2 * Copyright (C) 2017 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
9 #include <linux/module.h>
10 #include <linux/crc32.h>
12 #define DM_MSG_PREFIX "zoned metadata"
17 #define DMZ_META_VER 1
20 * On-disk super block magic.
22 #define DMZ_MAGIC ((((unsigned int)('D')) << 24) | \
23 (((unsigned int)('Z')) << 16) | \
24 (((unsigned int)('B')) << 8) | \
25 ((unsigned int)('D')))
28 * On disk super block.
29 * This uses only 512 B but uses on disk a full 4KB block. This block is
30 * followed on disk by the mapping table of chunks to zones and the bitmap
31 * blocks indicating zone block validity.
32 * The overall resulting metadata format is:
33 * (1) Super block (1 block)
34 * (2) Chunk mapping table (nr_map_blocks)
35 * (3) Bitmap blocks (nr_bitmap_blocks)
36 * All metadata blocks are stored in conventional zones, starting from the
37 * the first conventional zone found on disk.
43 /* Metadata version number */
44 __le32 version; /* 8 */
46 /* Generation number */
49 /* This block number */
50 __le64 sb_block; /* 24 */
52 /* The number of metadata blocks, including this super block */
53 __le32 nr_meta_blocks; /* 28 */
55 /* The number of sequential zones reserved for reclaim */
56 __le32 nr_reserved_seq; /* 32 */
58 /* The number of entries in the mapping table */
59 __le32 nr_chunks; /* 36 */
61 /* The number of blocks used for the chunk mapping table */
62 __le32 nr_map_blocks; /* 40 */
64 /* The number of blocks used for the block bitmaps */
65 __le32 nr_bitmap_blocks; /* 44 */
70 /* Padding to full 512B sector */
71 u8 reserved[464]; /* 512 */
75 * Chunk mapping entry: entries are indexed by chunk number
76 * and give the zone ID (dzone_id) mapping the chunk on disk.
77 * This zone may be sequential or random. If it is a sequential
78 * zone, a second zone (bzone_id) used as a write buffer may
79 * also be specified. This second zone will always be a randomly
88 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
90 #define DMZ_MAP_ENTRIES (DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
91 #define DMZ_MAP_ENTRIES_SHIFT (ilog2(DMZ_MAP_ENTRIES))
92 #define DMZ_MAP_ENTRIES_MASK (DMZ_MAP_ENTRIES - 1)
93 #define DMZ_MAP_UNMAPPED UINT_MAX
96 * Meta data block descriptor (for cached metadata blocks).
100 struct list_head link;
109 * Metadata block state flags.
119 * Super block information (one per metadata set).
123 struct dmz_mblock *mblk;
124 struct dmz_super *sb;
128 * In-memory metadata.
130 struct dmz_metadata {
133 sector_t zone_bitmap_size;
134 unsigned int zone_nr_bitmap_blocks;
136 unsigned int nr_bitmap_blocks;
137 unsigned int nr_map_blocks;
139 unsigned int nr_useable_zones;
140 unsigned int nr_meta_blocks;
141 unsigned int nr_meta_zones;
142 unsigned int nr_data_zones;
143 unsigned int nr_rnd_zones;
144 unsigned int nr_reserved_seq;
145 unsigned int nr_chunks;
147 /* Zone information array */
148 struct dm_zone *zones;
150 struct dm_zone *sb_zone;
152 unsigned int mblk_primary;
154 unsigned int min_nr_mblks;
155 unsigned int max_nr_mblks;
157 struct rw_semaphore mblk_sem;
158 struct mutex mblk_flush_lock;
159 spinlock_t mblk_lock;
160 struct rb_root mblk_rbtree;
161 struct list_head mblk_lru_list;
162 struct list_head mblk_dirty_list;
163 struct shrinker mblk_shrinker;
165 /* Zone allocation management */
166 struct mutex map_lock;
167 struct dmz_mblock **map_mblk;
169 atomic_t unmap_nr_rnd;
170 struct list_head unmap_rnd_list;
171 struct list_head map_rnd_list;
174 atomic_t unmap_nr_seq;
175 struct list_head unmap_seq_list;
176 struct list_head map_seq_list;
178 atomic_t nr_reserved_seq_zones;
179 struct list_head reserved_seq_zones_list;
181 wait_queue_head_t free_wq;
187 unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
189 return ((unsigned int)(zone - zmd->zones));
192 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
194 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
197 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
199 return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
202 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
204 return zmd->nr_chunks;
207 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
212 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
214 return atomic_read(&zmd->unmap_nr_rnd);
218 * Lock/unlock mapping table.
219 * The map lock also protects all the zone lists.
221 void dmz_lock_map(struct dmz_metadata *zmd)
223 mutex_lock(&zmd->map_lock);
226 void dmz_unlock_map(struct dmz_metadata *zmd)
228 mutex_unlock(&zmd->map_lock);
232 * Lock/unlock metadata access. This is a "read" lock on a semaphore
233 * that prevents metadata flush from running while metadata are being
234 * modified. The actual metadata write mutual exclusion is achieved with
235 * the map lock and zone styate management (active and reclaim state are
236 * mutually exclusive).
238 void dmz_lock_metadata(struct dmz_metadata *zmd)
240 down_read(&zmd->mblk_sem);
243 void dmz_unlock_metadata(struct dmz_metadata *zmd)
245 up_read(&zmd->mblk_sem);
249 * Lock/unlock flush: prevent concurrent executions
250 * of dmz_flush_metadata as well as metadata modification in reclaim
251 * while flush is being executed.
253 void dmz_lock_flush(struct dmz_metadata *zmd)
255 mutex_lock(&zmd->mblk_flush_lock);
258 void dmz_unlock_flush(struct dmz_metadata *zmd)
260 mutex_unlock(&zmd->mblk_flush_lock);
264 * Allocate a metadata block.
266 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
269 struct dmz_mblock *mblk = NULL;
271 /* See if we can reuse cached blocks */
272 if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
273 spin_lock(&zmd->mblk_lock);
274 mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
275 struct dmz_mblock, link);
277 list_del_init(&mblk->link);
278 rb_erase(&mblk->node, &zmd->mblk_rbtree);
281 spin_unlock(&zmd->mblk_lock);
286 /* Allocate a new block */
287 mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
291 mblk->page = alloc_page(GFP_NOIO);
297 RB_CLEAR_NODE(&mblk->node);
298 INIT_LIST_HEAD(&mblk->link);
302 mblk->data = page_address(mblk->page);
304 atomic_inc(&zmd->nr_mblks);
310 * Free a metadata block.
312 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
314 __free_pages(mblk->page, 0);
317 atomic_dec(&zmd->nr_mblks);
321 * Insert a metadata block in the rbtree.
