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;
135 unsigned int zone_bits_per_mblk;
137 unsigned int nr_bitmap_blocks;
138 unsigned int nr_map_blocks;
140 unsigned int nr_useable_zones;
141 unsigned int nr_meta_blocks;
142 unsigned int nr_meta_zones;
143 unsigned int nr_data_zones;
144 unsigned int nr_rnd_zones;
145 unsigned int nr_reserved_seq;
146 unsigned int nr_chunks;
148 /* Zone information array */
149 struct dm_zone *zones;
151 struct dm_zone *sb_zone;
153 unsigned int mblk_primary;
155 unsigned int min_nr_mblks;
156 unsigned int max_nr_mblks;
158 struct rw_semaphore mblk_sem;
159 struct mutex mblk_flush_lock;
160 spinlock_t mblk_lock;
161 struct rb_root mblk_rbtree;
162 struct list_head mblk_lru_list;
163 struct list_head mblk_dirty_list;
164 struct shrinker mblk_shrinker;
166 /* Zone allocation management */
167 struct mutex map_lock;
168 struct dmz_mblock **map_mblk;
170 atomic_t unmap_nr_rnd;
171 struct list_head unmap_rnd_list;
172 struct list_head map_rnd_list;
175 atomic_t unmap_nr_seq;
176 struct list_head unmap_seq_list;
177 struct list_head map_seq_list;
179 atomic_t nr_reserved_seq_zones;
180 struct list_head reserved_seq_zones_list;
182 wait_queue_head_t free_wq;
188 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
190 return ((unsigned int)(zone - zmd->zones));
193 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
195 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
198 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
200 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
203 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
205 return zmd->nr_chunks;
208 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
213 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
215 return atomic_read(&zmd->unmap_nr_rnd);
219 * Lock/unlock mapping table.
220 * The map lock also protects all the zone lists.
222 void dmz_lock_map(struct dmz_metadata *zmd)
224 mutex_lock(&zmd->map_lock);
227 void dmz_unlock_map(struct dmz_metadata *zmd)
229 mutex_unlock(&zmd->map_lock);
233 * Lock/unlock metadata access. This is a "read" lock on a semaphore
234 * that prevents metadata flush from running while metadata are being
235 * modified. The actual metadata write mutual exclusion is achieved with
236 * the map lock and zone styate management (active and reclaim state are
237 * mutually exclusive).
239 void dmz_lock_metadata(struct dmz_metadata *zmd)
241 down_read(&zmd->mblk_sem);
244 void dmz_unlock_metadata(struct dmz_metadata *zmd)
246 up_read(&zmd->mblk_sem);
250 * Lock/unlock flush: prevent concurrent executions
251 * of dmz_flush_metadata as well as metadata modification in reclaim
252 * while flush is being executed.
254 void dmz_lock_flush(struct dmz_metadata *zmd)
256 mutex_lock(&zmd->mblk_flush_lock);
259 void dmz_unlock_flush(struct dmz_metadata *zmd)
261 mutex_unlock(&zmd->mblk_flush_lock);
265 * Allocate a metadata block.
267 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
270 struct dmz_mblock *mblk = NULL;
272 /* See if we can reuse cached blocks */
273 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
274 spin_lock(&zmd->mblk_lock);
275 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
276 struct dmz_mblock, link);
278 list_del_init(&mblk->link);
279 rb_erase(&mblk->node, &zmd->mblk_rbtree);
282 spin_unlock(&zmd->mblk_lock);
287 /* Allocate a new block */
288 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
292 mblk->page = alloc_page(GFP_NOIO);
298 RB_CLEAR_NODE(&mblk->node);
299 INIT_LIST_HEAD(&mblk->link);
303 mblk->data = page_address(mblk->page);
305 atomic_inc(&zmd->nr_mblks);
311 * Free a metadata block.
313 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
315 __free_pages(mblk->page, 0);
318 atomic_dec(&zmd->nr_mblks);
322 * Insert a metadata block in the rbtree.
324 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
326 struct rb_root *root = &zmd->mblk_rbtree;
327 struct rb_node **new = &(root->rb_node), *parent = NULL;
328 struct dmz_mblock *b;
330 /* Figure out where to put the new node */
332 b = container_of(*new, struct dmz_mblock, node);
334 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
337 /* Add new node and rebalance tree */
338 rb_link_node(&mblk->node, parent, new);
339 rb_insert_color(&mblk->node, root);
343 * Lookup a metadata block in the rbtree. If the block is found, increment
344 * its reference count.
346 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
349 struct rb_root *root = &zmd->mblk_rbtree;
350 struct rb_node *node = root->rb_node;
351 struct dmz_mblock *mblk;
354 mblk = container_of(node, struct dmz_mblock, node);
355 if (mblk->no == mblk_no) {
357 * If this is the first reference to the block,
358 * remove it from the LRU list.
361 if (mblk->ref == 1 &&
362 !test_bit(DMZ_META_DIRTY, &mblk->state))
363 list_del_init(&mblk->link);
366 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
373 * Metadata block BIO end callback.
375 static void dmz_mblock_bio_end_io(struct bio *bio)
377 struct dmz_mblock *mblk = bio->bi_private;
381 set_bit(DMZ_META_ERROR, &mblk->state);
383 if (bio_op(bio) == REQ_OP_WRITE)
384 flag = DMZ_META_WRITING;
386 flag = DMZ_META_READING;
388 clear_bit_unlock(flag, &mblk->state);
389 smp_mb__after_atomic();
390 wake_up_bit(&mblk->state, flag);
396 * Read an uncached metadata block from disk and add it to the cache.
