Linux Audio

Check our new training course

Loading...
v6.2
   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(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 */
 589static 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 */
 615static 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 */
 626static 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 */
 642static 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 */
 669static 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 */
 702static 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 */
 713static 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 */
 740static 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 */
 767static 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 */
 816static 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 */
 859static 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 */
 884int 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++;
 959out:
 960	dmz_unlock_flush(zmd);
 961	up_write(&zmd->mblk_sem);
 962
 963	return ret;
 964
 965err:
 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 */
 979static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_sb *dsb,
 980			bool tertiary)
 981{
 982	struct dmz_super *sb = dsb->sb;
 983	struct dmz_dev *dev = dsb->dev;
 984	unsigned int nr_meta_zones, nr_data_zones;
 
 985	u32 crc, stored_crc;
 986	u64 gen, sb_block;
 987
 988	if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
 989		dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
 990			    DMZ_MAGIC, le32_to_cpu(sb->magic));
 991		return -ENXIO;
 992	}
 993
 994	zmd->sb_version = le32_to_cpu(sb->version);
 995	if (zmd->sb_version > DMZ_META_VER) {
 996		dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
 997			    DMZ_META_VER, zmd->sb_version);
 998		return -EINVAL;
 999	}
1000	if (zmd->sb_version < 2 && tertiary) {
1001		dmz_dev_err(dev, "Tertiary superblocks are not supported");
1002		return -EINVAL;
1003	}
1004
1005	gen = le64_to_cpu(sb->gen);
1006	stored_crc = le32_to_cpu(sb->crc);
1007	sb->crc = 0;
1008	crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
1009	if (crc != stored_crc) {
1010		dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
1011			    crc, stored_crc);
1012		return -ENXIO;
1013	}
1014
1015	sb_block = le64_to_cpu(sb->sb_block);
1016	if (sb_block != (u64)dsb->zone->id << zmd->zone_nr_blocks_shift ) {
1017		dmz_dev_err(dev, "Invalid superblock position "
1018			    "(is %llu expected %llu)",
1019			    sb_block,
1020			    (u64)dsb->zone->id << zmd->zone_nr_blocks_shift);
1021		return -EINVAL;
1022	}
1023	if (zmd->sb_version > 1) {
1024		uuid_t sb_uuid;
1025
1026		import_uuid(&sb_uuid, sb->dmz_uuid);
1027		if (uuid_is_null(&sb_uuid)) {
1028			dmz_dev_err(dev, "NULL DM-Zoned uuid");
1029			return -ENXIO;
1030		} else if (uuid_is_null(&zmd->uuid)) {
1031			uuid_copy(&zmd->uuid, &sb_uuid);
1032		} else if (!uuid_equal(&zmd->uuid, &sb_uuid)) {
1033			dmz_dev_err(dev, "mismatching DM-Zoned uuid, "
1034				    "is %pUl expected %pUl",
1035				    &sb_uuid, &zmd->uuid);
1036			return -ENXIO;
1037		}
1038		if (!strlen(zmd->label))
1039			memcpy(zmd->label, sb->dmz_label, BDEVNAME_SIZE);
1040		else if (memcmp(zmd->label, sb->dmz_label, BDEVNAME_SIZE)) {
1041			dmz_dev_err(dev, "mismatching DM-Zoned label, "
1042				    "is %s expected %s",
1043				    sb->dmz_label, zmd->label);
1044			return -ENXIO;
1045		}
1046		import_uuid(&dev->uuid, sb->dev_uuid);
1047		if (uuid_is_null(&dev->uuid)) {
1048			dmz_dev_err(dev, "NULL device uuid");
1049			return -ENXIO;
1050		}
1051
1052		if (tertiary) {
1053			/*
1054			 * Generation number should be 0, but it doesn't
1055			 * really matter if it isn't.
1056			 */
1057			if (gen != 0)
1058				dmz_dev_warn(dev, "Invalid generation %llu",
1059					    gen);
1060			return 0;
1061		}
1062	}
1063
1064	nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + zmd->zone_nr_blocks - 1)
1065		>> zmd->zone_nr_blocks_shift;
1066	if (!nr_meta_zones ||
1067	    (zmd->nr_devs <= 1 && nr_meta_zones >= zmd->nr_rnd_zones) ||
1068	    (zmd->nr_devs > 1 && nr_meta_zones >= zmd->nr_cache_zones)) {
1069		dmz_dev_err(dev, "Invalid number of metadata blocks");
1070		return -ENXIO;
1071	}
1072
1073	if (!le32_to_cpu(sb->nr_reserved_seq) ||
1074	    le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
1075		dmz_dev_err(dev, "Invalid number of reserved sequential zones");
1076		return -ENXIO;
1077	}
1078
1079	nr_data_zones = zmd->nr_useable_zones -
1080		(nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
1081	if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
1082		dmz_dev_err(dev, "Invalid number of chunks %u / %u",
1083			    le32_to_cpu(sb->nr_chunks), nr_data_zones);
1084		return -ENXIO;
1085	}
1086
1087	/* OK */
1088	zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
1089	zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
1090	zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
1091	zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
1092	zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
1093	zmd->nr_meta_zones = nr_meta_zones;
1094	zmd->nr_data_zones = nr_data_zones;
1095
1096	return 0;
1097}
1098
1099/*
1100 * Read the first or second super block from disk.
1101 */
1102static int dmz_read_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1103{
1104	dmz_zmd_debug(zmd, "read superblock set %d dev %pg block %llu",
1105		      set, sb->dev->bdev, sb->block);
1106
1107	return dmz_rdwr_block(sb->dev, REQ_OP_READ,
1108			      sb->block, sb->mblk->page);
1109}
1110
1111/*
1112 * Determine the position of the secondary super blocks on disk.
1113 * This is used only if a corruption of the primary super block
1114 * is detected.
1115 */
1116static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
1117{
1118	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
1119	struct dmz_mblock *mblk;
1120	unsigned int zone_id = zmd->sb[0].zone->id;
1121	int i;
1122
1123	/* Allocate a block */
1124	mblk = dmz_alloc_mblock(zmd, 0);
1125	if (!mblk)
1126		return -ENOMEM;
1127
1128	zmd->sb[1].mblk = mblk;
1129	zmd->sb[1].sb = mblk->data;
1130
1131	/* Bad first super block: search for the second one */
1132	zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
1133	zmd->sb[1].zone = dmz_get(zmd, zone_id + 1);
1134	zmd->sb[1].dev = zmd->sb[0].dev;
1135	for (i = 1; i < zmd->nr_rnd_zones; i++) {
1136		if (dmz_read_sb(zmd, &zmd->sb[1], 1) != 0)
1137			break;
1138		if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
1139			return 0;
1140		zmd->sb[1].block += zone_nr_blocks;
1141		zmd->sb[1].zone = dmz_get(zmd, zone_id + i);
1142	}
1143
1144	dmz_free_mblock(zmd, mblk);
1145	zmd->sb[1].mblk = NULL;
1146	zmd->sb[1].zone = NULL;
1147	zmd->sb[1].dev = NULL;
1148
1149	return -EIO;
1150}
1151
1152/*
1153 * Read a super block from disk.
1154 */
1155static int dmz_get_sb(struct dmz_metadata *zmd, struct dmz_sb *sb, int set)
1156{
1157	struct dmz_mblock *mblk;
1158	int ret;
1159
1160	/* Allocate a block */
1161	mblk = dmz_alloc_mblock(zmd, 0);
1162	if (!mblk)
1163		return -ENOMEM;
1164
1165	sb->mblk = mblk;
1166	sb->sb = mblk->data;
1167
1168	/* Read super block */
1169	ret = dmz_read_sb(zmd, sb, set);
1170	if (ret) {
1171		dmz_free_mblock(zmd, mblk);
1172		sb->mblk = NULL;
1173		return ret;
1174	}
1175
1176	return 0;
1177}
1178
1179/*
1180 * Recover a metadata set.
1181 */
1182static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
1183{
1184	unsigned int src_set = dst_set ^ 0x1;
1185	struct page *page;
1186	int i, ret;
1187
1188	dmz_dev_warn(zmd->sb[dst_set].dev,
1189		     "Metadata set %u invalid: recovering", dst_set);
1190
1191	if (dst_set == 0)
1192		zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1193	else
1194		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
 
 
1195
1196	page = alloc_page(GFP_NOIO);
1197	if (!page)
1198		return -ENOMEM;
1199
1200	/* Copy metadata blocks */
1201	for (i = 1; i < zmd->nr_meta_blocks; i++) {
1202		ret = dmz_rdwr_block(zmd->sb[src_set].dev, REQ_OP_READ,
1203				     zmd->sb[src_set].block + i, page);
1204		if (ret)
1205			goto out;
1206		ret = dmz_rdwr_block(zmd->sb[dst_set].dev, REQ_OP_WRITE,
1207				     zmd->sb[dst_set].block + i, page);
1208		if (ret)
1209			goto out;
1210	}
1211
1212	/* Finalize with the super block */
1213	if (!zmd->sb[dst_set].mblk) {
1214		zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
1215		if (!zmd->sb[dst_set].mblk) {
1216			ret = -ENOMEM;
1217			goto out;
1218		}
1219		zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
1220	}
1221
1222	ret = dmz_write_sb(zmd, dst_set);
1223out:
1224	__free_pages(page, 0);
1225
1226	return ret;
1227}
1228
1229/*
1230 * Get super block from disk.
1231 */
1232static int dmz_load_sb(struct dmz_metadata *zmd)
1233{
1234	bool sb_good[2] = {false, false};
1235	u64 sb_gen[2] = {0, 0};
1236	int ret;
1237
1238	if (!zmd->sb[0].zone) {
1239		dmz_zmd_err(zmd, "Primary super block zone not set");
1240		return -ENXIO;
1241	}
1242
1243	/* Read and check the primary super block */
1244	zmd->sb[0].block = dmz_start_block(zmd, zmd->sb[0].zone);
1245	zmd->sb[0].dev = zmd->sb[0].zone->dev;
1246	ret = dmz_get_sb(zmd, &zmd->sb[0], 0);
1247	if (ret) {
1248		dmz_dev_err(zmd->sb[0].dev, "Read primary super block failed");
1249		return ret;
1250	}
1251
1252	ret = dmz_check_sb(zmd, &zmd->sb[0], false);
1253
1254	/* Read and check secondary super block */
1255	if (ret == 0) {
1256		sb_good[0] = true;
1257		if (!zmd->sb[1].zone) {
1258			unsigned int zone_id =
1259				zmd->sb[0].zone->id + zmd->nr_meta_zones;
1260
1261			zmd->sb[1].zone = dmz_get(zmd, zone_id);
1262		}
1263		zmd->sb[1].block = dmz_start_block(zmd, zmd->sb[1].zone);
1264		zmd->sb[1].dev = zmd->sb[0].dev;
1265		ret = dmz_get_sb(zmd, &zmd->sb[1], 1);
1266	} else
1267		ret = dmz_lookup_secondary_sb(zmd);
1268
1269	if (ret) {
1270		dmz_dev_err(zmd->sb[1].dev, "Read secondary super block failed");
1271		return ret;
1272	}
1273
1274	ret = dmz_check_sb(zmd, &zmd->sb[1], false);
1275	if (ret == 0)
1276		sb_good[1] = true;
1277
1278	/* Use highest generation sb first */
1279	if (!sb_good[0] && !sb_good[1]) {
1280		dmz_zmd_err(zmd, "No valid super block found");
1281		return -EIO;
1282	}
1283
1284	if (sb_good[0])
1285		sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1286	else {
1287		ret = dmz_recover_mblocks(zmd, 0);
1288		if (ret) {
1289			dmz_dev_err(zmd->sb[0].dev,
1290				    "Recovery of superblock 0 failed");
1291			return -EIO;
1292		}
1293	}
1294
1295	if (sb_good[1])
1296		sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1297	else {
1298		ret = dmz_recover_mblocks(zmd, 1);
1299
1300		if (ret) {
1301			dmz_dev_err(zmd->sb[1].dev,
1302				    "Recovery of superblock 1 failed");
1303			return -EIO;
1304		}
1305	}
1306
1307	if (sb_gen[0] >= sb_gen[1]) {
1308		zmd->sb_gen = sb_gen[0];
1309		zmd->mblk_primary = 0;
1310	} else {
1311		zmd->sb_gen = sb_gen[1];
1312		zmd->mblk_primary = 1;
1313	}
1314
1315	dmz_dev_debug(zmd->sb[zmd->mblk_primary].dev,
1316		      "Using super block %u (gen %llu)",
1317		      zmd->mblk_primary, zmd->sb_gen);
1318
1319	if (zmd->sb_version > 1) {
1320		int i;
1321		struct dmz_sb *sb;
1322
1323		sb = kzalloc(sizeof(struct dmz_sb), GFP_KERNEL);
1324		if (!sb)
1325			return -ENOMEM;
1326		for (i = 1; i < zmd->nr_devs; i++) {
1327			sb->block = 0;
1328			sb->zone = dmz_get(zmd, zmd->dev[i].zone_offset);
1329			sb->dev = &zmd->dev[i];
1330			if (!dmz_is_meta(sb->zone)) {
1331				dmz_dev_err(sb->dev,
1332					    "Tertiary super block zone %u not marked as metadata zone",
1333					    sb->zone->id);
1334				ret = -EINVAL;
1335				goto out_kfree;
1336			}
1337			ret = dmz_get_sb(zmd, sb, i + 1);
1338			if (ret) {
1339				dmz_dev_err(sb->dev,
1340					    "Read tertiary super block failed");
1341				dmz_free_mblock(zmd, sb->mblk);
1342				goto out_kfree;
1343			}
1344			ret = dmz_check_sb(zmd, sb, true);
1345			dmz_free_mblock(zmd, sb->mblk);
1346			if (ret == -EINVAL)
1347				goto out_kfree;
1348		}
1349	out_kfree:
1350		kfree(sb);
1351	}
1352	return ret;
1353}
1354
1355/*
1356 * Initialize a zone descriptor.
1357 */
1358static int dmz_init_zone(struct blk_zone *blkz, unsigned int num, void *data)
 
1359{
1360	struct dmz_dev *dev = data;
1361	struct dmz_metadata *zmd = dev->metadata;
1362	int idx = num + dev->zone_offset;
1363	struct dm_zone *zone;
1364
1365	zone = dmz_insert(zmd, idx, dev);
1366	if (IS_ERR(zone))
1367		return PTR_ERR(zone);
1368
1369	if (blkz->len != zmd->zone_nr_sectors) {
1370		if (zmd->sb_version > 1) {
1371			/* Ignore the eventual runt (smaller) zone */
1372			set_bit(DMZ_OFFLINE, &zone->flags);
1373			return 0;
1374		} else if (blkz->start + blkz->len == dev->capacity)
1375			return 0;
1376		return -ENXIO;
1377	}
1378
1379	/*
1380	 * Devices that have zones with a capacity smaller than the zone size
1381	 * (e.g. NVMe zoned namespaces) are not supported.
1382	 */
1383	if (blkz->capacity != blkz->len)
1384		return -ENXIO;
1385
1386	switch (blkz->type) {
1387	case BLK_ZONE_TYPE_CONVENTIONAL:
1388		set_bit(DMZ_RND, &zone->flags);
1389		break;
1390	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1391	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1392		set_bit(DMZ_SEQ, &zone->flags);
1393		break;
1394	default:
1395		return -ENXIO;
1396	}
 
 
 
 
1397
1398	if (dmz_is_rnd(zone))
1399		zone->wp_block = 0;
1400	else
1401		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1402
1403	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1404		set_bit(DMZ_OFFLINE, &zone->flags);
1405	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1406		set_bit(DMZ_READ_ONLY, &zone->flags);
1407	else {
1408		zmd->nr_useable_zones++;
1409		if (dmz_is_rnd(zone)) {
1410			zmd->nr_rnd_zones++;
1411			if (zmd->nr_devs == 1 && !zmd->sb[0].zone) {
1412				/* Primary super block zone */
1413				zmd->sb[0].zone = zone;
1414			}
1415		}
1416		if (zmd->nr_devs > 1 && num == 0) {
1417			/*
1418			 * Tertiary superblock zones are always at the
1419			 * start of the zoned devices, so mark them
1420			 * as metadata zone.
1421			 */
1422			set_bit(DMZ_META, &zone->flags);
1423		}
1424	}
1425	return 0;
1426}
1427
1428static int dmz_emulate_zones(struct dmz_metadata *zmd, struct dmz_dev *dev)
1429{
1430	int idx;
1431	sector_t zone_offset = 0;
1432
1433	for(idx = 0; idx < dev->nr_zones; idx++) {
1434		struct dm_zone *zone;
1435
1436		zone = dmz_insert(zmd, idx, dev);
1437		if (IS_ERR(zone))
1438			return PTR_ERR(zone);
1439		set_bit(DMZ_CACHE, &zone->flags);
1440		zone->wp_block = 0;
1441		zmd->nr_cache_zones++;
1442		zmd->nr_useable_zones++;
1443		if (dev->capacity - zone_offset < zmd->zone_nr_sectors) {
1444			/* Disable runt zone */
1445			set_bit(DMZ_OFFLINE, &zone->flags);
1446			break;
1447		}
1448		zone_offset += zmd->zone_nr_sectors;
1449	}
1450	return 0;
1451}
1452
1453/*
1454 * Free zones descriptors.
1455 */
1456static void dmz_drop_zones(struct dmz_metadata *zmd)
1457{
1458	int idx;
1459
1460	for(idx = 0; idx < zmd->nr_zones; idx++) {
1461		struct dm_zone *zone = xa_load(&zmd->zones, idx);
1462
1463		kfree(zone);
1464		xa_erase(&zmd->zones, idx);
1465	}
1466	xa_destroy(&zmd->zones);
1467}
1468
1469/*
 
 
 
 
 
 
1470 * Allocate and initialize zone descriptors using the zone
1471 * information from disk.
1472 */
1473static int dmz_init_zones(struct dmz_metadata *zmd)
1474{
1475	int i, ret;
1476	struct dmz_dev *zoned_dev = &zmd->dev[0];
 
 
 
 
1477
1478	/* Init */
1479	zmd->zone_nr_sectors = zmd->dev[0].zone_nr_sectors;
1480	zmd->zone_nr_sectors_shift = ilog2(zmd->zone_nr_sectors);
1481	zmd->zone_nr_blocks = dmz_sect2blk(zmd->zone_nr_sectors);
1482	zmd->zone_nr_blocks_shift = ilog2(zmd->zone_nr_blocks);
1483	zmd->zone_bitmap_size = zmd->zone_nr_blocks >> 3;
1484	zmd->zone_nr_bitmap_blocks =
1485		max_t(sector_t, 1, zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT);
1486	zmd->zone_bits_per_mblk = min_t(sector_t, zmd->zone_nr_blocks,
1487					DMZ_BLOCK_SIZE_BITS);
1488
1489	/* Allocate zone array */
1490	zmd->nr_zones = 0;
1491	for (i = 0; i < zmd->nr_devs; i++) {
1492		struct dmz_dev *dev = &zmd->dev[i];
1493
1494		dev->metadata = zmd;
1495		zmd->nr_zones += dev->nr_zones;
1496
1497		atomic_set(&dev->unmap_nr_rnd, 0);
1498		INIT_LIST_HEAD(&dev->unmap_rnd_list);
1499		INIT_LIST_HEAD(&dev->map_rnd_list);
1500
1501		atomic_set(&dev->unmap_nr_seq, 0);
1502		INIT_LIST_HEAD(&dev->unmap_seq_list);
1503		INIT_LIST_HEAD(&dev->map_seq_list);
1504	}
1505
1506	if (!zmd->nr_zones) {
1507		DMERR("(%s): No zones found", zmd->devname);
1508		return -ENXIO;
1509	}
1510	xa_init(&zmd->zones);
1511
1512	DMDEBUG("(%s): Using %zu B for zone information",
1513		zmd->devname, sizeof(struct dm_zone) * zmd->nr_zones);
1514
1515	if (zmd->nr_devs > 1) {
1516		ret = dmz_emulate_zones(zmd, &zmd->dev[0]);
1517		if (ret < 0) {
1518			DMDEBUG("(%s): Failed to emulate zones, error %d",
1519				zmd->devname, ret);
1520			dmz_drop_zones(zmd);
1521			return ret;
 
