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, enum req_op 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 (is %llu expected %llu)",
1018 			    sb_block, (u64)dsb->zone->id << zmd->zone_nr_blocks_shift);
1019 		return -EINVAL;
1020 	}
1021 	if (zmd->sb_version > 1) {
1022 		uuid_t sb_uuid;
1023 
1024 		import_uuid(&sb_uuid, sb->dmz_uuid);
1025 		if (uuid_is_null(&sb_uuid)) {
1026 			dmz_dev_err(dev, "NULL DM-Zoned uuid");
1027 			return -ENXIO;
1028 		} else if (uuid_is_null(&zmd->uuid)) {
1029 			uuid_copy(&zmd->uuid, &sb_uuid);
1030 		} else if (!uuid_equal(&zmd->uuid, &sb_uuid)) {
1031 			dmz_dev_err(dev, "mismatching DM-Zoned uuid, is %pUl expected %pUl",
1032 				    &sb_uuid, &zmd->uuid);
1033 			return -ENXIO;
1034 		}
1035 		if (!strlen(zmd->label))
1036 			memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE);
1037 		else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) {
1038 			dmz_dev_err(dev, "mismatching DM-Zoned label, is %s expected %s",
1039 				    sb->dmz_label, zmd->label);
1040 			return -ENXIO;
1041 		}
1042 		import_uuid(&dev->uuid, sb->dev_uuid);
1043 		if (uuid_is_null(&dev->uuid)) {
1044 			dmz_dev_err(dev, "NULL device uuid");
1045 			return -ENXIO;
1046 		}
1047 
1048 		if (tertiary) {
1049 			/*
1050 			 * Generation number should be 0, but it doesn't
1051 			 * really matter if it isn't.
1052 			 */
1053 			if (gen != 0)
1054 				dmz_dev_warn(dev, "Invalid generation %llu",
1055 					    gen);
1056 			return 0;
1057 		}
1058 	}
1059 
1060 	nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1)
1061 		>> zmd->zone_nr_blocks_shift;
1062 	if (!nr_meta_zones ||
1063 	    (zmd->nr_devs <= 1 && nr_meta_zones >= zmd->nr_rnd_zones) ||
1064 	    (zmd->nr_devs > 1 && nr_meta_zones >= zmd->nr_cache_zones)) {
1065 		dmz_dev_err(dev, "Invalid number of metadata blocks");
1066 		return -ENXIO;
1067 	}
1068 
1069 	if (!le32_to_cpu(sb->nr_reserved_seq) ||
1070 	    le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
1071 		dmz_dev_err(dev, "Invalid number of reserved sequential zones");
1072 		return -ENXIO;
1073 	}
1074 
1075 	nr_data_zones = zmd->nr_useable_zones -
1076 		(nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
1077 	if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
1078 		dmz_dev_err(dev, "Invalid number of chunks %u / %u",
1079 			    le32_to_cpu(sb->nr_chunks), nr_data_zones);
1080 		return -ENXIO;
1081 	}
1082 
1083 	/* OK */
1084 	zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
1085 	zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
1086 	zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
1087 	zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
1088 	zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
1089 	zmd->nr_meta_zones = nr_meta_zones;
1090 	zmd->nr_data_zones = nr_data_zones;
1091 
1092 	return 0;
1093 }
1094 
1095 /*
1096  * Read the first or second super block from disk.
1097  */
1098 static int dmz_read_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1099 {
1100 	dmz_zmd_debug(zmd, "read superblock set %d dev %pg block %llu",
1101 		      set, sb->dev->bdev, sb->block);
1102 
1103 	return dmz_rdwr_block(sb->dev, REQ_OP_READ,
1104 			      sb->block, sb->mblk->page);
1105 }
1106 
1107 /*
1108  * Determine the position of the secondary super blocks on disk.
1109  * This is used only if a corruption of the primary super block
1110  * is detected.
1111  */
1112 static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
1113 {
1114 	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
1115 	struct dmz_mblock *mblk;
1116 	unsigned int zone_id = zmd->sb[0].zone->id;
1117 	int i;
1118 
1119 	/* Allocate a block */
1120 	mblk = dmz_alloc_mblock(zmd, 0);
1121 	if (!mblk)
1122 		return -ENOMEM;
1123 
1124 	zmd->sb[1].mblk = mblk;
1125 	zmd->sb[1].sb = mblk->data;
1126 
1127 	/* Bad first super block: search for the second one */
1128 	zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
1129 	zmd->sb[1].zone = dmz_get(zmd, zone_id + 1);
1130 	zmd->sb[1].dev = zmd->sb[0].dev;
1131 	for (i = 1; i < zmd->nr_rnd_zones; i++) {
1132 		if (dmz_read_sb(zmd, &zmd->sb[1], 1) != 0)
1133 			break;
1134 		if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
1135 			return 0;
1136 		zmd->sb[1].block += zone_nr_blocks;
1137 		zmd->sb[1].zone = dmz_get(zmd, zone_id + i);
1138 	}
1139 
1140 	dmz_free_mblock(zmd, mblk);
1141 	zmd->sb[1].mblk = NULL;
1142 	zmd->sb[1].zone = NULL;
1143 	zmd->sb[1].dev = NULL;
1144 
1145 	return -EIO;
1146 }
1147 
1148 /*
1149  * Read a super block from disk.
1150  */
1151 static int dmz_get_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1152 {
1153 	struct dmz_mblock *mblk;
1154 	int ret;
1155 
1156 	/* Allocate a block */
1157 	mblk = dmz_alloc_mblock(zmd, 0);
1158 	if (!mblk)
1159 		return -ENOMEM;
1160 
1161 	sb->mblk = mblk;
1162 	sb->sb = mblk->data;
1163 
1164 	/* Read super block */
1165 	ret = dmz_read_sb(zmd, sb, set);
1166 	if (ret) {
1167 		dmz_free_mblock(zmd, mblk);
1168 		sb->mblk = NULL;
1169 		return ret;
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 /*
1176  * Recover a metadata set.
1177  */
1178 static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
1179 {
1180 	unsigned int src_set = dst_set ^ 0x1;
1181 	struct page *page;
1182 	int i, ret;
1183 
1184 	dmz_dev_warn(zmd->sb[dst_set].dev,
1185 		     "Metadata set %u invalid: recovering", dst_set);
1186 
1187 	if (dst_set == 0)
1188 		zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1189 	else
1190 		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1191 
1192 	page = alloc_page(GFP_NOIO);
1193 	if (!page)
1194 		return -ENOMEM;
1195 
1196 	/* Copy metadata blocks */
1197 	for (i = 1; i < zmd->nr_meta_blocks; i++) {
1198 		ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ,
1199 				     zmd->sb[src_set].block + i, page);
1200 		if (ret)
1201 			goto out;
1202 		ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE,
1203 				     zmd->sb[dst_set].block + i, page);
1204 		if (ret)
1205 			goto out;
1206 	}
1207 
1208 	/* Finalize with the super block */
1209 	if (!zmd->sb[dst_set].mblk) {
1210 		zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1211 		if (!zmd->sb[dst_set].mblk) {
1212 			ret = -ENOMEM;
1213 			goto out;
1214 		}
1215 		zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1216 	}
1217 
1218 	ret = dmz_write_sb(zmd, dst_set);
1219 out:
1220 	__free_pages(page, 0);
1221 
1222 	return ret;
1223 }
1224 
1225 /*
1226  * Get super block from disk.