398 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
401 struct dmz_mblock *mblk, *m;
402 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
405 if (dmz_bdev_is_dying(zmd->dev))
406 return ERR_PTR(-EIO);
408 /* Get a new block and a BIO to read it */
409 mblk = dmz_alloc_mblock(zmd, mblk_no);
411 return ERR_PTR(-ENOMEM);
413 bio = bio_alloc(GFP_NOIO, 1);
415 dmz_free_mblock(zmd, mblk);
416 return ERR_PTR(-ENOMEM);
419 spin_lock(&zmd->mblk_lock);
422 * Make sure that another context did not start reading
425 m = dmz_get_mblock_fast(zmd, mblk_no);
427 spin_unlock(&zmd->mblk_lock);
428 dmz_free_mblock(zmd, mblk);
434 set_bit(DMZ_META_READING, &mblk->state);
435 dmz_insert_mblock(zmd, mblk);
437 spin_unlock(&zmd->mblk_lock);
439 /* Submit read BIO */
440 bio->bi_iter.bi_sector = dmz_blk2sect(block);
441 bio_set_dev(bio, zmd->dev->bdev);
442 bio->bi_private = mblk;
443 bio->bi_end_io = dmz_mblock_bio_end_io;
444 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
445 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
452 * Free metadata blocks.
454 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
457 struct dmz_mblock *mblk;
458 unsigned long count = 0;
460 if (!zmd->max_nr_mblks)
463 while (!list_empty(&zmd->mblk_lru_list) &&
464 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
466 mblk = list_first_entry(&zmd->mblk_lru_list,
467 struct dmz_mblock, link);
468 list_del_init(&mblk->link);
469 rb_erase(&mblk->node, &zmd->mblk_rbtree);
470 dmz_free_mblock(zmd, mblk);
478 * For mblock shrinker: get the number of unused metadata blocks in the cache.
480 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
481 struct shrink_control *sc)
483 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
485 return atomic_read(&zmd->nr_mblks);
489 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
491 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
492 struct shrink_control *sc)
494 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
497 spin_lock(&zmd->mblk_lock);
498 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
499 spin_unlock(&zmd->mblk_lock);
501 return count ? count : SHRINK_STOP;
505 * Release a metadata block.
507 static void dmz_release_mblock(struct dmz_metadata *zmd,
508 struct dmz_mblock *mblk)
514 spin_lock(&zmd->mblk_lock);
517 if (mblk->ref == 0) {
518 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
519 rb_erase(&mblk->node, &zmd->mblk_rbtree);
520 dmz_free_mblock(zmd, mblk);
521 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
522 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
523 dmz_shrink_mblock_cache(zmd, 1);
527 spin_unlock(&zmd->mblk_lock);
531 * Get a metadata block from the rbtree. If the block
532 * is not present, read it from disk.
534 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
537 struct dmz_mblock *mblk;
540 spin_lock(&zmd->mblk_lock);
541 mblk = dmz_get_mblock_fast(zmd, mblk_no);
542 spin_unlock(&zmd->mblk_lock);
545 /* Cache miss: read the block from disk */
546 mblk = dmz_get_mblock_slow(zmd, mblk_no);
551 /* Wait for on-going read I/O and check for error */
552 wait_on_bit_io(&mblk->state, DMZ_META_READING,
553 TASK_UNINTERRUPTIBLE);
554 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
555 dmz_release_mblock(zmd, mblk);
556 dmz_check_bdev(zmd->dev);
557 return ERR_PTR(-EIO);
564 * Mark a metadata block dirty.
566 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
568 spin_lock(&zmd->mblk_lock);
569 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
570 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
571 spin_unlock(&zmd->mblk_lock);
575 * Issue a metadata block write BIO.
577 static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
580 sector_t block = zmd->sb[set].block + mblk->no;
583 if (dmz_bdev_is_dying(zmd->dev))
586 bio = bio_alloc(GFP_NOIO, 1);
588 set_bit(DMZ_META_ERROR, &mblk->state);
592 set_bit(DMZ_META_WRITING, &mblk->state);
594 bio->bi_iter.bi_sector = dmz_blk2sect(block);
595 bio_set_dev(bio, zmd->dev->bdev);
596 bio->bi_private = mblk;
597 bio->bi_end_io = dmz_mblock_bio_end_io;
598 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
599 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
606 * Read/write a metadata block.
608 static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
614 if (dmz_bdev_is_dying(zmd->dev))
617 bio = bio_alloc(GFP_NOIO, 1);
621 bio->bi_iter.bi_sector = dmz_blk2sect(block);
622 bio_set_dev(bio, zmd->dev->bdev);
623 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
624 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
625 ret = submit_bio_wait(bio);
629 dmz_check_bdev(zmd->dev);
634 * Write super block of the specified metadata set.
636 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
638 sector_t block = zmd->sb[set].block;
639 struct dmz_mblock *mblk = zmd->sb[set].mblk;
640 struct dmz_super *sb = zmd->sb[set].sb;
641 u64 sb_gen = zmd->sb_gen + 1;
644 sb->magic = cpu_to_le32(DMZ_MAGIC);
645 sb->version = cpu_to_le32(DMZ_META_VER);
647 sb->gen = cpu_to_le64(sb_gen);
649 sb->sb_block = cpu_to_le64(block);
650 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
651 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
652 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
654 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
655 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
658 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
660 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
662 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
668 * Write dirty metadata blocks to the specified set.
670 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
671 struct list_head *write_list,
674 struct dmz_mblock *mblk;
675 struct blk_plug plug;
676 int ret = 0, nr_mblks_submitted = 0;
679 blk_start_plug(&plug);
680 list_for_each_entry(mblk, write_list, link) {
681 ret = dmz_write_mblock(zmd, mblk, set);
684 nr_mblks_submitted++;
686 blk_finish_plug(&plug);
688 /* Wait for completion */
689 list_for_each_entry(mblk, write_list, link) {
690 if (!nr_mblks_submitted)
692 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
693 TASK_UNINTERRUPTIBLE);
694 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
695 clear_bit(DMZ_META_ERROR, &mblk->state);
696 dmz_check_bdev(zmd->dev);
699 nr_mblks_submitted--;
702 /* Flush drive cache (this will also sync data) */
704 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
710 * Log dirty metadata blocks.
712 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
713 struct list_head *write_list)
715 unsigned int log_set = zmd->mblk_primary ^ 0x1;
718 /* Write dirty blocks to the log */
719 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
724 * No error so far: now validate the log by updating the
725 * log index super block generation.
727 ret = dmz_write_sb(zmd, log_set);
735 * Flush dirty metadata blocks.
