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 */
  41struct 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 */
  93struct 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 */
 109struct 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 */
 122enum {
 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 */
 132struct 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 */
 143struct 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 */
 221static 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
 229sector_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
 236sector_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
 243unsigned int dmz_zone_nr_blocks(struct dmz_metadata *zmd)
 244{
 245	return zmd->zone_nr_blocks;
 246}
 247
 248unsigned int dmz_zone_nr_blocks_shift(struct dmz_metadata *zmd)
 249{
 250	return zmd->zone_nr_blocks_shift;
 251}
 252
 253unsigned int dmz_zone_nr_sectors(struct dmz_metadata *zmd)
 254{
 255	return zmd->zone_nr_sectors;
 256}
 257
 258unsigned int dmz_zone_nr_sectors_shift(struct dmz_metadata *zmd)
 259{
 260	return zmd->zone_nr_sectors_shift;
 261}
 262
 263unsigned int dmz_nr_zones(struct dmz_metadata *zmd)
 264{
 265	return zmd->nr_zones;
 266}
 267
 268unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
 269{
 270	return zmd->nr_chunks;
 271}
 272
 273unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd, int idx)
 274{
 275	return zmd->dev[idx].nr_rnd;
 276}
 277
 278unsigned 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
 283unsigned int dmz_nr_cache_zones(struct dmz_metadata *zmd)
 284{
 285	return zmd->nr_cache;
 286}
 287
 288unsigned int dmz_nr_unmap_cache_zones(struct dmz_metadata *zmd)
 289{
 290	return atomic_read(&zmd->unmap_nr_cache);
 291}
 292
 293unsigned int dmz_nr_seq_zones(struct dmz_metadata *zmd, int idx)
 294{
 295	return zmd->dev[idx].nr_seq;
 296}
 297
 298unsigned 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
 303static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
 304{
 305	return xa_load(&zmd->zones, zone_id);
 306}
 307
 308static 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
 330const char *dmz_metadata_label(struct dmz_metadata *zmd)
 331{
 332	return (const char *)zmd->label;
 333}
 334
 335bool 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
 346bool 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 */
 361void dmz_lock_map(struct dmz_metadata *zmd)
 362{
 363	mutex_lock(&zmd->map_lock);
 364}
 365
 366void 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 */
 378void dmz_lock_metadata(struct dmz_metadata *zmd)
 379{
 380	down_read(&zmd->mblk_sem);
 381}
 382
 383void 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 */
 393void dmz_lock_flush(struct dmz_metadata *zmd)
 394{
 395	mutex_lock(&zmd->mblk_flush_lock);
 396}
 397
 398void dmz_unlock_flush(struct dmz_metadata *zmd)
 399{
 400	mutex_unlock(&zmd->mblk_flush_lock);
 401}
 402
 403/*
 404 * Allocate a metadata block.
 405 */
 406static 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 */
 452static 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 */
 463static 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 */
 485static 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 */
 514static 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 */
 537static 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(GFP_NOIO, 1);
 554	if (!bio) {
 555		dmz_free_mblock(zmd, mblk);
 556		return ERR_PTR(-ENOMEM);
 557	}
 558
 559	spin_lock(&zmd->mblk_lock);
 560
 561	/*
 562	 * Make sure that another context did not start reading
 563	 * the block already.
 564	 */
 565	m = dmz_get_mblock_fast(zmd, mblk_no);
 566	if (m) {
 567		spin_unlock(&zmd->mblk_lock);
 568		dmz_free_mblock(zmd, mblk);
 569		bio_put(bio);
 570		return m;
 571	}
 572
 573	mblk->ref++;
 574	set_bit(DMZ_META_READING, &mblk->state);
 575	dmz_insert_mblock(zmd, mblk);
 576
 577	spin_unlock(&zmd->mblk_lock);
 578
 579	/* Submit read BIO */
 580	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 581	bio_set_dev(bio, dev->bdev);
 582	bio->bi_private = mblk;
 583	bio->bi_end_io = dmz_mblock_bio_end_io;
 584	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
 585	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
 586	submit_bio(bio);
 587
 588	return mblk;
 589}
 590
 591/*
 592 * Free metadata blocks.
 593 */
 594static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
 595					     unsigned long limit)
 596{
 597	struct dmz_mblock *mblk;
 598	unsigned long count = 0;
 599
 600	if (!zmd->max_nr_mblks)
 601		return 0;
 602
 603	while (!list_empty(&zmd->mblk_lru_list) &&
 604	       atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
 605	       count < limit) {
 606		mblk = list_first_entry(&zmd->mblk_lru_list,
 607					struct dmz_mblock, link);
 608		list_del_init(&mblk->link);
 609		rb_erase(&mblk->node, &zmd->mblk_rbtree);
 610		dmz_free_mblock(zmd, mblk);
 611		count++;
 612	}
 613
 614	return count;
 615}
 616
 617/*
 618 * For mblock shrinker: get the number of unused metadata blocks in the cache.
 619 */
 620static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
 621					       struct shrink_control *sc)
 622{
 623	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
 624
 625	return atomic_read(&zmd->nr_mblks);
 626}
 627
 628/*
 629 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
 630 */
 631static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
 632					      struct shrink_control *sc)
 633{
 634	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
 635	unsigned long count;
 636
 637	spin_lock(&zmd->mblk_lock);
 638	count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
 639	spin_unlock(&zmd->mblk_lock);
 640
 641	return count ? count : SHRINK_STOP;
 642}
 643
 644/*
 645 * Release a metadata block.
 646 */
 647static void dmz_release_mblock(struct dmz_metadata *zmd,
 648			       struct dmz_mblock *mblk)
 649{
 650
 651	if (!mblk)
 652		return;
 653
 654	spin_lock(&zmd->mblk_lock);
 655
 656	mblk->ref--;
 657	if (mblk->ref == 0) {
 658		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 659			rb_erase(&mblk->node, &zmd->mblk_rbtree);
 660			dmz_free_mblock(zmd, mblk);
 661		} else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
 662			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
 663			dmz_shrink_mblock_cache(zmd, 1);
 664		}
 665	}
 666
 667	spin_unlock(&zmd->mblk_lock);
 668}
 669
 670/*
 671 * Get a metadata block from the rbtree. If the block
 672 * is not present, read it from disk.
 673 */
 674static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
 675					 sector_t mblk_no)
 676{
 677	struct dmz_mblock *mblk;
 678	struct dmz_dev *dev = zmd->sb[zmd->mblk_primary].dev;
 679
 680	/* Check rbtree */
 681	spin_lock(&zmd->mblk_lock);
 682	mblk = dmz_get_mblock_fast(zmd, mblk_no);
 683	spin_unlock(&zmd->mblk_lock);
 684
 685	if (!mblk) {
 686		/* Cache miss: read the block from disk */
 687		mblk = dmz_get_mblock_slow(zmd, mblk_no);
 688		if (IS_ERR(mblk))
 689			return mblk;
 690	}
 691
 692	/* Wait for on-going read I/O and check for error */
 693	wait_on_bit_io(&mblk->state, DMZ_META_READING,
 694		       TASK_UNINTERRUPTIBLE);
 695	if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 696		dmz_release_mblock(zmd, mblk);
 697		dmz_check_bdev(dev);
 698		return ERR_PTR(-EIO);
 699	}
 700
 701	return mblk;
 702}
 703
 704/*
 705 * Mark a metadata block dirty.
 706 */
 707static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
 708{
 709	spin_lock(&zmd->mblk_lock);
 710	if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
 711		list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
 712	spin_unlock(&zmd->mblk_lock);
 713}
 714
 715/*
 716 * Issue a metadata block write BIO.
 717 */
 718static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
 719			    unsigned int set)
 720{
 721	struct dmz_dev *dev = zmd->sb[set].dev;
 722	sector_t block = zmd->sb[set].block + mblk->no;
 723	struct bio *bio;
 724
 725	if (dmz_bdev_is_dying(dev))
 726		return -EIO;
 727
 728	bio = bio_alloc(GFP_NOIO, 1);
 729	if (!bio) {
 730		set_bit(DMZ_META_ERROR, &mblk->state);
 731		return -ENOMEM;
 732	}
 733
 734	set_bit(DMZ_META_WRITING, &mblk->state);
 735
 736	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 737	bio_set_dev(bio, dev->bdev);
 738	bio->bi_private = mblk;
 739	bio->bi_end_io = dmz_mblock_bio_end_io;
 740	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
 741	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
 742	submit_bio(bio);
 743
 744	return 0;
 745}
 746
 747/*
 748 * Read/write a metadata block.
 749 */
 750static int dmz_rdwr_block(struct dmz_dev *dev, int op,
 751			  sector_t block, struct page *page)
 752{
 753	struct bio *bio;
 754	int ret;
 755
 756	if (WARN_ON(!dev))
 757		return -EIO;
 758
 759	if (dmz_bdev_is_dying(dev))
 760		return -EIO;
 761
 762	bio = bio_alloc(GFP_NOIO, 1);
 763	if (!bio)
 764		return -ENOMEM;
 765
 766	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 767	bio_set_dev(bio, dev->bdev);
 768	bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
 769	bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
 770	ret = submit_bio_wait(bio);
 771	bio_put(bio);
 772
 773	if (ret)
 774		dmz_check_bdev(dev);
 775	return ret;
 776}
 777
 778/*
 779 * Write super block of the specified metadata set.
 780 */
 781static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
 782{
 783	struct dmz_mblock *mblk = zmd->sb[set].mblk;
 784	struct dmz_super *sb = zmd->sb[set].sb;
 785	struct dmz_dev *dev = zmd->sb[set].dev;
 786	sector_t sb_block;
 787	u64 sb_gen = zmd->sb_gen + 1;
 788	int ret;
 789
 790	sb->magic = cpu_to_le32(DMZ_MAGIC);
 791
 792	sb->version = cpu_to_le32(zmd->sb_version);
 793	if (zmd->sb_version > 1) {
 794		BUILD_BUG_ON(UUID_SIZE != 16);
 795		export_uuid(sb->dmz_uuid, &zmd->uuid);
 796		memcpy(sb->dmz_label, zmd->label, BDEVNAME_SIZE);
 797		export_uuid(sb->dev_uuid, &dev->uuid);
 798	}
 799
 800	sb->gen = cpu_to_le64(sb_gen);
 801
 802	/*
 803	 * The metadata always references the absolute block address,
 804	 * ie relative to the entire block range, not the per-device
 805	 * block address.