 
 
 
 
 
 
1522		}
1523
1524		/*
1525		 * Primary superblock zone is always at zone 0 when multiple
1526		 * drives are present.
1527		 */
1528		zmd->sb[0].zone = dmz_get(zmd, 0);
1529
1530		for (i = 1; i < zmd->nr_devs; i++) {
1531			zoned_dev = &zmd->dev[i];
1532
1533			ret = blkdev_report_zones(zoned_dev->bdev, 0,
1534						  BLK_ALL_ZONES,
1535						  dmz_init_zone, zoned_dev);
1536			if (ret < 0) {
1537				DMDEBUG("(%s): Failed to report zones, error %d",
1538					zmd->devname, ret);
1539				dmz_drop_zones(zmd);
1540				return ret;
1541			}
1542		}
1543		return 0;
1544	}
1545
1546	/*
1547	 * Get zone information and initialize zone descriptors.  At the same
1548	 * time, determine where the super block should be: first block of the
1549	 * first randomly writable zone.
1550	 */
1551	ret = blkdev_report_zones(zoned_dev->bdev, 0, BLK_ALL_ZONES,
1552				  dmz_init_zone, zoned_dev);
1553	if (ret < 0) {
1554		DMDEBUG("(%s): Failed to report zones, error %d",
1555			zmd->devname, ret);
1556		dmz_drop_zones(zmd);
1557		return ret;
1558	}
 
 
 
 
1559
1560	return 0;
1561}
1562
1563static int dmz_update_zone_cb(struct blk_zone *blkz, unsigned int idx,
1564			      void *data)
1565{
1566	struct dm_zone *zone = data;
1567
1568	clear_bit(DMZ_OFFLINE, &zone->flags);
1569	clear_bit(DMZ_READ_ONLY, &zone->flags);
1570	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1571		set_bit(DMZ_OFFLINE, &zone->flags);
1572	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1573		set_bit(DMZ_READ_ONLY, &zone->flags);
1574
1575	if (dmz_is_seq(zone))
1576		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1577	else
1578		zone->wp_block = 0;
1579	return 0;
1580}
1581
1582/*
1583 * Update a zone information.
1584 */
1585static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1586{
1587	struct dmz_dev *dev = zone->dev;
1588	unsigned int noio_flag;
 
1589	int ret;
1590
1591	if (dev->flags & DMZ_BDEV_REGULAR)
1592		return 0;
1593
1594	/*
1595	 * Get zone information from disk. Since blkdev_report_zones() uses
1596	 * GFP_KERNEL by default for memory allocations, set the per-task
1597	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1598	 * GFP_NOIO was specified.
1599	 */
1600	noio_flag = memalloc_noio_save();
1601	ret = blkdev_report_zones(dev->bdev, dmz_start_sect(zmd, zone), 1,
1602				  dmz_update_zone_cb, zone);
1603	memalloc_noio_restore(noio_flag);
1604
1605	if (ret == 0)
1606		ret = -EIO;
1607	if (ret < 0) {
1608		dmz_dev_err(dev, "Get zone %u report failed",
1609			    zone->id);
1610		dmz_check_bdev(dev);
1611		return ret;
1612	}
1613
 
 
 
 
 
 
 
 
 
 
 
 
1614	return 0;
1615}
1616
1617/*
1618 * Check a zone write pointer position when the zone is marked
1619 * with the sequential write error flag.
1620 */
1621static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1622				    struct dm_zone *zone)
1623{
1624	struct dmz_dev *dev = zone->dev;
1625	unsigned int wp = 0;
1626	int ret;
1627
1628	wp = zone->wp_block;
1629	ret = dmz_update_zone(zmd, zone);
1630	if (ret)
1631		return ret;
1632
1633	dmz_dev_warn(dev, "Processing zone %u write error (zone wp %u/%u)",
1634		     zone->id, zone->wp_block, wp);
1635
1636	if (zone->wp_block < wp) {
1637		dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1638				      wp - zone->wp_block);
1639	}
1640
1641	return 0;
1642}
1643
 
 
 
 
 
1644/*
1645 * Reset a zone write pointer.
1646 */
1647static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1648{
1649	int ret;
1650
1651	/*
1652	 * Ignore offline zones, read only zones,
1653	 * and conventional zones.
1654	 */
1655	if (dmz_is_offline(zone) ||
1656	    dmz_is_readonly(zone) ||
1657	    dmz_is_rnd(zone))
1658		return 0;
1659
1660	if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1661		struct dmz_dev *dev = zone->dev;
1662
1663		ret = blkdev_zone_mgmt(dev->bdev, REQ_OP_ZONE_RESET,
1664				       dmz_start_sect(zmd, zone),
1665				       zmd->zone_nr_sectors, GFP_NOIO);
1666		if (ret) {
1667			dmz_dev_err(dev, "Reset zone %u failed %d",
1668				    zone->id, ret);
1669			return ret;
1670		}
1671	}
1672
1673	/* Clear write error bit and rewind write pointer position */
1674	clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1675	zone->wp_block = 0;
1676
1677	return 0;
1678}
1679
1680static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1681
1682/*
1683 * Initialize chunk mapping.
1684 */
1685static int dmz_load_mapping(struct dmz_metadata *zmd)
1686{
 
1687	struct dm_zone *dzone, *bzone;
1688	struct dmz_mblock *dmap_mblk = NULL;
1689	struct dmz_map *dmap;
1690	unsigned int i = 0, e = 0, chunk = 0;
1691	unsigned int dzone_id;
1692	unsigned int bzone_id;
1693
1694	/* Metadata block array for the chunk mapping table */
1695	zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1696				sizeof(struct dmz_mblk *), GFP_KERNEL);
1697	if (!zmd->map_mblk)
1698		return -ENOMEM;
1699
1700	/* Get chunk mapping table blocks and initialize zone mapping */
1701	while (chunk < zmd->nr_chunks) {
1702		if (!dmap_mblk) {
1703			/* Get mapping block */
1704			dmap_mblk = dmz_get_mblock(zmd, i + 1);
1705			if (IS_ERR(dmap_mblk))
1706				return PTR_ERR(dmap_mblk);
1707			zmd->map_mblk[i] = dmap_mblk;
1708			dmap = (struct dmz_map *) dmap_mblk->data;
1709			i++;
1710			e = 0;
1711		}
1712
1713		/* Check data zone */
1714		dzone_id = le32_to_cpu(dmap[e].dzone_id);
1715		if (dzone_id == DMZ_MAP_UNMAPPED)
1716			goto next;
1717
1718		if (dzone_id >= zmd->nr_zones) {
1719			dmz_zmd_err(zmd, "Chunk %u mapping: invalid data zone ID %u",
1720				    chunk, dzone_id);
1721			return -EIO;
1722		}
1723
1724		dzone = dmz_get(zmd, dzone_id);
1725		if (!dzone) {
1726			dmz_zmd_err(zmd, "Chunk %u mapping: data zone %u not present",
1727				    chunk, dzone_id);
1728			return -EIO;
1729		}
1730		set_bit(DMZ_DATA, &dzone->flags);
1731		dzone->chunk = chunk;
1732		dmz_get_zone_weight(zmd, dzone);
1733
1734		if (dmz_is_cache(dzone))
1735			list_add_tail(&dzone->link, &zmd->map_cache_list);
1736		else if (dmz_is_rnd(dzone))
1737			list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
1738		else
1739			list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
1740
1741		/* Check buffer zone */
1742		bzone_id = le32_to_cpu(dmap[e].bzone_id);
1743		if (bzone_id == DMZ_MAP_UNMAPPED)
1744			goto next;
1745
1746		if (bzone_id >= zmd->nr_zones) {
1747			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone ID %u",
1748				    chunk, bzone_id);
1749			return -EIO;
1750		}
1751
1752		bzone = dmz_get(zmd, bzone_id);
1753		if (!bzone) {
1754			dmz_zmd_err(zmd, "Chunk %u mapping: buffer zone %u not present",
1755				    chunk, bzone_id);
1756			return -EIO;
1757		}
1758		if (!dmz_is_rnd(bzone) && !dmz_is_cache(bzone)) {
1759			dmz_zmd_err(zmd, "Chunk %u mapping: invalid buffer zone %u",
1760				    chunk, bzone_id);
1761			return -EIO;
1762		}
1763
1764		set_bit(DMZ_DATA, &bzone->flags);
1765		set_bit(DMZ_BUF, &bzone->flags);
1766		bzone->chunk = chunk;
1767		bzone->bzone = dzone;
1768		dzone->bzone = bzone;
1769		dmz_get_zone_weight(zmd, bzone);
1770		if (dmz_is_cache(bzone))
1771			list_add_tail(&bzone->link, &zmd->map_cache_list);
1772		else
1773			list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
1774next:
1775		chunk++;
1776		e++;
1777		if (e >= DMZ_MAP_ENTRIES)
1778			dmap_mblk = NULL;
1779	}
1780
1781	/*
1782	 * At this point, only meta zones and mapped data zones were
1783	 * fully initialized. All remaining zones are unmapped data
1784	 * zones. Finish initializing those here.
1785	 */
1786	for (i = 0; i < zmd->nr_zones; i++) {
1787		dzone = dmz_get(zmd, i);
1788		if (!dzone)
1789			continue;
1790		if (dmz_is_meta(dzone))
1791			continue;
1792		if (dmz_is_offline(dzone))
1793			continue;
1794
1795		if (dmz_is_cache(dzone))
1796			zmd->nr_cache++;
1797		else if (dmz_is_rnd(dzone))
1798			dzone->dev->nr_rnd++;
1799		else
1800			dzone->dev->nr_seq++;
1801
1802		if (dmz_is_data(dzone)) {
1803			/* Already initialized */
1804			continue;
1805		}
1806
1807		/* Unmapped data zone */
1808		set_bit(DMZ_DATA, &dzone->flags);
1809		dzone->chunk = DMZ_MAP_UNMAPPED;
1810		if (dmz_is_cache(dzone)) {
1811			list_add_tail(&dzone->link, &zmd->unmap_cache_list);
1812			atomic_inc(&zmd->unmap_nr_cache);
1813		} else if (dmz_is_rnd(dzone)) {
1814			list_add_tail(&dzone->link,
1815				      &dzone->dev->unmap_rnd_list);
1816			atomic_inc(&dzone->dev->unmap_nr_rnd);
1817		} else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1818			list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
1819			set_bit(DMZ_RESERVED, &dzone->flags);
1820			atomic_inc(&zmd->nr_reserved_seq_zones);
1821			dzone->dev->nr_seq--;
1822		} else {
1823			list_add_tail(&dzone->link,
1824				      &dzone->dev->unmap_seq_list);
1825			atomic_inc(&dzone->dev->unmap_nr_seq);
1826		}
1827	}
1828
1829	return 0;
1830}
1831
1832/*
1833 * Set a data chunk mapping.
1834 */
1835static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1836				  unsigned int dzone_id, unsigned int bzone_id)
1837{
1838	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1839	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1840	int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1841
1842	dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1843	dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1844	dmz_dirty_mblock(zmd, dmap_mblk);
1845}
1846
1847/*
1848 * The list of mapped zones is maintained in LRU order.
1849 * This rotates a zone at the end of its map list.
1850 */
1851static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1852{
1853	if (list_empty(&zone->link))
1854		return;
1855
1856	list_del_init(&zone->link);
1857	if (dmz_is_seq(zone)) {
1858		/* LRU rotate sequential zone */
1859		list_add_tail(&zone->link, &zone->dev->map_seq_list);
1860	} else if (dmz_is_cache(zone)) {
1861		/* LRU rotate cache zone */
1862		list_add_tail(&zone->link, &zmd->map_cache_list);
1863	} else {
1864		/* LRU rotate random zone */
1865		list_add_tail(&zone->link, &zone->dev->map_rnd_list);
1866	}
1867}
1868
1869/*
1870 * The list of mapped random zones is maintained
1871 * in LRU order. This rotates a zone at the end of the list.
1872 */
1873static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1874{
1875	__dmz_lru_zone(zmd, zone);
1876	if (zone->bzone)
1877		__dmz_lru_zone(zmd, zone->bzone);
1878}
1879
1880/*
1881 * Wait for any zone to be freed.
1882 */
1883static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1884{
1885	DEFINE_WAIT(wait);
1886
1887	prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1888	dmz_unlock_map(zmd);
1889	dmz_unlock_metadata(zmd);
1890
1891	io_schedule_timeout(HZ);
1892
1893	dmz_lock_metadata(zmd);
1894	dmz_lock_map(zmd);
1895	finish_wait(&zmd->free_wq, &wait);
1896}
1897
1898/*
1899 * Lock a zone for reclaim (set the zone RECLAIM bit).
1900 * Returns false if the zone cannot be locked or if it is already locked
1901 * and 1 otherwise.
1902 */
1903int dmz_lock_zone_reclaim(struct dm_zone *zone)
1904{
1905	/* Active zones cannot be reclaimed */
1906	if (dmz_is_active(zone))
1907		return 0;
1908
1909	return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1910}
1911
1912/*
1913 * Clear a zone reclaim flag.
1914 */
1915void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1916{
1917	WARN_ON(dmz_is_active(zone));
1918	WARN_ON(!dmz_in_reclaim(zone));
1919
1920	clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1921	smp_mb__after_atomic();
1922	wake_up_bit(&zone->flags, DMZ_RECLAIM);
1923}
1924
1925/*
1926 * Wait for a zone reclaim to complete.
1927 */
1928static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1929{
1930	dmz_unlock_map(zmd);
1931	dmz_unlock_metadata(zmd);
1932	set_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1933	wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
1934	clear_bit(DMZ_RECLAIM_TERMINATE, &zone->flags);
1935	dmz_lock_metadata(zmd);
1936	dmz_lock_map(zmd);
1937}
1938
1939/*
1940 * Select a cache or random write zone for reclaim.
1941 */
1942static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd,
1943						    unsigned int idx, bool idle)
1944{
1945	struct dm_zone *dzone = NULL;
1946	struct dm_zone *zone, *maxw_z = NULL;
1947	struct list_head *zone_list;
1948
1949	/* If we have cache zones select from the cache zone list */
1950	if (zmd->nr_cache) {
1951		zone_list = &zmd->map_cache_list;
1952		/* Try to relaim random zones, too, when idle */
1953		if (idle && list_empty(zone_list))
1954			zone_list = &zmd->dev[idx].map_rnd_list;
1955	} else
1956		zone_list = &zmd->dev[idx].map_rnd_list;
1957
1958	/*
1959	 * Find the buffer zone with the heaviest weight or the first (oldest)
1960	 * data zone that can be reclaimed.
1961	 */
1962	list_for_each_entry(zone, zone_list, link) {
1963		if (dmz_is_buf(zone)) {
1964			dzone = zone->bzone;
1965			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1966				continue;
1967			if (!maxw_z || maxw_z->weight < dzone->weight)
1968				maxw_z = dzone;
1969		} else {
1970			dzone = zone;
1971			if (dmz_lock_zone_reclaim(dzone))
1972				return dzone;
1973		}
1974	}
1975
1976	if (maxw_z && dmz_lock_zone_reclaim(maxw_z))
1977		return maxw_z;
1978
1979	/*
1980	 * If we come here, none of the zones inspected could be locked for
1981	 * reclaim. Try again, being more aggressive, that is, find the
1982	 * first zone that can be reclaimed regardless of its weitght.
1983	 */
1984	list_for_each_entry(zone, zone_list, link) {
1985		if (dmz_is_buf(zone)) {
1986			dzone = zone->bzone;
1987			if (dmz_is_rnd(dzone) && dzone->dev->dev_idx != idx)
1988				continue;
1989		} else
1990			dzone = zone;
1991		if (dmz_lock_zone_reclaim(dzone))
1992			return dzone;
1993	}
1994
1995	return NULL;
1996}
1997
1998/*
1999 * Select a buffered sequential zone for reclaim.
2000 */
2001static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd,
2002						    unsigned int idx)
2003{
2004	struct dm_zone *zone;
2005
2006	list_for_each_entry(zone, &zmd->dev[idx].map_seq_list, link) {
 
 
 
2007		if (!zone->bzone)
2008			continue;
2009		if (dmz_lock_zone_reclaim(zone))
2010			return zone;
2011	}
2012
2013	return NULL;
2014}
2015
2016/*
2017 * Select a zone for reclaim.
2018 */
2019struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd,
2020					 unsigned int dev_idx, bool idle)
2021{
2022	struct dm_zone *zone = NULL;
2023
2024	/*
2025	 * Search for a zone candidate to reclaim: 2 cases are possible.
2026	 * (1) There is no free sequential zones. Then a random data zone
2027	 *     cannot be reclaimed. So choose a sequential zone to reclaim so
2028	 *     that afterward a random zone can be reclaimed.
2029	 * (2) At least one free sequential zone is available, then choose
2030	 *     the oldest random zone (data or buffer) that can be locked.
2031	 */
2032	dmz_lock_map(zmd);
2033	if (list_empty(&zmd->reserved_seq_zones_list))
2034		zone = dmz_get_seq_zone_for_reclaim(zmd, dev_idx);
2035	if (!zone)
2036		zone = dmz_get_rnd_zone_for_reclaim(zmd, dev_idx, idle);
2037	dmz_unlock_map(zmd);
2038
2039	return zone;
2040}
2041
2042/*
2043 * Get the zone mapping a chunk, if the chunk is mapped already.
2044 * If no mapping exist and the operation is WRITE, a zone is
2045 * allocated and used to map the chunk.
2046 * The zone returned will be set to the active state.
2047 */
2048struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd,
2049				      unsigned int chunk, enum req_op op)
2050{
2051	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
2052	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
2053	int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
2054	unsigned int dzone_id;
2055	struct dm_zone *dzone = NULL;
2056	int ret = 0;
2057	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2058
2059	dmz_lock_map(zmd);
2060again:
2061	/* Get the chunk mapping */
2062	dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
2063	if (dzone_id == DMZ_MAP_UNMAPPED) {
2064		/*
2065		 * Read or discard in unmapped chunks are fine. But for
2066		 * writes, we need a mapping, so get one.
2067		 */
2068		if (op != REQ_OP_WRITE)
2069			goto out;
2070
2071		/* Allocate a random zone */
2072		dzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2073		if (!dzone) {
2074			if (dmz_dev_is_dying(zmd)) {
2075				dzone = ERR_PTR(-EIO);
2076				goto out;
2077			}
2078			dmz_wait_for_free_zones(zmd);
2079			goto again;
2080		}
2081
2082		dmz_map_zone(zmd, dzone, chunk);
2083
2084	} else {
2085		/* The chunk is already mapped: get the mapping zone */
2086		dzone = dmz_get(zmd, dzone_id);
2087		if (!dzone) {
2088			dzone = ERR_PTR(-EIO);
2089			goto out;
2090		}
2091		if (dzone->chunk != chunk) {
2092			dzone = ERR_PTR(-EIO);
2093			goto out;
2094		}
2095
2096		/* Repair write pointer if the sequential dzone has error */
2097		if (dmz_seq_write_err(dzone)) {
2098			ret = dmz_handle_seq_write_err(zmd, dzone);
2099			if (ret) {
2100				dzone = ERR_PTR(-EIO);
2101				goto out;
2102			}
2103			clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
2104		}
2105	}
2106
2107	/*
2108	 * If the zone is being reclaimed, the chunk mapping may change
2109	 * to a different zone. So wait for reclaim and retry. Otherwise,
2110	 * activate the zone (this will prevent reclaim from touching it).
2111	 */
2112	if (dmz_in_reclaim(dzone)) {
2113		dmz_wait_for_reclaim(zmd, dzone);
2114		goto again;
2115	}
2116	dmz_activate_zone(dzone);
2117	dmz_lru_zone(zmd, dzone);
2118out:
2119	dmz_unlock_map(zmd);
2120
2121	return dzone;
2122}
2123
2124/*
2125 * Write and discard change the block validity of data zones and their buffer
2126 * zones. Check here that valid blocks are still present. If all blocks are
2127 * invalid, the zones can be unmapped on the fly without waiting for reclaim
2128 * to do it.
2129 */
2130void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
2131{
2132	struct dm_zone *bzone;
2133
2134	dmz_lock_map(zmd);
2135
2136	bzone = dzone->bzone;
2137	if (bzone) {
2138		if (dmz_weight(bzone))
2139			dmz_lru_zone(zmd, bzone);
2140		else {
2141			/* Empty buffer zone: reclaim it */
2142			dmz_unmap_zone(zmd, bzone);
2143			dmz_free_zone(zmd, bzone);
2144			bzone = NULL;
2145		}
2146	}
2147
2148	/* Deactivate the data zone */
2149	dmz_deactivate_zone(dzone);
2150	if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
2151		dmz_lru_zone(zmd, dzone);
2152	else {
2153		/* Unbuffered inactive empty data zone: reclaim it */
2154		dmz_unmap_zone(zmd, dzone);
2155		dmz_free_zone(zmd, dzone);
2156	}
2157
2158	dmz_unlock_map(zmd);
2159}
2160
2161/*
2162 * Allocate and map a random zone to buffer a chunk
2163 * already mapped to a sequential zone.
2164 */
2165struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
2166				     struct dm_zone *dzone)
2167{
2168	struct dm_zone *bzone;
2169	int alloc_flags = zmd->nr_cache ? DMZ_ALLOC_CACHE : DMZ_ALLOC_RND;
2170
2171	dmz_lock_map(zmd);
2172again:
2173	bzone = dzone->bzone;
2174	if (bzone)
2175		goto out;
2176
2177	/* Allocate a random zone */
2178	bzone = dmz_alloc_zone(zmd, 0, alloc_flags);
2179	if (!bzone) {
2180		if (dmz_dev_is_dying(zmd)) {
2181			bzone = ERR_PTR(-EIO);
2182			goto out;
2183		}
2184		dmz_wait_for_free_zones(zmd);
2185		goto again;
2186	}
2187
2188	/* Update the chunk mapping */
2189	dmz_set_chunk_mapping(zmd, dzone->chunk, dzone->id, bzone->id);
 
