2 * Copyright (C) 2012 Red Hat, Inc.
4 * This file is released under the GPL.
7 #include "dm-cache-metadata.h"
9 #include "persistent-data/dm-array.h"
10 #include "persistent-data/dm-bitset.h"
11 #include "persistent-data/dm-space-map.h"
12 #include "persistent-data/dm-space-map-disk.h"
13 #include "persistent-data/dm-transaction-manager.h"
15 #include <linux/device-mapper.h>
17 /*----------------------------------------------------------------*/
19 #define DM_MSG_PREFIX "cache metadata"
21 #define CACHE_SUPERBLOCK_MAGIC 06142003
22 #define CACHE_SUPERBLOCK_LOCATION 0
25 * defines a range of metadata versions that this module can handle.
27 #define MIN_CACHE_VERSION 1
28 #define MAX_CACHE_VERSION 1
30 #define CACHE_METADATA_CACHE_SIZE 64
33 * 3 for btree insert +
34 * 2 for btree lookup used within space map
36 #define CACHE_MAX_CONCURRENT_LOCKS 5
37 #define SPACE_MAP_ROOT_SIZE 128
39 enum superblock_flag_bits {
40 /* for spotting crashes that would invalidate the dirty bitset */
42 /* metadata must be checked using the tools */
47 * Each mapping from cache block -> origin block carries a set of flags.
51 * A valid mapping. Because we're using an array we clear this
52 * flag for an non existant mapping.
57 * The data on the cache is different from that on the origin.
62 struct cache_disk_superblock {
71 __u8 policy_name[CACHE_POLICY_NAME_SIZE];
72 __le32 policy_hint_size;
74 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
79 __le64 discard_block_size;
80 __le64 discard_nr_blocks;
82 __le32 data_block_size;
83 __le32 metadata_block_size;
87 __le32 compat_ro_flags;
88 __le32 incompat_flags;
95 __le32 policy_version[CACHE_POLICY_VERSION_SIZE];
98 struct dm_cache_metadata {
100 struct list_head list;
102 struct block_device *bdev;
103 struct dm_block_manager *bm;
104 struct dm_space_map *metadata_sm;
105 struct dm_transaction_manager *tm;
107 struct dm_array_info info;
108 struct dm_array_info hint_info;
109 struct dm_disk_bitset discard_info;
111 struct rw_semaphore root_lock;
114 dm_block_t hint_root;
115 dm_block_t discard_root;
117 sector_t discard_block_size;
118 dm_dblock_t discard_nr_blocks;
120 sector_t data_block_size;
121 dm_cblock_t cache_blocks;
123 bool clean_when_opened:1;
125 char policy_name[CACHE_POLICY_NAME_SIZE];
126 unsigned policy_version[CACHE_POLICY_VERSION_SIZE];
127 size_t policy_hint_size;
128 struct dm_cache_statistics stats;
131 * Reading the space map root can fail, so we read it into this
132 * buffer before the superblock is locked and updated.
134 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
137 * Set if a transaction has to be aborted but the attempt to roll
138 * back to the previous (good) transaction failed. The only
139 * metadata operation permissible in this state is the closing of
145 /*-------------------------------------------------------------------
146 * superblock validator
147 *-----------------------------------------------------------------*/
149 #define SUPERBLOCK_CSUM_XOR 9031977
151 static void sb_prepare_for_write(struct dm_block_validator *v,
153 size_t sb_block_size)
155 struct cache_disk_superblock *disk_super = dm_block_data(b);
157 disk_super->blocknr = cpu_to_le64(dm_block_location(b));
158 disk_super->csum = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
159 sb_block_size - sizeof(__le32),
160 SUPERBLOCK_CSUM_XOR));
163 static int check_metadata_version(struct cache_disk_superblock *disk_super)
165 uint32_t metadata_version = le32_to_cpu(disk_super->version);
166 if (metadata_version < MIN_CACHE_VERSION || metadata_version > MAX_CACHE_VERSION) {
167 DMERR("Cache metadata version %u found, but only versions between %u and %u supported.",
168 metadata_version, MIN_CACHE_VERSION, MAX_CACHE_VERSION);
175 static int sb_check(struct dm_block_validator *v,
177 size_t sb_block_size)
179 struct cache_disk_superblock *disk_super = dm_block_data(b);
182 if (dm_block_location(b) != le64_to_cpu(disk_super->blocknr)) {
183 DMERR("sb_check failed: blocknr %llu: wanted %llu",
184 le64_to_cpu(disk_super->blocknr),
185 (unsigned long long)dm_block_location(b));
189 if (le64_to_cpu(disk_super->magic) != CACHE_SUPERBLOCK_MAGIC) {
190 DMERR("sb_check failed: magic %llu: wanted %llu",
191 le64_to_cpu(disk_super->magic),
192 (unsigned long long)CACHE_SUPERBLOCK_MAGIC);
196 csum_le = cpu_to_le32(dm_bm_checksum(&disk_super->flags,
197 sb_block_size - sizeof(__le32),
198 SUPERBLOCK_CSUM_XOR));
199 if (csum_le != disk_super->csum) {
200 DMERR("sb_check failed: csum %u: wanted %u",
201 le32_to_cpu(csum_le), le32_to_cpu(disk_super->csum));
205 return check_metadata_version(disk_super);
208 static struct dm_block_validator sb_validator = {
209 .name = "superblock",
