1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
5 * This file is released under the GPL.
10 #include <linux/module.h>
11 #include <linux/crc32.h>
12 #include <linux/sched/mm.h>
14 #define DM_MSG_PREFIX "zoned metadata"
19 #define DMZ_META_VER 1
22 * On-disk super block magic.
24 #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
25 (((unsigned int)('Z')) << 16) | \
26 (((unsigned int)('B')) << 8) | \
27 ((unsigned int)('D')))
30 * On disk super block.
31 * This uses only 512 B but uses on disk a full 4KB block. This block is
32 * followed on disk by the mapping table of chunks to zones and the bitmap
33 * blocks indicating zone block validity.
34 * The overall resulting metadata format is:
35 * (1) Super block (1 block)
36 * (2) Chunk mapping table (nr_map_blocks)
37 * (3) Bitmap blocks (nr_bitmap_blocks)
38 * All metadata blocks are stored in conventional zones, starting from
39 * the first conventional zone found on disk.
45 /* Metadata version number */
46 __le32 version; /* 8 */
48 /* Generation number */
51 /* This block number */
52 __le64 sb_block; /* 24 */
54 /* The number of metadata blocks, including this super block */
55 __le32 nr_meta_blocks; /* 28 */
57 /* The number of sequential zones reserved for reclaim */
58 __le32 nr_reserved_seq; /* 32 */
60 /* The number of entries in the mapping table */
61 __le32 nr_chunks; /* 36 */
63 /* The number of blocks used for the chunk mapping table */
64 __le32 nr_map_blocks; /* 40 */
66 /* The number of blocks used for the block bitmaps */
67 __le32 nr_bitmap_blocks; /* 44 */
72 /* Padding to full 512B sector */
73 u8 reserved[464]; /* 512 */
77 * Chunk mapping entry: entries are indexed by chunk number
78 * and give the zone ID (dzone_id) mapping the chunk on disk.
79 * This zone may be sequential or random. If it is a sequential
80 * zone, a second zone (bzone_id) used as a write buffer may
81 * also be specified. This second zone will always be a randomly
90 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
92 #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
93 #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
94 #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
95 #define DMZ_MAP_UNMAPPED UINT_MAX
98 * Meta data block descriptor (for cached metadata blocks).
102 struct list_head link;
111 * Metadata block state flags.
121 * Super block information (one per metadata set).
125 struct dmz_mblock *mblk;
126 struct dmz_super *sb;
130 * In-memory metadata.
132 struct dmz_metadata {
135 sector_t zone_bitmap_size;
136 unsigned int zone_nr_bitmap_blocks;
138 unsigned int nr_bitmap_blocks;
139 unsigned int nr_map_blocks;
141 unsigned int nr_useable_zones;
142 unsigned int nr_meta_blocks;
143 unsigned int nr_meta_zones;
144 unsigned int nr_data_zones;
145 unsigned int nr_rnd_zones;
146 unsigned int nr_reserved_seq;
147 unsigned int nr_chunks;
149 /* Zone information array */
150 struct dm_zone *zones;
152 struct dm_zone *sb_zone;
154 unsigned int mblk_primary;
156 unsigned int min_nr_mblks;
157 unsigned int max_nr_mblks;
159 struct rw_semaphore mblk_sem;
160 struct mutex mblk_flush_lock;
161 spinlock_t mblk_lock;
162 struct rb_root mblk_rbtree;
163 struct list_head mblk_lru_list;
164 struct list_head mblk_dirty_list;
165 struct shrinker mblk_shrinker;
167 /* Zone allocation management */
168 struct mutex map_lock;
169 struct dmz_mblock **map_mblk;
171 atomic_t unmap_nr_rnd;
172 struct list_head unmap_rnd_list;
173 struct list_head map_rnd_list;
176 atomic_t unmap_nr_seq;
177 struct list_head unmap_seq_list;
178 struct list_head map_seq_list;
180 atomic_t nr_reserved_seq_zones;
181 struct list_head reserved_seq_zones_list;
183 wait_queue_head_t free_wq;
189 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
191 return ((unsigned int)(zone - zmd->zones));
194 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
196 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
199 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
201 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
204 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
206 return zmd->nr_chunks;
209 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
214 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
216 return atomic_read(&zmd->unmap_nr_rnd);
220 * Lock/unlock mapping table.
221 * The map lock also protects all the zone lists.
223 void dmz_lock_map(struct dmz_metadata *zmd)
225 mutex_lock(&zmd->map_lock);
228 void dmz_unlock_map(struct dmz_metadata *zmd)
230 mutex_unlock(&zmd->map_lock);
234 * Lock/unlock metadata access. This is a "read" lock on a semaphore
235 * that prevents metadata flush from running while metadata are being
236 * modified. The actual metadata write mutual exclusion is achieved with
237 * the map lock and zone state management (active and reclaim state are
238 * mutually exclusive).
240 void dmz_lock_metadata(struct dmz_metadata *zmd)
242 down_read(&zmd->mblk_sem);
245 void dmz_unlock_metadata(struct dmz_metadata *zmd)
247 up_read(&zmd->mblk_sem);
251 * Lock/unlock flush: prevent concurrent executions
252 * of dmz_flush_metadata as well as metadata modification in reclaim
253 * while flush is being executed.
255 void dmz_lock_flush(struct dmz_metadata *zmd)
257 mutex_lock(&zmd->mblk_flush_lock);
260 void dmz_unlock_flush(struct dmz_metadata *zmd)
262 mutex_unlock(&zmd->mblk_flush_lock);
266 * Allocate a metadata block.
268 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
271 struct dmz_mblock *mblk = NULL;
273 /* See if we can reuse cached blocks */
274 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
275 spin_lock(&zmd->mblk_lock);
276 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
277 struct dmz_mblock, link);
279 list_del_init(&mblk->link);
280 rb_erase(&mblk->node, &zmd->mblk_rbtree);
283 spin_unlock(&zmd->mblk_lock);
288 /* Allocate a new block */
289 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
293 mblk->page = alloc_page(GFP_NOIO);
299 RB_CLEAR_NODE(&mblk->node);
300 INIT_LIST_HEAD(&mblk->link);
304 mblk->data = page_address(mblk->page);
306 atomic_inc(&zmd->nr_mblks);
312 * Free a metadata block.