2190
2191	set_bit(DMZ_BUF, &bzone->flags);
2192	bzone->chunk = dzone->chunk;
2193	bzone->bzone = dzone;
2194	dzone->bzone = bzone;
2195	if (dmz_is_cache(bzone))
2196		list_add_tail(&bzone->link, &zmd->map_cache_list);
2197	else
2198		list_add_tail(&bzone->link, &bzone->dev->map_rnd_list);
2199out:
2200	dmz_unlock_map(zmd);
2201
2202	return bzone;
2203}
2204
2205/*
2206 * Get an unmapped (free) zone.
2207 * This must be called with the mapping lock held.
2208 */
2209struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned int dev_idx,
2210			       unsigned long flags)
2211{
2212	struct list_head *list;
2213	struct dm_zone *zone;
2214	int i;
2215
2216	/* Schedule reclaim to ensure free zones are available */
2217	if (!(flags & DMZ_ALLOC_RECLAIM)) {
2218		for (i = 0; i < zmd->nr_devs; i++)
2219			dmz_schedule_reclaim(zmd->dev[i].reclaim);
2220	}
2221
2222	i = 0;
2223again:
2224	if (flags & DMZ_ALLOC_CACHE)
2225		list = &zmd->unmap_cache_list;
2226	else if (flags & DMZ_ALLOC_RND)
2227		list = &zmd->dev[dev_idx].unmap_rnd_list;
2228	else
2229		list = &zmd->dev[dev_idx].unmap_seq_list;
2230
2231	if (list_empty(list)) {
2232		/*
2233		 * No free zone: return NULL if this is for not reclaim.
 
2234		 */
2235		if (!(flags & DMZ_ALLOC_RECLAIM))
 
2236			return NULL;
2237		/*
2238		 * Try to allocate from other devices
2239		 */
2240		if (i < zmd->nr_devs) {
2241			dev_idx = (dev_idx + 1) % zmd->nr_devs;
2242			i++;
2243			goto again;
2244		}
2245
2246		/*
2247		 * Fallback to the reserved sequential zones
2248		 */
2249		zone = list_first_entry_or_null(&zmd->reserved_seq_zones_list,
2250						struct dm_zone, link);
2251		if (zone) {
2252			list_del_init(&zone->link);
2253			atomic_dec(&zmd->nr_reserved_seq_zones);
2254		}
2255		return zone;
2256	}
2257
2258	zone = list_first_entry(list, struct dm_zone, link);
2259	list_del_init(&zone->link);
2260
2261	if (dmz_is_cache(zone))
2262		atomic_dec(&zmd->unmap_nr_cache);
2263	else if (dmz_is_rnd(zone))
2264		atomic_dec(&zone->dev->unmap_nr_rnd);
2265	else
2266		atomic_dec(&zone->dev->unmap_nr_seq);
2267
2268	if (dmz_is_offline(zone)) {
2269		dmz_zmd_warn(zmd, "Zone %u is offline", zone->id);
2270		zone = NULL;
2271		goto again;
2272	}
2273	if (dmz_is_meta(zone)) {
2274		dmz_zmd_warn(zmd, "Zone %u has metadata", zone->id);
2275		zone = NULL;
2276		goto again;
2277	}
 
2278	return zone;
2279}
2280
2281/*
2282 * Free a zone.
2283 * This must be called with the mapping lock held.
2284 */
2285void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2286{
2287	/* If this is a sequential zone, reset it */
2288	if (dmz_is_seq(zone))
2289		dmz_reset_zone(zmd, zone);
2290
2291	/* Return the zone to its type unmap list */
2292	if (dmz_is_cache(zone)) {
2293		list_add_tail(&zone->link, &zmd->unmap_cache_list);
2294		atomic_inc(&zmd->unmap_nr_cache);
2295	} else if (dmz_is_rnd(zone)) {
2296		list_add_tail(&zone->link, &zone->dev->unmap_rnd_list);
2297		atomic_inc(&zone->dev->unmap_nr_rnd);
2298	} else if (dmz_is_reserved(zone)) {
2299		list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
2300		atomic_inc(&zmd->nr_reserved_seq_zones);
2301	} else {
2302		list_add_tail(&zone->link, &zone->dev->unmap_seq_list);
2303		atomic_inc(&zone->dev->unmap_nr_seq);
2304	}
2305
2306	wake_up_all(&zmd->free_wq);
2307}
2308
2309/*
2310 * Map a chunk to a zone.
2311 * This must be called with the mapping lock held.
2312 */
2313void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
2314		  unsigned int chunk)
2315{
2316	/* Set the chunk mapping */
2317	dmz_set_chunk_mapping(zmd, chunk, dzone->id,
2318			      DMZ_MAP_UNMAPPED);
2319	dzone->chunk = chunk;
2320	if (dmz_is_cache(dzone))
2321		list_add_tail(&dzone->link, &zmd->map_cache_list);
2322	else if (dmz_is_rnd(dzone))
2323		list_add_tail(&dzone->link, &dzone->dev->map_rnd_list);
2324	else
2325		list_add_tail(&dzone->link, &dzone->dev->map_seq_list);
2326}
2327
2328/*
2329 * Unmap a zone.
2330 * This must be called with the mapping lock held.
2331 */
2332void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
2333{
2334	unsigned int chunk = zone->chunk;
2335	unsigned int dzone_id;
2336
2337	if (chunk == DMZ_MAP_UNMAPPED) {
2338		/* Already unmapped */
2339		return;
2340	}
2341
2342	if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
2343		/*
2344		 * Unmapping the chunk buffer zone: clear only
2345		 * the chunk buffer mapping
2346		 */
2347		dzone_id = zone->bzone->id;
2348		zone->bzone->bzone = NULL;
2349		zone->bzone = NULL;
2350
2351	} else {
2352		/*
2353		 * Unmapping the chunk data zone: the zone must
2354		 * not be buffered.
2355		 */
2356		if (WARN_ON(zone->bzone)) {
2357			zone->bzone->bzone = NULL;
2358			zone->bzone = NULL;
2359		}
2360		dzone_id = DMZ_MAP_UNMAPPED;
2361	}
2362
2363	dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
2364
2365	zone->chunk = DMZ_MAP_UNMAPPED;
2366	list_del_init(&zone->link);
2367}
2368
2369/*
2370 * Set @nr_bits bits in @bitmap starting from @bit.
2371 * Return the number of bits changed from 0 to 1.
2372 */
2373static unsigned int dmz_set_bits(unsigned long *bitmap,
2374				 unsigned int bit, unsigned int nr_bits)
2375{
2376	unsigned long *addr;
2377	unsigned int end = bit + nr_bits;
2378	unsigned int n = 0;
2379
2380	while (bit < end) {
2381		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2382		    ((end - bit) >= BITS_PER_LONG)) {
2383			/* Try to set the whole word at once */
2384			addr = bitmap + BIT_WORD(bit);
2385			if (*addr == 0) {
2386				*addr = ULONG_MAX;
2387				n += BITS_PER_LONG;
2388				bit += BITS_PER_LONG;
2389				continue;
2390			}
2391		}
2392
2393		if (!test_and_set_bit(bit, bitmap))
2394			n++;
2395		bit++;
2396	}
2397
2398	return n;
2399}
2400
2401/*
2402 * Get the bitmap block storing the bit for chunk_block in zone.
2403 */
2404static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
2405					 struct dm_zone *zone,
2406					 sector_t chunk_block)
2407{
2408	sector_t bitmap_block = 1 + zmd->nr_map_blocks +
2409		(sector_t)(zone->id * zmd->zone_nr_bitmap_blocks) +
2410		(chunk_block >> DMZ_BLOCK_SHIFT_BITS);
2411
2412	return dmz_get_mblock(zmd, bitmap_block);
2413}
2414
2415/*
2416 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
2417 */
2418int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2419			  struct dm_zone *to_zone)
2420{
2421	struct dmz_mblock *from_mblk, *to_mblk;
2422	sector_t chunk_block = 0;
2423
2424	/* Get the zones bitmap blocks */
2425	while (chunk_block < zmd->zone_nr_blocks) {
2426		from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
2427		if (IS_ERR(from_mblk))
2428			return PTR_ERR(from_mblk);
2429		to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
2430		if (IS_ERR(to_mblk)) {
2431			dmz_release_mblock(zmd, from_mblk);
2432			return PTR_ERR(to_mblk);
2433		}
2434
2435		memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
2436		dmz_dirty_mblock(zmd, to_mblk);
2437
2438		dmz_release_mblock(zmd, to_mblk);
2439		dmz_release_mblock(zmd, from_mblk);
2440
2441		chunk_block += zmd->zone_bits_per_mblk;
2442	}
2443
2444	to_zone->weight = from_zone->weight;
2445
2446	return 0;
2447}
2448
2449/*
2450 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
2451 * starting from chunk_block.
2452 */
2453int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
2454			   struct dm_zone *to_zone, sector_t chunk_block)
2455{
2456	unsigned int nr_blocks;
2457	int ret;
2458
2459	/* Get the zones bitmap blocks */
2460	while (chunk_block < zmd->zone_nr_blocks) {
2461		/* Get a valid region from the source zone */
2462		ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2463		if (ret <= 0)
2464			return ret;
2465
2466		nr_blocks = ret;
2467		ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2468		if (ret)
2469			return ret;
2470
2471		chunk_block += nr_blocks;
2472	}
2473
2474	return 0;
2475}
2476
2477/*
2478 * Validate all the blocks in the range [block..block+nr_blocks-1].
2479 */
2480int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2481			sector_t chunk_block, unsigned int nr_blocks)
2482{
2483	unsigned int count, bit, nr_bits;
2484	unsigned int zone_nr_blocks = zmd->zone_nr_blocks;
2485	struct dmz_mblock *mblk;
2486	unsigned int n = 0;
2487
2488	dmz_zmd_debug(zmd, "=> VALIDATE zone %u, block %llu, %u blocks",
2489		      zone->id, (unsigned long long)chunk_block,
2490		      nr_blocks);
2491
2492	WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2493
2494	while (nr_blocks) {
2495		/* Get bitmap block */
2496		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2497		if (IS_ERR(mblk))
2498			return PTR_ERR(mblk);
2499
2500		/* Set bits */
2501		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2502		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2503
2504		count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2505		if (count) {
2506			dmz_dirty_mblock(zmd, mblk);
2507			n += count;
2508		}
2509		dmz_release_mblock(zmd, mblk);
2510
2511		nr_blocks -= nr_bits;
2512		chunk_block += nr_bits;
2513	}
2514
2515	if (likely(zone->weight + n <= zone_nr_blocks))
2516		zone->weight += n;
2517	else {
2518		dmz_zmd_warn(zmd, "Zone %u: weight %u should be <= %u",
2519			     zone->id, zone->weight,
2520			     zone_nr_blocks - n);
2521		zone->weight = zone_nr_blocks;
2522	}
2523
2524	return 0;
2525}
2526
2527/*
2528 * Clear nr_bits bits in bitmap starting from bit.
2529 * Return the number of bits cleared.
2530 */
2531static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2532{
2533	unsigned long *addr;
2534	int end = bit + nr_bits;
2535	int n = 0;
2536
2537	while (bit < end) {
2538		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2539		    ((end - bit) >= BITS_PER_LONG)) {
2540			/* Try to clear whole word at once */
2541			addr = bitmap + BIT_WORD(bit);
2542			if (*addr == ULONG_MAX) {
2543				*addr = 0;
2544				n += BITS_PER_LONG;
2545				bit += BITS_PER_LONG;
2546				continue;
2547			}
2548		}
2549
2550		if (test_and_clear_bit(bit, bitmap))
2551			n++;
2552		bit++;
2553	}
2554
2555	return n;
2556}
2557
2558/*
2559 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2560 */
2561int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2562			  sector_t chunk_block, unsigned int nr_blocks)
2563{
2564	unsigned int count, bit, nr_bits;
2565	struct dmz_mblock *mblk;
2566	unsigned int n = 0;
2567
2568	dmz_zmd_debug(zmd, "=> INVALIDATE zone %u, block %llu, %u blocks",
2569		      zone->id, (u64)chunk_block, nr_blocks);
2570
2571	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2572
2573	while (nr_blocks) {
2574		/* Get bitmap block */
2575		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2576		if (IS_ERR(mblk))
2577			return PTR_ERR(mblk);
2578
2579		/* Clear bits */
2580		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2581		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2582
2583		count = dmz_clear_bits((unsigned long *)mblk->data,
2584				       bit, nr_bits);
2585		if (count) {
2586			dmz_dirty_mblock(zmd, mblk);
2587			n += count;
2588		}
2589		dmz_release_mblock(zmd, mblk);
2590
2591		nr_blocks -= nr_bits;
2592		chunk_block += nr_bits;
2593	}
2594
2595	if (zone->weight >= n)
2596		zone->weight -= n;
2597	else {
2598		dmz_zmd_warn(zmd, "Zone %u: weight %u should be >= %u",
2599			     zone->id, zone->weight, n);
2600		zone->weight = 0;
2601	}
2602
2603	return 0;
2604}
2605
2606/*
2607 * Get a block bit value.
2608 */
2609static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2610			  sector_t chunk_block)
2611{
2612	struct dmz_mblock *mblk;
2613	int ret;
2614
2615	WARN_ON(chunk_block >= zmd->zone_nr_blocks);
2616
2617	/* Get bitmap block */
2618	mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2619	if (IS_ERR(mblk))
2620		return PTR_ERR(mblk);
2621
2622	/* Get offset */
2623	ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2624		       (unsigned long *) mblk->data) != 0;
2625
2626	dmz_release_mblock(zmd, mblk);
2627
2628	return ret;
2629}
2630
2631/*
2632 * Return the number of blocks from chunk_block to the first block with a bit
2633 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2634 */
2635static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2636				 sector_t chunk_block, unsigned int nr_blocks,
2637				 int set)
2638{
2639	struct dmz_mblock *mblk;
2640	unsigned int bit, set_bit, nr_bits;
2641	unsigned int zone_bits = zmd->zone_bits_per_mblk;
2642	unsigned long *bitmap;
2643	int n = 0;
2644
2645	WARN_ON(chunk_block + nr_blocks > zmd->zone_nr_blocks);
2646
2647	while (nr_blocks) {
2648		/* Get bitmap block */
2649		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2650		if (IS_ERR(mblk))
2651			return PTR_ERR(mblk);
2652
2653		/* Get offset */
2654		bitmap = (unsigned long *) mblk->data;
2655		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2656		nr_bits = min(nr_blocks, zone_bits - bit);
2657		if (set)
2658			set_bit = find_next_bit(bitmap, zone_bits, bit);
2659		else
2660			set_bit = find_next_zero_bit(bitmap, zone_bits, bit);
2661		dmz_release_mblock(zmd, mblk);
2662
2663		n += set_bit - bit;
2664		if (set_bit < zone_bits)
2665			break;
2666
2667		nr_blocks -= nr_bits;
2668		chunk_block += nr_bits;
2669	}
2670
2671	return n;
2672}
2673
2674/*
2675 * Test if chunk_block is valid. If it is, the number of consecutive
2676 * valid blocks from chunk_block will be returned.
2677 */
2678int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2679		    sector_t chunk_block)
2680{
2681	int valid;
2682
2683	valid = dmz_test_block(zmd, zone, chunk_block);
2684	if (valid <= 0)
2685		return valid;
2686
2687	/* The block is valid: get the number of valid blocks from block */
2688	return dmz_to_next_set_block(zmd, zone, chunk_block,
2689				     zmd->zone_nr_blocks - chunk_block, 0);
2690}
2691
2692/*
2693 * Find the first valid block from @chunk_block in @zone.
2694 * If such a block is found, its number is returned using
2695 * @chunk_block and the total number of valid blocks from @chunk_block
2696 * is returned.
2697 */
2698int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2699			  sector_t *chunk_block)
2700{
2701	sector_t start_block = *chunk_block;
2702	int ret;
2703
2704	ret = dmz_to_next_set_block(zmd, zone, start_block,
2705				    zmd->zone_nr_blocks - start_block, 1);
2706	if (ret < 0)
2707		return ret;
2708
2709	start_block += ret;
2710	*chunk_block = start_block;
2711
2712	return dmz_to_next_set_block(zmd, zone, start_block,
2713				     zmd->zone_nr_blocks - start_block, 0);
2714}
2715
2716/*
2717 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2718 */
2719static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2720{
2721	unsigned long *addr;
2722	int end = bit + nr_bits;
2723	int n = 0;
2724
2725	while (bit < end) {
2726		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2727		    ((end - bit) >= BITS_PER_LONG)) {
2728			addr = (unsigned long *)bitmap + BIT_WORD(bit);
2729			if (*addr == ULONG_MAX) {
2730				n += BITS_PER_LONG;
2731				bit += BITS_PER_LONG;
2732				continue;
2733			}
2734		}
2735
2736		if (test_bit(bit, bitmap))
2737			n++;
2738		bit++;
2739	}
2740
2741	return n;
2742}
2743
2744/*
2745 * Get a zone weight.
2746 */
2747static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2748{
2749	struct dmz_mblock *mblk;
2750	sector_t chunk_block = 0;
2751	unsigned int bit, nr_bits;
2752	unsigned int nr_blocks = zmd->zone_nr_blocks;
2753	void *bitmap;
2754	int n = 0;
2755
2756	while (nr_blocks) {
2757		/* Get bitmap block */
2758		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2759		if (IS_ERR(mblk)) {
2760			n = 0;
2761			break;
2762		}
2763
2764		/* Count bits in this block */
2765		bitmap = mblk->data;
2766		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2767		nr_bits = min(nr_blocks, zmd->zone_bits_per_mblk - bit);
2768		n += dmz_count_bits(bitmap, bit, nr_bits);
2769
2770		dmz_release_mblock(zmd, mblk);
2771
2772		nr_blocks -= nr_bits;
2773		chunk_block += nr_bits;
2774	}
2775
2776	zone->weight = n;
2777}
2778
2779/*
2780 * Cleanup the zoned metadata resources.
2781 */
2782static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2783{
2784	struct rb_root *root;
2785	struct dmz_mblock *mblk, *next;
2786	int i;
2787
2788	/* Release zone mapping resources */
2789	if (zmd->map_mblk) {
2790		for (i = 0; i < zmd->nr_map_blocks; i++)
2791			dmz_release_mblock(zmd, zmd->map_mblk[i]);
2792		kfree(zmd->map_mblk);
2793		zmd->map_mblk = NULL;
2794	}
2795
2796	/* Release super blocks */
2797	for (i = 0; i < 2; i++) {
2798		if (zmd->sb[i].mblk) {
2799			dmz_free_mblock(zmd, zmd->sb[i].mblk);
2800			zmd->sb[i].mblk = NULL;
2801		}
2802	}
2803
2804	/* Free cached blocks */
2805	while (!list_empty(&zmd->mblk_dirty_list)) {
2806		mblk = list_first_entry(&zmd->mblk_dirty_list,
2807					struct dmz_mblock, link);
2808		dmz_zmd_warn(zmd, "mblock %llu still in dirty list (ref %u)",
2809			     (u64)mblk->no, mblk->ref);
2810		list_del_init(&mblk->link);
2811		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2812		dmz_free_mblock(zmd, mblk);
2813	}
2814
2815	while (!list_empty(&zmd->mblk_lru_list)) {
2816		mblk = list_first_entry(&zmd->mblk_lru_list,
2817					struct dmz_mblock, link);
2818		list_del_init(&mblk->link);
2819		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2820		dmz_free_mblock(zmd, mblk);
2821	}
2822
2823	/* Sanity checks: the mblock rbtree should now be empty */
2824	root = &zmd->mblk_rbtree;
2825	rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2826		dmz_zmd_warn(zmd, "mblock %llu ref %u still in rbtree",
2827			     (u64)mblk->no, mblk->ref);
2828		mblk->ref = 0;
2829		dmz_free_mblock(zmd, mblk);
2830	}
2831
2832	/* Free the zone descriptors */
2833	dmz_drop_zones(zmd);
2834
2835	mutex_destroy(&zmd->mblk_flush_lock);
2836	mutex_destroy(&zmd->map_lock);
2837}
2838
2839static void dmz_print_dev(struct dmz_metadata *zmd, int num)
2840{
2841	struct dmz_dev *dev = &zmd->dev[num];
2842
2843	if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE)
2844		dmz_dev_info(dev, "Regular block device");
2845	else
2846		dmz_dev_info(dev, "Host-%s zoned block device",
2847			     bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2848			     "aware" : "managed");
2849	if (zmd->sb_version > 1) {
2850		sector_t sector_offset =
2851			dev->zone_offset << zmd->zone_nr_sectors_shift;
2852
2853		dmz_dev_info(dev, "  %llu 512-byte logical sectors (offset %llu)",
2854			     (u64)dev->capacity, (u64)sector_offset);
2855		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors (offset %llu)",
2856			     dev->nr_zones, (u64)zmd->zone_nr_sectors,
2857			     (u64)dev->zone_offset);
2858	} else {
2859		dmz_dev_info(dev, "  %llu 512-byte logical sectors",
2860			     (u64)dev->capacity);
2861		dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors",
2862			     dev->nr_zones, (u64)zmd->zone_nr_sectors);
2863	}
2864}
2865
2866/*
2867 * Initialize the zoned metadata.
2868 */
2869int dmz_ctr_metadata(struct dmz_dev *dev, int num_dev,
2870		     struct dmz_metadata **metadata,
2871		     const char *devname)
2872{
2873	struct dmz_metadata *zmd;
2874	unsigned int i;
2875	struct dm_zone *zone;
2876	int ret;
2877
2878	zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2879	if (!zmd)
2880		return -ENOMEM;
2881
2882	strcpy(zmd->devname, devname);
2883	zmd->dev = dev;
2884	zmd->nr_devs = num_dev;
2885	zmd->mblk_rbtree = RB_ROOT;
2886	init_rwsem(&zmd->mblk_sem);
2887	mutex_init(&zmd->mblk_flush_lock);
2888	spin_lock_init(&zmd->mblk_lock);
2889	INIT_LIST_HEAD(&zmd->mblk_lru_list);
2890	INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2891
2892	mutex_init(&zmd->map_lock);
2893
2894	atomic_set(&zmd->unmap_nr_cache, 0);
2895	INIT_LIST_HEAD(&zmd->unmap_cache_list);
2896	INIT_LIST_HEAD(&zmd->map_cache_list);
 