737 int dmz_flush_metadata(struct dmz_metadata *zmd)
739 struct dmz_mblock *mblk;
740 struct list_head write_list;
746 INIT_LIST_HEAD(&write_list);
749 * Make sure that metadata blocks are stable before logging: take
750 * the write lock on the metadata semaphore to prevent target BIOs
751 * from modifying metadata.
753 down_write(&zmd->mblk_sem);
756 * This is called from the target flush work and reclaim work.
757 * Concurrent execution is not allowed.
761 if (dmz_bdev_is_dying(zmd->dev)) {
766 /* Get dirty blocks */
767 spin_lock(&zmd->mblk_lock);
768 list_splice_init(&zmd->mblk_dirty_list, &write_list);
769 spin_unlock(&zmd->mblk_lock);
771 /* If there are no dirty metadata blocks, just flush the device cache */
772 if (list_empty(&write_list)) {
773 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
778 * The primary metadata set is still clean. Keep it this way until
779 * all updates are successful in the secondary set. That is, use
780 * the secondary set as a log.
782 ret = dmz_log_dirty_mblocks(zmd, &write_list);
787 * The log is on disk. It is now safe to update in place
788 * in the primary metadata set.
790 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
794 ret = dmz_write_sb(zmd, zmd->mblk_primary);
798 while (!list_empty(&write_list)) {
799 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
800 list_del_init(&mblk->link);
802 spin_lock(&zmd->mblk_lock);
803 clear_bit(DMZ_META_DIRTY, &mblk->state);
805 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
806 spin_unlock(&zmd->mblk_lock);
811 dmz_unlock_flush(zmd);
812 up_write(&zmd->mblk_sem);
817 if (!list_empty(&write_list)) {
818 spin_lock(&zmd->mblk_lock);
819 list_splice(&write_list, &zmd->mblk_dirty_list);
820 spin_unlock(&zmd->mblk_lock);
822 if (!dmz_check_bdev(zmd->dev))
830 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
832 unsigned int nr_meta_zones, nr_data_zones;
833 struct dmz_dev *dev = zmd->dev;
837 gen = le64_to_cpu(sb->gen);
838 stored_crc = le32_to_cpu(sb->crc);
840 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
841 if (crc != stored_crc) {
842 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
847 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
848 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
849 DMZ_MAGIC, le32_to_cpu(sb->magic));
853 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
854 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
855 DMZ_META_VER, le32_to_cpu(sb->version));
859 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
860 >> dev->zone_nr_blocks_shift;
861 if (!nr_meta_zones ||
862 nr_meta_zones >= zmd->nr_rnd_zones) {
863 dmz_dev_err(dev, "Invalid number of metadata blocks");
867 if (!le32_to_cpu(sb->nr_reserved_seq) ||
868 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
869 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
873 nr_data_zones = zmd->nr_useable_zones -
874 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
875 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
876 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
877 le32_to_cpu(sb->nr_chunks), nr_data_zones);
882 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
883 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
884 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
885 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
886 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
887 zmd->nr_meta_zones = nr_meta_zones;
888 zmd->nr_data_zones = nr_data_zones;
894 * Read the first or second super block from disk.
896 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
898 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
899 zmd->sb[set].mblk->page);
903 * Determine the position of the secondary super blocks on disk.
904 * This is used only if a corruption of the primary super block
907 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
909 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
910 struct dmz_mblock *mblk;
913 /* Allocate a block */
914 mblk = dmz_alloc_mblock(zmd, 0);
918 zmd->sb[1].mblk = mblk;
919 zmd->sb[1].sb = mblk->data;
921 /* Bad first super block: search for the second one */
922 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
923 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
924 if (dmz_read_sb(zmd, 1) != 0)
926 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
928 zmd->sb[1].block += zone_nr_blocks;
931 dmz_free_mblock(zmd, mblk);
932 zmd->sb[1].mblk = NULL;
938 * Read the first or second super block from disk.
940 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
942 struct dmz_mblock *mblk;
945 /* Allocate a block */
946 mblk = dmz_alloc_mblock(zmd, 0);
950 zmd->sb[set].mblk = mblk;
951 zmd->sb[set].sb = mblk->data;
953 /* Read super block */
954 ret = dmz_read_sb(zmd, set);
956 dmz_free_mblock(zmd, mblk);
957 zmd->sb[set].mblk = NULL;
965 * Recover a metadata set.
967 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
969 unsigned int src_set = dst_set ^ 0x1;
973 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
976 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
978 zmd->sb[1].block = zmd->sb[0].block +
979 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
982 page = alloc_page(GFP_NOIO);
986 /* Copy metadata blocks */
987 for (i = 1; i < zmd->nr_meta_blocks; i++) {
988 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
989 zmd->sb[src_set].block + i, page);
992 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
993 zmd->sb[dst_set].block + i, page);
998 /* Finalize with the super block */
999 if (!zmd->sb[dst_set].mblk) {
1000 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1001 if (!zmd->sb[dst_set].mblk) {
1005 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1008 ret = dmz_write_sb(zmd, dst_set);
1010 __free_pages(page, 0);
1016 * Get super block from disk.
1018 static int dmz_load_sb(struct dmz_metadata *zmd)
1020 bool sb_good[2] = {false, false};
1021 u64 sb_gen[2] = {0, 0};
1024 /* Read and check the primary super block */
1025 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1026 ret = dmz_get_sb(zmd, 0);
1028 dmz_dev_err(zmd->dev, "Read primary super block failed");
1032 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1034 /* Read and check secondary super block */
1037 zmd->sb[1].block = zmd->sb[0].block +
1038 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1039 ret = dmz_get_sb(zmd, 1);
1041 ret = dmz_lookup_secondary_sb(zmd);
1044 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1048 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1052 /* Use highest generation sb first */
1053 if (!sb_good[0] && !sb_good[1]) {
1054 dmz_dev_err(zmd->dev, "No valid super block found");
1059 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1061 ret = dmz_recover_mblocks(zmd, 0);
1064 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1066 ret = dmz_recover_mblocks(zmd, 1);
1069 dmz_dev_err(zmd->dev, "Recovery failed");
1073 if (sb_gen[0] >= sb_gen[1]) {
1074 zmd->sb_gen = sb_gen[0];
1075 zmd->mblk_primary = 0;
1077 zmd->sb_gen = sb_gen[1];
1078 zmd->mblk_primary = 1;
1081 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1082 zmd->mblk_primary, zmd->sb_gen);
1088 * Initialize a zone descriptor.