323 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
325 struct rb_root *root = &zmd->mblk_rbtree;
326 struct rb_node **new = &(root->rb_node), *parent = NULL;
327 struct dmz_mblock *b;
329 /* Figure out where to put the new node */
331 b = container_of(*new, struct dmz_mblock, node);
333 new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
336 /* Add new node and rebalance tree */
337 rb_link_node(&mblk->node, parent, new);
338 rb_insert_color(&mblk->node, root);
342 * Lookup a metadata block in the rbtree. If the block is found, increment
343 * its reference count.
345 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
348 struct rb_root *root = &zmd->mblk_rbtree;
349 struct rb_node *node = root->rb_node;
350 struct dmz_mblock *mblk;
353 mblk = container_of(node, struct dmz_mblock, node);
354 if (mblk->no == mblk_no) {
356 * If this is the first reference to the block,
357 * remove it from the LRU list.
360 if (mblk->ref == 1 &&
361 !test_bit(DMZ_META_DIRTY, &mblk->state))
362 list_del_init(&mblk->link);
365 node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
372 * Metadata block BIO end callback.
374 static void dmz_mblock_bio_end_io(struct bio *bio)
376 struct dmz_mblock *mblk = bio->bi_private;
380 set_bit(DMZ_META_ERROR, &mblk->state);
382 if (bio_op(bio) == REQ_OP_WRITE)
383 flag = DMZ_META_WRITING;
385 flag = DMZ_META_READING;
387 clear_bit_unlock(flag, &mblk->state);
388 smp_mb__after_atomic();
389 wake_up_bit(&mblk->state, flag);
395 * Read an uncached metadata block from disk and add it to the cache.
397 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
400 struct dmz_mblock *mblk, *m;
401 sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
404 /* Get a new block and a BIO to read it */
405 mblk = dmz_alloc_mblock(zmd, mblk_no);
409 bio = bio_alloc(GFP_NOIO, 1);
411 dmz_free_mblock(zmd, mblk);
415 spin_lock(&zmd->mblk_lock);
418 * Make sure that another context did not start reading
421 m = dmz_get_mblock_fast(zmd, mblk_no);
423 spin_unlock(&zmd->mblk_lock);
424 dmz_free_mblock(zmd, mblk);
430 set_bit(DMZ_META_READING, &mblk->state);
431 dmz_insert_mblock(zmd, mblk);
433 spin_unlock(&zmd->mblk_lock);
435 /* Submit read BIO */
436 bio->bi_iter.bi_sector = dmz_blk2sect(block);
437 bio_set_dev(bio, zmd->dev->bdev);
438 bio->bi_private = mblk;
439 bio->bi_end_io = dmz_mblock_bio_end_io;
440 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
441 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
448 * Free metadata blocks.
450 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
453 struct dmz_mblock *mblk;
454 unsigned long count = 0;
456 if (!zmd->max_nr_mblks)
459 while (!list_empty(&zmd->mblk_lru_list) &&
460 atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
462 mblk = list_first_entry(&zmd->mblk_lru_list,
463 struct dmz_mblock, link);
464 list_del_init(&mblk->link);
465 rb_erase(&mblk->node, &zmd->mblk_rbtree);
466 dmz_free_mblock(zmd, mblk);
474 * For mblock shrinker: get the number of unused metadata blocks in the cache.
476 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
477 struct shrink_control *sc)
479 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
481 return atomic_read(&zmd->nr_mblks);
485 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
487 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
488 struct shrink_control *sc)
490 struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
493 spin_lock(&zmd->mblk_lock);
494 count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
495 spin_unlock(&zmd->mblk_lock);
497 return count ? count : SHRINK_STOP;
501 * Release a metadata block.
503 static void dmz_release_mblock(struct dmz_metadata *zmd,
504 struct dmz_mblock *mblk)
510 spin_lock(&zmd->mblk_lock);
513 if (mblk->ref == 0) {
514 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
515 rb_erase(&mblk->node, &zmd->mblk_rbtree);
516 dmz_free_mblock(zmd, mblk);
517 } else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
518 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
519 dmz_shrink_mblock_cache(zmd, 1);
523 spin_unlock(&zmd->mblk_lock);
527 * Get a metadata block from the rbtree. If the block
528 * is not present, read it from disk.
530 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
533 struct dmz_mblock *mblk;
536 spin_lock(&zmd->mblk_lock);
537 mblk = dmz_get_mblock_fast(zmd, mblk_no);
538 spin_unlock(&zmd->mblk_lock);
541 /* Cache miss: read the block from disk */
542 mblk = dmz_get_mblock_slow(zmd, mblk_no);
544 return ERR_PTR(-ENOMEM);
547 /* Wait for on-going read I/O and check for error */
548 wait_on_bit_io(&mblk->state, DMZ_META_READING,
549 TASK_UNINTERRUPTIBLE);
550 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
551 dmz_release_mblock(zmd, mblk);
552 return ERR_PTR(-EIO);
559 * Mark a metadata block dirty.
561 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
563 spin_lock(&zmd->mblk_lock);
564 if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
565 list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
566 spin_unlock(&zmd->mblk_lock);
570 * Issue a metadata block write BIO.
572 static void dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
575 sector_t block = zmd->sb[set].block + mblk->no;
578 bio = bio_alloc(GFP_NOIO, 1);
580 set_bit(DMZ_META_ERROR, &mblk->state);
584 set_bit(DMZ_META_WRITING, &mblk->state);
586 bio->bi_iter.bi_sector = dmz_blk2sect(block);
587 bio_set_dev(bio, zmd->dev->bdev);
588 bio->bi_private = mblk;
589 bio->bi_end_io = dmz_mblock_bio_end_io;
590 bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
591 bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
596 * Read/write a metadata block.
598 static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
604 bio = bio_alloc(GFP_NOIO, 1);
608 bio->bi_iter.bi_sector = dmz_blk2sect(block);
609 bio_set_dev(bio, zmd->dev->bdev);
610 bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
611 bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
612 ret = submit_bio_wait(bio);
619 * Write super block of the specified metadata set.
621 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
623 sector_t block = zmd->sb[set].block;
624 struct dmz_mblock *mblk = zmd->sb[set].mblk;
625 struct dmz_super *sb = zmd->sb[set].sb;
626 u64 sb_gen = zmd->sb_gen + 1;
629 sb->magic = cpu_to_le32(DMZ_MAGIC);
630 sb->version = cpu_to_le32(DMZ_META_VER);
632 sb->gen = cpu_to_le64(sb_gen);
634 sb->sb_block = cpu_to_le64(block);
635 sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
636 sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
637 sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
639 sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
640 sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
643 sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
645 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
647 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
653 * Write dirty metadata blocks to the specified set.
