1 // SPDX-License-Identifier: GPL-2.0-only
3 #include "persistent-data/dm-transaction-manager.h"
4 #include "persistent-data/dm-bitset.h"
5 #include "persistent-data/dm-space-map.h"
7 #include <linux/dm-io.h>
8 #include <linux/dm-kcopyd.h>
9 #include <linux/init.h>
10 #include <linux/mempool.h>
11 #include <linux/module.h>
12 #include <linux/slab.h>
13 #include <linux/vmalloc.h>
15 #define DM_MSG_PREFIX "era"
17 #define SUPERBLOCK_LOCATION 0
18 #define SUPERBLOCK_MAGIC 2126579579
19 #define SUPERBLOCK_CSUM_XOR 146538381
20 #define MIN_ERA_VERSION 1
21 #define MAX_ERA_VERSION 1
22 #define INVALID_WRITESET_ROOT SUPERBLOCK_LOCATION
23 #define MIN_BLOCK_SIZE 8
25 /*----------------------------------------------------------------
27 *--------------------------------------------------------------*/
28 struct writeset_metadata {
34 struct writeset_metadata md;
37 * An in core copy of the bits to save constantly doing look ups on
44 * This does not free off the on disk bitset as this will normally be done
45 * after digesting into the era array.
47 static void writeset_free(struct writeset *ws)
53 static int setup_on_disk_bitset(struct dm_disk_bitset *info,
54 unsigned nr_bits, dm_block_t *root)
58 r = dm_bitset_empty(info, root);
62 return dm_bitset_resize(info, *root, 0, nr_bits, false, root);
65 static size_t bitset_size(unsigned nr_bits)
67 return sizeof(unsigned long) * dm_div_up(nr_bits, BITS_PER_LONG);
71 * Allocates memory for the in core bitset.
73 static int writeset_alloc(struct writeset *ws, dm_block_t nr_blocks)
75 ws->bits = vzalloc(bitset_size(nr_blocks));
77 DMERR("%s: couldn't allocate in memory bitset", __func__);
85 * Wipes the in-core bitset, and creates a new on disk bitset.
87 static int writeset_init(struct dm_disk_bitset *info, struct writeset *ws,
92 memset(ws->bits, 0, bitset_size(nr_blocks));
94 ws->md.nr_bits = nr_blocks;
95 r = setup_on_disk_bitset(info, ws->md.nr_bits, &ws->md.root);
97 DMERR("%s: setup_on_disk_bitset failed", __func__);
104 static bool writeset_marked(struct writeset *ws, dm_block_t block)
106 return test_bit(block, ws->bits);
109 static int writeset_marked_on_disk(struct dm_disk_bitset *info,
110 struct writeset_metadata *m, dm_block_t block,
113 dm_block_t old = m->root;
116 * The bitset was flushed when it was archived, so we know there'll
117 * be no change to the root.
119 int r = dm_bitset_test_bit(info, m->root, block, &m->root, result);
121 DMERR("%s: dm_bitset_test_bit failed", __func__);
125 BUG_ON(m->root != old);
131 * Returns < 0 on error, 0 if the bit wasn't previously set, 1 if it was.
133 static int writeset_test_and_set(struct dm_disk_bitset *info,
134 struct writeset *ws, uint32_t block)
138 if (!test_bit(block, ws->bits)) {
139 r = dm_bitset_set_bit(info, ws->md.root, block, &ws->md.root);
141 /* FIXME: fail mode */
151 /*----------------------------------------------------------------
152 * On disk metadata layout
153 *--------------------------------------------------------------*/
154 #define SPACE_MAP_ROOT_SIZE 128
157 struct writeset_disk {
162 struct superblock_disk {
171 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
173 __le32 data_block_size;
174 __le32 metadata_block_size;
178 struct writeset_disk current_writeset;
181 * Only these two fields are valid within the metadata snapshot.