1090 static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1091 struct blk_zone *blkz)
1093 struct dmz_dev *dev = zmd->dev;
1095 /* Ignore the eventual last runt (smaller) zone */
1096 if (blkz->len != dev->zone_nr_sectors) {
1097 if (blkz->start + blkz->len == dev->capacity)
1102 INIT_LIST_HEAD(&zone->link);
1103 atomic_set(&zone->refcount, 0);
1104 zone->chunk = DMZ_MAP_UNMAPPED;
1106 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1107 set_bit(DMZ_RND, &zone->flags);
1108 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1109 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1110 set_bit(DMZ_SEQ, &zone->flags);
1114 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1115 set_bit(DMZ_OFFLINE, &zone->flags);
1116 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1117 set_bit(DMZ_READ_ONLY, &zone->flags);
1119 if (dmz_is_rnd(zone))
1122 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1124 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1125 zmd->nr_useable_zones++;
1126 if (dmz_is_rnd(zone)) {
1127 zmd->nr_rnd_zones++;
1128 if (!zmd->sb_zone) {
1129 /* Super block zone */
1130 zmd->sb_zone = zone;
1139 * Free zones descriptors.
1141 static void dmz_drop_zones(struct dmz_metadata *zmd)
1148 * The size of a zone report in number of zones.
1149 * This results in 4096*64B=256KB report zones commands.
1151 #define DMZ_REPORT_NR_ZONES 4096
1154 * Allocate and initialize zone descriptors using the zone
1155 * information from disk.
1157 static int dmz_init_zones(struct dmz_metadata *zmd)
1159 struct dmz_dev *dev = zmd->dev;
1160 struct dm_zone *zone;
1161 struct blk_zone *blkz;
1162 unsigned int nr_blkz;
1163 sector_t sector = 0;
1167 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1168 zmd->zone_nr_bitmap_blocks =
1169 max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1170 zmd->zone_bits_per_mblk = min_t(sector_t, dev->zone_nr_blocks,
1171 DMZ_BLOCK_SIZE_BITS);
1173 /* Allocate zone array */
1174 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1178 dmz_dev_info(dev, "Using %zu B for zone information",
1179 sizeof(struct dm_zone) * dev->nr_zones);
1181 /* Get zone information */
1182 nr_blkz = DMZ_REPORT_NR_ZONES;
1183 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1190 * Get zone information and initialize zone descriptors.
1191 * At the same time, determine where the super block
1192 * should be: first block of the first randomly writable
1196 while (sector < dev->capacity) {
1197 /* Get zone information */
1198 nr_blkz = DMZ_REPORT_NR_ZONES;
1199 ret = blkdev_report_zones(dev->bdev, sector, blkz,
1200 &nr_blkz, GFP_KERNEL);
1202 dmz_dev_err(dev, "Report zones failed %d", ret);
1209 /* Process report */
1210 for (i = 0; i < nr_blkz; i++) {
1211 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1214 sector += dev->zone_nr_sectors;
1219 /* The entire zone configuration of the disk should now be known */
1220 if (sector < dev->capacity) {
1221 dmz_dev_err(dev, "Failed to get correct zone information");
1227 dmz_drop_zones(zmd);
1233 * Update a zone information.
1235 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1237 unsigned int nr_blkz = 1;
1238 struct blk_zone blkz;
1241 /* Get zone information from disk */
1242 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1243 &blkz, &nr_blkz, GFP_NOIO);
1247 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1249 dmz_check_bdev(zmd->dev);
1253 clear_bit(DMZ_OFFLINE, &zone->flags);
1254 clear_bit(DMZ_READ_ONLY, &zone->flags);
1255 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1256 set_bit(DMZ_OFFLINE, &zone->flags);
1257 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1258 set_bit(DMZ_READ_ONLY, &zone->flags);
1260 if (dmz_is_seq(zone))
1261 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1269 * Check a zone write pointer position when the zone is marked
1270 * with the sequential write error flag.
1272 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1273 struct dm_zone *zone)
1275 unsigned int wp = 0;
1278 wp = zone->wp_block;
1279 ret = dmz_update_zone(zmd, zone);
1283 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1284 dmz_id(zmd, zone), zone->wp_block, wp);
1286 if (zone->wp_block < wp) {
1287 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1288 wp - zone->wp_block);
1294 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1296 return &zmd->zones[zone_id];
1300 * Reset a zone write pointer.
1302 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1307 * Ignore offline zones, read only zones,
1308 * and conventional zones.
1310 if (dmz_is_offline(zone) ||
1311 dmz_is_readonly(zone) ||
1315 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1316 struct dmz_dev *dev = zmd->dev;
1318 ret = blkdev_reset_zones(dev->bdev,
1319 dmz_start_sect(zmd, zone),
1320 dev->zone_nr_sectors, GFP_NOIO);
1322 dmz_dev_err(dev, "Reset zone %u failed %d",
1323 dmz_id(zmd, zone), ret);
1328 /* Clear write error bit and rewind write pointer position */
1329 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1335 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1338 * Initialize chunk mapping.
