2 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
9 #include <linux/module.h>
10 #include <linux/crc32.h>
12 #define DM_MSG_PREFIX "zoned metadata"
17 #define DMZ_META_VER 1
20 * On-disk super block magic.
22 #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
23 (((unsigned int)('Z')) << 16) | \
24 (((unsigned int)('B')) << 8) | \
25 ((unsigned int)('D')))
28 * On disk super block.
29 * This uses only 512 B but uses on disk a full 4KB block. This block is
30 * followed on disk by the mapping table of chunks to zones and the bitmap
31 * blocks indicating zone block validity.
32 * The overall resulting metadata format is:
33 * (1) Super block (1 block)
34 * (2) Chunk mapping table (nr_map_blocks)
35 * (3) Bitmap blocks (nr_bitmap_blocks)
36 * All metadata blocks are stored in conventional zones, starting from the
37 * the first conventional zone found on disk.
43 /* Metadata version number */
44 __le32 version; /* 8 */
46 /* Generation number */
49 /* This block number */
50 __le64 sb_block; /* 24 */
52 /* The number of metadata blocks, including this super block */
53 __le32 nr_meta_blocks; /* 28 */
55 /* The number of sequential zones reserved for reclaim */
56 __le32 nr_reserved_seq; /* 32 */
58 /* The number of entries in the mapping table */
59 __le32 nr_chunks; /* 36 */
61 /* The number of blocks used for the chunk mapping table */
62 __le32 nr_map_blocks; /* 40 */
64 /* The number of blocks used for the block bitmaps */
65 __le32 nr_bitmap_blocks; /* 44 */
70 /* Padding to full 512B sector */
71 u8 reserved[464]; /* 512 */
75 * Chunk mapping entry: entries are indexed by chunk number
76 * and give the zone ID (dzone_id) mapping the chunk on disk.
77 * This zone may be sequential or random. If it is a sequential
78 * zone, a second zone (bzone_id) used as a write buffer may
79 * also be specified. This second zone will always be a randomly
88 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
90 #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
91 #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
92 #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
93 #define DMZ_MAP_UNMAPPED UINT_MAX
96 * Meta data block descriptor (for cached metadata blocks).
100 struct list_head link;
109 * Metadata block state flags.
119 * Super block information (one per metadata set).
123 struct dmz_mblock *mblk;
124 struct dmz_super *sb;
128 * In-memory metadata.
130 struct dmz_metadata {
133 sector_t zone_bitmap_size;
134 unsigned int zone_nr_bitmap_blocks;
136 unsigned int nr_bitmap_blocks;
137 unsigned int nr_map_blocks;
139 unsigned int nr_useable_zones;
140 unsigned int nr_meta_blocks;
141 unsigned int nr_meta_zones;
142 unsigned int nr_data_zones;
143 unsigned int nr_rnd_zones;
144 unsigned int nr_reserved_seq;
145 unsigned int nr_chunks;
147 /* Zone information array */
148 struct dm_zone *zones;
150 struct dm_zone *sb_zone;
152 unsigned int mblk_primary;
154 unsigned int min_nr_mblks;
155 unsigned int max_nr_mblks;
157 struct rw_semaphore mblk_sem;
158 struct mutex mblk_flush_lock;
159 spinlock_t mblk_lock;
160 struct rb_root mblk_rbtree;
161 struct list_head mblk_lru_list;
162 struct list_head mblk_dirty_list;
163 struct shrinker mblk_shrinker;
165 /* Zone allocation management */
166 struct mutex map_lock;
167 struct dmz_mblock **map_mblk;
169 atomic_t unmap_nr_rnd;
170 struct list_head unmap_rnd_list;
171 struct list_head map_rnd_list;
174 atomic_t unmap_nr_seq;
175 struct list_head unmap_seq_list;
176 struct list_head map_seq_list;
178 atomic_t nr_reserved_seq_zones;
179 struct list_head reserved_seq_zones_list;
181 wait_queue_head_t free_wq;
187 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
189 return ((unsigned int)(zone - zmd->zones));
192 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
194 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
197 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
199 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
202 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
204 return zmd->nr_chunks;
207 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
212 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
214 return atomic_read(&zmd->unmap_nr_rnd);
218 * Lock/unlock mapping table.
219 * The map lock also protects all the zone lists.
221 void dmz_lock_map(struct dmz_metadata *zmd)
223 mutex_lock(&zmd->map_lock);
226 void dmz_unlock_map(struct dmz_metadata *zmd)
228 mutex_unlock(&zmd->map_lock);
232 * Lock/unlock metadata access. This is a "read" lock on a semaphore
233 * that prevents metadata flush from running while metadata are being
234 * modified. The actual metadata write mutual exclusion is achieved with
235 * the map lock and zone styate management (active and reclaim state are
236 * mutually exclusive).
238 void dmz_lock_metadata(struct dmz_metadata *zmd)
240 down_read(&zmd->mblk_sem);
243 void dmz_unlock_metadata(struct dmz_metadata *zmd)
245 up_read(&zmd->mblk_sem);
249 * Lock/unlock flush: prevent concurrent executions
250 * of dmz_flush_metadata as well as metadata modification in reclaim
251 * while flush is being executed.
253 void dmz_lock_flush(struct dmz_metadata *zmd)
255 mutex_lock(&zmd->mblk_flush_lock);
258 void dmz_unlock_flush(struct dmz_metadata *zmd)
260 mutex_unlock(&zmd->mblk_flush_lock);
264 * Allocate a metadata block.
266 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
269 struct dmz_mblock *mblk = NULL;
271 /* See if we can reuse cached blocks */
272 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
273 spin_lock(&zmd->mblk_lock);
274 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
275 struct dmz_mblock, link);
277 list_del_init(&mblk->link);
278 rb_erase(&mblk->node, &zmd->mblk_rbtree);
281 spin_unlock(&zmd->mblk_lock);
286 /* Allocate a new block */
287 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
291 mblk->page = alloc_page(GFP_NOIO);
297 RB_CLEAR_NODE(&mblk->node);
298 INIT_LIST_HEAD(&mblk->link);
299 atomic_set(&mblk->ref, 0);
302 mblk->data = page_address(mblk->page);
304 atomic_inc(&zmd->nr_mblks);
310 * Free a metadata block.
312 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
314 __free_pages(mblk->page, 0);
317 atomic_dec(&zmd->nr_mblks);
321 * Insert a metadata block in the rbtree.
