if (separate_dirty_bits(cmd))
699 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
700 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
701 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
702 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
703 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
704 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
705 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
706 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
707 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
708 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
709 disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
711 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
712 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
713 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
714 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
715 __copy_sm_root(cmd, disk_super);
717 return dm_tm_commit(cmd->tm, sblock);
720 /*----------------------------------------------------------------*/
723 * The mappings are held in a dm-array that has 64-bit values stored in
724 * little-endian format. The index is the cblock, the high 48bits of the
725 * value are the oblock and the low 16 bit the flags.
727 #define FLAGS_MASK ((1 << 16) - 1)
729 static __le64 pack_value(dm_oblock_t block, unsigned flags)
731 uint64_t value = from_oblock(block);
733 value = value | (flags & FLAGS_MASK);
734 return cpu_to_le64(value);
737 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
739 uint64_t value = le64_to_cpu(value_le);
740 uint64_t b = value >> 16;
741 *block = to_oblock(b);
742 *flags = value & FLAGS_MASK;
745 /*----------------------------------------------------------------*/
747 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
748 sector_t data_block_size,
749 bool may_format_device,
750 size_t policy_hint_size,
751 unsigned metadata_version)
754 struct dm_cache_metadata *cmd;
756 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
758 DMERR("could not allocate metadata struct");
759 return ERR_PTR(-ENOMEM);
762 cmd->version = metadata_version;
763 atomic_set(&cmd->ref_count, 1);
764 init_rwsem(&cmd->root_lock);
766 cmd->data_block_size = data_block_size;
767 cmd->cache_blocks = 0;
768 cmd->policy_hint_size = policy_hint_size;
770 cmd->fail_io = false;
772 r = __create_persistent_data_objects(cmd, may_format_device);
778 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
780 dm_cache_metadata_close(cmd);
788 * We keep a little list of ref counted metadata objects to prevent two
789 * different target instances creating separate bufio instances. This is
790 * an issue if a table is reloaded before the suspend.
792 static DEFINE_MUTEX(table_lock);
793 static LIST_HEAD(table);
795 static struct dm_cache_metadata *lookup(struct block_device *bdev)
797 struct dm_cache_metadata *cmd;
799 list_for_each_entry(cmd, &table, list)
800 if (cmd->bdev == bdev) {
801 atomic_inc(&cmd->ref_count);
808 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
809 sector_t data_block_size,
810 bool may_format_device,
811 size_t policy_hint_size,
812 unsigned metadata_version)
814 struct dm_cache_metadata *cmd, *cmd2;
816 mutex_lock(&table_lock);
818 mutex_unlock(&table_lock);
823 cmd = metadata_open(bdev, data_block_size, may_format_device,
824 policy_hint_size, metadata_version);
826 mutex_lock(&table_lock);
829 mutex_unlock(&table_lock);
830 __destroy_persistent_data_objects(cmd, true);
834 list_add(&cmd->list, &table);
835 mutex_unlock(&table_lock);
841 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
843 if (cmd->data_block_size != data_block_size) {
844 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
845 (unsigned long long) data_block_size,
846 (unsigned long long) cmd->data_block_size);
853 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
854 sector_t data_block_size,
855 bool may_format_device,
856 size_t policy_hint_size,
857 unsigned metadata_version)
859 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
860 policy_hint_size, metadata_version);
862 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
863 dm_cache_metadata_close(cmd);
864 return ERR_PTR(-EINVAL);
870 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
872 if (atomic_dec_and_test(&cmd->ref_count)) {
873 mutex_lock(&table_lock);
874 list_del(&cmd->list);
875 mutex_unlock(&table_lock);
878 __destroy_persistent_data_objects(cmd, true);
884 * Checks that the given cache block is either unmapped or clean.
