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
30 #define CACHE_METADATA_CACHE_SIZE 64
33 * 3 for btree insert +
34 * 2 for btree lookup used within space map
36 #define CACHE_MAX_CONCURRENT_LOCKS 5
37 #define SPACE_MAP_ROOT_SIZE 128
39 enum superblock_flag_bits {
40 /* for spotting crashes that would invalidate the dirty bitset */
42 /* metadata must be checked using the tools */
47 * Each mapping from cache block -> origin block carries a set of flags.
51 * A valid mapping. Because we're using an array we clear this
52 * flag for an non existant mapping.
57 * The data on the cache is different from that on the origin.
58 * This flag is only used by metadata format 1.
63 struct cache_disk_superblock {
72 __u8 policy_name[CACHE_POLICY_NAME_SIZE];
73 __le32 policy_hint_size;
75 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
80 __le64 discard_block_size;
81 __le64 discard_nr_blocks;
83 __le32 data_block_size;
84 __le32 metadata_block_size;
88 __le32 compat_ro_flags;
89 __le32 incompat_flags;
96 __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
99 * Metadata format 2 fields.
104 struct dm_cache_metadata {
106 struct list_head list;
109 struct block_device *bdev;
110 struct dm_block_manager *bm;
111 struct dm_space_map *metadata_sm;
112 struct dm_transaction_manager *tm;
114 struct dm_array_info info;
115 struct dm_array_info hint_info;
116 struct dm_disk_bitset discard_info;
118 struct rw_semaphore root_lock;
121 dm_block_t hint_root;
122 dm_block_t discard_root;
124 sector_t discard_block_size;
125 dm_dblock_t discard_nr_blocks;
127 sector_t data_block_size;
128 dm_cblock_t cache_blocks;
130 bool clean_when_opened:1;
132 char policy_name[CACHE_POLICY_NAME_SIZE];
133 unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
134 size_t policy_hint_size;
135 struct dm_cache_statistics stats;
138 * Reading the space map root can fail, so we read it into this
139 * buffer before the superblock is locked and updated.
141 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
144 * Set if a transaction has to be aborted but the attempt to roll
145 * back to the previous (good) transaction failed. The only
146 * metadata operation permissible in this state is the closing of
152 * Metadata format 2 fields.
154 dm_block_t dirty_root;
155 struct dm_disk_bitset dirty_info;
158 * These structures are used when loading metadata. They're too
159 * big to put on the stack.
161 struct dm_array_cursor mapping_cursor;
162 struct dm_array_cursor hint_cursor;
163 struct dm_bitset_cursor dirty_cursor;
166 /*-------------------------------------------------------------------
167 * superblock validator
168 *-----------------------------------------------------------------*/
170 #define SUPERBLOCK_CSUM_XOR 9031977
172 static void sb_prepare_for_write(struct dm_block_validator *v,
174 size_t sb_block_size)
176 struct cache_disk_superblock *disk_super = dm_block_data(b);
178 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
179 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
180 sb_block_size - sizeof(__le32),
181 SUPERBLOCK_CSUM_XOR));
184 static int check_metadata_version(struct cache_disk_superblock *disk_super)
186 uint32_t metadata_version = le32_to_cpu(disk_super->version);
188 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
189 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
190 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
197 static int sb_check(struct dm_block_validator *v,
199 size_t sb_block_size)
201 struct cache_disk_superblock *disk_super = dm_block_data(b);
204 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
205 DMERR("sb_check failed: blocknr %llu: wanted %llu",
206 le64_to_cpu(disk_super->blocknr),
207 (unsigned long long)dm_block_location(b));
211 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
212 DMERR("sb_check failed: magic %llu: wanted %llu",
213 le64_to_cpu(disk_super->magic),
214 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
218 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
219 sb_block_size - sizeof(__le32),
220 SUPERBLOCK_CSUM_XOR));
221 if (csum_le != disk_super->csum) {
222 DMERR("sb_check failed: csum %u: wanted %u",
223 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
227 return check_metadata_version(disk_super);
230 static struct dm_block_validator sb_validator = {
231 .name = "superblock",
232 .prepare_for_write = sb_prepare_for_write,
236 /*----------------------------------------------------------------*/
238 static int superblock_read_lock(struct dm_cache_metadata *cmd,
239 struct dm_block **sblock)
241 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
242 &sb_validator, sblock);
245 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
246 struct dm_block **sblock)
248 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
249 &sb_validator, sblock);
252 static int superblock_lock(struct dm_cache_metadata *cmd,
253 struct dm_block **sblock)
255 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
256 &sb_validator, sblock);
259 /*----------------------------------------------------------------*/
261 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
266 __le64 *data_le, zero = cpu_to_le64(0);
267 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
270 * We can't use a validator here - it may be all zeroes.