323 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
325 struct rb_root *root = &zmd->mblk_rbtree;
326 struct rb_node **new = &(root->rb_node), *parent = NULL;
327 struct dmz_mblock *b;
329 /* Figure out where to put the new node */
331 b = container_of(*new, struct dmz_mblock, node);
333 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
336 /* Add new node and rebalance tree */
337 rb_link_node(&mblk->node, parent, new);
338 rb_insert_color(&mblk->node, root);
342 * Lookup a metadata block in the rbtree.
344 static struct dmz_mblock *dmz_lookup_mblock(struct dmz_metadata *zmd,
347 struct rb_root *root = &zmd->mblk_rbtree;
348 struct rb_node *node = root->rb_node;
349 struct dmz_mblock *mblk;
352 mblk = container_of(node, struct dmz_mblock, node);
353 if (mblk->no == mblk_no)
355 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
362 * Metadata block BIO end callback.
364 static void dmz_mblock_bio_end_io(struct bio *bio)
366 struct dmz_mblock *mblk = bio->bi_private;
370 set_bit(DMZ_META_ERROR, &mblk->state);
372 if (bio_op(bio) == REQ_OP_WRITE)
373 flag = DMZ_META_WRITING;
375 flag = DMZ_META_READING;
377 clear_bit_unlock(flag, &mblk->state);
378 smp_mb__after_atomic();
379 wake_up_bit(&mblk->state, flag);
385 * Read a metadata block from disk.
387 static struct dmz_mblock *dmz_fetch_mblock(struct dmz_metadata *zmd,
390 struct dmz_mblock *mblk;
391 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
394 /* Get block and insert it */
395 mblk = dmz_alloc_mblock(zmd, mblk_no);
399 spin_lock(&zmd->mblk_lock);
400 atomic_inc(&mblk->ref);
401 set_bit(DMZ_META_READING, &mblk->state);
402 dmz_insert_mblock(zmd, mblk);
403 spin_unlock(&zmd->mblk_lock);
405 bio = bio_alloc(GFP_NOIO, 1);
407 dmz_free_mblock(zmd, mblk);
411 bio->bi_iter.bi_sector = dmz_blk2sect(block);
412 bio->bi_bdev = zmd->dev->bdev;
413 bio->bi_private = mblk;
414 bio->bi_end_io = dmz_mblock_bio_end_io;
415 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
416 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
423 * Free metadata blocks.
425 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
428 struct dmz_mblock *mblk;
429 unsigned long count = 0;
431 if (!zmd->max_nr_mblks)
434 while (!list_empty(&zmd->mblk_lru_list) &&
435 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
437 mblk = list_first_entry(&zmd->mblk_lru_list,
438 struct dmz_mblock, link);
439 list_del_init(&mblk->link);
440 rb_erase(&mblk->node, &zmd->mblk_rbtree);
441 dmz_free_mblock(zmd, mblk);
449 * For mblock shrinker: get the number of unused metadata blocks in the cache.
451 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
452 struct shrink_control *sc)
454 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
456 return atomic_read(&zmd->nr_mblks);
460 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
462 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
463 struct shrink_control *sc)
465 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
468 spin_lock(&zmd->mblk_lock);
469 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
470 spin_unlock(&zmd->mblk_lock);
472 return count ? count : SHRINK_STOP;
476 * Release a metadata block.
478 static void dmz_release_mblock(struct dmz_metadata *zmd,
479 struct dmz_mblock *mblk)
485 spin_lock(&zmd->mblk_lock);
487 if (atomic_dec_and_test(&mblk->ref)) {
488 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
489 rb_erase(&mblk->node, &zmd->mblk_rbtree);
490 dmz_free_mblock(zmd, mblk);
491 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
492 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
493 dmz_shrink_mblock_cache(zmd, 1);
497 spin_unlock(&zmd->mblk_lock);
501 * Get a metadata block from the rbtree. If the block
502 * is not present, read it from disk.
504 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
507 struct dmz_mblock *mblk;
510 spin_lock(&zmd->mblk_lock);
511 mblk = dmz_lookup_mblock(zmd, mblk_no);
513 /* Cache hit: remove block from LRU list */
514 if (atomic_inc_return(&mblk->ref) == 1 &&
515 !test_bit(DMZ_META_DIRTY, &mblk->state))
516 list_del_init(&mblk->link);
518 spin_unlock(&zmd->mblk_lock);
521 /* Cache miss: read the block from disk */
522 mblk = dmz_fetch_mblock(zmd, mblk_no);
524 return ERR_PTR(-ENOMEM);
527 /* Wait for on-going read I/O and check for error */
528 wait_on_bit_io(&mblk->state, DMZ_META_READING,
529 TASK_UNINTERRUPTIBLE);
530 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
531 dmz_release_mblock(zmd, mblk);
532 return ERR_PTR(-EIO);
539 * Mark a metadata block dirty.
541 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
543 spin_lock(&zmd->mblk_lock);
544 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
545 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
546 spin_unlock(&zmd->mblk_lock);
550 * Issue a metadata block write BIO.
552 static void dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
555 sector_t block = zmd->sb[set].block + mblk->no;
558 bio = bio_alloc(GFP_NOIO, 1);
560 set_bit(DMZ_META_ERROR, &mblk->state);
564 set_bit(DMZ_META_WRITING, &mblk->state);
566 bio->bi_iter.bi_sector = dmz_blk2sect(block);
567 bio->bi_bdev = zmd->dev->bdev;
568 bio->bi_private = mblk;
569 bio->bi_end_io = dmz_mblock_bio_end_io;
570 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
571 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
576 * Read/write a metadata block.
578 static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
584 bio = bio_alloc(GFP_NOIO, 1);
588 bio->bi_iter.bi_sector = dmz_blk2sect(block);
589 bio->bi_bdev = zmd->dev->bdev;
590 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
591 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
592 ret = submit_bio_wait(bio);
599 * Write super block of the specified metadata set.
601 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
603 sector_t block = zmd->sb[set].block;
604 struct dmz_mblock *mblk = zmd->sb[set].mblk;
605 struct dmz_super *sb = zmd->sb[set].sb;
606 u64 sb_gen = zmd->sb_gen + 1;
609 sb->magic = cpu_to_le32(DMZ_MAGIC);
610 sb->version = cpu_to_le32(DMZ_META_VER);
612 sb->gen = cpu_to_le64(sb_gen);
614 sb->sb_block = cpu_to_le64(block);
615 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
616 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
617 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
619 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
620 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
623 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
625 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
627 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_KERNEL, NULL);
633 * Write dirty metadata blocks to the specified set.