183 __le64 writeset_tree_root;
184 __le64 era_array_root;
186 __le64 metadata_snap;
189 /*----------------------------------------------------------------
190 * Superblock validation
191 *--------------------------------------------------------------*/
192 static void sb_prepare_for_write(struct dm_block_validator *v,
194 size_t sb_block_size)
196 struct superblock_disk *disk = dm_block_data(b);
198 disk->blocknr = cpu_to_le64(dm_block_location(b));
199 disk->csum = cpu_to_le32(dm_bm_checksum(&disk->flags,
200 sb_block_size - sizeof(__le32),
201 SUPERBLOCK_CSUM_XOR));
204 static int check_metadata_version(struct superblock_disk *disk)
206 uint32_t metadata_version = le32_to_cpu(disk->version);
207 if (metadata_version < MIN_ERA_VERSION || metadata_version > MAX_ERA_VERSION) {
208 DMERR("Era metadata version %u found, but only versions between %u and %u supported.",
209 metadata_version, MIN_ERA_VERSION, MAX_ERA_VERSION);
216 static int sb_check(struct dm_block_validator *v,
218 size_t sb_block_size)
220 struct superblock_disk *disk = dm_block_data(b);
223 if (dm_block_location(b) != le64_to_cpu(disk->blocknr)) {
224 DMERR("sb_check failed: blocknr %llu: wanted %llu",
225 le64_to_cpu(disk->blocknr),
226 (unsigned long long)dm_block_location(b));
230 if (le64_to_cpu(disk->magic) != SUPERBLOCK_MAGIC) {
231 DMERR("sb_check failed: magic %llu: wanted %llu",
232 le64_to_cpu(disk->magic),
233 (unsigned long long) SUPERBLOCK_MAGIC);
237 csum_le = cpu_to_le32(dm_bm_checksum(&disk->flags,
238 sb_block_size - sizeof(__le32),
239 SUPERBLOCK_CSUM_XOR));
240 if (csum_le != disk->csum) {
241 DMERR("sb_check failed: csum %u: wanted %u",
242 le32_to_cpu(csum_le), le32_to_cpu(disk->csum));
246 return check_metadata_version(disk);
249 static struct dm_block_validator sb_validator = {
250 .name = "superblock",
251 .prepare_for_write = sb_prepare_for_write,
255 /*----------------------------------------------------------------
256 * Low level metadata handling
257 *--------------------------------------------------------------*/
258 #define DM_ERA_METADATA_BLOCK_SIZE 4096
259 #define ERA_MAX_CONCURRENT_LOCKS 5
261 struct era_metadata {
262 struct block_device *bdev;
263 struct dm_block_manager *bm;
264 struct dm_space_map *sm;
265 struct dm_transaction_manager *tm;
267 dm_block_t block_size;
270 uint32_t current_era;
273 * We preallocate 2 writesets. When an era rolls over we
274 * switch between them. This means the allocation is done at
275 * preresume time, rather than on the io path.
277 struct writeset writesets[2];
278 struct writeset *current_writeset;
280 dm_block_t writeset_tree_root;
281 dm_block_t era_array_root;
283 struct dm_disk_bitset bitset_info;
284 struct dm_btree_info writeset_tree_info;
285 struct dm_array_info era_array_info;
287 dm_block_t metadata_snap;
290 * A flag that is set whenever a writeset has been archived.
292 bool archived_writesets;
295 * Reading the space map root can fail, so we read it into this
296 * buffer before the superblock is locked and updated.
298 __u8 metadata_space_map_root[SPACE_MAP_ROOT_SIZE];
301 static int superblock_read_lock(struct era_metadata *md,
302 struct dm_block **sblock)
304 return dm_bm_read_lock(md->bm, SUPERBLOCK_LOCATION,
305 &sb_validator, sblock);
308 static int superblock_lock_zero(struct era_metadata *md,
309 struct dm_block **sblock)
311 return dm_bm_write_lock_zero(md->bm, SUPERBLOCK_LOCATION,
312 &sb_validator, sblock);
315 static int superblock_lock(struct era_metadata *md,
316 struct dm_block **sblock)
318 return dm_bm_write_lock(md->bm, SUPERBLOCK_LOCATION,
319 &sb_validator, sblock);
322 /* FIXME: duplication with cache and thin */
323 static int superblock_all_zeroes(struct dm_block_manager *bm, bool *result)
328 __le64 *data_le, zero = cpu_to_le64(0);
329 unsigned sb_block_size = dm_bm_block_size(bm) / sizeof(__le64);
332 * We can't use a validator here - it may be all zeroes.
334 r = dm_bm_read_lock(bm, SUPERBLOCK_LOCATION, NULL, &b);
338 data_le = dm_block_data(b);
340 for (i = 0; i < sb_block_size; i++) {
341 if (data_le[i] != zero) {
352 /*----------------------------------------------------------------*/
354 static void ws_pack(const struct writeset_metadata *core, struct writeset_disk *disk)
356 disk->nr_bits = cpu_to_le32(core->nr_bits);
357 disk->root = cpu_to_le64(core->root);
360 static void ws_unpack(const struct writeset_disk *disk, struct writeset_metadata *core)
362 core->nr_bits = le32_to_cpu(disk->nr_bits);
363 core->root = le64_to_cpu(disk->root);
366 static void ws_inc(void *context, const void *value)
368 struct era_metadata *md = context;
369 struct writeset_disk ws_d;
372 memcpy(&ws_d, value, sizeof(ws_d));
373 b = le64_to_cpu(ws_d.root);
375 dm_tm_inc(md->tm, b);
378 static void ws_dec(void *context, const void *value)
380 struct era_metadata *md = context;
381 struct writeset_disk ws_d;
384 memcpy(&ws_d, value, sizeof(ws_d));
385 b = le64_to_cpu(ws_d.root);
387 dm_bitset_del(&md->bitset_info, b);
390 static int ws_eq(void *context, const void *value1, const void *value2)
392 return !memcmp(value1, value2, sizeof(struct writeset_disk));
395 /*----------------------------------------------------------------*/
397 static void setup_writeset_tree_info(struct era_metadata *md)
399 struct dm_btree_value_type *vt = &md->writeset_tree_info.value_type;
400 md->writeset_tree_info.tm = md->tm;
401 md->writeset_tree_info.levels = 1;
403 vt->size = sizeof(struct writeset_disk);
409 static void setup_era_array_info(struct era_metadata *md)
412 struct dm_btree_value_type vt;
414 vt.size = sizeof(__le32);
419 dm_array_info_init(&md->era_array_info, md->tm, &vt);
422 static void setup_infos(struct era_metadata *md)
424 dm_disk_bitset_init(md->tm, &md->bitset_info);
425 setup_writeset_tree_info(md);
426 setup_era_array_info(md);
429 /*----------------------------------------------------------------*/
431 static int create_fresh_metadata(struct era_metadata *md)
435 r = dm_tm_create_with_sm(md->bm, SUPERBLOCK_LOCATION,
438 DMERR("dm_tm_create_with_sm failed");
444 r = dm_btree_empty(&md->writeset_tree_info, &md->writeset_tree_root);
446 DMERR("couldn't create new writeset tree");
450 r = dm_array_empty(&md->era_array_info, &md->era_array_root);
452 DMERR("couldn't create era array");
459 dm_sm_destroy(md->sm);
460 dm_tm_destroy(md->tm);
465 static int save_sm_root(struct era_metadata *md)
470 r = dm_sm_root_size(md->sm, &metadata_len);
474 return dm_sm_copy_root(md->sm, &md->metadata_space_map_root,
478 static void copy_sm_root(struct era_metadata *md, struct superblock_disk *disk)
480 memcpy(&disk->metadata_space_map_root,
481 &md->metadata_space_map_root,
482 sizeof(md->metadata_space_map_root));
486 * Writes a superblock, including the static fields that don't get updated
487 * with every commit (possible optimisation here). 'md' should be fully
488 * constructed when this is called.