1340 static int dmz_load_mapping(struct dmz_metadata *zmd)
1342 struct dmz_dev *dev = zmd->dev;
1343 struct dm_zone *dzone, *bzone;
1344 struct dmz_mblock *dmap_mblk = NULL;
1345 struct dmz_map *dmap;
1346 unsigned int i = 0, e = 0, chunk = 0;
1347 unsigned int dzone_id;
1348 unsigned int bzone_id;
1350 /* Metadata block array for the chunk mapping table */
1351 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1352 sizeof(struct dmz_mblk *), GFP_KERNEL);
1356 /* Get chunk mapping table blocks and initialize zone mapping */
1357 while (chunk < zmd->nr_chunks) {
1359 /* Get mapping block */
1360 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1361 if (IS_ERR(dmap_mblk))
1362 return PTR_ERR(dmap_mblk);
1363 zmd->map_mblk[i] = dmap_mblk;
1364 dmap = (struct dmz_map *) dmap_mblk->data;
1369 /* Check data zone */
1370 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1371 if (dzone_id == DMZ_MAP_UNMAPPED)
1374 if (dzone_id >= dev->nr_zones) {
1375 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1380 dzone = dmz_get(zmd, dzone_id);
1381 set_bit(DMZ_DATA, &dzone->flags);
1382 dzone->chunk = chunk;
1383 dmz_get_zone_weight(zmd, dzone);
1385 if (dmz_is_rnd(dzone))
1386 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1388 list_add_tail(&dzone->link, &zmd->map_seq_list);
1390 /* Check buffer zone */
1391 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1392 if (bzone_id == DMZ_MAP_UNMAPPED)
1395 if (bzone_id >= dev->nr_zones) {
1396 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1401 bzone = dmz_get(zmd, bzone_id);
1402 if (!dmz_is_rnd(bzone)) {
1403 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1408 set_bit(DMZ_DATA, &bzone->flags);
1409 set_bit(DMZ_BUF, &bzone->flags);
1410 bzone->chunk = chunk;
1411 bzone->bzone = dzone;
1412 dzone->bzone = bzone;
1413 dmz_get_zone_weight(zmd, bzone);
1414 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1418 if (e >= DMZ_MAP_ENTRIES)
1423 * At this point, only meta zones and mapped data zones were
1424 * fully initialized. All remaining zones are unmapped data
1425 * zones. Finish initializing those here.
1427 for (i = 0; i < dev->nr_zones; i++) {
1428 dzone = dmz_get(zmd, i);
1429 if (dmz_is_meta(dzone))
1432 if (dmz_is_rnd(dzone))
1437 if (dmz_is_data(dzone)) {
1438 /* Already initialized */
1442 /* Unmapped data zone */
1443 set_bit(DMZ_DATA, &dzone->flags);
1444 dzone->chunk = DMZ_MAP_UNMAPPED;
1445 if (dmz_is_rnd(dzone)) {
1446 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1447 atomic_inc(&zmd->unmap_nr_rnd);
1448 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1449 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1450 atomic_inc(&zmd->nr_reserved_seq_zones);
1453 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1454 atomic_inc(&zmd->unmap_nr_seq);
1462 * Set a data chunk mapping.
1464 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1465 unsigned int dzone_id, unsigned int bzone_id)
1467 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1468 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1469 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1471 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1472 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1473 dmz_dirty_mblock(zmd, dmap_mblk);
1477 * The list of mapped zones is maintained in LRU order.
1478 * This rotates a zone at the end of its map list.
1480 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1482 if (list_empty(&zone->link))
1485 list_del_init(&zone->link);
1486 if (dmz_is_seq(zone)) {
1487 /* LRU rotate sequential zone */
1488 list_add_tail(&zone->link, &zmd->map_seq_list);
1490 /* LRU rotate random zone */
1491 list_add_tail(&zone->link, &zmd->map_rnd_list);
1496 * The list of mapped random zones is maintained
1497 * in LRU order. This rotates a zone at the end of the list.
1499 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1501 __dmz_lru_zone(zmd, zone);
1503 __dmz_lru_zone(zmd, zone->bzone);
1507 * Wait for any zone to be freed.
1509 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1513 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1514 dmz_unlock_map(zmd);
1515 dmz_unlock_metadata(zmd);
1517 io_schedule_timeout(HZ);
1519 dmz_lock_metadata(zmd);
1521 finish_wait(&zmd->free_wq, &wait);
1525 * Lock a zone for reclaim (set the zone RECLAIM bit).
1526 * Returns false if the zone cannot be locked or if it is already locked
1529 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1531 /* Active zones cannot be reclaimed */
1532 if (dmz_is_active(zone))
1535 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1539 * Clear a zone reclaim flag.
1541 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1543 WARN_ON(dmz_is_active(zone));
1544 WARN_ON(!dmz_in_reclaim(zone));
1546 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1547 smp_mb__after_atomic();
1548 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1552 * Wait for a zone reclaim to complete.
1554 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1556 dmz_unlock_map(zmd);
1557 dmz_unlock_metadata(zmd);
1558 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1559 dmz_lock_metadata(zmd);
1564 * Select a random write zone for reclaim.
1566 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1568 struct dm_zone *dzone = NULL;
1569 struct dm_zone *zone;
1571 if (list_empty(&zmd->map_rnd_list))
1572 return ERR_PTR(-EBUSY);
1574 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1575 if (dmz_is_buf(zone))
1576 dzone = zone->bzone;
1579 if (dmz_lock_zone_reclaim(dzone))
1587 * Select a buffered sequential zone for reclaim.
1589 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1591 struct dm_zone *zone;
1593 if (list_empty(&zmd->map_seq_list))
1594 return ERR_PTR(-EBUSY);
1596 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1599 if (dmz_lock_zone_reclaim(zone))
1607 * Select a zone for reclaim.
1609 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1611 struct dm_zone *zone;
1614 * Search for a zone candidate to reclaim: 2 cases are possible.
1615 * (1) There is no free sequential zones. Then a random data zone
1616 * cannot be reclaimed. So choose a sequential zone to reclaim so
1617 * that afterward a random zone can be reclaimed.
1618 * (2) At least one free sequential zone is available, then choose
1619 * the oldest random zone (data or buffer) that can be locked.
1622 if (list_empty(&zmd->reserved_seq_zones_list))
1623 zone = dmz_get_seq_zone_for_reclaim(zmd);
1625 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1626 dmz_unlock_map(zmd);
1632 * Get the zone mapping a chunk, if the chunk is mapped already.
1633 * If no mapping exist and the operation is WRITE, a zone is
1634 * allocated and used to map the chunk.
1635 * The zone returned will be set to the active state.
1637 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1639 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1640 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1641 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1642 unsigned int dzone_id;
1643 struct dm_zone *dzone = NULL;
1648 /* Get the chunk mapping */
1649 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1650 if (dzone_id == DMZ_MAP_UNMAPPED) {
1652 * Read or discard in unmapped chunks are fine. But for
1653 * writes, we need a mapping, so get one.