635 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
636 struct list_head *write_list,
639 struct dmz_mblock *mblk;
640 struct blk_plug plug;
644 blk_start_plug(&plug);
645 list_for_each_entry(mblk, write_list, link)
646 dmz_write_mblock(zmd, mblk, set);
647 blk_finish_plug(&plug);
649 /* Wait for completion */
650 list_for_each_entry(mblk, write_list, link) {
651 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
652 TASK_UNINTERRUPTIBLE);
653 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
654 clear_bit(DMZ_META_ERROR, &mblk->state);
659 /* Flush drive cache (this will also sync data) */
661 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_KERNEL, NULL);
667 * Log dirty metadata blocks.
669 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
670 struct list_head *write_list)
672 unsigned int log_set = zmd->mblk_primary ^ 0x1;
675 /* Write dirty blocks to the log */
676 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
681 * No error so far: now validate the log by updating the
682 * log index super block generation.
684 ret = dmz_write_sb(zmd, log_set);
692 * Flush dirty metadata blocks.
694 int dmz_flush_metadata(struct dmz_metadata *zmd)
696 struct dmz_mblock *mblk;
697 struct list_head write_list;
703 INIT_LIST_HEAD(&write_list);
706 * Make sure that metadata blocks are stable before logging: take
707 * the write lock on the metadata semaphore to prevent target BIOs
708 * from modifying metadata.
710 down_write(&zmd->mblk_sem);
713 * This is called from the target flush work and reclaim work.
714 * Concurrent execution is not allowed.
718 /* Get dirty blocks */
719 spin_lock(&zmd->mblk_lock);
720 list_splice_init(&zmd->mblk_dirty_list, &write_list);
721 spin_unlock(&zmd->mblk_lock);
723 /* If there are no dirty metadata blocks, just flush the device cache */
724 if (list_empty(&write_list)) {
725 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_KERNEL, NULL);
730 * The primary metadata set is still clean. Keep it this way until
731 * all updates are successful in the secondary set. That is, use
732 * the secondary set as a log.
734 ret = dmz_log_dirty_mblocks(zmd, &write_list);
739 * The log is on disk. It is now safe to update in place
740 * in the primary metadata set.
742 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
746 ret = dmz_write_sb(zmd, zmd->mblk_primary);
750 while (!list_empty(&write_list)) {
751 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
752 list_del_init(&mblk->link);
754 spin_lock(&zmd->mblk_lock);
755 clear_bit(DMZ_META_DIRTY, &mblk->state);
756 if (atomic_read(&mblk->ref) == 0)
757 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
758 spin_unlock(&zmd->mblk_lock);
763 if (ret && !list_empty(&write_list)) {
764 spin_lock(&zmd->mblk_lock);
765 list_splice(&write_list, &zmd->mblk_dirty_list);
766 spin_unlock(&zmd->mblk_lock);
769 dmz_unlock_flush(zmd);
770 up_write(&zmd->mblk_sem);
778 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
780 unsigned int nr_meta_zones, nr_data_zones;
781 struct dmz_dev *dev = zmd->dev;
785 gen = le64_to_cpu(sb->gen);
786 stored_crc = le32_to_cpu(sb->crc);
788 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
789 if (crc != stored_crc) {
790 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
795 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
796 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
797 DMZ_MAGIC, le32_to_cpu(sb->magic));
801 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
802 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
803 DMZ_META_VER, le32_to_cpu(sb->version));
807 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
808 >> dev->zone_nr_blocks_shift;
809 if (!nr_meta_zones ||
810 nr_meta_zones >= zmd->nr_rnd_zones) {
811 dmz_dev_err(dev, "Invalid number of metadata blocks");
815 if (!le32_to_cpu(sb->nr_reserved_seq) ||
816 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
817 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
821 nr_data_zones = zmd->nr_useable_zones -
822 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
823 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
824 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
825 le32_to_cpu(sb->nr_chunks), nr_data_zones);
830 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
831 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
832 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
833 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
834 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
835 zmd->nr_meta_zones = nr_meta_zones;
836 zmd->nr_data_zones = nr_data_zones;
842 * Read the first or second super block from disk.
844 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
846 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
847 zmd->sb[set].mblk->page);
851 * Determine the position of the secondary super blocks on disk.
852 * This is used only if a corruption of the primary super block
855 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
857 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
858 struct dmz_mblock *mblk;
861 /* Allocate a block */
862 mblk = dmz_alloc_mblock(zmd, 0);
866 zmd->sb[1].mblk = mblk;
867 zmd->sb[1].sb = mblk->data;
869 /* Bad first super block: search for the second one */
870 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
871 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
872 if (dmz_read_sb(zmd, 1) != 0)
874 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
876 zmd->sb[1].block += zone_nr_blocks;
879 dmz_free_mblock(zmd, mblk);
880 zmd->sb[1].mblk = NULL;
886 * Read the first or second super block from disk.
888 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
890 struct dmz_mblock *mblk;
893 /* Allocate a block */
894 mblk = dmz_alloc_mblock(zmd, 0);
898 zmd->sb[set].mblk = mblk;
899 zmd->sb[set].sb = mblk->data;
901 /* Read super block */
902 ret = dmz_read_sb(zmd, set);
904 dmz_free_mblock(zmd, mblk);
905 zmd->sb[set].mblk = NULL;
913 * Recover a metadata set.
915 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
917 unsigned int src_set = dst_set ^ 0x1;
921 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
924 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
926 zmd->sb[1].block = zmd->sb[0].block +
927 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
930 page = alloc_page(GFP_KERNEL);
934 /* Copy metadata blocks */
935 for (i = 1; i < zmd->nr_meta_blocks; i++) {
936 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
937 zmd->sb[src_set].block + i, page);
940 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
941 zmd->sb[dst_set].block + i, page);
946 /* Finalize with the super block */
947 if (!zmd->sb[dst_set].mblk) {
948 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
949 if (!zmd->sb[dst_set].mblk) {
953 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
956 ret = dmz_write_sb(zmd, dst_set);
958 __free_pages(page, 0);
964 * Get super block from disk.