490 static void prepare_superblock(struct era_metadata *md, struct superblock_disk *disk)
492 disk->magic = cpu_to_le64(SUPERBLOCK_MAGIC);
493 disk->flags = cpu_to_le32(0ul);
495 /* FIXME: can't keep blanking the uuid (uuid is currently unused though) */
496 memset(disk->uuid, 0, sizeof(disk->uuid));
497 disk->version = cpu_to_le32(MAX_ERA_VERSION);
499 copy_sm_root(md, disk);
501 disk->data_block_size = cpu_to_le32(md->block_size);
502 disk->metadata_block_size = cpu_to_le32(DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT);
503 disk->nr_blocks = cpu_to_le32(md->nr_blocks);
504 disk->current_era = cpu_to_le32(md->current_era);
506 ws_pack(&md->current_writeset->md, &disk->current_writeset);
507 disk->writeset_tree_root = cpu_to_le64(md->writeset_tree_root);
508 disk->era_array_root = cpu_to_le64(md->era_array_root);
509 disk->metadata_snap = cpu_to_le64(md->metadata_snap);
512 static int write_superblock(struct era_metadata *md)
515 struct dm_block *sblock;
516 struct superblock_disk *disk;
518 r = save_sm_root(md);
520 DMERR("%s: save_sm_root failed", __func__);
524 r = superblock_lock_zero(md, &sblock);
528 disk = dm_block_data(sblock);
529 prepare_superblock(md, disk);
531 return dm_tm_commit(md->tm, sblock);
535 * Assumes block_size and the infos are set.
537 static int format_metadata(struct era_metadata *md)
541 r = create_fresh_metadata(md);
545 r = write_superblock(md);
547 dm_sm_destroy(md->sm);
548 dm_tm_destroy(md->tm);
555 static int open_metadata(struct era_metadata *md)
558 struct dm_block *sblock;
559 struct superblock_disk *disk;
561 r = superblock_read_lock(md, &sblock);
563 DMERR("couldn't read_lock superblock");
567 disk = dm_block_data(sblock);
569 /* Verify the data block size hasn't changed */
570 if (le32_to_cpu(disk->data_block_size) != md->block_size) {
571 DMERR("changing the data block size (from %u to %llu) is not supported",
572 le32_to_cpu(disk->data_block_size), md->block_size);
577 r = dm_tm_open_with_sm(md->bm, SUPERBLOCK_LOCATION,
578 disk->metadata_space_map_root,
579 sizeof(disk->metadata_space_map_root),
582 DMERR("dm_tm_open_with_sm failed");
588 md->nr_blocks = le32_to_cpu(disk->nr_blocks);
589 md->current_era = le32_to_cpu(disk->current_era);
591 ws_unpack(&disk->current_writeset, &md->current_writeset->md);
592 md->writeset_tree_root = le64_to_cpu(disk->writeset_tree_root);
593 md->era_array_root = le64_to_cpu(disk->era_array_root);
594 md->metadata_snap = le64_to_cpu(disk->metadata_snap);
595 md->archived_writesets = true;
597 dm_bm_unlock(sblock);
602 dm_bm_unlock(sblock);
606 static int open_or_format_metadata(struct era_metadata *md,
610 bool unformatted = false;
612 r = superblock_all_zeroes(md->bm, &unformatted);
617 return may_format ? format_metadata(md) : -EPERM;
619 return open_metadata(md);
622 static int create_persistent_data_objects(struct era_metadata *md,
627 md->bm = dm_block_manager_create(md->bdev, DM_ERA_METADATA_BLOCK_SIZE,
628 ERA_MAX_CONCURRENT_LOCKS);
629 if (IS_ERR(md->bm)) {
630 DMERR("could not create block manager");
631 return PTR_ERR(md->bm);
634 r = open_or_format_metadata(md, may_format);
636 dm_block_manager_destroy(md->bm);
641 static void destroy_persistent_data_objects(struct era_metadata *md)
643 dm_sm_destroy(md->sm);
644 dm_tm_destroy(md->tm);
645 dm_block_manager_destroy(md->bm);
649 * This waits until all era_map threads have picked up the new filter.