1227  */
1228 static int dmz_load_sb(struct dmz_metadata *zmd)
1229 {
1230 	bool sb_good[2] = {false, false};
1231 	u64 sb_gen[2] = {0, 0};
1232 	int ret;
1233 
1234 	if (!zmd->sb[0].zone) {
1235 		dmz_zmd_err(zmd, "Primary super block zone not set");
1236 		return -ENXIO;
1237 	}
1238 
1239 	/* Read and check the primary super block */
1240 	zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1241 	zmd->sb[0].dev = zmd->sb[0].zone->dev;
1242 	ret = dmz_get_sb(zmd, &zmd->sb[0], 0);
1243 	if (ret) {
1244 		dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed");
1245 		return ret;
1246 	}
1247 
1248 	ret = dmz_check_sb(zmd, &zmd->sb[0], false);
1249 
1250 	/* Read and check secondary super block */
1251 	if (ret == 0) {
1252 		sb_good[0] = true;
1253 		if (!zmd->sb[1].zone) {
1254 			unsigned int zone_id =
1255 				zmd->sb[0].zone->id + zmd->nr_meta_zones;
1256 
1257 			zmd->sb[1].zone = dmz_get(zmd, zone_id);
1258 		}
1259 		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1260 		zmd->sb[1].dev = zmd->sb[0].dev;
1261 		ret = dmz_get_sb(zmd, &zmd->sb[1], 1);
1262 	} else
1263 		ret = dmz_lookup_secondary_sb(zmd);
1264 
1265 	if (ret) {
1266 		dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed");
1267 		return ret;
1268 	}
1269 
1270 	ret = dmz_check_sb(zmd, &zmd->sb[1], false);
1271 	if (ret == 0)
1272 		sb_good[1] = true;
1273 
1274 	/* Use highest generation sb first */
1275 	if (!sb_good[0] && !sb_good[1]) {
1276 		dmz_zmd_err(zmd, "No valid super block found");
1277 		return -EIO;
1278 	}
1279 
1280 	if (sb_good[0])
1281 		sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1282 	else {
1283 		ret = dmz_recover_mblocks(zmd, 0);
1284 		if (ret) {
1285 			dmz_dev_err(zmd->sb[0].dev,
1286 				    "Recovery of superblock 0 failed");
1287 			return -EIO;
1288 		}
1289 	}
1290 
1291 	if (sb_good[1])
1292 		sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1293 	else {
1294 		ret = dmz_recover_mblocks(zmd, 1);
1295 
1296 		if (ret) {
1297 			dmz_dev_err(zmd->sb[1].dev,
1298 				    "Recovery of superblock 1 failed");
1299 			return -EIO;
1300 		}
1301 	}
1302 
1303 	if (sb_gen[0] >= sb_gen[1]) {
1304 		zmd->sb_gen = sb_gen[0];
1305 		zmd->mblk_primary = 0;
1306 	} else {
1307 		zmd->sb_gen = sb_gen[1];
1308 		zmd->mblk_primary = 1;
1309 	}
1310 
1311 	dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev,
1312 		      "Using super block %u (gen %llu)",
1313 		      zmd->mblk_primary, zmd->sb_gen);
1314 
1315 	if (zmd->sb_version > 1) {
1316 		int i;
1317 		struct dmz_sb *sb;
1318 
1319 		sb = kzalloc(sizeof(struct dmz_sb), GFP_KERNEL);
1320 		if (!sb)
1321 			return -ENOMEM;
1322 		for (i = 1; i < zmd->nr_devs; i++) {
1323 			sb->block = 0;
1324 			sb->zone = dmz_get(zmd, zmd->dev[i].zone_offset);
1325 			sb->dev = &zmd->dev[i];
1326 			if (!dmz_is_meta(sb->zone)) {
1327 				dmz_dev_err(sb->dev,
1328 					    "Tertiary super block zone %u not marked as metadata zone",
1329 					    sb->zone->id);
1330 				ret = -EINVAL;
1331 				goto out_kfree;
1332 			}
1333 			ret = dmz_get_sb(zmd, sb, i + 1);
1334 			if (ret) {
1335 				dmz_dev_err(sb->dev,
1336 					    "Read tertiary super block failed");
1337 				dmz_free_mblock(zmd, sb->mblk);
1338 				goto out_kfree;
1339 			}
1340 			ret = dmz_check_sb(zmd, sb, true);
1341 			dmz_free_mblock(zmd, sb->mblk);
1342 			if (ret == -EINVAL)
1343 				goto out_kfree;
1344 		}
1345 out_kfree:
1346 		kfree(sb);
1347 	}
1348 	return ret;
1349 }
1350 
1351 /*
1352  * Initialize a zone descriptor.
1353  */
1354 static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data)
1355 {
1356 	struct dmz_dev *dev = data;
1357 	struct dmz_metadata *zmd = dev->metadata;
1358 	int idx = num + dev->zone_offset;
1359 	struct dm_zone *zone;
1360 
1361 	zone = dmz_insert(zmd, idx, dev);
1362 	if (IS_ERR(zone))
1363 		return PTR_ERR(zone);
1364 
1365 	if (blkz->len != zmd->zone_nr_sectors) {
1366 		if (zmd->sb_version > 1) {
1367 			/* Ignore the eventual runt (smaller) zone */
1368 			set_bit(DMZ_OFFLINE, &zone->flags);
1369 			return 0;
1370 		} else if (blkz->start + blkz->len == dev->capacity)
1371 			return 0;
1372 		return -ENXIO;
1373 	}
1374 
1375 	/*
1376 	 * Devices that have zones with a capacity smaller than the zone size
1377 	 * (e.g. NVMe zoned namespaces) are not supported.
1378 	 */
1379 	if (blkz->capacity != blkz->len)
1380 		return -ENXIO;
1381 
1382 	switch (blkz->type) {
1383 	case BLK_ZONE_TYPE_CONVENTIONAL:
1384 		set_bit(DMZ_RND, &zone->flags);
1385 		break;
1386 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1387 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1388 		set_bit(DMZ_SEQ, &zone->flags);
1389 		break;
1390 	default:
1391 		return -ENXIO;
1392 	}
1393 
1394 	if (dmz_is_rnd(zone))
1395 		zone->wp_block = 0;
1396 	else
1397 		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1398 
1399 	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1400 		set_bit(DMZ_OFFLINE, &zone->flags);
1401 	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1402 		set_bit(DMZ_READ_ONLY, &zone->flags);
1403 	else {
1404 		zmd->nr_useable_zones++;
1405 		if (dmz_is_rnd(zone)) {
1406 			zmd->nr_rnd_zones++;
1407 			if (zmd->nr_devs == 1 && !zmd->sb[0].zone) {
1408 				/* Primary super block zone */
1409 				zmd->sb[0].zone = zone;
1410 			}
1411 		}
1412 		if (zmd->nr_devs > 1 && num == 0) {
1413 			/*
1414 			 * Tertiary superblock zones are always at the
1415 			 * start of the zoned devices, so mark them
1416 			 * as metadata zone.
1417 			 */
1418 			set_bit(DMZ_META, &zone->flags);
1419 		}
1420 	}
1421 	return 0;
1422 }
1423 
1424 static int dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev)
1425 {
1426 	int idx;
1427 	sector_t zone_offset = 0;
1428 
1429 	for (idx = 0; idx < dev->nr_zones; idx++) {
1430 		struct dm_zone *zone;
1431 
1432 		zone = dmz_insert(zmd, idx, dev);
1433 		if (IS_ERR(zone))
1434 			return PTR_ERR(zone);
1435 		set_bit(DMZ_CACHE, &zone->flags);
1436 		zone->wp_block = 0;
1437 		zmd->nr_cache_zones++;
1438 		zmd->nr_useable_zones++;
1439 		if (dev->capacity - zone_offset < zmd->zone_nr_sectors) {
1440 			/* Disable runt zone */
1441 			set_bit(DMZ_OFFLINE, &zone->flags);
1442 			break;
1443 		}
1444 		zone_offset += zmd->zone_nr_sectors;
1445 	}
1446 	return 0;
1447 }
1448 
1449 /*
1450  * Free zones descriptors.
1451  */
1452 static void dmz_drop_zones(struct dmz_metadata *zmd)
1453 {
1454 	int idx;
1455 
1456 	for (idx = 0; idx < zmd->nr_zones; idx++) {
1457 		struct dm_zone *zone = xa_load(&zmd->zones, idx);
1458 
1459 		kfree(zone);
1460 		xa_erase(&zmd->zones, idx);
1461 	}
1462 	xa_destroy(&zmd->zones);
1463 }
1464 
1465 /*
1466  * Allocate and initialize zone descriptors using the zone
1467  * information from disk.
1468  */
1469 static int dmz_init_zones(struct dmz_metadata *zmd)
1470 {
1471 	int i, ret;
1472 	struct dmz_dev *zoned_dev = &zmd->dev[0];
1473 
1474 	/* Init */
1475 	zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors;
1476 	zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors);
1477 	zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors);
1478 	zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks);
1479 	zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3;
1480 	zmd->zone_nr_bitmap_blocks =
1481 		max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1482 	zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks,
1483 					DMZ_BLOCK_SIZE_BITS);
1484 
1485 	/* Allocate zone array */
1486 	zmd->nr_zones = 0;
1487 	for (i = 0; i < zmd->nr_devs; i++) {
1488 		struct dmz_dev *dev = &zmd->dev[i];
1489 
1490 		dev->metadata = zmd;
1491 		zmd->nr_zones += dev->nr_zones;
1492 
1493 		atomic_set(&dev->unmap_nr_rnd, 0);
1494 		INIT_LIST_HEAD(&dev->unmap_rnd_list);
1495 		INIT_LIST_HEAD(&dev->map_rnd_list);
1496 
1497 		atomic_set(&dev->unmap_nr_seq, 0);
1498 		INIT_LIST_HEAD(&dev->unmap_seq_list);
1499 		INIT_LIST_HEAD(&dev->map_seq_list);
1500 	}
1501 
1502 	if (!zmd->nr_zones) {
1503 		DMERR("(%s): No zones found", zmd->devname);
1504 		return -ENXIO;
1505 	}
1506 	xa_init(&zmd->zones);
1507 
1508 	DMDEBUG("(%s): Using %zu B for zone information",
1509 		zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones);
1510 
1511 	if (zmd->nr_devs > 1) {
1512 		ret = dmz_emulate_zones(zmd, &zmd->dev[0]);
1513 		if (ret < 0) {
1514 			DMDEBUG("(%s): Failed to emulate zones, error %d",
1515 				zmd->devname, ret);
1516 			dmz_drop_zones(zmd);
1517 			return ret;
1518 		}
1519 
1520 		/*
1521 		 * Primary superblock zone is always at zone 0 when multiple
1522 		 * drives are present.
