1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017 Western Digital Corporation or its affiliates.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-zoned.h"
9 
10 #include <linux/module.h>
11 #include <linux/crc32.h>
12 #include <linux/sched/mm.h>
13 
14 #define	DM_MSG_PREFIX		"zoned metadata"
15 
16 /*
17  * Metadata version.
18  */
19 #define DMZ_META_VER	2
20 
21 /*
22  * On-disk super block magic.
23  */
24 #define DMZ_MAGIC	((((unsigned int)('D')) << 24) | \
25 			 (((unsigned int)('Z')) << 16) | \
26 			 (((unsigned int)('B')) <<  8) | \
27 			 ((unsigned int)('D')))
28 
29 /*
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.
40  */
41 struct dmz_super {
42 	/* Magic number */
43 	__le32		magic;			/*   4 */
44 
45 	/* Metadata version number */
46 	__le32		version;		/*   8 */
47 
48 	/* Generation number */
49 	__le64		gen;			/*  16 */
50 
51 	/* This block number */
52 	__le64		sb_block;		/*  24 */
53 
54 	/* The number of metadata blocks, including this super block */
55 	__le32		nr_meta_blocks;		/*  28 */
56 
57 	/* The number of sequential zones reserved for reclaim */
58 	__le32		nr_reserved_seq;	/*  32 */
59 
60 	/* The number of entries in the mapping table */
61 	__le32		nr_chunks;		/*  36 */
62 
63 	/* The number of blocks used for the chunk mapping table */
64 	__le32		nr_map_blocks;		/*  40 */
65 
66 	/* The number of blocks used for the block bitmaps */
67 	__le32		nr_bitmap_blocks;	/*  44 */
68 
69 	/* Checksum */
70 	__le32		crc;			/*  48 */
71 
72 	/* DM-Zoned label */
73 	u8		dmz_label[32];		/*  80 */
74 
75 	/* DM-Zoned UUID */
76 	u8		dmz_uuid[16];		/*  96 */
77 
78 	/* Device UUID */
79 	u8		dev_uuid[16];		/* 112 */
80 
81 	/* Padding to full 512B sector */
82 	u8		reserved[400];		/* 512 */
83 };
84 
85 /*
86  * Chunk mapping entry: entries are indexed by chunk number
87  * and give the zone ID (dzone_id) mapping the chunk on disk.
88  * This zone may be sequential or random. If it is a sequential
89  * zone, a second zone (bzone_id) used as a write buffer may
90  * also be specified. This second zone will always be a randomly
91  * writeable zone.
92  */
93 struct dmz_map {
94 	__le32			dzone_id;
95 	__le32			bzone_id;
96 };
97 
98 /*
99  * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
100  */
101 #define DMZ_MAP_ENTRIES		(DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
102 #define DMZ_MAP_ENTRIES_SHIFT	(ilog2(DMZ_MAP_ENTRIES))
103 #define DMZ_MAP_ENTRIES_MASK	(DMZ_MAP_ENTRIES - 1)
104 #define DMZ_MAP_UNMAPPED	UINT_MAX
105 
106 /*
107  * Meta data block descriptor (for cached metadata blocks).
108  */
109 struct dmz_mblock {
110 	struct rb_node		node;
111 	struct list_head	link;
112 	sector_t		no;
113 	unsigned int		ref;
114 	unsigned long		state;
115 	struct page		*page;
116 	void			*data;
117 };
118 
119 /*
120  * Metadata block state flags.
121  */
122 enum {
123 	DMZ_META_DIRTY,
124 	DMZ_META_READING,
125 	DMZ_META_WRITING,
126 	DMZ_META_ERROR,
127 };
128 
129 /*
130  * Super block information (one per metadata set).
131  */
132 struct dmz_sb {
133 	sector_t		block;
134 	struct dmz_dev		*dev;
135 	struct dmz_mblock	*mblk;
136 	struct dmz_super	*sb;
137 	struct dm_zone		*zone;
138 };
139 
140 /*
141  * In-memory metadata.
142  */
143 struct dmz_metadata {
144 	struct dmz_dev		*dev;
145 	unsigned int		nr_devs;
146 
147 	char			devname[BDEVNAME_SIZE];
148 	char			label[BDEVNAME_SIZE];
149 	uuid_t			uuid;
150 
151 	sector_t		zone_bitmap_size;
152 	unsigned int		zone_nr_bitmap_blocks;
153 	unsigned int		zone_bits_per_mblk;
154 
155 	sector_t		zone_nr_blocks;
156 	sector_t		zone_nr_blocks_shift;
157 
158 	sector_t		zone_nr_sectors;
159 	sector_t		zone_nr_sectors_shift;
160 
161 	unsigned int		nr_bitmap_blocks;
162 	unsigned int		nr_map_blocks;
163 
164 	unsigned int		nr_zones;
165 	unsigned int		nr_useable_zones;
166 	unsigned int		nr_meta_blocks;
167 	unsigned int		nr_meta_zones;
168 	unsigned int		nr_data_zones;
169 	unsigned int		nr_cache_zones;
170 	unsigned int		nr_rnd_zones;
171 	unsigned int		nr_reserved_seq;
172 	unsigned int		nr_chunks;
173 
174 	/* Zone information array */
175 	struct xarray		zones;
176 
177 	struct dmz_sb		sb[2];
178 	unsigned int		mblk_primary;
179 	unsigned int		sb_version;
180 	u64			sb_gen;
181 	unsigned int		min_nr_mblks;
182 	unsigned int		max_nr_mblks;
183 	atomic_t		nr_mblks;
184 	struct rw_semaphore	mblk_sem;
185 	struct mutex		mblk_flush_lock;
186 	spinlock_t		mblk_lock;
187 	struct rb_root		mblk_rbtree;
188 	struct list_head	mblk_lru_list;
189 	struct list_head	mblk_dirty_list;
190 	struct shrinker		mblk_shrinker;
191 
192 	/* Zone allocation management */
193 	struct mutex		map_lock;
194 	struct dmz_mblock	**map_mblk;
195 
196 	unsigned int		nr_cache;
197 	atomic_t		unmap_nr_cache;
198 	struct list_head	unmap_cache_list;
199 	struct list_head	map_cache_list;
200 
201 	atomic_t		nr_reserved_seq_zones;
202 	struct list_head	reserved_seq_zones_list;
203 
204 	wait_queue_head_t	free_wq;
205 };
206 
207 #define dmz_zmd_info(zmd, format, args...)	\
208 	DMINFO("(%s): " format, (zmd)->label, ## args)
209 
210 #define dmz_zmd_err(zmd, format, args...)	\
211 	DMERR("(%s): " format, (zmd)->label, ## args)
212 
213 #define dmz_zmd_warn(zmd, format, args...)	\
214 	DMWARN("(%s): " format, (zmd)->label, ## args)
215 
216 #define dmz_zmd_debug(zmd, format, args...)	\
217 	DMDEBUG("(%s): " format, (zmd)->label, ## args)
218 /*
219  * Various accessors
220  */
221 static unsigned int dmz_dev_zone_id(struct dmz_metadata *zmd, struct dm_zone *zone)
222 {
223 	if (WARN_ON(!zone))
224 		return 0;
225 
226 	return zone->id - zone->dev->zone_offset;
227 }
228 
229 sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
230 {
231 	unsigned int zone_id = dmz_dev_zone_id(zmd, zone);
232 
233 	return (sector_t)zone_id << zmd->zone_nr_sectors_shift;
234 }
235 
236 sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
237 {
238 	unsigned int zone_id = dmz_dev_zone_id(zmd, zone);
239 
240 	return (sector_t)zone_id << zmd->zone_nr_blocks_shift;
241 }
242 
243 unsigned int dmz_zone_nr_blocks(struct dmz_metadata *zmd)
244 {
245 	return zmd->zone_nr_blocks;
246 }
247 
248 unsigned int dmz_zone_nr_blocks_shift(struct dmz_metadata *zmd)
249 {
250 	return zmd->zone_nr_blocks_shift;
251 }
252 
253 unsigned int dmz_zone_nr_sectors(struct dmz_metadata *zmd)
254 {
255 	return zmd->zone_nr_sectors;
256 }
257 
258 unsigned int dmz_zone_nr_sectors_shift(struct dmz_metadata *zmd)
259 {
260 	return zmd->zone_nr_sectors_shift;
261 }
262 
263 unsigned int dmz_nr_zones(struct dmz_metadata *zmd)
264 {
265 	return zmd->nr_zones;
266 }
267 
268 unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
269 {
270 	return zmd->nr_chunks;
271 }
272 
273 unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd, int idx)
274 {
275 	return zmd->dev[idx].nr_rnd;
276 }
277 
278 unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd, int idx)
279 {
280 	return atomic_read(&zmd->dev[idx].unmap_nr_rnd);
281 }
282 
283 unsigned int dmz_nr_cache_zones(struct dmz_metadata *zmd)
284 {
285 	return zmd->nr_cache;
286 }
287 
288 unsigned int dmz_nr_unmap_cache_zones(struct dmz_metadata *zmd)
289 {
290 	return atomic_read(&zmd->unmap_nr_cache);
291 }
292 
293 unsigned int dmz_nr_seq_zones(struct dmz_metadata *zmd, int idx)
294 {
295 	return zmd->dev[idx].nr_seq;
296 }
297 
298 unsigned int dmz_nr_unmap_seq_zones(struct dmz_metadata *zmd, int idx)
299 {
300 	return atomic_read(&zmd->dev[idx].unmap_nr_seq);
301 }
302 
303 static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
304 {
305 	return xa_load(&zmd->zones, zone_id);
306 }
307 
308 static struct dm_zone *dmz_insert(struct dmz_metadata *zmd,
309 				  unsigned int zone_id, struct dmz_dev *dev)
310 {
311 	struct dm_zone *zone = kzalloc(sizeof(struct dm_zone), GFP_KERNEL);
312 
313 	if (!zone)
314 		return ERR_PTR(-ENOMEM);
315 
316 	if (xa_insert(&zmd->zones, zone_id, zone, GFP_KERNEL)) {
317 		kfree(zone);
318 		return ERR_PTR(-EBUSY);
319 	}
320 
321 	INIT_LIST_HEAD(&zone->link);
322 	atomic_set(&zone->refcount, 0);
323 	zone->id = zone_id;
324 	zone->chunk = DMZ_MAP_UNMAPPED;
325 	zone->dev = dev;
326 
327 	return zone;
328 }
329 
330 const char *dmz_metadata_label(struct dmz_metadata *zmd)
331 {
332 	return (const char *)zmd->label;
333 }
334 
335 bool dmz_check_dev(struct dmz_metadata *zmd)
336 {
337 	unsigned int i;
338 
339 	for (i = 0; i < zmd->nr_devs; i++) {
340 		if (!dmz_check_bdev(&zmd->dev[i]))
341 			return false;
342 	}
343 	return true;
344 }
345 
346 bool dmz_dev_is_dying(struct dmz_metadata *zmd)
347 {
348 	unsigned int i;
349 
350 	for (i = 0; i < zmd->nr_devs; i++) {
351 		if (dmz_bdev_is_dying(&zmd->dev[i]))
352 			return true;
353 	}
354 	return false;
355 }
356 
357 /*
358  * Lock/unlock mapping table.
359  * The map lock also protects all the zone lists.
360  */
361 void dmz_lock_map(struct dmz_metadata *zmd)
362 {
363 	mutex_lock(&zmd->map_lock);
364 }
365 
366 void dmz_unlock_map(struct dmz_metadata *zmd)
367 {
368 	mutex_unlock(&zmd->map_lock);
369 }
370 
371 /*
372  * Lock/unlock metadata access. This is a "read" lock on a semaphore
373  * that prevents metadata flush from running while metadata are being
374  * modified. The actual metadata write mutual exclusion is achieved with
375  * the map lock and zone state management (active and reclaim state are
376  * mutually exclusive).
377  */
378 void dmz_lock_metadata(struct dmz_metadata *zmd)
379 {
380 	down_read(&zmd->mblk_sem);
381 }
382 
383 void dmz_unlock_metadata(struct dmz_metadata *zmd)
384 {
385 	up_read(&zmd->mblk_sem);
386 }
387 
388 /*
389  * Lock/unlock flush: prevent concurrent executions
390  * of dmz_flush_metadata as well as metadata modification in reclaim
391  * while flush is being executed.
392  */
393 void dmz_lock_flush(struct dmz_metadata *zmd)
394 {
395 	mutex_lock(&zmd->mblk_flush_lock);
396 }
397 
398 void dmz_unlock_flush(struct dmz_metadata *zmd)
399 {
400 	mutex_unlock(&zmd->mblk_flush_lock);
401 }
402 
403 /*
404  * Allocate a metadata block.