966 static int dmz_load_sb(struct dmz_metadata *zmd)
968 bool sb_good[2] = {false, false};
969 u64 sb_gen[2] = {0, 0};
972 /* Read and check the primary super block */
973 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
974 ret = dmz_get_sb(zmd, 0);
976 dmz_dev_err(zmd->dev, "Read primary super block failed");
980 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
982 /* Read and check secondary super block */
985 zmd->sb[1].block = zmd->sb[0].block +
986 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
987 ret = dmz_get_sb(zmd, 1);
989 ret = dmz_lookup_secondary_sb(zmd);
992 dmz_dev_err(zmd->dev, "Read secondary super block failed");
996 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1000 /* Use highest generation sb first */
1001 if (!sb_good[0] && !sb_good[1]) {
1002 dmz_dev_err(zmd->dev, "No valid super block found");
1007 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1009 ret = dmz_recover_mblocks(zmd, 0);
1012 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1014 ret = dmz_recover_mblocks(zmd, 1);
1017 dmz_dev_err(zmd->dev, "Recovery failed");
1021 if (sb_gen[0] >= sb_gen[1]) {
1022 zmd->sb_gen = sb_gen[0];
1023 zmd->mblk_primary = 0;
1025 zmd->sb_gen = sb_gen[1];
1026 zmd->mblk_primary = 1;
1029 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1030 zmd->mblk_primary, zmd->sb_gen);
1036 * Initialize a zone descriptor.
1038 static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1039 struct blk_zone *blkz)
1041 struct dmz_dev *dev = zmd->dev;
1043 /* Ignore the eventual last runt (smaller) zone */
1044 if (blkz->len != dev->zone_nr_sectors) {
1045 if (blkz->start + blkz->len == dev->capacity)
1050 INIT_LIST_HEAD(&zone->link);
1051 atomic_set(&zone->refcount, 0);
1052 zone->chunk = DMZ_MAP_UNMAPPED;
1054 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1055 set_bit(DMZ_RND, &zone->flags);
1056 zmd->nr_rnd_zones++;
1057 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1058 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1059 set_bit(DMZ_SEQ, &zone->flags);
1063 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1064 set_bit(DMZ_OFFLINE, &zone->flags);
1065 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1066 set_bit(DMZ_READ_ONLY, &zone->flags);
1068 if (dmz_is_rnd(zone))
1071 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1073 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1074 zmd->nr_useable_zones++;
1075 if (dmz_is_rnd(zone)) {
1076 zmd->nr_rnd_zones++;
1077 if (!zmd->sb_zone) {
1078 /* Super block zone */
1079 zmd->sb_zone = zone;
1088 * Free zones descriptors.
1090 static void dmz_drop_zones(struct dmz_metadata *zmd)
1097 * The size of a zone report in number of zones.
1098 * This results in 4096*64B=256KB report zones commands.
1100 #define DMZ_REPORT_NR_ZONES 4096
1103 * Allocate and initialize zone descriptors using the zone
1104 * information from disk.
1106 static int dmz_init_zones(struct dmz_metadata *zmd)
1108 struct dmz_dev *dev = zmd->dev;
1109 struct dm_zone *zone;
1110 struct blk_zone *blkz;
1111 unsigned int nr_blkz;
1112 sector_t sector = 0;
1116 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1117 zmd->zone_nr_bitmap_blocks = zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT;
1119 /* Allocate zone array */
1120 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1124 dmz_dev_info(dev, "Using %zu B for zone information",
1125 sizeof(struct dm_zone) * dev->nr_zones);
1127 /* Get zone information */
1128 nr_blkz = DMZ_REPORT_NR_ZONES;
1129 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1136 * Get zone information and initialize zone descriptors.
1137 * At the same time, determine where the super block
1138 * should be: first block of the first randomly writable
1142 while (sector < dev->capacity) {
1143 /* Get zone information */
1144 nr_blkz = DMZ_REPORT_NR_ZONES;
1145 ret = blkdev_report_zones(dev->bdev, sector, blkz,
1146 &nr_blkz, GFP_KERNEL);
1148 dmz_dev_err(dev, "Report zones failed %d", ret);
1152 /* Process report */
1153 for (i = 0; i < nr_blkz; i++) {
1154 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1157 sector += dev->zone_nr_sectors;
1162 /* The entire zone configuration of the disk should now be known */
1163 if (sector < dev->capacity) {
1164 dmz_dev_err(dev, "Failed to get correct zone information");
1170 dmz_drop_zones(zmd);
1176 * Update a zone information.
1178 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1180 unsigned int nr_blkz = 1;
1181 struct blk_zone blkz;
1184 /* Get zone information from disk */
1185 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1186 &blkz, &nr_blkz, GFP_KERNEL);
1188 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1193 clear_bit(DMZ_OFFLINE, &zone->flags);
1194 clear_bit(DMZ_READ_ONLY, &zone->flags);
1195 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1196 set_bit(DMZ_OFFLINE, &zone->flags);
1197 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1198 set_bit(DMZ_READ_ONLY, &zone->flags);
1200 if (dmz_is_seq(zone))
1201 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1209 * Check a zone write pointer position when the zone is marked
1210 * with the sequential write error flag.
1212 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1213 struct dm_zone *zone)
1215 unsigned int wp = 0;
1218 wp = zone->wp_block;
1219 ret = dmz_update_zone(zmd, zone);
1223 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1224 dmz_id(zmd, zone), zone->wp_block, wp);
1226 if (zone->wp_block < wp) {
1227 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1228 wp - zone->wp_block);
1234 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1236 return &zmd->zones[zone_id];
1240 * Reset a zone write pointer.
1242 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1247 * Ignore offline zones, read only zones,
1248 * and conventional zones.
1250 if (dmz_is_offline(zone) ||
1251 dmz_is_readonly(zone) ||
1255 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1256 struct dmz_dev *dev = zmd->dev;
1258 ret = blkdev_reset_zones(dev->bdev,
1259 dmz_start_sect(zmd, zone),
1260 dev->zone_nr_sectors, GFP_KERNEL);
1262 dmz_dev_err(dev, "Reset zone %u failed %d",
1263 dmz_id(zmd, zone), ret);
1268 /* Clear write error bit and rewind write pointer position */
1269 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1275 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1278 * Initialize chunk mapping.