272 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
276 data_le = dm_block_data(b);
278 for (i = 0; i < sb_block_size; i++) {
279 if (data_le[i] != zero) {
290 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
292 struct dm_btree_value_type vt;
295 vt.size = sizeof(__le64);
299 dm_array_info_init(&cmd->info, cmd->tm, &vt);
301 if (cmd->policy_hint_size) {
302 vt.size = sizeof(__le32);
303 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
307 static int __save_sm_root(struct dm_cache_metadata *cmd)
312 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
316 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
320 static void __copy_sm_root(struct dm_cache_metadata *cmd,
321 struct cache_disk_superblock *disk_super)
323 memcpy(&disk_super->metadata_space_map_root,
324 &cmd->metadata_space_map_root,
325 sizeof(cmd->metadata_space_map_root));
328 static bool separate_dirty_bits(struct dm_cache_metadata *cmd)
330 return cmd->version >= 2;
333 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
336 struct dm_block *sblock;
337 struct cache_disk_superblock *disk_super;
338 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
340 /* FIXME: see if we can lose the max sectors limit */
341 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
342 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
344 r = dm_tm_pre_commit(cmd->tm);
349 * dm_sm_copy_root() can fail. So we need to do it before we start
350 * updating the superblock.
352 r = __save_sm_root(cmd);
356 r = superblock_lock_zero(cmd, &sblock);
360 disk_super = dm_block_data(sblock);
361 disk_super->flags = 0;
362 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
363 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
364 disk_super->version = cpu_to_le32(cmd->version);
365 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
366 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
367 disk_super->policy_hint_size = 0;
369 __copy_sm_root(cmd, disk_super);
371 disk_super->mapping_root = cpu_to_le64(cmd->root);
372 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
373 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
374 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
375 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
376 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
377 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
378 disk_super->cache_blocks = cpu_to_le32(0);
380 disk_super->read_hits = cpu_to_le32(0);
381 disk_super->read_misses = cpu_to_le32(0);
382 disk_super->write_hits = cpu_to_le32(0);
383 disk_super->write_misses = cpu_to_le32(0);
385 if (separate_dirty_bits(cmd))
386 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
388 return dm_tm_commit(cmd->tm, sblock);
391 static int __format_metadata(struct dm_cache_metadata *cmd)
395 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
396 &cmd->tm, &cmd->metadata_sm);
398 DMERR("tm_create_with_sm failed");
402 __setup_mapping_info(cmd);
404 r = dm_array_empty(&cmd->info, &cmd->root);
408 if (separate_dirty_bits(cmd)) {
409 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
410 r = dm_bitset_empty(&cmd->dirty_info, &cmd->dirty_root);
415 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
416 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
420 cmd->discard_block_size = 0;
421 cmd->discard_nr_blocks = 0;
423 r = __write_initial_superblock(cmd);
427 cmd->clean_when_opened = true;
431 dm_tm_destroy(cmd->tm);
432 dm_sm_destroy(cmd->metadata_sm);
437 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
438 struct dm_cache_metadata *cmd)
440 uint32_t incompat_flags, features;
442 incompat_flags = le32_to_cpu(disk_super->incompat_flags);
443 features = incompat_flags & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
445 DMERR("could not access metadata due to unsupported optional features (%lx).",
446 (unsigned long)features);
451 * Check for read-only metadata to skip the following RDWR checks.