405  */
406 static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
407 					   sector_t mblk_no)
408 {
409 	struct dmz_mblock *mblk = NULL;
410 
411 	/* See if we can reuse cached blocks */
412 	if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
413 		spin_lock(&zmd->mblk_lock);
414 		mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
415 						struct dmz_mblock, link);
416 		if (mblk) {
417 			list_del_init(&mblk->link);
418 			rb_erase(&mblk->node, &zmd->mblk_rbtree);
419 			mblk->no = mblk_no;
420 		}
421 		spin_unlock(&zmd->mblk_lock);
422 		if (mblk)
423 			return mblk;
424 	}
425 
426 	/* Allocate a new block */
427 	mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
428 	if (!mblk)
429 		return NULL;
430 
431 	mblk->page = alloc_page(GFP_NOIO);
432 	if (!mblk->page) {
433 		kfree(mblk);
434 		return NULL;
435 	}
436 
437 	RB_CLEAR_NODE(&mblk->node);
438 	INIT_LIST_HEAD(&mblk->link);
439 	mblk->ref = 0;
440 	mblk->state = 0;
441 	mblk->no = mblk_no;
442 	mblk->data = page_address(mblk->page);
443 
444 	atomic_inc(&zmd->nr_mblks);
445 
446 	return mblk;
447 }
448 
449 /*
450  * Free a metadata block.
451  */
452 static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
453 {
454 	__free_pages(mblk->page, 0);
455 	kfree(mblk);
456 
457 	atomic_dec(&zmd->nr_mblks);
458 }
459 
460 /*
461  * Insert a metadata block in the rbtree.
462  */
463 static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
464 {
465 	struct rb_root *root = &zmd->mblk_rbtree;
466 	struct rb_node **new = &(root->rb_node), *parent = NULL;
467 	struct dmz_mblock *b;
468 
469 	/* Figure out where to put the new node */
470 	while (*new) {
471 		b = container_of(*new, struct dmz_mblock, node);
472 		parent = *new;
473 		new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
474 	}
475 
476 	/* Add new node and rebalance tree */
477 	rb_link_node(&mblk->node, parent, new);
478 	rb_insert_color(&mblk->node, root);
479 }
480 
481 /*
482  * Lookup a metadata block in the rbtree. If the block is found, increment
483  * its reference count.
484  */
485 static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
486 					      sector_t mblk_no)
487 {
488 	struct rb_root *root = &zmd->mblk_rbtree;
489 	struct rb_node *node = root->rb_node;
490 	struct dmz_mblock *mblk;
491 
492 	while (node) {
493 		mblk = container_of(node, struct dmz_mblock, node);
494 		if (mblk->no == mblk_no) {
495 			/*
496 			 * If this is the first reference to the block,
497 			 * remove it from the LRU list.
498 			 */
499 			mblk->ref++;
500 			if (mblk->ref == 1 &&
501 			    !test_bit(DMZ_META_DIRTY, &mblk->state))
502 				list_del_init(&mblk->link);
503 			return mblk;
504 		}
505 		node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
506 	}
507 
508 	return NULL;
509 }
510 
511 /*
512  * Metadata block BIO end callback.
513  */
514 static void dmz_mblock_bio_end_io(struct bio *bio)
515 {
516 	struct dmz_mblock *mblk = bio->bi_private;
517 	int flag;
518 
519 	if (bio->bi_status)
520 		set_bit(DMZ_META_ERROR, &mblk->state);
521 
522 	if (bio_op(bio) == REQ_OP_WRITE)
523 		flag = DMZ_META_WRITING;
524 	else
525 		flag = DMZ_META_READING;
526 
527 	clear_bit_unlock(flag, &mblk->state);
528 	smp_mb__after_atomic();
529 	wake_up_bit(&mblk->state, flag);
530 
531 	bio_put(bio);
532 }
533 
534 /*
535  * Read an uncached metadata block from disk and add it to the cache.
536  */
537 static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
538 					      sector_t mblk_no)
539 {
540 	struct dmz_mblock *mblk, *m;
541 	sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
542 	struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev;
543 	struct bio *bio;
544 
545 	if (dmz_bdev_is_dying(dev))
546 		return ERR_PTR(-EIO);
547 
548 	/* Get a new block and a BIO to read it */
549 	mblk = dmz_alloc_mblock(zmd, mblk_no);
550 	if (!mblk)
551 		return ERR_PTR(-ENOMEM);
552 
553 	bio = bio_alloc(dev->bdev, 1, REQ_OP_READ | REQ_META | REQ_PRIO,
554 			GFP_NOIO);
555 
556 	spin_lock(&zmd->mblk_lock);
557 
558 	/*
559 	 * Make sure that another context did not start reading
560 	 * the block already.
561 	 */
562 	m = dmz_get_mblock_fast(zmd, mblk_no);
563 	if (m) {
564 		spin_unlock(&zmd->mblk_lock);
565 		dmz_free_mblock(zmd, mblk);
566 		bio_put(bio);
567 		return m;
568 	}
569 
570 	mblk->ref++;
571 	set_bit(DMZ_META_READING, &mblk->state);
572 	dmz_insert_mblock(zmd, mblk);
573 
574 	spin_unlock(&zmd->mblk_lock);
575 
576 	/* Submit read BIO */
577 	bio->bi_iter.bi_sector = dmz_blk2sect(block);
578 	bio->bi_private = mblk;
579 	bio->bi_end_io = dmz_mblock_bio_end_io;
580 	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
581 	submit_bio(bio);
582 
583 	return mblk;
584 }
585 
586 /*
587  * Free metadata blocks.
588  */
589 static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
590 					     unsigned long limit)
591 {
592 	struct dmz_mblock *mblk;
593 	unsigned long count = 0;
594 
595 	if (!zmd->max_nr_mblks)
596 		return 0;
597 
598 	while (!list_empty(&zmd->mblk_lru_list) &&
599 	       atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
600 	       count < limit) {
601 		mblk = list_first_entry(&zmd->mblk_lru_list,
602 					struct dmz_mblock, link);
603 		list_del_init(&mblk->link);
604 		rb_erase(&mblk->node, &zmd->mblk_rbtree);
605 		dmz_free_mblock(zmd, mblk);
606 		count++;
607 	}
608 
609 	return count;
610 }
611 
612 /*
613  * For mblock shrinker: get the number of unused metadata blocks in the cache.
614  */
615 static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
616 					       struct shrink_control *sc)
617 {
618 	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
619 
620 	return atomic_read(&zmd->nr_mblks);
621 }
622 
623 /*
624  * For mblock shrinker: scan unused metadata blocks and shrink the cache.
625  */
626 static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
627 					      struct shrink_control *sc)
628 {
629 	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
630 	unsigned long count;
631 
632 	spin_lock(&zmd->mblk_lock);
633 	count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
634 	spin_unlock(&zmd->mblk_lock);
635 
636 	return count ? count : SHRINK_STOP;
637 }
638 
639 /*
640  * Release a metadata block.
641  */
642 static void dmz_release_mblock(struct dmz_metadata *zmd,
643 			       struct dmz_mblock *mblk)
644 {
645 
646 	if (!mblk)
647 		return;
648 
649 	spin_lock(&zmd->mblk_lock);
650 
651 	mblk->ref--;
652 	if (mblk->ref == 0) {
653 		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
654 			rb_erase(&mblk->node, &zmd->mblk_rbtree);
655 			dmz_free_mblock(zmd, mblk);
656 		} else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
657 			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
658 			dmz_shrink_mblock_cache(zmd, 1);
659 		}
660 	}
661 
662 	spin_unlock(&zmd->mblk_lock);
663 }
664 
665 /*
666  * Get a metadata block from the rbtree. If the block
667  * is not present, read it from disk.
668  */
669 static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
670 					 sector_t mblk_no)
671 {
672 	struct dmz_mblock *mblk;
673 	struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev;
674 
675 	/* Check rbtree */
676 	spin_lock(&zmd->mblk_lock);
677 	mblk = dmz_get_mblock_fast(zmd, mblk_no);
678 	spin_unlock(&zmd->mblk_lock);
679 
680 	if (!mblk) {
681 		/* Cache miss: read the block from disk */
682 		mblk = dmz_get_mblock_slow(zmd, mblk_no);
683 		if (IS_ERR(mblk))
684 			return mblk;
685 	}
686 
687 	/* Wait for on-going read I/O and check for error */
688 	wait_on_bit_io(&mblk->state, DMZ_META_READING,
689 		       TASK_UNINTERRUPTIBLE);
690 	if (test_bit(DMZ_META_ERROR, &mblk->state)) {
691 		dmz_release_mblock(zmd, mblk);
692 		dmz_check_bdev(dev);
693 		return ERR_PTR(-EIO);
694 	}
695 
696 	return mblk;
697 }
698 
699 /*
700  * Mark a metadata block dirty.
701  */
702 static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
703 {
704 	spin_lock(&zmd->mblk_lock);
705 	if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
706 		list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
707 	spin_unlock(&zmd->mblk_lock);
708 }
709 
710 /*
711  * Issue a metadata block write BIO.
712  */
713 static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
714 			    unsigned int set)
715 {
716 	struct dmz_dev *dev = zmd->sb[set].dev;
717 	sector_t block = zmd->sb[set].block + mblk->no;
718 	struct bio *bio;
719 
720 	if (dmz_bdev_is_dying(dev))
721 		return -EIO;
722 
723 	bio = bio_alloc(dev->bdev, 1, REQ_OP_WRITE | REQ_META | REQ_PRIO,
724 			GFP_NOIO);
725 
726 	set_bit(DMZ_META_WRITING, &mblk->state);
727 
728 	bio->bi_iter.bi_sector = dmz_blk2sect(block);
729 	bio->bi_private = mblk;
730 	bio->bi_end_io = dmz_mblock_bio_end_io;
731 	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
732 	submit_bio(bio);
733 
734 	return 0;
735 }
736 
737 /*
738  * Read/write a metadata block.
739  */
740 static int dmz_rdwr_block(struct dmz_dev *dev, int op,
741 			  sector_t block, struct page *page)
742 {
743 	struct bio *bio;
744 	int ret;
745 
746 	if (WARN_ON(!dev))
747 		return -EIO;
748 
749 	if (dmz_bdev_is_dying(dev))
750 		return -EIO;
751 
752 	bio = bio_alloc(dev->bdev, 1, op | REQ_SYNC | REQ_META | REQ_PRIO,
753 			GFP_NOIO);
754 	bio->bi_iter.bi_sector = dmz_blk2sect(block);
755 	bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
756 	ret = submit_bio_wait(bio);
757 	bio_put(bio);
758 
759 	if (ret)
760 		dmz_check_bdev(dev);
761 	return ret;
762 }
763 
764 /*
765  * Write super block of the specified metadata set.
766  */
767 static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
768 {
769 	struct dmz_mblock *mblk = zmd->sb[set].mblk;
770 	struct dmz_super *sb = zmd->sb[set].sb;
771 	struct dmz_dev *dev = zmd->sb[set].dev;
772 	sector_t sb_block;
773 	u64 sb_gen = zmd->sb_gen + 1;
774 	int ret;
775 
776 	sb->magic = cpu_to_le32(DMZ_MAGIC);
777 
778 	sb->version = cpu_to_le32(zmd->sb_version);
779 	if (zmd->sb_version > 1) {
780 		BUILD_BUG_ON(UUID_SIZE != 16);
781 		export_uuid(sb->dmz_uuid, &zmd->uuid);
782 		memcpy(sb->dmz_label, zmd->label, BDEVNAME_SIZE);
783 		export_uuid(sb->dev_uuid, &dev->uuid);
784 	}
785 
786 	sb->gen = cpu_to_le64(sb_gen);
787 
788 	/*
789 	 * The metadata always references the absolute block address,
790 	 * ie relative to the entire block range, not the per-device
791 	 * block address.
792 	 */
793 	sb_block = zmd->sb[set].zone->id << zmd->zone_nr_blocks_shift;
794 	sb->sb_block = cpu_to_le64(sb_block);
795 	sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
796 	sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
797 	sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
798 
799 	sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
800 	sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
801 
802 	sb->crc = 0;
803 	sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
804 
805 	ret = dmz_rdwr_block(dev, REQ_OP_WRITE, zmd->sb[set].block,
806 			     mblk->page);
807 	if (ret == 0)
808 		ret = blkdev_issue_flush(dev->bdev);
809 
810 	return ret;
811 }
812 
813 /*
814  * Write dirty metadata blocks to the specified set.