210 .prepare_for_write = sb_prepare_for_write,
214 /*----------------------------------------------------------------*/
216 static int superblock_read_lock(struct dm_cache_metadata *cmd,
217 struct dm_block **sblock)
219 return dm_bm_read_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
220 &sb_validator, sblock);
223 static int superblock_lock_zero(struct dm_cache_metadata *cmd,
224 struct dm_block **sblock)
226 return dm_bm_write_lock_zero(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
227 &sb_validator, sblock);
230 static int superblock_lock(struct dm_cache_metadata *cmd,
231 struct dm_block **sblock)
233 return dm_bm_write_lock(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
234 &sb_validator, sblock);
237 /*----------------------------------------------------------------*/
239 static int __superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
244 __le64 *data_le, zero = cpu_to_le64(0);
245 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
248 * We can't use a validator here - it may be all zeroes.
250 r = dm_bm_read_lock(bm, CACHE_SUPERBLOCK_LOCATION, NULL, &b);
254 data_le = dm_block_data(b);
256 for (i = 0; i < sb_block_size; i++) {
257 if (data_le[i] != zero) {
268 static void __setup_mapping_info(struct dm_cache_metadata *cmd)
270 struct dm_btree_value_type vt;
273 vt.size = sizeof(__le64);
277 dm_array_info_init(&cmd->info, cmd->tm, &vt);
279 if (cmd->policy_hint_size) {
280 vt.size = sizeof(__le32);
281 dm_array_info_init(&cmd->hint_info, cmd->tm, &vt);
285 static int __save_sm_root(struct dm_cache_metadata *cmd)
290 r = dm_sm_root_size(cmd->metadata_sm, &metadata_len);
294 return dm_sm_copy_root(cmd->metadata_sm, &cmd->metadata_space_map_root,
298 static void __copy_sm_root(struct dm_cache_metadata *cmd,
299 struct cache_disk_superblock *disk_super)
301 memcpy(&disk_super->metadata_space_map_root,
302 &cmd->metadata_space_map_root,
303 sizeof(cmd->metadata_space_map_root));
306 static int __write_initial_superblock(struct dm_cache_metadata *cmd)
309 struct dm_block *sblock;
310 struct cache_disk_superblock *disk_super;
311 sector_t bdev_size = i_size_read(cmd->bdev->bd_inode) >> SECTOR_SHIFT;
313 /* FIXME: see if we can lose the max sectors limit */
314 if (bdev_size > DM_CACHE_METADATA_MAX_SECTORS)
315 bdev_size = DM_CACHE_METADATA_MAX_SECTORS;
317 r = dm_tm_pre_commit(cmd->tm);
322 * dm_sm_copy_root() can fail. So we need to do it before we start
323 * updating the superblock.
325 r = __save_sm_root(cmd);
329 r = superblock_lock_zero(cmd, &sblock);
333 disk_super = dm_block_data(sblock);
334 disk_super->flags = 0;
335 memset(disk_super->uuid, 0, sizeof(disk_super->uuid));
336 disk_super->magic = cpu_to_le64(CACHE_SUPERBLOCK_MAGIC);
337 disk_super->version = cpu_to_le32(MAX_CACHE_VERSION);
338 memset(disk_super->policy_name, 0, sizeof(disk_super->policy_name));
339 memset(disk_super->policy_version, 0, sizeof(disk_super->policy_version));
340 disk_super->policy_hint_size = cpu_to_le32(0);
342 __copy_sm_root(cmd, disk_super);
344 disk_super->mapping_root = cpu_to_le64(cmd->root);
345 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
346 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
347 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
348 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
349 disk_super->metadata_block_size = cpu_to_le32(DM_CACHE_METADATA_BLOCK_SIZE);
350 disk_super->data_block_size = cpu_to_le32(cmd->data_block_size);
351 disk_super->cache_blocks = cpu_to_le32(0);
353 disk_super->read_hits = cpu_to_le32(0);
354 disk_super->read_misses = cpu_to_le32(0);
355 disk_super->write_hits = cpu_to_le32(0);
356 disk_super->write_misses = cpu_to_le32(0);
358 return dm_tm_commit(cmd->tm, sblock);
361 static int __format_metadata(struct dm_cache_metadata *cmd)
365 r = dm_tm_create_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
366 &cmd->tm, &cmd->metadata_sm);
368 DMERR("tm_create_with_sm failed");
372 __setup_mapping_info(cmd);
374 r = dm_array_empty(&cmd->info, &cmd->root);
378 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
380 r = dm_bitset_empty(&cmd->discard_info, &cmd->discard_root);
384 cmd->discard_block_size = 0;
385 cmd->discard_nr_blocks = 0;
387 r = __write_initial_superblock(cmd);
391 cmd->clean_when_opened = true;
395 dm_tm_destroy(cmd->tm);
396 dm_sm_destroy(cmd->metadata_sm);
401 static int __check_incompat_features(struct cache_disk_superblock *disk_super,
402 struct dm_cache_metadata *cmd)
406 features = le32_to_cpu(disk_super->incompat_flags) & ~DM_CACHE_FEATURE_INCOMPAT_SUPP;
408 DMERR("could not access metadata due to unsupported optional features (%lx).",
409 (unsigned long)features);
414 * Check for read-only metadata to skip the following RDWR checks.