 806	 */
 807	sb_block = zmd->sb[set].zone->id << zmd->zone_nr_blocks_shift;
 808	sb->sb_block = cpu_to_le64(sb_block);
 809	sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
 810	sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
 811	sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
 812
 813	sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
 814	sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
 815
 816	sb->crc = 0;
 817	sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
 818
 819	ret = dmz_rdwr_block(dev, REQ_OP_WRITE, zmd->sb[set].block,
 820			     mblk->page);
 821	if (ret == 0)
 822		ret = blkdev_issue_flush(dev->bdev, GFP_NOIO);
 823
 824	return ret;
 825}
 826
 827/*
 828 * Write dirty metadata blocks to the specified set.
 829 */
 830static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
 831				   struct list_head *write_list,
 832				   unsigned int set)
 833{
 834	struct dmz_mblock *mblk;
 835	struct dmz_dev *dev = zmd->sb[set].dev;
 836	struct blk_plug plug;
 837	int ret = 0, nr_mblks_submitted = 0;
 838
 839	/* Issue writes */
 840	blk_start_plug(&plug);
 841	list_for_each_entry(mblk, write_list, link) {
 842		ret = dmz_write_mblock(zmd, mblk, set);
 843		if (ret)
 844			break;
 845		nr_mblks_submitted++;
 846	}
 847	blk_finish_plug(&plug);
 848
 849	/* Wait for completion */
 850	list_for_each_entry(mblk, write_list, link) {
 851		if (!nr_mblks_submitted)
 852			break;
 853		wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
 854			       TASK_UNINTERRUPTIBLE);
 855		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 856			clear_bit(DMZ_META_ERROR, &mblk->state);
 857			dmz_check_bdev(dev);
 858			ret = -EIO;
 859		}
 860		nr_mblks_submitted--;
 861	}
 862
 863	/* Flush drive cache (this will also sync data) */
 864	if (ret == 0)
 865		ret = blkdev_issue_flush(dev->bdev, GFP_NOIO);
 866
 867	return ret;
 868}
 869
 870/*
 871 * Log dirty metadata blocks.
 872 */
 873static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
 874				 struct list_head *write_list)
 875{
 876	unsigned int log_set = zmd->mblk_primary ^ 0x1;
 877	int ret;
 878
 879	/* Write dirty blocks to the log */
 880	ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
 881	if (ret)
 882		return ret;
 883
 884	/*
 885	 * No error so far: now validate the log by updating the
 886	 * log index super block generation.
 887	 */
 888	ret = dmz_write_sb(zmd, log_set);
 889	if (ret)
 890		return ret;
 891
 892	return 0;
 893}
 894
 895/*
 896 * Flush dirty metadata blocks.
 897 */
 898int dmz_flush_metadata(struct dmz_metadata *zmd)
 899{
 900	struct dmz_mblock *mblk;
 901	struct list_head write_list;
 902	struct dmz_dev *dev;
 903	int ret;
 904
 905	if (WARN_ON(!zmd))
 906		return 0;
 907
 908	INIT_LIST_HEAD(&write_list);
 909
 910	/*
 911	 * Make sure that metadata blocks are stable before logging: take
 912	 * the write lock on the metadata semaphore to prevent target BIOs
 913	 * from modifying metadata.
 914	 */
 915	down_write(&zmd->mblk_sem);
 916	dev = zmd->sb[zmd->mblk_primary].dev;
 917
 918	/*
 919	 * This is called from the target flush work and reclaim work.
 920	 * Concurrent execution is not allowed.
 921	 */
 922	dmz_lock_flush(zmd);
 923
 924	if (dmz_bdev_is_dying(dev)) {
 925		ret = -EIO;
 926		goto out;
 927	}
 928
 929	/* Get dirty blocks */
 930	spin_lock(&zmd->mblk_lock);
 931	list_splice_init(&zmd->mblk_dirty_list, &write_list);
 932	spin_unlock(&zmd->mblk_lock);
 933
 934	/* If there are no dirty metadata blocks, just flush the device cache */
 935	if (list_empty(&write_list)) {
 936		ret = blkdev_issue_flush(dev->bdev, GFP_NOIO);
 937		goto err;
 938	}
 939
 940	/*
 941	 * The primary metadata set is still clean. Keep it this way until
 942	 * all updates are successful in the secondary set. That is, use
 943	 * the secondary set as a log.
 944	 */
 945	ret = dmz_log_dirty_mblocks(zmd, &write_list);
 946	if (ret)
 947		goto err;
 948
 949	/*
 950	 * The log is on disk. It is now safe to update in place
 951	 * in the primary metadata set.
 952	 */
 953	ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
 954	if (ret)
 955		goto err;
 956
 957	ret = dmz_write_sb(zmd, zmd->mblk_primary);
 958	if (ret)
 959		goto err;
 960
 961	while (!list_empty(&write_list)) {
 962		mblk = list_first_entry(&write_list, struct dmz_mblock, link);
 963		list_del_init(&mblk->link);
 964
 965		spin_lock(&zmd->mblk_lock);
 966		clear_bit(DMZ_META_DIRTY, &mblk->state);
 967		if (mblk->ref == 0)
 968			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
 969		spin_unlock(&zmd->mblk_lock);
 970	}
 971
 972	zmd->sb_gen++;
 973out:
 974	dmz_unlock_flush(zmd);
 975	up_write(&zmd->mblk_sem);
 976
 977	return ret;
 978
 979err:
 980	if (!list_empty(&write_list)) {
 981		spin_lock(&zmd->mblk_lock);
 982		list_splice(&write_list, &zmd->mblk_dirty_list);
 983		spin_unlock(&zmd->mblk_lock);
 984	}
 985	if (!dmz_check_bdev(dev))
 986		ret = -EIO;
 987	goto out;
 988}
 989
 990/*
 991 * Check super block.
 992 */
 993static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_sb *dsb,
 994			bool tertiary)
 995{
 996	struct dmz_super *sb = dsb->sb;
 997	struct dmz_dev *dev = dsb->dev;
 998	unsigned int nr_meta_zones, nr_data_zones;
 999	u32 crc, stored_crc;
1000	u64 gen, sb_block;
1001
1002	if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
1003		dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
1004			    DMZ_MAGIC, le32_to_cpu(sb->magic));
1005		return -ENXIO;
1006	}
1007
1008	zmd->sb_version = le32_to_cpu(sb->version);
1009	if (zmd->sb_version > DMZ_META_VER) {
1010		dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
1011			    DMZ_META_VER, zmd->sb_version);
1012		return -EINVAL;
1013	}
1014	if (zmd->sb_version < 2 && tertiary) {
1015		dmz_dev_err(dev, "Tertiary superblocks are not supported");
1016		return -EINVAL;
1017	}
1018
1019	gen = le64_to_cpu(sb->gen);
1020	stored_crc = le32_to_cpu(sb->crc);
1021	sb->crc = 0;
1022	crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
1023	if (crc != stored_crc) {
1024		dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
1025			    crc, stored_crc);
1026		return -ENXIO;
1027	}
1028
1029	sb_block = le64_to_cpu(sb->sb_block);
1030	if (sb_block != (u64)dsb->zone->id << zmd->zone_nr_blocks_shift ) {
1031		dmz_dev_err(dev, "Invalid superblock position "
1032			    "(is %llu expected %llu)",
1033			    sb_block,
1034			    (u64)dsb->zone->id << zmd->zone_nr_blocks_shift);
1035		return -EINVAL;
1036	}
1037	if (zmd->sb_version > 1) {
1038		uuid_t sb_uuid;
1039
1040		import_uuid(&sb_uuid, sb->dmz_uuid);
1041		if (uuid_is_null(&sb_uuid)) {
1042			dmz_dev_err(dev, "NULL DM-Zoned uuid");
1043			return -ENXIO;
1044		} else if (uuid_is_null(&zmd->uuid)) {
1045			uuid_copy(&zmd->uuid, &sb_uuid);
1046		} else if (!uuid_equal(&zmd->uuid, &sb_uuid)) {
1047			dmz_dev_err(dev, "mismatching DM-Zoned uuid, "
1048				    "is %pUl expected %pUl",
1049				    &sb_uuid, &zmd->uuid);
1050			return -ENXIO;
1051		}
1052		if (!strlen(zmd->label))
1053			memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE);
1054		else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) {
1055			dmz_dev_err(dev, "mismatching DM-Zoned label, "
1056				    "is %s expected %s",
1057				    sb->dmz_label, zmd->label);
1058			return -ENXIO;
1059		}
1060		import_uuid(&dev->uuid, sb->dev_uuid);
1061		if (uuid_is_null(&dev->uuid)) {
1062			dmz_dev_err(dev, "NULL device uuid");
1063			return -ENXIO;
1064		}
1065
1066		if (tertiary) {
1067			/*
1068			 * Generation number should be 0, but it doesn't
1069			 * really matter if it isn't.
1070			 */
1071			if (gen != 0)
1072				dmz_dev_warn(dev, "Invalid generation %llu",
1073					    gen);
1074			return 0;
1075		}
1076	}
1077
1078	nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1)
1079		>> zmd->zone_nr_blocks_shift;
1080	if (!nr_meta_zones ||
1081	    (zmd->nr_devs <= 1 && nr_meta_zones >= zmd->nr_rnd_zones) ||
1082	    (zmd->nr_devs > 1 && nr_meta_zones >= zmd->nr_cache_zones)) {
1083		dmz_dev_err(dev, "Invalid number of metadata blocks");
1084		return -ENXIO;
1085	}
1086
1087	if (!le32_to_cpu(sb->nr_reserved_seq) ||
1088	    le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
1089		dmz_dev_err(dev, "Invalid number of reserved sequential zones");
1090		return -ENXIO;
1091	}
1092
1093	nr_data_zones = zmd->nr_useable_zones -
1094		(nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
1095	if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
1096		dmz_dev_err(dev, "Invalid number of chunks %u / %u",
1097			    le32_to_cpu(sb->nr_chunks), nr_data_zones);
1098		return -ENXIO;
1099	}
1100
1101	/* OK */
1102	zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
1103	zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
1104	zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
1105	zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
1106	zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
1107	zmd->nr_meta_zones = nr_meta_zones;
1108	zmd->nr_data_zones = nr_data_zones;
1109
1110	return 0;
1111}
1112
1113/*
1114 * Read the first or second super block from disk.