314 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
316 __free_pages(mblk->page, 0);
319 atomic_dec(&zmd->nr_mblks);
323 * Insert a metadata block in the rbtree.
325 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
327 struct rb_root *root = &zmd->mblk_rbtree;
328 struct rb_node **new = &(root->rb_node), *parent = NULL;
329 struct dmz_mblock *b;
331 /* Figure out where to put the new node */
333 b = container_of(*new, struct dmz_mblock, node);
335 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
338 /* Add new node and rebalance tree */
339 rb_link_node(&mblk->node, parent, new);
340 rb_insert_color(&mblk->node, root);
344 * Lookup a metadata block in the rbtree. If the block is found, increment
345 * its reference count.
347 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
350 struct rb_root *root = &zmd->mblk_rbtree;
351 struct rb_node *node = root->rb_node;
352 struct dmz_mblock *mblk;
355 mblk = container_of(node, struct dmz_mblock, node);
356 if (mblk->no == mblk_no) {
358 * If this is the first reference to the block,
359 * remove it from the LRU list.
362 if (mblk->ref == 1 &&
363 !test_bit(DMZ_META_DIRTY, &mblk->state))
364 list_del_init(&mblk->link);
367 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
374 * Metadata block BIO end callback.
376 static void dmz_mblock_bio_end_io(struct bio *bio)
378 struct dmz_mblock *mblk = bio->bi_private;
382 set_bit(DMZ_META_ERROR, &mblk->state);
384 if (bio_op(bio) == REQ_OP_WRITE)
385 flag = DMZ_META_WRITING;
387 flag = DMZ_META_READING;
389 clear_bit_unlock(flag, &mblk->state);
390 smp_mb__after_atomic();
391 wake_up_bit(&mblk->state, flag);
397 * Read an uncached metadata block from disk and add it to the cache.
399 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
402 struct dmz_mblock *mblk, *m;
403 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
406 if (dmz_bdev_is_dying(zmd->dev))
407 return ERR_PTR(-EIO);
409 /* Get a new block and a BIO to read it */
410 mblk = dmz_alloc_mblock(zmd, mblk_no);
412 return ERR_PTR(-ENOMEM);
414 bio = bio_alloc(GFP_NOIO, 1);
416 dmz_free_mblock(zmd, mblk);
417 return ERR_PTR(-ENOMEM);
420 spin_lock(&zmd->mblk_lock);
423 * Make sure that another context did not start reading
426 m = dmz_get_mblock_fast(zmd, mblk_no);
428 spin_unlock(&zmd->mblk_lock);
429 dmz_free_mblock(zmd, mblk);
435 set_bit(DMZ_META_READING, &mblk->state);
436 dmz_insert_mblock(zmd, mblk);
438 spin_unlock(&zmd->mblk_lock);
440 /* Submit read BIO */
441 bio->bi_iter.bi_sector = dmz_blk2sect(block);
442 bio_set_dev(bio, zmd->dev->bdev);
443 bio->bi_private = mblk;
444 bio->bi_end_io = dmz_mblock_bio_end_io;
445 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
446 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
453 * Free metadata blocks.
455 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
458 struct dmz_mblock *mblk;
459 unsigned long count = 0;
461 if (!zmd->max_nr_mblks)
464 while (!list_empty(&zmd->mblk_lru_list) &&
465 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
467 mblk = list_first_entry(&zmd->mblk_lru_list,
468 struct dmz_mblock, link);
469 list_del_init(&mblk->link);
470 rb_erase(&mblk->node, &zmd->mblk_rbtree);
471 dmz_free_mblock(zmd, mblk);
479 * For mblock shrinker: get the number of unused metadata blocks in the cache.
481 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
482 struct shrink_control *sc)
484 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
486 return atomic_read(&zmd->nr_mblks);
490 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
492 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
493 struct shrink_control *sc)
495 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
498 spin_lock(&zmd->mblk_lock);
499 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
500 spin_unlock(&zmd->mblk_lock);
502 return count ? count : SHRINK_STOP;
506 * Release a metadata block.
508 static void dmz_release_mblock(struct dmz_metadata *zmd,
509 struct dmz_mblock *mblk)
515 spin_lock(&zmd->mblk_lock);
518 if (mblk->ref == 0) {
519 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
520 rb_erase(&mblk->node, &zmd->mblk_rbtree);
521 dmz_free_mblock(zmd, mblk);
522 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
523 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
524 dmz_shrink_mblock_cache(zmd, 1);
528 spin_unlock(&zmd->mblk_lock);
532 * Get a metadata block from the rbtree. If the block
533 * is not present, read it from disk.
535 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
538 struct dmz_mblock *mblk;
541 spin_lock(&zmd->mblk_lock);
542 mblk = dmz_get_mblock_fast(zmd, mblk_no);
543 spin_unlock(&zmd->mblk_lock);
546 /* Cache miss: read the block from disk */
547 mblk = dmz_get_mblock_slow(zmd, mblk_no);
552 /* Wait for on-going read I/O and check for error */
553 wait_on_bit_io(&mblk->state, DMZ_META_READING,
554 TASK_UNINTERRUPTIBLE);
555 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
556 dmz_release_mblock(zmd, mblk);
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);
632 * Write super block of the specified metadata set.
634 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
636 sector_t block = zmd->sb[set].block;
637 struct dmz_mblock *mblk = zmd->sb[set].mblk;
638 struct dmz_super *sb = zmd->sb[set].sb;
639 u64 sb_gen = zmd->sb_gen + 1;
642 sb->magic = cpu_to_le32(DMZ_MAGIC);
643 sb->version = cpu_to_le32(DMZ_META_VER);
645 sb->gen = cpu_to_le64(sb_gen);
647 sb->sb_block = cpu_to_le64(block);
648 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
649 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
650 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
652 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
653 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
656 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
658 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
660 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
666 * Write dirty metadata blocks to the specified set.
668 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
669 struct list_head *write_list,
672 struct dmz_mblock *mblk;
673 struct blk_plug plug;
674 int ret = 0, nr_mblks_submitted = 0;
677 blk_start_plug(&plug);
678 list_for_each_entry(mblk, write_list, link) {
679 ret = dmz_write_mblock(zmd, mblk, set);
682 nr_mblks_submitted++;
684 blk_finish_plug(&plug);
686 /* Wait for completion */
687 list_for_each_entry(mblk, write_list, link) {
688 if (!nr_mblks_submitted)
690 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
691 TASK_UNINTERRUPTIBLE);
692 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
693 clear_bit(DMZ_META_ERROR, &mblk->state);
696 nr_mblks_submitted--;
699 /* Flush drive cache (this will also sync data) */
701 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
707 * Log dirty metadata blocks.