 
 
2897
2898	atomic_set(&zmd->nr_reserved_seq_zones, 0);
2899	INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2900
2901	init_waitqueue_head(&zmd->free_wq);
2902
2903	/* Initialize zone descriptors */
2904	ret = dmz_init_zones(zmd);
2905	if (ret)
2906		goto err;
2907
2908	/* Get super block */
2909	ret = dmz_load_sb(zmd);
2910	if (ret)
2911		goto err;
2912
2913	/* Set metadata zones starting from sb_zone */
 
2914	for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2915		zone = dmz_get(zmd, zmd->sb[0].zone->id + i);
2916		if (!zone) {
2917			dmz_zmd_err(zmd,
2918				    "metadata zone %u not present", i);
2919			ret = -ENXIO;
2920			goto err;
2921		}
2922		if (!dmz_is_rnd(zone) && !dmz_is_cache(zone)) {
2923			dmz_zmd_err(zmd,
2924				    "metadata zone %d is not random", i);
2925			ret = -ENXIO;
2926			goto err;
2927		}
2928		set_bit(DMZ_META, &zone->flags);
2929	}
 
2930	/* Load mapping table */
2931	ret = dmz_load_mapping(zmd);
2932	if (ret)
2933		goto err;
2934
2935	/*
2936	 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2937	 * blocks and enough blocks to be able to cache the bitmap blocks of
2938	 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2939	 * the cache to add 512 more metadata blocks.
2940	 */
2941	zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2942	zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2943	zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2944	zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2945	zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2946
2947	/* Metadata cache shrinker */
2948	ret = register_shrinker(&zmd->mblk_shrinker, "md-meta:(%u:%u)",
2949				MAJOR(dev->bdev->bd_dev),
2950				MINOR(dev->bdev->bd_dev));
2951	if (ret) {
2952		dmz_zmd_err(zmd, "Register metadata cache shrinker failed");
2953		goto err;
2954	}
2955
2956	dmz_zmd_info(zmd, "DM-Zoned metadata version %d", zmd->sb_version);
2957	for (i = 0; i < zmd->nr_devs; i++)
2958		dmz_print_dev(zmd, i);
2959
2960	dmz_zmd_info(zmd, "  %u zones of %llu 512-byte logical sectors",
2961		     zmd->nr_zones, (u64)zmd->zone_nr_sectors);
2962	dmz_zmd_debug(zmd, "  %u metadata zones",
2963		      zmd->nr_meta_zones * 2);
2964	dmz_zmd_debug(zmd, "  %u data zones for %u chunks",
2965		      zmd->nr_data_zones, zmd->nr_chunks);
2966	dmz_zmd_debug(zmd, "    %u cache zones (%u unmapped)",
2967		      zmd->nr_cache, atomic_read(&zmd->unmap_nr_cache));
2968	for (i = 0; i < zmd->nr_devs; i++) {
2969		dmz_zmd_debug(zmd, "    %u random zones (%u unmapped)",
2970			      dmz_nr_rnd_zones(zmd, i),
2971			      dmz_nr_unmap_rnd_zones(zmd, i));
2972		dmz_zmd_debug(zmd, "    %u sequential zones (%u unmapped)",
2973			      dmz_nr_seq_zones(zmd, i),
2974			      dmz_nr_unmap_seq_zones(zmd, i));
2975	}
2976	dmz_zmd_debug(zmd, "  %u reserved sequential data zones",
2977		      zmd->nr_reserved_seq);
2978	dmz_zmd_debug(zmd, "Format:");
2979	dmz_zmd_debug(zmd, "%u metadata blocks per set (%u max cache)",
2980		      zmd->nr_meta_blocks, zmd->max_nr_mblks);
2981	dmz_zmd_debug(zmd, "  %u data zone mapping blocks",
2982		      zmd->nr_map_blocks);
2983	dmz_zmd_debug(zmd, "  %u bitmap blocks",
2984		      zmd->nr_bitmap_blocks);
2985
2986	*metadata = zmd;
2987
2988	return 0;
2989err:
2990	dmz_cleanup_metadata(zmd);
2991	kfree(zmd);
2992	*metadata = NULL;
2993
2994	return ret;
2995}
2996
2997/*
2998 * Cleanup the zoned metadata resources.
2999 */
3000void dmz_dtr_metadata(struct dmz_metadata *zmd)
3001{
3002	unregister_shrinker(&zmd->mblk_shrinker);
3003	dmz_cleanup_metadata(zmd);
3004	kfree(zmd);
3005}
3006
3007/*
3008 * Check zone information on resume.
3009 */
3010int dmz_resume_metadata(struct dmz_metadata *zmd)
3011{
 
3012	struct dm_zone *zone;
3013	sector_t wp_block;
3014	unsigned int i;
3015	int ret;
3016
3017	/* Check zones */
3018	for (i = 0; i < zmd->nr_zones; i++) {
3019		zone = dmz_get(zmd, i);
3020		if (!zone) {
3021			dmz_zmd_err(zmd, "Unable to get zone %u", i);
3022			return -EIO;
3023		}
 
3024		wp_block = zone->wp_block;
3025
3026		ret = dmz_update_zone(zmd, zone);
3027		if (ret) {
3028			dmz_zmd_err(zmd, "Broken zone %u", i);
3029			return ret;
3030		}
3031
3032		if (dmz_is_offline(zone)) {
3033			dmz_zmd_warn(zmd, "Zone %u is offline", i);
3034			continue;
3035		}
3036
3037		/* Check write pointer */
3038		if (!dmz_is_seq(zone))
3039			zone->wp_block = 0;
3040		else if (zone->wp_block != wp_block) {
3041			dmz_zmd_err(zmd, "Zone %u: Invalid wp (%llu / %llu)",
3042				    i, (u64)zone->wp_block, (u64)wp_block);
3043			zone->wp_block = wp_block;
3044			dmz_invalidate_blocks(zmd, zone, zone->wp_block,
3045					      zmd->zone_nr_blocks - zone->wp_block);
3046		}
3047	}
3048
3049	return 0;
3050}
v5.4
   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	1
  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	/* Padding to full 512B sector */
  73	u8		reserved[464];		/* 512 */
  74};
  75
  76/*
  77 * Chunk mapping entry: entries are indexed by chunk number
  78 * and give the zone ID (dzone_id) mapping the chunk on disk.
  79 * This zone may be sequential or random. If it is a sequential
  80 * zone, a second zone (bzone_id) used as a write buffer may
  81 * also be specified. This second zone will always be a randomly
  82 * writeable zone.
  83 */
  84struct dmz_map {
  85	__le32			dzone_id;
  86	__le32			bzone_id;
  87};
  88
  89/*
  90 * Chunk mapping table metadata: 512 8-bytes entries per 4KB block.
  91 */
  92#define DMZ_MAP_ENTRIES		(DMZ_BLOCK_SIZE / sizeof(struct dmz_map))
  93#define DMZ_MAP_ENTRIES_SHIFT	(ilog2(DMZ_MAP_ENTRIES))
  94#define DMZ_MAP_ENTRIES_MASK	(DMZ_MAP_ENTRIES - 1)
  95#define DMZ_MAP_UNMAPPED	UINT_MAX
  96
  97/*
  98 * Meta data block descriptor (for cached metadata blocks).
  99 */
 100struct dmz_mblock {
 101	struct rb_node		node;
 102	struct list_head	link;
 103	sector_t		no;
 104	unsigned int		ref;
 105	unsigned long		state;
 106	struct page		*page;
 107	void			*data;
 108};
 109
 110/*
 111 * Metadata block state flags.
 112 */
 113enum {
 114	DMZ_META_DIRTY,
 115	DMZ_META_READING,
 116	DMZ_META_WRITING,
 117	DMZ_META_ERROR,
 118};
 119
 120/*
 121 * Super block information (one per metadata set).
 122 */
 123struct dmz_sb {
 124	sector_t		block;
 
 125	struct dmz_mblock	*mblk;
 126	struct dmz_super	*sb;
 
 127};
 128
 129/*
 130 * In-memory metadata.
 131 */
 132struct dmz_metadata {
 133	struct dmz_dev		*dev;
 
 
 
 
 
 134
 135	sector_t		zone_bitmap_size;
 136	unsigned int		zone_nr_bitmap_blocks;
 
 
 
 
 
 
 
 137
 138	unsigned int		nr_bitmap_blocks;
 139	unsigned int		nr_map_blocks;
 140
 
 141	unsigned int		nr_useable_zones;
 142	unsigned int		nr_meta_blocks;
 143	unsigned int		nr_meta_zones;
 144	unsigned int		nr_data_zones;
 
 145	unsigned int		nr_rnd_zones;
 146	unsigned int		nr_reserved_seq;
 147	unsigned int		nr_chunks;
 148
 149	/* Zone information array */
 150	struct dm_zone		*zones;
 151
 152	struct dm_zone		*sb_zone;
 153	struct dmz_sb		sb[2];
 154	unsigned int		mblk_primary;
 
 155	u64			sb_gen;
 156	unsigned int		min_nr_mblks;
 157	unsigned int		max_nr_mblks;
 158	atomic_t		nr_mblks;
 159	struct rw_semaphore	mblk_sem;
 160	struct mutex		mblk_flush_lock;
 161	spinlock_t		mblk_lock;
 162	struct rb_root		mblk_rbtree;
 163	struct list_head	mblk_lru_list;
 164	struct list_head	mblk_dirty_list;
 165	struct shrinker		mblk_shrinker;
 166
 167	/* Zone allocation management */
 168	struct mutex		map_lock;
 169	struct dmz_mblock	**map_mblk;
 170	unsigned int		nr_rnd;
 171	atomic_t		unmap_nr_rnd;
 172	struct list_head	unmap_rnd_list;
 173	struct list_head	map_rnd_list;
 174
 175	unsigned int		nr_seq;
 176	atomic_t		unmap_nr_seq;
 177	struct list_head	unmap_seq_list;
 178	struct list_head	map_seq_list;
 179
 180	atomic_t		nr_reserved_seq_zones;
 181	struct list_head	reserved_seq_zones_list;
 182
 183	wait_queue_head_t	free_wq;
 184};
 185
 
 
 
 
 
 
 
 
 
 
 
 186/*
 187 * Various accessors
 188 */
 189unsigned int dmz_id(struct dmz_metadata *zmd, struct dm_zone *zone)
 190{
 191	return ((unsigned int)(zone - zmd->zones));
 
 
 
 192}
 193
 194sector_t dmz_start_sect(struct dmz_metadata *zmd, struct dm_zone *zone)
 195{
 196	return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_sectors_shift;
 
 
 197}
 198
 199sector_t dmz_start_block(struct dmz_metadata *zmd, struct dm_zone *zone)
 200{
 201	return (sector_t)dmz_id(zmd, zone) << zmd->dev->zone_nr_blocks_shift;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 202}
 203
 204unsigned int dmz_nr_chunks(struct dmz_metadata *zmd)
 205{
 206	return zmd->nr_chunks;
 207}
 208
 209unsigned int dmz_nr_rnd_zones(struct dmz_metadata *zmd)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 210{
 211	return zmd->nr_rnd;
 212}
 213
 214unsigned int dmz_nr_unmap_rnd_zones(struct dmz_metadata *zmd)
 215{
 216	return atomic_read(&zmd->unmap_nr_rnd);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 217}
 218
 219/*
 220 * Lock/unlock mapping table.
 221 * The map lock also protects all the zone lists.
 222 */
 223void dmz_lock_map(struct dmz_metadata *zmd)
 224{
 225	mutex_lock(&zmd->map_lock);
 226}
 227
 228void dmz_unlock_map(struct dmz_metadata *zmd)
 229{
 230	mutex_unlock(&zmd->map_lock);
 231}
 232
 233/*
 234 * Lock/unlock metadata access. This is a "read" lock on a semaphore
 235 * that prevents metadata flush from running while metadata are being
 236 * modified. The actual metadata write mutual exclusion is achieved with
 237 * the map lock and zone state management (active and reclaim state are
 238 * mutually exclusive).
 239 */
 240void dmz_lock_metadata(struct dmz_metadata *zmd)
 241{
 242	down_read(&zmd->mblk_sem);
 243}
 244
 245void dmz_unlock_metadata(struct dmz_metadata *zmd)
 246{
 247	up_read(&zmd->mblk_sem);
 248}
 249
 250/*
 251 * Lock/unlock flush: prevent concurrent executions
 252 * of dmz_flush_metadata as well as metadata modification in reclaim
 253 * while flush is being executed.
 254 */
 255void dmz_lock_flush(struct dmz_metadata *zmd)
 256{
 257	mutex_lock(&zmd->mblk_flush_lock);
 258}
 259
 260void dmz_unlock_flush(struct dmz_metadata *zmd)
 261{
 262	mutex_unlock(&zmd->mblk_flush_lock);
 263}
 264
 265/*
 266 * Allocate a metadata block.
 267 */
 268static struct dmz_mblock *dmz_alloc_mblock(struct dmz_metadata *zmd,
 269					   sector_t mblk_no)
 270{
 271	struct dmz_mblock *mblk = NULL;
 272
 273	/* See if we can reuse cached blocks */
 274	if (zmd->max_nr_mblks && atomic_read(&zmd->nr_mblks) > zmd->max_nr_mblks) {
 275		spin_lock(&zmd->mblk_lock);
 276		mblk = list_first_entry_or_null(&zmd->mblk_lru_list,
 277						struct dmz_mblock, link);
 278		if (mblk) {
 279			list_del_init(&mblk->link);
 280			rb_erase(&mblk->node, &zmd->mblk_rbtree);
 281			mblk->no = mblk_no;
 282		}
 283		spin_unlock(&zmd->mblk_lock);
 284		if (mblk)
 285			return mblk;
 286	}
 287
 288	/* Allocate a new block */
 289	mblk = kmalloc(sizeof(struct dmz_mblock), GFP_NOIO);
 290	if (!mblk)
 291		return NULL;
 292
 293	mblk->page = alloc_page(GFP_NOIO);
 294	if (!mblk->page) {
 295		kfree(mblk);
 296		return NULL;
 297	}
 298
 299	RB_CLEAR_NODE(&mblk->node);
 300	INIT_LIST_HEAD(&mblk->link);
 301	mblk->ref = 0;
 302	mblk->state = 0;
 303	mblk->no = mblk_no;
 304	mblk->data = page_address(mblk->page);
 305
 306	atomic_inc(&zmd->nr_mblks);
 307
 308	return mblk;
 309}
 310
 311/*
 312 * Free a metadata block.
 313 */
 314static void dmz_free_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
 315{
 316	__free_pages(mblk->page, 0);
 317	kfree(mblk);
 318
 319	atomic_dec(&zmd->nr_mblks);
 320}
 321
 322/*
 323 * Insert a metadata block in the rbtree.
 324 */
 325static void dmz_insert_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
 326{
 327	struct rb_root *root = &zmd->mblk_rbtree;
 328	struct rb_node **new = &(root->rb_node), *parent = NULL;
 329	struct dmz_mblock *b;
 330
 331	/* Figure out where to put the new node */
 332	while (*new) {
 333		b = container_of(*new, struct dmz_mblock, node);
 334		parent = *new;
 335		new = (b->no < mblk->no) ? &((*new)->rb_left) : &((*new)->rb_right);
 336	}
 337
 338	/* Add new node and rebalance tree */
 339	rb_link_node(&mblk->node, parent, new);
 340	rb_insert_color(&mblk->node, root);
 341}
 342
 343/*
 344 * Lookup a metadata block in the rbtree. If the block is found, increment
 345 * its reference count.
 346 */
 347static struct dmz_mblock *dmz_get_mblock_fast(struct dmz_metadata *zmd,
 348					      sector_t mblk_no)
 349{
 350	struct rb_root *root = &zmd->mblk_rbtree;
 351	struct rb_node *node = root->rb_node;
 352	struct dmz_mblock *mblk;
 353
 354	while (node) {
 355		mblk = container_of(node, struct dmz_mblock, node);
 356		if (mblk->no == mblk_no) {
 357			/*
 358			 * If this is the first reference to the block,
 359			 * remove it from the LRU list.
 360			 */
 361			mblk->ref++;
 362			if (mblk->ref == 1 &&
 363			    !test_bit(DMZ_META_DIRTY, &mblk->state))
 364				list_del_init(&mblk->link);
 365			return mblk;
 366		}
 367		node = (mblk->no < mblk_no) ? node->rb_left : node->rb_right;
 368	}
 369
 370	return NULL;
 371}
 372
 373/*
 374 * Metadata block BIO end callback.
 375 */
 376static void dmz_mblock_bio_end_io(struct bio *bio)
 377{
 378	struct dmz_mblock *mblk = bio->bi_private;
 379	int flag;
 380
 381	if (bio->bi_status)
 382		set_bit(DMZ_META_ERROR, &mblk->state);
 383
 384	if (bio_op(bio) == REQ_OP_WRITE)
 385		flag = DMZ_META_WRITING;
 386	else
 387		flag = DMZ_META_READING;
 388
 389	clear_bit_unlock(flag, &mblk->state);
 390	smp_mb__after_atomic();
 391	wake_up_bit(&mblk->state, flag);
 392
 393	bio_put(bio);
 394}
 395
 396/*
 397 * Read an uncached metadata block from disk and add it to the cache.
 398 */
 399static struct dmz_mblock *dmz_get_mblock_slow(struct dmz_metadata *zmd,
 400					      sector_t mblk_no)
 401{
 402	struct dmz_mblock *mblk, *m;
 403	sector_t block = zmd->sb[zmd->mblk_primary].block + mblk_no;
 