1115 */
1116static int dmz_read_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1117{
1118	dmz_zmd_debug(zmd, "read superblock set %d dev %s block %llu",
1119		      set, sb->dev->name, sb->block);
1120
1121	return dmz_rdwr_block(sb->dev, REQ_OP_READ,
1122			      sb->block, sb->mblk->page);
1123}
1124
1125/*
1126 * Determine the position of the secondary super blocks on disk.
1127 * This is used only if a corruption of the primary super block
1128 * is detected.
1129 */
1130static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
1131{
1132	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
1133	struct dmz_mblock *mblk;
1134	unsigned int zone_id = zmd->sb[0].zone->id;
1135	int i;
1136
1137	/* Allocate a block */
1138	mblk = dmz_alloc_mblock(zmd, 0);
1139	if (!mblk)
1140		return -ENOMEM;
1141
1142	zmd->sb[1].mblk = mblk;
1143	zmd->sb[1].sb = mblk->data;
1144
1145	/* Bad first super block: search for the second one */
1146	zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
1147	zmd->sb[1].zone = dmz_get(zmd, zone_id + 1);
1148	zmd->sb[1].dev = zmd->sb[0].dev;
1149	for (i = 1; i < zmd->nr_rnd_zones; i++) {
1150		if (dmz_read_sb(zmd, &zmd->sb[1], 1) != 0)
1151			break;
1152		if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
1153			return 0;
1154		zmd->sb[1].block += zone_nr_blocks;
1155		zmd->sb[1].zone = dmz_get(zmd, zone_id + i);
1156	}
1157
1158	dmz_free_mblock(zmd, mblk);
1159	zmd->sb[1].mblk = NULL;
1160	zmd->sb[1].zone = NULL;
1161	zmd->sb[1].dev = NULL;
1162
1163	return -EIO;
1164}
1165
1166/*
1167 * Read a super block from disk.
1168 */
1169static int dmz_get_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1170{
1171	struct dmz_mblock *mblk;
1172	int ret;
1173
1174	/* Allocate a block */
1175	mblk = dmz_alloc_mblock(zmd, 0);
1176	if (!mblk)
1177		return -ENOMEM;
1178
1179	sb->mblk = mblk;
1180	sb->sb = mblk->data;
1181
1182	/* Read super block */
1183	ret = dmz_read_sb(zmd, sb, set);
1184	if (ret) {
1185		dmz_free_mblock(zmd, mblk);
1186		sb->mblk = NULL;
1187		return ret;
1188	}
1189
1190	return 0;
1191}
1192
1193/*
1194 * Recover a metadata set.
1195 */
1196static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
1197{
1198	unsigned int src_set = dst_set ^ 0x1;
1199	struct page *page;
1200	int i, ret;
1201
1202	dmz_dev_warn(zmd->sb[dst_set].dev,
1203		     "Metadata set %u invalid: recovering", dst_set);
1204
1205	if (dst_set == 0)
1206		zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1207	else
1208		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1209
1210	page = alloc_page(GFP_NOIO);
1211	if (!page)
1212		return -ENOMEM;
1213
1214	/* Copy metadata blocks */
1215	for (i = 1; i < zmd->nr_meta_blocks; i++) {
1216		ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ,
1217				     zmd->sb[src_set].block + i, page);
1218		if (ret)
1219			goto out;
1220		ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE,
1221				     zmd->sb[dst_set].block + i, page);
1222		if (ret)
1223			goto out;
1224	}
1225
1226	/* Finalize with the super block */
1227	if (!zmd->sb[dst_set].mblk) {
1228		zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1229		if (!zmd->sb[dst_set].mblk) {
1230			ret = -ENOMEM;
1231			goto out;
1232		}
1233		zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1234	}
1235
1236	ret = dmz_write_sb(zmd, dst_set);
1237out:
1238	__free_pages(page, 0);
1239
1240	return ret;
1241}
1242
1243/*
1244 * Get super block from disk.
1245 */
1246static int dmz_load_sb(struct dmz_metadata *zmd)
1247{
1248	bool sb_good[2] = {false, false};
1249	u64 sb_gen[2] = {0, 0};
1250	int ret;
1251
1252	if (!zmd->sb[0].zone) {
1253		dmz_zmd_err(zmd, "Primary super block zone not set");
1254		return -ENXIO;
1255	}
1256
1257	/* Read and check the primary super block */
1258	zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1259	zmd->sb[0].dev = zmd->sb[0].zone->dev;
1260	ret = dmz_get_sb(zmd, &zmd->sb[0], 0);
1261	if (ret) {
1262		dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed");
1263		return ret;
1264	}
1265
1266	ret = dmz_check_sb(zmd, &zmd->sb[0], false);
1267
1268	/* Read and check secondary super block */
1269	if (ret == 0) {
1270		sb_good[0] = true;
1271		if (!zmd->sb[1].zone) {
1272			unsigned int zone_id =
1273				zmd->sb[0].zone->id + zmd->nr_meta_zones;
1274
1275			zmd->sb[1].zone = dmz_get(zmd, zone_id);
1276		}
1277		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1278		zmd->sb[1].dev = zmd->sb[0].dev;
1279		ret = dmz_get_sb(zmd, &zmd->sb[1], 1);
1280	} else
1281		ret = dmz_lookup_secondary_sb(zmd);
1282
1283	if (ret) {
1284		dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed");
1285		return ret;
1286	}
1287
1288	ret = dmz_check_sb(zmd, &zmd->sb[1], false);
1289	if (ret == 0)
1290		sb_good[1] = true;
1291
1292	/* Use highest generation sb first */
1293	if (!sb_good[0] && !sb_good[1]) {
1294		dmz_zmd_err(zmd, "No valid super block found");
1295		return -EIO;
1296	}
1297
1298	if (sb_good[0])
1299		sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1300	else {
1301		ret = dmz_recover_mblocks(zmd, 0);
1302		if (ret) {
1303			dmz_dev_err(zmd->sb[0].dev,
1304				    "Recovery of superblock 0 failed");
1305			return -EIO;
1306		}
1307	}
1308
1309	if (sb_good[1])
1310		sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1311	else {
1312		ret = dmz_recover_mblocks(zmd, 1);
1313
1314		if (ret) {
1315			dmz_dev_err(zmd->sb[1].dev,
1316				    "Recovery of superblock 1 failed");
1317			return -EIO;
1318		}
1319	}
1320
1321	if (sb_gen[0] >= sb_gen[1]) {
1322		zmd->sb_gen = sb_gen[0];
1323		zmd->mblk_primary = 0;
1324	} else {
1325		zmd->sb_gen = sb_gen[1];
1326		zmd->mblk_primary = 1;
1327	}
1328
1329	dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev,
1330		      "Using super block %u (gen %llu)",
1331		      zmd->mblk_primary, zmd->sb_gen);
1332
1333	if (zmd->sb_version > 1) {
1334		int i;
1335		struct dmz_sb *sb;
1336
1337		sb = kzalloc(sizeof(struct dmz_sb), GFP_KERNEL);
1338		if (!sb)
1339			return -ENOMEM;
1340		for (i = 1; i < zmd->nr_devs; i++) {
1341			sb->block = 0;
1342			sb->zone = dmz_get(zmd, zmd->dev[i].zone_offset);
1343			sb->dev = &zmd->dev[i];
1344			if (!dmz_is_meta(sb->zone)) {
1345				dmz_dev_err(sb->dev,
1346					    "Tertiary super block zone %u not marked as metadata zone",
1347					    sb->zone->id);
1348				ret = -EINVAL;
1349				goto out_kfree;
1350			}
1351			ret = dmz_get_sb(zmd, sb, i + 1);
1352			if (ret) {
1353				dmz_dev_err(sb->dev,
1354					    "Read tertiary super block failed");
1355				dmz_free_mblock(zmd, sb->mblk);
1356				goto out_kfree;
1357			}
1358			ret = dmz_check_sb(zmd, sb, true);
1359			dmz_free_mblock(zmd, sb->mblk);
1360			if (ret == -EINVAL)
1361				goto out_kfree;
1362		}
1363	out_kfree:
1364		kfree(sb);
1365	}
1366	return ret;
1367}
1368
1369/*
1370 * Initialize a zone descriptor.
1371 */
1372static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data)
1373{
1374	struct dmz_dev *dev = data;
1375	struct dmz_metadata *zmd = dev->metadata;
1376	int idx = num + dev->zone_offset;
1377	struct dm_zone *zone;
1378
1379	zone = dmz_insert(zmd, idx, dev);
1380	if (IS_ERR(zone))
1381		return PTR_ERR(zone);
1382
1383	if (blkz->len != zmd->zone_nr_sectors) {
1384		if (zmd->sb_version > 1) {
1385			/* Ignore the eventual runt (smaller) zone */
1386			set_bit(DMZ_OFFLINE, &zone->flags);
1387			return 0;
1388		} else if (blkz->start + blkz->len == dev->capacity)
1389			return 0;
1390		return -ENXIO;
1391	}
1392
1393	switch (blkz->type) {
1394	case BLK_ZONE_TYPE_CONVENTIONAL:
1395		set_bit(DMZ_RND, &zone->flags);
1396		break;
1397	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1398	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1399		set_bit(DMZ_SEQ, &zone->flags);
1400		break;
1401	default:
1402		return -ENXIO;
1403	}
1404
1405	if (dmz_is_rnd(zone))
1406		zone->wp_block = 0;
1407	else
1408		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1409
1410	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1411		set_bit(DMZ_OFFLINE, &zone->flags);
1412	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1413		set_bit(DMZ_READ_ONLY, &zone->flags);
1414	else {
1415		zmd->nr_useable_zones++;
1416		if (dmz_is_rnd(zone)) {
1417			zmd->nr_rnd_zones++;
1418			if (zmd->nr_devs == 1 && !zmd->sb[0].zone) {
1419				/* Primary super block zone */
1420				zmd->sb[0].zone = zone;
1421			}
1422		}
1423		if (zmd->nr_devs > 1 && num == 0) {
1424			/*
1425			 * Tertiary superblock zones are always at the
1426			 * start of the zoned devices, so mark them
1427			 * as metadata zone.