709 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
710 struct list_head *write_list)
712 unsigned int log_set = zmd->mblk_primary ^ 0x1;
715 /* Write dirty blocks to the log */
716 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
721 * No error so far: now validate the log by updating the
722 * log index super block generation.
724 ret = dmz_write_sb(zmd, log_set);
732 * Flush dirty metadata blocks.
734 int dmz_flush_metadata(struct dmz_metadata *zmd)
736 struct dmz_mblock *mblk;
737 struct list_head write_list;
743 INIT_LIST_HEAD(&write_list);
746 * Make sure that metadata blocks are stable before logging: take
747 * the write lock on the metadata semaphore to prevent target BIOs
748 * from modifying metadata.
750 down_write(&zmd->mblk_sem);
753 * This is called from the target flush work and reclaim work.
754 * Concurrent execution is not allowed.
758 if (dmz_bdev_is_dying(zmd->dev)) {
763 /* Get dirty blocks */
764 spin_lock(&zmd->mblk_lock);
765 list_splice_init(&zmd->mblk_dirty_list, &write_list);
766 spin_unlock(&zmd->mblk_lock);
768 /* If there are no dirty metadata blocks, just flush the device cache */
769 if (list_empty(&write_list)) {
770 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
775 * The primary metadata set is still clean. Keep it this way until
776 * all updates are successful in the secondary set. That is, use
777 * the secondary set as a log.
779 ret = dmz_log_dirty_mblocks(zmd, &write_list);
784 * The log is on disk. It is now safe to update in place
785 * in the primary metadata set.
787 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
791 ret = dmz_write_sb(zmd, zmd->mblk_primary);
795 while (!list_empty(&write_list)) {
796 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
797 list_del_init(&mblk->link);
799 spin_lock(&zmd->mblk_lock);
800 clear_bit(DMZ_META_DIRTY, &mblk->state);
802 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
803 spin_unlock(&zmd->mblk_lock);
808 if (ret && !list_empty(&write_list)) {
809 spin_lock(&zmd->mblk_lock);
810 list_splice(&write_list, &zmd->mblk_dirty_list);
811 spin_unlock(&zmd->mblk_lock);
814 dmz_unlock_flush(zmd);
815 up_write(&zmd->mblk_sem);
823 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
825 unsigned int nr_meta_zones, nr_data_zones;
826 struct dmz_dev *dev = zmd->dev;
830 gen = le64_to_cpu(sb->gen);
831 stored_crc = le32_to_cpu(sb->crc);
833 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
834 if (crc != stored_crc) {
835 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
840 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
841 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
842 DMZ_MAGIC, le32_to_cpu(sb->magic));
846 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
847 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
848 DMZ_META_VER, le32_to_cpu(sb->version));
852 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
853 >> dev->zone_nr_blocks_shift;
854 if (!nr_meta_zones ||
855 nr_meta_zones >= zmd->nr_rnd_zones) {
856 dmz_dev_err(dev, "Invalid number of metadata blocks");
860 if (!le32_to_cpu(sb->nr_reserved_seq) ||
861 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
862 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
866 nr_data_zones = zmd->nr_useable_zones -
867 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
868 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
869 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
870 le32_to_cpu(sb->nr_chunks), nr_data_zones);
875 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
876 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
877 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
878 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
879 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
880 zmd->nr_meta_zones = nr_meta_zones;
881 zmd->nr_data_zones = nr_data_zones;
887 * Read the first or second super block from disk.
889 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
891 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
892 zmd->sb[set].mblk->page);
896 * Determine the position of the secondary super blocks on disk.
897 * This is used only if a corruption of the primary super block
900 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
902 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
903 struct dmz_mblock *mblk;
906 /* Allocate a block */
907 mblk = dmz_alloc_mblock(zmd, 0);
911 zmd->sb[1].mblk = mblk;
912 zmd->sb[1].sb = mblk->data;
914 /* Bad first super block: search for the second one */
915 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
916 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
917 if (dmz_read_sb(zmd, 1) != 0)
919 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
921 zmd->sb[1].block += zone_nr_blocks;
924 dmz_free_mblock(zmd, mblk);
925 zmd->sb[1].mblk = NULL;
931 * Read the first or second super block from disk.
933 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
935 struct dmz_mblock *mblk;
938 /* Allocate a block */
939 mblk = dmz_alloc_mblock(zmd, 0);
943 zmd->sb[set].mblk = mblk;
944 zmd->sb[set].sb = mblk->data;
946 /* Read super block */
947 ret = dmz_read_sb(zmd, set);
949 dmz_free_mblock(zmd, mblk);
950 zmd->sb[set].mblk = NULL;
958 * Recover a metadata set.
960 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
962 unsigned int src_set = dst_set ^ 0x1;
966 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
969 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
971 zmd->sb[1].block = zmd->sb[0].block +
972 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
975 page = alloc_page(GFP_NOIO);
979 /* Copy metadata blocks */
980 for (i = 1; i < zmd->nr_meta_blocks; i++) {
981 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
982 zmd->sb[src_set].block + i, page);
985 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
986 zmd->sb[dst_set].block + i, page);
991 /* Finalize with the super block */
992 if (!zmd->sb[dst_set].mblk) {
993 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
994 if (!zmd->sb[dst_set].mblk) {
998 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1001 ret = dmz_write_sb(zmd, dst_set);
1003 __free_pages(page, 0);
1009 * Get super block from disk.