815  */
816 static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
817 				   struct list_head *write_list,
818 				   unsigned int set)
819 {
820 	struct dmz_mblock *mblk;
821 	struct dmz_dev *dev = zmd->sb[set].dev;
822 	struct blk_plug plug;
823 	int ret = 0, nr_mblks_submitted = 0;
824 
825 	/* Issue writes */
826 	blk_start_plug(&plug);
827 	list_for_each_entry(mblk, write_list, link) {
828 		ret = dmz_write_mblock(zmd, mblk, set);
829 		if (ret)
830 			break;
831 		nr_mblks_submitted++;
832 	}
833 	blk_finish_plug(&plug);
834 
835 	/* Wait for completion */
836 	list_for_each_entry(mblk, write_list, link) {
837 		if (!nr_mblks_submitted)
838 			break;
839 		wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
840 			       TASK_UNINTERRUPTIBLE);
841 		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
842 			clear_bit(DMZ_META_ERROR, &mblk->state);
843 			dmz_check_bdev(dev);
844 			ret = -EIO;
845 		}
846 		nr_mblks_submitted--;
847 	}
848 
849 	/* Flush drive cache (this will also sync data) */
850 	if (ret == 0)
851 		ret = blkdev_issue_flush(dev->bdev);
852 
853 	return ret;
854 }
855 
856 /*
857  * Log dirty metadata blocks.
858  */
859 static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
860 				 struct list_head *write_list)
861 {
862 	unsigned int log_set = zmd->mblk_primary ^ 0x1;
863 	int ret;
864 
865 	/* Write dirty blocks to the log */
866 	ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
867 	if (ret)
868 		return ret;
869 
870 	/*
871 	 * No error so far: now validate the log by updating the
872 	 * log index super block generation.
873 	 */
874 	ret = dmz_write_sb(zmd, log_set);
875 	if (ret)
876 		return ret;
877 
878 	return 0;
879 }
880 
881 /*
882  * Flush dirty metadata blocks.
883  */
884 int dmz_flush_metadata(struct dmz_metadata *zmd)
885 {
886 	struct dmz_mblock *mblk;
887 	struct list_head write_list;
888 	struct dmz_dev *dev;
889 	int ret;
890 
891 	if (WARN_ON(!zmd))
892 		return 0;
893 
894 	INIT_LIST_HEAD(&write_list);
895 
896 	/*
897 	 * Make sure that metadata blocks are stable before logging: take
898 	 * the write lock on the metadata semaphore to prevent target BIOs
899 	 * from modifying metadata.
900 	 */
901 	down_write(&zmd->mblk_sem);
902 	dev = zmd->sb[zmd->mblk_primary].dev;
903 
904 	/*
905 	 * This is called from the target flush work and reclaim work.
906 	 * Concurrent execution is not allowed.
907 	 */
908 	dmz_lock_flush(zmd);
909 
910 	if (dmz_bdev_is_dying(dev)) {
911 		ret = -EIO;
912 		goto out;
913 	}
914 
915 	/* Get dirty blocks */
916 	spin_lock(&zmd->mblk_lock);
917 	list_splice_init(&zmd->mblk_dirty_list, &write_list);
918 	spin_unlock(&zmd->mblk_lock);
919 
920 	/* If there are no dirty metadata blocks, just flush the device cache */
921 	if (list_empty(&write_list)) {
922 		ret = blkdev_issue_flush(dev->bdev);
923 		goto err;
924 	}
925 
926 	/*
927 	 * The primary metadata set is still clean. Keep it this way until
928 	 * all updates are successful in the secondary set. That is, use
929 	 * the secondary set as a log.
930 	 */
931 	ret = dmz_log_dirty_mblocks(zmd, &write_list);
932 	if (ret)
933 		goto err;
934 
935 	/*
936 	 * The log is on disk. It is now safe to update in place
937 	 * in the primary metadata set.
938 	 */
939 	ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
940 	if (ret)
941 		goto err;
942 
943 	ret = dmz_write_sb(zmd, zmd->mblk_primary);
944 	if (ret)
945 		goto err;
946 
947 	while (!list_empty(&write_list)) {
948 		mblk = list_first_entry(&write_list, struct dmz_mblock, link);
949 		list_del_init(&mblk->link);
950 
951 		spin_lock(&zmd->mblk_lock);
952 		clear_bit(DMZ_META_DIRTY, &mblk->state);
953 		if (mblk->ref == 0)
954 			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
955 		spin_unlock(&zmd->mblk_lock);
956 	}
957 
958 	zmd->sb_gen++;
959 out:
960 	dmz_unlock_flush(zmd);
961 	up_write(&zmd->mblk_sem);
962 
963 	return ret;
964 
965 err:
966 	if (!list_empty(&write_list)) {
967 		spin_lock(&zmd->mblk_lock);
968 		list_splice(&write_list, &zmd->mblk_dirty_list);
969 		spin_unlock(&zmd->mblk_lock);
970 	}
971 	if (!dmz_check_bdev(dev))
972 		ret = -EIO;
973 	goto out;
974 }
975 
976 /*
977  * Check super block.
978  */
979 static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_sb *dsb,
980 			bool tertiary)
981 {
982 	struct dmz_super *sb = dsb->sb;
983 	struct dmz_dev *dev = dsb->dev;
984 	unsigned int nr_meta_zones, nr_data_zones;
985 	u32 crc, stored_crc;
986 	u64 gen, sb_block;
987 
988 	if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
989 		dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
990 			    DMZ_MAGIC, le32_to_cpu(sb->magic));
991 		return -ENXIO;
992 	}
993 
994 	zmd->sb_version = le32_to_cpu(sb->version);
995 	if (zmd->sb_version > DMZ_META_VER) {
996 		dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
997 			    DMZ_META_VER, zmd->sb_version);
998 		return -EINVAL;
999 	}
1000 	if (zmd->sb_version < 2 && tertiary) {
1001 		dmz_dev_err(dev, "Tertiary superblocks are not supported");
1002 		return -EINVAL;
1003 	}
1004 
1005 	gen = le64_to_cpu(sb->gen);
1006 	stored_crc = le32_to_cpu(sb->crc);
1007 	sb->crc = 0;
1008 	crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
1009 	if (crc != stored_crc) {
1010 		dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
1011 			    crc, stored_crc);
1012 		return -ENXIO;
1013 	}
1014 
1015 	sb_block = le64_to_cpu(sb->sb_block);
1016 	if (sb_block != (u64)dsb->zone->id << zmd->zone_nr_blocks_shift ) {
1017 		dmz_dev_err(dev, "Invalid superblock position "
1018 			    "(is %llu expected %llu)",
1019 			    sb_block,
1020 			    (u64)dsb->zone->id << zmd->zone_nr_blocks_shift);
1021 		return -EINVAL;
1022 	}
1023 	if (zmd->sb_version > 1) {
1024 		uuid_t sb_uuid;
1025 
1026 		import_uuid(&sb_uuid, sb->dmz_uuid);
1027 		if (uuid_is_null(&sb_uuid)) {
1028 			dmz_dev_err(dev, "NULL DM-Zoned uuid");
1029 			return -ENXIO;
1030 		} else if (uuid_is_null(&zmd->uuid)) {
1031 			uuid_copy(&zmd->uuid, &sb_uuid);
1032 		} else if (!uuid_equal(&zmd->uuid, &sb_uuid)) {
1033 			dmz_dev_err(dev, "mismatching DM-Zoned uuid, "
1034 				    "is %pUl expected %pUl",
1035 				    &sb_uuid, &zmd->uuid);
1036 			return -ENXIO;
1037 		}
1038 		if (!strlen(zmd->label))
1039 			memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE);
1040 		else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) {
1041 			dmz_dev_err(dev, "mismatching DM-Zoned label, "
1042 				    "is %s expected %s",
1043 				    sb->dmz_label, zmd->label);
1044 			return -ENXIO;
1045 		}
1046 		import_uuid(&dev->uuid, sb->dev_uuid);
1047 		if (uuid_is_null(&dev->uuid)) {
1048 			dmz_dev_err(dev, "NULL device uuid");
1049 			return -ENXIO;
1050 		}
1051 
1052 		if (tertiary) {
1053 			/*
1054 			 * Generation number should be 0, but it doesn't
1055 			 * really matter if it isn't.
1056 			 */
1057 			if (gen != 0)
1058 				dmz_dev_warn(dev, "Invalid generation %llu",
1059 					    gen);
1060 			return 0;
1061 		}
1062 	}
1063 
1064 	nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1)
1065 		>> zmd->zone_nr_blocks_shift;
1066 	if (!nr_meta_zones ||
1067 	    (zmd->nr_devs <= 1 && nr_meta_zones >= zmd->nr_rnd_zones) ||
1068 	    (zmd->nr_devs > 1 && nr_meta_zones >= zmd->nr_cache_zones)) {
1069 		dmz_dev_err(dev, "Invalid number of metadata blocks");
1070 		return -ENXIO;
1071 	}
1072 
1073 	if (!le32_to_cpu(sb->nr_reserved_seq) ||
1074 	    le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
1075 		dmz_dev_err(dev, "Invalid number of reserved sequential zones");
1076 		return -ENXIO;
1077 	}
1078 
1079 	nr_data_zones = zmd->nr_useable_zones -
1080 		(nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
1081 	if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
1082 		dmz_dev_err(dev, "Invalid number of chunks %u / %u",
1083 			    le32_to_cpu(sb->nr_chunks), nr_data_zones);
1084 		return -ENXIO;
1085 	}
1086 
1087 	/* OK */
1088 	zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
1089 	zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
1090 	zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
1091 	zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
1092 	zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
1093 	zmd->nr_meta_zones = nr_meta_zones;
1094 	zmd->nr_data_zones = nr_data_zones;
1095 
1096 	return 0;
1097 }
1098 
1099 /*
1100  * Read the first or second super block from disk.
1101  */
1102 static int dmz_read_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1103 {
1104 	dmz_zmd_debug(zmd, "read superblock set %d dev %pg block %llu",
1105 		      set, sb->dev->bdev, sb->block);
1106 
1107 	return dmz_rdwr_block(sb->dev, REQ_OP_READ,
1108 			      sb->block, sb->mblk->page);
1109 }
1110 
1111 /*
1112  * Determine the position of the secondary super blocks on disk.
1113  * This is used only if a corruption of the primary super block
1114  * is detected.
1115  */
1116 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
1117 {
1118 	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
1119 	struct dmz_mblock *mblk;
1120 	unsigned int zone_id = zmd->sb[0].zone->id;
1121 	int i;
1122 
1123 	/* Allocate a block */
1124 	mblk = dmz_alloc_mblock(zmd, 0);
1125 	if (!mblk)
1126 		return -ENOMEM;
1127 
1128 	zmd->sb[1].mblk = mblk;
1129 	zmd->sb[1].sb = mblk->data;
1130 
1131 	/* Bad first super block: search for the second one */
1132 	zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
1133 	zmd->sb[1].zone = dmz_get(zmd, zone_id + 1);
1134 	zmd->sb[1].dev = zmd->sb[0].dev;
1135 	for (i = 1; i < zmd->nr_rnd_zones; i++) {
1136 		if (dmz_read_sb(zmd, &zmd->sb[1], 1) != 0)
1137 			break;
1138 		if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
1139 			return 0;
1140 		zmd->sb[1].block += zone_nr_blocks;
1141 		zmd->sb[1].zone = dmz_get(zmd, zone_id + i);
1142 	}
1143 
1144 	dmz_free_mblock(zmd, mblk);
1145 	zmd->sb[1].mblk = NULL;
1146 	zmd->sb[1].zone = NULL;
1147 	zmd->sb[1].dev = NULL;
1148 
1149 	return -EIO;
1150 }
1151 
1152 /*
1153  * Read a super block from disk.
1154  */
1155 static int dmz_get_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1156 {
1157 	struct dmz_mblock *mblk;
1158 	int ret;
1159 
1160 	/* Allocate a block */
1161 	mblk = dmz_alloc_mblock(zmd, 0);
1162 	if (!mblk)
1163 		return -ENOMEM;
1164 
1165 	sb->mblk = mblk;
1166 	sb->sb = mblk->data;
1167 
1168 	/* Read super block */
1169 	ret = dmz_read_sb(zmd, sb, set);
1170 	if (ret) {
1171 		dmz_free_mblock(zmd, mblk);
1172 		sb->mblk = NULL;
1173 		return ret;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
1179 /*
1180  * Recover a metadata set.