453 if (get_disk_ro(cmd->bdev->bd_disk))
456 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
458 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
459 (unsigned long)features);
466 static int __open_metadata(struct dm_cache_metadata *cmd)
469 struct dm_block *sblock;
470 struct cache_disk_superblock *disk_super;
471 unsigned long sb_flags;
473 r = superblock_read_lock(cmd, &sblock);
475 DMERR("couldn't read lock superblock");
479 disk_super = dm_block_data(sblock);
481 /* Verify the data block size hasn't changed */
482 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
483 DMERR("changing the data block size (from %u to %llu) is not supported",
484 le32_to_cpu(disk_super->data_block_size),
485 (unsigned long long)cmd->data_block_size);
490 r = __check_incompat_features(disk_super, cmd);
494 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
495 disk_super->metadata_space_map_root,
496 sizeof(disk_super->metadata_space_map_root),
497 &cmd->tm, &cmd->metadata_sm);
499 DMERR("tm_open_with_sm failed");
503 __setup_mapping_info(cmd);
504 dm_disk_bitset_init(cmd->tm, &cmd->dirty_info);
505 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
506 sb_flags = le32_to_cpu(disk_super->flags);
507 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
508 dm_bm_unlock(sblock);
513 dm_bm_unlock(sblock);
517 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
521 bool unformatted = false;
523 r = __superblock_all_zeroes(cmd->bm, &unformatted);
528 return format_device ? __format_metadata(cmd) : -EPERM;
530 return __open_metadata(cmd);
533 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
534 bool may_format_device)
537 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
538 CACHE_METADATA_CACHE_SIZE,
539 CACHE_MAX_CONCURRENT_LOCKS);
540 if (IS_ERR(cmd->bm)) {
541 DMERR("could not create block manager");
542 return PTR_ERR(cmd->bm);
545 r = __open_or_format_metadata(cmd, may_format_device);
547 dm_block_manager_destroy(cmd->bm);
552 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
554 dm_sm_destroy(cmd->metadata_sm);
555 dm_tm_destroy(cmd->tm);
556 dm_block_manager_destroy(cmd->bm);
559 typedef unsigned long (*flags_mutator)(unsigned long);
561 static void update_flags(struct cache_disk_superblock *disk_super,
562 flags_mutator mutator)
564 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
565 disk_super->flags = cpu_to_le32(sb_flags);
568 static unsigned long set_clean_shutdown(unsigned long flags)
570 set_bit(CLEAN_SHUTDOWN, &flags);
574 static unsigned long clear_clean_shutdown(unsigned long flags)
576 clear_bit(CLEAN_SHUTDOWN, &flags);
580 static void read_superblock_fields(struct dm_cache_metadata *cmd,
581 struct cache_disk_superblock *disk_super)
583 cmd->version = le32_to_cpu(disk_super->version);
584 cmd->flags = le32_to_cpu(disk_super->flags);
585 cmd->root = le64_to_cpu(disk_super->mapping_root);
586 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
587 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
588 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
589 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
590 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
591 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
592 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
593 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
594 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
595 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
596 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
598 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
599 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
600 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
601 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
603 if (separate_dirty_bits(cmd))
604 cmd->dirty_root = le64_to_cpu(disk_super->dirty_root);
606 cmd->changed = false;
610 * The mutator updates the superblock flags.