1655 if (op != REQ_OP_WRITE)
1658 /* Alloate a random zone */
1659 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1661 if (dmz_bdev_is_dying(zmd->dev)) {
1662 dzone = ERR_PTR(-EIO);
1665 dmz_wait_for_free_zones(zmd);
1669 dmz_map_zone(zmd, dzone, chunk);
1672 /* The chunk is already mapped: get the mapping zone */
1673 dzone = dmz_get(zmd, dzone_id);
1674 if (dzone->chunk != chunk) {
1675 dzone = ERR_PTR(-EIO);
1679 /* Repair write pointer if the sequential dzone has error */
1680 if (dmz_seq_write_err(dzone)) {
1681 ret = dmz_handle_seq_write_err(zmd, dzone);
1683 dzone = ERR_PTR(-EIO);
1686 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1691 * If the zone is being reclaimed, the chunk mapping may change
1692 * to a different zone. So wait for reclaim and retry. Otherwise,
1693 * activate the zone (this will prevent reclaim from touching it).
1695 if (dmz_in_reclaim(dzone)) {
1696 dmz_wait_for_reclaim(zmd, dzone);
1699 dmz_activate_zone(dzone);
1700 dmz_lru_zone(zmd, dzone);
1702 dmz_unlock_map(zmd);
1708 * Write and discard change the block validity of data zones and their buffer
1709 * zones. Check here that valid blocks are still present. If all blocks are
1710 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1713 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1715 struct dm_zone *bzone;
1719 bzone = dzone->bzone;
1721 if (dmz_weight(bzone))
1722 dmz_lru_zone(zmd, bzone);
1724 /* Empty buffer zone: reclaim it */
1725 dmz_unmap_zone(zmd, bzone);
1726 dmz_free_zone(zmd, bzone);
1731 /* Deactivate the data zone */
1732 dmz_deactivate_zone(dzone);
1733 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1734 dmz_lru_zone(zmd, dzone);
1736 /* Unbuffered inactive empty data zone: reclaim it */
1737 dmz_unmap_zone(zmd, dzone);
1738 dmz_free_zone(zmd, dzone);
1741 dmz_unlock_map(zmd);
1745 * Allocate and map a random zone to buffer a chunk
1746 * already mapped to a sequential zone.
1748 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1749 struct dm_zone *dzone)
1751 struct dm_zone *bzone;
1755 bzone = dzone->bzone;
1759 /* Alloate a random zone */
1760 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1762 if (dmz_bdev_is_dying(zmd->dev)) {
1763 bzone = ERR_PTR(-EIO);
1766 dmz_wait_for_free_zones(zmd);
1770 /* Update the chunk mapping */
1771 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1772 dmz_id(zmd, bzone));
1774 set_bit(DMZ_BUF, &bzone->flags);
1775 bzone->chunk = dzone->chunk;
1776 bzone->bzone = dzone;
1777 dzone->bzone = bzone;
1778 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1780 dmz_unlock_map(zmd);
1786 * Get an unmapped (free) zone.
1787 * This must be called with the mapping lock held.
1789 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1791 struct list_head *list;
1792 struct dm_zone *zone;
1794 if (flags & DMZ_ALLOC_RND)
1795 list = &zmd->unmap_rnd_list;
1797 list = &zmd->unmap_seq_list;
1799 if (list_empty(list)) {
1801 * No free zone: if this is for reclaim, allow using the
1802 * reserved sequential zones.
1804 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1805 list_empty(&zmd->reserved_seq_zones_list))
1808 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1809 struct dm_zone, link);
1810 list_del_init(&zone->link);
1811 atomic_dec(&zmd->nr_reserved_seq_zones);
1815 zone = list_first_entry(list, struct dm_zone, link);
1816 list_del_init(&zone->link);
1818 if (dmz_is_rnd(zone))
1819 atomic_dec(&zmd->unmap_nr_rnd);
1821 atomic_dec(&zmd->unmap_nr_seq);
1823 if (dmz_is_offline(zone)) {
1824 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1834 * This must be called with the mapping lock held.
1836 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1838 /* If this is a sequential zone, reset it */
1839 if (dmz_is_seq(zone))
1840 dmz_reset_zone(zmd, zone);
1842 /* Return the zone to its type unmap list */
1843 if (dmz_is_rnd(zone)) {
1844 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1845 atomic_inc(&zmd->unmap_nr_rnd);
1846 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1847 zmd->nr_reserved_seq) {
1848 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1849 atomic_inc(&zmd->nr_reserved_seq_zones);
1851 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1852 atomic_inc(&zmd->unmap_nr_seq);
1855 wake_up_all(&zmd->free_wq);
1859 * Map a chunk to a zone.
1860 * This must be called with the mapping lock held.
1862 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1865 /* Set the chunk mapping */
1866 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1868 dzone->chunk = chunk;
1869 if (dmz_is_rnd(dzone))
1870 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1872 list_add_tail(&dzone->link, &zmd->map_seq_list);
1877 * This must be called with the mapping lock held.
1879 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1881 unsigned int chunk = zone->chunk;
1882 unsigned int dzone_id;
1884 if (chunk == DMZ_MAP_UNMAPPED) {
1885 /* Already unmapped */
1889 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1891 * Unmapping the chunk buffer zone: clear only
1892 * the chunk buffer mapping
1894 dzone_id = dmz_id(zmd, zone->bzone);
1895 zone->bzone->bzone = NULL;
1900 * Unmapping the chunk data zone: the zone must
1903 if (WARN_ON(zone->bzone)) {
1904 zone->bzone->bzone = NULL;
1907 dzone_id = DMZ_MAP_UNMAPPED;
1910 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1912 zone->chunk = DMZ_MAP_UNMAPPED;
1913 list_del_init(&zone->link);
1917 * Set @nr_bits bits in @bitmap starting from @bit.
1918 * Return the number of bits changed from 0 to 1.
1920 static unsigned int dmz_set_bits(unsigned long *bitmap,
1921 unsigned int bit, unsigned int nr_bits)
1923 unsigned long *addr;
1924 unsigned int end = bit + nr_bits;
1928 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1929 ((end - bit) >= BITS_PER_LONG)) {
1930 /* Try to set the whole word at once */
1931 addr = bitmap + BIT_WORD(bit);
1935 bit += BITS_PER_LONG;
1940 if (!test_and_set_bit(bit, bitmap))
1949 * Get the bitmap block storing the bit for chunk_block in zone.