1011 static int dmz_load_sb(struct dmz_metadata *zmd)
1013 bool sb_good[2] = {false, false};
1014 u64 sb_gen[2] = {0, 0};
1017 /* Read and check the primary super block */
1018 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1019 ret = dmz_get_sb(zmd, 0);
1021 dmz_dev_err(zmd->dev, "Read primary super block failed");
1025 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1027 /* Read and check secondary super block */
1030 zmd->sb[1].block = zmd->sb[0].block +
1031 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1032 ret = dmz_get_sb(zmd, 1);
1034 ret = dmz_lookup_secondary_sb(zmd);
1037 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1041 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1045 /* Use highest generation sb first */
1046 if (!sb_good[0] && !sb_good[1]) {
1047 dmz_dev_err(zmd->dev, "No valid super block found");
1052 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1054 ret = dmz_recover_mblocks(zmd, 0);
1057 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1059 ret = dmz_recover_mblocks(zmd, 1);
1062 dmz_dev_err(zmd->dev, "Recovery failed");
1066 if (sb_gen[0] >= sb_gen[1]) {
1067 zmd->sb_gen = sb_gen[0];
1068 zmd->mblk_primary = 0;
1070 zmd->sb_gen = sb_gen[1];
1071 zmd->mblk_primary = 1;
1074 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1075 zmd->mblk_primary, zmd->sb_gen);
1081 * Initialize a zone descriptor.
1083 static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1084 struct blk_zone *blkz)
1086 struct dmz_dev *dev = zmd->dev;
1088 /* Ignore the eventual last runt (smaller) zone */
1089 if (blkz->len != dev->zone_nr_sectors) {
1090 if (blkz->start + blkz->len == dev->capacity)
1095 INIT_LIST_HEAD(&zone->link);
1096 atomic_set(&zone->refcount, 0);
1097 zone->chunk = DMZ_MAP_UNMAPPED;
1099 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1100 set_bit(DMZ_RND, &zone->flags);
1101 zmd->nr_rnd_zones++;
1102 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1103 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1104 set_bit(DMZ_SEQ, &zone->flags);
1108 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1109 set_bit(DMZ_OFFLINE, &zone->flags);
1110 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1111 set_bit(DMZ_READ_ONLY, &zone->flags);
1113 if (dmz_is_rnd(zone))
1116 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1118 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1119 zmd->nr_useable_zones++;
1120 if (dmz_is_rnd(zone)) {
1121 zmd->nr_rnd_zones++;
1122 if (!zmd->sb_zone) {
1123 /* Super block zone */
1124 zmd->sb_zone = zone;
1133 * Free zones descriptors.
1135 static void dmz_drop_zones(struct dmz_metadata *zmd)
1142 * The size of a zone report in number of zones.
1143 * This results in 4096*64B=256KB report zones commands.
1145 #define DMZ_REPORT_NR_ZONES 4096
1148 * Allocate and initialize zone descriptors using the zone
1149 * information from disk.
1151 static int dmz_init_zones(struct dmz_metadata *zmd)
1153 struct dmz_dev *dev = zmd->dev;
1154 struct dm_zone *zone;
1155 struct blk_zone *blkz;
1156 unsigned int nr_blkz;
1157 sector_t sector = 0;
1161 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1162 zmd->zone_nr_bitmap_blocks = zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT;
1164 /* Allocate zone array */
1165 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1169 dmz_dev_info(dev, "Using %zu B for zone information",
1170 sizeof(struct dm_zone) * dev->nr_zones);
1172 /* Get zone information */
1173 nr_blkz = DMZ_REPORT_NR_ZONES;
1174 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1181 * Get zone information and initialize zone descriptors.
1182 * At the same time, determine where the super block
1183 * should be: first block of the first randomly writable
1187 while (sector < dev->capacity) {
1188 /* Get zone information */
1189 nr_blkz = DMZ_REPORT_NR_ZONES;
1190 ret = blkdev_report_zones(dev->bdev, sector, blkz, &nr_blkz);
1192 dmz_dev_err(dev, "Report zones failed %d", ret);
1199 /* Process report */
1200 for (i = 0; i < nr_blkz; i++) {
1201 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1204 sector += dev->zone_nr_sectors;
1209 /* The entire zone configuration of the disk should now be known */
1210 if (sector < dev->capacity) {
1211 dmz_dev_err(dev, "Failed to get correct zone information");
1217 dmz_drop_zones(zmd);
1223 * Update a zone information.
1225 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1227 unsigned int nr_blkz = 1;
1228 unsigned int noio_flag;
1229 struct blk_zone blkz;
1233 * Get zone information from disk. Since blkdev_report_zones() uses
1234 * GFP_KERNEL by default for memory allocations, set the per-task
1235 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1236 * GFP_NOIO was specified.
1238 noio_flag = memalloc_noio_save();
1239 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1241 memalloc_noio_restore(noio_flag);
1245 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1250 clear_bit(DMZ_OFFLINE, &zone->flags);
1251 clear_bit(DMZ_READ_ONLY, &zone->flags);
1252 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1253 set_bit(DMZ_OFFLINE, &zone->flags);
1254 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1255 set_bit(DMZ_READ_ONLY, &zone->flags);
1257 if (dmz_is_seq(zone))
1258 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1266 * Check a zone write pointer position when the zone is marked
1267 * with the sequential write error flag.
1269 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1270 struct dm_zone *zone)
1272 unsigned int wp = 0;
1275 wp = zone->wp_block;
1276 ret = dmz_update_zone(zmd, zone);
1280 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1281 dmz_id(zmd, zone), zone->wp_block, wp);
1283 if (zone->wp_block < wp) {
1284 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1285 wp - zone->wp_block);
1291 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1293 return &zmd->zones[zone_id];
1297 * Reset a zone write pointer.
1299 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1304 * Ignore offline zones, read only zones,
1305 * and conventional zones.
1307 if (dmz_is_offline(zone) ||
1308 dmz_is_readonly(zone) ||
1312 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1313 struct dmz_dev *dev = zmd->dev;
1315 ret = blkdev_reset_zones(dev->bdev,
1316 dmz_start_sect(zmd, zone),
1317 dev->zone_nr_sectors, GFP_NOIO);
1319 dmz_dev_err(dev, "Reset zone %u failed %d",
1320 dmz_id(zmd, zone), ret);
1325 /* Clear write error bit and rewind write pointer position */
1326 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1332 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1335 * Initialize chunk mapping.