416 if (get_disk_ro(cmd->bdev->bd_disk))
419 features = le32_to_cpu(disk_super->compat_ro_flags) & ~DM_CACHE_FEATURE_COMPAT_RO_SUPP;
421 DMERR("could not access metadata RDWR due to unsupported optional features (%lx).",
422 (unsigned long)features);
429 static int __open_metadata(struct dm_cache_metadata *cmd)
432 struct dm_block *sblock;
433 struct cache_disk_superblock *disk_super;
434 unsigned long sb_flags;
436 r = superblock_read_lock(cmd, &sblock);
438 DMERR("couldn't read lock superblock");
442 disk_super = dm_block_data(sblock);
444 /* Verify the data block size hasn't changed */
445 if (le32_to_cpu(disk_super->data_block_size) != cmd->data_block_size) {
446 DMERR("changing the data block size (from %u to %llu) is not supported",
447 le32_to_cpu(disk_super->data_block_size),
448 (unsigned long long)cmd->data_block_size);
453 r = __check_incompat_features(disk_super, cmd);
457 r = dm_tm_open_with_sm(cmd->bm, CACHE_SUPERBLOCK_LOCATION,
458 disk_super->metadata_space_map_root,
459 sizeof(disk_super->metadata_space_map_root),
460 &cmd->tm, &cmd->metadata_sm);
462 DMERR("tm_open_with_sm failed");
466 __setup_mapping_info(cmd);
467 dm_disk_bitset_init(cmd->tm, &cmd->discard_info);
468 sb_flags = le32_to_cpu(disk_super->flags);
469 cmd->clean_when_opened = test_bit(CLEAN_SHUTDOWN, &sb_flags);
470 dm_bm_unlock(sblock);
475 dm_bm_unlock(sblock);
479 static int __open_or_format_metadata(struct dm_cache_metadata *cmd,
483 bool unformatted = false;
485 r = __superblock_all_zeroes(cmd->bm, &unformatted);
490 return format_device ? __format_metadata(cmd) : -EPERM;
492 return __open_metadata(cmd);
495 static int __create_persistent_data_objects(struct dm_cache_metadata *cmd,
496 bool may_format_device)
499 cmd->bm = dm_block_manager_create(cmd->bdev, DM_CACHE_METADATA_BLOCK_SIZE << SECTOR_SHIFT,
500 CACHE_METADATA_CACHE_SIZE,
501 CACHE_MAX_CONCURRENT_LOCKS);
502 if (IS_ERR(cmd->bm)) {
503 DMERR("could not create block manager");
504 r = PTR_ERR(cmd->bm);
509 r = __open_or_format_metadata(cmd, may_format_device);
511 dm_block_manager_destroy(cmd->bm);
518 static void __destroy_persistent_data_objects(struct dm_cache_metadata *cmd)
520 dm_sm_destroy(cmd->metadata_sm);
521 dm_tm_destroy(cmd->tm);
522 dm_block_manager_destroy(cmd->bm);
525 typedef unsigned long (*flags_mutator)(unsigned long);
527 static void update_flags(struct cache_disk_superblock *disk_super,
528 flags_mutator mutator)
530 uint32_t sb_flags = mutator(le32_to_cpu(disk_super->flags));
531 disk_super->flags = cpu_to_le32(sb_flags);
534 static unsigned long set_clean_shutdown(unsigned long flags)
536 set_bit(CLEAN_SHUTDOWN, &flags);
540 static unsigned long clear_clean_shutdown(unsigned long flags)
542 clear_bit(CLEAN_SHUTDOWN, &flags);
546 static void read_superblock_fields(struct dm_cache_metadata *cmd,
547 struct cache_disk_superblock *disk_super)
549 cmd->flags = le32_to_cpu(disk_super->flags);
550 cmd->root = le64_to_cpu(disk_super->mapping_root);
551 cmd->hint_root = le64_to_cpu(disk_super->hint_root);
552 cmd->discard_root = le64_to_cpu(disk_super->discard_root);
553 cmd->discard_block_size = le64_to_cpu(disk_super->discard_block_size);
554 cmd->discard_nr_blocks = to_dblock(le64_to_cpu(disk_super->discard_nr_blocks));
555 cmd->data_block_size = le32_to_cpu(disk_super->data_block_size);
556 cmd->cache_blocks = to_cblock(le32_to_cpu(disk_super->cache_blocks));
557 strncpy(cmd->policy_name, disk_super->policy_name, sizeof(cmd->policy_name));