1428			 */
1429			set_bit(DMZ_META, &zone->flags);
1430		}
1431	}
1432	return 0;
1433}
1434
1435static int dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev)
1436{
1437	int idx;
1438	sector_t zone_offset = 0;
1439
1440	for(idx = 0; idx < dev->nr_zones; idx++) {
1441		struct dm_zone *zone;
1442
1443		zone = dmz_insert(zmd, idx, dev);
1444		if (IS_ERR(zone))
1445			return PTR_ERR(zone);
1446		set_bit(DMZ_CACHE, &zone->flags);
1447		zone->wp_block = 0;
1448		zmd->nr_cache_zones++;
1449		zmd->nr_useable_zones++;
1450		if (dev->capacity - zone_offset < zmd->zone_nr_sectors) {
1451			/* Disable runt zone */
1452			set_bit(DMZ_OFFLINE, &zone->flags);
1453			break;
1454		}
1455		zone_offset += zmd->zone_nr_sectors;
1456	}
1457	return 0;
1458}
1459
1460/*
1461 * Free zones descriptors.
1462 */
1463static void dmz_drop_zones(struct dmz_metadata *zmd)
1464{
1465	int idx;
1466
1467	for(idx = 0; idx < zmd->nr_zones; idx++) {
1468		struct dm_zone *zone = xa_load(&zmd->zones, idx);
1469
1470		kfree(zone);
1471		xa_erase(&zmd->zones, idx);
1472	}
1473	xa_destroy(&zmd->zones);
1474}
1475
1476/*
1477 * Allocate and initialize zone descriptors using the zone
1478 * information from disk.
1479 */
1480static int dmz_init_zones(struct dmz_metadata *zmd)
1481{
1482	int i, ret;
1483	struct dmz_dev *zoned_dev = &zmd->dev[0];
1484
1485	/* Init */
1486	zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors;
1487	zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors);
1488	zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors);
1489	zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks);
1490	zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3;
1491	zmd->zone_nr_bitmap_blocks =
1492		max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1493	zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks,
1494					DMZ_BLOCK_SIZE_BITS);
1495
1496	/* Allocate zone array */
1497	zmd->nr_zones = 0;
1498	for (i = 0; i < zmd->nr_devs; i++) {
1499		struct dmz_dev *dev = &zmd->dev[i];
1500
1501		dev->metadata = zmd;
1502		zmd->nr_zones += dev->nr_zones;
1503
1504		atomic_set(&dev->unmap_nr_rnd, 0);
1505		INIT_LIST_HEAD(&dev->unmap_rnd_list);
1506		INIT_LIST_HEAD(&dev->map_rnd_list);
1507
1508		atomic_set(&dev->unmap_nr_seq, 0);
1509		INIT_LIST_HEAD(&dev->unmap_seq_list);
1510		INIT_LIST_HEAD(&dev->map_seq_list);
1511	}
1512
1513	if (!zmd->nr_zones) {
1514		DMERR("(%s): No zones found", zmd->devname);
1515		return -ENXIO;
1516	}
1517	xa_init(&zmd->zones);
1518
1519	DMDEBUG("(%s): Using %zu B for zone information",
1520		zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones);
1521
1522	if (zmd->nr_devs > 1) {
1523		ret = dmz_emulate_zones(zmd, &zmd->dev[0]);
1524		if (ret < 0) {
1525			DMDEBUG("(%s): Failed to emulate zones, error %d",
1526				zmd->devname, ret);
1527			dmz_drop_zones(zmd);
1528			return ret;
1529		}
1530
1531		/*
1532		 * Primary superblock zone is always at zone 0 when multiple
1533		 * drives are present.
1534		 */
1535		zmd->sb[0].zone = dmz_get(zmd, 0);
1536
1537		for (i = 1; i < zmd->nr_devs; i++) {
1538			zoned_dev = &zmd->dev[i];
1539
1540			ret = blkdev_report_zones(zoned_dev->bdev, 0,
1541						  BLK_ALL_ZONES,
1542						  dmz_init_zone, zoned_dev);
1543			if (ret < 0) {
1544				DMDEBUG("(%s): Failed to report zones, error %d",
1545					zmd->devname, ret);
1546				dmz_drop_zones(zmd);
1547				return ret;
1548			}
1549		}
1550		return 0;
1551	}
1552
1553	/*
1554	 * Get zone information and initialize zone descriptors.  At the same
1555	 * time, determine where the super block should be: first block of the
1556	 * first randomly writable zone.
1557	 */
1558	ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES,
1559				  dmz_init_zone, zoned_dev);
1560	if (ret < 0) {
1561		DMDEBUG("(%s): Failed to report zones, error %d",
1562			zmd->devname, ret);
1563		dmz_drop_zones(zmd);
1564		return ret;
1565	}
1566
1567	return 0;
1568}
1569
1570static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1571			      void *data)
1572{
1573	struct dm_zone *zone = data;
1574
1575	clear_bit(DMZ_OFFLINE, &zone->flags);
1576	clear_bit(DMZ_READ_ONLY, &zone->flags);
1577	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1578		set_bit(DMZ_OFFLINE, &zone->flags);
1579	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1580		set_bit(DMZ_READ_ONLY, &zone->flags);
1581
1582	if (dmz_is_seq(zone))
1583		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1584	else
1585		zone->wp_block = 0;
1586	return 0;
1587}
1588
1589/*
1590 * Update a zone information.
1591 */
1592static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1593{
1594	struct dmz_dev *dev = zone->dev;
1595	unsigned int noio_flag;
1596	int ret;
1597
1598	if (dev->flags & DMZ_BDEV_REGULAR)
1599		return 0;
1600
1601	/*
1602	 * Get zone information from disk. Since blkdev_report_zones() uses
1603	 * GFP_KERNEL by default for memory allocations, set the per-task
1604	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1605	 * GFP_NOIO was specified.
1606	 */
1607	noio_flag = memalloc_noio_save();
1608	ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1,
1609				  dmz_update_zone_cb, zone);
1610	memalloc_noio_restore(noio_flag);
1611
1612	if (ret == 0)
1613		ret = -EIO;
1614	if (ret < 0) {
1615		dmz_dev_err(dev, "Get zone %u report failed",
1616			    zone->id);
1617		dmz_check_bdev(dev);
1618		return ret;
1619	}
1620
1621	return 0;
1622}
1623
1624/*
1625 * Check a zone write pointer position when the zone is marked
1626 * with the sequential write error flag.
1627 */
1628static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1629				    struct dm_zone *zone)
1630{
1631	struct dmz_dev *dev = zone->dev;
1632	unsigned int wp = 0;
1633	int ret;
1634
1635	wp = zone->wp_block;
1636	ret = dmz_update_zone(zmd, zone);
1637	if (ret)
1638		return ret;
1639
1640	dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)",
1641		     zone->id, zone->wp_block, wp);
1642
1643	if (zone->wp_block < wp) {
1644		dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1645				      wp - zone->wp_block);
1646	}
1647
1648	return 0;
1649}
1650
1651/*
1652 * Reset a zone write pointer.
1653 */
1654static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1655{
1656	int ret;
1657
1658	/*
1659	 * Ignore offline zones, read only zones,
1660	 * and conventional zones.
1661	 */
1662	if (dmz_is_offline(zone) ||
1663	    dmz_is_readonly(zone) ||
1664	    dmz_is_rnd(zone))
1665		return 0;
1666
1667	if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1668		struct dmz_dev *dev = zone->dev;
1669
1670		ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1671				       dmz_start_sect(zmd, zone),
1672				       zmd->zone_nr_sectors, GFP_NOIO);
1673		if (ret) {
1674			dmz_dev_err(dev, "Reset zone %u failed %d",
1675				    zone->id, ret);
1676			return ret;
1677		}
1678	}
1679
1680	/* Clear write error bit and rewind write pointer position */
1681	clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1682	zone->wp_block = 0;
1683
1684	return 0;
1685}
1686
1687static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1688
1689/*
1690 * Initialize chunk mapping.
1691 */
1692static int dmz_load_mapping(struct dmz_metadata *zmd)
1693{
1694	struct dm_zone *dzone, *bzone;
1695	struct dmz_mblock *dmap_mblk = NULL;
1696	struct dmz_map *dmap;
1697	unsigned int i = 0, e = 0, chunk = 0;
1698	unsigned int dzone_id;
1699	unsigned int bzone_id;
1700
1701	/* Metadata block array for the chunk mapping table */
1702	zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1703				sizeof(struct dmz_mblk *), GFP_KERNEL);
1704	if (!zmd->map_mblk)
1705		return -ENOMEM;
1706
1707	/* Get chunk mapping table blocks and initialize zone mapping */
1708	while (chunk < zmd->nr_chunks) {
1709		if (!dmap_mblk) {
1710			/* Get mapping block */
1711			dmap_mblk = dmz_get_mblock(zmd, i + 1);
1712			if (IS_ERR(dmap_mblk))
1713				return PTR_ERR(dmap_mblk);
1714			zmd->map_mblk[i] = dmap_mblk;
1715			dmap = (struct dmz_map *) dmap_mblk->data;
1716			i++;
1717			e = 0;
1718		}
1719
1720		/* Check data zone */
1721		dzone_id = le32_to_cpu(dmap[e].dzone_id);
1722		if (dzone_id == DMZ_MAP_UNMAPPED)
1723			goto next;
1724
1725		if (dzone_id >= zmd->nr_zones) {
1726			dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u",
1727				    chunk, dzone_id);
1728			return -EIO;
1729		}
1730
1731		dzone = dmz_get(zmd, dzone_id);
1732		if (!dzone) {
1733			dmz_zmd_err(zmd, "Chunk %u mapping: data zone %u not present",
1734				    chunk, dzone_id);
1735			return -EIO;
1736		}
1737		set_bit(DMZ_DATA, &dzone->flags);
1738		dzone->chunk = chunk;
1739		dmz_get_zone_weight(zmd, dzone);
1740
1741		if (dmz_is_cache(dzone))
1742			list_add_tail(&dzone->link, &zmd->map_cache_list);
1743		else if (dmz_is_rnd(dzone))
1744			list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
1745		else
1746			list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
1747
1748		/* Check buffer zone */
1749		bzone_id = le32_to_cpu(dmap[e].bzone_id);
1750		if (bzone_id == DMZ_MAP_UNMAPPED)
1751			goto next;
1752
1753		if (bzone_id >= zmd->nr_zones) {
1754			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u",
1755				    chunk, bzone_id);
1756			return -EIO;
1757		}
1758
1759		bzone = dmz_get(zmd, bzone_id);
1760		if (!bzone) {
1761			dmz_zmd_err(zmd, "Chunk %u mapping: buffer zone %u not present",
1762				    chunk, bzone_id);
1763			return -EIO;
1764		}
1765		if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) {
1766			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u",
1767				    chunk, bzone_id);
1768			return -EIO;
1769		}
1770
1771		set_bit(DMZ_DATA, &bzone->flags);
1772		set_bit(DMZ_BUF, &bzone->flags);
1773		bzone->chunk = chunk;
1774		bzone->bzone = dzone;
1775		dzone->bzone = bzone;
1776		dmz_get_zone_weight(zmd, bzone);
1777		if (dmz_is_cache(bzone))
1778			list_add_tail(&bzone->link, &zmd->map_cache_list);
1779		else
1780			list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
1781next:
1782		chunk++;
1783		e++;
1784		if (e >= DMZ_MAP_ENTRIES)
1785			dmap_mblk = NULL;
1786	}
1787
1788	/*
1789	 * At this point, only meta zones and mapped data zones were
1790	 * fully initialized. All remaining zones are unmapped data
1791	 * zones. Finish initializing those here.