1280 static int dmz_load_mapping(struct dmz_metadata *zmd)
1282 struct dmz_dev *dev = zmd->dev;
1283 struct dm_zone *dzone, *bzone;
1284 struct dmz_mblock *dmap_mblk = NULL;
1285 struct dmz_map *dmap;
1286 unsigned int i = 0, e = 0, chunk = 0;
1287 unsigned int dzone_id;
1288 unsigned int bzone_id;
1290 /* Metadata block array for the chunk mapping table */
1291 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1292 sizeof(struct dmz_mblk *), GFP_KERNEL);
1296 /* Get chunk mapping table blocks and initialize zone mapping */
1297 while (chunk < zmd->nr_chunks) {
1299 /* Get mapping block */
1300 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1301 if (IS_ERR(dmap_mblk))
1302 return PTR_ERR(dmap_mblk);
1303 zmd->map_mblk[i] = dmap_mblk;
1304 dmap = (struct dmz_map *) dmap_mblk->data;
1309 /* Check data zone */
1310 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1311 if (dzone_id == DMZ_MAP_UNMAPPED)
1314 if (dzone_id >= dev->nr_zones) {
1315 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1320 dzone = dmz_get(zmd, dzone_id);
1321 set_bit(DMZ_DATA, &dzone->flags);
1322 dzone->chunk = chunk;
1323 dmz_get_zone_weight(zmd, dzone);
1325 if (dmz_is_rnd(dzone))
1326 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1328 list_add_tail(&dzone->link, &zmd->map_seq_list);
1330 /* Check buffer zone */
1331 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1332 if (bzone_id == DMZ_MAP_UNMAPPED)
1335 if (bzone_id >= dev->nr_zones) {
1336 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1341 bzone = dmz_get(zmd, bzone_id);
1342 if (!dmz_is_rnd(bzone)) {
1343 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1348 set_bit(DMZ_DATA, &bzone->flags);
1349 set_bit(DMZ_BUF, &bzone->flags);
1350 bzone->chunk = chunk;
1351 bzone->bzone = dzone;
1352 dzone->bzone = bzone;
1353 dmz_get_zone_weight(zmd, bzone);
1354 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1358 if (e >= DMZ_MAP_ENTRIES)
1363 * At this point, only meta zones and mapped data zones were
1364 * fully initialized. All remaining zones are unmapped data
1365 * zones. Finish initializing those here.
1367 for (i = 0; i < dev->nr_zones; i++) {
1368 dzone = dmz_get(zmd, i);
1369 if (dmz_is_meta(dzone))
1372 if (dmz_is_rnd(dzone))
1377 if (dmz_is_data(dzone)) {
1378 /* Already initialized */
1382 /* Unmapped data zone */
1383 set_bit(DMZ_DATA, &dzone->flags);
1384 dzone->chunk = DMZ_MAP_UNMAPPED;
1385 if (dmz_is_rnd(dzone)) {
1386 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1387 atomic_inc(&zmd->unmap_nr_rnd);
1388 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1389 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1390 atomic_inc(&zmd->nr_reserved_seq_zones);
1393 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1394 atomic_inc(&zmd->unmap_nr_seq);
1402 * Set a data chunk mapping.
1404 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1405 unsigned int dzone_id, unsigned int bzone_id)
1407 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1408 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1409 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1411 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1412 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1413 dmz_dirty_mblock(zmd, dmap_mblk);
1417 * The list of mapped zones is maintained in LRU order.
1418 * This rotates a zone at the end of its map list.
1420 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1422 if (list_empty(&zone->link))
1425 list_del_init(&zone->link);
1426 if (dmz_is_seq(zone)) {
1427 /* LRU rotate sequential zone */
1428 list_add_tail(&zone->link, &zmd->map_seq_list);
1430 /* LRU rotate random zone */
1431 list_add_tail(&zone->link, &zmd->map_rnd_list);
1436 * The list of mapped random zones is maintained
1437 * in LRU order. This rotates a zone at the end of the list.
1439 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1441 __dmz_lru_zone(zmd, zone);
1443 __dmz_lru_zone(zmd, zone->bzone);
1447 * Wait for any zone to be freed.
1449 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1453 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1454 dmz_unlock_map(zmd);
1455 dmz_unlock_metadata(zmd);
1457 io_schedule_timeout(HZ);
1459 dmz_lock_metadata(zmd);
1461 finish_wait(&zmd->free_wq, &wait);
1465 * Lock a zone for reclaim (set the zone RECLAIM bit).
1466 * Returns false if the zone cannot be locked or if it is already locked
1469 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1471 /* Active zones cannot be reclaimed */
1472 if (dmz_is_active(zone))
1475 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1479 * Clear a zone reclaim flag.
1481 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1483 WARN_ON(dmz_is_active(zone));
1484 WARN_ON(!dmz_in_reclaim(zone));
1486 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1487 smp_mb__after_atomic();
1488 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1492 * Wait for a zone reclaim to complete.
1494 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1496 dmz_unlock_map(zmd);
1497 dmz_unlock_metadata(zmd);
1498 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1499 dmz_lock_metadata(zmd);
1504 * Select a random write zone for reclaim.
1506 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1508 struct dm_zone *dzone = NULL;
1509 struct dm_zone *zone;
1511 if (list_empty(&zmd->map_rnd_list))
1514 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1515 if (dmz_is_buf(zone))
1516 dzone = zone->bzone;
1519 if (dmz_lock_zone_reclaim(dzone))
1527 * Select a buffered sequential zone for reclaim.
1529 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1531 struct dm_zone *zone;
1533 if (list_empty(&zmd->map_seq_list))
1536 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1539 if (dmz_lock_zone_reclaim(zone))
1547 * Select a zone for reclaim.
1549 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1551 struct dm_zone *zone;
1554 * Search for a zone candidate to reclaim: 2 cases are possible.
1555 * (1) There is no free sequential zones. Then a random data zone
1556 * cannot be reclaimed. So choose a sequential zone to reclaim so
1557 * that afterward a random zone can be reclaimed.
1558 * (2) At least one free sequential zone is available, then choose
1559 * the oldest random zone (data or buffer) that can be locked.
1562 if (list_empty(&zmd->reserved_seq_zones_list))
1563 zone = dmz_get_seq_zone_for_reclaim(zmd);
1565 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1566 dmz_unlock_map(zmd);
1572 * Activate a zone (increment its reference count).
1574 void dmz_activate_zone(struct dm_zone *zone)
1576 set_bit(DMZ_ACTIVE, &zone->flags);
1577 atomic_inc(&zone->refcount);
1581 * Deactivate a zone. This decrement the zone reference counter
1582 * and clears the active state of the zone once the count reaches 0,
1583 * indicating that all BIOs to the zone have completed. Returns
1584 * true if the zone was deactivated.