 404	struct bio *bio;
 405
 406	if (dmz_bdev_is_dying(zmd->dev))
 407		return ERR_PTR(-EIO);
 408
 409	/* Get a new block and a BIO to read it */
 410	mblk = dmz_alloc_mblock(zmd, mblk_no);
 411	if (!mblk)
 412		return ERR_PTR(-ENOMEM);
 413
 414	bio = bio_alloc(GFP_NOIO, 1);
 415	if (!bio) {
 416		dmz_free_mblock(zmd, mblk);
 417		return ERR_PTR(-ENOMEM);
 418	}
 419
 420	spin_lock(&zmd->mblk_lock);
 421
 422	/*
 423	 * Make sure that another context did not start reading
 424	 * the block already.
 425	 */
 426	m = dmz_get_mblock_fast(zmd, mblk_no);
 427	if (m) {
 428		spin_unlock(&zmd->mblk_lock);
 429		dmz_free_mblock(zmd, mblk);
 430		bio_put(bio);
 431		return m;
 432	}
 433
 434	mblk->ref++;
 435	set_bit(DMZ_META_READING, &mblk->state);
 436	dmz_insert_mblock(zmd, mblk);
 437
 438	spin_unlock(&zmd->mblk_lock);
 439
 440	/* Submit read BIO */
 441	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 442	bio_set_dev(bio, zmd->dev->bdev);
 443	bio->bi_private = mblk;
 444	bio->bi_end_io = dmz_mblock_bio_end_io;
 445	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META | REQ_PRIO);
 446	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
 447	submit_bio(bio);
 448
 449	return mblk;
 450}
 451
 452/*
 453 * Free metadata blocks.
 454 */
 455static unsigned long dmz_shrink_mblock_cache(struct dmz_metadata *zmd,
 456					     unsigned long limit)
 457{
 458	struct dmz_mblock *mblk;
 459	unsigned long count = 0;
 460
 461	if (!zmd->max_nr_mblks)
 462		return 0;
 463
 464	while (!list_empty(&zmd->mblk_lru_list) &&
 465	       atomic_read(&zmd->nr_mblks) > zmd->min_nr_mblks &&
 466	       count < limit) {
 467		mblk = list_first_entry(&zmd->mblk_lru_list,
 468					struct dmz_mblock, link);
 469		list_del_init(&mblk->link);
 470		rb_erase(&mblk->node, &zmd->mblk_rbtree);
 471		dmz_free_mblock(zmd, mblk);
 472		count++;
 473	}
 474
 475	return count;
 476}
 477
 478/*
 479 * For mblock shrinker: get the number of unused metadata blocks in the cache.
 480 */
 481static unsigned long dmz_mblock_shrinker_count(struct shrinker *shrink,
 482					       struct shrink_control *sc)
 483{
 484	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
 485
 486	return atomic_read(&zmd->nr_mblks);
 487}
 488
 489/*
 490 * For mblock shrinker: scan unused metadata blocks and shrink the cache.
 491 */
 492static unsigned long dmz_mblock_shrinker_scan(struct shrinker *shrink,
 493					      struct shrink_control *sc)
 494{
 495	struct dmz_metadata *zmd = container_of(shrink, struct dmz_metadata, mblk_shrinker);
 496	unsigned long count;
 497
 498	spin_lock(&zmd->mblk_lock);
 499	count = dmz_shrink_mblock_cache(zmd, sc->nr_to_scan);
 500	spin_unlock(&zmd->mblk_lock);
 501
 502	return count ? count : SHRINK_STOP;
 503}
 504
 505/*
 506 * Release a metadata block.
 507 */
 508static void dmz_release_mblock(struct dmz_metadata *zmd,
 509			       struct dmz_mblock *mblk)
 510{
 511
 512	if (!mblk)
 513		return;
 514
 515	spin_lock(&zmd->mblk_lock);
 516
 517	mblk->ref--;
 518	if (mblk->ref == 0) {
 519		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 520			rb_erase(&mblk->node, &zmd->mblk_rbtree);
 521			dmz_free_mblock(zmd, mblk);
 522		} else if (!test_bit(DMZ_META_DIRTY, &mblk->state)) {
 523			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
 524			dmz_shrink_mblock_cache(zmd, 1);
 525		}
 526	}
 527
 528	spin_unlock(&zmd->mblk_lock);
 529}
 530
 531/*
 532 * Get a metadata block from the rbtree. If the block
 533 * is not present, read it from disk.
 534 */
 535static struct dmz_mblock *dmz_get_mblock(struct dmz_metadata *zmd,
 536					 sector_t mblk_no)
 537{
 538	struct dmz_mblock *mblk;
 
 539
 540	/* Check rbtree */
 541	spin_lock(&zmd->mblk_lock);
 542	mblk = dmz_get_mblock_fast(zmd, mblk_no);
 543	spin_unlock(&zmd->mblk_lock);
 544
 545	if (!mblk) {
 546		/* Cache miss: read the block from disk */
 547		mblk = dmz_get_mblock_slow(zmd, mblk_no);
 548		if (IS_ERR(mblk))
 549			return mblk;
 550	}
 551
 552	/* Wait for on-going read I/O and check for error */
 553	wait_on_bit_io(&mblk->state, DMZ_META_READING,
 554		       TASK_UNINTERRUPTIBLE);
 555	if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 556		dmz_release_mblock(zmd, mblk);
 
 557		return ERR_PTR(-EIO);
 558	}
 559
 560	return mblk;
 561}
 562
 563/*
 564 * Mark a metadata block dirty.
 565 */
 566static void dmz_dirty_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk)
 567{
 568	spin_lock(&zmd->mblk_lock);
 569	if (!test_and_set_bit(DMZ_META_DIRTY, &mblk->state))
 570		list_add_tail(&mblk->link, &zmd->mblk_dirty_list);
 571	spin_unlock(&zmd->mblk_lock);
 572}
 573
 574/*
 575 * Issue a metadata block write BIO.
 576 */
 577static int dmz_write_mblock(struct dmz_metadata *zmd, struct dmz_mblock *mblk,
 578			    unsigned int set)
 579{
 
 580	sector_t block = zmd->sb[set].block + mblk->no;
 581	struct bio *bio;
 582
 583	if (dmz_bdev_is_dying(zmd->dev))
 584		return -EIO;
 585
 586	bio = bio_alloc(GFP_NOIO, 1);
 587	if (!bio) {
 588		set_bit(DMZ_META_ERROR, &mblk->state);
 589		return -ENOMEM;
 590	}
 591
 592	set_bit(DMZ_META_WRITING, &mblk->state);
 593
 594	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 595	bio_set_dev(bio, zmd->dev->bdev);
 596	bio->bi_private = mblk;
 597	bio->bi_end_io = dmz_mblock_bio_end_io;
 598	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_META | REQ_PRIO);
 599	bio_add_page(bio, mblk->page, DMZ_BLOCK_SIZE, 0);
 600	submit_bio(bio);
 601
 602	return 0;
 603}
 604
 605/*
 606 * Read/write a metadata block.
 607 */
 608static int dmz_rdwr_block(struct dmz_metadata *zmd, int op, sector_t block,
 609			  struct page *page)
 610{
 611	struct bio *bio;
 612	int ret;
 613
 614	if (dmz_bdev_is_dying(zmd->dev))
 615		return -EIO;
 616
 617	bio = bio_alloc(GFP_NOIO, 1);
 618	if (!bio)
 619		return -ENOMEM;
 620
 
 
 621	bio->bi_iter.bi_sector = dmz_blk2sect(block);
 622	bio_set_dev(bio, zmd->dev->bdev);
 623	bio_set_op_attrs(bio, op, REQ_SYNC | REQ_META | REQ_PRIO);
 624	bio_add_page(bio, page, DMZ_BLOCK_SIZE, 0);
 625	ret = submit_bio_wait(bio);
 626	bio_put(bio);
 627
 
 
 628	return ret;
 629}
 630
 631/*
 632 * Write super block of the specified metadata set.
 633 */
 634static int dmz_write_sb(struct dmz_metadata *zmd, unsigned int set)
 635{
 636	sector_t block = zmd->sb[set].block;
 637	struct dmz_mblock *mblk = zmd->sb[set].mblk;
 638	struct dmz_super *sb = zmd->sb[set].sb;
 
 
 639	u64 sb_gen = zmd->sb_gen + 1;
 640	int ret;
 641
 642	sb->magic = cpu_to_le32(DMZ_MAGIC);
 643	sb->version = cpu_to_le32(DMZ_META_VER);
 
 
 
 
 
 
 
 644
 645	sb->gen = cpu_to_le64(sb_gen);
 646
 647	sb->sb_block = cpu_to_le64(block);
 
 
 
 
 
 
 648	sb->nr_meta_blocks = cpu_to_le32(zmd->nr_meta_blocks);
 649	sb->nr_reserved_seq = cpu_to_le32(zmd->nr_reserved_seq);
 650	sb->nr_chunks = cpu_to_le32(zmd->nr_chunks);
 651
 652	sb->nr_map_blocks = cpu_to_le32(zmd->nr_map_blocks);
 653	sb->nr_bitmap_blocks = cpu_to_le32(zmd->nr_bitmap_blocks);
 654
 655	sb->crc = 0;
 656	sb->crc = cpu_to_le32(crc32_le(sb_gen, (unsigned char *)sb, DMZ_BLOCK_SIZE));
 657
 658	ret = dmz_rdwr_block(zmd, REQ_OP_WRITE, block, mblk->page);
 
 659	if (ret == 0)
 660		ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
 661
 662	return ret;
 663}
 664
 665/*
 666 * Write dirty metadata blocks to the specified set.
 667 */
 668static int dmz_write_dirty_mblocks(struct dmz_metadata *zmd,
 669				   struct list_head *write_list,
 670				   unsigned int set)
 671{
 672	struct dmz_mblock *mblk;
 
 673	struct blk_plug plug;
 674	int ret = 0, nr_mblks_submitted = 0;
 675
 676	/* Issue writes */
 677	blk_start_plug(&plug);
 678	list_for_each_entry(mblk, write_list, link) {
 679		ret = dmz_write_mblock(zmd, mblk, set);
 680		if (ret)
 681			break;
 682		nr_mblks_submitted++;
 683	}
 684	blk_finish_plug(&plug);
 685
 686	/* Wait for completion */
 687	list_for_each_entry(mblk, write_list, link) {
 688		if (!nr_mblks_submitted)
 689			break;
 690		wait_on_bit_io(&mblk->state, DMZ_META_WRITING,
 691			       TASK_UNINTERRUPTIBLE);
 692		if (test_bit(DMZ_META_ERROR, &mblk->state)) {
 693			clear_bit(DMZ_META_ERROR, &mblk->state);
 
 694			ret = -EIO;
 695		}
 696		nr_mblks_submitted--;
 697	}
 698
 699	/* Flush drive cache (this will also sync data) */
 700	if (ret == 0)
 701		ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
 702
 703	return ret;
 704}
 705
 706/*
 707 * Log dirty metadata blocks.
 708 */
 709static int dmz_log_dirty_mblocks(struct dmz_metadata *zmd,
 710				 struct list_head *write_list)
 711{
 712	unsigned int log_set = zmd->mblk_primary ^ 0x1;
 713	int ret;
 714
 715	/* Write dirty blocks to the log */
 716	ret = dmz_write_dirty_mblocks(zmd, write_list, log_set);
 717	if (ret)
 718		return ret;
 719
 720	/*
 721	 * No error so far: now validate the log by updating the
 722	 * log index super block generation.
 723	 */
 724	ret = dmz_write_sb(zmd, log_set);
 725	if (ret)
 726		return ret;
 727
 728	return 0;
 729}
 730
 731/*
 732 * Flush dirty metadata blocks.
 733 */
 734int dmz_flush_metadata(struct dmz_metadata *zmd)
 735{
 736	struct dmz_mblock *mblk;
 737	struct list_head write_list;
 
 738	int ret;
 739
 740	if (WARN_ON(!zmd))
 741		return 0;
 742
 743	INIT_LIST_HEAD(&write_list);
 744
 745	/*
 746	 * Make sure that metadata blocks are stable before logging: take
 747	 * the write lock on the metadata semaphore to prevent target BIOs
 748	 * from modifying metadata.
 749	 */
 750	down_write(&zmd->mblk_sem);
 
 751
 752	/*
 753	 * This is called from the target flush work and reclaim work.
 754	 * Concurrent execution is not allowed.
 755	 */
 756	dmz_lock_flush(zmd);
 757
 758	if (dmz_bdev_is_dying(zmd->dev)) {
 759		ret = -EIO;
 760		goto out;
 761	}
 762
 763	/* Get dirty blocks */
 764	spin_lock(&zmd->mblk_lock);
 765	list_splice_init(&zmd->mblk_dirty_list, &write_list);
 766	spin_unlock(&zmd->mblk_lock);
 767
 768	/* If there are no dirty metadata blocks, just flush the device cache */
 769	if (list_empty(&write_list)) {
 770		ret = blkdev_issue_flush(zmd->dev->bdev, GFP_NOIO, NULL);
 771		goto out;
 772	}
 773
 774	/*
 775	 * The primary metadata set is still clean. Keep it this way until
 776	 * all updates are successful in the secondary set. That is, use
 777	 * the secondary set as a log.
 778	 */
 779	ret = dmz_log_dirty_mblocks(zmd, &write_list);
 780	if (ret)
 781		goto out;
 782
 783	/*
 784	 * The log is on disk. It is now safe to update in place
 785	 * in the primary metadata set.
 786	 */
 787	ret = dmz_write_dirty_mblocks(zmd, &write_list, zmd->mblk_primary);
 788	if (ret)
 789		goto out;
 790
 791	ret = dmz_write_sb(zmd, zmd->mblk_primary);
 792	if (ret)
 793		goto out;
 794
 795	while (!list_empty(&write_list)) {
 796		mblk = list_first_entry(&write_list, struct dmz_mblock, link);
 797		list_del_init(&mblk->link);
 798
 799		spin_lock(&zmd->mblk_lock);
 800		clear_bit(DMZ_META_DIRTY, &mblk->state);
 801		if (mblk->ref == 0)
 802			list_add_tail(&mblk->link, &zmd->mblk_lru_list);
 803		spin_unlock(&zmd->mblk_lock);
 804	}
 805
 806	zmd->sb_gen++;
 807out:
 808	if (ret && !list_empty(&write_list)) {
 
 
 
 
 
 
 809		spin_lock(&zmd->mblk_lock);
 810		list_splice(&write_list, &zmd->mblk_dirty_list);
 811		spin_unlock(&zmd->mblk_lock);
 812	}
 813
 814	dmz_unlock_flush(zmd);
 815	up_write(&zmd->mblk_sem);
 816
 817	return ret;
 818}
 819
 820/*
 821 * Check super block.
 822 */
 823static int dmz_check_sb(struct dmz_metadata *zmd, struct dmz_super *sb)
 
 824{
 
 
 825	unsigned int nr_meta_zones, nr_data_zones;
 826	struct dmz_dev *dev = zmd->dev;
 827	u32 crc, stored_crc;
 828	u64 gen;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 829
 830	gen = le64_to_cpu(sb->gen);
 831	stored_crc = le32_to_cpu(sb->crc);
 832	sb->crc = 0;
 833	crc = crc32_le(gen, (unsigned char *)sb, DMZ_BLOCK_SIZE);
 834	if (crc != stored_crc) {
 835		dmz_dev_err(dev, "Invalid checksum (needed 0x%08x, got 0x%08x)",
 836			    crc, stored_crc);
 837		return -ENXIO;
 838	}
 839
 840	if (le32_to_cpu(sb->magic) != DMZ_MAGIC) {
 841		dmz_dev_err(dev, "Invalid meta magic (needed 0x%08x, got 0x%08x)",
 842			    DMZ_MAGIC, le32_to_cpu(sb->magic));
 843		return -ENXIO;
 844	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 845
 846	if (le32_to_cpu(sb->version) != DMZ_META_VER) {
 847		dmz_dev_err(dev, "Invalid meta version (needed %d, got %d)",
 848			    DMZ_META_VER, le32_to_cpu(sb->version));
 849		return -ENXIO;
 
 
 
 
 
 
 850	}
 851
 852	nr_meta_zones = (le32_to_cpu(sb->nr_meta_blocks) + dev->zone_nr_blocks - 1)
 853		>> dev->zone_nr_blocks_shift;
 854	if (!nr_meta_zones ||
 855	    nr_meta_zones >= zmd->nr_rnd_zones) {
 
 856		dmz_dev_err(dev, "Invalid number of metadata blocks");
 857		return -ENXIO;
 858	}
 859
 860	if (!le32_to_cpu(sb->nr_reserved_seq) ||
 861	    le32_to_cpu(sb->nr_reserved_seq) >= (zmd->nr_useable_zones - nr_meta_zones)) {
 862		dmz_dev_err(dev, "Invalid number of reserved sequential zones");
 863		return -ENXIO;
 864	}
 865
 866	nr_data_zones = zmd->nr_useable_zones -
 867		(nr_meta_zones * 2 + le32_to_cpu(sb->nr_reserved_seq));
 868	if (le32_to_cpu(sb->nr_chunks) > nr_data_zones) {
 869		dmz_dev_err(dev, "Invalid number of chunks %u / %u",
 870			    le32_to_cpu(sb->nr_chunks), nr_data_zones);
 871		return -ENXIO;
 872	}
 873
 874	/* OK */
 875	zmd->nr_meta_blocks = le32_to_cpu(sb->nr_meta_blocks);
 876	zmd->nr_reserved_seq = le32_to_cpu(sb->nr_reserved_seq);
 877	zmd->nr_chunks = le32_to_cpu(sb->nr_chunks);
 878	zmd->nr_map_blocks = le32_to_cpu(sb->nr_map_blocks);
 879	zmd->nr_bitmap_blocks = le32_to_cpu(sb->nr_bitmap_blocks);
 880	zmd->nr_meta_zones = nr_meta_zones;
 881	zmd->nr_data_zones = nr_data_zones;
 882
 883	return 0;
 884}
 885
 886/*
 887 * Read the first or second super block from disk.
 888 */
 889static int dmz_read_sb(struct dmz_metadata *zmd, unsigned int set)
 890{
 891	return dmz_rdwr_block(zmd, REQ_OP_READ, zmd->sb[set].block,
 892			      zmd->sb[set].mblk->page);
 
 
 
 893}
 894
 895/*
 896 * Determine the position of the secondary super blocks on disk.
 897 * This is used only if a corruption of the primary super block
 898 * is detected.
 899 */
 900static int dmz_lookup_secondary_sb(struct dmz_metadata *zmd)
 901{
 902	unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
 903	struct dmz_mblock *mblk;
 