651 static void swap_writeset(struct era_metadata *md, struct writeset *new_writeset)
653 rcu_assign_pointer(md->current_writeset, new_writeset);
657 /*----------------------------------------------------------------
658 * Writesets get 'digested' into the main era array.
660 * We're using a coroutine here so the worker thread can do the digestion,
661 * thus avoiding synchronisation of the metadata. Digesting a whole
662 * writeset in one go would cause too much latency.
663 *--------------------------------------------------------------*/
666 unsigned nr_bits, current_bit;
667 struct writeset_metadata writeset;
669 struct dm_disk_bitset info;
671 int (*step)(struct era_metadata *, struct digest *);
674 static int metadata_digest_lookup_writeset(struct era_metadata *md,
677 static int metadata_digest_remove_writeset(struct era_metadata *md,
681 uint64_t key = d->era;
683 r = dm_btree_remove(&md->writeset_tree_info, md->writeset_tree_root,
684 &key, &md->writeset_tree_root);
686 DMERR("%s: dm_btree_remove failed", __func__);
690 d->step = metadata_digest_lookup_writeset;
694 #define INSERTS_PER_STEP 100
696 static int metadata_digest_transcribe_writeset(struct era_metadata *md,
701 unsigned b, e = min(d->current_bit + INSERTS_PER_STEP, d->nr_bits);
703 for (b = d->current_bit; b < e; b++) {
704 r = writeset_marked_on_disk(&d->info, &d->writeset, b, &marked);
706 DMERR("%s: writeset_marked_on_disk failed", __func__);
713 __dm_bless_for_disk(&d->value);
714 r = dm_array_set_value(&md->era_array_info, md->era_array_root,
715 b, &d->value, &md->era_array_root);
717 DMERR("%s: dm_array_set_value failed", __func__);
723 d->step = metadata_digest_remove_writeset;
730 static int metadata_digest_lookup_writeset(struct era_metadata *md,
735 struct writeset_disk disk;
737 r = dm_btree_find_lowest_key(&md->writeset_tree_info,
738 md->writeset_tree_root, &key);
744 r = dm_btree_lookup(&md->writeset_tree_info,
745 md->writeset_tree_root, &key, &disk);
752 DMERR("%s: dm_btree_lookup failed", __func__);
756 ws_unpack(&disk, &d->writeset);
757 d->value = cpu_to_le32(key);
760 * We initialise another bitset info to avoid any caching side effects
761 * with the previous one.
763 dm_disk_bitset_init(md->tm, &d->info);
765 d->nr_bits = min(d->writeset.nr_bits, md->nr_blocks);
767 d->step = metadata_digest_transcribe_writeset;
772 static int metadata_digest_start(struct era_metadata *md, struct digest *d)
777 memset(d, 0, sizeof(*d));
778 d->step = metadata_digest_lookup_writeset;
783 /*----------------------------------------------------------------
784 * High level metadata interface. Target methods should use these, and not
785 * the lower level ones.
786 *--------------------------------------------------------------*/
787 static struct era_metadata *metadata_open(struct block_device *bdev,
792 struct era_metadata *md = kzalloc(sizeof(*md), GFP_KERNEL);
798 md->block_size = block_size;
800 md->writesets[0].md.root = INVALID_WRITESET_ROOT;
801 md->writesets[1].md.root = INVALID_WRITESET_ROOT;
802 md->current_writeset = &md->writesets[0];
804 r = create_persistent_data_objects(md, may_format);
813 static void metadata_close(struct era_metadata *md)
815 writeset_free(&md->writesets[0]);
816 writeset_free(&md->writesets[1]);
817 destroy_persistent_data_objects(md);
821 static bool valid_nr_blocks(dm_block_t n)
824 * dm_bitset restricts us to 2^32. test_bit & co. restrict us
825 * further to 2^31 - 1
827 return n < (1ull << 31);
830 static int metadata_resize(struct era_metadata *md, void *arg)
833 dm_block_t *new_size = arg;
836 if (!valid_nr_blocks(*new_size)) {
837 DMERR("Invalid number of origin blocks %llu",
838 (unsigned long long) *new_size);
842 writeset_free(&md->writesets[0]);
843 writeset_free(&md->writesets[1]);
845 r = writeset_alloc(&md->writesets[0], *new_size);
847 DMERR("%s: writeset_alloc failed for writeset 0", __func__);
851 r = writeset_alloc(&md->writesets[1], *new_size);
853 DMERR("%s: writeset_alloc failed for writeset 1", __func__);
854 writeset_free(&md->writesets[0]);
858 value = cpu_to_le32(0u);
859 __dm_bless_for_disk(&value);
860 r = dm_array_resize(&md->era_array_info, md->era_array_root,
861 md->nr_blocks, *new_size,
862 &value, &md->era_array_root);
864 DMERR("%s: dm_array_resize failed", __func__);
865 writeset_free(&md->writesets[0]);
866 writeset_free(&md->writesets[1]);
870 md->nr_blocks = *new_size;
874 static int metadata_era_archive(struct era_metadata *md)
878 struct writeset_disk value;
880 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
881 &md->current_writeset->md.root);
883 DMERR("%s: dm_bitset_flush failed", __func__);
887 ws_pack(&md->current_writeset->md, &value);
889 keys[0] = md->current_era;
890 __dm_bless_for_disk(&value);
891 r = dm_btree_insert(&md->writeset_tree_info, md->writeset_tree_root,
892 keys, &value, &md->writeset_tree_root);
894 DMERR("%s: couldn't insert writeset into btree", __func__);
895 /* FIXME: fail mode */
899 md->current_writeset->md.root = INVALID_WRITESET_ROOT;
900 md->archived_writesets = true;
905 static struct writeset *next_writeset(struct era_metadata *md)
907 return (md->current_writeset == &md->writesets[0]) ?