1523 		 */
1524 		zmd->sb[0].zone = dmz_get(zmd, 0);
1525 
1526 		for (i = 1; i < zmd->nr_devs; i++) {
1527 			zoned_dev = &zmd->dev[i];
1528 
1529 			ret = blkdev_report_zones(zoned_dev->bdev, 0,
1530 						  BLK_ALL_ZONES,
1531 						  dmz_init_zone, zoned_dev);
1532 			if (ret < 0) {
1533 				DMDEBUG("(%s): Failed to report zones, error %d",
1534 					zmd->devname, ret);
1535 				dmz_drop_zones(zmd);
1536 				return ret;
1537 			}
1538 		}
1539 		return 0;
1540 	}
1541 
1542 	/*
1543 	 * Get zone information and initialize zone descriptors.  At the same
1544 	 * time, determine where the super block should be: first block of the
1545 	 * first randomly writable zone.
1546 	 */
1547 	ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES,
1548 				  dmz_init_zone, zoned_dev);
1549 	if (ret < 0) {
1550 		DMDEBUG("(%s): Failed to report zones, error %d",
1551 			zmd->devname, ret);
1552 		dmz_drop_zones(zmd);
1553 		return ret;
1554 	}
1555 
1556 	return 0;
1557 }
1558 
1559 static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1560 			      void *data)
1561 {
1562 	struct dm_zone *zone = data;
1563 
1564 	clear_bit(DMZ_OFFLINE, &zone->flags);
1565 	clear_bit(DMZ_READ_ONLY, &zone->flags);
1566 	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1567 		set_bit(DMZ_OFFLINE, &zone->flags);
1568 	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1569 		set_bit(DMZ_READ_ONLY, &zone->flags);
1570 
1571 	if (dmz_is_seq(zone))
1572 		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1573 	else
1574 		zone->wp_block = 0;
1575 	return 0;
1576 }
1577 
1578 /*
1579  * Update a zone information.
1580  */
1581 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1582 {
1583 	struct dmz_dev *dev = zone->dev;
1584 	unsigned int noio_flag;
1585 	int ret;
1586 
1587 	if (dev->flags & DMZ_BDEV_REGULAR)
1588 		return 0;
1589 
1590 	/*
1591 	 * Get zone information from disk. Since blkdev_report_zones() uses
1592 	 * GFP_KERNEL by default for memory allocations, set the per-task
1593 	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1594 	 * GFP_NOIO was specified.
1595 	 */
1596 	noio_flag = memalloc_noio_save();
1597 	ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1,
1598 				  dmz_update_zone_cb, zone);
1599 	memalloc_noio_restore(noio_flag);
1600 
1601 	if (ret == 0)
1602 		ret = -EIO;
1603 	if (ret < 0) {
1604 		dmz_dev_err(dev, "Get zone %u report failed",
1605 			    zone->id);
1606 		dmz_check_bdev(dev);
1607 		return ret;
1608 	}
1609 
1610 	return 0;
1611 }
1612 
1613 /*
1614  * Check a zone write pointer position when the zone is marked
1615  * with the sequential write error flag.
1616  */
1617 static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1618 				    struct dm_zone *zone)
1619 {
1620 	struct dmz_dev *dev = zone->dev;
1621 	unsigned int wp = 0;
1622 	int ret;
1623 
1624 	wp = zone->wp_block;
1625 	ret = dmz_update_zone(zmd, zone);
1626 	if (ret)
1627 		return ret;
1628 
1629 	dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)",
1630 		     zone->id, zone->wp_block, wp);
1631 
1632 	if (zone->wp_block < wp) {
1633 		dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1634 				      wp - zone->wp_block);
1635 	}
1636 
1637 	return 0;
1638 }
1639 
1640 /*
1641  * Reset a zone write pointer.
1642  */
1643 static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1644 {
1645 	int ret;
1646 
1647 	/*
1648 	 * Ignore offline zones, read only zones,
1649 	 * and conventional zones.
1650 	 */
1651 	if (dmz_is_offline(zone) ||
1652 	    dmz_is_readonly(zone) ||
1653 	    dmz_is_rnd(zone))
1654 		return 0;
1655 
1656 	if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1657 		struct dmz_dev *dev = zone->dev;
1658 
1659 		ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1660 				       dmz_start_sect(zmd, zone),
1661 				       zmd->zone_nr_sectors, GFP_NOIO);
1662 		if (ret) {
1663 			dmz_dev_err(dev, "Reset zone %u failed %d",
1664 				    zone->id, ret);
1665 			return ret;
1666 		}
1667 	}
1668 
1669 	/* Clear write error bit and rewind write pointer position */
1670 	clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1671 	zone->wp_block = 0;
1672 
1673 	return 0;
1674 }
1675 
1676 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1677 
1678 /*
1679  * Initialize chunk mapping.
1680  */
1681 static int dmz_load_mapping(struct dmz_metadata *zmd)
1682 {
1683 	struct dm_zone *dzone, *bzone;
1684 	struct dmz_mblock *dmap_mblk = NULL;
1685 	struct dmz_map *dmap;
1686 	unsigned int i = 0, e = 0, chunk = 0;
1687 	unsigned int dzone_id;
1688 	unsigned int bzone_id;
1689 
1690 	/* Metadata block array for the chunk mapping table */
1691 	zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1692 				sizeof(struct dmz_mblk *), GFP_KERNEL);
1693 	if (!zmd->map_mblk)
1694 		return -ENOMEM;
1695 
1696 	/* Get chunk mapping table blocks and initialize zone mapping */
1697 	while (chunk < zmd->nr_chunks) {
1698 		if (!dmap_mblk) {
1699 			/* Get mapping block */
1700 			dmap_mblk = dmz_get_mblock(zmd, i + 1);
1701 			if (IS_ERR(dmap_mblk))
1702 				return PTR_ERR(dmap_mblk);
1703 			zmd->map_mblk[i] = dmap_mblk;
1704 			dmap = dmap_mblk->data;
1705 			i++;
1706 			e = 0;
1707 		}
1708 
1709 		/* Check data zone */
1710 		dzone_id = le32_to_cpu(dmap[e].dzone_id);
1711 		if (dzone_id == DMZ_MAP_UNMAPPED)
1712 			goto next;
1713 
1714 		if (dzone_id >= zmd->nr_zones) {
1715 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u",
1716 				    chunk, dzone_id);
1717 			return -EIO;
1718 		}
1719 
1720 		dzone = dmz_get(zmd, dzone_id);
1721 		if (!dzone) {
1722 			dmz_zmd_err(zmd, "Chunk %u mapping: data zone %u not present",
1723 				    chunk, dzone_id);
1724 			return -EIO;
1725 		}
1726 		set_bit(DMZ_DATA, &dzone->flags);
1727 		dzone->chunk = chunk;
1728 		dmz_get_zone_weight(zmd, dzone);
1729 
1730 		if (dmz_is_cache(dzone))
1731 			list_add_tail(&dzone->link, &zmd->map_cache_list);
1732 		else if (dmz_is_rnd(dzone))
1733 			list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
1734 		else
1735 			list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
1736 
1737 		/* Check buffer zone */
1738 		bzone_id = le32_to_cpu(dmap[e].bzone_id);
1739 		if (bzone_id == DMZ_MAP_UNMAPPED)
1740 			goto next;
1741 
1742 		if (bzone_id >= zmd->nr_zones) {
1743 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u",
1744 				    chunk, bzone_id);
1745 			return -EIO;
1746 		}
1747 
1748 		bzone = dmz_get(zmd, bzone_id);
1749 		if (!bzone) {
1750 			dmz_zmd_err(zmd, "Chunk %u mapping: buffer zone %u not present",
1751 				    chunk, bzone_id);
1752 			return -EIO;
1753 		}
1754 		if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) {
1755 			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u",
1756 				    chunk, bzone_id);
1757 			return -EIO;
1758 		}
1759 
1760 		set_bit(DMZ_DATA, &bzone->flags);
1761 		set_bit(DMZ_BUF, &bzone->flags);
1762 		bzone->chunk = chunk;
1763 		bzone->bzone = dzone;
1764 		dzone->bzone = bzone;
1765 		dmz_get_zone_weight(zmd, bzone);
1766 		if (dmz_is_cache(bzone))
1767 			list_add_tail(&bzone->link, &zmd->map_cache_list);
1768 		else
1769 			list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
1770 next:
1771 		chunk++;
1772 		e++;
1773 		if (e >= DMZ_MAP_ENTRIES)
1774 			dmap_mblk = NULL;
1775 	}
1776 
1777 	/*
1778 	 * At this point, only meta zones and mapped data zones were
1779 	 * fully initialized. All remaining zones are unmapped data
1780 	 * zones. Finish initializing those here.
