1/*
   2 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   3 * Copyright (C) 2016-2017 Milan Broz
   4 * Copyright (C) 2016-2017 Mikulas Patocka
   5 *
   6 * This file is released under the GPL.
   7 */
   8
   9#include <linux/compiler.h>
  10#include <linux/module.h>
  11#include <linux/device-mapper.h>
  12#include <linux/dm-io.h>
  13#include <linux/vmalloc.h>
  14#include <linux/sort.h>
  15#include <linux/rbtree.h>
  16#include <linux/delay.h>
  17#include <linux/random.h>
  18#include <linux/reboot.h>
  19#include <crypto/hash.h>
  20#include <crypto/skcipher.h>
  21#include <linux/async_tx.h>
  22#include <linux/dm-bufio.h>
  23
  24#define DM_MSG_PREFIX "integrity"
  25
  26#define DEFAULT_INTERLEAVE_SECTORS	32768
  27#define DEFAULT_JOURNAL_SIZE_FACTOR	7
  28#define DEFAULT_SECTORS_PER_BITMAP_BIT	32768
  29#define DEFAULT_BUFFER_SECTORS		128
  30#define DEFAULT_JOURNAL_WATERMARK	50
  31#define DEFAULT_SYNC_MSEC		10000
  32#define DEFAULT_MAX_JOURNAL_SECTORS	131072
  33#define MIN_LOG2_INTERLEAVE_SECTORS	3
  34#define MAX_LOG2_INTERLEAVE_SECTORS	31
  35#define METADATA_WORKQUEUE_MAX_ACTIVE	16
  36#define RECALC_SECTORS			8192
  37#define RECALC_WRITE_SUPER		16
  38#define BITMAP_BLOCK_SIZE		4096	/* don't change it */
  39#define BITMAP_FLUSH_INTERVAL		(10 * HZ)
  40
  41/*
  42 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  43 * so it should not be enabled in the official kernel
  44 */
  45//#define DEBUG_PRINT
  46//#define INTERNAL_VERIFY
  47
  48/*
  49 * On disk structures
  50 */
  51
  52#define SB_MAGIC			"integrt"
  53#define SB_VERSION_1			1
  54#define SB_VERSION_2			2
  55#define SB_VERSION_3			3
  56#define SB_SECTORS			8
  57#define MAX_SECTORS_PER_BLOCK		8
  58
  59struct superblock {
  60	__u8 magic[8];
  61	__u8 version;
  62	__u8 log2_interleave_sectors;
  63	__u16 integrity_tag_size;
  64	__u32 journal_sections;
  65	__u64 provided_data_sectors;	/* userspace uses this value */
  66	__u32 flags;
  67	__u8 log2_sectors_per_block;
  68	__u8 log2_blocks_per_bitmap_bit;
  69	__u8 pad[2];
  70	__u64 recalc_sector;
  71};
  72
  73#define SB_FLAG_HAVE_JOURNAL_MAC	0x1
  74#define SB_FLAG_RECALCULATING		0x2
  75#define SB_FLAG_DIRTY_BITMAP		0x4
  76
  77#define	JOURNAL_ENTRY_ROUNDUP		8
  78
  79typedef __u64 commit_id_t;
  80#define JOURNAL_MAC_PER_SECTOR		8
  81
  82struct journal_entry {
  83	union {
  84		struct {
  85			__u32 sector_lo;
  86			__u32 sector_hi;
  87		} s;
  88		__u64 sector;
  89	} u;
  90	commit_id_t last_bytes[0];
  91	/* __u8 tag[0]; */
  92};
  93
  94#define journal_entry_tag(ic, je)		((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  95
  96#if BITS_PER_LONG == 64
  97#define journal_entry_set_sector(je, x)		do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
  98#else
  99#define journal_entry_set_sector(je, x)		do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
 100#endif
 101#define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
 102#define journal_entry_is_unused(je)		((je)->u.s.sector_hi == cpu_to_le32(-1))
 103#define journal_entry_set_unused(je)		do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 104#define journal_entry_is_inprogress(je)		((je)->u.s.sector_hi == cpu_to_le32(-2))
 105#define journal_entry_set_inprogress(je)	do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 106
 107#define JOURNAL_BLOCK_SECTORS		8
 108#define JOURNAL_SECTOR_DATA		((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 109#define JOURNAL_MAC_SIZE		(JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 110
 111struct journal_sector {
 112	__u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 113	__u8 mac[JOURNAL_MAC_PER_SECTOR];
 114	commit_id_t commit_id;
 115};
 116
 117#define MAX_TAG_SIZE			(JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 118
 119#define METADATA_PADDING_SECTORS	8
 120
 121#define N_COMMIT_IDS			4
 122
 123static unsigned char prev_commit_seq(unsigned char seq)
 124{
 125	return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 126}
 127
 128static unsigned char next_commit_seq(unsigned char seq)
 129{
 130	return (seq + 1) % N_COMMIT_IDS;
 131}
 132
 133/*