1181  */
1182 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
1183 {
1184 	unsigned int src_set = dst_set ^ 0x1;
1185 	struct page *page;
1186 	int i, ret;
1187 
1188 	dmz_dev_warn(zmd->sb[dst_set].dev,
1189 		     "Metadata set %u invalid: recovering", dst_set);
1190 
1191 	if (dst_set == 0)
1192 		zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1193 	else
1194 		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1195 
1196 	page = alloc_page(GFP_NOIO);
1197 	if (!page)
1198 		return -ENOMEM;
1199 
1200 	/* Copy metadata blocks */
1201 	for (i = 1; i < zmd->nr_meta_blocks; i++) {
1202 		ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ,
1203 				     zmd->sb[src_set].block + i, page);
1204 		if (ret)
1205 			goto out;
1206 		ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE,
1207 				     zmd->sb[dst_set].block + i, page);
1208 		if (ret)
1209 			goto out;
1210 	}
1211 
1212 	/* Finalize with the super block */
1213 	if (!zmd->sb[dst_set].mblk) {
1214 		zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1215 		if (!zmd->sb[dst_set].mblk) {
1216 			ret = -ENOMEM;
1217 			goto out;
1218 		}
1219 		zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1220 	}
1221 
1222 	ret = dmz_write_sb(zmd, dst_set);
1223 out:
1224 	__free_pages(page, 0);
1225 
1226 	return ret;
1227 }
1228 
1229 /*
1230  * Get super block from disk.
1231  */
1232 static int dmz_load_sb(struct dmz_metadata *zmd)
1233 {
1234 	bool sb_good[2] = {false, false};
1235 	u64 sb_gen[2] = {0, 0};
1236 	int ret;
1237 
1238 	if (!zmd->sb[0].zone) {
1239 		dmz_zmd_err(zmd, "Primary super block zone not set");
1240 		return -ENXIO;
1241 	}
1242 
1243 	/* Read and check the primary super block */
1244 	zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1245 	zmd->sb[0].dev = zmd->sb[0].zone->dev;
1246 	ret = dmz_get_sb(zmd, &zmd->sb[0], 0);
1247 	if (ret) {
1248 		dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed");
1249 		return ret;
1250 	}
1251 
1252 	ret = dmz_check_sb(zmd, &zmd->sb[0], false);
1253 
1254 	/* Read and check secondary super block */
1255 	if (ret == 0) {
1256 		sb_good[0] = true;
1257 		if (!zmd->sb[1].zone) {
1258 			unsigned int zone_id =
1259 				zmd->sb[0].zone->id + zmd->nr_meta_zones;
1260 
1261 			zmd->sb[1].zone = dmz_get(zmd, zone_id);
1262 		}
1263 		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1264 		zmd->sb[1].dev = zmd->sb[0].dev;
1265 		ret = dmz_get_sb(zmd, &zmd->sb[1], 1);
1266 	} else
1267 		ret = dmz_lookup_secondary_sb(zmd);
1268 
1269 	if (ret) {
1270 		dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed");
1271 		return ret;
1272 	}
1273 
1274 	ret = dmz_check_sb(zmd, &zmd->sb[1], false);
1275 	if (ret == 0)
1276 		sb_good[1] = true;
1277 
1278 	/* Use highest generation sb first */
1279 	if (!sb_good[0] && !sb_good[1]) {
1280 		dmz_zmd_err(zmd, "No valid super block found");
1281 		return -EIO;
1282 	}
1283 
1284 	if (sb_good[0])
1285 		sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1286 	else {
1287 		ret = dmz_recover_mblocks(zmd, 0);
1288 		if (ret) {
1289 			dmz_dev_err(zmd->sb[0].dev,
1290 				    "Recovery of superblock 0 failed");
1291 			return -EIO;
1292 		}
1293 	}
1294 
1295 	if (sb_good[1])
1296 		sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1297 	else {
1298 		ret = dmz_recover_mblocks(zmd, 1);
1299 
1300 		if (ret) {
1301 			dmz_dev_err(zmd->sb[1].dev,
1302 				    "Recovery of superblock 1 failed");
1303 			return -EIO;
1304 		}
1305 	}
1306 
1307 	if (sb_gen[0] >= sb_gen[1]) {
1308 		zmd->sb_gen = sb_gen[0];
1309 		zmd->mblk_primary = 0;
1310 	} else {
1311 		zmd->sb_gen = sb_gen[1];
1312 		zmd->mblk_primary = 1;
1313 	}
1314 
1315 	dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev,
1316 		      "Using super block %u (gen %llu)",
1317 		      zmd->mblk_primary, zmd->sb_gen);
1318 
1319 	if (zmd->sb_version > 1) {
1320 		int i;
1321 		struct dmz_sb *sb;
1322 
1323 		sb = kzalloc(sizeof(struct dmz_sb), GFP_KERNEL);
1324 		if (!sb)
1325 			return -ENOMEM;
1326 		for (i = 1; i < zmd->nr_devs; i++) {
1327 			sb->block = 0;
1328 			sb->zone = dmz_get(zmd, zmd->dev[i].zone_offset);
1329 			sb->dev = &zmd->dev[i];
1330 			if (!dmz_is_meta(sb->zone)) {
1331 				dmz_dev_err(sb->dev,
1332 					    "Tertiary super block zone %u not marked as metadata zone",
1333 					    sb->zone->id);
1334 				ret = -EINVAL;
1335 				goto out_kfree;
1336 			}
1337 			ret = dmz_get_sb(zmd, sb, i + 1);
1338 			if (ret) {
1339 				dmz_dev_err(sb->dev,
1340 					    "Read tertiary super block failed");
1341 				dmz_free_mblock(zmd, sb->mblk);
1342 				goto out_kfree;
1343 			}
1344 			ret = dmz_check_sb(zmd, sb, true);
1345 			dmz_free_mblock(zmd, sb->mblk);
1346 			if (ret == -EINVAL)
1347 				goto out_kfree;
1348 		}
1349 	out_kfree:
1350 		kfree(sb);
1351 	}
1352 	return ret;
1353 }
1354 
1355 /*
1356  * Initialize a zone descriptor.
1357  */
1358 static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data)
1359 {
1360 	struct dmz_dev *dev = data;
1361 	struct dmz_metadata *zmd = dev->metadata;
1362 	int idx = num + dev->zone_offset;
1363 	struct dm_zone *zone;
1364 
1365 	zone = dmz_insert(zmd, idx, dev);
1366 	if (IS_ERR(zone))
1367 		return PTR_ERR(zone);
1368 
1369 	if (blkz->len != zmd->zone_nr_sectors) {
1370 		if (zmd->sb_version > 1) {
1371 			/* Ignore the eventual runt (smaller) zone */
1372 			set_bit(DMZ_OFFLINE, &zone->flags);
1373 			return 0;
1374 		} else if (blkz->start + blkz->len == dev->capacity)
1375 			return 0;
1376 		return -ENXIO;
1377 	}
1378 
1379 	/*
1380 	 * Devices that have zones with a capacity smaller than the zone size
1381 	 * (e.g. NVMe zoned namespaces) are not supported.
1382 	 */
1383 	if (blkz->capacity != blkz->len)
1384 		return -ENXIO;
1385 
1386 	switch (blkz->type) {
1387 	case BLK_ZONE_TYPE_CONVENTIONAL:
1388 		set_bit(DMZ_RND, &zone->flags);
1389 		break;
1390 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1391 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1392 		set_bit(DMZ_SEQ, &zone->flags);
1393 		break;
1394 	default:
1395 		return -ENXIO;
1396 	}
1397 
1398 	if (dmz_is_rnd(zone))
1399 		zone->wp_block = 0;
1400 	else
1401 		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1402 
1403 	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1404 		set_bit(DMZ_OFFLINE, &zone->flags);
1405 	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1406 		set_bit(DMZ_READ_ONLY, &zone->flags);
1407 	else {
1408 		zmd->nr_useable_zones++;
1409 		if (dmz_is_rnd(zone)) {
1410 			zmd->nr_rnd_zones++;
1411 			if (zmd->nr_devs == 1 && !zmd->sb[0].zone) {
1412 				/* Primary super block zone */
1413 				zmd->sb[0].zone = zone;
1414 			}
1415 		}
1416 		if (zmd->nr_devs > 1 && num == 0) {
1417 			/*
1418 			 * Tertiary superblock zones are always at the
1419 			 * start of the zoned devices, so mark them
1420 			 * as metadata zone.
1421 			 */
1422 			set_bit(DMZ_META, &zone->flags);
1423 		}
1424 	}
1425 	return 0;
1426 }
1427 
1428 static int dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev)
1429 {
1430 	int idx;
1431 	sector_t zone_offset = 0;
1432 
1433 	for(idx = 0; idx < dev->nr_zones; idx++) {
1434 		struct dm_zone *zone;
1435 
1436 		zone = dmz_insert(zmd, idx, dev);
1437 		if (IS_ERR(zone))
1438 			return PTR_ERR(zone);
1439 		set_bit(DMZ_CACHE, &zone->flags);
1440 		zone->wp_block = 0;
1441 		zmd->nr_cache_zones++;
1442 		zmd->nr_useable_zones++;
1443 		if (dev->capacity - zone_offset < zmd->zone_nr_sectors) {
1444 			/* Disable runt zone */
1445 			set_bit(DMZ_OFFLINE, &zone->flags);
1446 			break;
1447 		}
1448 		zone_offset += zmd->zone_nr_sectors;
1449 	}
1450 	return 0;
1451 }
1452 
1453 /*
1454  * Free zones descriptors.
1455  */
1456 static void dmz_drop_zones(struct dmz_metadata *zmd)
1457 {
1458 	int idx;
1459 
1460 	for(idx = 0; idx < zmd->nr_zones; idx++) {
1461 		struct dm_zone *zone = xa_load(&zmd->zones, idx);
1462 
1463 		kfree(zone);
1464 		xa_erase(&zmd->zones, idx);
1465 	}
1466 	xa_destroy(&zmd->zones);
1467 }
1468 
1469 /*
1470  * Allocate and initialize zone descriptors using the zone
1471  * information from disk.
1472  */
1473 static int dmz_init_zones(struct dmz_metadata *zmd)
1474 {
1475 	int i, ret;
1476 	struct dmz_dev *zoned_dev = &zmd->dev[0];
1477 
1478 	/* Init */
1479 	zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors;
1480 	zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors);
1481 	zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors);
1482 	zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks);
1483 	zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3;
1484 	zmd->zone_nr_bitmap_blocks =
1485 		max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1486 	zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks,
1487 					DMZ_BLOCK_SIZE_BITS);
1488 
1489 	/* Allocate zone array */
1490 	zmd->nr_zones = 0;
1491 	for (i = 0; i < zmd->nr_devs; i++) {
1492 		struct dmz_dev *dev = &zmd->dev[i];
1493 
1494 		dev->metadata = zmd;
1495 		zmd->nr_zones += dev->nr_zones;
1496 
1497 		atomic_set(&dev->unmap_nr_rnd, 0);
1498 		INIT_LIST_HEAD(&dev->unmap_rnd_list);
1499 		INIT_LIST_HEAD(&dev->map_rnd_list);
1500 
1501 		atomic_set(&dev->unmap_nr_seq, 0);
1502 		INIT_LIST_HEAD(&dev->unmap_seq_list);
1503 		INIT_LIST_HEAD(&dev->map_seq_list);
1504 	}
1505 
1506 	if (!zmd->nr_zones) {
1507 		DMERR("(%s): No zones found", zmd->devname);
1508 		return -ENXIO;
1509 	}
1510 	xa_init(&zmd->zones);
1511 
1512 	DMDEBUG("(%s): Using %zu B for zone information",
1513 		zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones);
1514 
1515 	if (zmd->nr_devs > 1) {
1516 		ret = dmz_emulate_zones(zmd, &zmd->dev[0]);
1517 		if (ret < 0) {
1518 			DMDEBUG("(%s): Failed to emulate zones, error %d",
1519 				zmd->devname, ret);
1520 			dmz_drop_zones(zmd);
1521 			return ret;
1522 		}
1523 
1524 		/*
1525 		 * Primary superblock zone is always at zone 0 when multiple
1526 		 * drives are present.
1527 		 */
1528 		zmd->sb[0].zone = dmz_get(zmd, 0);
1529 
1530 		for (i = 1; i < zmd->nr_devs; i++) {
1531 			zoned_dev = &zmd->dev[i];
1532 
1533 			ret = blkdev_report_zones(zoned_dev->bdev, 0,
1534 						  BLK_ALL_ZONES,
1535 						  dmz_init_zone, zoned_dev);
1536 			if (ret < 0) {
1537 				DMDEBUG("(%s): Failed to report zones, error %d",
1538 					zmd->devname, ret);
1539 				dmz_drop_zones(zmd);
1540 				return ret;
1541 			}
1542 		}
1543 		return 0;
1544 	}
1545 
1546 	/*
1547 	 * Get zone information and initialize zone descriptors.  At the same
1548 	 * time, determine where the super block should be: first block of the
1549 	 * first randomly writable zone.