1586 void dmz_deactivate_zone(struct dm_zone *zone)
1588 if (atomic_dec_and_test(&zone->refcount)) {
1589 WARN_ON(!test_bit(DMZ_ACTIVE, &zone->flags));
1590 clear_bit_unlock(DMZ_ACTIVE, &zone->flags);
1591 smp_mb__after_atomic();
1596 * Get the zone mapping a chunk, if the chunk is mapped already.
1597 * If no mapping exist and the operation is WRITE, a zone is
1598 * allocated and used to map the chunk.
1599 * The zone returned will be set to the active state.
1601 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1603 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1604 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1605 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1606 unsigned int dzone_id;
1607 struct dm_zone *dzone = NULL;
1612 /* Get the chunk mapping */
1613 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1614 if (dzone_id == DMZ_MAP_UNMAPPED) {
1616 * Read or discard in unmapped chunks are fine. But for
1617 * writes, we need a mapping, so get one.
1619 if (op != REQ_OP_WRITE)
1622 /* Alloate a random zone */
1623 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1625 dmz_wait_for_free_zones(zmd);
1629 dmz_map_zone(zmd, dzone, chunk);
1632 /* The chunk is already mapped: get the mapping zone */
1633 dzone = dmz_get(zmd, dzone_id);
1634 if (dzone->chunk != chunk) {
1635 dzone = ERR_PTR(-EIO);
1639 /* Repair write pointer if the sequential dzone has error */
1640 if (dmz_seq_write_err(dzone)) {
1641 ret = dmz_handle_seq_write_err(zmd, dzone);
1643 dzone = ERR_PTR(-EIO);
1646 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1651 * If the zone is being reclaimed, the chunk mapping may change
1652 * to a different zone. So wait for reclaim and retry. Otherwise,
1653 * activate the zone (this will prevent reclaim from touching it).
1655 if (dmz_in_reclaim(dzone)) {
1656 dmz_wait_for_reclaim(zmd, dzone);
1659 dmz_activate_zone(dzone);
1660 dmz_lru_zone(zmd, dzone);
1662 dmz_unlock_map(zmd);
1668 * Write and discard change the block validity of data zones and their buffer
1669 * zones. Check here that valid blocks are still present. If all blocks are
1670 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1673 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1675 struct dm_zone *bzone;
1679 bzone = dzone->bzone;
1681 if (dmz_weight(bzone))
1682 dmz_lru_zone(zmd, bzone);
1684 /* Empty buffer zone: reclaim it */
1685 dmz_unmap_zone(zmd, bzone);
1686 dmz_free_zone(zmd, bzone);
1691 /* Deactivate the data zone */
1692 dmz_deactivate_zone(dzone);
1693 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1694 dmz_lru_zone(zmd, dzone);
1696 /* Unbuffered inactive empty data zone: reclaim it */
1697 dmz_unmap_zone(zmd, dzone);
1698 dmz_free_zone(zmd, dzone);
1701 dmz_unlock_map(zmd);
1705 * Allocate and map a random zone to buffer a chunk
1706 * already mapped to a sequential zone.
1708 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1709 struct dm_zone *dzone)
1711 struct dm_zone *bzone;
1715 bzone = dzone->bzone;
1719 /* Alloate a random zone */
1720 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1722 dmz_wait_for_free_zones(zmd);
1726 /* Update the chunk mapping */
1727 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1728 dmz_id(zmd, bzone));
1730 set_bit(DMZ_BUF, &bzone->flags);
1731 bzone->chunk = dzone->chunk;
1732 bzone->bzone = dzone;
1733 dzone->bzone = bzone;
1734 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1736 dmz_unlock_map(zmd);
1742 * Get an unmapped (free) zone.
1743 * This must be called with the mapping lock held.
1745 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1747 struct list_head *list;
1748 struct dm_zone *zone;
1750 if (flags & DMZ_ALLOC_RND)
1751 list = &zmd->unmap_rnd_list;
1753 list = &zmd->unmap_seq_list;
1755 if (list_empty(list)) {
1757 * No free zone: if this is for reclaim, allow using the
1758 * reserved sequential zones.
1760 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1761 list_empty(&zmd->reserved_seq_zones_list))
1764 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1765 struct dm_zone, link);
1766 list_del_init(&zone->link);
1767 atomic_dec(&zmd->nr_reserved_seq_zones);
1771 zone = list_first_entry(list, struct dm_zone, link);
1772 list_del_init(&zone->link);
1774 if (dmz_is_rnd(zone))
1775 atomic_dec(&zmd->unmap_nr_rnd);
1777 atomic_dec(&zmd->unmap_nr_seq);
1779 if (dmz_is_offline(zone)) {
1780 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1790 * This must be called with the mapping lock held.
1792 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1794 /* If this is a sequential zone, reset it */
1795 if (dmz_is_seq(zone))
1796 dmz_reset_zone(zmd, zone);
1798 /* Return the zone to its type unmap list */
1799 if (dmz_is_rnd(zone)) {
1800 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1801 atomic_inc(&zmd->unmap_nr_rnd);
1802 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1803 zmd->nr_reserved_seq) {
1804 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1805 atomic_inc(&zmd->nr_reserved_seq_zones);
1807 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1808 atomic_inc(&zmd->unmap_nr_seq);
1811 wake_up_all(&zmd->free_wq);
1815 * Map a chunk to a zone.
1816 * This must be called with the mapping lock held.
1818 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1821 /* Set the chunk mapping */
1822 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1824 dzone->chunk = chunk;
1825 if (dmz_is_rnd(dzone))
1826 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1828 list_add_tail(&dzone->link, &zmd->map_seq_list);
1833 * This must be called with the mapping lock held.
1835 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1837 unsigned int chunk = zone->chunk;
1838 unsigned int dzone_id;
1840 if (chunk == DMZ_MAP_UNMAPPED) {
1841 /* Already unmapped */
1845 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1847 * Unmapping the chunk buffer zone: clear only
1848 * the chunk buffer mapping
1850 dzone_id = dmz_id(zmd, zone->bzone);
1851 zone->bzone->bzone = NULL;
1856 * Unmapping the chunk data zone: the zone must
1859 if (WARN_ON(zone->bzone)) {
1860 zone->bzone->bzone = NULL;
1863 dzone_id = DMZ_MAP_UNMAPPED;
1866 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1868 zone->chunk = DMZ_MAP_UNMAPPED;
1869 list_del_init(&zone->link);
1873 * Set @nr_bits bits in @bitmap starting from @bit.