558 cmd->policy_version[0] = le32_to_cpu(disk_super->policy_version[0]);
559 cmd->policy_version[1] = le32_to_cpu(disk_super->policy_version[1]);
560 cmd->policy_version[2] = le32_to_cpu(disk_super->policy_version[2]);
561 cmd->policy_hint_size = le32_to_cpu(disk_super->policy_hint_size);
563 cmd->stats.read_hits = le32_to_cpu(disk_super->read_hits);
564 cmd->stats.read_misses = le32_to_cpu(disk_super->read_misses);
565 cmd->stats.write_hits = le32_to_cpu(disk_super->write_hits);
566 cmd->stats.write_misses = le32_to_cpu(disk_super->write_misses);
568 cmd->changed = false;
572 * The mutator updates the superblock flags.
574 static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
575 flags_mutator mutator)
578 struct cache_disk_superblock *disk_super;
579 struct dm_block *sblock;
581 r = superblock_lock(cmd, &sblock);
585 disk_super = dm_block_data(sblock);
586 update_flags(disk_super, mutator);
587 read_superblock_fields(cmd, disk_super);
588 dm_bm_unlock(sblock);
590 return dm_bm_flush(cmd->bm);
593 static int __begin_transaction(struct dm_cache_metadata *cmd)
596 struct cache_disk_superblock *disk_super;
597 struct dm_block *sblock;
600 * We re-read the superblock every time. Shouldn't need to do this
603 r = superblock_read_lock(cmd, &sblock);
607 disk_super = dm_block_data(sblock);
608 read_superblock_fields(cmd, disk_super);
609 dm_bm_unlock(sblock);
614 static int __commit_transaction(struct dm_cache_metadata *cmd,
615 flags_mutator mutator)
618 struct cache_disk_superblock *disk_super;
619 struct dm_block *sblock;
622 * We need to know if the cache_disk_superblock exceeds a 512-byte sector.
624 BUILD_BUG_ON(sizeof(struct cache_disk_superblock) > 512);
626 r = dm_bitset_flush(&cmd->discard_info, cmd->discard_root,
631 r = dm_tm_pre_commit(cmd->tm);
635 r = __save_sm_root(cmd);
639 r = superblock_lock(cmd, &sblock);
643 disk_super = dm_block_data(sblock);
645 disk_super->flags = cpu_to_le32(cmd->flags);
647 update_flags(disk_super, mutator);
649 disk_super->mapping_root = cpu_to_le64(cmd->root);
650 disk_super->hint_root = cpu_to_le64(cmd->hint_root);
651 disk_super->discard_root = cpu_to_le64(cmd->discard_root);
652 disk_super->discard_block_size = cpu_to_le64(cmd->discard_block_size);
653 disk_super->discard_nr_blocks = cpu_to_le64(from_dblock(cmd->discard_nr_blocks));
654 disk_super->cache_blocks = cpu_to_le32(from_cblock(cmd->cache_blocks));
655 strncpy(disk_super->policy_name, cmd->policy_name, sizeof(disk_super->policy_name));
656 disk_super->policy_version[0] = cpu_to_le32(cmd->policy_version[0]);
657 disk_super->policy_version[1] = cpu_to_le32(cmd->policy_version[1]);
658 disk_super->policy_version[2] = cpu_to_le32(cmd->policy_version[2]);
659 disk_super->policy_hint_size = cpu_to_le32(cmd->policy_hint_size);
661 disk_super->read_hits = cpu_to_le32(cmd->stats.read_hits);
662 disk_super->read_misses = cpu_to_le32(cmd->stats.read_misses);
663 disk_super->write_hits = cpu_to_le32(cmd->stats.write_hits);
664 disk_super->write_misses = cpu_to_le32(cmd->stats.write_misses);
665 __copy_sm_root(cmd, disk_super);
667 return dm_tm_commit(cmd->tm, sblock);
670 /*----------------------------------------------------------------*/
673 * The mappings are held in a dm-array that has 64-bit values stored in
674 * little-endian format. The index is the cblock, the high 48bits of the
675 * value are the oblock and the low 16 bit the flags.