655 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
656 struct list_head *write_list,
659 struct dmz_mblock *mblk;
660 struct blk_plug plug;
664 blk_start_plug(&plug);
665 list_for_each_entry(mblk, write_list, link)
666 dmz_write_mblock(zmd, mblk, set);
667 blk_finish_plug(&plug);
669 /* Wait for completion */
670 list_for_each_entry(mblk, write_list, link) {
671 wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
672 TASK_UNINTERRUPTIBLE);
673 if (test_bit(DMZ_META_ERROR, &mblk->state)) {
674 clear_bit(DMZ_META_ERROR, &mblk->state);
679 /* Flush drive cache (this will also sync data) */
681 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
687 * Log dirty metadata blocks.
689 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
690 struct list_head *write_list)
692 unsigned int log_set = zmd->mblk_primary ^ 0x1;
695 /* Write dirty blocks to the log */
696 ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
701 * No error so far: now validate the log by updating the
702 * log index super block generation.
704 ret = dmz_write_sb(zmd, log_set);
712 * Flush dirty metadata blocks.
714 int dmz_flush_metadata(struct dmz_metadata *zmd)
716 struct dmz_mblock *mblk;
717 struct list_head write_list;
723 INIT_LIST_HEAD(&write_list);
726 * Make sure that metadata blocks are stable before logging: take
727 * the write lock on the metadata semaphore to prevent target BIOs
728 * from modifying metadata.
730 down_write(&zmd->mblk_sem);
733 * This is called from the target flush work and reclaim work.
734 * Concurrent execution is not allowed.
738 /* Get dirty blocks */
739 spin_lock(&zmd->mblk_lock);
740 list_splice_init(&zmd->mblk_dirty_list, &write_list);
741 spin_unlock(&zmd->mblk_lock);
743 /* If there are no dirty metadata blocks, just flush the device cache */
744 if (list_empty(&write_list)) {
745 ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
750 * The primary metadata set is still clean. Keep it this way until
751 * all updates are successful in the secondary set. That is, use
752 * the secondary set as a log.
754 ret = dmz_log_dirty_mblocks(zmd, &write_list);
759 * The log is on disk. It is now safe to update in place
760 * in the primary metadata set.
762 ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
766 ret = dmz_write_sb(zmd, zmd->mblk_primary);
770 while (!list_empty(&write_list)) {
771 mblk = list_first_entry(&write_list, struct dmz_mblock, link);
772 list_del_init(&mblk->link);
774 spin_lock(&zmd->mblk_lock);
775 clear_bit(DMZ_META_DIRTY, &mblk->state);
777 list_add_tail(&mblk->link, &zmd->mblk_lru_list);
778 spin_unlock(&zmd->mblk_lock);
783 if (ret && !list_empty(&write_list)) {
784 spin_lock(&zmd->mblk_lock);
785 list_splice(&write_list, &zmd->mblk_dirty_list);
786 spin_unlock(&zmd->mblk_lock);
789 dmz_unlock_flush(zmd);
790 up_write(&zmd->mblk_sem);
798 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
800 unsigned int nr_meta_zones, nr_data_zones;
801 struct dmz_dev *dev = zmd->dev;
805 gen = le64_to_cpu(sb->gen);
806 stored_crc = le32_to_cpu(sb->crc);
808 crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
809 if (crc != stored_crc) {
810 dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
815 if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
816 dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
817 DMZ_MAGIC, le32_to_cpu(sb->magic));
821 if (le32_to_cpu(sb->version) != DMZ_META_VER) {
822 dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
823 DMZ_META_VER, le32_to_cpu(sb->version));
827 nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
828 >> dev->zone_nr_blocks_shift;
829 if (!nr_meta_zones ||
830 nr_meta_zones >= zmd->nr_rnd_zones) {
831 dmz_dev_err(dev, "Invalid number of metadata blocks");
835 if (!le32_to_cpu(sb->nr_reserved_seq) ||
836 le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
837 dmz_dev_err(dev, "Invalid number of reserved sequential zones");
841 nr_data_zones = zmd->nr_useable_zones -
842 (nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
843 if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
844 dmz_dev_err(dev, "Invalid number of chunks %u / %u",
845 le32_to_cpu(sb->nr_chunks), nr_data_zones);
850 zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
851 zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
852 zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
853 zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
854 zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
855 zmd->nr_meta_zones = nr_meta_zones;
856 zmd->nr_data_zones = nr_data_zones;
862 * Read the first or second super block from disk.
864 static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
866 return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
867 zmd->sb[set].mblk->page);
871 * Determine the position of the secondary super blocks on disk.
872 * This is used only if a corruption of the primary super block
875 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
877 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
878 struct dmz_mblock *mblk;
881 /* Allocate a block */
882 mblk = dmz_alloc_mblock(zmd, 0);
886 zmd->sb[1].mblk = mblk;
887 zmd->sb[1].sb = mblk->data;
889 /* Bad first super block: search for the second one */
890 zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
891 for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
892 if (dmz_read_sb(zmd, 1) != 0)
894 if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
896 zmd->sb[1].block += zone_nr_blocks;
899 dmz_free_mblock(zmd, mblk);
900 zmd->sb[1].mblk = NULL;
906 * Read the first or second super block from disk.
908 static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
910 struct dmz_mblock *mblk;
913 /* Allocate a block */
914 mblk = dmz_alloc_mblock(zmd, 0);
918 zmd->sb[set].mblk = mblk;
919 zmd->sb[set].sb = mblk->data;
921 /* Read super block */
922 ret = dmz_read_sb(zmd, set);
924 dmz_free_mblock(zmd, mblk);
925 zmd->sb[set].mblk = NULL;
933 * Recover a metadata set.
935 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
937 unsigned int src_set = dst_set ^ 0x1;
941 dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
944 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
946 zmd->sb[1].block = zmd->sb[0].block +
947 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
950 page = alloc_page(GFP_NOIO);
954 /* Copy metadata blocks */
955 for (i = 1; i < zmd->nr_meta_blocks; i++) {
956 ret = dmz_rdwr_block(zmd, REQ_OP_READ,
957 zmd->sb[src_set].block + i, page);
960 ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
961 zmd->sb[dst_set].block + i, page);
966 /* Finalize with the super block */
967 if (!zmd->sb[dst_set].mblk) {
968 zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
969 if (!zmd->sb[dst_set].mblk) {
973 zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
976 ret = dmz_write_sb(zmd, dst_set);
978 __free_pages(page, 0);
984 * Get super block from disk.