886 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
894 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
898 unpack_value(value, &ob, &flags);
899 *result = !((flags & M_VALID) && (flags & M_DIRTY));
904 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
905 dm_cblock_t begin, dm_cblock_t end,
911 while (begin != end) {
912 r = block_clean_combined_dirty(cmd, begin, result);
914 DMERR("block_clean_combined_dirty failed");
919 DMERR("cache block %llu is dirty",
920 (unsigned long long) from_cblock(begin));
924 begin = to_cblock(from_cblock(begin) + 1);
930 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
931 dm_cblock_t begin, dm_cblock_t end,
938 if (from_cblock(cmd->cache_blocks) == 0)
942 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
943 from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
945 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
949 r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
951 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
952 dm_bitset_cursor_end(&cmd->dirty_cursor);
956 while (begin != end) {
958 * We assume that unmapped blocks have their dirty bit
961 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
963 DMERR("%s: cache block %llu is dirty", __func__,
964 (unsigned long long) from_cblock(begin));
965 dm_bitset_cursor_end(&cmd->dirty_cursor);
970 begin = to_cblock(from_cblock(begin) + 1);
974 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
976 DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
977 dm_bitset_cursor_end(&cmd->dirty_cursor);
982 dm_bitset_cursor_end(&cmd->dirty_cursor);
987 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
988 dm_cblock_t begin, dm_cblock_t end,
991 if (separate_dirty_bits(cmd))
992 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
994 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
997 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
999 down_write(&cmd->root_lock);
1000 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
1001 up_write(&cmd->root_lock);
1007 #define WRITE_LOCK(cmd) \
1009 if (!cmd_write_lock((cmd))) \
1013 #define WRITE_LOCK_VOID(cmd) \
1015 if (!cmd_write_lock((cmd))) \
1019 #define WRITE_UNLOCK(cmd) \
1020 up_write(&(cmd)->root_lock)
1022 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1024 down_read(&cmd->root_lock);
1026 up_read(&cmd->root_lock);
1032 #define READ_LOCK(cmd) \
1034 if (!cmd_read_lock((cmd))) \
1038 #define READ_LOCK_VOID(cmd) \
1040 if (!cmd_read_lock((cmd))) \
1044 #define READ_UNLOCK(cmd) \
1045 up_read(&(cmd)->root_lock)
1047 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1051 __le64 null_mapping = pack_value(0, 0);
1054 __dm_bless_for_disk(&null_mapping);
1056 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1057 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1059 __dm_unbless_for_disk(&null_mapping);
1064 DMERR("unable to shrink cache due to dirty blocks");
1066 __dm_unbless_for_disk(&null_mapping);
1071 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1072 from_cblock(new_cache_size),
1073 &null_mapping, &cmd->root);
1077 if (separate_dirty_bits(cmd)) {
1078 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1079 from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1080 false, &cmd->dirty_root);
1085 cmd->cache_blocks = new_cache_size;
1086 cmd->changed = true;
1094 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1095 sector_t discard_block_size,
1096 dm_dblock_t new_nr_entries)
1101 r = dm_bitset_resize(&cmd->discard_info,
1103 from_dblock(cmd->discard_nr_blocks),
1104 from_dblock(new_nr_entries),
1105 false, &cmd->discard_root);
1107 cmd->discard_block_size = discard_block_size;
1108 cmd->discard_nr_blocks = new_nr_entries;
1111 cmd->changed = true;
1117 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1119 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1120 from_dblock(b), &cmd->discard_root);
1123 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1125 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1126 from_dblock(b), &cmd->discard_root);
1129 static int __discard(struct dm_cache_metadata *cmd,
1130 dm_dblock_t dblock, bool discard)
1134 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1138 cmd->changed = true;
1142 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1143 dm_dblock_t dblock, bool discard)
1148 r = __discard(cmd, dblock, discard);
1154 static int __load_discards(struct dm_cache_metadata *cmd,
1155 load_discard_fn fn, void *context)
1159 struct dm_bitset_cursor c;
1161 if (from_dblock(cmd->discard_nr_blocks) == 0)
1165 if (cmd->clean_when_opened) {
1166 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1170 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1171 from_dblock(cmd->discard_nr_blocks), &c);
1175 for (b = 0; ; b++) {
1176 r = fn(context, cmd->discard_block_size, to_dblock(b),
1177 dm_bitset_cursor_get_value(&c));
1181 if (b >= (from_dblock(cmd->discard_nr_blocks) - 1))
1184 r = dm_bitset_cursor_next(&c);
1189 dm_bitset_cursor_end(&c);
1192 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1193 r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1202 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1203 load_discard_fn fn, void *context)
1208 r = __load_discards(cmd, fn, context);
1214 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1217 *result = cmd->cache_blocks;
1223 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1226 __le64 value = pack_value(0, 0);
1228 __dm_bless_for_disk(&value);
1229 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1230 &value, &cmd->root);
1234 cmd->changed = true;
1238 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1243 r = __remove(cmd, cblock);
1249 static int __insert(struct dm_cache_metadata *cmd,
1250 dm_cblock_t cblock, dm_oblock_t oblock)
1253 __le64 value = pack_value(oblock, M_VALID);
1254 __dm_bless_for_disk(&value);
1256 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1257 &value, &cmd->root);
1261 cmd->changed = true;
1265 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1266 dm_cblock_t cblock, dm_oblock_t oblock)
1271 r = __insert(cmd, cblock, oblock);
1281 struct dm_cache_metadata *cmd;
1282 bool respect_dirty_flags;
1286 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1287 struct dm_cache_policy *policy)
1289 const char *policy_name = dm_cache_policy_get_name(policy);
1290 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1291 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1294 * Ensure policy names match.