 904	int i;
 905
 906	/* Allocate a block */
 907	mblk = dmz_alloc_mblock(zmd, 0);
 908	if (!mblk)
 909		return -ENOMEM;
 910
 911	zmd->sb[1].mblk = mblk;
 912	zmd->sb[1].sb = mblk->data;
 913
 914	/* Bad first super block: search for the second one */
 915	zmd->sb[1].block = zmd->sb[0].block + zone_nr_blocks;
 916	for (i = 0; i < zmd->nr_rnd_zones - 1; i++) {
 917		if (dmz_read_sb(zmd, 1) != 0)
 
 
 918			break;
 919		if (le32_to_cpu(zmd->sb[1].sb->magic) == DMZ_MAGIC)
 920			return 0;
 921		zmd->sb[1].block += zone_nr_blocks;
 
 922	}
 923
 924	dmz_free_mblock(zmd, mblk);
 925	zmd->sb[1].mblk = NULL;
 
 
 926
 927	return -EIO;
 928}
 929
 930/*
 931 * Read the first or second super block from disk.
 932 */
 933static int dmz_get_sb(struct dmz_metadata *zmd, unsigned int set)
 934{
 935	struct dmz_mblock *mblk;
 936	int ret;
 937
 938	/* Allocate a block */
 939	mblk = dmz_alloc_mblock(zmd, 0);
 940	if (!mblk)
 941		return -ENOMEM;
 942
 943	zmd->sb[set].mblk = mblk;
 944	zmd->sb[set].sb = mblk->data;
 945
 946	/* Read super block */
 947	ret = dmz_read_sb(zmd, set);
 948	if (ret) {
 949		dmz_free_mblock(zmd, mblk);
 950		zmd->sb[set].mblk = NULL;
 951		return ret;
 952	}
 953
 954	return 0;
 955}
 956
 957/*
 958 * Recover a metadata set.
 959 */
 960static int dmz_recover_mblocks(struct dmz_metadata *zmd, unsigned int dst_set)
 961{
 962	unsigned int src_set = dst_set ^ 0x1;
 963	struct page *page;
 964	int i, ret;
 965
 966	dmz_dev_warn(zmd->dev, "Metadata set %u invalid: recovering", dst_set);
 
 967
 968	if (dst_set == 0)
 969		zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
 970	else {
 971		zmd->sb[1].block = zmd->sb[0].block +
 972			(zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
 973	}
 974
 975	page = alloc_page(GFP_NOIO);
 976	if (!page)
 977		return -ENOMEM;
 978
 979	/* Copy metadata blocks */
 980	for (i = 1; i < zmd->nr_meta_blocks; i++) {
 981		ret = dmz_rdwr_block(zmd, REQ_OP_READ,
 982				     zmd->sb[src_set].block + i, page);
 983		if (ret)
 984			goto out;
 985		ret = dmz_rdwr_block(zmd, REQ_OP_WRITE,
 986				     zmd->sb[dst_set].block + i, page);
 987		if (ret)
 988			goto out;
 989	}
 990
 991	/* Finalize with the super block */
 992	if (!zmd->sb[dst_set].mblk) {
 993		zmd->sb[dst_set].mblk = dmz_alloc_mblock(zmd, 0);
 994		if (!zmd->sb[dst_set].mblk) {
 995			ret = -ENOMEM;
 996			goto out;
 997		}
 998		zmd->sb[dst_set].sb = zmd->sb[dst_set].mblk->data;
 999	}
1000
1001	ret = dmz_write_sb(zmd, dst_set);
1002out:
1003	__free_pages(page, 0);
1004
1005	return ret;
1006}
1007
1008/*
1009 * Get super block from disk.
1010 */
1011static int dmz_load_sb(struct dmz_metadata *zmd)
1012{
1013	bool sb_good[2] = {false, false};
1014	u64 sb_gen[2] = {0, 0};
1015	int ret;
1016
 
 
 
 
 
1017	/* Read and check the primary super block */
1018	zmd->sb[0].block = dmz_start_block(zmd, zmd->sb_zone);
1019	ret = dmz_get_sb(zmd, 0);
 
1020	if (ret) {
1021		dmz_dev_err(zmd->dev, "Read primary super block failed");
1022		return ret;
1023	}
1024
1025	ret = dmz_check_sb(zmd, zmd->sb[0].sb);
1026
1027	/* Read and check secondary super block */
1028	if (ret == 0) {
1029		sb_good[0] = true;
1030		zmd->sb[1].block = zmd->sb[0].block +
1031			(zmd->nr_meta_zones << zmd->dev->zone_nr_blocks_shift);
1032		ret = dmz_get_sb(zmd, 1);
 
 
 
 
 
 
1033	} else
1034		ret = dmz_lookup_secondary_sb(zmd);
1035
1036	if (ret) {
1037		dmz_dev_err(zmd->dev, "Read secondary super block failed");
1038		return ret;
1039	}
1040
1041	ret = dmz_check_sb(zmd, zmd->sb[1].sb);
1042	if (ret == 0)
1043		sb_good[1] = true;
1044
1045	/* Use highest generation sb first */
1046	if (!sb_good[0] && !sb_good[1]) {
1047		dmz_dev_err(zmd->dev, "No valid super block found");
1048		return -EIO;
1049	}
1050
1051	if (sb_good[0])
1052		sb_gen[0] = le64_to_cpu(zmd->sb[0].sb->gen);
1053	else
1054		ret = dmz_recover_mblocks(zmd, 0);
 
 
 
 
 
 
1055
1056	if (sb_good[1])
1057		sb_gen[1] = le64_to_cpu(zmd->sb[1].sb->gen);
1058	else
1059		ret = dmz_recover_mblocks(zmd, 1);
1060
1061	if (ret) {
1062		dmz_dev_err(zmd->dev, "Recovery failed");
1063		return -EIO;
 
 
1064	}
1065
1066	if (sb_gen[0] >= sb_gen[1]) {
1067		zmd->sb_gen = sb_gen[0];
1068		zmd->mblk_primary = 0;
1069	} else {
1070		zmd->sb_gen = sb_gen[1];
1071		zmd->mblk_primary = 1;
1072	}
1073
1074	dmz_dev_debug(zmd->dev, "Using super block %u (gen %llu)",
 
1075		      zmd->mblk_primary, zmd->sb_gen);
1076
1077	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1078}
1079
1080/*
1081 * Initialize a zone descriptor.
1082 */
1083static int dmz_init_zone(struct dmz_metadata *zmd, struct dm_zone *zone,
1084			 struct blk_zone *blkz)
1085{
1086	struct dmz_dev *dev = zmd->dev;
 
 
 
1087
1088	/* Ignore the eventual last runt (smaller) zone */
1089	if (blkz->len != dev->zone_nr_sectors) {
1090		if (blkz->start + blkz->len == dev->capacity)
 
 
 
 
 
 
 
1091			return 0;
1092		return -ENXIO;
1093	}
1094
1095	INIT_LIST_HEAD(&zone->link);
1096	atomic_set(&zone->refcount, 0);
1097	zone->chunk = DMZ_MAP_UNMAPPED;
 
 
 
1098
1099	if (blkz->type == BLK_ZONE_TYPE_CONVENTIONAL) {
 
1100		set_bit(DMZ_RND, &zone->flags);
1101		zmd->nr_rnd_zones++;
1102	} else if (blkz->type == BLK_ZONE_TYPE_SEQWRITE_REQ ||
1103		   blkz->type == BLK_ZONE_TYPE_SEQWRITE_PREF) {
1104		set_bit(DMZ_SEQ, &zone->flags);
1105	} else
 
1106		return -ENXIO;
1107
1108	if (blkz->cond == BLK_ZONE_COND_OFFLINE)
1109		set_bit(DMZ_OFFLINE, &zone->flags);
1110	else if (blkz->cond == BLK_ZONE_COND_READONLY)
1111		set_bit(DMZ_READ_ONLY, &zone->flags);
1112
1113	if (dmz_is_rnd(zone))
1114		zone->wp_block = 0;
1115	else
1116		zone->wp_block = dmz_sect2blk(blkz->wp - blkz->start);
1117
1118	if (!dmz_is_offline(zone) && !dmz_is_readonly(zone)) {
 
 
 
 
1119		zmd->nr_useable_zones++;
1120		if (dmz_is_rnd(zone)) {
1121			zmd->nr_rnd_zones++;
1122			if (!zmd->sb_zone) {
1123				/* Super block zone */
1124				zmd->sb_zone = zone;
1125			}
1126		}
 
 
 
 
 
 
 
 
1127	}
 
 
1128
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1129	return 0;
1130}
1131
1132/*
1133 * Free zones descriptors.
1134 */
1135static void dmz_drop_zones(struct dmz_metadata *zmd)
1136{
1137	kfree(zmd->zones);
1138	zmd->zones = NULL;
 
 
 
 
 
 
 
1139}
1140
1141/*
1142 * The size of a zone report in number of zones.
1143 * This results in 4096*64B=256KB report zones commands.
1144 */
1145#define DMZ_REPORT_NR_ZONES	4096
1146
1147/*
1148 * Allocate and initialize zone descriptors using the zone
1149 * information from disk.
1150 */
1151static int dmz_init_zones(struct dmz_metadata *zmd)
1152{
1153	struct dmz_dev *dev = zmd->dev;
1154	struct dm_zone *zone;
1155	struct blk_zone *blkz;
1156	unsigned int nr_blkz;
1157	sector_t sector = 0;
1158	int i, ret = 0;
1159
1160	/* Init */
1161	zmd->zone_bitmap_size = dev->zone_nr_blocks >> 3;
1162	zmd->zone_nr_bitmap_blocks = zmd->zone_bitmap_size >> DMZ_BLOCK_SHIFT;
 
 
 
 
 
 
 
1163
1164	/* Allocate zone array */
1165	zmd->zones = kcalloc(dev->nr_zones, sizeof(struct dm_zone), GFP_KERNEL);
1166	if (!zmd->zones)
1167		return -ENOMEM;
 
 
 
1168
1169	dmz_dev_info(dev, "Using %zu B for zone information",
1170		     sizeof(struct dm_zone) * dev->nr_zones);
 
1171
1172	/* Get zone information */
1173	nr_blkz = DMZ_REPORT_NR_ZONES;
1174	blkz = kcalloc(nr_blkz, sizeof(struct blk_zone), GFP_KERNEL);
1175	if (!blkz) {
1176		ret = -ENOMEM;
1177		goto out;
 
 
1178	}
 
 
 
 
1179
1180	/*
1181	 * Get zone information and initialize zone descriptors.
1182	 * At the same time, determine where the super block
1183	 * should be: first block of the first randomly writable
1184	 * zone.
1185	 */
1186	zone = zmd->zones;
1187	while (sector < dev->capacity) {
1188		/* Get zone information */
1189		nr_blkz = DMZ_REPORT_NR_ZONES;
1190		ret = blkdev_report_zones(dev->bdev, sector, blkz, &nr_blkz);
1191		if (ret) {
1192			dmz_dev_err(dev, "Report zones failed %d", ret);
1193			goto out;
1194		}
1195
1196		if (!nr_blkz)
1197			break;
 
 
 
 
 
 
1198
1199		/* Process report */
1200		for (i = 0; i < nr_blkz; i++) {
1201			ret = dmz_init_zone(zmd, zone, &blkz[i]);
1202			if (ret)
1203				goto out;
1204			sector += dev->zone_nr_sectors;
1205			zone++;
 
 
1206		}
 
1207	}
1208
1209	/* The entire zone configuration of the disk should now be known */
1210	if (sector < dev->capacity) {
1211		dmz_dev_err(dev, "Failed to get correct zone information");
1212		ret = -ENXIO;
 
 
 
 
 
 
 
 
1213	}
1214out:
1215	kfree(blkz);
1216	if (ret)
1217		dmz_drop_zones(zmd);
1218
1219	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1220}
1221
1222/*
1223 * Update a zone information.
1224 */
1225static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1226{
1227	unsigned int nr_blkz = 1;
1228	unsigned int noio_flag;
1229	struct blk_zone blkz;
1230	int ret;
1231
 
 
 
1232	/*
1233	 * Get zone information from disk. Since blkdev_report_zones() uses
1234	 * GFP_KERNEL by default for memory allocations, set the per-task
1235	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
1236	 * GFP_NOIO was specified.
1237	 */
1238	noio_flag = memalloc_noio_save();
1239	ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
1240				  &blkz, &nr_blkz);
1241	memalloc_noio_restore(noio_flag);
1242	if (!nr_blkz)
 
1243		ret = -EIO;
1244	if (ret) {
1245		dmz_dev_err(zmd->dev, "Get zone %u report failed",
1246			    dmz_id(zmd, zone));
 
1247		return ret;
1248	}
1249
1250	clear_bit(DMZ_OFFLINE, &zone->flags);
1251	clear_bit(DMZ_READ_ONLY, &zone->flags);
1252	if (blkz.cond == BLK_ZONE_COND_OFFLINE)
1253		set_bit(DMZ_OFFLINE, &zone->flags);
1254	else if (blkz.cond == BLK_ZONE_COND_READONLY)
1255		set_bit(DMZ_READ_ONLY, &zone->flags);
1256
1257	if (dmz_is_seq(zone))
1258		zone->wp_block = dmz_sect2blk(blkz.wp - blkz.start);
1259	else
1260		zone->wp_block = 0;
1261
1262	return 0;
1263}
1264
1265/*
1266 * Check a zone write pointer position when the zone is marked
1267 * with the sequential write error flag.
1268 */
1269static int dmz_handle_seq_write_err(struct dmz_metadata *zmd,
1270				    struct dm_zone *zone)
1271{
 
1272	unsigned int wp = 0;
1273	int ret;
1274
1275	wp = zone->wp_block;
1276	ret = dmz_update_zone(zmd, zone);
1277	if (ret)
1278		return ret;
1279
1280	dmz_dev_warn(zmd->dev, "Processing zone %u write error (zone wp %u/%u)",
1281		     dmz_id(zmd, zone), zone->wp_block, wp);
1282
1283	if (zone->wp_block < wp) {
1284		dmz_invalidate_blocks(zmd, zone, zone->wp_block,
1285				      wp - zone->wp_block);
1286	}
1287
1288	return 0;
1289}
1290
1291static struct dm_zone *dmz_get(struct dmz_metadata *zmd, unsigned int zone_id)
1292{
1293	return &zmd->zones[zone_id];
1294}
1295
1296/*
1297 * Reset a zone write pointer.
1298 */
1299static int dmz_reset_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1300{
1301	int ret;
1302
1303	/*
1304	 * Ignore offline zones, read only zones,
1305	 * and conventional zones.
1306	 */
1307	if (dmz_is_offline(zone) ||
1308	    dmz_is_readonly(zone) ||
1309	    dmz_is_rnd(zone))
1310		return 0;
1311
1312	if (!dmz_is_empty(zone) || dmz_seq_write_err(zone)) {
1313		struct dmz_dev *dev = zmd->dev;
1314
1315		ret = blkdev_reset_zones(dev->bdev,
1316					 dmz_start_sect(zmd, zone),
1317					 dev->zone_nr_sectors, GFP_NOIO);
1318		if (ret) {
1319			dmz_dev_err(dev, "Reset zone %u failed %d",
1320				    dmz_id(zmd, zone), ret);
1321			return ret;
1322		}
1323	}
1324
1325	/* Clear write error bit and rewind write pointer position */
1326	clear_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
1327	zone->wp_block = 0;
1328
1329	return 0;
1330}
1331
1332static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone);
1333
1334/*
1335 * Initialize chunk mapping.
1336 */
1337static int dmz_load_mapping(struct dmz_metadata *zmd)
1338{
1339	struct dmz_dev *dev = zmd->dev;
1340	struct dm_zone *dzone, *bzone;
1341	struct dmz_mblock *dmap_mblk = NULL;
1342	struct dmz_map *dmap;
1343	unsigned int i = 0, e = 0, chunk = 0;
1344	unsigned int dzone_id;
1345	unsigned int bzone_id;
1346
1347	/* Metadata block array for the chunk mapping table */
1348	zmd->map_mblk = kcalloc(zmd->nr_map_blocks,
1349				sizeof(struct dmz_mblk *), GFP_KERNEL);
1350	if (!zmd->map_mblk)
1351		return -ENOMEM;
1352
1353	/* Get chunk mapping table blocks and initialize zone mapping */
1354	while (chunk < zmd->nr_chunks) {
1355		if (!dmap_mblk) {
1356			/* Get mapping block */
1357			dmap_mblk = dmz_get_mblock(zmd, i + 1);
1358			if (IS_ERR(dmap_mblk))
1359				return PTR_ERR(dmap_mblk);
1360			zmd->map_mblk[i] = dmap_mblk;
1361			dmap = (struct dmz_map *) dmap_mblk->data;
1362			i++;
1363			e = 0;
1364		}
1365
1366		/* Check data zone */
1367		dzone_id = le32_to_cpu(dmap[e].dzone_id);
1368		if (dzone_id == DMZ_MAP_UNMAPPED)
1369			goto next;
1370
1371		if (dzone_id >= dev->nr_zones) {
1372			dmz_dev_err(dev, "Chunk %u mapping: invalid data zone ID %u",
1373				    chunk, dzone_id);
1374			return -EIO;
1375		}
1376
1377		dzone = dmz_get(zmd, dzone_id);
 
 
 
 
 
1378		set_bit(DMZ_DATA, &dzone->flags);
1379		dzone->chunk = chunk;
1380		dmz_get_zone_weight(zmd, dzone);
1381
1382		if (dmz_is_rnd(dzone))
1383			list_add_tail(&dzone->link, &zmd->map_rnd_list);
 
 
1384		else
1385			list_add_tail(&dzone->link, &zmd->map_seq_list);
1386
1387		/* Check buffer zone */
1388		bzone_id = le32_to_cpu(dmap[e].bzone_id);
1389		if (bzone_id == DMZ_MAP_UNMAPPED)
1390			goto next;
1391
1392		if (bzone_id >= dev->nr_zones) {
1393			dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone ID %u",
1394				    chunk, bzone_id);
1395			return -EIO;
1396		}
1397
1398		bzone = dmz_get(zmd, bzone_id);
1399		if (!dmz_is_rnd(bzone)) {
1400			dmz_dev_err(dev, "Chunk %u mapping: invalid buffer zone %u",
 
 
 
 
 
1401				    chunk, bzone_id);
1402			return -EIO;
1403		}
1404
1405		set_bit(DMZ_DATA, &bzone->flags);
1406		set_bit(DMZ_BUF, &bzone->flags);
1407		bzone->chunk = chunk;
1408		bzone->bzone = dzone;
1409		dzone->bzone = bzone;
1410		dmz_get_zone_weight(zmd, bzone);
1411		list_add_tail(&bzone->link, &zmd->map_rnd_list);
 
 
 
1412next:
1413		chunk++;
1414		e++;
1415		if (e >= DMZ_MAP_ENTRIES)
1416			dmap_mblk = NULL;
1417	}
1418
1419	/*
1420	 * At this point, only meta zones and mapped data zones were
1421	 * fully initialized. All remaining zones are unmapped data
1422	 * zones. Finish initializing those here.
1423	 */
1424	for (i = 0; i < dev->nr_zones; i++) {
1425		dzone = dmz_get(zmd, i);
 
 
1426		if (dmz_is_meta(dzone))
1427			continue;
 
 
1428
1429		if (dmz_is_rnd(dzone))
1430			zmd->nr_rnd++;
 
 
1431		else
1432			zmd->nr_seq++;
1433
1434		if (dmz_is_data(dzone)) {
1435			/* Already initialized */
1436			continue;
1437		}
1438
1439		/* Unmapped data zone */
1440		set_bit(DMZ_DATA, &dzone->flags);
1441		dzone->chunk = DMZ_MAP_UNMAPPED;
1442		if (dmz_is_rnd(dzone)) {
1443			list_add_tail(&dzone->link, &zmd->unmap_rnd_list);
1444			atomic_inc(&zmd->unmap_nr_rnd);
 
 
 
 
1445		} else if (atomic_read(&zmd->nr_reserved_seq_zones) < zmd->nr_reserved_seq) {
1446			list_add_tail(&dzone->link, &zmd->reserved_seq_zones_list);
 
1447			atomic_inc(&zmd->nr_reserved_seq_zones);
1448			zmd->nr_seq--;
1449		} else {
1450			list_add_tail(&dzone->link, &zmd->unmap_seq_list);
1451			atomic_inc(&zmd->unmap_nr_seq);
 