1550 	 */
1551 	ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES,
1552 				  dmz_init_zone, zoned_dev);
1553 	if (ret < 0) {
1554 		DMDEBUG("(%s): Failed to report zones, error %d",
1555 			zmd->devname, ret);
1556 		dmz_drop_zones(zmd);
1557 		return ret;
1558 	}
1559 
1560 	return 0;
1561 }
1562 
1563 static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1564 			      void *data)
1565 {
1566 	struct dm_zone *zone = data;
1567 
1568 	clear_bit(DMZ_OFFLINE, &zone->flags);
1569 	clear_bit(DMZ_READ_ONLY, &zone->flags);
1570 	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1571 		set_bit(DMZ_OFFLINE, &zone->flags);
1572 	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1573 		set_bit(DMZ_READ_ONLY, &zone->flags);
1574 
1575 	if (dmz_is_seq(zone))
1576 		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1577 	else
1578 		zone->wp_block = 0;
1579 	return 0;
1580 }
1581 
1582 /*
1583  * Update a zone information.
1584  */
1585 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1586 {
1587 	struct dmz_dev *dev = zone->dev;
1588 	unsigned int noio_flag;
1589 	int ret;
1590 
1591 	if (dev->flags & DMZ_BDEV_REGULAR)
1592 		return 0;
1593 
1594 	/*
1595 	 * Get zone information from disk. Since blkdev_report_zones() uses
1596 	 * GFP_KERNEL by default for memory allocations, set the per-task
1597 	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1598 	 * GFP_NOIO was specified.
1599 	 */
1600 	noio_flag = memalloc_noio_save();
1601 	ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1,
1602 				  dmz_update_zone_cb, zone);
1603 	memalloc_noio_restore(noio_flag);
1604 
1605 	if (ret == 0)
1606 		ret = -EIO;
1607 	if (ret < 0) {
1608 		dmz_dev_err(dev, "Get zone %u report failed",
1609 			    zone->id);
1610 		dmz_check_bdev(dev);
1611 		return ret;
1612 	}
1613 
1614 	return 0;
1615 }
1616 
1617 /*
1618  * Check a zone write pointer position when the zone is marked
1619  * with the sequential write error flag.
1620  */
1621 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1622 				    struct dm_zone *zone)
1623 {
1624 	struct dmz_dev *dev = zone->dev;
1625 	unsigned int wp = 0;
1626 	int ret;
1627 
1628 	wp = zone->wp_block;
1629 	ret = dmz_update_zone(zmd, zone);
1630 	if (ret)
1631 		return ret;
1632 
1633 	dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)",
1634 		     zone->id, zone->wp_block, wp);
1635 
1636 	if (zone->wp_block < wp) {
1637 		dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1638 				      wp - zone->wp_block);
1639 	}
1640 
1641 	return 0;
1642 }
1643 
1644 /*
1645  * Reset a zone write pointer.
1646  */
1647 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1648 {
1649 	int ret;
1650 
1651 	/*
1652 	 * Ignore offline zones, read only zones,
1653 	 * and conventional zones.
1654 	 */
1655 	if (dmz_is_offline(zone) ||
1656 	    dmz_is_readonly(zone) ||
1657 	    dmz_is_rnd(zone))
1658 		return 0;
1659 
1660 	if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1661 		struct dmz_dev *dev = zone->dev;
1662 
1663 		ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1664 				       dmz_start_sect(zmd, zone),
1665 				       zmd->zone_nr_sectors, GFP_NOIO);
1666 		if (ret) {
1667 			dmz_dev_err(dev, "Reset zone %u failed %d",
1668 				    zone->id, ret);
1669 			return ret;
1670 		}
1671 	}
1672 
1673 	/* Clear write error bit and rewind write pointer position */
1674 	clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1675 	zone->wp_block = 0;
1676 
1677 	return 0;
1678 }
1679 
1680 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1681 
1682 /*
1683  * Initialize chunk mapping.
1684  */
1685 static int dmz_load_mapping(struct dmz_metadata *zmd)
1686 {
1687 	struct dm_zone *dzone, *bzone;
1688 	struct dmz_mblock *dmap_mblk = NULL;
1689 	struct dmz_map *dmap;
1690 	unsigned int i = 0, e = 0, chunk = 0;
1691 	unsigned int dzone_id;
1692 	unsigned int bzone_id;
1693 
1694 	/* Metadata block array for the chunk mapping table */
1695 	zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1696 				sizeof(struct dmz_mblk *), GFP_KERNEL);
1697 	if (!zmd->map_mblk)
1698 		return -ENOMEM;
1699 
1700 	/* Get chunk mapping table blocks and initialize zone mapping */
1701 	while (chunk < zmd->nr_chunks) {
1702 		if (!dmap_mblk) {
1703 			/* Get mapping block */
1704 			dmap_mblk = dmz_get_mblock(zmd, i + 1);
1705 			if (IS_ERR(dmap_mblk))
1706 				return PTR_ERR(dmap_mblk);
1707 			zmd->map_mblk[i] = dmap_mblk;
1708 			dmap = (struct dmz_map *) dmap_mblk->data;
1709 			i++;
1710 			e = 0;
1711 		}
1712 
1713 		/* Check data zone */
1714 		dzone_id = le32_to_cpu(dmap[e].dzone_id);
1715 		if (dzone_id == DMZ_MAP_UNMAPPED)
1716 			goto next;
1717 
1718 		if (dzone_id >= zmd->nr_zones) {
1719 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u",
1720 				    chunk, dzone_id);
1721 			return -EIO;
1722 		}
1723 
1724 		dzone = dmz_get(zmd, dzone_id);
1725 		if (!dzone) {
1726 			dmz_zmd_err(zmd, "Chunk %u mapping: data zone %u not present",
1727 				    chunk, dzone_id);
1728 			return -EIO;
1729 		}
1730 		set_bit(DMZ_DATA, &dzone->flags);
1731 		dzone->chunk = chunk;
1732 		dmz_get_zone_weight(zmd, dzone);
1733 
1734 		if (dmz_is_cache(dzone))
1735 			list_add_tail(&dzone->link, &zmd->map_cache_list);
1736 		else if (dmz_is_rnd(dzone))
1737 			list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
1738 		else
1739 			list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
1740 
1741 		/* Check buffer zone */
1742 		bzone_id = le32_to_cpu(dmap[e].bzone_id);
1743 		if (bzone_id == DMZ_MAP_UNMAPPED)
1744 			goto next;
1745 
1746 		if (bzone_id >= zmd->nr_zones) {
1747 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u",
1748 				    chunk, bzone_id);
1749 			return -EIO;
1750 		}
1751 
1752 		bzone = dmz_get(zmd, bzone_id);
1753 		if (!bzone) {
1754 			dmz_zmd_err(zmd, "Chunk %u mapping: buffer zone %u not present",
1755 				    chunk, bzone_id);
1756 			return -EIO;
1757 		}
1758 		if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) {
1759 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u",
1760 				    chunk, bzone_id);
1761 			return -EIO;
1762 		}
1763 
1764 		set_bit(DMZ_DATA, &bzone->flags);
1765 		set_bit(DMZ_BUF, &bzone->flags);
1766 		bzone->chunk = chunk;
1767 		bzone->bzone = dzone;
1768 		dzone->bzone = bzone;
1769 		dmz_get_zone_weight(zmd, bzone);
1770 		if (dmz_is_cache(bzone))
1771 			list_add_tail(&bzone->link, &zmd->map_cache_list);
1772 		else
1773 			list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
1774 next:
1775 		chunk++;
1776 		e++;
1777 		if (e >= DMZ_MAP_ENTRIES)
1778 			dmap_mblk = NULL;
1779 	}
1780 
1781 	/*
1782 	 * At this point, only meta zones and mapped data zones were
1783 	 * fully initialized. All remaining zones are unmapped data
1784 	 * zones. Finish initializing those here.
1785 	 */
1786 	for (i = 0; i < zmd->nr_zones; i++) {
1787 		dzone = dmz_get(zmd, i);
1788 		if (!dzone)
1789 			continue;
1790 		if (dmz_is_meta(dzone))
1791 			continue;
1792 		if (dmz_is_offline(dzone))
1793 			continue;
1794 
1795 		if (dmz_is_cache(dzone))
1796 			zmd->nr_cache++;
1797 		else if (dmz_is_rnd(dzone))
1798 			dzone->dev->nr_rnd++;
1799 		else
1800 			dzone->dev->nr_seq++;
1801 
1802 		if (dmz_is_data(dzone)) {
1803 			/* Already initialized */
1804 			continue;
1805 		}
1806 
1807 		/* Unmapped data zone */
1808 		set_bit(DMZ_DATA, &dzone->flags);
1809 		dzone->chunk = DMZ_MAP_UNMAPPED;
1810 		if (dmz_is_cache(dzone)) {
1811 			list_add_tail(&dzone->link, &zmd->unmap_cache_list);
1812 			atomic_inc(&zmd->unmap_nr_cache);
1813 		} else if (dmz_is_rnd(dzone)) {
1814 			list_add_tail(&dzone->link,
1815 				      &dzone->dev->unmap_rnd_list);
1816 			atomic_inc(&dzone->dev->unmap_nr_rnd);
1817 		} else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1818 			list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1819 			set_bit(DMZ_RESERVED, &dzone->flags);
1820 			atomic_inc(&zmd->nr_reserved_seq_zones);
1821 			dzone->dev->nr_seq--;
1822 		} else {
1823 			list_add_tail(&dzone->link,
1824 				      &dzone->dev->unmap_seq_list);
1825 			atomic_inc(&dzone->dev->unmap_nr_seq);
1826 		}
1827 	}
1828 
1829 	return 0;
1830 }
1831 
1832 /*
1833  * Set a data chunk mapping.
1834  */
1835 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1836 				  unsigned int dzone_id, unsigned int bzone_id)
1837 {
1838 	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1839 	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1840 	int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1841 
1842 	dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1843 	dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1844 	dmz_dirty_mblock(zmd, dmap_mblk);
1845 }
1846 
1847 /*
1848  * The list of mapped zones is maintained in LRU order.
1849  * This rotates a zone at the end of its map list.
1850  */
1851 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1852 {
1853 	if (list_empty(&zone->link))
1854 		return;
1855 
1856 	list_del_init(&zone->link);
1857 	if (dmz_is_seq(zone)) {
1858 		/* LRU rotate sequential zone */
1859 		list_add_tail(&zone->link, &zone->dev->map_seq_list);
1860 	} else if (dmz_is_cache(zone)) {
1861 		/* LRU rotate cache zone */
1862 		list_add_tail(&zone->link, &zmd->map_cache_list);
1863 	} else {
1864 		/* LRU rotate random zone */
1865 		list_add_tail(&zone->link, &zone->dev->map_rnd_list);
1866 	}
1867 }
1868 
1869 /*
1870  * The list of mapped random zones is maintained
1871  * in LRU order. This rotates a zone at the end of the list.
1872  */
1873 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1874 {
1875 	__dmz_lru_zone(zmd, zone);
1876 	if (zone->bzone)
1877 		__dmz_lru_zone(zmd, zone->bzone);
1878 }
1879 
1880 /*
1881  * Wait for any zone to be freed.
1882  */
1883 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1884 {
1885 	DEFINE_WAIT(wait);
1886 
1887 	prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1888 	dmz_unlock_map(zmd);
1889 	dmz_unlock_metadata(zmd);
1890 
1891 	io_schedule_timeout(HZ);
1892 
1893 	dmz_lock_metadata(zmd);
1894 	dmz_lock_map(zmd);
1895 	finish_wait(&zmd->free_wq, &wait);
1896 }
1897 
1898 /*
1899  * Lock a zone for reclaim (set the zone RECLAIM bit).
1900  * Returns false if the zone cannot be locked or if it is already locked
1901  * and 1 otherwise.
1902  */
1903 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1904 {
1905 	/* Active zones cannot be reclaimed */
1906 	if (dmz_is_active(zone))
1907 		return 0;
1908 
1909 	return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1910 }
1911 
1912 /*
1913  * Clear a zone reclaim flag.