1337 static int dmz_load_mapping(struct dmz_metadata *zmd)
1339 struct dmz_dev *dev = zmd->dev;
1340 struct dm_zone *dzone, *bzone;
1341 struct dmz_mblock *dmap_mblk = NULL;
1342 struct dmz_map *dmap;
1343 unsigned int i = 0, e = 0, chunk = 0;
1344 unsigned int dzone_id;
1345 unsigned int bzone_id;
1347 /* Metadata block array for the chunk mapping table */
1348 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1349 sizeof(struct dmz_mblk *), GFP_KERNEL);
1353 /* Get chunk mapping table blocks and initialize zone mapping */
1354 while (chunk < zmd->nr_chunks) {
1356 /* Get mapping block */
1357 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1358 if (IS_ERR(dmap_mblk))
1359 return PTR_ERR(dmap_mblk);
1360 zmd->map_mblk[i] = dmap_mblk;
1361 dmap = (struct dmz_map *) dmap_mblk->data;
1366 /* Check data zone */
1367 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1368 if (dzone_id == DMZ_MAP_UNMAPPED)
1371 if (dzone_id >= dev->nr_zones) {
1372 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1377 dzone = dmz_get(zmd, dzone_id);
1378 set_bit(DMZ_DATA, &dzone->flags);
1379 dzone->chunk = chunk;
1380 dmz_get_zone_weight(zmd, dzone);
1382 if (dmz_is_rnd(dzone))
1383 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1385 list_add_tail(&dzone->link, &zmd->map_seq_list);
1387 /* Check buffer zone */
1388 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1389 if (bzone_id == DMZ_MAP_UNMAPPED)
1392 if (bzone_id >= dev->nr_zones) {
1393 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1398 bzone = dmz_get(zmd, bzone_id);
1399 if (!dmz_is_rnd(bzone)) {
1400 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1405 set_bit(DMZ_DATA, &bzone->flags);
1406 set_bit(DMZ_BUF, &bzone->flags);
1407 bzone->chunk = chunk;
1408 bzone->bzone = dzone;
1409 dzone->bzone = bzone;
1410 dmz_get_zone_weight(zmd, bzone);
1411 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1415 if (e >= DMZ_MAP_ENTRIES)
1420 * At this point, only meta zones and mapped data zones were
1421 * fully initialized. All remaining zones are unmapped data
1422 * zones. Finish initializing those here.
1424 for (i = 0; i < dev->nr_zones; i++) {
1425 dzone = dmz_get(zmd, i);
1426 if (dmz_is_meta(dzone))
1429 if (dmz_is_rnd(dzone))
1434 if (dmz_is_data(dzone)) {
1435 /* Already initialized */
1439 /* Unmapped data zone */
1440 set_bit(DMZ_DATA, &dzone->flags);
1441 dzone->chunk = DMZ_MAP_UNMAPPED;
1442 if (dmz_is_rnd(dzone)) {
1443 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1444 atomic_inc(&zmd->unmap_nr_rnd);
1445 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1446 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1447 atomic_inc(&zmd->nr_reserved_seq_zones);
1450 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1451 atomic_inc(&zmd->unmap_nr_seq);
1459 * Set a data chunk mapping.
1461 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1462 unsigned int dzone_id, unsigned int bzone_id)
1464 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1465 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1466 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1468 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1469 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1470 dmz_dirty_mblock(zmd, dmap_mblk);
1474 * The list of mapped zones is maintained in LRU order.
1475 * This rotates a zone at the end of its map list.
1477 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1479 if (list_empty(&zone->link))
1482 list_del_init(&zone->link);
1483 if (dmz_is_seq(zone)) {
1484 /* LRU rotate sequential zone */
1485 list_add_tail(&zone->link, &zmd->map_seq_list);
1487 /* LRU rotate random zone */
1488 list_add_tail(&zone->link, &zmd->map_rnd_list);
1493 * The list of mapped random zones is maintained
1494 * in LRU order. This rotates a zone at the end of the list.
1496 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1498 __dmz_lru_zone(zmd, zone);
1500 __dmz_lru_zone(zmd, zone->bzone);
1504 * Wait for any zone to be freed.
1506 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1510 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1511 dmz_unlock_map(zmd);
1512 dmz_unlock_metadata(zmd);
1514 io_schedule_timeout(HZ);
1516 dmz_lock_metadata(zmd);
1518 finish_wait(&zmd->free_wq, &wait);
1522 * Lock a zone for reclaim (set the zone RECLAIM bit).
1523 * Returns false if the zone cannot be locked or if it is already locked
1526 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1528 /* Active zones cannot be reclaimed */
1529 if (dmz_is_active(zone))
1532 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1536 * Clear a zone reclaim flag.
1538 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1540 WARN_ON(dmz_is_active(zone));
1541 WARN_ON(!dmz_in_reclaim(zone));
1543 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1544 smp_mb__after_atomic();
1545 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1549 * Wait for a zone reclaim to complete.
1551 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1553 dmz_unlock_map(zmd);
1554 dmz_unlock_metadata(zmd);
1555 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1556 dmz_lock_metadata(zmd);
1561 * Select a random write zone for reclaim.
1563 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1565 struct dm_zone *dzone = NULL;
1566 struct dm_zone *zone;
1568 if (list_empty(&zmd->map_rnd_list))
1569 return ERR_PTR(-EBUSY);
1571 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1572 if (dmz_is_buf(zone))
1573 dzone = zone->bzone;
1576 if (dmz_lock_zone_reclaim(dzone))
1580 return ERR_PTR(-EBUSY);
1584 * Select a buffered sequential zone for reclaim.
1586 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1588 struct dm_zone *zone;
1590 if (list_empty(&zmd->map_seq_list))
1591 return ERR_PTR(-EBUSY);
1593 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1596 if (dmz_lock_zone_reclaim(zone))
1600 return ERR_PTR(-EBUSY);
1604 * Select a zone for reclaim.
1606 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1608 struct dm_zone *zone;
1611 * Search for a zone candidate to reclaim: 2 cases are possible.
1612 * (1) There is no free sequential zones. Then a random data zone
1613 * cannot be reclaimed. So choose a sequential zone to reclaim so
1614 * that afterward a random zone can be reclaimed.
1615 * (2) At least one free sequential zone is available, then choose
1616 * the oldest random zone (data or buffer) that can be locked.
1619 if (list_empty(&zmd->reserved_seq_zones_list))
1620 zone = dmz_get_seq_zone_for_reclaim(zmd);
1622 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1623 dmz_unlock_map(zmd);
1629 * Get the zone mapping a chunk, if the chunk is mapped already.
1630 * If no mapping exist and the operation is WRITE, a zone is
1631 * allocated and used to map the chunk.