1452		}
1453	}
1454
1455	return 0;
1456}
1457
1458/*
1459 * Set a data chunk mapping.
1460 */
1461static void dmz_set_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk,
1462				  unsigned int dzone_id, unsigned int bzone_id)
1463{
1464	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1465	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1466	int map_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1467
1468	dmap[map_idx].dzone_id = cpu_to_le32(dzone_id);
1469	dmap[map_idx].bzone_id = cpu_to_le32(bzone_id);
1470	dmz_dirty_mblock(zmd, dmap_mblk);
1471}
1472
1473/*
1474 * The list of mapped zones is maintained in LRU order.
1475 * This rotates a zone at the end of its map list.
1476 */
1477static void __dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1478{
1479	if (list_empty(&zone->link))
1480		return;
1481
1482	list_del_init(&zone->link);
1483	if (dmz_is_seq(zone)) {
1484		/* LRU rotate sequential zone */
1485		list_add_tail(&zone->link, &zmd->map_seq_list);
 
 
 
1486	} else {
1487		/* LRU rotate random zone */
1488		list_add_tail(&zone->link, &zmd->map_rnd_list);
1489	}
1490}
1491
1492/*
1493 * The list of mapped random zones is maintained
1494 * in LRU order. This rotates a zone at the end of the list.
1495 */
1496static void dmz_lru_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1497{
1498	__dmz_lru_zone(zmd, zone);
1499	if (zone->bzone)
1500		__dmz_lru_zone(zmd, zone->bzone);
1501}
1502
1503/*
1504 * Wait for any zone to be freed.
1505 */
1506static void dmz_wait_for_free_zones(struct dmz_metadata *zmd)
1507{
1508	DEFINE_WAIT(wait);
1509
1510	prepare_to_wait(&zmd->free_wq, &wait, TASK_UNINTERRUPTIBLE);
1511	dmz_unlock_map(zmd);
1512	dmz_unlock_metadata(zmd);
1513
1514	io_schedule_timeout(HZ);
1515
1516	dmz_lock_metadata(zmd);
1517	dmz_lock_map(zmd);
1518	finish_wait(&zmd->free_wq, &wait);
1519}
1520
1521/*
1522 * Lock a zone for reclaim (set the zone RECLAIM bit).
1523 * Returns false if the zone cannot be locked or if it is already locked
1524 * and 1 otherwise.
1525 */
1526int dmz_lock_zone_reclaim(struct dm_zone *zone)
1527{
1528	/* Active zones cannot be reclaimed */
1529	if (dmz_is_active(zone))
1530		return 0;
1531
1532	return !test_and_set_bit(DMZ_RECLAIM, &zone->flags);
1533}
1534
1535/*
1536 * Clear a zone reclaim flag.
1537 */
1538void dmz_unlock_zone_reclaim(struct dm_zone *zone)
1539{
1540	WARN_ON(dmz_is_active(zone));
1541	WARN_ON(!dmz_in_reclaim(zone));
1542
1543	clear_bit_unlock(DMZ_RECLAIM, &zone->flags);
1544	smp_mb__after_atomic();
1545	wake_up_bit(&zone->flags, DMZ_RECLAIM);
1546}
1547
1548/*
1549 * Wait for a zone reclaim to complete.
1550 */
1551static void dmz_wait_for_reclaim(struct dmz_metadata *zmd, struct dm_zone *zone)
1552{
1553	dmz_unlock_map(zmd);
1554	dmz_unlock_metadata(zmd);
 
1555	wait_on_bit_timeout(&zone->flags, DMZ_RECLAIM, TASK_UNINTERRUPTIBLE, HZ);
 
1556	dmz_lock_metadata(zmd);
1557	dmz_lock_map(zmd);
1558}
1559
1560/*
1561 * Select a random write zone for reclaim.
1562 */
1563static struct dm_zone *dmz_get_rnd_zone_for_reclaim(struct dmz_metadata *zmd)
 
1564{
1565	struct dm_zone *dzone = NULL;
1566	struct dm_zone *zone;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1567
1568	if (list_empty(&zmd->map_rnd_list))
1569		return ERR_PTR(-EBUSY);
1570
1571	list_for_each_entry(zone, &zmd->map_rnd_list, link) {
1572		if (dmz_is_buf(zone))
 
 
 
 
 
1573			dzone = zone->bzone;
1574		else
 
 
1575			dzone = zone;
1576		if (dmz_lock_zone_reclaim(dzone))
1577			return dzone;
1578	}
1579
1580	return ERR_PTR(-EBUSY);
1581}
1582
1583/*
1584 * Select a buffered sequential zone for reclaim.
1585 */
1586static struct dm_zone *dmz_get_seq_zone_for_reclaim(struct dmz_metadata *zmd)
 
1587{
1588	struct dm_zone *zone;
1589
1590	if (list_empty(&zmd->map_seq_list))
1591		return ERR_PTR(-EBUSY);
1592
1593	list_for_each_entry(zone, &zmd->map_seq_list, link) {
1594		if (!zone->bzone)
1595			continue;
1596		if (dmz_lock_zone_reclaim(zone))
1597			return zone;
1598	}
1599
1600	return ERR_PTR(-EBUSY);
1601}
1602
1603/*
1604 * Select a zone for reclaim.
1605 */
1606struct dm_zone *dmz_get_zone_for_reclaim(struct dmz_metadata *zmd)
 
1607{
1608	struct dm_zone *zone;
1609
1610	/*
1611	 * Search for a zone candidate to reclaim: 2 cases are possible.
1612	 * (1) There is no free sequential zones. Then a random data zone
1613	 *     cannot be reclaimed. So choose a sequential zone to reclaim so
1614	 *     that afterward a random zone can be reclaimed.
1615	 * (2) At least one free sequential zone is available, then choose
1616	 *     the oldest random zone (data or buffer) that can be locked.
1617	 */
1618	dmz_lock_map(zmd);
1619	if (list_empty(&zmd->reserved_seq_zones_list))
1620		zone = dmz_get_seq_zone_for_reclaim(zmd);
1621	else
1622		zone = dmz_get_rnd_zone_for_reclaim(zmd);
1623	dmz_unlock_map(zmd);
1624
1625	return zone;
1626}
1627
1628/*
1629 * Get the zone mapping a chunk, if the chunk is mapped already.
1630 * If no mapping exist and the operation is WRITE, a zone is
1631 * allocated and used to map the chunk.
1632 * The zone returned will be set to the active state.
1633 */
1634struct dm_zone *dmz_get_chunk_mapping(struct dmz_metadata *zmd, unsigned int chunk, int op)
 
1635{
1636	struct dmz_mblock *dmap_mblk = zmd->map_mblk[chunk >> DMZ_MAP_ENTRIES_SHIFT];
1637	struct dmz_map *dmap = (struct dmz_map *) dmap_mblk->data;
1638	int dmap_idx = chunk & DMZ_MAP_ENTRIES_MASK;
1639	unsigned int dzone_id;
1640	struct dm_zone *dzone = NULL;
1641	int ret = 0;
 
1642
1643	dmz_lock_map(zmd);
1644again:
1645	/* Get the chunk mapping */
1646	dzone_id = le32_to_cpu(dmap[dmap_idx].dzone_id);
1647	if (dzone_id == DMZ_MAP_UNMAPPED) {
1648		/*
1649		 * Read or discard in unmapped chunks are fine. But for
1650		 * writes, we need a mapping, so get one.
1651		 */
1652		if (op != REQ_OP_WRITE)
1653			goto out;
1654
1655		/* Allocate a random zone */
1656		dzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1657		if (!dzone) {
1658			if (dmz_bdev_is_dying(zmd->dev)) {
1659				dzone = ERR_PTR(-EIO);
1660				goto out;
1661			}
1662			dmz_wait_for_free_zones(zmd);
1663			goto again;
1664		}
1665
1666		dmz_map_zone(zmd, dzone, chunk);
1667
1668	} else {
1669		/* The chunk is already mapped: get the mapping zone */
1670		dzone = dmz_get(zmd, dzone_id);
 
 
 
 
1671		if (dzone->chunk != chunk) {
1672			dzone = ERR_PTR(-EIO);
1673			goto out;
1674		}
1675
1676		/* Repair write pointer if the sequential dzone has error */
1677		if (dmz_seq_write_err(dzone)) {
1678			ret = dmz_handle_seq_write_err(zmd, dzone);
1679			if (ret) {
1680				dzone = ERR_PTR(-EIO);
1681				goto out;
1682			}
1683			clear_bit(DMZ_SEQ_WRITE_ERR, &dzone->flags);
1684		}
1685	}
1686
1687	/*
1688	 * If the zone is being reclaimed, the chunk mapping may change
1689	 * to a different zone. So wait for reclaim and retry. Otherwise,
1690	 * activate the zone (this will prevent reclaim from touching it).
1691	 */
1692	if (dmz_in_reclaim(dzone)) {
1693		dmz_wait_for_reclaim(zmd, dzone);
1694		goto again;
1695	}
1696	dmz_activate_zone(dzone);
1697	dmz_lru_zone(zmd, dzone);
1698out:
1699	dmz_unlock_map(zmd);
1700
1701	return dzone;
1702}
1703
1704/*
1705 * Write and discard change the block validity of data zones and their buffer
1706 * zones. Check here that valid blocks are still present. If all blocks are
1707 * invalid, the zones can be unmapped on the fly without waiting for reclaim
1708 * to do it.
1709 */
1710void dmz_put_chunk_mapping(struct dmz_metadata *zmd, struct dm_zone *dzone)
1711{
1712	struct dm_zone *bzone;
1713
1714	dmz_lock_map(zmd);
1715
1716	bzone = dzone->bzone;
1717	if (bzone) {
1718		if (dmz_weight(bzone))
1719			dmz_lru_zone(zmd, bzone);
1720		else {
1721			/* Empty buffer zone: reclaim it */
1722			dmz_unmap_zone(zmd, bzone);
1723			dmz_free_zone(zmd, bzone);
1724			bzone = NULL;
1725		}
1726	}
1727
1728	/* Deactivate the data zone */
1729	dmz_deactivate_zone(dzone);
1730	if (dmz_is_active(dzone) || bzone || dmz_weight(dzone))
1731		dmz_lru_zone(zmd, dzone);
1732	else {
1733		/* Unbuffered inactive empty data zone: reclaim it */
1734		dmz_unmap_zone(zmd, dzone);
1735		dmz_free_zone(zmd, dzone);
1736	}
1737
1738	dmz_unlock_map(zmd);
1739}
1740
1741/*
1742 * Allocate and map a random zone to buffer a chunk
1743 * already mapped to a sequential zone.
1744 */
1745struct dm_zone *dmz_get_chunk_buffer(struct dmz_metadata *zmd,
1746				     struct dm_zone *dzone)
1747{
1748	struct dm_zone *bzone;
 
1749
1750	dmz_lock_map(zmd);
1751again:
1752	bzone = dzone->bzone;
1753	if (bzone)
1754		goto out;
1755
1756	/* Allocate a random zone */
1757	bzone = dmz_alloc_zone(zmd, DMZ_ALLOC_RND);
1758	if (!bzone) {
1759		if (dmz_bdev_is_dying(zmd->dev)) {
1760			bzone = ERR_PTR(-EIO);
1761			goto out;
1762		}
1763		dmz_wait_for_free_zones(zmd);
1764		goto again;
1765	}
1766
1767	/* Update the chunk mapping */
1768	dmz_set_chunk_mapping(zmd, dzone->chunk, dmz_id(zmd, dzone),
1769			      dmz_id(zmd, bzone));
1770
1771	set_bit(DMZ_BUF, &bzone->flags);
1772	bzone->chunk = dzone->chunk;
1773	bzone->bzone = dzone;
1774	dzone->bzone = bzone;
1775	list_add_tail(&bzone->link, &zmd->map_rnd_list);
 
 
 
1776out:
1777	dmz_unlock_map(zmd);
1778
1779	return bzone;
1780}
1781
1782/*
1783 * Get an unmapped (free) zone.
1784 * This must be called with the mapping lock held.
1785 */
1786struct dm_zone *dmz_alloc_zone(struct dmz_metadata *zmd, unsigned long flags)
 
1787{
1788	struct list_head *list;
1789	struct dm_zone *zone;
 
 
 
 
 
 
 
1790
1791	if (flags & DMZ_ALLOC_RND)
1792		list = &zmd->unmap_rnd_list;
 
 
 
 
1793	else
1794		list = &zmd->unmap_seq_list;
1795again:
1796	if (list_empty(list)) {
1797		/*
1798		 * No free zone: if this is for reclaim, allow using the
1799		 * reserved sequential zones.
1800		 */
1801		if (!(flags & DMZ_ALLOC_RECLAIM) ||
1802		    list_empty(&zmd->reserved_seq_zones_list))
1803			return NULL;
 
 
 
 
 
 
 
 
1804
1805		zone = list_first_entry(&zmd->reserved_seq_zones_list,
1806					struct dm_zone, link);
1807		list_del_init(&zone->link);
1808		atomic_dec(&zmd->nr_reserved_seq_zones);
 
 
 
 
 
1809		return zone;
1810	}
1811
1812	zone = list_first_entry(list, struct dm_zone, link);
1813	list_del_init(&zone->link);
1814
1815	if (dmz_is_rnd(zone))
1816		atomic_dec(&zmd->unmap_nr_rnd);
 
 
1817	else
1818		atomic_dec(&zmd->unmap_nr_seq);
1819
1820	if (dmz_is_offline(zone)) {
1821		dmz_dev_warn(zmd->dev, "Zone %u is offline", dmz_id(zmd, zone));
 
 
 
 
 
1822		zone = NULL;
1823		goto again;
1824	}
1825
1826	return zone;
1827}
1828
1829/*
1830 * Free a zone.
1831 * This must be called with the mapping lock held.
1832 */
1833void dmz_free_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1834{
1835	/* If this is a sequential zone, reset it */
1836	if (dmz_is_seq(zone))
1837		dmz_reset_zone(zmd, zone);
1838
1839	/* Return the zone to its type unmap list */
1840	if (dmz_is_rnd(zone)) {
1841		list_add_tail(&zone->link, &zmd->unmap_rnd_list);
1842		atomic_inc(&zmd->unmap_nr_rnd);
1843	} else if (atomic_read(&zmd->nr_reserved_seq_zones) <
1844		   zmd->nr_reserved_seq) {
 
 
1845		list_add_tail(&zone->link, &zmd->reserved_seq_zones_list);
1846		atomic_inc(&zmd->nr_reserved_seq_zones);
1847	} else {
1848		list_add_tail(&zone->link, &zmd->unmap_seq_list);
1849		atomic_inc(&zmd->unmap_nr_seq);
1850	}
1851
1852	wake_up_all(&zmd->free_wq);
1853}
1854
1855/*
1856 * Map a chunk to a zone.
1857 * This must be called with the mapping lock held.
1858 */
1859void dmz_map_zone(struct dmz_metadata *zmd, struct dm_zone *dzone,
1860		  unsigned int chunk)
1861{
1862	/* Set the chunk mapping */
1863	dmz_set_chunk_mapping(zmd, chunk, dmz_id(zmd, dzone),
1864			      DMZ_MAP_UNMAPPED);
1865	dzone->chunk = chunk;
1866	if (dmz_is_rnd(dzone))
1867		list_add_tail(&dzone->link, &zmd->map_rnd_list);
 