1781 	 */
1782 	for (i = 0; i < zmd->nr_zones; i++) {
1783 		dzone = dmz_get(zmd, i);
1784 		if (!dzone)
1785 			continue;
1786 		if (dmz_is_meta(dzone))
1787 			continue;
1788 		if (dmz_is_offline(dzone))
1789 			continue;
1790 
1791 		if (dmz_is_cache(dzone))
1792 			zmd->nr_cache++;
1793 		else if (dmz_is_rnd(dzone))
1794 			dzone->dev->nr_rnd++;
1795 		else
1796 			dzone->dev->nr_seq++;
1797 
1798 		if (dmz_is_data(dzone)) {
1799 			/* Already initialized */
1800 			continue;
1801 		}
1802 
1803 		/* Unmapped data zone */
1804 		set_bit(DMZ_DATA, &dzone->flags);
1805 		dzone->chunk = DMZ_MAP_UNMAPPED;
1806 		if (dmz_is_cache(dzone)) {
1807 			list_add_tail(&dzone->link, &zmd->unmap_cache_list);
1808 			atomic_inc(&zmd->unmap_nr_cache);
1809 		} else if (dmz_is_rnd(dzone)) {
1810 			list_add_tail(&dzone->link,
1811 				      &dzone->dev->unmap_rnd_list);
1812 			atomic_inc(&dzone->dev->unmap_nr_rnd);
1813 		} else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1814 			list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1815 			set_bit(DMZ_RESERVED, &dzone->flags);
1816 			atomic_inc(&zmd->nr_reserved_seq_zones);
1817 			dzone->dev->nr_seq--;
1818 		} else {
1819 			list_add_tail(&dzone->link,
1820 				      &dzone->dev->unmap_seq_list);
1821 			atomic_inc(&dzone->dev->unmap_nr_seq);
1822 		}
1823 	}
1824 
1825 	return 0;
1826 }
1827 
1828 /*
1829  * Set a data chunk mapping.
1830  */
1831 static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1832 				  unsigned int dzone_id, unsigned int bzone_id)
1833 {
1834 	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1835 	struct dmz_map *dmap = dmap_mblk->data;
1836 	int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1837 
1838 	dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1839 	dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1840 	dmz_dirty_mblock(zmd, dmap_mblk);
1841 }
1842 
1843 /*
1844  * The list of mapped zones is maintained in LRU order.
1845  * This rotates a zone at the end of its map list.
1846  */
1847 static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1848 {
1849 	if (list_empty(&zone->link))
1850 		return;
1851 
1852 	list_del_init(&zone->link);
1853 	if (dmz_is_seq(zone)) {
1854 		/* LRU rotate sequential zone */
1855 		list_add_tail(&zone->link, &zone->dev->map_seq_list);
1856 	} else if (dmz_is_cache(zone)) {
1857 		/* LRU rotate cache zone */
1858 		list_add_tail(&zone->link, &zmd->map_cache_list);
1859 	} else {
1860 		/* LRU rotate random zone */
1861 		list_add_tail(&zone->link, &zone->dev->map_rnd_list);
1862 	}
1863 }
1864 
1865 /*
1866  * The list of mapped random zones is maintained
1867  * in LRU order. This rotates a zone at the end of the list.
1868  */
1869 static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1870 {
1871 	__dmz_lru_zone(zmd, zone);
1872 	if (zone->bzone)
1873 		__dmz_lru_zone(zmd, zone->bzone);
1874 }
1875 
1876 /*
1877  * Wait for any zone to be freed.
1878  */
1879 static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1880 {
1881 	DEFINE_WAIT(wait);
1882 
1883 	prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1884 	dmz_unlock_map(zmd);
1885 	dmz_unlock_metadata(zmd);
1886 
1887 	io_schedule_timeout(HZ);
1888 
1889 	dmz_lock_metadata(zmd);
1890 	dmz_lock_map(zmd);
1891 	finish_wait(&zmd->free_wq, &wait);
1892 }
1893 
1894 /*
1895  * Lock a zone for reclaim (set the zone RECLAIM bit).
1896  * Returns false if the zone cannot be locked or if it is already locked
1897  * and 1 otherwise.
1898  */
1899 int dmz_lock_zone_reclaim(struct dm_zone *zone)
1900 {
1901 	/* Active zones cannot be reclaimed */
1902 	if (dmz_is_active(zone))
1903 		return 0;
1904 
1905 	return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1906 }
1907 
1908 /*
1909  * Clear a zone reclaim flag.
1910  */
1911 void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1912 {
1913 	WARN_ON(dmz_is_active(zone));
1914 	WARN_ON(!dmz_in_reclaim(zone));
1915 
1916 	clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1917 	smp_mb__after_atomic();
1918 	wake_up_bit(&zone->flags, DMZ_RECLAIM);
1919 }
1920 
1921 /*
1922  * Wait for a zone reclaim to complete.
1923  */
1924 static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1925 {
1926 	dmz_unlock_map(zmd);
1927 	dmz_unlock_metadata(zmd);
1928 	set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1929 	wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1930 	clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1931 	dmz_lock_metadata(zmd);
1932 	dmz_lock_map(zmd);
1933 }
1934 
1935 /*
1936  * Select a cache or random write zone for reclaim.
1937  */
1938 static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd,
1939 						    unsigned int idx, bool idle)
1940 {
1941 	struct dm_zone *dzone = NULL;
1942 	struct dm_zone *zone, *maxw_z = NULL;
1943 	struct list_head *zone_list;
1944 
1945 	/* If we have cache zones select from the cache zone list */
1946 	if (zmd->nr_cache) {
1947 		zone_list = &zmd->map_cache_list;
1948 		/* Try to relaim random zones, too, when idle */
1949 		if (idle && list_empty(zone_list))
1950 			zone_list = &zmd->dev[idx].map_rnd_list;
1951 	} else
1952 		zone_list = &zmd->dev[idx].map_rnd_list;
1953 
1954 	/*
1955 	 * Find the buffer zone with the heaviest weight or the first (oldest)
1956 	 * data zone that can be reclaimed.
1957 	 */
1958 	list_for_each_entry(zone, zone_list, link) {
1959 		if (dmz_is_buf(zone)) {
1960 			dzone = zone->bzone;
1961 			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1962 				continue;
1963 			if (!maxw_z || maxw_z->weight < dzone->weight)
1964 				maxw_z = dzone;
1965 		} else {
1966 			dzone = zone;
1967 			if (dmz_lock_zone_reclaim(dzone))
1968 				return dzone;
1969 		}
1970 	}
1971 
1972 	if (maxw_z && dmz_lock_zone_reclaim(maxw_z))
1973 		return maxw_z;
1974 
1975 	/*
1976 	 * If we come here, none of the zones inspected could be locked for
1977 	 * reclaim. Try again, being more aggressive, that is, find the
1978 	 * first zone that can be reclaimed regardless of its weitght.
1979 	 */
1980 	list_for_each_entry(zone, zone_list, link) {
1981 		if (dmz_is_buf(zone)) {
1982 			dzone = zone->bzone;
1983 			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1984 				continue;
1985 		} else
1986 			dzone = zone;
1987 		if (dmz_lock_zone_reclaim(dzone))
1988 			return dzone;
1989 	}
1990 
1991 	return NULL;
1992 }
1993 
1994 /*
1995  * Select a buffered sequential zone for reclaim.
1996  */
1997 static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd,
1998 						    unsigned int idx)
1999 {
2000 	struct dm_zone *zone;
2001 
2002 	list_for_each_entry(zone, &zmd->dev[idx].map_seq_list, link) {
2003 		if (!zone->bzone)
2004 			continue;
2005 		if (dmz_lock_zone_reclaim(zone))
2006 			return zone;
2007 	}
2008 
2009 	return NULL;
2010 }
2011 
2012 /*
2013  * Select a zone for reclaim.
2014  */
2015 struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd,
2016 					 unsigned int dev_idx, bool idle)
2017 {
2018 	struct dm_zone *zone = NULL;
2019 
2020 	/*
2021 	 * Search for a zone candidate to reclaim: 2 cases are possible.