1792	 */
1793	for (i = 0; i < zmd->nr_zones; i++) {
1794		dzone = dmz_get(zmd, i);
1795		if (!dzone)
1796			continue;
1797		if (dmz_is_meta(dzone))
1798			continue;
1799		if (dmz_is_offline(dzone))
1800			continue;
1801
1802		if (dmz_is_cache(dzone))
1803			zmd->nr_cache++;
1804		else if (dmz_is_rnd(dzone))
1805			dzone->dev->nr_rnd++;
1806		else
1807			dzone->dev->nr_seq++;
1808
1809		if (dmz_is_data(dzone)) {
1810			/* Already initialized */
1811			continue;
1812		}
1813
1814		/* Unmapped data zone */
1815		set_bit(DMZ_DATA, &dzone->flags);
1816		dzone->chunk = DMZ_MAP_UNMAPPED;
1817		if (dmz_is_cache(dzone)) {
1818			list_add_tail(&dzone->link, &zmd->unmap_cache_list);
1819			atomic_inc(&zmd->unmap_nr_cache);
1820		} else if (dmz_is_rnd(dzone)) {
1821			list_add_tail(&dzone->link,
1822				      &dzone->dev->unmap_rnd_list);
1823			atomic_inc(&dzone->dev->unmap_nr_rnd);
1824		} else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1825			list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1826			set_bit(DMZ_RESERVED, &dzone->flags);
1827			atomic_inc(&zmd->nr_reserved_seq_zones);
1828			dzone->dev->nr_seq--;
1829		} else {
1830			list_add_tail(&dzone->link,
1831				      &dzone->dev->unmap_seq_list);
1832			atomic_inc(&dzone->dev->unmap_nr_seq);
1833		}
1834	}
1835
1836	return 0;
1837}
1838
1839/*
1840 * Set a data chunk mapping.
1841 */
1842static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1843				  unsigned int dzone_id, unsigned int bzone_id)
1844{
1845	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1846	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1847	int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1848
1849	dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1850	dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1851	dmz_dirty_mblock(zmd, dmap_mblk);
1852}
1853
1854/*
1855 * The list of mapped zones is maintained in LRU order.
1856 * This rotates a zone at the end of its map list.
1857 */
1858static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1859{
1860	if (list_empty(&zone->link))
1861		return;
1862
1863	list_del_init(&zone->link);
1864	if (dmz_is_seq(zone)) {
1865		/* LRU rotate sequential zone */
1866		list_add_tail(&zone->link, &zone->dev->map_seq_list);
1867	} else if (dmz_is_cache(zone)) {
1868		/* LRU rotate cache zone */
1869		list_add_tail(&zone->link, &zmd->map_cache_list);
1870	} else {
1871		/* LRU rotate random zone */
1872		list_add_tail(&zone->link, &zone->dev->map_rnd_list);
1873	}
1874}
1875
1876/*
1877 * The list of mapped random zones is maintained
1878 * in LRU order. This rotates a zone at the end of the list.
1879 */
1880static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1881{
1882	__dmz_lru_zone(zmd, zone);
1883	if (zone->bzone)
1884		__dmz_lru_zone(zmd, zone->bzone);
1885}
1886
1887/*
1888 * Wait for any zone to be freed.
1889 */
1890static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1891{
1892	DEFINE_WAIT(wait);
1893
1894	prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1895	dmz_unlock_map(zmd);
1896	dmz_unlock_metadata(zmd);
1897
1898	io_schedule_timeout(HZ);
1899
1900	dmz_lock_metadata(zmd);
1901	dmz_lock_map(zmd);
1902	finish_wait(&zmd->free_wq, &wait);
1903}
1904
1905/*
1906 * Lock a zone for reclaim (set the zone RECLAIM bit).
1907 * Returns false if the zone cannot be locked or if it is already locked
1908 * and 1 otherwise.
1909 */
1910int dmz_lock_zone_reclaim(struct dm_zone *zone)
1911{
1912	/* Active zones cannot be reclaimed */
1913	if (dmz_is_active(zone))
1914		return 0;
1915
1916	return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1917}
1918
1919/*
1920 * Clear a zone reclaim flag.
1921 */
1922void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1923{
1924	WARN_ON(dmz_is_active(zone));
1925	WARN_ON(!dmz_in_reclaim(zone));
1926
1927	clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1928	smp_mb__after_atomic();
1929	wake_up_bit(&zone->flags, DMZ_RECLAIM);
1930}
1931
1932/*
1933 * Wait for a zone reclaim to complete.
1934 */
1935static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1936{
1937	dmz_unlock_map(zmd);
1938	dmz_unlock_metadata(zmd);
1939	set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1940	wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1941	clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1942	dmz_lock_metadata(zmd);
1943	dmz_lock_map(zmd);
1944}
1945
1946/*
1947 * Select a cache or random write zone for reclaim.
1948 */
1949static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd,
1950						    unsigned int idx, bool idle)
1951{
1952	struct dm_zone *dzone = NULL;
1953	struct dm_zone *zone, *maxw_z = NULL;
1954	struct list_head *zone_list;
1955
1956	/* If we have cache zones select from the cache zone list */
1957	if (zmd->nr_cache) {
1958		zone_list = &zmd->map_cache_list;
1959		/* Try to relaim random zones, too, when idle */
1960		if (idle && list_empty(zone_list))
1961			zone_list = &zmd->dev[idx].map_rnd_list;
1962	} else
1963		zone_list = &zmd->dev[idx].map_rnd_list;
1964
1965	/*
1966	 * Find the buffer zone with the heaviest weight or the first (oldest)
1967	 * data zone that can be reclaimed.
1968	 */
1969	list_for_each_entry(zone, zone_list, link) {
1970		if (dmz_is_buf(zone)) {
1971			dzone = zone->bzone;
1972			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1973				continue;
1974			if (!maxw_z || maxw_z->weight < dzone->weight)
1975				maxw_z = dzone;
1976		} else {
1977			dzone = zone;
1978			if (dmz_lock_zone_reclaim(dzone))
1979				return dzone;
1980		}
1981	}
1982
1983	if (maxw_z && dmz_lock_zone_reclaim(maxw_z))
1984		return maxw_z;
1985
1986	/*
1987	 * If we come here, none of the zones inspected could be locked for
1988	 * reclaim. Try again, being more aggressive, that is, find the
1989	 * first zone that can be reclaimed regardless of its weitght.
1990	 */
1991	list_for_each_entry(zone, zone_list, link) {
1992		if (dmz_is_buf(zone)) {
1993			dzone = zone->bzone;
1994			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1995				continue;
1996		} else
1997			dzone = zone;
1998		if (dmz_lock_zone_reclaim(dzone))
1999			return dzone;
2000	}
2001
2002	return NULL;
2003}
2004
2005/*
2006 * Select a buffered sequential zone for reclaim.
2007 */
2008static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd,
2009						    unsigned int idx)
2010{
2011	struct dm_zone *zone;
2012
2013	list_for_each_entry(zone, &zmd->dev[idx].map_seq_list, link) {
2014		if (!zone->bzone)
2015			continue;
2016		if (dmz_lock_zone_reclaim(zone))
2017			return zone;
2018	}
2019
2020	return NULL;
2021}
2022
2023/*
2024 * Select a zone for reclaim.
2025 */
2026struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd,
2027					 unsigned int dev_idx, bool idle)
2028{
2029	struct dm_zone *zone = NULL;
2030
2031	/*
2032	 * Search for a zone candidate to reclaim: 2 cases are possible.
2033	 * (1) There is no free sequential zones. Then a random data zone
2034	 *     cannot be reclaimed. So choose a sequential zone to reclaim so
2035	 *     that afterward a random zone can be reclaimed.
2036	 * (2) At least one free sequential zone is available, then choose
2037	 *     the oldest random zone (data or buffer) that can be locked.
2038	 */
2039	dmz_lock_map(zmd);
2040	if (list_empty(&zmd->reserved_seq_zones_list))
2041		zone = dmz_get_seq_zone_for_reclaim(zmd, dev_idx);
2042	if (!zone)
2043		zone = dmz_get_rnd_zone_for_reclaim(zmd, dev_idx, idle);
2044	dmz_unlock_map(zmd);
2045
2046	return zone;
2047}
2048
2049/*
2050 * Get the zone mapping a chunk, if the chunk is mapped already.
2051 * If no mapping exist and the operation is WRITE, a zone is
2052 * allocated and used to map the chunk.
2053 * The zone returned will be set to the active state.
2054 */
2055struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
2056{
2057	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
2058	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
2059	int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
2060	unsigned int dzone_id;
2061	struct dm_zone *dzone = NULL;
2062	int ret = 0;
2063	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2064
2065	dmz_lock_map(zmd);
2066again:
2067	/* Get the chunk mapping */
2068	dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
2069	if (dzone_id == DMZ_MAP_UNMAPPED) {
2070		/*
2071		 * Read or discard in unmapped chunks are fine. But for
2072		 * writes, we need a mapping, so get one.