1874 * Return the number of bits changed from 0 to 1.
1876 static unsigned int dmz_set_bits(unsigned long *bitmap,
1877 unsigned int bit, unsigned int nr_bits)
1879 unsigned long *addr;
1880 unsigned int end = bit + nr_bits;
1884 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1885 ((end - bit) >= BITS_PER_LONG)) {
1886 /* Try to set the whole word at once */
1887 addr = bitmap + BIT_WORD(bit);
1891 bit += BITS_PER_LONG;
1896 if (!test_and_set_bit(bit, bitmap))
1905 * Get the bitmap block storing the bit for chunk_block in zone.
1907 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1908 struct dm_zone *zone,
1909 sector_t chunk_block)
1911 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1912 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1913 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1915 return dmz_get_mblock(zmd, bitmap_block);
1919 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1921 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1922 struct dm_zone *to_zone)
1924 struct dmz_mblock *from_mblk, *to_mblk;
1925 sector_t chunk_block = 0;
1927 /* Get the zones bitmap blocks */
1928 while (chunk_block < zmd->dev->zone_nr_blocks) {
1929 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1930 if (IS_ERR(from_mblk))
1931 return PTR_ERR(from_mblk);
1932 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1933 if (IS_ERR(to_mblk)) {
1934 dmz_release_mblock(zmd, from_mblk);
1935 return PTR_ERR(to_mblk);
1938 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1939 dmz_dirty_mblock(zmd, to_mblk);
1941 dmz_release_mblock(zmd, to_mblk);
1942 dmz_release_mblock(zmd, from_mblk);
1944 chunk_block += DMZ_BLOCK_SIZE_BITS;
1947 to_zone->weight = from_zone->weight;
1953 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1954 * starting from chunk_block.
1956 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1957 struct dm_zone *to_zone, sector_t chunk_block)
1959 unsigned int nr_blocks;
1962 /* Get the zones bitmap blocks */
1963 while (chunk_block < zmd->dev->zone_nr_blocks) {
1964 /* Get a valid region from the source zone */
1965 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1970 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
1974 chunk_block += nr_blocks;
1981 * Validate all the blocks in the range [block..block+nr_blocks-1].
1983 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
1984 sector_t chunk_block, unsigned int nr_blocks)
1986 unsigned int count, bit, nr_bits;
1987 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
1988 struct dmz_mblock *mblk;
1991 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
1992 dmz_id(zmd, zone), (unsigned long long)chunk_block,
1995 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
1998 /* Get bitmap block */
1999 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2001 return PTR_ERR(mblk);
2004 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2005 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2007 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2009 dmz_dirty_mblock(zmd, mblk);
2012 dmz_release_mblock(zmd, mblk);
2014 nr_blocks -= nr_bits;
2015 chunk_block += nr_bits;
2018 if (likely(zone->weight + n <= zone_nr_blocks))
2021 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2022 dmz_id(zmd, zone), zone->weight,
2023 zone_nr_blocks - n);
2024 zone->weight = zone_nr_blocks;
2031 * Clear nr_bits bits in bitmap starting from bit.
2032 * Return the number of bits cleared.
2034 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2036 unsigned long *addr;
2037 int end = bit + nr_bits;
2041 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2042 ((end - bit) >= BITS_PER_LONG)) {
2043 /* Try to clear whole word at once */
2044 addr = bitmap + BIT_WORD(bit);
2045 if (*addr == ULONG_MAX) {
2048 bit += BITS_PER_LONG;
2053 if (test_and_clear_bit(bit, bitmap))
2062 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2064 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2065 sector_t chunk_block, unsigned int nr_blocks)
2067 unsigned int count, bit, nr_bits;
2068 struct dmz_mblock *mblk;
2071 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2072 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2074 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2077 /* Get bitmap block */
2078 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2080 return PTR_ERR(mblk);
2083 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2084 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2086 count = dmz_clear_bits((unsigned long *)mblk->data,
2089 dmz_dirty_mblock(zmd, mblk);
2092 dmz_release_mblock(zmd, mblk);
2094 nr_blocks -= nr_bits;
2095 chunk_block += nr_bits;
2098 if (zone->weight >= n)
2101 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2102 dmz_id(zmd, zone), zone->weight, n);
2110 * Get a block bit value.
2112 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2113 sector_t chunk_block)
2115 struct dmz_mblock *mblk;
2118 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2120 /* Get bitmap block */
2121 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2123 return PTR_ERR(mblk);
2126 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2127 (unsigned long *) mblk->data) != 0;
2129 dmz_release_mblock(zmd, mblk);
2135 * Return the number of blocks from chunk_block to the first block with a bit
2136 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2138 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2139 sector_t chunk_block, unsigned int nr_blocks,
2142 struct dmz_mblock *mblk;
2143 unsigned int bit, set_bit, nr_bits;
2144 unsigned long *bitmap;
2147 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2150 /* Get bitmap block */
2151 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2153 return PTR_ERR(mblk);
2156 bitmap = (unsigned long *) mblk->data;
2157 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2158 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2160 set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2162 set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2163 dmz_release_mblock(zmd, mblk);
2166 if (set_bit < DMZ_BLOCK_SIZE_BITS)
2169 nr_blocks -= nr_bits;
2170 chunk_block += nr_bits;
2177 * Test if chunk_block is valid. If it is, the number of consecutive
2178 * valid blocks from chunk_block will be returned.
2180 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2181 sector_t chunk_block)
2185 valid = dmz_test_block(zmd, zone, chunk_block);
2189 /* The block is valid: get the number of valid blocks from block */
2190 return dmz_to_next_set_block(zmd, zone, chunk_block,
2191 zmd->dev->zone_nr_blocks - chunk_block, 0);
2195 * Find the first valid block from @chunk_block in @zone.
2196 * If such a block is found, its number is returned using
2197 * @chunk_block and the total number of valid blocks from @chunk_block
2200 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2201 sector_t *chunk_block)
2203 sector_t start_block = *chunk_block;
2206 ret = dmz_to_next_set_block(zmd, zone, start_block,
2207 zmd->dev->zone_nr_blocks - start_block, 1);
2212 *chunk_block = start_block;
2214 return dmz_to_next_set_block(zmd, zone, start_block,
2215 zmd->dev->zone_nr_blocks - start_block, 0);
2219 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2221 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2223 unsigned long *addr;
2224 int end = bit + nr_bits;
2228 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2229 ((end - bit) >= BITS_PER_LONG)) {
2230 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2231 if (*addr == ULONG_MAX) {
2233 bit += BITS_PER_LONG;
2238 if (test_bit(bit, bitmap))
2247 * Get a zone weight.