2022 	 * (1) There is no free sequential zones. Then a random data zone
2023 	 *     cannot be reclaimed. So choose a sequential zone to reclaim so
2024 	 *     that afterward a random zone can be reclaimed.
2025 	 * (2) At least one free sequential zone is available, then choose
2026 	 *     the oldest random zone (data or buffer) that can be locked.
2027 	 */
2028 	dmz_lock_map(zmd);
2029 	if (list_empty(&zmd->reserved_seq_zones_list))
2030 		zone = dmz_get_seq_zone_for_reclaim(zmd, dev_idx);
2031 	if (!zone)
2032 		zone = dmz_get_rnd_zone_for_reclaim(zmd, dev_idx, idle);
2033 	dmz_unlock_map(zmd);
2034 
2035 	return zone;
2036 }
2037 
2038 /*
2039  * Get the zone mapping a chunk, if the chunk is mapped already.
2040  * If no mapping exist and the operation is WRITE, a zone is
2041  * allocated and used to map the chunk.
2042  * The zone returned will be set to the active state.
2043  */
2044 struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd,
2045 				      unsigned int chunk, enum req_op op)
2046 {
2047 	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
2048 	struct dmz_map *dmap = dmap_mblk->data;
2049 	int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
2050 	unsigned int dzone_id;
2051 	struct dm_zone *dzone = NULL;
2052 	int ret = 0;
2053 	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2054 
2055 	dmz_lock_map(zmd);
2056 again:
2057 	/* Get the chunk mapping */
2058 	dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
2059 	if (dzone_id == DMZ_MAP_UNMAPPED) {
2060 		/*
2061 		 * Read or discard in unmapped chunks are fine. But for
2062 		 * writes, we need a mapping, so get one.
2063 		 */
2064 		if (op != REQ_OP_WRITE)
2065 			goto out;
2066 
2067 		/* Allocate a random zone */
2068 		dzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2069 		if (!dzone) {
2070 			if (dmz_dev_is_dying(zmd)) {
2071 				dzone = ERR_PTR(-EIO);
2072 				goto out;
2073 			}
2074 			dmz_wait_for_free_zones(zmd);
2075 			goto again;
2076 		}
2077 
2078 		dmz_map_zone(zmd, dzone, chunk);
2079 
2080 	} else {
2081 		/* The chunk is already mapped: get the mapping zone */
2082 		dzone = dmz_get(zmd, dzone_id);
2083 		if (!dzone) {
2084 			dzone = ERR_PTR(-EIO);
2085 			goto out;
2086 		}
2087 		if (dzone->chunk != chunk) {
2088 			dzone = ERR_PTR(-EIO);
2089 			goto out;
2090 		}
2091 
2092 		/* Repair write pointer if the sequential dzone has error */
2093 		if (dmz_seq_write_err(dzone)) {
2094 			ret = dmz_handle_seq_write_err(zmd, dzone);
2095 			if (ret) {
2096 				dzone = ERR_PTR(-EIO);
2097 				goto out;
2098 			}
2099 			clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
2100 		}
2101 	}
2102 
2103 	/*
2104 	 * If the zone is being reclaimed, the chunk mapping may change
2105 	 * to a different zone. So wait for reclaim and retry. Otherwise,
2106 	 * activate the zone (this will prevent reclaim from touching it).
2107 	 */
2108 	if (dmz_in_reclaim(dzone)) {
2109 		dmz_wait_for_reclaim(zmd, dzone);
2110 		goto again;
2111 	}
2112 	dmz_activate_zone(dzone);
2113 	dmz_lru_zone(zmd, dzone);
2114 out:
2115 	dmz_unlock_map(zmd);
2116 
2117 	return dzone;
2118 }
2119 
2120 /*
2121  * Write and discard change the block validity of data zones and their buffer
2122  * zones. Check here that valid blocks are still present. If all blocks are
2123  * invalid, the zones can be unmapped on the fly without waiting for reclaim
2124  * to do it.
2125  */
2126 void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
2127 {
2128 	struct dm_zone *bzone;
2129 
2130 	dmz_lock_map(zmd);
2131 
2132 	bzone = dzone->bzone;
2133 	if (bzone) {
2134 		if (dmz_weight(bzone))
2135 			dmz_lru_zone(zmd, bzone);
2136 		else {
2137 			/* Empty buffer zone: reclaim it */
2138 			dmz_unmap_zone(zmd, bzone);
2139 			dmz_free_zone(zmd, bzone);
2140 			bzone = NULL;
2141 		}
2142 	}
2143 
2144 	/* Deactivate the data zone */
2145 	dmz_deactivate_zone(dzone);
2146 	if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
2147 		dmz_lru_zone(zmd, dzone);
2148 	else {
2149 		/* Unbuffered inactive empty data zone: reclaim it */
2150 		dmz_unmap_zone(zmd, dzone);
2151 		dmz_free_zone(zmd, dzone);
2152 	}
2153 
2154 	dmz_unlock_map(zmd);
2155 }
2156 
2157 /*
2158  * Allocate and map a random zone to buffer a chunk
2159  * already mapped to a sequential zone.
2160  */
2161 struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
2162 				     struct dm_zone *dzone)
2163 {
2164 	struct dm_zone *bzone;
2165 	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2166 
2167 	dmz_lock_map(zmd);
2168 again:
2169 	bzone = dzone->bzone;
2170 	if (bzone)
2171 		goto out;
2172 
2173 	/* Allocate a random zone */
2174 	bzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2175 	if (!bzone) {
2176 		if (dmz_dev_is_dying(zmd)) {
2177 			bzone = ERR_PTR(-EIO);
2178 			goto out;
2179 		}
2180 		dmz_wait_for_free_zones(zmd);
2181 		goto again;
2182 	}
2183 
2184 	/* Update the chunk mapping */
2185 	dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id);
2186 
2187 	set_bit(DMZ_BUF, &bzone->flags);
2188 	bzone->chunk = dzone->chunk;
2189 	bzone->bzone = dzone;
2190 	dzone->bzone = bzone;
2191 	if (dmz_is_cache(bzone))
2192 		list_add_tail(&bzone->link, &zmd->map_cache_list);
2193 	else
2194 		list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
2195 out:
2196 	dmz_unlock_map(zmd);
2197 
2198 	return bzone;
2199 }
2200 
2201 /*
2202  * Get an unmapped (free) zone.
2203  * This must be called with the mapping lock held.
2204  */
2205 struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned int dev_idx,
2206 			       unsigned long flags)
2207 {
2208 	struct list_head *list;
2209 	struct dm_zone *zone;
2210 	int i;
2211 
2212 	/* Schedule reclaim to ensure free zones are available */
2213 	if (!(flags & DMZ_ALLOC_RECLAIM)) {
2214 		for (i = 0; i < zmd->nr_devs; i++)
2215 			dmz_schedule_reclaim(zmd->dev[i].reclaim);
2216 	}
2217 
2218 	i = 0;
2219 again:
2220 	if (flags & DMZ_ALLOC_CACHE)
2221 		list = &zmd->unmap_cache_list;
2222 	else if (flags & DMZ_ALLOC_RND)
2223 		list = &zmd->dev[dev_idx].unmap_rnd_list;
2224 	else
2225 		list = &zmd->dev[dev_idx].unmap_seq_list;
2226 
2227 	if (list_empty(list)) {
2228 		/*
2229 		 * No free zone: return NULL if this is for not reclaim.
2230 		 */
2231 		if (!(flags & DMZ_ALLOC_RECLAIM))
2232 			return NULL;
2233 		/*
2234 		 * Try to allocate from other devices
2235 		 */
2236 		if (i < zmd->nr_devs) {
2237 			dev_idx = (dev_idx + 1) % zmd->nr_devs;
2238 			i++;
2239 			goto again;
2240 		}
2241 
2242 		/*
2243 		 * Fallback to the reserved sequential zones
2244 		 */
2245 		zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list,
2246 						struct dm_zone, link);
2247 		if (zone) {
2248 			list_del_init(&zone->link);
2249 			atomic_dec(&zmd->nr_reserved_seq_zones);
2250 		}
2251 		return zone;
2252 	}
2253 
2254 	zone = list_first_entry(list, struct dm_zone, link);
2255 	list_del_init(&zone->link);
2256 
2257 	if (dmz_is_cache(zone))
2258 		atomic_dec(&zmd->unmap_nr_cache);
2259 	else if (dmz_is_rnd(zone))
2260 		atomic_dec(&zone->dev->unmap_nr_rnd);
2261 	else
2262 		atomic_dec(&zone->dev->unmap_nr_seq);
2263 
2264 	if (dmz_is_offline(zone)) {
2265 		dmz_zmd_warn(zmd, "Zone %u is offline", zone->id);
2266 		zone = NULL;
2267 		goto again;
2268 	}
2269 	if (dmz_is_meta(zone)) {
2270 		dmz_zmd_warn(zmd, "Zone %u has metadata", zone->id);
2271 		zone = NULL;
2272 		goto again;
2273 	}
2274 	return zone;
2275 }
2276 
2277 /*
2278  * Free a zone.