2073		 */
2074		if (op != REQ_OP_WRITE)
2075			goto out;
2076
2077		/* Allocate a random zone */
2078		dzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2079		if (!dzone) {
2080			if (dmz_dev_is_dying(zmd)) {
2081				dzone = ERR_PTR(-EIO);
2082				goto out;
2083			}
2084			dmz_wait_for_free_zones(zmd);
2085			goto again;
2086		}
2087
2088		dmz_map_zone(zmd, dzone, chunk);
2089
2090	} else {
2091		/* The chunk is already mapped: get the mapping zone */
2092		dzone = dmz_get(zmd, dzone_id);
2093		if (!dzone) {
2094			dzone = ERR_PTR(-EIO);
2095			goto out;
2096		}
2097		if (dzone->chunk != chunk) {
2098			dzone = ERR_PTR(-EIO);
2099			goto out;
2100		}
2101
2102		/* Repair write pointer if the sequential dzone has error */
2103		if (dmz_seq_write_err(dzone)) {
2104			ret = dmz_handle_seq_write_err(zmd, dzone);
2105			if (ret) {
2106				dzone = ERR_PTR(-EIO);
2107				goto out;
2108			}
2109			clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
2110		}
2111	}
2112
2113	/*
2114	 * If the zone is being reclaimed, the chunk mapping may change
2115	 * to a different zone. So wait for reclaim and retry. Otherwise,
2116	 * activate the zone (this will prevent reclaim from touching it).
2117	 */
2118	if (dmz_in_reclaim(dzone)) {
2119		dmz_wait_for_reclaim(zmd, dzone);
2120		goto again;
2121	}
2122	dmz_activate_zone(dzone);
2123	dmz_lru_zone(zmd, dzone);
2124out:
2125	dmz_unlock_map(zmd);
2126
2127	return dzone;
2128}
2129
2130/*
2131 * Write and discard change the block validity of data zones and their buffer
2132 * zones. Check here that valid blocks are still present. If all blocks are
2133 * invalid, the zones can be unmapped on the fly without waiting for reclaim
2134 * to do it.
2135 */
2136void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
2137{
2138	struct dm_zone *bzone;
2139
2140	dmz_lock_map(zmd);
2141
2142	bzone = dzone->bzone;
2143	if (bzone) {
2144		if (dmz_weight(bzone))
2145			dmz_lru_zone(zmd, bzone);
2146		else {
2147			/* Empty buffer zone: reclaim it */
2148			dmz_unmap_zone(zmd, bzone);
2149			dmz_free_zone(zmd, bzone);
2150			bzone = NULL;
2151		}
2152	}
2153
2154	/* Deactivate the data zone */
2155	dmz_deactivate_zone(dzone);
2156	if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
2157		dmz_lru_zone(zmd, dzone);
2158	else {
2159		/* Unbuffered inactive empty data zone: reclaim it */
2160		dmz_unmap_zone(zmd, dzone);
2161		dmz_free_zone(zmd, dzone);
2162	}
2163
2164	dmz_unlock_map(zmd);
2165}
2166
2167/*
2168 * Allocate and map a random zone to buffer a chunk
2169 * already mapped to a sequential zone.
2170 */
2171struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
2172				     struct dm_zone *dzone)
2173{
2174	struct dm_zone *bzone;
2175	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2176
2177	dmz_lock_map(zmd);
2178again:
2179	bzone = dzone->bzone;
2180	if (bzone)
2181		goto out;
2182
2183	/* Allocate a random zone */
2184	bzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2185	if (!bzone) {
2186		if (dmz_dev_is_dying(zmd)) {
2187			bzone = ERR_PTR(-EIO);
2188			goto out;
2189		}
2190		dmz_wait_for_free_zones(zmd);
2191		goto again;
2192	}
2193
2194	/* Update the chunk mapping */
2195	dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id);
2196
2197	set_bit(DMZ_BUF, &bzone->flags);
2198	bzone->chunk = dzone->chunk;
2199	bzone->bzone = dzone;
2200	dzone->bzone = bzone;
2201	if (dmz_is_cache(bzone))
2202		list_add_tail(&bzone->link, &zmd->map_cache_list);
2203	else
2204		list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
2205out:
2206	dmz_unlock_map(zmd);
2207
2208	return bzone;
2209}
2210
2211/*
2212 * Get an unmapped (free) zone.
2213 * This must be called with the mapping lock held.
2214 */
2215struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned int dev_idx,
2216			       unsigned long flags)
2217{
2218	struct list_head *list;
2219	struct dm_zone *zone;
2220	int i;
2221
2222	/* Schedule reclaim to ensure free zones are available */
2223	if (!(flags & DMZ_ALLOC_RECLAIM)) {
2224		for (i = 0; i < zmd->nr_devs; i++)
2225			dmz_schedule_reclaim(zmd->dev[i].reclaim);
2226	}
2227
2228	i = 0;
2229again:
2230	if (flags & DMZ_ALLOC_CACHE)
2231		list = &zmd->unmap_cache_list;
2232	else if (flags & DMZ_ALLOC_RND)
2233		list = &zmd->dev[dev_idx].unmap_rnd_list;
2234	else
2235		list = &zmd->dev[dev_idx].unmap_seq_list;
2236
2237	if (list_empty(list)) {
2238		/*
2239		 * No free zone: return NULL if this is for not reclaim.
2240		 */
2241		if (!(flags & DMZ_ALLOC_RECLAIM))
2242			return NULL;
2243		/*
2244		 * Try to allocate from other devices
2245		 */
2246		if (i < zmd->nr_devs) {
2247			dev_idx = (dev_idx + 1) % zmd->nr_devs;
2248			i++;
2249			goto again;
2250		}
2251
2252		/*
2253		 * Fallback to the reserved sequential zones
2254		 */
2255		zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list,
2256						struct dm_zone, link);
2257		if (zone) {
2258			list_del_init(&zone->link);
2259			atomic_dec(&zmd->nr_reserved_seq_zones);
2260		}
2261		return zone;
2262	}
2263
2264	zone = list_first_entry(list, struct dm_zone, link);
2265	list_del_init(&zone->link);
2266
2267	if (dmz_is_cache(zone))
2268		atomic_dec(&zmd->unmap_nr_cache);
2269	else if (dmz_is_rnd(zone))
2270		atomic_dec(&zone->dev->unmap_nr_rnd);
2271	else
2272		atomic_dec(&zone->dev->unmap_nr_seq);
2273
2274	if (dmz_is_offline(zone)) {
2275		dmz_zmd_warn(zmd, "Zone %u is offline", zone->id);
2276		zone = NULL;
2277		goto again;
2278	}
2279	if (dmz_is_meta(zone)) {
2280		dmz_zmd_warn(zmd, "Zone %u has metadata", zone->id);
2281		zone = NULL;
2282		goto again;
2283	}
2284	return zone;
2285}
2286
2287/*
2288 * Free a zone.
2289 * This must be called with the mapping lock held.
2290 */
2291void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2292{
2293	/* If this is a sequential zone, reset it */
2294	if (dmz_is_seq(zone))
2295		dmz_reset_zone(zmd, zone);
2296
2297	/* Return the zone to its type unmap list */
2298	if (dmz_is_cache(zone)) {
2299		list_add_tail(&zone->link, &zmd->unmap_cache_list);
2300		atomic_inc(&zmd->unmap_nr_cache);
2301	} else if (dmz_is_rnd(zone)) {
2302		list_add_tail(&zone->link, &zone->dev->unmap_rnd_list);
2303		atomic_inc(&zone->dev->unmap_nr_rnd);
2304	} else if (dmz_is_reserved(zone)) {
2305		list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
2306		atomic_inc(&zmd->nr_reserved_seq_zones);
2307	} else {
2308		list_add_tail(&zone->link, &zone->dev->unmap_seq_list);
2309		atomic_inc(&zone->dev->unmap_nr_seq);
2310	}
2311
2312	wake_up_all(&zmd->free_wq);
2313}
2314
2315/*
2316 * Map a chunk to a zone.
2317 * This must be called with the mapping lock held.
2318 */
2319void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
2320		  unsigned int chunk)
2321{
2322	/* Set the chunk mapping */
2323	dmz_set_chunk_mapping(zmd, chunk, dzone->id,
2324			      DMZ_MAP_UNMAPPED);
2325	dzone->chunk = chunk;
2326	if (dmz_is_cache(dzone))
2327		list_add_tail(&dzone->link, &zmd->map_cache_list);
2328	else if (dmz_is_rnd(dzone))
2329		list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
2330	else
2331		list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
2332}
2333
2334/*
2335 * Unmap a zone.
2336 * This must be called with the mapping lock held.
2337 */
2338void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2339{
2340	unsigned int chunk = zone->chunk;
2341	unsigned int dzone_id;
2342
2343	if (chunk == DMZ_MAP_UNMAPPED) {
2344		/* Already unmapped */
2345		return;
2346	}
2347
2348	if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
2349		/*
2350		 * Unmapping the chunk buffer zone: clear only
2351		 * the chunk buffer mapping
2352		 */
2353		dzone_id = zone->bzone->id;
2354		zone->bzone->bzone = NULL;
2355		zone->bzone = NULL;
2356
2357	} else {
2358		/*
2359		 * Unmapping the chunk data zone: the zone must
2360		 * not be buffered.
2361		 */
2362		if (WARN_ON(zone->bzone)) {
2363			zone->bzone->bzone = NULL;
2364			zone->bzone = NULL;
2365		}
2366		dzone_id = DMZ_MAP_UNMAPPED;
2367	}
2368
2369	dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
2370
2371	zone->chunk = DMZ_MAP_UNMAPPED;
2372	list_del_init(&zone->link);
2373}
2374
2375/*
2376 * Set @nr_bits bits in @bitmap starting from @bit.
2377 * Return the number of bits changed from 0 to 1.