986 static int dmz_load_sb(struct dmz_metadata *zmd)
988 bool sb_good[2] = {false, false};
989 u64 sb_gen[2] = {0, 0};
992 /* Read and check the primary super block */
993 zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
994 ret = dmz_get_sb(zmd, 0);
996 dmz_dev_err(zmd->dev, "Read primary super block failed");
1000 ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1002 /* Read and check secondary super block */
1005 zmd->sb[1].block = zmd->sb[0].block +
1006 (zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1007 ret = dmz_get_sb(zmd, 1);
1009 ret = dmz_lookup_secondary_sb(zmd);
1012 dmz_dev_err(zmd->dev, "Read secondary super block failed");
1016 ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1020 /* Use highest generation sb first */
1021 if (!sb_good[0] && !sb_good[1]) {
1022 dmz_dev_err(zmd->dev, "No valid super block found");
1027 sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1029 ret = dmz_recover_mblocks(zmd, 0);
1032 sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1034 ret = dmz_recover_mblocks(zmd, 1);
1037 dmz_dev_err(zmd->dev, "Recovery failed");
1041 if (sb_gen[0] >= sb_gen[1]) {
1042 zmd->sb_gen = sb_gen[0];
1043 zmd->mblk_primary = 0;
1045 zmd->sb_gen = sb_gen[1];
1046 zmd->mblk_primary = 1;
1049 dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
1050 zmd->mblk_primary, zmd->sb_gen);
1056 * Initialize a zone descriptor.
1058 static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1059 struct blk_zone *blkz)
1061 struct dmz_dev *dev = zmd->dev;
1063 /* Ignore the eventual last runt (smaller) zone */
1064 if (blkz->len != dev->zone_nr_sectors) {
1065 if (blkz->start + blkz->len == dev->capacity)
1070 INIT_LIST_HEAD(&zone->link);
1071 atomic_set(&zone->refcount, 0);
1072 zone->chunk = DMZ_MAP_UNMAPPED;
1074 if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
1075 set_bit(DMZ_RND, &zone->flags);
1076 zmd->nr_rnd_zones++;
1077 } else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1078 blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1079 set_bit(DMZ_SEQ, &zone->flags);
1083 if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1084 set_bit(DMZ_OFFLINE, &zone->flags);
1085 else if (blkz->cond == BLK_ZONE_COND_READONLY)
1086 set_bit(DMZ_READ_ONLY, &zone->flags);
1088 if (dmz_is_rnd(zone))
1091 zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1093 if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
1094 zmd->nr_useable_zones++;
1095 if (dmz_is_rnd(zone)) {
1096 zmd->nr_rnd_zones++;
1097 if (!zmd->sb_zone) {
1098 /* Super block zone */
1099 zmd->sb_zone = zone;
1108 * Free zones descriptors.
1110 static void dmz_drop_zones(struct dmz_metadata *zmd)
1117 * The size of a zone report in number of zones.
1118 * This results in 4096*64B=256KB report zones commands.
1120 #define DMZ_REPORT_NR_ZONES 4096
1123 * Allocate and initialize zone descriptors using the zone
1124 * information from disk.
1126 static int dmz_init_zones(struct dmz_metadata *zmd)
1128 struct dmz_dev *dev = zmd->dev;
1129 struct dm_zone *zone;
1130 struct blk_zone *blkz;
1131 unsigned int nr_blkz;
1132 sector_t sector = 0;
1136 zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1137 zmd->zone_nr_bitmap_blocks = zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT;
1139 /* Allocate zone array */
1140 zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1144 dmz_dev_info(dev, "Using %zu B for zone information",
1145 sizeof(struct dm_zone) * dev->nr_zones);
1147 /* Get zone information */
1148 nr_blkz = DMZ_REPORT_NR_ZONES;
1149 blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1156 * Get zone information and initialize zone descriptors.
1157 * At the same time, determine where the super block
1158 * should be: first block of the first randomly writable
1162 while (sector < dev->capacity) {
1163 /* Get zone information */
1164 nr_blkz = DMZ_REPORT_NR_ZONES;
1165 ret = blkdev_report_zones(dev->bdev, sector, blkz,
1166 &nr_blkz, GFP_KERNEL);
1168 dmz_dev_err(dev, "Report zones failed %d", ret);
1175 /* Process report */
1176 for (i = 0; i < nr_blkz; i++) {
1177 ret = dmz_init_zone(zmd, zone, &blkz[i]);
1180 sector += dev->zone_nr_sectors;
1185 /* The entire zone configuration of the disk should now be known */
1186 if (sector < dev->capacity) {
1187 dmz_dev_err(dev, "Failed to get correct zone information");
1193 dmz_drop_zones(zmd);
1199 * Update a zone information.
1201 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1203 unsigned int nr_blkz = 1;
1204 struct blk_zone blkz;
1207 /* Get zone information from disk */
1208 ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1209 &blkz, &nr_blkz, GFP_NOIO);
1213 dmz_dev_err(zmd->dev, "Get zone %u report failed",
1218 clear_bit(DMZ_OFFLINE, &zone->flags);
1219 clear_bit(DMZ_READ_ONLY, &zone->flags);
1220 if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1221 set_bit(DMZ_OFFLINE, &zone->flags);
1222 else if (blkz.cond == BLK_ZONE_COND_READONLY)
1223 set_bit(DMZ_READ_ONLY, &zone->flags);
1225 if (dmz_is_seq(zone))
1226 zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1234 * Check a zone write pointer position when the zone is marked
1235 * with the sequential write error flag.
1237 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1238 struct dm_zone *zone)
1240 unsigned int wp = 0;
1243 wp = zone->wp_block;
1244 ret = dmz_update_zone(zmd, zone);
1248 dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1249 dmz_id(zmd, zone), zone->wp_block, wp);
1251 if (zone->wp_block < wp) {
1252 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1253 wp - zone->wp_block);
1259 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1261 return &zmd->zones[zone_id];
1265 * Reset a zone write pointer.
1267 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1272 * Ignore offline zones, read only zones,
1273 * and conventional zones.
1275 if (dmz_is_offline(zone) ||
1276 dmz_is_readonly(zone) ||
1280 if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1281 struct dmz_dev *dev = zmd->dev;
1283 ret = blkdev_reset_zones(dev->bdev,
1284 dmz_start_sect(zmd, zone),
1285 dev->zone_nr_sectors, GFP_NOIO);
1287 dmz_dev_err(dev, "Reset zone %u failed %d",
1288 dmz_id(zmd, zone), ret);
1293 /* Clear write error bit and rewind write pointer position */
1294 clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1300 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1303 * Initialize chunk mapping.