2279  * This must be called with the mapping lock held.
2280  */
2281 void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2282 {
2283 	/* If this is a sequential zone, reset it */
2284 	if (dmz_is_seq(zone))
2285 		dmz_reset_zone(zmd, zone);
2286 
2287 	/* Return the zone to its type unmap list */
2288 	if (dmz_is_cache(zone)) {
2289 		list_add_tail(&zone->link, &zmd->unmap_cache_list);
2290 		atomic_inc(&zmd->unmap_nr_cache);
2291 	} else if (dmz_is_rnd(zone)) {
2292 		list_add_tail(&zone->link, &zone->dev->unmap_rnd_list);
2293 		atomic_inc(&zone->dev->unmap_nr_rnd);
2294 	} else if (dmz_is_reserved(zone)) {
2295 		list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
2296 		atomic_inc(&zmd->nr_reserved_seq_zones);
2297 	} else {
2298 		list_add_tail(&zone->link, &zone->dev->unmap_seq_list);
2299 		atomic_inc(&zone->dev->unmap_nr_seq);
2300 	}
2301 
2302 	wake_up_all(&zmd->free_wq);
2303 }
2304 
2305 /*
2306  * Map a chunk to a zone.
2307  * This must be called with the mapping lock held.
2308  */
2309 void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
2310 		  unsigned int chunk)
2311 {
2312 	/* Set the chunk mapping */
2313 	dmz_set_chunk_mapping(zmd, chunk, dzone->id,
2314 			      DMZ_MAP_UNMAPPED);
2315 	dzone->chunk = chunk;
2316 	if (dmz_is_cache(dzone))
2317 		list_add_tail(&dzone->link, &zmd->map_cache_list);
2318 	else if (dmz_is_rnd(dzone))
2319 		list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
2320 	else
2321 		list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
2322 }
2323 
2324 /*
2325  * Unmap a zone.
2326  * This must be called with the mapping lock held.
2327  */
2328 void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2329 {
2330 	unsigned int chunk = zone->chunk;
2331 	unsigned int dzone_id;
2332 
2333 	if (chunk == DMZ_MAP_UNMAPPED) {
2334 		/* Already unmapped */
2335 		return;
2336 	}
2337 
2338 	if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
2339 		/*
2340 		 * Unmapping the chunk buffer zone: clear only
2341 		 * the chunk buffer mapping
2342 		 */
2343 		dzone_id = zone->bzone->id;
2344 		zone->bzone->bzone = NULL;
2345 		zone->bzone = NULL;
2346 
2347 	} else {
2348 		/*
2349 		 * Unmapping the chunk data zone: the zone must
2350 		 * not be buffered.
2351 		 */
2352 		if (WARN_ON(zone->bzone)) {
2353 			zone->bzone->bzone = NULL;
2354 			zone->bzone = NULL;
2355 		}
2356 		dzone_id = DMZ_MAP_UNMAPPED;
2357 	}
2358 
2359 	dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
2360 
2361 	zone->chunk = DMZ_MAP_UNMAPPED;
2362 	list_del_init(&zone->link);
2363 }
2364 
2365 /*
2366  * Set @nr_bits bits in @bitmap starting from @bit.
2367  * Return the number of bits changed from 0 to 1.
2368  */
2369 static unsigned int dmz_set_bits(unsigned long *bitmap,
2370 				 unsigned int bit, unsigned int nr_bits)
2371 {
2372 	unsigned long *addr;
2373 	unsigned int end = bit + nr_bits;
2374 	unsigned int n = 0;
2375 
2376 	while (bit < end) {
2377 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2378 		    ((end - bit) >= BITS_PER_LONG)) {
2379 			/* Try to set the whole word at once */
2380 			addr = bitmap + BIT_WORD(bit);
2381 			if (*addr == 0) {
2382 				*addr = ULONG_MAX;
2383 				n += BITS_PER_LONG;
2384 				bit += BITS_PER_LONG;
2385 				continue;
2386 			}
2387 		}
2388 
2389 		if (!test_and_set_bit(bit, bitmap))
2390 			n++;
2391 		bit++;
2392 	}
2393 
2394 	return n;
2395 }
2396 
2397 /*
2398  * Get the bitmap block storing the bit for chunk_block in zone.
2399  */
2400 static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
2401 					 struct dm_zone *zone,
2402 					 sector_t chunk_block)
2403 {
2404 	sector_t bitmap_block = 1 + zmd->nr_map_blocks +
2405 		(sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) +
2406 		(chunk_block >> DMZ_BLOCK_SHIFT_BITS);
2407 
2408 	return dmz_get_mblock(zmd, bitmap_block);
2409 }
2410 
2411 /*
2412  * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
2413  */
2414 int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2415 			  struct dm_zone *to_zone)
2416 {
2417 	struct dmz_mblock *from_mblk, *to_mblk;
2418 	sector_t chunk_block = 0;
2419 
2420 	/* Get the zones bitmap blocks */
2421 	while (chunk_block < zmd->zone_nr_blocks) {
2422 		from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
2423 		if (IS_ERR(from_mblk))
2424 			return PTR_ERR(from_mblk);
2425 		to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
2426 		if (IS_ERR(to_mblk)) {
2427 			dmz_release_mblock(zmd, from_mblk);
2428 			return PTR_ERR(to_mblk);
2429 		}
2430 
2431 		memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
2432 		dmz_dirty_mblock(zmd, to_mblk);
2433 
2434 		dmz_release_mblock(zmd, to_mblk);
2435 		dmz_release_mblock(zmd, from_mblk);
2436 
2437 		chunk_block += zmd->zone_bits_per_mblk;
2438 	}
2439 
2440 	to_zone->weight = from_zone->weight;
2441 
2442 	return 0;
2443 }
2444 
2445 /*
2446  * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
2447  * starting from chunk_block.
2448  */
2449 int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2450 			   struct dm_zone *to_zone, sector_t chunk_block)
2451 {
2452 	unsigned int nr_blocks;
2453 	int ret;
2454 
2455 	/* Get the zones bitmap blocks */
2456 	while (chunk_block < zmd->zone_nr_blocks) {
2457 		/* Get a valid region from the source zone */
2458 		ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2459 		if (ret <= 0)
2460 			return ret;
2461 
2462 		nr_blocks = ret;
2463 		ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2464 		if (ret)
2465 			return ret;
2466 
2467 		chunk_block += nr_blocks;
2468 	}
2469 
2470 	return 0;
2471 }
2472 
2473 /*
2474  * Validate all the blocks in the range [block..block+nr_blocks-1].
2475  */
2476 int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2477 			sector_t chunk_block, unsigned int nr_blocks)
2478 {
2479 	unsigned int count, bit, nr_bits;
2480 	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
2481 	struct dmz_mblock *mblk;
2482 	unsigned int n = 0;
2483 
2484 	dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks",
2485 		      zone->id, (unsigned long long)chunk_block,
2486 		      nr_blocks);
2487 
2488 	WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2489 
2490 	while (nr_blocks) {
2491 		/* Get bitmap block */
2492 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2493 		if (IS_ERR(mblk))
2494 			return PTR_ERR(mblk);
2495 
2496 		/* Set bits */
2497 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2498 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2499 
2500 		count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2501 		if (count) {
2502 			dmz_dirty_mblock(zmd, mblk);
2503 			n += count;
2504 		}
2505 		dmz_release_mblock(zmd, mblk);
2506 
2507 		nr_blocks -= nr_bits;
2508 		chunk_block += nr_bits;
2509 	}
2510 
2511 	if (likely(zone->weight + n <= zone_nr_blocks))
2512 		zone->weight += n;
2513 	else {
2514 		dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u",
2515 			     zone->id, zone->weight,
2516 			     zone_nr_blocks - n);
2517 		zone->weight = zone_nr_blocks;
2518 	}
2519 
2520 	return 0;
2521 }
2522 
2523 /*
2524  * Clear nr_bits bits in bitmap starting from bit.
2525  * Return the number of bits cleared.