1296 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1300 * Ensure policy major versions match.
1302 if (cmd->policy_version[0] != policy_version[0])
1306 * Ensure policy hint sizes match.
1308 if (cmd->policy_hint_size != policy_hint_size)
1314 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1316 return cmd->hint_root && cmd->policy_hint_size;
1319 static bool hints_array_available(struct dm_cache_metadata *cmd,
1320 struct dm_cache_policy *policy)
1322 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1323 hints_array_initialized(cmd);
1326 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1327 uint64_t cb, bool hints_valid,
1328 struct dm_array_cursor *mapping_cursor,
1329 struct dm_array_cursor *hint_cursor,
1330 load_mapping_fn fn, void *context)
1337 __le64 *mapping_value_le;
1338 __le32 *hint_value_le;
1344 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1345 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1346 unpack_value(mapping, &oblock, &flags);
1348 if (flags & M_VALID) {
1350 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1351 memcpy(&hint, hint_value_le, sizeof(hint));
1353 if (cmd->clean_when_opened)
1354 dirty = flags & M_DIRTY;
1356 r = fn(context, oblock, to_cblock(cb), dirty,
1357 le32_to_cpu(hint), hints_valid);
1359 DMERR("policy couldn't load cache block %llu",
1360 (unsigned long long) from_cblock(to_cblock(cb)));
1367 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1368 uint64_t cb, bool hints_valid,
1369 struct dm_array_cursor *mapping_cursor,
1370 struct dm_array_cursor *hint_cursor,
1371 struct dm_bitset_cursor *dirty_cursor,
1372 load_mapping_fn fn, void *context)
1379 __le64 *mapping_value_le;
1380 __le32 *hint_value_le;
1386 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1387 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1388 unpack_value(mapping, &oblock, &flags);
1390 if (flags & M_VALID) {
1392 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1393 memcpy(&hint, hint_value_le, sizeof(hint));
1395 if (cmd->clean_when_opened)
1396 dirty = dm_bitset_cursor_get_value(dirty_cursor);
1398 r = fn(context, oblock, to_cblock(cb), dirty,
1399 le32_to_cpu(hint), hints_valid);
1401 DMERR("policy couldn't load cache block %llu",
1402 (unsigned long long) from_cblock(to_cblock(cb)));
1409 static int __load_mappings(struct dm_cache_metadata *cmd,
1410 struct dm_cache_policy *policy,
1411 load_mapping_fn fn, void *context)
1416 bool hints_valid = hints_array_available(cmd, policy);
1418 if (from_cblock(cmd->cache_blocks) == 0)
1422 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1427 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1429 dm_array_cursor_end(&cmd->mapping_cursor);
1434 if (separate_dirty_bits(cmd)) {
1435 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1436 from_cblock(cmd->cache_blocks),
1437 &cmd->dirty_cursor);
1439 dm_array_cursor_end(&cmd->hint_cursor);
1440 dm_array_cursor_end(&cmd->mapping_cursor);
1445 for (cb = 0; ; cb++) {
1446 if (separate_dirty_bits(cmd))
1447 r = __load_mapping_v2(cmd, cb, hints_valid,
1448 &cmd->mapping_cursor,
1453 r = __load_mapping_v1(cmd, cb, hints_valid,
1454 &cmd->mapping_cursor, &cmd->hint_cursor,
1460 * We need to break out before we move the cursors.
1462 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1465 r = dm_array_cursor_next(&cmd->mapping_cursor);
1467 DMERR("dm_array_cursor_next for mapping failed");
1472 r = dm_array_cursor_next(&cmd->hint_cursor);
1474 dm_array_cursor_end(&cmd->hint_cursor);
1475 hints_valid = false;
1479 if (separate_dirty_bits(cmd)) {
1480 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1482 DMERR("dm_bitset_cursor_next for dirty failed");
1488 dm_array_cursor_end(&cmd->mapping_cursor);
1490 dm_array_cursor_end(&cmd->hint_cursor);
1492 if (separate_dirty_bits(cmd))
1493 dm_bitset_cursor_end(&cmd->dirty_cursor);
1498 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1499 struct dm_cache_policy *policy,
1500 load_mapping_fn fn, void *context)
1505 r = __load_mappings(cmd, policy, fn, context);
1511 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 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 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,
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 *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);