1305 static int dmz_load_mapping(struct dmz_metadata *zmd)
1307 struct dmz_dev *dev = zmd->dev;
1308 struct dm_zone *dzone, *bzone;
1309 struct dmz_mblock *dmap_mblk = NULL;
1310 struct dmz_map *dmap;
1311 unsigned int i = 0, e = 0, chunk = 0;
1312 unsigned int dzone_id;
1313 unsigned int bzone_id;
1315 /* Metadata block array for the chunk mapping table */
1316 zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1317 sizeof(struct dmz_mblk *), GFP_KERNEL);
1321 /* Get chunk mapping table blocks and initialize zone mapping */
1322 while (chunk < zmd->nr_chunks) {
1324 /* Get mapping block */
1325 dmap_mblk = dmz_get_mblock(zmd, i + 1);
1326 if (IS_ERR(dmap_mblk))
1327 return PTR_ERR(dmap_mblk);
1328 zmd->map_mblk[i] = dmap_mblk;
1329 dmap = (struct dmz_map *) dmap_mblk->data;
1334 /* Check data zone */
1335 dzone_id = le32_to_cpu(dmap[e].dzone_id);
1336 if (dzone_id == DMZ_MAP_UNMAPPED)
1339 if (dzone_id >= dev->nr_zones) {
1340 dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1345 dzone = dmz_get(zmd, dzone_id);
1346 set_bit(DMZ_DATA, &dzone->flags);
1347 dzone->chunk = chunk;
1348 dmz_get_zone_weight(zmd, dzone);
1350 if (dmz_is_rnd(dzone))
1351 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1353 list_add_tail(&dzone->link, &zmd->map_seq_list);
1355 /* Check buffer zone */
1356 bzone_id = le32_to_cpu(dmap[e].bzone_id);
1357 if (bzone_id == DMZ_MAP_UNMAPPED)
1360 if (bzone_id >= dev->nr_zones) {
1361 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1366 bzone = dmz_get(zmd, bzone_id);
1367 if (!dmz_is_rnd(bzone)) {
1368 dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
1373 set_bit(DMZ_DATA, &bzone->flags);
1374 set_bit(DMZ_BUF, &bzone->flags);
1375 bzone->chunk = chunk;
1376 bzone->bzone = dzone;
1377 dzone->bzone = bzone;
1378 dmz_get_zone_weight(zmd, bzone);
1379 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1383 if (e >= DMZ_MAP_ENTRIES)
1388 * At this point, only meta zones and mapped data zones were
1389 * fully initialized. All remaining zones are unmapped data
1390 * zones. Finish initializing those here.
1392 for (i = 0; i < dev->nr_zones; i++) {
1393 dzone = dmz_get(zmd, i);
1394 if (dmz_is_meta(dzone))
1397 if (dmz_is_rnd(dzone))
1402 if (dmz_is_data(dzone)) {
1403 /* Already initialized */
1407 /* Unmapped data zone */
1408 set_bit(DMZ_DATA, &dzone->flags);
1409 dzone->chunk = DMZ_MAP_UNMAPPED;
1410 if (dmz_is_rnd(dzone)) {
1411 list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1412 atomic_inc(&zmd->unmap_nr_rnd);
1413 } else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1414 list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1415 atomic_inc(&zmd->nr_reserved_seq_zones);
1418 list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1419 atomic_inc(&zmd->unmap_nr_seq);
1427 * Set a data chunk mapping.
1429 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1430 unsigned int dzone_id, unsigned int bzone_id)
1432 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1433 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1434 int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1436 dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1437 dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1438 dmz_dirty_mblock(zmd, dmap_mblk);
1442 * The list of mapped zones is maintained in LRU order.
1443 * This rotates a zone at the end of its map list.
1445 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1447 if (list_empty(&zone->link))
1450 list_del_init(&zone->link);
1451 if (dmz_is_seq(zone)) {
1452 /* LRU rotate sequential zone */
1453 list_add_tail(&zone->link, &zmd->map_seq_list);
1455 /* LRU rotate random zone */
1456 list_add_tail(&zone->link, &zmd->map_rnd_list);
1461 * The list of mapped random zones is maintained
1462 * in LRU order. This rotates a zone at the end of the list.
1464 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1466 __dmz_lru_zone(zmd, zone);
1468 __dmz_lru_zone(zmd, zone->bzone);
1472 * Wait for any zone to be freed.
1474 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1478 prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1479 dmz_unlock_map(zmd);
1480 dmz_unlock_metadata(zmd);
1482 io_schedule_timeout(HZ);
1484 dmz_lock_metadata(zmd);
1486 finish_wait(&zmd->free_wq, &wait);
1490 * Lock a zone for reclaim (set the zone RECLAIM bit).
1491 * Returns false if the zone cannot be locked or if it is already locked
1494 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1496 /* Active zones cannot be reclaimed */
1497 if (dmz_is_active(zone))
1500 return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1504 * Clear a zone reclaim flag.
1506 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1508 WARN_ON(dmz_is_active(zone));
1509 WARN_ON(!dmz_in_reclaim(zone));
1511 clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1512 smp_mb__after_atomic();
1513 wake_up_bit(&zone->flags, DMZ_RECLAIM);
1517 * Wait for a zone reclaim to complete.
1519 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1521 dmz_unlock_map(zmd);
1522 dmz_unlock_metadata(zmd);
1523 wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1524 dmz_lock_metadata(zmd);
1529 * Select a random write zone for reclaim.
1531 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
1533 struct dm_zone *dzone = NULL;
1534 struct dm_zone *zone;
1536 if (list_empty(&zmd->map_rnd_list))
1539 list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1540 if (dmz_is_buf(zone))
1541 dzone = zone->bzone;
1544 if (dmz_lock_zone_reclaim(dzone))
1552 * Select a buffered sequential zone for reclaim.
1554 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
1556 struct dm_zone *zone;
1558 if (list_empty(&zmd->map_seq_list))
1561 list_for_each_entry(zone, &zmd->map_seq_list, link) {
1564 if (dmz_lock_zone_reclaim(zone))
1572 * Select a zone for reclaim.
1574 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
1576 struct dm_zone *zone;
1579 * Search for a zone candidate to reclaim: 2 cases are possible.
1580 * (1) There is no free sequential zones. Then a random data zone
1581 * cannot be reclaimed. So choose a sequential zone to reclaim so
1582 * that afterward a random zone can be reclaimed.
1583 * (2) At least one free sequential zone is available, then choose
1584 * the oldest random zone (data or buffer) that can be locked.
1587 if (list_empty(&zmd->reserved_seq_zones_list))
1588 zone = dmz_get_seq_zone_for_reclaim(zmd);
1590 zone = dmz_get_rnd_zone_for_reclaim(zmd);
1591 dmz_unlock_map(zmd);
1597 * Activate a zone (increment its reference count).
1599 void dmz_activate_zone(struct dm_zone *zone)
1601 set_bit(DMZ_ACTIVE, &zone->flags);
1602 atomic_inc(&zone->refcount);
1606 * Deactivate a zone. This decrement the zone reference counter
1607 * and clears the active state of the zone once the count reaches 0,
1608 * indicating that all BIOs to the zone have completed. Returns
1609 * true if the zone was deactivated.
1611 void dmz_deactivate_zone(struct dm_zone *zone)
1613 if (atomic_dec_and_test(&zone->refcount)) {
1614 WARN_ON(!test_bit(DMZ_ACTIVE, &zone->flags));
1615 clear_bit_unlock(DMZ_ACTIVE, &zone->flags);
1616 smp_mb__after_atomic();
1621 * Get the zone mapping a chunk, if the chunk is mapped already.