2526  */
2527 static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2528 {
2529 	unsigned long *addr;
2530 	int end = bit + nr_bits;
2531 	int n = 0;
2532 
2533 	while (bit < end) {
2534 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2535 		    ((end - bit) >= BITS_PER_LONG)) {
2536 			/* Try to clear whole word at once */
2537 			addr = bitmap + BIT_WORD(bit);
2538 			if (*addr == ULONG_MAX) {
2539 				*addr = 0;
2540 				n += BITS_PER_LONG;
2541 				bit += BITS_PER_LONG;
2542 				continue;
2543 			}
2544 		}
2545 
2546 		if (test_and_clear_bit(bit, bitmap))
2547 			n++;
2548 		bit++;
2549 	}
2550 
2551 	return n;
2552 }
2553 
2554 /*
2555  * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2556  */
2557 int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2558 			  sector_t chunk_block, unsigned int nr_blocks)
2559 {
2560 	unsigned int count, bit, nr_bits;
2561 	struct dmz_mblock *mblk;
2562 	unsigned int n = 0;
2563 
2564 	dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks",
2565 		      zone->id, (u64)chunk_block, nr_blocks);
2566 
2567 	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2568 
2569 	while (nr_blocks) {
2570 		/* Get bitmap block */
2571 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2572 		if (IS_ERR(mblk))
2573 			return PTR_ERR(mblk);
2574 
2575 		/* Clear bits */
2576 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2577 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2578 
2579 		count = dmz_clear_bits((unsigned long *)mblk->data,
2580 				       bit, nr_bits);
2581 		if (count) {
2582 			dmz_dirty_mblock(zmd, mblk);
2583 			n += count;
2584 		}
2585 		dmz_release_mblock(zmd, mblk);
2586 
2587 		nr_blocks -= nr_bits;
2588 		chunk_block += nr_bits;
2589 	}
2590 
2591 	if (zone->weight >= n)
2592 		zone->weight -= n;
2593 	else {
2594 		dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u",
2595 			     zone->id, zone->weight, n);
2596 		zone->weight = 0;
2597 	}
2598 
2599 	return 0;
2600 }
2601 
2602 /*
2603  * Get a block bit value.
2604  */
2605 static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2606 			  sector_t chunk_block)
2607 {
2608 	struct dmz_mblock *mblk;
2609 	int ret;
2610 
2611 	WARN_ON(chunk_block >= zmd->zone_nr_blocks);
2612 
2613 	/* Get bitmap block */
2614 	mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2615 	if (IS_ERR(mblk))
2616 		return PTR_ERR(mblk);
2617 
2618 	/* Get offset */
2619 	ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2620 		       (unsigned long *) mblk->data) != 0;
2621 
2622 	dmz_release_mblock(zmd, mblk);
2623 
2624 	return ret;
2625 }
2626 
2627 /*
2628  * Return the number of blocks from chunk_block to the first block with a bit
2629  * value specified by set. Search at most nr_blocks blocks from chunk_block.
2630  */
2631 static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2632 				 sector_t chunk_block, unsigned int nr_blocks,
2633 				 int set)
2634 {
2635 	struct dmz_mblock *mblk;
2636 	unsigned int bit, set_bit, nr_bits;
2637 	unsigned int zone_bits = zmd->zone_bits_per_mblk;
2638 	unsigned long *bitmap;
2639 	int n = 0;
2640 
2641 	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2642 
2643 	while (nr_blocks) {
2644 		/* Get bitmap block */
2645 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2646 		if (IS_ERR(mblk))
2647 			return PTR_ERR(mblk);
2648 
2649 		/* Get offset */
2650 		bitmap = (unsigned long *) mblk->data;
2651 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2652 		nr_bits = min(nr_blocks, zone_bits - bit);
2653 		if (set)
2654 			set_bit = find_next_bit(bitmap, zone_bits, bit);
2655 		else
2656 			set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2657 		dmz_release_mblock(zmd, mblk);
2658 
2659 		n += set_bit - bit;
2660 		if (set_bit < zone_bits)
2661 			break;
2662 
2663 		nr_blocks -= nr_bits;
2664 		chunk_block += nr_bits;
2665 	}
2666 
2667 	return n;
2668 }
2669 
2670 /*
2671  * Test if chunk_block is valid. If it is, the number of consecutive
2672  * valid blocks from chunk_block will be returned.
2673  */
2674 int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2675 		    sector_t chunk_block)
2676 {
2677 	int valid;
2678 
2679 	valid = dmz_test_block(zmd, zone, chunk_block);
2680 	if (valid <= 0)
2681 		return valid;
2682 
2683 	/* The block is valid: get the number of valid blocks from block */
2684 	return dmz_to_next_set_block(zmd, zone, chunk_block,
2685 				     zmd->zone_nr_blocks - chunk_block, 0);
2686 }
2687 
2688 /*
2689  * Find the first valid block from @chunk_block in @zone.
2690  * If such a block is found, its number is returned using
2691  * @chunk_block and the total number of valid blocks from @chunk_block
2692  * is returned.
2693  */
2694 int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2695 			  sector_t *chunk_block)
2696 {
2697 	sector_t start_block = *chunk_block;
2698 	int ret;
2699 
2700 	ret = dmz_to_next_set_block(zmd, zone, start_block,
2701 				    zmd->zone_nr_blocks - start_block, 1);
2702 	if (ret < 0)
2703 		return ret;
2704 
2705 	start_block += ret;
2706 	*chunk_block = start_block;
2707 
2708 	return dmz_to_next_set_block(zmd, zone, start_block,
2709 				     zmd->zone_nr_blocks - start_block, 0);
2710 }
2711 
2712 /*
2713  * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2714  */
2715 static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2716 {
2717 	unsigned long *addr;
2718 	int end = bit + nr_bits;
2719 	int n = 0;
2720 
2721 	while (bit < end) {
2722 		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2723 		    ((end - bit) >= BITS_PER_LONG)) {
2724 			addr = (unsigned long *)bitmap + BIT_WORD(bit);
2725 			if (*addr == ULONG_MAX) {
2726 				n += BITS_PER_LONG;
2727 				bit += BITS_PER_LONG;
2728 				continue;
2729 			}
2730 		}
2731 
2732 		if (test_bit(bit, bitmap))
2733 			n++;
2734 		bit++;
2735 	}
2736 
2737 	return n;
2738 }
2739 
2740 /*
2741  * Get a zone weight.
2742  */
2743 static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2744 {
2745 	struct dmz_mblock *mblk;
2746 	sector_t chunk_block = 0;
2747 	unsigned int bit, nr_bits;
2748 	unsigned int nr_blocks = zmd->zone_nr_blocks;
2749 	void *bitmap;
2750 	int n = 0;
2751 
2752 	while (nr_blocks) {
2753 		/* Get bitmap block */
2754 		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2755 		if (IS_ERR(mblk)) {
2756 			n = 0;
2757 			break;
2758 		}
2759 
2760 		/* Count bits in this block */
2761 		bitmap = mblk->data;
2762 		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2763 		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2764 		n += dmz_count_bits(bitmap, bit, nr_bits);
2765 
2766 		dmz_release_mblock(zmd, mblk);
2767 
2768 		nr_blocks -= nr_bits;
2769 		chunk_block += nr_bits;
2770 	}
2771 
2772 	zone->weight = n;
2773 }
2774 
2775 /*
2776  * Cleanup the zoned metadata resources.
2777  */
2778 static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2779 {
2780 	struct rb_root *root;
2781 	struct dmz_mblock *mblk, *next;
2782 	int i;
2783 
2784 	/* Release zone mapping resources */
2785 	if (zmd->map_mblk) {
2786 		for (i = 0; i < zmd->nr_map_blocks; i++)
2787 			dmz_release_mblock(zmd, zmd->map_mblk[i]);
2788 		kfree(zmd->map_mblk);
2789 		zmd->map_mblk = NULL;
2790 	}
2791 
2792 	/* Release super blocks */
2793 	for (i = 0; i < 2; i++) {
2794 		if (zmd->sb[i].mblk) {
2795 			dmz_free_mblock(zmd, zmd->sb[i].mblk);
2796 			zmd->sb[i].mblk = NULL;
2797 		}
2798 	}
2799 
2800 	/* Free cached blocks */
2801 	while (!list_empty(&zmd->mblk_dirty_list)) {
2802 		mblk = list_first_entry(&zmd->mblk_dirty_list,
2803 					struct dmz_mblock, link);
2804 		dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)",
2805 			     (u64)mblk->no, mblk->ref);
2806 		list_del_init(&mblk->link);
2807 		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2808 		dmz_free_mblock(zmd, mblk);
2809 	}
2810 
2811 	while (!list_empty(&zmd->mblk_lru_list)) {
2812 		mblk = list_first_entry(&zmd->mblk_lru_list,
2813 					struct dmz_mblock, link);
2814 		list_del_init(&mblk->link);
2815 		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2816 		dmz_free_mblock(zmd, mblk);
2817 	}
2818 
2819 	/* Sanity checks: the mblock rbtree should now be empty */
2820 	root = &zmd->mblk_rbtree;
2821 	rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2822 		dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree",
2823 			     (u64)mblk->no, mblk->ref);
2824 		mblk->ref = 0;
2825 		dmz_free_mblock(zmd, mblk);
2826 	}
2827 
2828 	/* Free the zone descriptors */
2829 	dmz_drop_zones(zmd);
2830 
2831 	mutex_destroy(&zmd->mblk_flush_lock);
2832 	mutex_destroy(&zmd->map_lock);
2833 }
2834 
2835 static void dmz_print_dev(struct dmz_metadata *zmd, int num)
2836 {
2837 	struct dmz_dev *dev = &zmd->dev[num];
2838 
2839 	if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE)
2840 		dmz_dev_info(dev, "Regular block device");
2841 	else
2842 		dmz_dev_info(dev, "Host-%s zoned block device",
2843 			     bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2844 			     "aware" : "managed");
2845 	if (zmd->sb_version > 1) {
2846 		sector_t sector_offset =
2847 			dev->zone_offset << zmd->zone_nr_sectors_shift;
2848 
2849 		dmz_dev_info(dev, "  %llu 512-byte logical sectors (offset %llu)",
2850 			     (u64)dev->capacity, (u64)sector_offset);
2851 		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors (offset %llu)",
2852 			     dev->nr_zones, (u64)zmd->zone_nr_sectors,
2853 			     (u64)dev->zone_offset);
2854 	} else {
2855 		dmz_dev_info(dev, "  %llu 512-byte logical sectors",
2856 			     (u64)dev->capacity);
2857 		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors",
2858 			     dev->nr_zones, (u64)zmd->zone_nr_sectors);
2859 	}
2860 }
2861 
2862 /*
2863  * Initialize the zoned metadata.