612 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
613 flags_mutator mutator)
616 struct cache_disk_superblock *disk_super;
617 struct dm_block *sblock;
619 r = superblock_lock(cmd, &sblock);
623 disk_super = dm_block_data(sblock);
624 update_flags(disk_super, mutator);
625 read_superblock_fields(cmd, disk_super);
626 dm_bm_unlock(sblock);
628 return dm_bm_flush(cmd->bm);
631 static int __begin_transaction(struct dm_cache_metadata *cmd)
634 struct cache_disk_superblock *disk_super;
635 struct dm_block *sblock;
638 * We re-read the superblock every time. Shouldn't need to do this
641 r = superblock_read_lock(cmd, &sblock);
645 disk_super = dm_block_data(sblock);
646 read_superblock_fields(cmd, disk_super);
647 dm_bm_unlock(sblock);
652 static int __commit_transaction(struct dm_cache_metadata *cmd,
653 flags_mutator mutator)
656 struct cache_disk_superblock *disk_super;
657 struct dm_block *sblock;
660 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
662 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
664 if (separate_dirty_bits(cmd)) {
665 r = dm_bitset_flush(&cmd->dirty_info, cmd->dirty_root,
671 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
676 r = dm_tm_pre_commit(cmd->tm);
680 r = __save_sm_root(cmd);
684 r = superblock_lock(cmd, &sblock);
688 disk_super = dm_block_data(sblock);
690 disk_super->flags = cpu_to_le32(cmd->flags);
692 update_flags(disk_super, mutator);
694 disk_super->mapping_root = cpu_to_le64(cmd->root);
695 if (separate_dirty_bits(cmd))
696 disk_super->dirty_root = cpu_to_le64(cmd->dirty_root);
697 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
698 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
699 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
700 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
701 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
702 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
703 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
704 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
705 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
707 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
708 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
709 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
710 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
711 __copy_sm_root(cmd, disk_super);
713 return dm_tm_commit(cmd->tm, sblock);
716 /*----------------------------------------------------------------*/
719 * The mappings are held in a dm-array that has 64-bit values stored in
720 * little-endian format. The index is the cblock, the high 48bits of the
721 * value are the oblock and the low 16 bit the flags.
723 #define FLAGS_MASK ((1 << 16) - 1)
725 static __le64 pack_value(dm_oblock_t block, unsigned flags)
727 uint64_t value = from_oblock(block);
729 value = value | (flags & FLAGS_MASK);
730 return cpu_to_le64(value);
733 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
735 uint64_t value = le64_to_cpu(value_le);
736 uint64_t b = value >> 16;
737 *block = to_oblock(b);
738 *flags = value & FLAGS_MASK;
741 /*----------------------------------------------------------------*/
743 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
744 sector_t data_block_size,
745 bool may_format_device,
746 size_t policy_hint_size,
747 unsigned metadata_version)
750 struct dm_cache_metadata *cmd;
752 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
754 DMERR("could not allocate metadata struct");
755 return ERR_PTR(-ENOMEM);
758 cmd->version = metadata_version;
759 atomic_set(&cmd->ref_count, 1);
760 init_rwsem(&cmd->root_lock);
762 cmd->data_block_size = data_block_size;
763 cmd->cache_blocks = 0;
764 cmd->policy_hint_size = policy_hint_size;
766 cmd->fail_io = false;
768 r = __create_persistent_data_objects(cmd, may_format_device);
774 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
776 dm_cache_metadata_close(cmd);
784 * We keep a little list of ref counted metadata objects to prevent two
785 * different target instances creating separate bufio instances. This is
786 * an issue if a table is reloaded before the suspend.
788 static DEFINE_MUTEX(table_lock);
789 static LIST_HEAD(table);
791 static struct dm_cache_metadata *lookup(struct block_device *bdev)
793 struct dm_cache_metadata *cmd;
795 list_for_each_entry(cmd, &table, list)
796 if (cmd->bdev == bdev) {
797 atomic_inc(&cmd->ref_count);
804 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
805 sector_t data_block_size,
806 bool may_format_device,
807 size_t policy_hint_size,
808 unsigned metadata_version)
810 struct dm_cache_metadata *cmd, *cmd2;
812 mutex_lock(&table_lock);
814 mutex_unlock(&table_lock);
819 cmd = metadata_open(bdev, data_block_size, may_format_device,
820 policy_hint_size, metadata_version);
822 mutex_lock(&table_lock);
825 mutex_unlock(&table_lock);
826 __destroy_persistent_data_objects(cmd);
830 list_add(&cmd->list, &table);
831 mutex_unlock(&table_lock);
837 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
839 if (cmd->data_block_size != data_block_size) {
840 DMERR("data_block_size (%llu) different from that in metadata (%llu)",
841 (unsigned long long) data_block_size,
842 (unsigned long long) cmd->data_block_size);
849 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
850 sector_t data_block_size,
851 bool may_format_device,
852 size_t policy_hint_size,
853 unsigned metadata_version)
855 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size, may_format_device,
856 policy_hint_size, metadata_version);
858 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
859 dm_cache_metadata_close(cmd);
860 return ERR_PTR(-EINVAL);
866 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
868 if (atomic_dec_and_test(&cmd->ref_count)) {
869 mutex_lock(&table_lock);
870 list_del(&cmd->list);
871 mutex_unlock(&table_lock);
874 __destroy_persistent_data_objects(cmd);
880 * Checks that the given cache block is either unmapped or clean.