1622 * If no mapping exist and the operation is WRITE, a zone is
1623 * allocated and used to map the chunk.
1624 * The zone returned will be set to the active state.
1626 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
1628 struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1629 struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1630 int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1631 unsigned int dzone_id;
1632 struct dm_zone *dzone = NULL;
1637 /* Get the chunk mapping */
1638 dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1639 if (dzone_id == DMZ_MAP_UNMAPPED) {
1641 * Read or discard in unmapped chunks are fine. But for
1642 * writes, we need a mapping, so get one.
1644 if (op != REQ_OP_WRITE)
1647 /* Alloate a random zone */
1648 dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1650 dmz_wait_for_free_zones(zmd);
1654 dmz_map_zone(zmd, dzone, chunk);
1657 /* The chunk is already mapped: get the mapping zone */
1658 dzone = dmz_get(zmd, dzone_id);
1659 if (dzone->chunk != chunk) {
1660 dzone = ERR_PTR(-EIO);
1664 /* Repair write pointer if the sequential dzone has error */
1665 if (dmz_seq_write_err(dzone)) {
1666 ret = dmz_handle_seq_write_err(zmd, dzone);
1668 dzone = ERR_PTR(-EIO);
1671 clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1676 * If the zone is being reclaimed, the chunk mapping may change
1677 * to a different zone. So wait for reclaim and retry. Otherwise,
1678 * activate the zone (this will prevent reclaim from touching it).
1680 if (dmz_in_reclaim(dzone)) {
1681 dmz_wait_for_reclaim(zmd, dzone);
1684 dmz_activate_zone(dzone);
1685 dmz_lru_zone(zmd, dzone);
1687 dmz_unlock_map(zmd);
1693 * Write and discard change the block validity of data zones and their buffer
1694 * zones. Check here that valid blocks are still present. If all blocks are
1695 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1698 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1700 struct dm_zone *bzone;
1704 bzone = dzone->bzone;
1706 if (dmz_weight(bzone))
1707 dmz_lru_zone(zmd, bzone);
1709 /* Empty buffer zone: reclaim it */
1710 dmz_unmap_zone(zmd, bzone);
1711 dmz_free_zone(zmd, bzone);
1716 /* Deactivate the data zone */
1717 dmz_deactivate_zone(dzone);
1718 if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1719 dmz_lru_zone(zmd, dzone);
1721 /* Unbuffered inactive empty data zone: reclaim it */
1722 dmz_unmap_zone(zmd, dzone);
1723 dmz_free_zone(zmd, dzone);
1726 dmz_unlock_map(zmd);
1730 * Allocate and map a random zone to buffer a chunk
1731 * already mapped to a sequential zone.
1733 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1734 struct dm_zone *dzone)
1736 struct dm_zone *bzone;
1740 bzone = dzone->bzone;
1744 /* Alloate a random zone */
1745 bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1747 dmz_wait_for_free_zones(zmd);
1751 /* Update the chunk mapping */
1752 dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1753 dmz_id(zmd, bzone));
1755 set_bit(DMZ_BUF, &bzone->flags);
1756 bzone->chunk = dzone->chunk;
1757 bzone->bzone = dzone;
1758 dzone->bzone = bzone;
1759 list_add_tail(&bzone->link, &zmd->map_rnd_list);
1761 dmz_unlock_map(zmd);
1767 * Get an unmapped (free) zone.
1768 * This must be called with the mapping lock held.
1770 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
1772 struct list_head *list;
1773 struct dm_zone *zone;
1775 if (flags & DMZ_ALLOC_RND)
1776 list = &zmd->unmap_rnd_list;
1778 list = &zmd->unmap_seq_list;
1780 if (list_empty(list)) {
1782 * No free zone: if this is for reclaim, allow using the
1783 * reserved sequential zones.
1785 if (!(flags & DMZ_ALLOC_RECLAIM) ||
1786 list_empty(&zmd->reserved_seq_zones_list))
1789 zone = list_first_entry(&zmd->reserved_seq_zones_list,
1790 struct dm_zone, link);
1791 list_del_init(&zone->link);
1792 atomic_dec(&zmd->nr_reserved_seq_zones);
1796 zone = list_first_entry(list, struct dm_zone, link);
1797 list_del_init(&zone->link);
1799 if (dmz_is_rnd(zone))
1800 atomic_dec(&zmd->unmap_nr_rnd);
1802 atomic_dec(&zmd->unmap_nr_seq);
1804 if (dmz_is_offline(zone)) {
1805 dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
1815 * This must be called with the mapping lock held.
1817 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1819 /* If this is a sequential zone, reset it */
1820 if (dmz_is_seq(zone))
1821 dmz_reset_zone(zmd, zone);
1823 /* Return the zone to its type unmap list */
1824 if (dmz_is_rnd(zone)) {
1825 list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1826 atomic_inc(&zmd->unmap_nr_rnd);
1827 } else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1828 zmd->nr_reserved_seq) {
1829 list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1830 atomic_inc(&zmd->nr_reserved_seq_zones);
1832 list_add_tail(&zone->link, &zmd->unmap_seq_list);
1833 atomic_inc(&zmd->unmap_nr_seq);
1836 wake_up_all(&zmd->free_wq);
1840 * Map a chunk to a zone.
1841 * This must be called with the mapping lock held.
1843 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1846 /* Set the chunk mapping */
1847 dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1849 dzone->chunk = chunk;
1850 if (dmz_is_rnd(dzone))
1851 list_add_tail(&dzone->link, &zmd->map_rnd_list);
1853 list_add_tail(&dzone->link, &zmd->map_seq_list);
1858 * This must be called with the mapping lock held.
1860 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1862 unsigned int chunk = zone->chunk;
1863 unsigned int dzone_id;
1865 if (chunk == DMZ_MAP_UNMAPPED) {
1866 /* Already unmapped */
1870 if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1872 * Unmapping the chunk buffer zone: clear only
1873 * the chunk buffer mapping
1875 dzone_id = dmz_id(zmd, zone->bzone);
1876 zone->bzone->bzone = NULL;
1881 * Unmapping the chunk data zone: the zone must
1884 if (WARN_ON(zone->bzone)) {
1885 zone->bzone->bzone = NULL;
1888 dzone_id = DMZ_MAP_UNMAPPED;
1891 dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1893 zone->chunk = DMZ_MAP_UNMAPPED;
1894 list_del_init(&zone->link);
1898 * Set @nr_bits bits in @bitmap starting from @bit.
1899 * Return the number of bits changed from 0 to 1.