2864  */
2865 int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev,
2866 		     struct dmz_metadata **metadata,
2867 		     const char *devname)
2868 {
2869 	struct dmz_metadata *zmd;
2870 	unsigned int i;
2871 	struct dm_zone *zone;
2872 	int ret;
2873 
2874 	zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2875 	if (!zmd)
2876 		return -ENOMEM;
2877 
2878 	strcpy(zmd->devname, devname);
2879 	zmd->dev = dev;
2880 	zmd->nr_devs = num_dev;
2881 	zmd->mblk_rbtree = RB_ROOT;
2882 	init_rwsem(&zmd->mblk_sem);
2883 	mutex_init(&zmd->mblk_flush_lock);
2884 	spin_lock_init(&zmd->mblk_lock);
2885 	INIT_LIST_HEAD(&zmd->mblk_lru_list);
2886 	INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2887 
2888 	mutex_init(&zmd->map_lock);
2889 
2890 	atomic_set(&zmd->unmap_nr_cache, 0);
2891 	INIT_LIST_HEAD(&zmd->unmap_cache_list);
2892 	INIT_LIST_HEAD(&zmd->map_cache_list);
2893 
2894 	atomic_set(&zmd->nr_reserved_seq_zones, 0);
2895 	INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2896 
2897 	init_waitqueue_head(&zmd->free_wq);
2898 
2899 	/* Initialize zone descriptors */
2900 	ret = dmz_init_zones(zmd);
2901 	if (ret)
2902 		goto err;
2903 
2904 	/* Get super block */
2905 	ret = dmz_load_sb(zmd);
2906 	if (ret)
2907 		goto err;
2908 
2909 	/* Set metadata zones starting from sb_zone */
2910 	for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2911 		zone = dmz_get(zmd, zmd->sb[0].zone->id + i);
2912 		if (!zone) {
2913 			dmz_zmd_err(zmd,
2914 				    "metadata zone %u not present", i);
2915 			ret = -ENXIO;
2916 			goto err;
2917 		}
2918 		if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) {
2919 			dmz_zmd_err(zmd,
2920 				    "metadata zone %d is not random", i);
2921 			ret = -ENXIO;
2922 			goto err;
2923 		}
2924 		set_bit(DMZ_META, &zone->flags);
2925 	}
2926 	/* Load mapping table */
2927 	ret = dmz_load_mapping(zmd);
2928 	if (ret)
2929 		goto err;
2930 
2931 	/*
2932 	 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2933 	 * blocks and enough blocks to be able to cache the bitmap blocks of
2934 	 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2935 	 * the cache to add 512 more metadata blocks.
2936 	 */
2937 	zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2938 	zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2939 	zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2940 	zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2941 	zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2942 
2943 	/* Metadata cache shrinker */
2944 	ret = register_shrinker(&zmd->mblk_shrinker, "dm-zoned-meta:(%u:%u)",
2945 				MAJOR(dev->bdev->bd_dev),
2946 				MINOR(dev->bdev->bd_dev));
2947 	if (ret) {
2948 		dmz_zmd_err(zmd, "Register metadata cache shrinker failed");
2949 		goto err;
2950 	}
2951 
2952 	dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version);
2953 	for (i = 0; i < zmd->nr_devs; i++)
2954 		dmz_print_dev(zmd, i);
2955 
2956 	dmz_zmd_info(zmd, "  %u zones of %llu 512-byte logical sectors",
2957 		     zmd->nr_zones, (u64)zmd->zone_nr_sectors);
2958 	dmz_zmd_debug(zmd, "  %u metadata zones",
2959 		      zmd->nr_meta_zones * 2);
2960 	dmz_zmd_debug(zmd, "  %u data zones for %u chunks",
2961 		      zmd->nr_data_zones, zmd->nr_chunks);
2962 	dmz_zmd_debug(zmd, "    %u cache zones (%u unmapped)",
2963 		      zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache));
2964 	for (i = 0; i < zmd->nr_devs; i++) {
2965 		dmz_zmd_debug(zmd, "    %u random zones (%u unmapped)",
2966 			      dmz_nr_rnd_zones(zmd, i),
2967 			      dmz_nr_unmap_rnd_zones(zmd, i));
2968 		dmz_zmd_debug(zmd, "    %u sequential zones (%u unmapped)",
2969 			      dmz_nr_seq_zones(zmd, i),
2970 			      dmz_nr_unmap_seq_zones(zmd, i));
2971 	}
2972 	dmz_zmd_debug(zmd, "  %u reserved sequential data zones",
2973 		      zmd->nr_reserved_seq);
2974 	dmz_zmd_debug(zmd, "Format:");
2975 	dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)",
2976 		      zmd->nr_meta_blocks, zmd->max_nr_mblks);
2977 	dmz_zmd_debug(zmd, "  %u data zone mapping blocks",
2978 		      zmd->nr_map_blocks);
2979 	dmz_zmd_debug(zmd, "  %u bitmap blocks",
2980 		      zmd->nr_bitmap_blocks);
2981 
2982 	*metadata = zmd;
2983 
2984 	return 0;
2985 err:
2986 	dmz_cleanup_metadata(zmd);
2987 	kfree(zmd);
2988 	*metadata = NULL;
2989 
2990 	return ret;
2991 }
2992 
2993 /*
2994  * Cleanup the zoned metadata resources.
2995  */
2996 void dmz_dtr_metadata(struct dmz_metadata *zmd)
2997 {
2998 	unregister_shrinker(&zmd->mblk_shrinker);
2999 	dmz_cleanup_metadata(zmd);
3000 	kfree(zmd);
3001 }
3002 
3003 /*
3004  * Check zone information on resume.
3005  */
3006 int dmz_resume_metadata(struct dmz_metadata *zmd)
3007 {
3008 	struct dm_zone *zone;
3009 	sector_t wp_block;
3010 	unsigned int i;
3011 	int ret;
3012 
3013 	/* Check zones */
3014 	for (i = 0; i < zmd->nr_zones; i++) {
3015 		zone = dmz_get(zmd, i);
3016 		if (!zone) {
3017 			dmz_zmd_err(zmd, "Unable to get zone %u", i);
3018 			return -EIO;
3019 		}
3020 		wp_block = zone->wp_block;
3021 
3022 		ret = dmz_update_zone(zmd, zone);
3023 		if (ret) {
3024 			dmz_zmd_err(zmd, "Broken zone %u", i);
3025 			return ret;
3026 		}
3027 
3028 		if (dmz_is_offline(zone)) {
3029 			dmz_zmd_warn(zmd, "Zone %u is offline", i);
3030 			continue;
3031 		}
3032 
3033 		/* Check write pointer */
3034 		if (!dmz_is_seq(zone))
3035 			zone->wp_block = 0;
3036 		else if (zone->wp_block != wp_block) {
3037 			dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)",
3038 				    i, (u64)zone->wp_block, (u64)wp_block);
3039 			zone->wp_block = wp_block;
3040 			dmz_invalidate_blocks(zmd, zone, zone->wp_block,
3041 					      zmd->zone_nr_blocks - zone->wp_block);
3042 		}
3043 	}
3044 
3045 	return 0;
3046 }
3047