2249 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2251 struct dmz_mblock *mblk;
2252 sector_t chunk_block = 0;
2253 unsigned int bit, nr_bits;
2254 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2259 /* Get bitmap block */
2260 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2266 /* Count bits in this block */
2267 bitmap = mblk->data;
2268 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2269 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2270 n += dmz_count_bits(bitmap, bit, nr_bits);
2272 dmz_release_mblock(zmd, mblk);
2274 nr_blocks -= nr_bits;
2275 chunk_block += nr_bits;
2282 * Cleanup the zoned metadata resources.
2284 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2286 struct rb_root *root;
2287 struct dmz_mblock *mblk, *next;
2290 /* Release zone mapping resources */
2291 if (zmd->map_mblk) {
2292 for (i = 0; i < zmd->nr_map_blocks; i++)
2293 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2294 kfree(zmd->map_mblk);
2295 zmd->map_mblk = NULL;
2298 /* Release super blocks */
2299 for (i = 0; i < 2; i++) {
2300 if (zmd->sb[i].mblk) {
2301 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2302 zmd->sb[i].mblk = NULL;
2306 /* Free cached blocks */
2307 while (!list_empty(&zmd->mblk_dirty_list)) {
2308 mblk = list_first_entry(&zmd->mblk_dirty_list,
2309 struct dmz_mblock, link);
2310 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2311 (u64)mblk->no, atomic_read(&mblk->ref));
2312 list_del_init(&mblk->link);
2313 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2314 dmz_free_mblock(zmd, mblk);
2317 while (!list_empty(&zmd->mblk_lru_list)) {
2318 mblk = list_first_entry(&zmd->mblk_lru_list,
2319 struct dmz_mblock, link);
2320 list_del_init(&mblk->link);
2321 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2322 dmz_free_mblock(zmd, mblk);
2325 /* Sanity checks: the mblock rbtree should now be empty */
2326 root = &zmd->mblk_rbtree;
2327 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2328 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2329 (u64)mblk->no, atomic_read(&mblk->ref));
2330 atomic_set(&mblk->ref, 0);
2331 dmz_free_mblock(zmd, mblk);
2334 /* Free the zone descriptors */
2335 dmz_drop_zones(zmd);
2339 * Initialize the zoned metadata.
2341 int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2343 struct dmz_metadata *zmd;
2344 unsigned int i, zid;
2345 struct dm_zone *zone;
2348 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2353 zmd->mblk_rbtree = RB_ROOT;
2354 init_rwsem(&zmd->mblk_sem);
2355 mutex_init(&zmd->mblk_flush_lock);
2356 spin_lock_init(&zmd->mblk_lock);
2357 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2358 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2360 mutex_init(&zmd->map_lock);
2361 atomic_set(&zmd->unmap_nr_rnd, 0);
2362 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2363 INIT_LIST_HEAD(&zmd->map_rnd_list);
2365 atomic_set(&zmd->unmap_nr_seq, 0);
2366 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2367 INIT_LIST_HEAD(&zmd->map_seq_list);
2369 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2370 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2372 init_waitqueue_head(&zmd->free_wq);
2374 /* Initialize zone descriptors */
2375 ret = dmz_init_zones(zmd);
2379 /* Get super block */
2380 ret = dmz_load_sb(zmd);
2384 /* Set metadata zones starting from sb_zone */
2385 zid = dmz_id(zmd, zmd->sb_zone);
2386 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2387 zone = dmz_get(zmd, zid + i);
2388 if (!dmz_is_rnd(zone))
2390 set_bit(DMZ_META, &zone->flags);
2393 /* Load mapping table */
2394 ret = dmz_load_mapping(zmd);
2399 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2400 * blocks and enough blocks to be able to cache the bitmap blocks of
2401 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2402 * the cache to add 512 more metadata blocks.
2404 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2405 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2406 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2407 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2408 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2410 /* Metadata cache shrinker */
2411 ret = register_shrinker(&zmd->mblk_shrinker);
2413 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2417 dmz_dev_info(dev, "Host-%s zoned block device",
2418 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2419 "aware" : "managed");
2420 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2421 (u64)dev->capacity);
2422 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2423 dev->nr_zones, (u64)dev->zone_nr_sectors);
2424 dmz_dev_info(dev, " %u metadata zones",
2425 zmd->nr_meta_zones * 2);
2426 dmz_dev_info(dev, " %u data zones for %u chunks",
2427 zmd->nr_data_zones, zmd->nr_chunks);
2428 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2429 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2430 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2431 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2432 dmz_dev_info(dev, " %u reserved sequential data zones",
2433 zmd->nr_reserved_seq);
2435 dmz_dev_debug(dev, "Format:");
2436 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2437 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2438 dmz_dev_debug(dev, " %u data zone mapping blocks",
2439 zmd->nr_map_blocks);
2440 dmz_dev_debug(dev, " %u bitmap blocks",
2441 zmd->nr_bitmap_blocks);
2447 dmz_cleanup_metadata(zmd);
2455 * Cleanup the zoned metadata resources.
2457 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2459 unregister_shrinker(&zmd->mblk_shrinker);
2460 dmz_cleanup_metadata(zmd);
2465 * Check zone information on resume.
2467 int dmz_resume_metadata(struct dmz_metadata *zmd)
2469 struct dmz_dev *dev = zmd->dev;
2470 struct dm_zone *zone;
2476 for (i = 0; i < dev->nr_zones; i++) {
2477 zone = dmz_get(zmd, i);
2479 dmz_dev_err(dev, "Unable to get zone %u", i);
2483 wp_block = zone->wp_block;
2485 ret = dmz_update_zone(zmd, zone);
2487 dmz_dev_err(dev, "Broken zone %u", i);
2491 if (dmz_is_offline(zone)) {
2492 dmz_dev_warn(dev, "Zone %u is offline", i);
2496 /* Check write pointer */
2497 if (!dmz_is_seq(zone))
2499 else if (zone->wp_block != wp_block) {
2500 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2501 i, (u64)zone->wp_block, (u64)wp_block);
2502 zone->wp_block = wp_block;
2503 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2504 dev->zone_nr_blocks - zone->wp_block);