1901 static unsigned int dmz_set_bits(unsigned long *bitmap,
1902 unsigned int bit, unsigned int nr_bits)
1904 unsigned long *addr;
1905 unsigned int end = bit + nr_bits;
1909 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1910 ((end - bit) >= BITS_PER_LONG)) {
1911 /* Try to set the whole word at once */
1912 addr = bitmap + BIT_WORD(bit);
1916 bit += BITS_PER_LONG;
1921 if (!test_and_set_bit(bit, bitmap))
1930 * Get the bitmap block storing the bit for chunk_block in zone.
1932 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1933 struct dm_zone *zone,
1934 sector_t chunk_block)
1936 sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1937 (sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1938 (chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1940 return dmz_get_mblock(zmd, bitmap_block);
1944 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1946 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1947 struct dm_zone *to_zone)
1949 struct dmz_mblock *from_mblk, *to_mblk;
1950 sector_t chunk_block = 0;
1952 /* Get the zones bitmap blocks */
1953 while (chunk_block < zmd->dev->zone_nr_blocks) {
1954 from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1955 if (IS_ERR(from_mblk))
1956 return PTR_ERR(from_mblk);
1957 to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1958 if (IS_ERR(to_mblk)) {
1959 dmz_release_mblock(zmd, from_mblk);
1960 return PTR_ERR(to_mblk);
1963 memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1964 dmz_dirty_mblock(zmd, to_mblk);
1966 dmz_release_mblock(zmd, to_mblk);
1967 dmz_release_mblock(zmd, from_mblk);
1969 chunk_block += DMZ_BLOCK_SIZE_BITS;
1972 to_zone->weight = from_zone->weight;
1978 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1979 * starting from chunk_block.
1981 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1982 struct dm_zone *to_zone, sector_t chunk_block)
1984 unsigned int nr_blocks;
1987 /* Get the zones bitmap blocks */
1988 while (chunk_block < zmd->dev->zone_nr_blocks) {
1989 /* Get a valid region from the source zone */
1990 ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
1995 ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
1999 chunk_block += nr_blocks;
2006 * Validate all the blocks in the range [block..block+nr_blocks-1].
2008 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2009 sector_t chunk_block, unsigned int nr_blocks)
2011 unsigned int count, bit, nr_bits;
2012 unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2013 struct dmz_mblock *mblk;
2016 dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2017 dmz_id(zmd, zone), (unsigned long long)chunk_block,
2020 WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2023 /* Get bitmap block */
2024 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2026 return PTR_ERR(mblk);
2029 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2030 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2032 count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2034 dmz_dirty_mblock(zmd, mblk);
2037 dmz_release_mblock(zmd, mblk);
2039 nr_blocks -= nr_bits;
2040 chunk_block += nr_bits;
2043 if (likely(zone->weight + n <= zone_nr_blocks))
2046 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2047 dmz_id(zmd, zone), zone->weight,
2048 zone_nr_blocks - n);
2049 zone->weight = zone_nr_blocks;
2056 * Clear nr_bits bits in bitmap starting from bit.
2057 * Return the number of bits cleared.
2059 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2061 unsigned long *addr;
2062 int end = bit + nr_bits;
2066 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2067 ((end - bit) >= BITS_PER_LONG)) {
2068 /* Try to clear whole word at once */
2069 addr = bitmap + BIT_WORD(bit);
2070 if (*addr == ULONG_MAX) {
2073 bit += BITS_PER_LONG;
2078 if (test_and_clear_bit(bit, bitmap))
2087 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2089 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2090 sector_t chunk_block, unsigned int nr_blocks)
2092 unsigned int count, bit, nr_bits;
2093 struct dmz_mblock *mblk;
2096 dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2097 dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2099 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2102 /* Get bitmap block */
2103 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2105 return PTR_ERR(mblk);
2108 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2109 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2111 count = dmz_clear_bits((unsigned long *)mblk->data,
2114 dmz_dirty_mblock(zmd, mblk);
2117 dmz_release_mblock(zmd, mblk);
2119 nr_blocks -= nr_bits;
2120 chunk_block += nr_bits;
2123 if (zone->weight >= n)
2126 dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2127 dmz_id(zmd, zone), zone->weight, n);
2135 * Get a block bit value.
2137 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2138 sector_t chunk_block)
2140 struct dmz_mblock *mblk;
2143 WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2145 /* Get bitmap block */
2146 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2148 return PTR_ERR(mblk);
2151 ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2152 (unsigned long *) mblk->data) != 0;
2154 dmz_release_mblock(zmd, mblk);
2160 * Return the number of blocks from chunk_block to the first block with a bit
2161 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2163 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2164 sector_t chunk_block, unsigned int nr_blocks,
2167 struct dmz_mblock *mblk;
2168 unsigned int bit, set_bit, nr_bits;
2169 unsigned long *bitmap;
2172 WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2175 /* Get bitmap block */
2176 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2178 return PTR_ERR(mblk);
2181 bitmap = (unsigned long *) mblk->data;
2182 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2183 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2185 set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2187 set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2188 dmz_release_mblock(zmd, mblk);
2191 if (set_bit < DMZ_BLOCK_SIZE_BITS)
2194 nr_blocks -= nr_bits;
2195 chunk_block += nr_bits;
2202 * Test if chunk_block is valid. If it is, the number of consecutive
2203 * valid blocks from chunk_block will be returned.
2205 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2206 sector_t chunk_block)
2210 valid = dmz_test_block(zmd, zone, chunk_block);
2214 /* The block is valid: get the number of valid blocks from block */
2215 return dmz_to_next_set_block(zmd, zone, chunk_block,
2216 zmd->dev->zone_nr_blocks - chunk_block, 0);
2220 * Find the first valid block from @chunk_block in @zone.
2221 * If such a block is found, its number is returned using
2222 * @chunk_block and the total number of valid blocks from @chunk_block
2225 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2226 sector_t *chunk_block)
2228 sector_t start_block = *chunk_block;
2231 ret = dmz_to_next_set_block(zmd, zone, start_block,
2232 zmd->dev->zone_nr_blocks - start_block, 1);
2237 *chunk_block = start_block;
2239 return dmz_to_next_set_block(zmd, zone, start_block,
2240 zmd->dev->zone_nr_blocks - start_block, 0);
2244 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2246 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2248 unsigned long *addr;
2249 int end = bit + nr_bits;
2253 if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2254 ((end - bit) >= BITS_PER_LONG)) {
2255 addr = (unsigned long *)bitmap + BIT_WORD(bit);
2256 if (*addr == ULONG_MAX) {
2258 bit += BITS_PER_LONG;
2263 if (test_bit(bit, bitmap))
2272 * Get a zone weight.