908 &md->writesets[1] : &md->writesets[0];
911 static int metadata_new_era(struct era_metadata *md)
914 struct writeset *new_writeset = next_writeset(md);
916 r = writeset_init(&md->bitset_info, new_writeset, md->nr_blocks);
918 DMERR("%s: writeset_init failed", __func__);
922 swap_writeset(md, new_writeset);
928 static int metadata_era_rollover(struct era_metadata *md)
932 if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
933 r = metadata_era_archive(md);
935 DMERR("%s: metadata_archive_era failed", __func__);
936 /* FIXME: fail mode? */
941 r = metadata_new_era(md);
943 DMERR("%s: new era failed", __func__);
944 /* FIXME: fail mode */
951 static bool metadata_current_marked(struct era_metadata *md, dm_block_t block)
957 ws = rcu_dereference(md->current_writeset);
958 r = writeset_marked(ws, block);
964 static int metadata_commit(struct era_metadata *md)
967 struct dm_block *sblock;
969 if (md->current_writeset->md.root != INVALID_WRITESET_ROOT) {
970 r = dm_bitset_flush(&md->bitset_info, md->current_writeset->md.root,
971 &md->current_writeset->md.root);
973 DMERR("%s: bitset flush failed", __func__);
978 r = dm_tm_pre_commit(md->tm);
980 DMERR("%s: pre commit failed", __func__);
984 r = save_sm_root(md);
986 DMERR("%s: save_sm_root failed", __func__);
990 r = superblock_lock(md, &sblock);
992 DMERR("%s: superblock lock failed", __func__);
996 prepare_superblock(md, dm_block_data(sblock));
998 return dm_tm_commit(md->tm, sblock);
1001 static int metadata_checkpoint(struct era_metadata *md)
1004 * For now we just rollover, but later I want to put a check in to
1005 * avoid this if the filter is still pretty fresh.
1007 return metadata_era_rollover(md);
1011 * Metadata snapshots allow userland to access era data.
1013 static int metadata_take_snap(struct era_metadata *md)
1016 struct dm_block *clone;
1018 if (md->metadata_snap != SUPERBLOCK_LOCATION) {
1019 DMERR("%s: metadata snapshot already exists", __func__);
1023 r = metadata_era_rollover(md);
1025 DMERR("%s: era rollover failed", __func__);
1029 r = metadata_commit(md);
1031 DMERR("%s: pre commit failed", __func__);
1035 r = dm_sm_inc_block(md->sm, SUPERBLOCK_LOCATION);
1037 DMERR("%s: couldn't increment superblock", __func__);
1041 r = dm_tm_shadow_block(md->tm, SUPERBLOCK_LOCATION,
1042 &sb_validator, &clone, &inc);
1044 DMERR("%s: couldn't shadow superblock", __func__);
1045 dm_sm_dec_block(md->sm, SUPERBLOCK_LOCATION);
1050 r = dm_sm_inc_block(md->sm, md->writeset_tree_root);
1052 DMERR("%s: couldn't inc writeset tree root", __func__);
1053 dm_tm_unlock(md->tm, clone);
1057 r = dm_sm_inc_block(md->sm, md->era_array_root);
1059 DMERR("%s: couldn't inc era tree root", __func__);
1060 dm_sm_dec_block(md->sm, md->writeset_tree_root);
1061 dm_tm_unlock(md->tm, clone);
1065 md->metadata_snap = dm_block_location(clone);
1067 dm_tm_unlock(md->tm, clone);
1072 static int metadata_drop_snap(struct era_metadata *md)
1075 dm_block_t location;
1076 struct dm_block *clone;
1077 struct superblock_disk *disk;
1079 if (md->metadata_snap == SUPERBLOCK_LOCATION) {
1080 DMERR("%s: no snap to drop", __func__);
1084 r = dm_tm_read_lock(md->tm, md->metadata_snap, &sb_validator, &clone);
1086 DMERR("%s: couldn't read lock superblock clone", __func__);
1091 * Whatever happens now we'll commit with no record of the metadata
1094 md->metadata_snap = SUPERBLOCK_LOCATION;
1096 disk = dm_block_data(clone);
1097 r = dm_btree_del(&md->writeset_tree_info,
1098 le64_to_cpu(disk->writeset_tree_root));
1100 DMERR("%s: error deleting writeset tree clone", __func__);
1101 dm_tm_unlock(md->tm, clone);
1105 r = dm_array_del(&md->era_array_info, le64_to_cpu(disk->era_array_root));
1107 DMERR("%s: error deleting era array clone", __func__);
1108 dm_tm_unlock(md->tm, clone);
1112 location = dm_block_location(clone);
1113 dm_tm_unlock(md->tm, clone);
1115 return dm_sm_dec_block(md->sm, location);
1118 struct metadata_stats {
1125 static int metadata_get_stats(struct era_metadata *md, void *ptr)
1128 struct metadata_stats *s = ptr;
1129 dm_block_t nr_free, nr_total;
1131 r = dm_sm_get_nr_free(md->sm, &nr_free);
1133 DMERR("dm_sm_get_nr_free returned %d", r);
1137 r = dm_sm_get_nr_blocks(md->sm, &nr_total);
1139 DMERR("dm_pool_get_metadata_dev_size returned %d", r);
1143 s->used = nr_total - nr_free;