1632 * The zone returned will be set to the active state.
1634 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1636 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1637 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1638 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1639 unsigned int dzone_id;
1640 struct dm_zone *dzone = NULL;
1645 /* Get the chunk mapping */
1646 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1647 if (dzone_id == DMZ_MAP_UNMAPPED) {
1649 * Read or discard in unmapped chunks are fine. But for
1650 * writes, we need a mapping, so get one.
1652 if (op != REQ_OP_WRITE)
1655 /* Allocate a random zone */
1656 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1658 if (dmz_bdev_is_dying(zmd->dev)) {
1659 dzone = ERR_PTR(-EIO);
1662 dmz_wait_for_free_zones(zmd);
1666 dmz_map_zone(zmd, dzone, chunk);
1669 /* The chunk is already mapped: get the mapping zone */
1670 dzone = dmz_get(zmd, dzone_id);
1671 if (dzone->chunk != chunk) {
1672 dzone = ERR_PTR(-EIO);
1676 /* Repair write pointer if the sequential dzone has error */
1677 if (dmz_seq_write_err(dzone)) {
1678 ret = dmz_handle_seq_write_err(zmd, dzone);
1680 dzone = ERR_PTR(-EIO);
1683 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1688 * If the zone is being reclaimed, the chunk mapping may change
1689 * to a different zone. So wait for reclaim and retry. Otherwise,
1690 * activate the zone (this will prevent reclaim from touching it).
1692 if (dmz_in_reclaim(dzone)) {
1693 dmz_wait_for_reclaim(zmd, dzone);
1696 dmz_activate_zone(dzone);
1697 dmz_lru_zone(zmd, dzone);
1699 dmz_unlock_map(zmd);
1705 * Write and discard change the block validity of data zones and their buffer
1706 * zones. Check here that valid blocks are still present. If all blocks are
1707 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1710 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1712 struct dm_zone *bzone;
1716 bzone = dzone->bzone;
1718 if (dmz_weight(bzone))
1719 dmz_lru_zone(zmd, bzone);
1721 /* Empty buffer zone: reclaim it */
1722 dmz_unmap_zone(zmd, bzone);
1723 dmz_free_zone(zmd, bzone);
1728 /* Deactivate the data zone */
1729 dmz_deactivate_zone(dzone);
1730 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1731 dmz_lru_zone(zmd, dzone);
1733 /* Unbuffered inactive empty data zone: reclaim it */
1734 dmz_unmap_zone(zmd, dzone);
1735 dmz_free_zone(zmd, dzone);
1738 dmz_unlock_map(zmd);
1742 * Allocate and map a random zone to buffer a chunk
1743 * already mapped to a sequential zone.
1745 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1746 struct dm_zone *dzone)
1748 struct dm_zone *bzone;
1752 bzone = dzone->bzone;
1756 /* Allocate a random zone */
1757 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1759 if (dmz_bdev_is_dying(zmd->dev)) {
1760 bzone = ERR_PTR(-EIO);
1763 dmz_wait_for_free_zones(zmd);
1767 /* Update the chunk mapping */
1768 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1769 dmz_id(zmd, bzone));
1771 set_bit(DMZ_BUF, &bzone->flags);
1772 bzone->chunk = dzone->chunk;
1773 bzone->bzone = dzone;
1774 dzone->bzone = bzone;
1775 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1777 dmz_unlock_map(zmd);
1783 * Get an unmapped (free) zone.
1784 * This must be called with the mapping lock held.
1786 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1788 struct list_head *list;
1789 struct dm_zone *zone;
1791 if (flags & DMZ_ALLOC_RND)
1792 list = &zmd->unmap_rnd_list;
1794 list = &zmd->unmap_seq_list;
1796 if (list_empty(list)) {
1798 * No free zone: if this is for reclaim, allow using the
1799 * reserved sequential zones.
1801 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1802 list_empty(&zmd->reserved_seq_zones_list))
1805 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1806 struct dm_zone, link);
1807 list_del_init(&zone->link);
1808 atomic_dec(&zmd->nr_reserved_seq_zones);
1812 zone = list_first_entry(list, struct dm_zone, link);
1813 list_del_init(&zone->link);
1815 if (dmz_is_rnd(zone))
1816 atomic_dec(&zmd->unmap_nr_rnd);
1818 atomic_dec(&zmd->unmap_nr_seq);
1820 if (dmz_is_offline(zone)) {
1821 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1831 * This must be called with the mapping lock held.
1833 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1835 /* If this is a sequential zone, reset it */
1836 if (dmz_is_seq(zone))
1837 dmz_reset_zone(zmd, zone);
1839 /* Return the zone to its type unmap list */
1840 if (dmz_is_rnd(zone)) {
1841 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1842 atomic_inc(&zmd->unmap_nr_rnd);
1843 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1844 zmd->nr_reserved_seq) {
1845 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1846 atomic_inc(&zmd->nr_reserved_seq_zones);
1848 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1849 atomic_inc(&zmd->unmap_nr_seq);
1852 wake_up_all(&zmd->free_wq);
1856 * Map a chunk to a zone.
1857 * This must be called with the mapping lock held.
1859 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1862 /* Set the chunk mapping */
1863 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1865 dzone->chunk = chunk;
1866 if (dmz_is_rnd(dzone))
1867 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1869 list_add_tail(&dzone->link, &zmd->map_seq_list);
1874 * This must be called with the mapping lock held.
1876 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1878 unsigned int chunk = zone->chunk;
1879 unsigned int dzone_id;
1881 if (chunk == DMZ_MAP_UNMAPPED) {
1882 /* Already unmapped */
1886 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1888 * Unmapping the chunk buffer zone: clear only
1889 * the chunk buffer mapping
1891 dzone_id = dmz_id(zmd, zone->bzone);
1892 zone->bzone->bzone = NULL;
1897 * Unmapping the chunk data zone: the zone must
1900 if (WARN_ON(zone->bzone)) {
1901 zone->bzone->bzone = NULL;
1904 dzone_id = DMZ_MAP_UNMAPPED;
1907 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1909 zone->chunk = DMZ_MAP_UNMAPPED;
1910 list_del_init(&zone->link);
1914 * Set @nr_bits bits in @bitmap starting from @bit.
1915 * Return the number of bits changed from 0 to 1.