2274 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2276 struct dmz_mblock *mblk;
2277 sector_t chunk_block = 0;
2278 unsigned int bit, nr_bits;
2279 unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2284 /* Get bitmap block */
2285 mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2291 /* Count bits in this block */
2292 bitmap = mblk->data;
2293 bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2294 nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2295 n += dmz_count_bits(bitmap, bit, nr_bits);
2297 dmz_release_mblock(zmd, mblk);
2299 nr_blocks -= nr_bits;
2300 chunk_block += nr_bits;
2307 * Cleanup the zoned metadata resources.
2309 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2311 struct rb_root *root;
2312 struct dmz_mblock *mblk, *next;
2315 /* Release zone mapping resources */
2316 if (zmd->map_mblk) {
2317 for (i = 0; i < zmd->nr_map_blocks; i++)
2318 dmz_release_mblock(zmd, zmd->map_mblk[i]);
2319 kfree(zmd->map_mblk);
2320 zmd->map_mblk = NULL;
2323 /* Release super blocks */
2324 for (i = 0; i < 2; i++) {
2325 if (zmd->sb[i].mblk) {
2326 dmz_free_mblock(zmd, zmd->sb[i].mblk);
2327 zmd->sb[i].mblk = NULL;
2331 /* Free cached blocks */
2332 while (!list_empty(&zmd->mblk_dirty_list)) {
2333 mblk = list_first_entry(&zmd->mblk_dirty_list,
2334 struct dmz_mblock, link);
2335 dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2336 (u64)mblk->no, mblk->ref);
2337 list_del_init(&mblk->link);
2338 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2339 dmz_free_mblock(zmd, mblk);
2342 while (!list_empty(&zmd->mblk_lru_list)) {
2343 mblk = list_first_entry(&zmd->mblk_lru_list,
2344 struct dmz_mblock, link);
2345 list_del_init(&mblk->link);
2346 rb_erase(&mblk->node, &zmd->mblk_rbtree);
2347 dmz_free_mblock(zmd, mblk);
2350 /* Sanity checks: the mblock rbtree should now be empty */
2351 root = &zmd->mblk_rbtree;
2352 rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2353 dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2354 (u64)mblk->no, mblk->ref);
2356 dmz_free_mblock(zmd, mblk);
2359 /* Free the zone descriptors */
2360 dmz_drop_zones(zmd);
2362 mutex_destroy(&zmd->mblk_flush_lock);
2363 mutex_destroy(&zmd->map_lock);
2367 * Initialize the zoned metadata.
2369 int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
2371 struct dmz_metadata *zmd;
2372 unsigned int i, zid;
2373 struct dm_zone *zone;
2376 zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2381 zmd->mblk_rbtree = RB_ROOT;
2382 init_rwsem(&zmd->mblk_sem);
2383 mutex_init(&zmd->mblk_flush_lock);
2384 spin_lock_init(&zmd->mblk_lock);
2385 INIT_LIST_HEAD(&zmd->mblk_lru_list);
2386 INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2388 mutex_init(&zmd->map_lock);
2389 atomic_set(&zmd->unmap_nr_rnd, 0);
2390 INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2391 INIT_LIST_HEAD(&zmd->map_rnd_list);
2393 atomic_set(&zmd->unmap_nr_seq, 0);
2394 INIT_LIST_HEAD(&zmd->unmap_seq_list);
2395 INIT_LIST_HEAD(&zmd->map_seq_list);
2397 atomic_set(&zmd->nr_reserved_seq_zones, 0);
2398 INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2400 init_waitqueue_head(&zmd->free_wq);
2402 /* Initialize zone descriptors */
2403 ret = dmz_init_zones(zmd);
2407 /* Get super block */
2408 ret = dmz_load_sb(zmd);
2412 /* Set metadata zones starting from sb_zone */
2413 zid = dmz_id(zmd, zmd->sb_zone);
2414 for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2415 zone = dmz_get(zmd, zid + i);
2416 if (!dmz_is_rnd(zone))
2418 set_bit(DMZ_META, &zone->flags);
2421 /* Load mapping table */
2422 ret = dmz_load_mapping(zmd);
2427 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2428 * blocks and enough blocks to be able to cache the bitmap blocks of
2429 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2430 * the cache to add 512 more metadata blocks.
2432 zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2433 zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2434 zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2435 zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2436 zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2438 /* Metadata cache shrinker */
2439 ret = register_shrinker(&zmd->mblk_shrinker);
2441 dmz_dev_err(dev, "Register metadata cache shrinker failed");
2445 dmz_dev_info(dev, "Host-%s zoned block device",
2446 bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2447 "aware" : "managed");
2448 dmz_dev_info(dev, " %llu 512-byte logical sectors",
2449 (u64)dev->capacity);
2450 dmz_dev_info(dev, " %u zones of %llu 512-byte logical sectors",
2451 dev->nr_zones, (u64)dev->zone_nr_sectors);
2452 dmz_dev_info(dev, " %u metadata zones",
2453 zmd->nr_meta_zones * 2);
2454 dmz_dev_info(dev, " %u data zones for %u chunks",
2455 zmd->nr_data_zones, zmd->nr_chunks);
2456 dmz_dev_info(dev, " %u random zones (%u unmapped)",
2457 zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2458 dmz_dev_info(dev, " %u sequential zones (%u unmapped)",
2459 zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2460 dmz_dev_info(dev, " %u reserved sequential data zones",
2461 zmd->nr_reserved_seq);
2463 dmz_dev_debug(dev, "Format:");
2464 dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
2465 zmd->nr_meta_blocks, zmd->max_nr_mblks);
2466 dmz_dev_debug(dev, " %u data zone mapping blocks",
2467 zmd->nr_map_blocks);
2468 dmz_dev_debug(dev, " %u bitmap blocks",
2469 zmd->nr_bitmap_blocks);
2475 dmz_cleanup_metadata(zmd);
2483 * Cleanup the zoned metadata resources.
2485 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2487 unregister_shrinker(&zmd->mblk_shrinker);
2488 dmz_cleanup_metadata(zmd);
2493 * Check zone information on resume.
2495 int dmz_resume_metadata(struct dmz_metadata *zmd)
2497 struct dmz_dev *dev = zmd->dev;
2498 struct dm_zone *zone;
2504 for (i = 0; i < dev->nr_zones; i++) {
2505 zone = dmz_get(zmd, i);
2507 dmz_dev_err(dev, "Unable to get zone %u", i);
2511 wp_block = zone->wp_block;
2513 ret = dmz_update_zone(zmd, zone);
2515 dmz_dev_err(dev, "Broken zone %u", i);
2519 if (dmz_is_offline(zone)) {
2520 dmz_dev_warn(dev, "Zone %u is offline", i);
2524 /* Check write pointer */
2525 if (!dmz_is_seq(zone))
2527 else if (zone->wp_block != wp_block) {
2528 dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2529 i, (u64)zone->wp_block, (u64)wp_block);
2530 zone->wp_block = wp_block;
2531 dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2532 dev->zone_nr_blocks - zone->wp_block);