 
1868	else
1869		list_add_tail(&dzone->link, &zmd->map_seq_list);
1870}
1871
1872/*
1873 * Unmap a zone.
1874 * This must be called with the mapping lock held.
1875 */
1876void dmz_unmap_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
1877{
1878	unsigned int chunk = zone->chunk;
1879	unsigned int dzone_id;
1880
1881	if (chunk == DMZ_MAP_UNMAPPED) {
1882		/* Already unmapped */
1883		return;
1884	}
1885
1886	if (test_and_clear_bit(DMZ_BUF, &zone->flags)) {
1887		/*
1888		 * Unmapping the chunk buffer zone: clear only
1889		 * the chunk buffer mapping
1890		 */
1891		dzone_id = dmz_id(zmd, zone->bzone);
1892		zone->bzone->bzone = NULL;
1893		zone->bzone = NULL;
1894
1895	} else {
1896		/*
1897		 * Unmapping the chunk data zone: the zone must
1898		 * not be buffered.
1899		 */
1900		if (WARN_ON(zone->bzone)) {
1901			zone->bzone->bzone = NULL;
1902			zone->bzone = NULL;
1903		}
1904		dzone_id = DMZ_MAP_UNMAPPED;
1905	}
1906
1907	dmz_set_chunk_mapping(zmd, chunk, dzone_id, DMZ_MAP_UNMAPPED);
1908
1909	zone->chunk = DMZ_MAP_UNMAPPED;
1910	list_del_init(&zone->link);
1911}
1912
1913/*
1914 * Set @nr_bits bits in @bitmap starting from @bit.
1915 * Return the number of bits changed from 0 to 1.
1916 */
1917static unsigned int dmz_set_bits(unsigned long *bitmap,
1918				 unsigned int bit, unsigned int nr_bits)
1919{
1920	unsigned long *addr;
1921	unsigned int end = bit + nr_bits;
1922	unsigned int n = 0;
1923
1924	while (bit < end) {
1925		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
1926		    ((end - bit) >= BITS_PER_LONG)) {
1927			/* Try to set the whole word at once */
1928			addr = bitmap + BIT_WORD(bit);
1929			if (*addr == 0) {
1930				*addr = ULONG_MAX;
1931				n += BITS_PER_LONG;
1932				bit += BITS_PER_LONG;
1933				continue;
1934			}
1935		}
1936
1937		if (!test_and_set_bit(bit, bitmap))
1938			n++;
1939		bit++;
1940	}
1941
1942	return n;
1943}
1944
1945/*
1946 * Get the bitmap block storing the bit for chunk_block in zone.
1947 */
1948static struct dmz_mblock *dmz_get_bitmap(struct dmz_metadata *zmd,
1949					 struct dm_zone *zone,
1950					 sector_t chunk_block)
1951{
1952	sector_t bitmap_block = 1 + zmd->nr_map_blocks +
1953		(sector_t)(dmz_id(zmd, zone) * zmd->zone_nr_bitmap_blocks) +
1954		(chunk_block >> DMZ_BLOCK_SHIFT_BITS);
1955
1956	return dmz_get_mblock(zmd, bitmap_block);
1957}
1958
1959/*
1960 * Copy the valid blocks bitmap of from_zone to the bitmap of to_zone.
1961 */
1962int dmz_copy_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1963			  struct dm_zone *to_zone)
1964{
1965	struct dmz_mblock *from_mblk, *to_mblk;
1966	sector_t chunk_block = 0;
1967
1968	/* Get the zones bitmap blocks */
1969	while (chunk_block < zmd->dev->zone_nr_blocks) {
1970		from_mblk = dmz_get_bitmap(zmd, from_zone, chunk_block);
1971		if (IS_ERR(from_mblk))
1972			return PTR_ERR(from_mblk);
1973		to_mblk = dmz_get_bitmap(zmd, to_zone, chunk_block);
1974		if (IS_ERR(to_mblk)) {
1975			dmz_release_mblock(zmd, from_mblk);
1976			return PTR_ERR(to_mblk);
1977		}
1978
1979		memcpy(to_mblk->data, from_mblk->data, DMZ_BLOCK_SIZE);
1980		dmz_dirty_mblock(zmd, to_mblk);
1981
1982		dmz_release_mblock(zmd, to_mblk);
1983		dmz_release_mblock(zmd, from_mblk);
1984
1985		chunk_block += DMZ_BLOCK_SIZE_BITS;
1986	}
1987
1988	to_zone->weight = from_zone->weight;
1989
1990	return 0;
1991}
1992
1993/*
1994 * Merge the valid blocks bitmap of from_zone into the bitmap of to_zone,
1995 * starting from chunk_block.
1996 */
1997int dmz_merge_valid_blocks(struct dmz_metadata *zmd, struct dm_zone *from_zone,
1998			   struct dm_zone *to_zone, sector_t chunk_block)
1999{
2000	unsigned int nr_blocks;
2001	int ret;
2002
2003	/* Get the zones bitmap blocks */
2004	while (chunk_block < zmd->dev->zone_nr_blocks) {
2005		/* Get a valid region from the source zone */
2006		ret = dmz_first_valid_block(zmd, from_zone, &chunk_block);
2007		if (ret <= 0)
2008			return ret;
2009
2010		nr_blocks = ret;
2011		ret = dmz_validate_blocks(zmd, to_zone, chunk_block, nr_blocks);
2012		if (ret)
2013			return ret;
2014
2015		chunk_block += nr_blocks;
2016	}
2017
2018	return 0;
2019}
2020
2021/*
2022 * Validate all the blocks in the range [block..block+nr_blocks-1].
2023 */
2024int dmz_validate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2025			sector_t chunk_block, unsigned int nr_blocks)
2026{
2027	unsigned int count, bit, nr_bits;
2028	unsigned int zone_nr_blocks = zmd->dev->zone_nr_blocks;
2029	struct dmz_mblock *mblk;
2030	unsigned int n = 0;
2031
2032	dmz_dev_debug(zmd->dev, "=> VALIDATE zone %u, block %llu, %u blocks",
2033		      dmz_id(zmd, zone), (unsigned long long)chunk_block,
2034		      nr_blocks);
2035
2036	WARN_ON(chunk_block + nr_blocks > zone_nr_blocks);
2037
2038	while (nr_blocks) {
2039		/* Get bitmap block */
2040		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2041		if (IS_ERR(mblk))
2042			return PTR_ERR(mblk);
2043
2044		/* Set bits */
2045		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2046		nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2047
2048		count = dmz_set_bits((unsigned long *)mblk->data, bit, nr_bits);
2049		if (count) {
2050			dmz_dirty_mblock(zmd, mblk);
2051			n += count;
2052		}
2053		dmz_release_mblock(zmd, mblk);
2054
2055		nr_blocks -= nr_bits;
2056		chunk_block += nr_bits;
2057	}
2058
2059	if (likely(zone->weight + n <= zone_nr_blocks))
2060		zone->weight += n;
2061	else {
2062		dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be <= %u",
2063			     dmz_id(zmd, zone), zone->weight,
2064			     zone_nr_blocks - n);
2065		zone->weight = zone_nr_blocks;
2066	}
2067
2068	return 0;
2069}
2070
2071/*
2072 * Clear nr_bits bits in bitmap starting from bit.
2073 * Return the number of bits cleared.
2074 */
2075static int dmz_clear_bits(unsigned long *bitmap, int bit, int nr_bits)
2076{
2077	unsigned long *addr;
2078	int end = bit + nr_bits;
2079	int n = 0;
2080
2081	while (bit < end) {
2082		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2083		    ((end - bit) >= BITS_PER_LONG)) {
2084			/* Try to clear whole word at once */
2085			addr = bitmap + BIT_WORD(bit);
2086			if (*addr == ULONG_MAX) {
2087				*addr = 0;
2088				n += BITS_PER_LONG;
2089				bit += BITS_PER_LONG;
2090				continue;
2091			}
2092		}
2093
2094		if (test_and_clear_bit(bit, bitmap))
2095			n++;
2096		bit++;
2097	}
2098
2099	return n;
2100}
2101
2102/*
2103 * Invalidate all the blocks in the range [block..block+nr_blocks-1].
2104 */
2105int dmz_invalidate_blocks(struct dmz_metadata *zmd, struct dm_zone *zone,
2106			  sector_t chunk_block, unsigned int nr_blocks)
2107{
2108	unsigned int count, bit, nr_bits;
2109	struct dmz_mblock *mblk;
2110	unsigned int n = 0;
2111
2112	dmz_dev_debug(zmd->dev, "=> INVALIDATE zone %u, block %llu, %u blocks",
2113		      dmz_id(zmd, zone), (u64)chunk_block, nr_blocks);
2114
2115	WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2116
2117	while (nr_blocks) {
2118		/* Get bitmap block */
2119		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2120		if (IS_ERR(mblk))
2121			return PTR_ERR(mblk);
2122
2123		/* Clear bits */
2124		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2125		nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2126
2127		count = dmz_clear_bits((unsigned long *)mblk->data,
2128				       bit, nr_bits);
2129		if (count) {
2130			dmz_dirty_mblock(zmd, mblk);
2131			n += count;
2132		}
2133		dmz_release_mblock(zmd, mblk);
2134
2135		nr_blocks -= nr_bits;
2136		chunk_block += nr_bits;
2137	}
2138
2139	if (zone->weight >= n)
2140		zone->weight -= n;
2141	else {
2142		dmz_dev_warn(zmd->dev, "Zone %u: weight %u should be >= %u",
2143			     dmz_id(zmd, zone), zone->weight, n);
2144		zone->weight = 0;
2145	}
2146
2147	return 0;
2148}
2149
2150/*
2151 * Get a block bit value.
2152 */
2153static int dmz_test_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2154			  sector_t chunk_block)
2155{
2156	struct dmz_mblock *mblk;
2157	int ret;
2158
2159	WARN_ON(chunk_block >= zmd->dev->zone_nr_blocks);
2160
2161	/* Get bitmap block */
2162	mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2163	if (IS_ERR(mblk))
2164		return PTR_ERR(mblk);
2165
2166	/* Get offset */
2167	ret = test_bit(chunk_block & DMZ_BLOCK_MASK_BITS,
2168		       (unsigned long *) mblk->data) != 0;
2169
2170	dmz_release_mblock(zmd, mblk);
2171
2172	return ret;
2173}
2174
2175/*
2176 * Return the number of blocks from chunk_block to the first block with a bit
2177 * value specified by set. Search at most nr_blocks blocks from chunk_block.
2178 */
2179static int dmz_to_next_set_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2180				 sector_t chunk_block, unsigned int nr_blocks,
2181				 int set)
2182{
2183	struct dmz_mblock *mblk;
2184	unsigned int bit, set_bit, nr_bits;
 
2185	unsigned long *bitmap;
2186	int n = 0;
2187
2188	WARN_ON(chunk_block + nr_blocks > zmd->dev->zone_nr_blocks);
2189
2190	while (nr_blocks) {
2191		/* Get bitmap block */
2192		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2193		if (IS_ERR(mblk))
2194			return PTR_ERR(mblk);
2195
2196		/* Get offset */
2197		bitmap = (unsigned long *) mblk->data;
2198		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2199		nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2200		if (set)
2201			set_bit = find_next_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2202		else
2203			set_bit = find_next_zero_bit(bitmap, DMZ_BLOCK_SIZE_BITS, bit);
2204		dmz_release_mblock(zmd, mblk);
2205
2206		n += set_bit - bit;
2207		if (set_bit < DMZ_BLOCK_SIZE_BITS)
2208			break;
2209
2210		nr_blocks -= nr_bits;
2211		chunk_block += nr_bits;
2212	}
2213
2214	return n;
2215}
2216
2217/*
2218 * Test if chunk_block is valid. If it is, the number of consecutive
2219 * valid blocks from chunk_block will be returned.
2220 */
2221int dmz_block_valid(struct dmz_metadata *zmd, struct dm_zone *zone,
2222		    sector_t chunk_block)
2223{
2224	int valid;
2225
2226	valid = dmz_test_block(zmd, zone, chunk_block);
2227	if (valid <= 0)
2228		return valid;
2229
2230	/* The block is valid: get the number of valid blocks from block */
2231	return dmz_to_next_set_block(zmd, zone, chunk_block,
2232				     zmd->dev->zone_nr_blocks - chunk_block, 0);
2233}
2234
2235/*
2236 * Find the first valid block from @chunk_block in @zone.
2237 * If such a block is found, its number is returned using
2238 * @chunk_block and the total number of valid blocks from @chunk_block
2239 * is returned.
2240 */
2241int dmz_first_valid_block(struct dmz_metadata *zmd, struct dm_zone *zone,
2242			  sector_t *chunk_block)
2243{
2244	sector_t start_block = *chunk_block;
2245	int ret;
2246
2247	ret = dmz_to_next_set_block(zmd, zone, start_block,
2248				    zmd->dev->zone_nr_blocks - start_block, 1);
2249	if (ret < 0)
2250		return ret;
2251
2252	start_block += ret;
2253	*chunk_block = start_block;
2254
2255	return dmz_to_next_set_block(zmd, zone, start_block,
2256				     zmd->dev->zone_nr_blocks - start_block, 0);
2257}
2258
2259/*
2260 * Count the number of bits set starting from bit up to bit + nr_bits - 1.
2261 */
2262static int dmz_count_bits(void *bitmap, int bit, int nr_bits)
2263{
2264	unsigned long *addr;
2265	int end = bit + nr_bits;
2266	int n = 0;
2267
2268	while (bit < end) {
2269		if (((bit & (BITS_PER_LONG - 1)) == 0) &&
2270		    ((end - bit) >= BITS_PER_LONG)) {
2271			addr = (unsigned long *)bitmap + BIT_WORD(bit);
2272			if (*addr == ULONG_MAX) {
2273				n += BITS_PER_LONG;
2274				bit += BITS_PER_LONG;
2275				continue;
2276			}
2277		}
2278
2279		if (test_bit(bit, bitmap))
2280			n++;
2281		bit++;
2282	}
2283
2284	return n;
2285}
2286
2287/*
2288 * Get a zone weight.
2289 */
2290static void dmz_get_zone_weight(struct dmz_metadata *zmd, struct dm_zone *zone)
2291{
2292	struct dmz_mblock *mblk;
2293	sector_t chunk_block = 0;
2294	unsigned int bit, nr_bits;
2295	unsigned int nr_blocks = zmd->dev->zone_nr_blocks;
2296	void *bitmap;
2297	int n = 0;
2298
2299	while (nr_blocks) {
2300		/* Get bitmap block */
2301		mblk = dmz_get_bitmap(zmd, zone, chunk_block);
2302		if (IS_ERR(mblk)) {
2303			n = 0;
2304			break;
2305		}
2306
2307		/* Count bits in this block */
2308		bitmap = mblk->data;
2309		bit = chunk_block & DMZ_BLOCK_MASK_BITS;
2310		nr_bits = min(nr_blocks, DMZ_BLOCK_SIZE_BITS - bit);
2311		n += dmz_count_bits(bitmap, bit, nr_bits);
2312
2313		dmz_release_mblock(zmd, mblk);
2314
2315		nr_blocks -= nr_bits;
2316		chunk_block += nr_bits;
2317	}
2318
2319	zone->weight = n;
2320}
2321
2322/*
2323 * Cleanup the zoned metadata resources.
2324 */
2325static void dmz_cleanup_metadata(struct dmz_metadata *zmd)
2326{
2327	struct rb_root *root;
2328	struct dmz_mblock *mblk, *next;
2329	int i;
2330
2331	/* Release zone mapping resources */
2332	if (zmd->map_mblk) {
2333		for (i = 0; i < zmd->nr_map_blocks; i++)
2334			dmz_release_mblock(zmd, zmd->map_mblk[i]);
2335		kfree(zmd->map_mblk);
2336		zmd->map_mblk = NULL;
2337	}
2338
2339	/* Release super blocks */
2340	for (i = 0; i < 2; i++) {
2341		if (zmd->sb[i].mblk) {
2342			dmz_free_mblock(zmd, zmd->sb[i].mblk);
2343			zmd->sb[i].mblk = NULL;
2344		}
2345	}
2346
2347	/* Free cached blocks */
2348	while (!list_empty(&zmd->mblk_dirty_list)) {
2349		mblk = list_first_entry(&zmd->mblk_dirty_list,
2350					struct dmz_mblock, link);
2351		dmz_dev_warn(zmd->dev, "mblock %llu still in dirty list (ref %u)",
2352			     (u64)mblk->no, mblk->ref);
2353		list_del_init(&mblk->link);
2354		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2355		dmz_free_mblock(zmd, mblk);
2356	}
2357
2358	while (!list_empty(&zmd->mblk_lru_list)) {
2359		mblk = list_first_entry(&zmd->mblk_lru_list,
2360					struct dmz_mblock, link);
2361		list_del_init(&mblk->link);
2362		rb_erase(&mblk->node, &zmd->mblk_rbtree);
2363		dmz_free_mblock(zmd, mblk);
2364	}
2365
2366	/* Sanity checks: the mblock rbtree should now be empty */
2367	root = &zmd->mblk_rbtree;
2368	rbtree_postorder_for_each_entry_safe(mblk, next, root, node) {
2369		dmz_dev_warn(zmd->dev, "mblock %llu ref %u still in rbtree",
2370			     (u64)mblk->no, mblk->ref);
2371		mblk->ref = 0;
2372		dmz_free_mblock(zmd, mblk);
2373	}
2374
2375	/* Free the zone descriptors */
2376	dmz_drop_zones(zmd);
2377
2378	mutex_destroy(&zmd->mblk_flush_lock);
2379	mutex_destroy(&zmd->map_lock);
2380}
2381
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2382/*
2383 * Initialize the zoned metadata.
2384 */
2385int dmz_ctr_metadata(struct dmz_dev *dev, struct dmz_metadata **metadata)
 
 
2386{
2387	struct dmz_metadata *zmd;
2388	unsigned int i, zid;
2389	struct dm_zone *zone;
2390	int ret;
2391
2392	zmd = kzalloc(sizeof(struct dmz_metadata), GFP_KERNEL);
2393	if (!zmd)
2394		return -ENOMEM;
2395
 
2396	zmd->dev = dev;
 
2397	zmd->mblk_rbtree = RB_ROOT;
2398	init_rwsem(&zmd->mblk_sem);
2399	mutex_init(&zmd->mblk_flush_lock);
2400	spin_lock_init(&zmd->mblk_lock);
2401	INIT_LIST_HEAD(&zmd->mblk_lru_list);
2402	INIT_LIST_HEAD(&zmd->mblk_dirty_list);
2403
2404	mutex_init(&zmd->map_lock);
2405	atomic_set(&zmd->unmap_nr_rnd, 0);
2406	INIT_LIST_HEAD(&zmd->unmap_rnd_list);
2407	INIT_LIST_HEAD(&zmd->map_rnd_list);
2408
2409	atomic_set(&zmd->unmap_nr_seq, 0);
2410	INIT_LIST_HEAD(&zmd->unmap_seq_list);
2411	INIT_LIST_HEAD(&zmd->map_seq_list);
2412
2413	atomic_set(&zmd->nr_reserved_seq_zones, 0);
2414	INIT_LIST_HEAD(&zmd->reserved_seq_zones_list);
2415
2416	init_waitqueue_head(&zmd->free_wq);
2417
2418	/* Initialize zone descriptors */
2419	ret = dmz_init_zones(zmd);
2420	if (ret)
2421		goto err;
2422
2423	/* Get super block */
2424	ret = dmz_load_sb(zmd);
2425	if (ret)
2426		goto err;
2427
2428	/* Set metadata zones starting from sb_zone */
2429	zid = dmz_id(zmd, zmd->sb_zone);
2430	for (i = 0; i < zmd->nr_meta_zones << 1; i++) {
2431		zone = dmz_get(zmd, zid + i);
2432		if (!dmz_is_rnd(zone))
 
 
 
 
 
 
 
 
 
2433			goto err;
 
2434		set_bit(DMZ_META, &zone->flags);
2435	}
2436
2437	/* Load mapping table */
2438	ret = dmz_load_mapping(zmd);
2439	if (ret)
2440		goto err;
2441
2442	/*
2443	 * Cache size boundaries: allow at least 2 super blocks, the chunk map
2444	 * blocks and enough blocks to be able to cache the bitmap blocks of
2445	 * up to 16 zones when idle (min_nr_mblks). Otherwise, if busy, allow
2446	 * the cache to add 512 more metadata blocks.
2447	 */
2448	zmd->min_nr_mblks = 2 + zmd->nr_map_blocks + zmd->zone_nr_bitmap_blocks * 16;
2449	zmd->max_nr_mblks = zmd->min_nr_mblks + 512;
2450	zmd->mblk_shrinker.count_objects = dmz_mblock_shrinker_count;
2451	zmd->mblk_shrinker.scan_objects = dmz_mblock_shrinker_scan;
2452	zmd->mblk_shrinker.seeks = DEFAULT_SEEKS;
2453
2454	/* Metadata cache shrinker */
2455	ret = register_shrinker(&zmd->mblk_shrinker);
 
 
2456	if (ret) {
2457		dmz_dev_err(dev, "Register metadata cache shrinker failed");
2458		goto err;
2459	}
2460
2461	dmz_dev_info(dev, "Host-%s zoned block device",
2462		     bdev_zoned_model(dev->bdev) == BLK_ZONED_HA ?
2463		     "aware" : "managed");
2464	dmz_dev_info(dev, "  %llu 512-byte logical sectors",
2465		     (u64)dev->capacity);
2466	dmz_dev_info(dev, "  %u zones of %llu 512-byte logical sectors",
2467		     dev->nr_zones, (u64)dev->zone_nr_sectors);
2468	dmz_dev_info(dev, "  %u metadata zones",
2469		     zmd->nr_meta_zones * 2);
2470	dmz_dev_info(dev, "  %u data zones for %u chunks",
2471		     zmd->nr_data_zones, zmd->nr_chunks);
2472	dmz_dev_info(dev, "    %u random zones (%u unmapped)",
2473		     zmd->nr_rnd, atomic_read(&zmd->unmap_nr_rnd));
2474	dmz_dev_info(dev, "    %u sequential zones (%u unmapped)",
2475		     zmd->nr_seq, atomic_read(&zmd->unmap_nr_seq));
2476	dmz_dev_info(dev, "  %u reserved sequential data zones",
2477		     zmd->nr_reserved_seq);
2478
2479	dmz_dev_debug(dev, "Format:");
2480	dmz_dev_debug(dev, "%u metadata blocks per set (%u max cache)",
 
 
 
 
2481		      zmd->nr_meta_blocks, zmd->max_nr_mblks);
2482	dmz_dev_debug(dev, "  %u data zone mapping blocks",
2483		      zmd->nr_map_blocks);
2484	dmz_dev_debug(dev, "  %u bitmap blocks",
2485		      zmd->nr_bitmap_blocks);
2486
2487	*metadata = zmd;
2488
2489	return 0;
2490err:
2491	dmz_cleanup_metadata(zmd);
2492	kfree(zmd);
2493	*metadata = NULL;
2494
2495	return ret;
2496}
2497
2498/*
2499 * Cleanup the zoned metadata resources.
2500 */
2501void dmz_dtr_metadata(struct dmz_metadata *zmd)
2502{
2503	unregister_shrinker(&zmd->mblk_shrinker);
2504	dmz_cleanup_metadata(zmd);
2505	kfree(zmd);
2506}
2507
2508/*
2509 * Check zone information on resume.
2510 */
2511int dmz_resume_metadata(struct dmz_metadata *zmd)
2512{
2513	struct dmz_dev *dev = zmd->dev;
2514	struct dm_zone *zone;
2515	sector_t wp_block;
2516	unsigned int i;
2517	int ret;
2518
2519	/* Check zones */
2520	for (i = 0; i < dev->nr_zones; i++) {
2521		zone = dmz_get(zmd, i);
2522		if (!zone) {
2523			dmz_dev_err(dev, "Unable to get zone %u", i);
2524			return -EIO;
2525		}
2526
2527		wp_block = zone->wp_block;
2528
2529		ret = dmz_update_zone(zmd, zone);
2530		if (ret) {
2531			dmz_dev_err(dev, "Broken zone %u", i);
2532			return ret;
2533		}
2534
2535		if (dmz_is_offline(zone)) {
2536			dmz_dev_warn(dev, "Zone %u is offline", i);
2537			continue;
2538		}
2539
2540		/* Check write pointer */
2541		if (!dmz_is_seq(zone))
2542			zone->wp_block = 0;
2543		else if (zone->wp_block != wp_block) {
2544			dmz_dev_err(dev, "Zone %u: Invalid wp (%llu / %llu)",
2545				    i, (u64)zone->wp_block, (u64)wp_block);
2546			zone->wp_block = wp_block;
2547			dmz_invalidate_blocks(zmd, zone, zone->wp_block,
2548					      dev->zone_nr_blocks - zone->wp_block);
2549		}
2550	}
2551
2552	return 0;
2553}