2378 */
2379static unsigned int dmz_set_bits(unsigned long *bitmap,
2380				 unsigned int bit, unsigned int nr_bits)
2381{
2382	unsigned long *addr;
2383	unsigned int end = bit + nr_bits;
2384	unsigned int n = 0;
2385
2386	while (bit < end) {
2387		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2388		    ((end - bit) >= BITS_PER_LONG)) {
2389			/* Try to set the whole word at once */
2390			addr = bitmap + BIT_WORD(bit);
2391			if (*addr == 0) {
2392				*addr = ULONG_MAX;
2393				n += BITS_PER_LONG;
2394				bit += BITS_PER_LONG;
2395				continue;
2396			}
2397		}
2398
2399		if (!test_and_set_bit(bit, bitmap))
2400			n++;
2401		bit++;
2402	}
2403
2404	return n;
2405}
2406
2407/*
2408 * Get the bitmap block storing the bit for chunk_block in zone.
2409 */
2410static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
2411					 struct dm_zone *zone,
2412					 sector_t chunk_block)
2413{
2414	sector_t bitmap_block = 1 + zmd->nr_map_blocks +
2415		(sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) +
2416		(chunk_block >> DMZ_BLOCK_SHIFT_BITS);
2417
2418	return dmz_get_mblock(zmd, bitmap_block);
2419}
2420
2421/*
2422 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
2423 */
2424int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2425			  struct dm_zone *to_zone)
2426{
2427	struct dmz_mblock *from_mblk, *to_mblk;
2428	sector_t chunk_block = 0;
2429
2430	/* Get the zones bitmap blocks */
2431	while (chunk_block < zmd->zone_nr_blocks) {
2432		from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
2433		if (IS_ERR(from_mblk))
2434			return PTR_ERR(from_mblk);
2435		to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
2436		if (IS_ERR(to_mblk)) {
2437			dmz_release_mblock(zmd, from_mblk);
2438			return PTR_ERR(to_mblk);
2439		}
2440
2441		memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
2442		dmz_dirty_mblock(zmd, to_mblk);
2443
2444		dmz_release_mblock(zmd, to_mblk);
2445		dmz_release_mblock(zmd, from_mblk);
2446
2447		chunk_block += zmd->zone_bits_per_mblk;
2448	}
2449
2450	to_zone->weight = from_zone->weight;
2451
2452	return 0;
2453}
2454
2455/*
2456 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
2457 * starting from chunk_block.
2458 */
2459int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2460			   struct dm_zone *to_zone, sector_t chunk_block)
2461{
2462	unsigned int nr_blocks;
2463	int ret;
2464
2465	/* Get the zones bitmap blocks */
2466	while (chunk_block < zmd->zone_nr_blocks) {
2467		/* Get a valid region from the source zone */
2468		ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2469		if (ret <= 0)
2470			return ret;
2471
2472		nr_blocks = ret;
2473		ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2474		if (ret)
2475			return ret;
2476
2477		chunk_block += nr_blocks;
2478	}
2479
2480	return 0;
2481}
2482
2483/*
2484 * Validate all the blocks in the range [block..block+nr_blocks-1].
2485 */
2486int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2487			sector_t chunk_block, unsigned int nr_blocks)
2488{
2489	unsigned int count, bit, nr_bits;
2490	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
2491	struct dmz_mblock *mblk;
2492	unsigned int n = 0;
2493
2494	dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks",
2495		      zone->id, (unsigned long long)chunk_block,
2496		      nr_blocks);
2497
2498	WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2499
2500	while (nr_blocks) {
2501		/* Get bitmap block */
2502		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2503		if (IS_ERR(mblk))
2504			return PTR_ERR(mblk);
2505
2506		/* Set bits */
2507		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2508		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2509
2510		count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2511		if (count) {
2512			dmz_dirty_mblock(zmd, mblk);
2513			n += count;
2514		}
2515		dmz_release_mblock(zmd, mblk);
2516
2517		nr_blocks -= nr_bits;
2518		chunk_block += nr_bits;
2519	}
2520
2521	if (likely(zone->weight + n <= zone_nr_blocks))
2522		zone->weight += n;
2523	else {
2524		dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u",
2525			     zone->id, zone->weight,
2526			     zone_nr_blocks - n);
2527		zone->weight = zone_nr_blocks;
2528	}
2529
2530	return 0;
2531}
2532
2533/*
2534 * Clear nr_bits bits in bitmap starting from bit.
2535 * Return the number of bits cleared.
2536 */
2537static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2538{
2539	unsigned long *addr;
2540	int end = bit + nr_bits;
2541	int n = 0;
2542
2543	while (bit < end) {
2544		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2545		    ((end - bit) >= BITS_PER_LONG)) {
2546			/* Try to clear whole word at once */
2547			addr = bitmap + BIT_WORD(bit);
2548			if (*addr == ULONG_MAX) {
2549				*addr = 0;
2550				n += BITS_PER_LONG;
2551				bit += BITS_PER_LONG;
2552				continue;
2553			}
2554		}
2555
2556		if (test_and_clear_bit(bit, bitmap))
2557			n++;
2558		bit++;
2559	}
2560
2561	return n;
2562}
2563
2564/*
2565 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2566 */
2567int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2568			  sector_t chunk_block, unsigned int nr_blocks)
2569{
2570	unsigned int count, bit, nr_bits;
2571	struct dmz_mblock *mblk;
2572	unsigned int n = 0;
2573
2574	dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks",
2575		      zone->id, (u64)chunk_block, nr_blocks);
2576
2577	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2578
2579	while (nr_blocks) {
2580		/* Get bitmap block */
2581		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2582		if (IS_ERR(mblk))
2583			return PTR_ERR(mblk);
2584
2585		/* Clear bits */
2586		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2587		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2588
2589		count = dmz_clear_bits((unsigned long *)mblk->data,
2590				       bit, nr_bits);
2591		if (count) {
2592			dmz_dirty_mblock(zmd, mblk);
2593			n += count;
2594		}
2595		dmz_release_mblock(zmd, mblk);
2596
2597		nr_blocks -= nr_bits;
2598		chunk_block += nr_bits;
2599	}
2600
2601	if (zone->weight >= n)
2602		zone->weight -= n;
2603	else {
2604		dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u",
2605			     zone->id, zone->weight, n);
2606		zone->weight = 0;
2607	}
2608
2609	return 0;
2610}
2611
2612/*
2613 * Get a block bit value.
2614 */
2615static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2616			  sector_t chunk_block)
2617{
2618	struct dmz_mblock *mblk;
2619	int ret;
2620
2621	WARN_ON(chunk_block >= zmd->zone_nr_blocks);
2622
2623	/* Get bitmap block */
2624	mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2625	if (IS_ERR(mblk))
2626		return PTR_ERR(mblk);
2627
2628	/* Get offset */
2629	ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2630		       (unsigned long *) mblk->data) != 0;
2631
2632	dmz_release_mblock(zmd, mblk);
2633
2634	return ret;
2635}
2636
2637/*
2638 * Return the number of blocks from chunk_block to the first block with a bit
2639 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2640 */
2641static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2642				 sector_t chunk_block, unsigned int nr_blocks,
2643				 int set)
2644{
2645	struct dmz_mblock *mblk;
2646	unsigned int bit, set_bit, nr_bits;
2647	unsigned int zone_bits = zmd->zone_bits_per_mblk;
2648	unsigned long *bitmap;
2649	int n = 0;
2650
2651	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2652
2653	while (nr_blocks) {
2654		/* Get bitmap block */
2655		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2656		if (IS_ERR(mblk))
2657			return PTR_ERR(mblk);
2658
2659		/* Get offset */
2660		bitmap = (unsigned long *) mblk->data;
2661		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2662		nr_bits = min(nr_blocks, zone_bits - bit);
2663		if (set)
2664			set_bit = find_next_bit(bitmap, zone_bits, bit);
2665		else
2666			set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2667		dmz_release_mblock(zmd, mblk);
2668
2669		n += set_bit - bit;
2670		if (set_bit < zone_bits)
2671			break;
2672
2673		nr_blocks -= nr_bits;
2674		chunk_block += nr_bits;
2675	}
2676
2677	return n;
2678}
2679
2680/*
2681 * Test if chunk_block is valid. If it is, the number of consecutive
2682 * valid blocks from chunk_block will be returned.
2683 */
2684int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2685		    sector_t chunk_block)
2686{
2687	int valid;
2688
2689	valid = dmz_test_block(zmd, zone, chunk_block);
2690	if (valid <= 0)
2691		return valid;
2692
2693	/* The block is valid: get the number of valid blocks from block */
2694	return dmz_to_next_set_block(zmd, zone, chunk_block,
2695				     zmd->zone_nr_blocks - chunk_block, 0);
2696}
2697
2698/*
2699 * Find the first valid block from @chunk_block in @zone.
2700 * If such a block is found, its number is returned using
2701 * @chunk_block and the total number of valid blocks from @chunk_block
2702 * is returned.
2703 */
2704int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2705			  sector_t *chunk_block)
2706{
2707	sector_t start_block = *chunk_block;
2708	int ret;
2709
2710	ret = dmz_to_next_set_block(zmd, zone, start_block,
2711				    zmd->zone_nr_blocks - start_block, 1);
2712	if (ret < 0)
2713		return ret;
2714
2715	start_block += ret;
2716	*chunk_block = start_block;
2717
2718	return dmz_to_next_set_block(zmd, zone, start_block,
2719				     zmd->zone_nr_blocks - start_block, 0);
2720}
2721
2722/*
2723 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2724 */
2725static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2726{
2727	unsigned long *addr;
2728	int end = bit + nr_bits;
2729	int n = 0;
2730
2731	while (bit < end) {
2732		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2733		    ((end - bit) >= BITS_PER_LONG)) {
2734			addr = (unsigned long *)bitmap + BIT_WORD(bit);
2735			if (*addr == ULONG_MAX) {
2736				n += BITS_PER_LONG;
2737				bit += BITS_PER_LONG;
2738				continue;
2739			}
2740		}
2741
2742		if (test_bit(bit, bitmap))
2743			n++;
2744		bit++;
2745	}
2746
2747	return n;
2748}
2749
2750/*
2751 * Get a zone weight.