1144 s->total = nr_total;
1145 s->snap = md->metadata_snap;
1146 s->era = md->current_era;
1151 /*----------------------------------------------------------------*/
1154 struct dm_target *ti;
1156 struct dm_dev *metadata_dev;
1157 struct dm_dev *origin_dev;
1159 dm_block_t nr_blocks;
1160 uint32_t sectors_per_block;
1161 int sectors_per_block_shift;
1162 struct era_metadata *md;
1164 struct workqueue_struct *wq;
1165 struct work_struct worker;
1167 spinlock_t deferred_lock;
1168 struct bio_list deferred_bios;
1170 spinlock_t rpc_lock;
1171 struct list_head rpc_calls;
1173 struct digest digest;
1178 struct list_head list;
1180 int (*fn0)(struct era_metadata *);
1181 int (*fn1)(struct era_metadata *, void *);
1185 struct completion complete;
1188 /*----------------------------------------------------------------
1190 *---------------------------------------------------------------*/
1191 static bool block_size_is_power_of_two(struct era *era)
1193 return era->sectors_per_block_shift >= 0;
1196 static dm_block_t get_block(struct era *era, struct bio *bio)
1198 sector_t block_nr = bio->bi_iter.bi_sector;
1200 if (!block_size_is_power_of_two(era))
1201 (void) sector_div(block_nr, era->sectors_per_block);
1203 block_nr >>= era->sectors_per_block_shift;
1208 static void remap_to_origin(struct era *era, struct bio *bio)
1210 bio_set_dev(bio, era->origin_dev->bdev);
1213 /*----------------------------------------------------------------
1215 *--------------------------------------------------------------*/
1216 static void wake_worker(struct era *era)
1218 if (!atomic_read(&era->suspended))
1219 queue_work(era->wq, &era->worker);
1222 static void process_old_eras(struct era *era)
1226 if (!era->digest.step)
1229 r = era->digest.step(era->md, &era->digest);
1231 DMERR("%s: digest step failed, stopping digestion", __func__);
1232 era->digest.step = NULL;
1234 } else if (era->digest.step)
1238 static void process_deferred_bios(struct era *era)
1241 struct bio_list deferred_bios, marked_bios;
1243 struct blk_plug plug;
1244 bool commit_needed = false;
1245 bool failed = false;
1246 struct writeset *ws = era->md->current_writeset;
1248 bio_list_init(&deferred_bios);
1249 bio_list_init(&marked_bios);
1251 spin_lock(&era->deferred_lock);
1252 bio_list_merge(&deferred_bios, &era->deferred_bios);
1253 bio_list_init(&era->deferred_bios);
1254 spin_unlock(&era->deferred_lock);
1256 if (bio_list_empty(&deferred_bios))
1259 while ((bio = bio_list_pop(&deferred_bios))) {
1260 r = writeset_test_and_set(&era->md->bitset_info, ws,
1261 get_block(era, bio));
1264 * This is bad news, we need to rollback.
1269 commit_needed = true;
1271 bio_list_add(&marked_bios, bio);
1274 if (commit_needed) {
1275 r = metadata_commit(era->md);
1281 while ((bio = bio_list_pop(&marked_bios)))
1284 blk_start_plug(&plug);
1285 while ((bio = bio_list_pop(&marked_bios))) {
1287 * Only update the in-core writeset if the on-disk one
1291 set_bit(get_block(era, bio), ws->bits);
1292 submit_bio_noacct(bio);
1294 blk_finish_plug(&plug);
1298 static void process_rpc_calls(struct era *era)
1301 bool need_commit = false;
1302 struct list_head calls;
1303 struct rpc *rpc, *tmp;
1305 INIT_LIST_HEAD(&calls);
1306 spin_lock(&era->rpc_lock);
1307 list_splice_init(&era->rpc_calls, &calls);
1308 spin_unlock(&era->rpc_lock);
1310 list_for_each_entry_safe(rpc, tmp, &calls, list) {
1311 rpc->result = rpc->fn0 ? rpc->fn0(era->md) : rpc->fn1(era->md, rpc->arg);
1316 r = metadata_commit(era->md);
1318 list_for_each_entry_safe(rpc, tmp, &calls, list)
1322 list_for_each_entry_safe(rpc, tmp, &calls, list)
1323 complete(&rpc->complete);
1326 static void kick_off_digest(struct era *era)
1328 if (era->md->archived_writesets) {
1329 era->md->archived_writesets = false;
1330 metadata_digest_start(era->md, &era->digest);
1334 static void do_work(struct work_struct *ws)
1336 struct era *era = container_of(ws, struct era, worker);
1338 kick_off_digest(era);
1339 process_old_eras(era);
1340 process_deferred_bios(era);
1341 process_rpc_calls(era);
1344 static void defer_bio(struct era *era, struct bio *bio)
1346 spin_lock(&era->deferred_lock);
1347 bio_list_add(&era->deferred_bios, bio);
1348 spin_unlock(&era->deferred_lock);
1354 * Make an rpc call to the worker to change the metadata.