1951 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1952 struct dm_zone *zone,
1953 sector_t chunk_block)
1955 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1956 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1957 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1959 return dmz_get_mblock(zmd, bitmap_block);
1963 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1965 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1966 struct dm_zone *to_zone)
1968 struct dmz_mblock *from_mblk, *to_mblk;
1969 sector_t chunk_block = 0;
1971 /* Get the zones bitmap blocks */
1972 while (chunk_block < zmd->dev->zone_nr_blocks) {
1973 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1974 if (IS_ERR(from_mblk))
1975 return PTR_ERR(from_mblk);
1976 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1977 if (IS_ERR(to_mblk)) {
1978 dmz_release_mblock(zmd, from_mblk);
1979 return PTR_ERR(to_mblk);
1982 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1983 dmz_dirty_mblock(zmd, to_mblk);
1985 dmz_release_mblock(zmd, to_mblk);
1986 dmz_release_mblock(zmd, from_mblk);
1988 chunk_block += zmd->zone_bits_per_mblk;
1991 to_zone->weight = from_zone->weight;
1997 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1998 * starting from chunk_block.
2000 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2001 struct dm_zone *to_zone, sector_t chunk_block)
2003 unsigned int nr_blocks;
2006 /* Get the zones bitmap blocks */
2007 while (chunk_block < zmd->dev->zone_nr_blocks) {
2008 /* Get a valid region from the source zone */
2009 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2014 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2018 chunk_block += nr_blocks;
2025 * Validate all the blocks in the range [block..block+nr_blocks-1].
2027 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2028 sector_t chunk_block, unsigned int nr_blocks)
2030 unsigned int count, bit, nr_bits;
2031 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2032 struct dmz_mblock *mblk;
2035 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2036 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2039 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2042 /* Get bitmap block */
2043 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2045 return PTR_ERR(mblk);
2048 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2049 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2051 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2053 dmz_dirty_mblock(zmd, mblk);
2056 dmz_release_mblock(zmd, mblk);
2058 nr_blocks -= nr_bits;
2059 chunk_block += nr_bits;
2062 if (likely(zone->weight + n <= zone_nr_blocks))
2065 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2066 dmz_id(zmd, zone), zone->weight,
2067 zone_nr_blocks - n);
2068 zone->weight = zone_nr_blocks;
2075 * Clear nr_bits bits in bitmap starting from bit.
2076 * Return the number of bits cleared.
2078 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2080 unsigned long *addr;
2081 int end = bit + nr_bits;
2085 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2086 ((end - bit) >= BITS_PER_LONG)) {
2087 /* Try to clear whole word at once */
2088 addr = bitmap + BIT_WORD(bit);
2089 if (*addr == ULONG_MAX) {
2092 bit += BITS_PER_LONG;
2097 if (test_and_clear_bit(bit, bitmap))
2106 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2108 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2109 sector_t chunk_block, unsigned int nr_blocks)
2111 unsigned int count, bit, nr_bits;
2112 struct dmz_mblock *mblk;
2115 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2116 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2118 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2121 /* Get bitmap block */
2122 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2124 return PTR_ERR(mblk);
2127 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2128 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2130 count = dmz_clear_bits((unsigned long *)mblk->data,
2133 dmz_dirty_mblock(zmd, mblk);
2136 dmz_release_mblock(zmd, mblk);
2138 nr_blocks -= nr_bits;
2139 chunk_block += nr_bits;
2142 if (zone->weight >= n)
2145 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2146 dmz_id(zmd, zone), zone->weight, n);
2154 * Get a block bit value.
2156 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2157 sector_t chunk_block)
2159 struct dmz_mblock *mblk;
2162 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2164 /* Get bitmap block */
2165 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2167 return PTR_ERR(mblk);
2170 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2171 (unsigned long *) mblk->data) != 0;
2173 dmz_release_mblock(zmd, mblk);
2179 * Return the number of blocks from chunk_block to the first block with a bit
2180 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2182 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2183 sector_t chunk_block, unsigned int nr_blocks,
2186 struct dmz_mblock *mblk;
2187 unsigned int bit, set_bit, nr_bits;
2188 unsigned int zone_bits = zmd->zone_bits_per_mblk;
2189 unsigned long *bitmap;
2192 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2195 /* Get bitmap block */
2196 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2198 return PTR_ERR(mblk);
2201 bitmap = (unsigned long *) mblk->data;
2202 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2203 nr_bits = min(nr_blocks, zone_bits - bit);
2205 set_bit = find_next_bit(bitmap, zone_bits, bit);
2207 set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2208 dmz_release_mblock(zmd, mblk);
2211 if (set_bit < zone_bits)
2214 nr_blocks -= nr_bits;
2215 chunk_block += nr_bits;
2222 * Test if chunk_block is valid. If it is, the number of consecutive
2223 * valid blocks from chunk_block will be returned.
2225 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2226 sector_t chunk_block)
2230 valid = dmz_test_block(zmd, zone, chunk_block);
2234 /* The block is valid: get the number of valid blocks from block */
2235 return dmz_to_next_set_block(zmd, zone, chunk_block,
2236 zmd->dev->zone_nr_blocks - chunk_block, 0);
2240 * Find the first valid block from @chunk_block in @zone.
2241 * If such a block is found, its number is returned using
2242 * @chunk_block and the total number of valid blocks from @chunk_block
2245 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2246 sector_t *chunk_block)
2248 sector_t start_block = *chunk_block;
2251 ret = dmz_to_next_set_block(zmd, zone, start_block,
2252 zmd->dev->zone_nr_blocks - start_block, 1);
2257 *chunk_block = start_block;
2259 return dmz_to_next_set_block(zmd, zone, start_block,
2260 zmd->dev->zone_nr_blocks - start_block, 0);
2264 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2266 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2268 unsigned long *addr;
2269 int end = bit + nr_bits;
2273 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2274 ((end - bit) >= BITS_PER_LONG)) {
2275 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2276 if (*addr == ULONG_MAX) {
2278 bit += BITS_PER_LONG;
2283 if (test_bit(bit, bitmap))
2292 * Get a zone weight.