1917 static unsigned int dmz_set_bits(unsigned long *bitmap,
1918 unsigned int bit, unsigned int nr_bits)
1920 unsigned long *addr;
1921 unsigned int end = bit + nr_bits;
1925 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1926 ((end - bit) >= BITS_PER_LONG)) {
1927 /* Try to set the whole word at once */
1928 addr = bitmap + BIT_WORD(bit);
1932 bit += BITS_PER_LONG;
1937 if (!test_and_set_bit(bit, bitmap))
1946 * Get the bitmap block storing the bit for chunk_block in zone.
1948 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1949 struct dm_zone *zone,
1950 sector_t chunk_block)
1952 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1953 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1954 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1956 return dmz_get_mblock(zmd, bitmap_block);
1960 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1962 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1963 struct dm_zone *to_zone)
1965 struct dmz_mblock *from_mblk, *to_mblk;
1966 sector_t chunk_block = 0;
1968 /* Get the zones bitmap blocks */
1969 while (chunk_block < zmd->dev->zone_nr_blocks) {
1970 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1971 if (IS_ERR(from_mblk))
1972 return PTR_ERR(from_mblk);
1973 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1974 if (IS_ERR(to_mblk)) {
1975 dmz_release_mblock(zmd, from_mblk);
1976 return PTR_ERR(to_mblk);
1979 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1980 dmz_dirty_mblock(zmd, to_mblk);
1982 dmz_release_mblock(zmd, to_mblk);
1983 dmz_release_mblock(zmd, from_mblk);
1985 chunk_block += DMZ_BLOCK_SIZE_BITS;
1988 to_zone->weight = from_zone->weight;
1994 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1995 * starting from chunk_block.
1997 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1998 struct dm_zone *to_zone, sector_t chunk_block)
2000 unsigned int nr_blocks;
2003 /* Get the zones bitmap blocks */
2004 while (chunk_block < zmd->dev->zone_nr_blocks) {
2005 /* Get a valid region from the source zone */
2006 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2011 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2015 chunk_block += nr_blocks;
2022 * Validate all the blocks in the range [block..block+nr_blocks-1].
2024 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2025 sector_t chunk_block, unsigned int nr_blocks)
2027 unsigned int count, bit, nr_bits;
2028 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2029 struct dmz_mblock *mblk;
2032 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2033 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2036 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2039 /* Get bitmap block */
2040 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2042 return PTR_ERR(mblk);
2045 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2046 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2048 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2050 dmz_dirty_mblock(zmd, mblk);
2053 dmz_release_mblock(zmd, mblk);
2055 nr_blocks -= nr_bits;
2056 chunk_block += nr_bits;
2059 if (likely(zone->weight + n <= zone_nr_blocks))
2062 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2063 dmz_id(zmd, zone), zone->weight,
2064 zone_nr_blocks - n);
2065 zone->weight = zone_nr_blocks;
2072 * Clear nr_bits bits in bitmap starting from bit.
2073 * Return the number of bits cleared.
2075 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2077 unsigned long *addr;
2078 int end = bit + nr_bits;
2082 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2083 ((end - bit) >= BITS_PER_LONG)) {
2084 /* Try to clear whole word at once */
2085 addr = bitmap + BIT_WORD(bit);
2086 if (*addr == ULONG_MAX) {
2089 bit += BITS_PER_LONG;
2094 if (test_and_clear_bit(bit, bitmap))
2103 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2105 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2106 sector_t chunk_block, unsigned int nr_blocks)
2108 unsigned int count, bit, nr_bits;
2109 struct dmz_mblock *mblk;
2112 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2113 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2115 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2118 /* Get bitmap block */
2119 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2121 return PTR_ERR(mblk);
2124 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2125 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2127 count = dmz_clear_bits((unsigned long *)mblk->data,
2130 dmz_dirty_mblock(zmd, mblk);
2133 dmz_release_mblock(zmd, mblk);
2135 nr_blocks -= nr_bits;
2136 chunk_block += nr_bits;
2139 if (zone->weight >= n)
2142 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2143 dmz_id(zmd, zone), zone->weight, n);
2151 * Get a block bit value.
2153 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2154 sector_t chunk_block)
2156 struct dmz_mblock *mblk;
2159 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2161 /* Get bitmap block */
2162 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2164 return PTR_ERR(mblk);
2167 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2168 (unsigned long *) mblk->data) != 0;
2170 dmz_release_mblock(zmd, mblk);
2176 * Return the number of blocks from chunk_block to the first block with a bit
2177 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2179 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2180 sector_t chunk_block, unsigned int nr_blocks,
2183 struct dmz_mblock *mblk;
2184 unsigned int bit, set_bit, nr_bits;
2185 unsigned long *bitmap;
2188 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2191 /* Get bitmap block */
2192 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2194 return PTR_ERR(mblk);
2197 bitmap = (unsigned long *) mblk->data;
2198 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2199 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2201 set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2203 set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2204 dmz_release_mblock(zmd, mblk);
2207 if (set_bit < DMZ_BLOCK_SIZE_BITS)
2210 nr_blocks -= nr_bits;
2211 chunk_block += nr_bits;
2218 * Test if chunk_block is valid. If it is, the number of consecutive
2219 * valid blocks from chunk_block will be returned.
2221 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2222 sector_t chunk_block)
2226 valid = dmz_test_block(zmd, zone, chunk_block);
2230 /* The block is valid: get the number of valid blocks from block */
2231 return dmz_to_next_set_block(zmd, zone, chunk_block,
2232 zmd->dev->zone_nr_blocks - chunk_block, 0);
2236 * Find the first valid block from @chunk_block in @zone.
2237 * If such a block is found, its number is returned using
2238 * @chunk_block and the total number of valid blocks from @chunk_block
2241 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2242 sector_t *chunk_block)
2244 sector_t start_block = *chunk_block;
2247 ret = dmz_to_next_set_block(zmd, zone, start_block,
2248 zmd->dev->zone_nr_blocks - start_block, 1);
2253 *chunk_block = start_block;
2255 return dmz_to_next_set_block(zmd, zone, start_block,
2256 zmd->dev->zone_nr_blocks - start_block, 0);
2260 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2262 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2264 unsigned long *addr;
2265 int end = bit + nr_bits;
2269 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2270 ((end - bit) >= BITS_PER_LONG)) {
2271 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2272 if (*addr == ULONG_MAX) {
2274 bit += BITS_PER_LONG;
2279 if (test_bit(bit, bitmap))
2288 * Get a zone weight.