677 #define FLAGS_MASK ((1 << 16) - 1)
679 static __le64 pack_value(dm_oblock_t block, unsigned flags)
681 uint64_t value = from_oblock(block);
683 value = value | (flags & FLAGS_MASK);
684 return cpu_to_le64(value);
687 static void unpack_value(__le64 value_le, dm_oblock_t *block, unsigned *flags)
689 uint64_t value = le64_to_cpu(value_le);
690 uint64_t b = value >> 16;
691 *block = to_oblock(b);
692 *flags = value & FLAGS_MASK;
695 /*----------------------------------------------------------------*/
697 static struct dm_cache_metadata *metadata_open(struct block_device *bdev,
698 sector_t data_block_size,
699 bool may_format_device,
700 size_t policy_hint_size)
703 struct dm_cache_metadata *cmd;
705 cmd = kzalloc(sizeof(*cmd), GFP_KERNEL);
707 DMERR("could not allocate metadata struct");
708 return ERR_PTR(-ENOMEM);
711 atomic_set(&cmd->ref_count, 1);
712 init_rwsem(&cmd->root_lock);
714 cmd->data_block_size = data_block_size;
715 cmd->cache_blocks = 0;
716 cmd->policy_hint_size = policy_hint_size;
718 cmd->fail_io = false;
720 r = __create_persistent_data_objects(cmd, may_format_device);
726 r = __begin_transaction_flags(cmd, clear_clean_shutdown);
728 dm_cache_metadata_close(cmd);
736 * We keep a little list of ref counted metadata objects to prevent two
737 * different target instances creating separate bufio instances. This is
738 * an issue if a table is reloaded before the suspend.
740 static DEFINE_MUTEX(table_lock);
741 static LIST_HEAD(table);
743 static struct dm_cache_metadata *lookup(struct block_device *bdev)
745 struct dm_cache_metadata *cmd;
747 list_for_each_entry(cmd, &table, list)
748 if (cmd->bdev == bdev) {
749 atomic_inc(&cmd->ref_count);
756 static struct dm_cache_metadata *lookup_or_open(struct block_device *bdev,
757 sector_t data_block_size,
758 bool may_format_device,
759 size_t policy_hint_size)
761 struct dm_cache_metadata *cmd, *cmd2;
763 mutex_lock(&table_lock);
765 mutex_unlock(&table_lock);
770 cmd = metadata_open(bdev, data_block_size, may_format_device, policy_hint_size);
772 mutex_lock(&table_lock);
775 mutex_unlock(&table_lock);
776 __destroy_persistent_data_objects(cmd);
780 list_add(&cmd->list, &table);
781 mutex_unlock(&table_lock);
787 static bool same_params(struct dm_cache_metadata *cmd, sector_t data_block_size)
789 if (cmd->data_block_size != data_block_size) {
790 DMERR("data_block_size (%llu) different from that in metadata (%llu)\n",
791 (unsigned long long) data_block_size,
792 (unsigned long long) cmd->data_block_size);
799 struct dm_cache_metadata *dm_cache_metadata_open(struct block_device *bdev,
800 sector_t data_block_size,
801 bool may_format_device,
802 size_t policy_hint_size)
804 struct dm_cache_metadata *cmd = lookup_or_open(bdev, data_block_size,
805 may_format_device, policy_hint_size);
807 if (!IS_ERR(cmd) && !same_params(cmd, data_block_size)) {
808 dm_cache_metadata_close(cmd);
809 return ERR_PTR(-EINVAL);
815 void dm_cache_metadata_close(struct dm_cache_metadata *cmd)
817 if (atomic_dec_and_test(&cmd->ref_count)) {
818 mutex_lock(&table_lock);
819 list_del(&cmd->list);
820 mutex_unlock(&table_lock);
823 __destroy_persistent_data_objects(cmd);
829 * Checks that the given cache block is either unmapped or clean.