882 static int block_clean_combined_dirty(struct dm_cache_metadata *cmd, dm_cblock_t b,
890 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
894 unpack_value(value, &ob, &flags);
895 *result = !((flags & M_VALID) && (flags & M_DIRTY));
900 static int blocks_are_clean_combined_dirty(struct dm_cache_metadata *cmd,
901 dm_cblock_t begin, dm_cblock_t end,
907 while (begin != end) {
908 r = block_clean_combined_dirty(cmd, begin, result);
910 DMERR("block_clean_combined_dirty failed");
915 DMERR("cache block %llu is dirty",
916 (unsigned long long) from_cblock(begin));
920 begin = to_cblock(from_cblock(begin) + 1);
926 static int blocks_are_clean_separate_dirty(struct dm_cache_metadata *cmd,
927 dm_cblock_t begin, dm_cblock_t end,
934 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
935 from_cblock(cmd->cache_blocks), &cmd->dirty_cursor);
937 DMERR("%s: dm_bitset_cursor_begin for dirty failed", __func__);
941 r = dm_bitset_cursor_skip(&cmd->dirty_cursor, from_cblock(begin));
943 DMERR("%s: dm_bitset_cursor_skip for dirty failed", __func__);
944 dm_bitset_cursor_end(&cmd->dirty_cursor);
948 while (begin != end) {
950 * We assume that unmapped blocks have their dirty bit
953 dirty_flag = dm_bitset_cursor_get_value(&cmd->dirty_cursor);
955 DMERR("%s: cache block %llu is dirty", __func__,
956 (unsigned long long) from_cblock(begin));
957 dm_bitset_cursor_end(&cmd->dirty_cursor);
962 begin = to_cblock(from_cblock(begin) + 1);
966 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
968 DMERR("%s: dm_bitset_cursor_next for dirty failed", __func__);
969 dm_bitset_cursor_end(&cmd->dirty_cursor);
974 dm_bitset_cursor_end(&cmd->dirty_cursor);
979 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
980 dm_cblock_t begin, dm_cblock_t end,
983 if (separate_dirty_bits(cmd))
984 return blocks_are_clean_separate_dirty(cmd, begin, end, result);
986 return blocks_are_clean_combined_dirty(cmd, begin, end, result);
989 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
991 down_write(&cmd->root_lock);
992 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
993 up_write(&cmd->root_lock);
999 #define WRITE_LOCK(cmd) \
1001 if (!cmd_write_lock((cmd))) \
1005 #define WRITE_LOCK_VOID(cmd) \
1007 if (!cmd_write_lock((cmd))) \
1011 #define WRITE_UNLOCK(cmd) \
1012 up_write(&(cmd)->root_lock)
1014 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
1016 down_read(&cmd->root_lock);
1018 up_read(&cmd->root_lock);
1024 #define READ_LOCK(cmd) \
1026 if (!cmd_read_lock((cmd))) \
1030 #define READ_LOCK_VOID(cmd) \
1032 if (!cmd_read_lock((cmd))) \
1036 #define READ_UNLOCK(cmd) \
1037 up_read(&(cmd)->root_lock)
1039 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
1043 __le64 null_mapping = pack_value(0, 0);
1046 __dm_bless_for_disk(&null_mapping);
1048 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
1049 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
1051 __dm_unbless_for_disk(&null_mapping);
1056 DMERR("unable to shrink cache due to dirty blocks");
1058 __dm_unbless_for_disk(&null_mapping);
1063 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
1064 from_cblock(new_cache_size),
1065 &null_mapping, &cmd->root);
1069 if (separate_dirty_bits(cmd)) {
1070 r = dm_bitset_resize(&cmd->dirty_info, cmd->dirty_root,
1071 from_cblock(cmd->cache_blocks), from_cblock(new_cache_size),
1072 false, &cmd->dirty_root);
1077 cmd->cache_blocks = new_cache_size;