2294 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2296 struct dmz_mblock *mblk;
2297 sector_t chunk_block = 0;
2298 unsigned int bit, nr_bits;
2299 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2304 /* Get bitmap block */
2305 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2311 /* Count bits in this block */
2312 bitmap = mblk->data;
2313 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2314 nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2315 n += dmz_count_bits(bitmap, bit, nr_bits);
2317 dmz_release_mblock(zmd, mblk);
2319 nr_blocks -= nr_bits;
2320 chunk_block += nr_bits;
2327 * Cleanup the zoned metadata resources.
2329 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2331 struct rb_root *root;
2332 struct dmz_mblock *mblk, *next;
2335 /* Release zone mapping resources */
2336 if (zmd->map_mblk) {
2337 for (i = 0; i < zmd->nr_map_blocks; i++)
2338 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2339 kfree(zmd->map_mblk);
2340 zmd->map_mblk = NULL;
2343 /* Release super blocks */
2344 for (i = 0; i < 2; i++) {
2345 if (zmd->sb[i].mblk) {
2346 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2347 zmd->sb[i].mblk = NULL;
2351 /* Free cached blocks */
2352 while (!list_empty(&zmd->mblk_dirty_list)) {
2353 mblk = list_first_entry(&zmd->mblk_dirty_list,
2354 struct dmz_mblock, link);
2355 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2356 (u64)mblk->no, mblk->ref);
2357 list_del_init(&mblk->link);
2358 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2359 dmz_free_mblock(zmd, mblk);
2362 while (!list_empty(&zmd->mblk_lru_list)) {
2363 mblk = list_first_entry(&zmd->mblk_lru_list,
2364 struct dmz_mblock, link);
2365 list_del_init(&mblk->link);
2366 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2367 dmz_free_mblock(zmd, mblk);
2370 /* Sanity checks: the mblock rbtree should now be empty */
2371 root = &zmd->mblk_rbtree;
2372 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2373 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2374 (u64)mblk->no, mblk->ref);
2376 dmz_free_mblock(zmd, mblk);
2379 /* Free the zone descriptors */
2380 dmz_drop_zones(zmd);
2382 mutex_destroy(&zmd->mblk_flush_lock);
2383 mutex_destroy(&zmd->map_lock);
2387 * Initialize the zoned metadata.
2389 int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2391 struct dmz_metadata *zmd;
2392 unsigned int i, zid;
2393 struct dm_zone *zone;
2396 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2401 zmd->mblk_rbtree = RB_ROOT;
2402 init_rwsem(&zmd->mblk_sem);
2403 mutex_init(&zmd->mblk_flush_lock);
2404 spin_lock_init(&zmd->mblk_lock);
2405 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2406 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2408 mutex_init(&zmd->map_lock);
2409 atomic_set(&zmd->unmap_nr_rnd, 0);
2410 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2411 INIT_LIST_HEAD(&zmd->map_rnd_list);
2413 atomic_set(&zmd->unmap_nr_seq, 0);
2414 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2415 INIT_LIST_HEAD(&zmd->map_seq_list);
2417 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2418 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2420 init_waitqueue_head(&zmd->free_wq);
2422 /* Initialize zone descriptors */
2423 ret = dmz_init_zones(zmd);
2427 /* Get super block */
2428 ret = dmz_load_sb(zmd);
2432 /* Set metadata zones starting from sb_zone */
2433 zid = dmz_id(zmd, zmd->sb_zone);
2434 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2435 zone = dmz_get(zmd, zid + i);
2436 if (!dmz_is_rnd(zone))
2438 set_bit(DMZ_META, &zone->flags);
2441 /* Load mapping table */
2442 ret = dmz_load_mapping(zmd);
2447 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2448 * blocks and enough blocks to be able to cache the bitmap blocks of
2449 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2450 * the cache to add 512 more metadata blocks.
2452 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2453 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2454 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2455 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2456 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2458 /* Metadata cache shrinker */
2459 ret = register_shrinker(&zmd->mblk_shrinker);
2461 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2465 dmz_dev_info(dev, "Host-%s zoned block device",
2466 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2467 "aware" : "managed");
2468 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2469 (u64)dev->capacity);
2470 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2471 dev->nr_zones, (u64)dev->zone_nr_sectors);
2472 dmz_dev_info(dev, " %u metadata zones",
2473 zmd->nr_meta_zones * 2);
2474 dmz_dev_info(dev, " %u data zones for %u chunks",
2475 zmd->nr_data_zones, zmd->nr_chunks);
2476 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2477 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2478 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2479 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2480 dmz_dev_info(dev, " %u reserved sequential data zones",
2481 zmd->nr_reserved_seq);
2483 dmz_dev_debug(dev, "Format:");
2484 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2485 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2486 dmz_dev_debug(dev, " %u data zone mapping blocks",
2487 zmd->nr_map_blocks);
2488 dmz_dev_debug(dev, " %u bitmap blocks",
2489 zmd->nr_bitmap_blocks);
2495 dmz_cleanup_metadata(zmd);
2503 * Cleanup the zoned metadata resources.
2505 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2507 unregister_shrinker(&zmd->mblk_shrinker);
2508 dmz_cleanup_metadata(zmd);
2513 * Check zone information on resume.
2515 int dmz_resume_metadata(struct dmz_metadata *zmd)
2517 struct dmz_dev *dev = zmd->dev;
2518 struct dm_zone *zone;
2524 for (i = 0; i < dev->nr_zones; i++) {
2525 zone = dmz_get(zmd, i);
2527 dmz_dev_err(dev, "Unable to get zone %u", i);
2531 wp_block = zone->wp_block;
2533 ret = dmz_update_zone(zmd, zone);
2535 dmz_dev_err(dev, "Broken zone %u", i);
2539 if (dmz_is_offline(zone)) {
2540 dmz_dev_warn(dev, "Zone %u is offline", i);
2544 /* Check write pointer */
2545 if (!dmz_is_seq(zone))
2547 else if (zone->wp_block != wp_block) {
2548 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2549 i, (u64)zone->wp_block, (u64)wp_block);
2550 zone->wp_block = wp_block;
2551 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2552 dev->zone_nr_blocks - zone->wp_block);