1914  */
1915 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1916 {
1917 	WARN_ON(dmz_is_active(zone));
1918 	WARN_ON(!dmz_in_reclaim(zone));
1919 
1920 	clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1921 	smp_mb__after_atomic();
1922 	wake_up_bit(&zone->flags, DMZ_RECLAIM);
1923 }
1924 
1925 /*
1926  * Wait for a zone reclaim to complete.
1927  */
1928 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1929 {
1930 	dmz_unlock_map(zmd);
1931 	dmz_unlock_metadata(zmd);
1932 	set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1933 	wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1934 	clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1935 	dmz_lock_metadata(zmd);
1936 	dmz_lock_map(zmd);
1937 }
1938 
1939 /*
1940  * Select a cache or random write zone for reclaim.
1941  */
1942 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd,
1943 						    unsigned int idx, bool idle)
1944 {
1945 	struct dm_zone *dzone = NULL;
1946 	struct dm_zone *zone, *maxw_z = NULL;
1947 	struct list_head *zone_list;
1948 
1949 	/* If we have cache zones select from the cache zone list */
1950 	if (zmd->nr_cache) {
1951 		zone_list = &zmd->map_cache_list;
1952 		/* Try to relaim random zones, too, when idle */
1953 		if (idle && list_empty(zone_list))
1954 			zone_list = &zmd->dev[idx].map_rnd_list;
1955 	} else
1956 		zone_list = &zmd->dev[idx].map_rnd_list;
1957 
1958 	/*
1959 	 * Find the buffer zone with the heaviest weight or the first (oldest)
1960 	 * data zone that can be reclaimed.
1961 	 */
1962 	list_for_each_entry(zone, zone_list, link) {
1963 		if (dmz_is_buf(zone)) {
1964 			dzone = zone->bzone;
1965 			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1966 				continue;
1967 			if (!maxw_z || maxw_z->weight < dzone->weight)
1968 				maxw_z = dzone;
1969 		} else {
1970 			dzone = zone;
1971 			if (dmz_lock_zone_reclaim(dzone))
1972 				return dzone;
1973 		}
1974 	}
1975 
1976 	if (maxw_z && dmz_lock_zone_reclaim(maxw_z))
1977 		return maxw_z;
1978 
1979 	/*
1980 	 * If we come here, none of the zones inspected could be locked for
1981 	 * reclaim. Try again, being more aggressive, that is, find the
1982 	 * first zone that can be reclaimed regardless of its weitght.
1983 	 */
1984 	list_for_each_entry(zone, zone_list, link) {
1985 		if (dmz_is_buf(zone)) {
1986 			dzone = zone->bzone;
1987 			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1988 				continue;
1989 		} else
1990 			dzone = zone;
1991 		if (dmz_lock_zone_reclaim(dzone))
1992 			return dzone;
1993 	}
1994 
1995 	return NULL;
1996 }
1997 
1998 /*
1999  * Select a buffered sequential zone for reclaim.
2000  */
2001 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd,
2002 						    unsigned int idx)
2003 {
2004 	struct dm_zone *zone;
2005 
2006 	list_for_each_entry(zone, &zmd->dev[idx].map_seq_list, link) {
2007 		if (!zone->bzone)
2008 			continue;
2009 		if (dmz_lock_zone_reclaim(zone))
2010 			return zone;
2011 	}
2012 
2013 	return NULL;
2014 }
2015 
2016 /*
2017  * Select a zone for reclaim.
2018  */
2019 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd,
2020 					 unsigned int dev_idx, bool idle)
2021 {
2022 	struct dm_zone *zone = NULL;
2023 
2024 	/*
2025 	 * Search for a zone candidate to reclaim: 2 cases are possible.
2026 	 * (1) There is no free sequential zones. Then a random data zone
2027 	 *     cannot be reclaimed. So choose a sequential zone to reclaim so
2028 	 *     that afterward a random zone can be reclaimed.
2029 	 * (2) At least one free sequential zone is available, then choose
2030 	 *     the oldest random zone (data or buffer) that can be locked.
2031 	 */
2032 	dmz_lock_map(zmd);
2033 	if (list_empty(&zmd->reserved_seq_zones_list))
2034 		zone = dmz_get_seq_zone_for_reclaim(zmd, dev_idx);
2035 	if (!zone)
2036 		zone = dmz_get_rnd_zone_for_reclaim(zmd, dev_idx, idle);
2037 	dmz_unlock_map(zmd);
2038 
2039 	return zone;
2040 }
2041 
2042 /*
2043  * Get the zone mapping a chunk, if the chunk is mapped already.
2044  * If no mapping exist and the operation is WRITE, a zone is
2045  * allocated and used to map the chunk.
2046  * The zone returned will be set to the active state.
2047  */
2048 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
2049 {
2050 	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
2051 	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
2052 	int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
2053 	unsigned int dzone_id;
2054 	struct dm_zone *dzone = NULL;
2055 	int ret = 0;
2056 	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2057 
2058 	dmz_lock_map(zmd);
2059 again:
2060 	/* Get the chunk mapping */
2061 	dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
2062 	if (dzone_id == DMZ_MAP_UNMAPPED) {
2063 		/*
2064 		 * Read or discard in unmapped chunks are fine. But for
2065 		 * writes, we need a mapping, so get one.
2066 		 */
2067 		if (op != REQ_OP_WRITE)
2068 			goto out;
2069 
2070 		/* Allocate a random zone */
2071 		dzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2072 		if (!dzone) {
2073 			if (dmz_dev_is_dying(zmd)) {
2074 				dzone = ERR_PTR(-EIO);
2075 				goto out;
2076 			}
2077 			dmz_wait_for_free_zones(zmd);
2078 			goto again;
2079 		}
2080 
2081 		dmz_map_zone(zmd, dzone, chunk);
2082 
2083 	} else {
2084 		/* The chunk is already mapped: get the mapping zone */
2085 		dzone = dmz_get(zmd, dzone_id);
2086 		if (!dzone) {
2087 			dzone = ERR_PTR(-EIO);
2088 			goto out;
2089 		}
2090 		if (dzone->chunk != chunk) {
2091 			dzone = ERR_PTR(-EIO);
2092 			goto out;
2093 		}
2094 
2095 		/* Repair write pointer if the sequential dzone has error */
2096 		if (dmz_seq_write_err(dzone)) {
2097 			ret = dmz_handle_seq_write_err(zmd, dzone);
2098 			if (ret) {
2099 				dzone = ERR_PTR(-EIO);
2100 				goto out;
2101 			}
2102 			clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
2103 		}
2104 	}
2105 
2106 	/*
2107 	 * If the zone is being reclaimed, the chunk mapping may change
2108 	 * to a different zone. So wait for reclaim and retry. Otherwise,
2109 	 * activate the zone (this will prevent reclaim from touching it).
2110 	 */
2111 	if (dmz_in_reclaim(dzone)) {
2112 		dmz_wait_for_reclaim(zmd, dzone);
2113 		goto again;
2114 	}
2115 	dmz_activate_zone(dzone);
2116 	dmz_lru_zone(zmd, dzone);
2117 out:
2118 	dmz_unlock_map(zmd);
2119 
2120 	return dzone;
2121 }
2122 
2123 /*
2124  * Write and discard change the block validity of data zones and their buffer
2125  * zones. Check here that valid blocks are still present. If all blocks are
2126  * invalid, the zones can be unmapped on the fly without waiting for reclaim
2127  * to do it.
2128  */
2129 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
2130 {
2131 	struct dm_zone *bzone;
2132 
2133 	dmz_lock_map(zmd);
2134 
2135 	bzone = dzone->bzone;
2136 	if (bzone) {
2137 		if (dmz_weight(bzone))
2138 			dmz_lru_zone(zmd, bzone);
2139 		else {
2140 			/* Empty buffer zone: reclaim it */
2141 			dmz_unmap_zone(zmd, bzone);
2142 			dmz_free_zone(zmd, bzone);
2143 			bzone = NULL;
2144 		}
2145 	}
2146 
2147 	/* Deactivate the data zone */
2148 	dmz_deactivate_zone(dzone);
2149 	if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
2150 		dmz_lru_zone(zmd, dzone);
2151 	else {
2152 		/* Unbuffered inactive empty data zone: reclaim it */
2153 		dmz_unmap_zone(zmd, dzone);
2154 		dmz_free_zone(zmd, dzone);
2155 	}
2156 
2157 	dmz_unlock_map(zmd);
2158 }
2159 
2160 /*
2161  * Allocate and map a random zone to buffer a chunk
2162  * already mapped to a sequential zone.
2163  */
2164 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
2165 				     struct dm_zone *dzone)
2166 {
2167 	struct dm_zone *bzone;
2168 	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2169 
2170 	dmz_lock_map(zmd);
2171 again:
2172 	bzone = dzone->bzone;
2173 	if (bzone)
2174 		goto out;
2175 
2176 	/* Allocate a random zone */
2177 	bzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2178 	if (!bzone) {
2179 		if (dmz_dev_is_dying(zmd)) {
2180 			bzone = ERR_PTR(-EIO);
2181 			goto out;
2182 		}
2183 		dmz_wait_for_free_zones(zmd);
2184 		goto again;
2185 	}
2186 
2187 	/* Update the chunk mapping */
2188 	dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id);
2189 
2190 	set_bit(DMZ_BUF, &bzone->flags);
2191 	bzone->chunk = dzone->chunk;
2192 	bzone->bzone = dzone;
2193 	dzone->bzone = bzone;
2194 	if (dmz_is_cache(bzone))
2195 		list_add_tail(&bzone->link, &zmd->map_cache_list);
2196 	else
2197 		list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
2198 out:
2199 	dmz_unlock_map(zmd);
2200 
2201 	return bzone;
2202 }
2203 
2204 /*
2205  * Get an unmapped (free) zone.
2206  * This must be called with the mapping lock held.
2207  */
2208 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned int dev_idx,
2209 			       unsigned long flags)
2210 {
2211 	struct list_head *list;
2212 	struct dm_zone *zone;
2213 	int i;
2214 
2215 	/* Schedule reclaim to ensure free zones are available */
2216 	if (!(flags & DMZ_ALLOC_RECLAIM)) {
2217 		for (i = 0; i < zmd->nr_devs; i++)
2218 			dmz_schedule_reclaim(zmd->dev[i].reclaim);
2219 	}
2220 
2221 	i = 0;
2222 again:
2223 	if (flags & DMZ_ALLOC_CACHE)
2224 		list = &zmd->unmap_cache_list;
2225 	else if (flags & DMZ_ALLOC_RND)
2226 		list = &zmd->dev[dev_idx].unmap_rnd_list;
2227 	else
2228 		list = &zmd->dev[dev_idx].unmap_seq_list;
2229 
2230 	if (list_empty(list)) {
2231 		/*
2232 		 * No free zone: return NULL if this is for not reclaim.
2233 		 */
2234 		if (!(flags & DMZ_ALLOC_RECLAIM))
2235 			return NULL;
2236 		/*
2237 		 * Try to allocate from other devices
2238 		 */
2239 		if (i < zmd->nr_devs) {
2240 			dev_idx = (dev_idx + 1) % zmd->nr_devs;
2241 			i++;
2242 			goto again;
2243 		}
2244 
2245 		/*
2246 		 * Fallback to the reserved sequential zones
2247 		 */
2248 		zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list,
2249 						struct dm_zone, link);
2250 		if (zone) {
2251 			list_del_init(&zone->link);
2252 			atomic_dec(&zmd->nr_reserved_seq_zones);
2253 		}
2254 		return zone;
2255 	}
2256 
2257 	zone = list_first_entry(list, struct dm_zone, link);
2258 	list_del_init(&zone->link);
2259 
2260 	if (dmz_is_cache(zone))
2261 		atomic_dec(&zmd->unmap_nr_cache);
2262 	else if (dmz_is_rnd(zone))
2263 		atomic_dec(&zone->dev->unmap_nr_rnd);
2264 	else
2265 		atomic_dec(&zone->dev->unmap_nr_seq);
2266 
2267 	if (dmz_is_offline(zone)) {
2268 		dmz_zmd_warn(zmd, "Zone %u is offline", zone->id);
2269 		zone = NULL;
2270 		goto again;
2271 	}
2272 	if (dmz_is_meta(zone)) {
2273 		dmz_zmd_warn(zmd, "Zone %u has metadata", zone->id);
2274 		zone = NULL;
2275 		goto again;
2276 	}
2277 	return zone;
2278 }
2279 
2280 /*
2281  * Free a zone.
2282  * This must be called with the mapping lock held.