831 static int block_unmapped_or_clean(struct dm_cache_metadata *cmd, dm_cblock_t b,
839 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(b), &value);
841 DMERR("block_unmapped_or_clean failed");
845 unpack_value(value, &ob, &flags);
846 *result = !((flags & M_VALID) && (flags & M_DIRTY));
851 static int blocks_are_unmapped_or_clean(struct dm_cache_metadata *cmd,
852 dm_cblock_t begin, dm_cblock_t end,
858 while (begin != end) {
859 r = block_unmapped_or_clean(cmd, begin, result);
864 DMERR("cache block %llu is dirty",
865 (unsigned long long) from_cblock(begin));
869 begin = to_cblock(from_cblock(begin) + 1);
875 static bool cmd_write_lock(struct dm_cache_metadata *cmd)
877 down_write(&cmd->root_lock);
878 if (cmd->fail_io || dm_bm_is_read_only(cmd->bm)) {
879 up_write(&cmd->root_lock);
885 #define WRITE_LOCK(cmd) \
887 if (!cmd_write_lock((cmd))) \
891 #define WRITE_LOCK_VOID(cmd) \
893 if (!cmd_write_lock((cmd))) \
897 #define WRITE_UNLOCK(cmd) \
898 up_write(&(cmd)->root_lock)
900 static bool cmd_read_lock(struct dm_cache_metadata *cmd)
902 down_read(&cmd->root_lock);
904 up_read(&cmd->root_lock);
910 #define READ_LOCK(cmd) \
912 if (!cmd_read_lock((cmd))) \
916 #define READ_LOCK_VOID(cmd) \
918 if (!cmd_read_lock((cmd))) \
922 #define READ_UNLOCK(cmd) \
923 up_read(&(cmd)->root_lock)
925 int dm_cache_resize(struct dm_cache_metadata *cmd, dm_cblock_t new_cache_size)
929 __le64 null_mapping = pack_value(0, 0);
932 __dm_bless_for_disk(&null_mapping);
934 if (from_cblock(new_cache_size) < from_cblock(cmd->cache_blocks)) {
935 r = blocks_are_unmapped_or_clean(cmd, new_cache_size, cmd->cache_blocks, &clean);
937 __dm_unbless_for_disk(&null_mapping);
942 DMERR("unable to shrink cache due to dirty blocks");
944 __dm_unbless_for_disk(&null_mapping);
949 r = dm_array_resize(&cmd->info, cmd->root, from_cblock(cmd->cache_blocks),
950 from_cblock(new_cache_size),
951 &null_mapping, &cmd->root);
953 cmd->cache_blocks = new_cache_size;
962 int dm_cache_discard_bitset_resize(struct dm_cache_metadata *cmd,
963 sector_t discard_block_size,
964 dm_dblock_t new_nr_entries)
969 r = dm_bitset_resize(&cmd->discard_info,
971 from_dblock(cmd->discard_nr_blocks),
972 from_dblock(new_nr_entries),
973 false, &cmd->discard_root);
975 cmd->discard_block_size = discard_block_size;
976 cmd->discard_nr_blocks = new_nr_entries;
985 static int __set_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
987 return dm_bitset_set_bit(&cmd->discard_info, cmd->discard_root,
988 from_dblock(b), &cmd->discard_root);
991 static int __clear_discard(struct dm_cache_metadata *cmd, dm_dblock_t b)
993 return dm_bitset_clear_bit(&cmd->discard_info, cmd->discard_root,
994 from_dblock(b), &cmd->discard_root);
997 static int __is_discarded(struct dm_cache_metadata *cmd, dm_dblock_t b,
1000 return dm_bitset_test_bit(&cmd->discard_info, cmd->discard_root,
1001 from_dblock(b), &cmd->discard_root,
1005 static int __discard(struct dm_cache_metadata *cmd,
1006 dm_dblock_t dblock, bool discard)
1010 r = (discard ? __set_discard : __clear_discard)(cmd, dblock);
1014 cmd->changed = true;
1018 int dm_cache_set_discard(struct dm_cache_metadata *cmd,
1019 dm_dblock_t dblock, bool discard)
1024 r = __discard(cmd, dblock, discard);
1030 static int __load_discards(struct dm_cache_metadata *cmd,
1031 load_discard_fn fn, void *context)
1037 for (b = 0; b < from_dblock(cmd->discard_nr_blocks); b++) {
1038 dm_dblock_t dblock = to_dblock(b);
1040 if (cmd->clean_when_opened) {
1041 r = __is_discarded(cmd, dblock, &discard);
1047 r = fn(context, cmd->discard_block_size, dblock, discard);
1055 int dm_cache_load_discards(struct dm_cache_metadata *cmd,
1056 load_discard_fn fn, void *context)
1061 r = __load_discards(cmd, fn, context);
1067 int dm_cache_size(struct dm_cache_metadata *cmd, dm_cblock_t *result)
1070 *result = cmd->cache_blocks;
1076 static int __remove(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1079 __le64 value = pack_value(0, 0);
1081 __dm_bless_for_disk(&value);
1082 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1083 &value, &cmd->root);
1087 cmd->changed = true;
1091 int dm_cache_remove_mapping(struct dm_cache_metadata *cmd, dm_cblock_t cblock)
1096 r = __remove(cmd, cblock);
1102 static int __insert(struct dm_cache_metadata *cmd,
1103 dm_cblock_t cblock, dm_oblock_t oblock)
1106 __le64 value = pack_value(oblock, M_VALID);
1107 __dm_bless_for_disk(&value);
1109 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1110 &value, &cmd->root);
1114 cmd->changed = true;
1118 int dm_cache_insert_mapping(struct dm_cache_metadata *cmd,
1119 dm_cblock_t cblock, dm_oblock_t oblock)
1124 r = __insert(cmd, cblock, oblock);
1134 struct dm_cache_metadata *cmd;
1135 bool respect_dirty_flags;
1139 static bool policy_unchanged(struct dm_cache_metadata *cmd,
1140 struct dm_cache_policy *policy)
1142 const char *policy_name = dm_cache_policy_get_name(policy);
1143 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1144 size_t policy_hint_size = dm_cache_policy_get_hint_size(policy);
1147 * Ensure policy names match.