1356 static int perform_rpc(struct era *era, struct rpc *rpc)
1359 init_completion(&rpc->complete);
1361 spin_lock(&era->rpc_lock);
1362 list_add(&rpc->list, &era->rpc_calls);
1363 spin_unlock(&era->rpc_lock);
1366 wait_for_completion(&rpc->complete);
1371 static int in_worker0(struct era *era, int (*fn)(struct era_metadata *))
1377 return perform_rpc(era, &rpc);
1380 static int in_worker1(struct era *era,
1381 int (*fn)(struct era_metadata *, void *), void *arg)
1388 return perform_rpc(era, &rpc);
1391 static void start_worker(struct era *era)
1393 atomic_set(&era->suspended, 0);
1396 static void stop_worker(struct era *era)
1398 atomic_set(&era->suspended, 1);
1399 drain_workqueue(era->wq);
1402 /*----------------------------------------------------------------
1404 *--------------------------------------------------------------*/
1405 static void era_destroy(struct era *era)
1408 metadata_close(era->md);
1411 destroy_workqueue(era->wq);
1413 if (era->origin_dev)
1414 dm_put_device(era->ti, era->origin_dev);
1416 if (era->metadata_dev)
1417 dm_put_device(era->ti, era->metadata_dev);
1422 static dm_block_t calc_nr_blocks(struct era *era)
1424 return dm_sector_div_up(era->ti->len, era->sectors_per_block);
1427 static bool valid_block_size(dm_block_t block_size)
1429 bool greater_than_zero = block_size > 0;
1430 bool multiple_of_min_block_size = (block_size & (MIN_BLOCK_SIZE - 1)) == 0;
1432 return greater_than_zero && multiple_of_min_block_size;
1436 * <metadata dev> <data dev> <data block size (sectors)>
1438 static int era_ctr(struct dm_target *ti, unsigned argc, char **argv)
1443 struct era_metadata *md;
1446 ti->error = "Invalid argument count";
1450 era = kzalloc(sizeof(*era), GFP_KERNEL);
1452 ti->error = "Error allocating era structure";
1458 r = dm_get_device(ti, argv[0], FMODE_READ | FMODE_WRITE, &era->metadata_dev);
1460 ti->error = "Error opening metadata device";
1465 r = dm_get_device(ti, argv[1], FMODE_READ | FMODE_WRITE, &era->origin_dev);
1467 ti->error = "Error opening data device";
1472 r = sscanf(argv[2], "%u%c", &era->sectors_per_block, &dummy);
1474 ti->error = "Error parsing block size";
1479 r = dm_set_target_max_io_len(ti, era->sectors_per_block);
1481 ti->error = "could not set max io len";
1486 if (!valid_block_size(era->sectors_per_block)) {
1487 ti->error = "Invalid block size";
1491 if (era->sectors_per_block & (era->sectors_per_block - 1))
1492 era->sectors_per_block_shift = -1;
1494 era->sectors_per_block_shift = __ffs(era->sectors_per_block);
1496 md = metadata_open(era->metadata_dev->bdev, era->sectors_per_block, true);
1498 ti->error = "Error reading metadata";
1504 era->wq = alloc_ordered_workqueue("dm-" DM_MSG_PREFIX, WQ_MEM_RECLAIM);
1506 ti->error = "could not create workqueue for metadata object";
1510 INIT_WORK(&era->worker, do_work);
1512 spin_lock_init(&era->deferred_lock);
1513 bio_list_init(&era->deferred_bios);
1515 spin_lock_init(&era->rpc_lock);
1516 INIT_LIST_HEAD(&era->rpc_calls);
1519 ti->num_flush_bios = 1;
1520 ti->flush_supported = true;
1522 ti->num_discard_bios = 1;
1527 static void era_dtr(struct dm_target *ti)
1529 era_destroy(ti->private);
1532 static int era_map(struct dm_target *ti, struct bio *bio)
1534 struct era *era = ti->private;
1535 dm_block_t block = get_block(era, bio);
1538 * All bios get remapped to the origin device. We do this now, but
1539 * it may not get issued until later. Depending on whether the
1540 * block is marked in this era.