1078 cmd->changed = true;
1086 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
1087 sector_t discard_block_size,
1088 dm_dblock_t new_nr_entries)
1093 r = dm_bitset_resize(&cmd->discard_info,
1095 from_dblock(cmd->discard_nr_blocks),
1096 from_dblock(new_nr_entries),
1097 false, &cmd->discard_root);
1099 cmd->discard_block_size = discard_block_size;
1100 cmd->discard_nr_blocks = new_nr_entries;
1103 cmd->changed = true;
1109 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1111 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
1112 from_dblock(b), &cmd->discard_root);
1115 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
1117 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
1118 from_dblock(b), &cmd->discard_root);
1121 static int __discard(struct dm_cache_metadata *cmd,
1122 dm_dblock_t dblock, bool discard)
1126 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1130 cmd->changed = true;
1134 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1135 dm_dblock_t dblock, bool discard)
1140 r = __discard(cmd, dblock, discard);
1146 static int __load_discards(struct dm_cache_metadata *cmd,
1147 load_discard_fn fn, void *context)
1151 struct dm_bitset_cursor c;
1153 if (from_dblock(cmd->discard_nr_blocks) == 0)
1157 if (cmd->clean_when_opened) {
1158 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root, &cmd->discard_root);
1162 r = dm_bitset_cursor_begin(&cmd->discard_info, cmd->discard_root,
1163 from_dblock(cmd->discard_nr_blocks), &c);
1167 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1168 r = fn(context, cmd->discard_block_size, to_dblock(b),
1169 dm_bitset_cursor_get_value(&c));
1174 dm_bitset_cursor_end(&c);
1177 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1178 r = fn(context, cmd->discard_block_size, to_dblock(b), false);
1187 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1188 load_discard_fn fn, void *context)
1193 r = __load_discards(cmd, fn, context);
1199 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1202 *result = cmd->cache_blocks;
1208 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1211 __le64 value = pack_value(0, 0);
1213 __dm_bless_for_disk(&value);
1214 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1215 &value, &cmd->root);
1219 cmd->changed = true;
1223 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1228 r = __remove(cmd, cblock);
1234 static int __insert(struct dm_cache_metadata *cmd,
1235 dm_cblock_t cblock, dm_oblock_t oblock)
1238 __le64 value = pack_value(oblock, M_VALID);
1239 __dm_bless_for_disk(&value);
1241 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1242 &value, &cmd->root);
1246 cmd->changed = true;
1250 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1251 dm_cblock_t cblock, dm_oblock_t oblock)
1256 r = __insert(cmd, cblock, oblock);
1266 struct dm_cache_metadata *cmd;
1267 bool respect_dirty_flags;
1271 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1272 struct dm_cache_policy *policy)
1274 const char *policy_name = dm_cache_policy_get_name(policy);
1275 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1276 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1279 * Ensure policy names match.
1281 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1285 * Ensure policy major versions match.
1287 if (cmd->policy_version[0] != policy_version[0])
1291 * Ensure policy hint sizes match.