1149 if (strncmp(cmd->policy_name, policy_name, sizeof(cmd->policy_name)))
1153 * Ensure policy major versions match.
1155 if (cmd->policy_version[0] != policy_version[0])
1159 * Ensure policy hint sizes match.
1161 if (cmd->policy_hint_size != policy_hint_size)
1167 static bool hints_array_initialized(struct dm_cache_metadata *cmd)
1169 return cmd->hint_root && cmd->policy_hint_size;
1172 static bool hints_array_available(struct dm_cache_metadata *cmd,
1173 struct dm_cache_policy *policy)
1175 return cmd->clean_when_opened && policy_unchanged(cmd, policy) &&
1176 hints_array_initialized(cmd);
1179 static int __load_mapping(void *context, uint64_t cblock, void *leaf)
1184 __le32 hint_value = 0;
1187 struct thunk *thunk = context;
1188 struct dm_cache_metadata *cmd = thunk->cmd;
1190 memcpy(&value, leaf, sizeof(value));
1191 unpack_value(value, &oblock, &flags);
1193 if (flags & M_VALID) {
1194 if (thunk->hints_valid) {
1195 r = dm_array_get_value(&cmd->hint_info, cmd->hint_root,
1196 cblock, &hint_value);
1197 if (r && r != -ENODATA)
1201 dirty = thunk->respect_dirty_flags ? (flags & M_DIRTY) : true;
1202 r = thunk->fn(thunk->context, oblock, to_cblock(cblock),
1203 dirty, le32_to_cpu(hint_value), thunk->hints_valid);
1209 static int __load_mappings(struct dm_cache_metadata *cmd,
1210 struct dm_cache_policy *policy,
1211 load_mapping_fn fn, void *context)
1216 thunk.context = context;
1219 thunk.respect_dirty_flags = cmd->clean_when_opened;
1220 thunk.hints_valid = hints_array_available(cmd, policy);
1222 return dm_array_walk(&cmd->info, cmd->root, __load_mapping, &thunk);
1225 int dm_cache_load_mappings(struct dm_cache_metadata *cmd,
1226 struct dm_cache_policy *policy,
1227 load_mapping_fn fn, void *context)
1232 r = __load_mappings(cmd, policy, fn, context);
1238 static int __dump_mapping(void *context, uint64_t cblock, void *leaf)
1245 memcpy(&value, leaf, sizeof(value));
1246 unpack_value(value, &oblock, &flags);
1251 static int __dump_mappings(struct dm_cache_metadata *cmd)
1253 return dm_array_walk(&cmd->info, cmd->root, __dump_mapping, NULL);
1256 void dm_cache_dump(struct dm_cache_metadata *cmd)
1258 READ_LOCK_VOID(cmd);
1259 __dump_mappings(cmd);
1263 int dm_cache_changed_this_transaction(struct dm_cache_metadata *cmd)
1274 static int __dirty(struct dm_cache_metadata *cmd, dm_cblock_t cblock, bool dirty)
1281 r = dm_array_get_value(&cmd->info, cmd->root, from_cblock(cblock), &value);
1285 unpack_value(value, &oblock, &flags);
1287 if (((flags & M_DIRTY) && dirty) || (!(flags & M_DIRTY) && !dirty))
1288 /* nothing to be done */
1291 value = pack_value(oblock, (flags & ~M_DIRTY) | (dirty ? M_DIRTY : 0));
1292 __dm_bless_for_disk(&value);
1294 r = dm_array_set_value(&cmd->info, cmd->root, from_cblock(cblock),
1295 &value, &cmd->root);
1299 cmd->changed = true;
1304 int dm_cache_set_dirty(struct dm_cache_metadata *cmd,
1305 dm_cblock_t cblock, bool dirty)
1310 r = __dirty(cmd, cblock, dirty);
1316 void dm_cache_metadata_get_stats(struct dm_cache_metadata *cmd,
1317 struct dm_cache_statistics *stats)
1319 READ_LOCK_VOID(cmd);
1320 *stats = cmd->stats;
1324 void dm_cache_metadata_set_stats(struct dm_cache_metadata *cmd,
1325 struct dm_cache_statistics *stats)
1327 WRITE_LOCK_VOID(cmd);
1328 cmd->stats = *stats;
1332 int dm_cache_commit(struct dm_cache_metadata *cmd, bool clean_shutdown)