2752 */
2753static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2754{
2755	struct dmz_mblock *mblk;
2756	sector_t chunk_block = 0;
2757	unsigned int bit, nr_bits;
2758	unsigned int nr_blocks = zmd->zone_nr_blocks;
2759	void *bitmap;
2760	int n = 0;
2761
2762	while (nr_blocks) {
2763		/* Get bitmap block */
2764		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2765		if (IS_ERR(mblk)) {
2766			n = 0;
2767			break;
2768		}
2769
2770		/* Count bits in this block */
2771		bitmap = mblk->data;
2772		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2773		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2774		n += dmz_count_bits(bitmap, bit, nr_bits);
2775
2776		dmz_release_mblock(zmd, mblk);
2777
2778		nr_blocks -= nr_bits;
2779		chunk_block += nr_bits;
2780	}
2781
2782	zone->weight = n;
2783}
2784
2785/*
2786 * Cleanup the zoned metadata resources.
2787 */
2788static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2789{
2790	struct rb_root *root;
2791	struct dmz_mblock *mblk, *next;
2792	int i;
2793
2794	/* Release zone mapping resources */
2795	if (zmd->map_mblk) {
2796		for (i = 0; i < zmd->nr_map_blocks; i++)
2797			dmz_release_mblock(zmd, zmd->map_mblk[i]);
2798		kfree(zmd->map_mblk);
2799		zmd->map_mblk = NULL;
2800	}
2801
2802	/* Release super blocks */
2803	for (i = 0; i < 2; i++) {
2804		if (zmd->sb[i].mblk) {
2805			dmz_free_mblock(zmd, zmd->sb[i].mblk);
2806			zmd->sb[i].mblk = NULL;
2807		}
2808	}
2809
2810	/* Free cached blocks */
2811	while (!list_empty(&zmd->mblk_dirty_list)) {
2812		mblk = list_first_entry(&zmd->mblk_dirty_list,
2813					struct dmz_mblock, link);
2814		dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)",
2815			     (u64)mblk->no, mblk->ref);
2816		list_del_init(&mblk->link);
2817		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2818		dmz_free_mblock(zmd, mblk);
2819	}
2820
2821	while (!list_empty(&zmd->mblk_lru_list)) {
2822		mblk = list_first_entry(&zmd->mblk_lru_list,
2823					struct dmz_mblock, link);
2824		list_del_init(&mblk->link);
2825		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2826		dmz_free_mblock(zmd, mblk);
2827	}
2828
2829	/* Sanity checks: the mblock rbtree should now be empty */
2830	root = &zmd->mblk_rbtree;
2831	rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2832		dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree",
2833			     (u64)mblk->no, mblk->ref);
2834		mblk->ref = 0;
2835		dmz_free_mblock(zmd, mblk);
2836	}
2837
2838	/* Free the zone descriptors */
2839	dmz_drop_zones(zmd);
2840
2841	mutex_destroy(&zmd->mblk_flush_lock);
2842	mutex_destroy(&zmd->map_lock);
2843}
2844
2845static void dmz_print_dev(struct dmz_metadata *zmd, int num)
2846{
2847	struct dmz_dev *dev = &zmd->dev[num];
2848
2849	if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE)
2850		dmz_dev_info(dev, "Regular block device");
2851	else
2852		dmz_dev_info(dev, "Host-%s zoned block device",
2853			     bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2854			     "aware" : "managed");
2855	if (zmd->sb_version > 1) {
2856		sector_t sector_offset =
2857			dev->zone_offset << zmd->zone_nr_sectors_shift;
2858
2859		dmz_dev_info(dev, "  %llu 512-byte logical sectors (offset %llu)",
2860			     (u64)dev->capacity, (u64)sector_offset);
2861		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors (offset %llu)",
2862			     dev->nr_zones, (u64)zmd->zone_nr_sectors,
2863			     (u64)dev->zone_offset);
2864	} else {
2865		dmz_dev_info(dev, "  %llu 512-byte logical sectors",
2866			     (u64)dev->capacity);
2867		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors",
2868			     dev->nr_zones, (u64)zmd->zone_nr_sectors);
2869	}
2870}
2871
2872/*
2873 * Initialize the zoned metadata.
2874 */
2875int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev,
2876		     struct dmz_metadata **metadata,
2877		     const char *devname)
2878{
2879	struct dmz_metadata *zmd;
2880	unsigned int i;
2881	struct dm_zone *zone;
2882	int ret;
2883
2884	zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2885	if (!zmd)
2886		return -ENOMEM;
2887
2888	strcpy(zmd->devname, devname);
2889	zmd->dev = dev;
2890	zmd->nr_devs = num_dev;
2891	zmd->mblk_rbtree = RB_ROOT;
2892	init_rwsem(&zmd->mblk_sem);
2893	mutex_init(&zmd->mblk_flush_lock);
2894	spin_lock_init(&zmd->mblk_lock);
2895	INIT_LIST_HEAD(&zmd->mblk_lru_list);
2896	INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2897
2898	mutex_init(&zmd->map_lock);
2899
2900	atomic_set(&zmd->unmap_nr_cache, 0);
2901	INIT_LIST_HEAD(&zmd->unmap_cache_list);
2902	INIT_LIST_HEAD(&zmd->map_cache_list);
2903
2904	atomic_set(&zmd->nr_reserved_seq_zones, 0);
2905	INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2906
2907	init_waitqueue_head(&zmd->free_wq);
2908
2909	/* Initialize zone descriptors */
2910	ret = dmz_init_zones(zmd);
2911	if (ret)
2912		goto err;
2913
2914	/* Get super block */
2915	ret = dmz_load_sb(zmd);
2916	if (ret)
2917		goto err;
2918
2919	/* Set metadata zones starting from sb_zone */
2920	for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2921		zone = dmz_get(zmd, zmd->sb[0].zone->id + i);
2922		if (!zone) {
2923			dmz_zmd_err(zmd,
2924				    "metadata zone %u not present", i);
2925			ret = -ENXIO;
2926			goto err;
2927		}
2928		if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) {
2929			dmz_zmd_err(zmd,
2930				    "metadata zone %d is not random", i);
2931			ret = -ENXIO;
2932			goto err;
2933		}
2934		set_bit(DMZ_META, &zone->flags);
2935	}
2936	/* Load mapping table */
2937	ret = dmz_load_mapping(zmd);
2938	if (ret)
2939		goto err;
2940
2941	/*
2942	 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2943	 * blocks and enough blocks to be able to cache the bitmap blocks of
2944	 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2945	 * the cache to add 512 more metadata blocks.
2946	 */
2947	zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2948	zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2949	zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2950	zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2951	zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2952
2953	/* Metadata cache shrinker */
2954	ret = register_shrinker(&zmd->mblk_shrinker);
2955	if (ret) {
2956		dmz_zmd_err(zmd, "Register metadata cache shrinker failed");
2957		goto err;
2958	}
2959
2960	dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version);
2961	for (i = 0; i < zmd->nr_devs; i++)
2962		dmz_print_dev(zmd, i);
2963
2964	dmz_zmd_info(zmd, "  %u zones of %llu 512-byte logical sectors",
2965		     zmd->nr_zones, (u64)zmd->zone_nr_sectors);
2966	dmz_zmd_debug(zmd, "  %u metadata zones",
2967		      zmd->nr_meta_zones * 2);
2968	dmz_zmd_debug(zmd, "  %u data zones for %u chunks",
2969		      zmd->nr_data_zones, zmd->nr_chunks);
2970	dmz_zmd_debug(zmd, "    %u cache zones (%u unmapped)",
2971		      zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache));
2972	for (i = 0; i < zmd->nr_devs; i++) {
2973		dmz_zmd_debug(zmd, "    %u random zones (%u unmapped)",
2974			      dmz_nr_rnd_zones(zmd, i),
2975			      dmz_nr_unmap_rnd_zones(zmd, i));
2976		dmz_zmd_debug(zmd, "    %u sequential zones (%u unmapped)",
2977			      dmz_nr_seq_zones(zmd, i),
2978			      dmz_nr_unmap_seq_zones(zmd, i));
2979	}
2980	dmz_zmd_debug(zmd, "  %u reserved sequential data zones",
2981		      zmd->nr_reserved_seq);
2982	dmz_zmd_debug(zmd, "Format:");
2983	dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)",
2984		      zmd->nr_meta_blocks, zmd->max_nr_mblks);
2985	dmz_zmd_debug(zmd, "  %u data zone mapping blocks",
2986		      zmd->nr_map_blocks);
2987	dmz_zmd_debug(zmd, "  %u bitmap blocks",
2988		      zmd->nr_bitmap_blocks);
2989
2990	*metadata = zmd;
2991
2992	return 0;
2993err:
2994	dmz_cleanup_metadata(zmd);
2995	kfree(zmd);
2996	*metadata = NULL;
2997
2998	return ret;
2999}
3000
3001/*
3002 * Cleanup the zoned metadata resources.
3003 */
3004void dmz_dtr_metadata(struct dmz_metadata *zmd)
3005{
3006	unregister_shrinker(&zmd->mblk_shrinker);
3007	dmz_cleanup_metadata(zmd);
3008	kfree(zmd);
3009}
3010
3011/*
3012 * Check zone information on resume.
3013 */
3014int dmz_resume_metadata(struct dmz_metadata *zmd)
3015{
3016	struct dm_zone *zone;
3017	sector_t wp_block;
3018	unsigned int i;
3019	int ret;
3020
3021	/* Check zones */
3022	for (i = 0; i < zmd->nr_zones; i++) {
3023		zone = dmz_get(zmd, i);
3024		if (!zone) {
3025			dmz_zmd_err(zmd, "Unable to get zone %u", i);
3026			return -EIO;
3027		}
3028		wp_block = zone->wp_block;
3029
3030		ret = dmz_update_zone(zmd, zone);
3031		if (ret) {
3032			dmz_zmd_err(zmd, "Broken zone %u", i);
3033			return ret;
3034		}
3035
3036		if (dmz_is_offline(zone)) {
3037			dmz_zmd_warn(zmd, "Zone %u is offline", i);
3038			continue;
3039		}
3040
3041		/* Check write pointer */
3042		if (!dmz_is_seq(zone))
3043			zone->wp_block = 0;
3044		else if (zone->wp_block != wp_block) {
3045			dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)",
3046				    i, (u64)zone->wp_block, (u64)wp_block);
3047			zone->wp_block = wp_block;
3048			dmz_invalidate_blocks(zmd, zone, zone->wp_block,
3049					      zmd->zone_nr_blocks - zone->wp_block);
3050		}
3051	}
3052
3053	return 0;
3054}