1542 remap_to_origin(era, bio);
1545 * REQ_PREFLUSH bios carry no data, so we're not interested in them.
1547 if (!(bio->bi_opf & REQ_PREFLUSH) &&
1548 (bio_data_dir(bio) == WRITE) &&
1549 !metadata_current_marked(era->md, block)) {
1550 defer_bio(era, bio);
1551 return DM_MAPIO_SUBMITTED;
1554 return DM_MAPIO_REMAPPED;
1557 static void era_postsuspend(struct dm_target *ti)
1560 struct era *era = ti->private;
1562 r = in_worker0(era, metadata_era_archive);
1564 DMERR("%s: couldn't archive current era", __func__);
1565 /* FIXME: fail mode */
1570 r = metadata_commit(era->md);
1572 DMERR("%s: metadata_commit failed", __func__);
1573 /* FIXME: fail mode */
1577 static int era_preresume(struct dm_target *ti)
1580 struct era *era = ti->private;
1581 dm_block_t new_size = calc_nr_blocks(era);
1583 if (era->nr_blocks != new_size) {
1584 r = metadata_resize(era->md, &new_size);
1586 DMERR("%s: metadata_resize failed", __func__);
1590 r = metadata_commit(era->md);
1592 DMERR("%s: metadata_commit failed", __func__);
1596 era->nr_blocks = new_size;
1601 r = in_worker0(era, metadata_era_rollover);
1603 DMERR("%s: metadata_era_rollover failed", __func__);
1613 * <metadata block size> <#used metadata blocks>/<#total metadata blocks>
1614 * <current era> <held metadata root | '-'>
1616 static void era_status(struct dm_target *ti, status_type_t type,
1617 unsigned status_flags, char *result, unsigned maxlen)
1620 struct era *era = ti->private;
1622 struct metadata_stats stats;
1623 char buf[BDEVNAME_SIZE];
1626 case STATUSTYPE_INFO:
1627 r = in_worker1(era, metadata_get_stats, &stats);
1631 DMEMIT("%u %llu/%llu %u",
1632 (unsigned) (DM_ERA_METADATA_BLOCK_SIZE >> SECTOR_SHIFT),
1633 (unsigned long long) stats.used,
1634 (unsigned long long) stats.total,
1635 (unsigned) stats.era);
1637 if (stats.snap != SUPERBLOCK_LOCATION)
1638 DMEMIT(" %llu", stats.snap);
1643 case STATUSTYPE_TABLE:
1644 format_dev_t(buf, era->metadata_dev->bdev->bd_dev);
1646 format_dev_t(buf, era->origin_dev->bdev->bd_dev);
1647 DMEMIT("%s %u", buf, era->sectors_per_block);
1657 static int era_message(struct dm_target *ti, unsigned argc, char **argv,
1658 char *result, unsigned maxlen)
1660 struct era *era = ti->private;
1663 DMERR("incorrect number of message arguments");
1667 if (!strcasecmp(argv[0], "checkpoint"))
1668 return in_worker0(era, metadata_checkpoint);
1670 if (!strcasecmp(argv[0], "take_metadata_snap"))
1671 return in_worker0(era, metadata_take_snap);
1673 if (!strcasecmp(argv[0], "drop_metadata_snap"))
1674 return in_worker0(era, metadata_drop_snap);
1676 DMERR("unsupported message '%s'", argv[0]);
1680 static sector_t get_dev_size(struct dm_dev *dev)
1682 return i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
1685 static int era_iterate_devices(struct dm_target *ti,
1686 iterate_devices_callout_fn fn, void *data)
1688 struct era *era = ti->private;
1689 return fn(ti, era->origin_dev, 0, get_dev_size(era->origin_dev), data);
1692 static void era_io_hints(struct dm_target *ti, struct queue_limits *limits)
1694 struct era *era = ti->private;
1695 uint64_t io_opt_sectors = limits->io_opt >> SECTOR_SHIFT;
1698 * If the system-determined stacked limits are compatible with the
1699 * era device's blocksize (io_opt is a factor) do not override them.
1701 if (io_opt_sectors < era->sectors_per_block ||
1702 do_div(io_opt_sectors, era->sectors_per_block)) {
1703 blk_limits_io_min(limits, 0);
1704 blk_limits_io_opt(limits, era->sectors_per_block << SECTOR_SHIFT);
1708 /*----------------------------------------------------------------*/
1710 static struct target_type era_target = {
1712 .version = {1, 0, 0},
1713 .module = THIS_MODULE,
1717 .postsuspend = era_postsuspend,
1718 .preresume = era_preresume,
1719 .status = era_status,
1720 .message = era_message,
1721 .iterate_devices = era_iterate_devices,
1722 .io_hints = era_io_hints
1725 static int __init dm_era_init(void)
1729 r = dm_register_target(&era_target);
1731 DMERR("era target registration failed: %d", r);
1738 static void __exit dm_era_exit(void)
1740 dm_unregister_target(&era_target);
1743 module_init(dm_era_init);
1744 module_exit(dm_era_exit);
1746 MODULE_DESCRIPTION(DM_NAME " era target");
1747 MODULE_AUTHOR("Joe Thornber <ejt@redhat.com>");
1748 MODULE_LICENSE("GPL");