2290 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2292 struct dmz_mblock *mblk;
2293 sector_t chunk_block = 0;
2294 unsigned int bit, nr_bits;
2295 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2300 /* Get bitmap block */
2301 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2307 /* Count bits in this block */
2308 bitmap = mblk->data;
2309 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2310 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2311 n += dmz_count_bits(bitmap, bit, nr_bits);
2313 dmz_release_mblock(zmd, mblk);
2315 nr_blocks -= nr_bits;
2316 chunk_block += nr_bits;
2323 * Cleanup the zoned metadata resources.
2325 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2327 struct rb_root *root;
2328 struct dmz_mblock *mblk, *next;
2331 /* Release zone mapping resources */
2332 if (zmd->map_mblk) {
2333 for (i = 0; i < zmd->nr_map_blocks; i++)
2334 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2335 kfree(zmd->map_mblk);
2336 zmd->map_mblk = NULL;
2339 /* Release super blocks */
2340 for (i = 0; i < 2; i++) {
2341 if (zmd->sb[i].mblk) {
2342 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2343 zmd->sb[i].mblk = NULL;
2347 /* Free cached blocks */
2348 while (!list_empty(&zmd->mblk_dirty_list)) {
2349 mblk = list_first_entry(&zmd->mblk_dirty_list,
2350 struct dmz_mblock, link);
2351 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2352 (u64)mblk->no, mblk->ref);
2353 list_del_init(&mblk->link);
2354 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2355 dmz_free_mblock(zmd, mblk);
2358 while (!list_empty(&zmd->mblk_lru_list)) {
2359 mblk = list_first_entry(&zmd->mblk_lru_list,
2360 struct dmz_mblock, link);
2361 list_del_init(&mblk->link);
2362 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2363 dmz_free_mblock(zmd, mblk);
2366 /* Sanity checks: the mblock rbtree should now be empty */
2367 root = &zmd->mblk_rbtree;
2368 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2369 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2370 (u64)mblk->no, mblk->ref);
2372 dmz_free_mblock(zmd, mblk);
2375 /* Free the zone descriptors */
2376 dmz_drop_zones(zmd);
2378 mutex_destroy(&zmd->mblk_flush_lock);
2379 mutex_destroy(&zmd->map_lock);
2383 * Initialize the zoned metadata.
2385 int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2387 struct dmz_metadata *zmd;
2388 unsigned int i, zid;
2389 struct dm_zone *zone;
2392 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2397 zmd->mblk_rbtree = RB_ROOT;
2398 init_rwsem(&zmd->mblk_sem);
2399 mutex_init(&zmd->mblk_flush_lock);
2400 spin_lock_init(&zmd->mblk_lock);
2401 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2402 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2404 mutex_init(&zmd->map_lock);
2405 atomic_set(&zmd->unmap_nr_rnd, 0);
2406 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2407 INIT_LIST_HEAD(&zmd->map_rnd_list);
2409 atomic_set(&zmd->unmap_nr_seq, 0);
2410 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2411 INIT_LIST_HEAD(&zmd->map_seq_list);
2413 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2414 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2416 init_waitqueue_head(&zmd->free_wq);
2418 /* Initialize zone descriptors */
2419 ret = dmz_init_zones(zmd);
2423 /* Get super block */
2424 ret = dmz_load_sb(zmd);
2428 /* Set metadata zones starting from sb_zone */
2429 zid = dmz_id(zmd, zmd->sb_zone);
2430 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2431 zone = dmz_get(zmd, zid + i);
2432 if (!dmz_is_rnd(zone))
2434 set_bit(DMZ_META, &zone->flags);
2437 /* Load mapping table */
2438 ret = dmz_load_mapping(zmd);
2443 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2444 * blocks and enough blocks to be able to cache the bitmap blocks of
2445 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2446 * the cache to add 512 more metadata blocks.
2448 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2449 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2450 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2451 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2452 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2454 /* Metadata cache shrinker */
2455 ret = register_shrinker(&zmd->mblk_shrinker);
2457 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2461 dmz_dev_info(dev, "Host-%s zoned block device",
2462 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2463 "aware" : "managed");
2464 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2465 (u64)dev->capacity);
2466 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2467 dev->nr_zones, (u64)dev->zone_nr_sectors);
2468 dmz_dev_info(dev, " %u metadata zones",
2469 zmd->nr_meta_zones * 2);
2470 dmz_dev_info(dev, " %u data zones for %u chunks",
2471 zmd->nr_data_zones, zmd->nr_chunks);
2472 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2473 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2474 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2475 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2476 dmz_dev_info(dev, " %u reserved sequential data zones",
2477 zmd->nr_reserved_seq);
2479 dmz_dev_debug(dev, "Format:");
2480 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2481 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2482 dmz_dev_debug(dev, " %u data zone mapping blocks",
2483 zmd->nr_map_blocks);
2484 dmz_dev_debug(dev, " %u bitmap blocks",
2485 zmd->nr_bitmap_blocks);
2491 dmz_cleanup_metadata(zmd);
2499 * Cleanup the zoned metadata resources.
2501 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2503 unregister_shrinker(&zmd->mblk_shrinker);
2504 dmz_cleanup_metadata(zmd);
2509 * Check zone information on resume.
2511 int dmz_resume_metadata(struct dmz_metadata *zmd)
2513 struct dmz_dev *dev = zmd->dev;
2514 struct dm_zone *zone;
2520 for (i = 0; i < dev->nr_zones; i++) {
2521 zone = dmz_get(zmd, i);
2523 dmz_dev_err(dev, "Unable to get zone %u", i);
2527 wp_block = zone->wp_block;
2529 ret = dmz_update_zone(zmd, zone);
2531 dmz_dev_err(dev, "Broken zone %u", i);
2535 if (dmz_is_offline(zone)) {
2536 dmz_dev_warn(dev, "Zone %u is offline", i);
2540 /* Check write pointer */
2541 if (!dmz_is_seq(zone))
2543 else if (zone->wp_block != wp_block) {
2544 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2545 i, (u64)zone->wp_block, (u64)wp_block);
2546 zone->wp_block = wp_block;
2547 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2548 dev->zone_nr_blocks - zone->wp_block);