2283  */
2284 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2285 {
2286 	/* If this is a sequential zone, reset it */
2287 	if (dmz_is_seq(zone))
2288 		dmz_reset_zone(zmd, zone);
2289 
2290 	/* Return the zone to its type unmap list */
2291 	if (dmz_is_cache(zone)) {
2292 		list_add_tail(&zone->link, &zmd->unmap_cache_list);
2293 		atomic_inc(&zmd->unmap_nr_cache);
2294 	} else if (dmz_is_rnd(zone)) {
2295 		list_add_tail(&zone->link, &zone->dev->unmap_rnd_list);
2296 		atomic_inc(&zone->dev->unmap_nr_rnd);
2297 	} else if (dmz_is_reserved(zone)) {
2298 		list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
2299 		atomic_inc(&zmd->nr_reserved_seq_zones);
2300 	} else {
2301 		list_add_tail(&zone->link, &zone->dev->unmap_seq_list);
2302 		atomic_inc(&zone->dev->unmap_nr_seq);
2303 	}
2304 
2305 	wake_up_all(&zmd->free_wq);
2306 }
2307 
2308 /*
2309  * Map a chunk to a zone.
2310  * This must be called with the mapping lock held.
2311  */
2312 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
2313 		  unsigned int chunk)
2314 {
2315 	/* Set the chunk mapping */
2316 	dmz_set_chunk_mapping(zmd, chunk, dzone->id,
2317 			      DMZ_MAP_UNMAPPED);
2318 	dzone->chunk = chunk;
2319 	if (dmz_is_cache(dzone))
2320 		list_add_tail(&dzone->link, &zmd->map_cache_list);
2321 	else if (dmz_is_rnd(dzone))
2322 		list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
2323 	else
2324 		list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
2325 }
2326 
2327 /*
2328  * Unmap a zone.
2329  * This must be called with the mapping lock held.
2330  */
2331 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2332 {
2333 	unsigned int chunk = zone->chunk;
2334 	unsigned int dzone_id;
2335 
2336 	if (chunk == DMZ_MAP_UNMAPPED) {
2337 		/* Already unmapped */
2338 		return;
2339 	}
2340 
2341 	if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
2342 		/*
2343 		 * Unmapping the chunk buffer zone: clear only
2344 		 * the chunk buffer mapping
2345 		 */
2346 		dzone_id = zone->bzone->id;
2347 		zone->bzone->bzone = NULL;
2348 		zone->bzone = NULL;
2349 
2350 	} else {
2351 		/*
2352 		 * Unmapping the chunk data zone: the zone must
2353 		 * not be buffered.
2354 		 */
2355 		if (WARN_ON(zone->bzone)) {
2356 			zone->bzone->bzone = NULL;
2357 			zone->bzone = NULL;
2358 		}
2359 		dzone_id = DMZ_MAP_UNMAPPED;
2360 	}
2361 
2362 	dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
2363 
2364 	zone->chunk = DMZ_MAP_UNMAPPED;
2365 	list_del_init(&zone->link);
2366 }
2367 
2368 /*
2369  * Set @nr_bits bits in @bitmap starting from @bit.
2370  * Return the number of bits changed from 0 to 1.
2371  */
2372 static unsigned int dmz_set_bits(unsigned long *bitmap,
2373 				 unsigned int bit, unsigned int nr_bits)
2374 {
2375 	unsigned long *addr;
2376 	unsigned int end = bit + nr_bits;
2377 	unsigned int n = 0;
2378 
2379 	while (bit < end) {
2380 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2381 		    ((end - bit) >= BITS_PER_LONG)) {
2382 			/* Try to set the whole word at once */
2383 			addr = bitmap + BIT_WORD(bit);
2384 			if (*addr == 0) {
2385 				*addr = ULONG_MAX;
2386 				n += BITS_PER_LONG;
2387 				bit += BITS_PER_LONG;
2388 				continue;
2389 			}
2390 		}
2391 
2392 		if (!test_and_set_bit(bit, bitmap))
2393 			n++;
2394 		bit++;
2395 	}
2396 
2397 	return n;
2398 }
2399 
2400 /*
2401  * Get the bitmap block storing the bit for chunk_block in zone.
2402  */
2403 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
2404 					 struct dm_zone *zone,
2405 					 sector_t chunk_block)
2406 {
2407 	sector_t bitmap_block = 1 + zmd->nr_map_blocks +
2408 		(sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) +
2409 		(chunk_block >> DMZ_BLOCK_SHIFT_BITS);
2410 
2411 	return dmz_get_mblock(zmd, bitmap_block);
2412 }
2413 
2414 /*
2415  * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
2416  */
2417 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2418 			  struct dm_zone *to_zone)
2419 {
2420 	struct dmz_mblock *from_mblk, *to_mblk;
2421 	sector_t chunk_block = 0;
2422 
2423 	/* Get the zones bitmap blocks */
2424 	while (chunk_block < zmd->zone_nr_blocks) {
2425 		from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
2426 		if (IS_ERR(from_mblk))
2427 			return PTR_ERR(from_mblk);
2428 		to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
2429 		if (IS_ERR(to_mblk)) {
2430 			dmz_release_mblock(zmd, from_mblk);
2431 			return PTR_ERR(to_mblk);
2432 		}
2433 
2434 		memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
2435 		dmz_dirty_mblock(zmd, to_mblk);
2436 
2437 		dmz_release_mblock(zmd, to_mblk);
2438 		dmz_release_mblock(zmd, from_mblk);
2439 
2440 		chunk_block += zmd->zone_bits_per_mblk;
2441 	}
2442 
2443 	to_zone->weight = from_zone->weight;
2444 
2445 	return 0;
2446 }
2447 
2448 /*
2449  * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
2450  * starting from chunk_block.
2451  */
2452 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2453 			   struct dm_zone *to_zone, sector_t chunk_block)
2454 {
2455 	unsigned int nr_blocks;
2456 	int ret;
2457 
2458 	/* Get the zones bitmap blocks */
2459 	while (chunk_block < zmd->zone_nr_blocks) {
2460 		/* Get a valid region from the source zone */
2461 		ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2462 		if (ret <= 0)
2463 			return ret;
2464 
2465 		nr_blocks = ret;
2466 		ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2467 		if (ret)
2468 			return ret;
2469 
2470 		chunk_block += nr_blocks;
2471 	}
2472 
2473 	return 0;
2474 }
2475 
2476 /*
2477  * Validate all the blocks in the range [block..block+nr_blocks-1].
2478  */
2479 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2480 			sector_t chunk_block, unsigned int nr_blocks)
2481 {
2482 	unsigned int count, bit, nr_bits;
2483 	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
2484 	struct dmz_mblock *mblk;
2485 	unsigned int n = 0;
2486 
2487 	dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks",
2488 		      zone->id, (unsigned long long)chunk_block,
2489 		      nr_blocks);
2490 
2491 	WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2492 
2493 	while (nr_blocks) {
2494 		/* Get bitmap block */
2495 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2496 		if (IS_ERR(mblk))
2497 			return PTR_ERR(mblk);
2498 
2499 		/* Set bits */
2500 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2501 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2502 
2503 		count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2504 		if (count) {
2505 			dmz_dirty_mblock(zmd, mblk);
2506 			n += count;
2507 		}
2508 		dmz_release_mblock(zmd, mblk);
2509 
2510 		nr_blocks -= nr_bits;
2511 		chunk_block += nr_bits;
2512 	}
2513 
2514 	if (likely(zone->weight + n <= zone_nr_blocks))
2515 		zone->weight += n;
2516 	else {
2517 		dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u",
2518 			     zone->id, zone->weight,
2519 			     zone_nr_blocks - n);
2520 		zone->weight = zone_nr_blocks;
2521 	}
2522 
2523 	return 0;
2524 }
2525 
2526 /*
2527  * Clear nr_bits bits in bitmap starting from bit.
2528  * Return the number of bits cleared.
2529  */
2530 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2531 {
2532 	unsigned long *addr;
2533 	int end = bit + nr_bits;
2534 	int n = 0;
2535 
2536 	while (bit < end) {
2537 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2538 		    ((end - bit) >= BITS_PER_LONG)) {
2539 			/* Try to clear whole word at once */
2540 			addr = bitmap + BIT_WORD(bit);
2541 			if (*addr == ULONG_MAX) {
2542 				*addr = 0;
2543 				n += BITS_PER_LONG;
2544 				bit += BITS_PER_LONG;
2545 				continue;
2546 			}
2547 		}
2548 
2549 		if (test_and_clear_bit(bit, bitmap))
2550 			n++;
2551 		bit++;
2552 	}
2553 
2554 	return n;
2555 }
2556 
2557 /*
2558  * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2559  */
2560 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2561 			  sector_t chunk_block, unsigned int nr_blocks)
2562 {
2563 	unsigned int count, bit, nr_bits;
2564 	struct dmz_mblock *mblk;
2565 	unsigned int n = 0;
2566 
2567 	dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks",
2568 		      zone->id, (u64)chunk_block, nr_blocks);
2569 
2570 	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2571 
2572 	while (nr_blocks) {
2573 		/* Get bitmap block */
2574 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2575 		if (IS_ERR(mblk))
2576 			return PTR_ERR(mblk);
2577 
2578 		/* Clear bits */
2579 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2580 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2581 
2582 		count = dmz_clear_bits((unsigned long *)mblk->data,
2583 				       bit, nr_bits);
2584 		if (count) {
2585 			dmz_dirty_mblock(zmd, mblk);
2586 			n += count;
2587 		}
2588 		dmz_release_mblock(zmd, mblk);
2589 
2590 		nr_blocks -= nr_bits;
2591 		chunk_block += nr_bits;
2592 	}
2593 
2594 	if (zone->weight >= n)
2595 		zone->weight -= n;
2596 	else {
2597 		dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u",
2598 			     zone->id, zone->weight, n);
2599 		zone->weight = 0;
2600 	}
2601 
2602 	return 0;
2603 }
2604 
2605 /*
2606  * Get a block bit value.
2607  */
2608 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2609 			  sector_t chunk_block)
2610 {
2611 	struct dmz_mblock *mblk;
2612 	int ret;
2613 
2614 	WARN_ON(chunk_block >= zmd->zone_nr_blocks);
2615 
2616 	/* Get bitmap block */
2617 	mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2618 	if (IS_ERR(mblk))
2619 		return PTR_ERR(mblk);
2620 
2621 	/* Get offset */
2622 	ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2623 		       (unsigned long *) mblk->data) != 0;
2624 
2625 	dmz_release_mblock(zmd, mblk);
2626 
2627 	return ret;
2628 }
2629 
2630 /*
2631  * Return the number of blocks from chunk_block to the first block with a bit
2632  * value specified by set. Search at most nr_blocks blocks from chunk_block.
2633  */
2634 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2635 				 sector_t chunk_block, unsigned int nr_blocks,
2636 				 int set)
2637 {
2638 	struct dmz_mblock *mblk;
2639 	unsigned int bit, set_bit, nr_bits;
2640 	unsigned int zone_bits = zmd->zone_bits_per_mblk;
2641 	unsigned long *bitmap;
2642 	int n = 0;
2643 
2644 	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2645 
2646 	while (nr_blocks) {
2647 		/* Get bitmap block */
2648 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2649 		if (IS_ERR(mblk))
2650 			return PTR_ERR(mblk);
2651 
2652 		/* Get offset */
2653 		bitmap = (unsigned long *) mblk->data;
2654 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2655 		nr_bits = min(nr_blocks, zone_bits - bit);
2656 		if (set)
2657 			set_bit = find_next_bit(bitmap, zone_bits, bit);
2658 		else
2659 			set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2660 		dmz_release_mblock(zmd, mblk);
2661 
2662 		n += set_bit - bit;
2663 		if (set_bit < zone_bits)
2664 			break;
2665 
2666 		nr_blocks -= nr_bits;
2667 		chunk_block += nr_bits;
2668 	}
2669 
2670 	return n;
2671 }
2672 
2673 /*
2674  * Test if chunk_block is valid. If it is, the number of consecutive
2675  * valid blocks from chunk_block will be returned.
2676  */
2677 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2678 		    sector_t chunk_block)
2679 {
2680 	int valid;
2681 
2682 	valid = dmz_test_block(zmd, zone, chunk_block);
2683 	if (valid <= 0)
2684 		return valid;
2685 
2686 	/* The block is valid: get the number of valid blocks from block */
2687 	return dmz_to_next_set_block(zmd, zone, chunk_block,
2688 				     zmd->zone_nr_blocks - chunk_block, 0);
2689 }
2690 
2691 /*
2692  * Find the first valid block from @chunk_block in @zone.
2693  * If such a block is found, its number is returned using
2694  * @chunk_block and the total number of valid blocks from @chunk_block
2695  * is returned.