1293 if (cmd->policy_hint_size != policy_hint_size)
1299 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1301 return cmd->hint_root && cmd->policy_hint_size;
1304 static bool hints_array_available(struct dm_cache_metadata *cmd,
1305 struct dm_cache_policy *policy)
1307 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1308 hints_array_initialized(cmd);
1311 static int __load_mapping_v1(struct dm_cache_metadata *cmd,
1312 uint64_t cb, bool hints_valid,
1313 struct dm_array_cursor *mapping_cursor,
1314 struct dm_array_cursor *hint_cursor,
1315 load_mapping_fn fn, void *context)
1322 __le64 *mapping_value_le;
1323 __le32 *hint_value_le;
1328 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1329 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1330 unpack_value(mapping, &oblock, &flags);
1332 if (flags & M_VALID) {
1334 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1335 memcpy(&hint, hint_value_le, sizeof(hint));
1338 r = fn(context, oblock, to_cblock(cb), flags & M_DIRTY,
1339 le32_to_cpu(hint), hints_valid);
1341 DMERR("policy couldn't load cache block %llu",
1342 (unsigned long long) from_cblock(to_cblock(cb)));
1349 static int __load_mapping_v2(struct dm_cache_metadata *cmd,
1350 uint64_t cb, bool hints_valid,
1351 struct dm_array_cursor *mapping_cursor,
1352 struct dm_array_cursor *hint_cursor,
1353 struct dm_bitset_cursor *dirty_cursor,
1354 load_mapping_fn fn, void *context)
1361 __le64 *mapping_value_le;
1362 __le32 *hint_value_le;
1368 dm_array_cursor_get_value(mapping_cursor, (void **) &mapping_value_le);
1369 memcpy(&mapping, mapping_value_le, sizeof(mapping));
1370 unpack_value(mapping, &oblock, &flags);
1372 if (flags & M_VALID) {
1374 dm_array_cursor_get_value(hint_cursor, (void **) &hint_value_le);
1375 memcpy(&hint, hint_value_le, sizeof(hint));
1378 dirty = dm_bitset_cursor_get_value(dirty_cursor);
1379 r = fn(context, oblock, to_cblock(cb), dirty,
1380 le32_to_cpu(hint), hints_valid);
1382 DMERR("policy couldn't load cache block %llu",
1383 (unsigned long long) from_cblock(to_cblock(cb)));
1390 static int __load_mappings(struct dm_cache_metadata *cmd,
1391 struct dm_cache_policy *policy,
1392 load_mapping_fn fn, void *context)
1397 bool hints_valid = hints_array_available(cmd, policy);
1399 if (from_cblock(cmd->cache_blocks) == 0)
1403 r = dm_array_cursor_begin(&cmd->info, cmd->root, &cmd->mapping_cursor);
1408 r = dm_array_cursor_begin(&cmd->hint_info, cmd->hint_root, &cmd->hint_cursor);
1410 dm_array_cursor_end(&cmd->mapping_cursor);
1415 if (separate_dirty_bits(cmd)) {
1416 r = dm_bitset_cursor_begin(&cmd->dirty_info, cmd->dirty_root,
1417 from_cblock(cmd->cache_blocks),
1418 &cmd->dirty_cursor);
1420 dm_array_cursor_end(&cmd->hint_cursor);
1421 dm_array_cursor_end(&cmd->mapping_cursor);
1426 for (cb = 0; ; cb++) {
1427 if (separate_dirty_bits(cmd))
1428 r = __load_mapping_v2(cmd, cb, hints_valid,
1429 &cmd->mapping_cursor,
1434 r = __load_mapping_v1(cmd, cb, hints_valid,
1435 &cmd->mapping_cursor, &cmd->hint_cursor,
1441 * We need to break out before we move the cursors.