1335 flags_mutator mutator = (clean_shutdown ? set_clean_shutdown :
1336 clear_clean_shutdown);
1342 r = __commit_transaction(cmd, mutator);
1346 r = __begin_transaction(cmd);
1352 int dm_cache_get_free_metadata_block_count(struct dm_cache_metadata *cmd,
1359 r = dm_sm_get_nr_free(cmd->metadata_sm, result);
1365 int dm_cache_get_metadata_dev_size(struct dm_cache_metadata *cmd,
1372 r = dm_sm_get_nr_blocks(cmd->metadata_sm, result);
1378 /*----------------------------------------------------------------*/
1380 static int begin_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1385 const char *policy_name = dm_cache_policy_get_name(policy);
1386 const unsigned *policy_version = dm_cache_policy_get_version(policy);
1388 if (!policy_name[0] ||
1389 (strlen(policy_name) > sizeof(cmd->policy_name) - 1))
1392 if (!policy_unchanged(cmd, policy)) {
1393 strncpy(cmd->policy_name, policy_name, sizeof(cmd->policy_name));
1394 memcpy(cmd->policy_version, policy_version, sizeof(cmd->policy_version));
1396 hint_size = dm_cache_policy_get_hint_size(policy);
1398 return 0; /* short-circuit hints initialization */
1399 cmd->policy_hint_size = hint_size;
1401 if (cmd->hint_root) {
1402 r = dm_array_del(&cmd->hint_info, cmd->hint_root);
1407 r = dm_array_empty(&cmd->hint_info, &cmd->hint_root);
1411 value = cpu_to_le32(0);
1412 __dm_bless_for_disk(&value);
1413 r = dm_array_resize(&cmd->hint_info, cmd->hint_root, 0,
1414 from_cblock(cmd->cache_blocks),
1415 &value, &cmd->hint_root);
1423 static int save_hint(void *context, dm_cblock_t cblock, dm_oblock_t oblock, uint32_t hint)
1425 struct dm_cache_metadata *cmd = context;
1426 __le32 value = cpu_to_le32(hint);
1429 __dm_bless_for_disk(&value);
1431 r = dm_array_set_value(&cmd->hint_info, cmd->hint_root,
1432 from_cblock(cblock), &value, &cmd->hint_root);
1433 cmd->changed = true;
1438 static int write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1442 r = begin_hints(cmd, policy);
1444 DMERR("begin_hints failed");
1448 return policy_walk_mappings(policy, save_hint, cmd);
1451 int dm_cache_write_hints(struct dm_cache_metadata *cmd, struct dm_cache_policy *policy)
1456 r = write_hints(cmd, policy);
1462 int dm_cache_metadata_all_clean(struct dm_cache_metadata *cmd, bool *result)
1467 r = blocks_are_unmapped_or_clean(cmd, 0, cmd->cache_blocks, result);
1473 void dm_cache_metadata_set_read_only(struct dm_cache_metadata *cmd)
1475 WRITE_LOCK_VOID(cmd);
1476 dm_bm_set_read_only(cmd->bm);
1480 void dm_cache_metadata_set_read_write(struct dm_cache_metadata *cmd)
1482 WRITE_LOCK_VOID(cmd);
1483 dm_bm_set_read_write(cmd->bm);
1487 int dm_cache_metadata_set_needs_check(struct dm_cache_metadata *cmd)
1490 struct dm_block *sblock;
1491 struct cache_disk_superblock *disk_super;
1494 set_bit(NEEDS_CHECK, &cmd->flags);
1496 r = superblock_lock(cmd, &sblock);
1498 DMERR("couldn't read superblock");
1502 disk_super = dm_block_data(sblock);
1503 disk_super->flags = cpu_to_le32(cmd->flags);
1505 dm_bm_unlock(sblock);
1512 int dm_cache_metadata_needs_check(struct dm_cache_metadata *cmd, bool *result)
1515 *result = !!test_bit(NEEDS_CHECK, &cmd->flags);
1521 int dm_cache_metadata_abort(struct dm_cache_metadata *cmd)
1526 __destroy_persistent_data_objects(cmd);
1527 r = __create_persistent_data_objects(cmd, false);
1529 cmd->fail_io = true;