2696  */
2697 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2698 			  sector_t *chunk_block)
2699 {
2700 	sector_t start_block = *chunk_block;
2701 	int ret;
2702 
2703 	ret = dmz_to_next_set_block(zmd, zone, start_block,
2704 				    zmd->zone_nr_blocks - start_block, 1);
2705 	if (ret < 0)
2706 		return ret;
2707 
2708 	start_block += ret;
2709 	*chunk_block = start_block;
2710 
2711 	return dmz_to_next_set_block(zmd, zone, start_block,
2712 				     zmd->zone_nr_blocks - start_block, 0);
2713 }
2714 
2715 /*
2716  * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2717  */
2718 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2719 {
2720 	unsigned long *addr;
2721 	int end = bit + nr_bits;
2722 	int n = 0;
2723 
2724 	while (bit < end) {
2725 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2726 		    ((end - bit) >= BITS_PER_LONG)) {
2727 			addr = (unsigned long *)bitmap + BIT_WORD(bit);
2728 			if (*addr == ULONG_MAX) {
2729 				n += BITS_PER_LONG;
2730 				bit += BITS_PER_LONG;
2731 				continue;
2732 			}
2733 		}
2734 
2735 		if (test_bit(bit, bitmap))
2736 			n++;
2737 		bit++;
2738 	}
2739 
2740 	return n;
2741 }
2742 
2743 /*
2744  * Get a zone weight.
2745  */
2746 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2747 {
2748 	struct dmz_mblock *mblk;
2749 	sector_t chunk_block = 0;
2750 	unsigned int bit, nr_bits;
2751 	unsigned int nr_blocks = zmd->zone_nr_blocks;
2752 	void *bitmap;
2753 	int n = 0;
2754 
2755 	while (nr_blocks) {
2756 		/* Get bitmap block */
2757 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2758 		if (IS_ERR(mblk)) {
2759 			n = 0;
2760 			break;
2761 		}
2762 
2763 		/* Count bits in this block */
2764 		bitmap = mblk->data;
2765 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2766 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2767 		n += dmz_count_bits(bitmap, bit, nr_bits);
2768 
2769 		dmz_release_mblock(zmd, mblk);
2770 
2771 		nr_blocks -= nr_bits;
2772 		chunk_block += nr_bits;
2773 	}
2774 
2775 	zone->weight = n;
2776 }
2777 
2778 /*
2779  * Cleanup the zoned metadata resources.
2780  */
2781 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2782 {
2783 	struct rb_root *root;
2784 	struct dmz_mblock *mblk, *next;
2785 	int i;
2786 
2787 	/* Release zone mapping resources */
2788 	if (zmd->map_mblk) {
2789 		for (i = 0; i < zmd->nr_map_blocks; i++)
2790 			dmz_release_mblock(zmd, zmd->map_mblk[i]);
2791 		kfree(zmd->map_mblk);
2792 		zmd->map_mblk = NULL;
2793 	}
2794 
2795 	/* Release super blocks */
2796 	for (i = 0; i < 2; i++) {
2797 		if (zmd->sb[i].mblk) {
2798 			dmz_free_mblock(zmd, zmd->sb[i].mblk);
2799 			zmd->sb[i].mblk = NULL;
2800 		}
2801 	}
2802 
2803 	/* Free cached blocks */
2804 	while (!list_empty(&zmd->mblk_dirty_list)) {
2805 		mblk = list_first_entry(&zmd->mblk_dirty_list,
2806 					struct dmz_mblock, link);
2807 		dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)",
2808 			     (u64)mblk->no, mblk->ref);
2809 		list_del_init(&mblk->link);
2810 		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2811 		dmz_free_mblock(zmd, mblk);
2812 	}
2813 
2814 	while (!list_empty(&zmd->mblk_lru_list)) {
2815 		mblk = list_first_entry(&zmd->mblk_lru_list,
2816 					struct dmz_mblock, link);
2817 		list_del_init(&mblk->link);
2818 		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2819 		dmz_free_mblock(zmd, mblk);
2820 	}
2821 
2822 	/* Sanity checks: the mblock rbtree should now be empty */
2823 	root = &zmd->mblk_rbtree;
2824 	rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2825 		dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree",
2826 			     (u64)mblk->no, mblk->ref);
2827 		mblk->ref = 0;
2828 		dmz_free_mblock(zmd, mblk);
2829 	}
2830 
2831 	/* Free the zone descriptors */
2832 	dmz_drop_zones(zmd);
2833 
2834 	mutex_destroy(&zmd->mblk_flush_lock);
2835 	mutex_destroy(&zmd->map_lock);
2836 }
2837 
2838 static void dmz_print_dev(struct dmz_metadata *zmd, int num)
2839 {
2840 	struct dmz_dev *dev = &zmd->dev[num];
2841 
2842 	if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE)
2843 		dmz_dev_info(dev, "Regular block device");
2844 	else
2845 		dmz_dev_info(dev, "Host-%s zoned block device",
2846 			     bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2847 			     "aware" : "managed");
2848 	if (zmd->sb_version > 1) {
2849 		sector_t sector_offset =
2850 			dev->zone_offset << zmd->zone_nr_sectors_shift;
2851 
2852 		dmz_dev_info(dev, "  %llu 512-byte logical sectors (offset %llu)",
2853 			     (u64)dev->capacity, (u64)sector_offset);
2854 		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors (offset %llu)",
2855 			     dev->nr_zones, (u64)zmd->zone_nr_sectors,
2856 			     (u64)dev->zone_offset);
2857 	} else {
2858 		dmz_dev_info(dev, "  %llu 512-byte logical sectors",
2859 			     (u64)dev->capacity);
2860 		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors",
2861 			     dev->nr_zones, (u64)zmd->zone_nr_sectors);
2862 	}
2863 }
2864 
2865 /*
2866  * Initialize the zoned metadata.
2867  */
2868 int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev,
2869 		     struct dmz_metadata **metadata,
2870 		     const char *devname)
2871 {
2872 	struct dmz_metadata *zmd;
2873 	unsigned int i;
2874 	struct dm_zone *zone;
2875 	int ret;
2876 
2877 	zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2878 	if (!zmd)
2879 		return -ENOMEM;
2880 
2881 	strcpy(zmd->devname, devname);
2882 	zmd->dev = dev;
2883 	zmd->nr_devs = num_dev;
2884 	zmd->mblk_rbtree = RB_ROOT;
2885 	init_rwsem(&zmd->mblk_sem);
2886 	mutex_init(&zmd->mblk_flush_lock);
2887 	spin_lock_init(&zmd->mblk_lock);
2888 	INIT_LIST_HEAD(&zmd->mblk_lru_list);
2889 	INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2890 
2891 	mutex_init(&zmd->map_lock);
2892 
2893 	atomic_set(&zmd->unmap_nr_cache, 0);
2894 	INIT_LIST_HEAD(&zmd->unmap_cache_list);
2895 	INIT_LIST_HEAD(&zmd->map_cache_list);
2896 
2897 	atomic_set(&zmd->nr_reserved_seq_zones, 0);
2898 	INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2899 
2900 	init_waitqueue_head(&zmd->free_wq);
2901 
2902 	/* Initialize zone descriptors */
2903 	ret = dmz_init_zones(zmd);
2904 	if (ret)
2905 		goto err;
2906 
2907 	/* Get super block */
2908 	ret = dmz_load_sb(zmd);
2909 	if (ret)
2910 		goto err;
2911 
2912 	/* Set metadata zones starting from sb_zone */
2913 	for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2914 		zone = dmz_get(zmd, zmd->sb[0].zone->id + i);
2915 		if (!zone) {
2916 			dmz_zmd_err(zmd,
2917 				    "metadata zone %u not present", i);
2918 			ret = -ENXIO;
2919 			goto err;
2920 		}
2921 		if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) {
2922 			dmz_zmd_err(zmd,
2923 				    "metadata zone %d is not random", i);
2924 			ret = -ENXIO;
2925 			goto err;
2926 		}
2927 		set_bit(DMZ_META, &zone->flags);
2928 	}
2929 	/* Load mapping table */
2930 	ret = dmz_load_mapping(zmd);
2931 	if (ret)
2932 		goto err;
2933 
2934 	/*
2935 	 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2936 	 * blocks and enough blocks to be able to cache the bitmap blocks of
2937 	 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2938 	 * the cache to add 512 more metadata blocks.
2939 	 */
2940 	zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2941 	zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2942 	zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2943 	zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2944 	zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2945 
2946 	/* Metadata cache shrinker */
2947 	ret = register_shrinker(&zmd->mblk_shrinker);
2948 	if (ret) {
2949 		dmz_zmd_err(zmd, "Register metadata cache shrinker failed");
2950 		goto err;
2951 	}
2952 
2953 	dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version);
2954 	for (i = 0; i < zmd->nr_devs; i++)
2955 		dmz_print_dev(zmd, i);
2956 
2957 	dmz_zmd_info(zmd, "  %u zones of %llu 512-byte logical sectors",
2958 		     zmd->nr_zones, (u64)zmd->zone_nr_sectors);
2959 	dmz_zmd_debug(zmd, "  %u metadata zones",
2960 		      zmd->nr_meta_zones * 2);
2961 	dmz_zmd_debug(zmd, "  %u data zones for %u chunks",
2962 		      zmd->nr_data_zones, zmd->nr_chunks);
2963 	dmz_zmd_debug(zmd, "    %u cache zones (%u unmapped)",
2964 		      zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache));
2965 	for (i = 0; i < zmd->nr_devs; i++) {
2966 		dmz_zmd_debug(zmd, "    %u random zones (%u unmapped)",
2967 			      dmz_nr_rnd_zones(zmd, i),
2968 			      dmz_nr_unmap_rnd_zones(zmd, i));
2969 		dmz_zmd_debug(zmd, "    %u sequential zones (%u unmapped)",
2970 			      dmz_nr_seq_zones(zmd, i),
2971 			      dmz_nr_unmap_seq_zones(zmd, i));
2972 	}
2973 	dmz_zmd_debug(zmd, "  %u reserved sequential data zones",
2974 		      zmd->nr_reserved_seq);
2975 	dmz_zmd_debug(zmd, "Format:");
2976 	dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)",
2977 		      zmd->nr_meta_blocks, zmd->max_nr_mblks);
2978 	dmz_zmd_debug(zmd, "  %u data zone mapping blocks",
2979 		      zmd->nr_map_blocks);
2980 	dmz_zmd_debug(zmd, "  %u bitmap blocks",
2981 		      zmd->nr_bitmap_blocks);
2982 
2983 	*metadata = zmd;
2984 
2985 	return 0;
2986 err:
2987 	dmz_cleanup_metadata(zmd);
2988 	kfree(zmd);
2989 	*metadata = NULL;
2990 
2991 	return ret;
2992 }
2993 
2994 /*
2995  * Cleanup the zoned metadata resources.
2996  */
2997 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2998 {
2999 	unregister_shrinker(&zmd->mblk_shrinker);
3000 	dmz_cleanup_metadata(zmd);
3001 	kfree(zmd);
3002 }
3003 
3004 /*
3005  * Check zone information on resume.
3006  */
3007 int dmz_resume_metadata(struct dmz_metadata *zmd)
3008 {
3009 	struct dm_zone *zone;
3010 	sector_t wp_block;
3011 	unsigned int i;
3012 	int ret;
3013 
3014 	/* Check zones */
3015 	for (i = 0; i < zmd->nr_zones; i++) {
3016 		zone = dmz_get(zmd, i);
3017 		if (!zone) {
3018 			dmz_zmd_err(zmd, "Unable to get zone %u", i);
3019 			return -EIO;
3020 		}
3021 		wp_block = zone->wp_block;
3022 
3023 		ret = dmz_update_zone(zmd, zone);
3024 		if (ret) {
3025 			dmz_zmd_err(zmd, "Broken zone %u", i);
3026 			return ret;
3027 		}
3028 
3029 		if (dmz_is_offline(zone)) {
3030 			dmz_zmd_warn(zmd, "Zone %u is offline", i);
3031 			continue;
3032 		}
3033 
3034 		/* Check write pointer */
3035 		if (!dmz_is_seq(zone))
3036 			zone->wp_block = 0;
3037 		else if (zone->wp_block != wp_block) {
3038 			dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)",
3039 				    i, (u64)zone->wp_block, (u64)wp_block);
3040 			zone->wp_block = wp_block;
3041 			dmz_invalidate_blocks(zmd, zone, zone->wp_block,
3042 					      zmd->zone_nr_blocks - zone->wp_block);
3043 		}
3044 	}
3045 
3046 	return 0;
3047 }
3048