1443 if (cb >= (from_cblock(cmd->cache_blocks) - 1))
1446 r = dm_array_cursor_next(&cmd->mapping_cursor);
1448 DMERR("dm_array_cursor_next for mapping failed");
1453 r = dm_array_cursor_next(&cmd->hint_cursor);
1455 DMERR("dm_array_cursor_next for hint failed");
1460 if (separate_dirty_bits(cmd)) {
1461 r = dm_bitset_cursor_next(&cmd->dirty_cursor);
1463 DMERR("dm_bitset_cursor_next for dirty failed");
1469 dm_array_cursor_end(&cmd->mapping_cursor);
1471 dm_array_cursor_end(&cmd->hint_cursor);
1473 if (separate_dirty_bits(cmd))
1474 dm_bitset_cursor_end(&cmd->dirty_cursor);
1479 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1480 struct dm_cache_policy *policy,
1481 load_mapping_fn fn, void *context)
1486 r = __load_mappings(cmd, policy, fn, context);
1492 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1499 memcpy(&value, leaf, sizeof(value));
1500 unpack_value(value, &oblock, &flags);
1505 static int __dump_mappings(struct dm_cache_metadata *cmd)
1507 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1510 void dm_cache_dump(struct dm_cache_metadata *cmd)
1512 READ_LOCK_VOID(cmd);
1513 __dump_mappings(cmd);
1517 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1528 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1535 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1539 unpack_value(value, &oblock, &flags);
1541 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1542 /* nothing to be done */
1545 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1546 __dm_bless_for_disk(&value);
1548 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1549 &value, &cmd->root);
1553 cmd->changed = true;
1558 static int __set_dirty_bits_v1(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1562 for (i = 0; i < nr_bits; i++) {
1563 r = __dirty(cmd, to_cblock(i), test_bit(i, bits));
1571 static int is_dirty_callback(uint32_t index, bool *value, void *context)
1573 unsigned long *bits = context;
1574 *value = test_bit(index, bits);
1578 static int __set_dirty_bits_v2(struct dm_cache_metadata *cmd, unsigned nr_bits, unsigned long *bits)
1582 /* nr_bits is really just a sanity check */
1583 if (nr_bits != from_cblock(cmd->cache_blocks)) {
1584 DMERR("dirty bitset is wrong size");
1588 r = dm_bitset_del(&cmd->dirty_info, cmd->dirty_root);
1592 cmd->changed = true;
1593 return dm_bitset_new(&cmd->dirty_info, &cmd->dirty_root, nr_bits, is_dirty_callback, bits);
1596 int dm_cache_set_dirty_bits(struct dm_cache_metadata *cmd,
1598 unsigned long *bits)
1603 if (separate_dirty_bits(cmd))
1604 r = __set_dirty_bits_v2(cmd, nr_bits, bits);
1606 r = __set_dirty_bits_v1(cmd, nr_bits, bits);
1612 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1613 struct dm_cache_statistics *stats)
1615 READ_LOCK_VOID(cmd);
1616 *stats = cmd->stats;
1620 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1621 struct dm_cache_statistics *stats)
1623 WRITE_LOCK_VOID(cmd);
1624 cmd->stats = *stats;
1628 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1631 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1632 clear_clean_shutdown);
1635 r = __commit_transaction(cmd, mutator);
1639 r = __begin_transaction(cmd);
1646 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1652 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1658 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1664 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1670 /*----------------------------------------------------------------*/
1672 static int get_hint(uint32_t index, void *value_le, void *context)
1675 struct dm_cache_policy *policy = context;
1677 value = policy_get_hint(policy, to_cblock(index));
1678 *((__le32 *) value_le) = cpu_to_le32(value);
1684 * It's quicker to always delete the hint array, and recreate with
1687 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1691 const char *policy_name = dm_cache_policy_get_name(policy);
1692 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1694 if (!policy_name[0] ||
1695 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1698 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1699 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1701 hint_size = dm_cache_policy_get_hint_size(policy);
1703 return 0; /* short-circuit hints initialization */
1704 cmd->policy_hint_size = hint_size;
1706 if (cmd->hint_root) {
1707 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1712 return dm_array_new(&cmd->hint_info, &cmd->hint_root,
1713 from_cblock(cmd->cache_blocks),
1717 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1722 r = write_hints(cmd, policy);
1728 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1733 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1739 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1741 WRITE_LOCK_VOID(cmd);
1742 dm_bm_set_read_only(cmd->bm);
1746 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1748 WRITE_LOCK_VOID(cmd);
1749 dm_bm_set_read_write(cmd->bm);
1753 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1756 struct dm_block *sblock;
1757 struct cache_disk_superblock *disk_super;
1760 set_bit(NEEDS_CHECK, &cmd->flags);
1762 r = superblock_lock(cmd, &sblock);
1764 DMERR("couldn't read superblock");
1768 disk_super = dm_block_data(sblock);
1769 disk_super->flags = cpu_to_le32(cmd->flags);
1771 dm_bm_unlock(sblock);
1778 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1781 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1787 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1792 __destroy_persistent_data_objects(cmd);
1793 r = __create_persistent_data_objects(cmd, false);
1795 cmd->fail_io = true;