 134 * In-memory structures
 135 */
 136
 137struct journal_node {
 138	struct rb_node node;
 139	sector_t sector;
 140};
 141
 142struct alg_spec {
 143	char *alg_string;
 144	char *key_string;
 145	__u8 *key;
 146	unsigned key_size;
 147};
 148
 149struct dm_integrity_c {
 150	struct dm_dev *dev;
 151	struct dm_dev *meta_dev;
 152	unsigned tag_size;
 153	__s8 log2_tag_size;
 154	sector_t start;
 155	mempool_t journal_io_mempool;
 156	struct dm_io_client *io;
 157	struct dm_bufio_client *bufio;
 158	struct workqueue_struct *metadata_wq;
 159	struct superblock *sb;
 160	unsigned journal_pages;
 161	unsigned n_bitmap_blocks;
 162
 163	struct page_list *journal;
 164	struct page_list *journal_io;
 165	struct page_list *journal_xor;
 166	struct page_list *recalc_bitmap;
 167	struct page_list *may_write_bitmap;
 168	struct bitmap_block_status *bbs;
 169	unsigned bitmap_flush_interval;
 170	int synchronous_mode;
 171	struct bio_list synchronous_bios;
 172	struct delayed_work bitmap_flush_work;
 173
 174	struct crypto_skcipher *journal_crypt;
 175	struct scatterlist **journal_scatterlist;
 176	struct scatterlist **journal_io_scatterlist;
 177	struct skcipher_request **sk_requests;
 178
 179	struct crypto_shash *journal_mac;
 180
 181	struct journal_node *journal_tree;
 182	struct rb_root journal_tree_root;
 183
 184	sector_t provided_data_sectors;
 185
 186	unsigned short journal_entry_size;
 187	unsigned char journal_entries_per_sector;
 188	unsigned char journal_section_entries;
 189	unsigned short journal_section_sectors;
 190	unsigned journal_sections;
 191	unsigned journal_entries;
 192	sector_t data_device_sectors;
 193	sector_t meta_device_sectors;
 194	unsigned initial_sectors;
 195	unsigned metadata_run;
 196	__s8 log2_metadata_run;
 197	__u8 log2_buffer_sectors;
 198	__u8 sectors_per_block;
 199	__u8 log2_blocks_per_bitmap_bit;
 200
 201	unsigned char mode;
 202	int suspending;
 203
 204	int failed;
 205
 206	struct crypto_shash *internal_hash;
 207
 208	/* these variables are locked with endio_wait.lock */
 209	struct rb_root in_progress;
 210	struct list_head wait_list;
 211	wait_queue_head_t endio_wait;
 212	struct workqueue_struct *wait_wq;
 213
 214	unsigned char commit_seq;
 215	commit_id_t commit_ids[N_COMMIT_IDS];
 216
 217	unsigned committed_section;
 218	unsigned n_committed_sections;
 219
 220	unsigned uncommitted_section;
 221	unsigned n_uncommitted_sections;
 222
 223	unsigned free_section;
 224	unsigned char free_section_entry;
 225	unsigned free_sectors;
 226
 227	unsigned free_sectors_threshold;
 228
 229	struct workqueue_struct *commit_wq;
 230	struct work_struct commit_work;
 231
 232	struct workqueue_struct *writer_wq;
 233	struct work_struct writer_work;
 234
 235	struct workqueue_struct *recalc_wq;
 236	struct work_struct recalc_work;
 237	u8 *recalc_buffer;
 238	u8 *recalc_tags;
 239
 240	struct bio_list flush_bio_list;
 241
 242	unsigned long autocommit_jiffies;
 243	struct timer_list autocommit_timer;
 244	unsigned autocommit_msec;
 245
 246	wait_queue_head_t copy_to_journal_wait;
 247
 248	struct completion crypto_backoff;
 249
 250	bool journal_uptodate;
 251	bool just_formatted;
 252	bool recalculate_flag;
 253
 254	struct alg_spec internal_hash_alg;
 255	struct alg_spec journal_crypt_alg;
 256	struct alg_spec journal_mac_alg;
 257
 258	atomic64_t number_of_mismatches;
 259
 260	struct notifier_block reboot_notifier;
 261};
 262
 263struct dm_integrity_range {
 264	sector_t logical_sector;
 265	sector_t n_sectors;
 266	bool waiting;
 267	union {
 268		struct rb_node node;
 269		struct {
 270			struct task_struct *task;
 271			struct list_head wait_entry;
 272		};
 273	};
 274};
 275
 276struct dm_integrity_io {
 277	struct work_struct work;
 278
 279	struct dm_integrity_c *ic;
 280	bool write;
 281	bool fua;
 282
 283	struct dm_integrity_range range;
 284
 285	sector_t metadata_block;
 286	unsigned metadata_offset;
 287
 288	atomic_t in_flight;
 289	blk_status_t bi_status;
 290
 291	struct completion *completion;
 292
 293	struct gendisk *orig_bi_disk;
 294	u8 orig_bi_partno;
 295	bio_end_io_t *orig_bi_end_io;
 296	struct bio_integrity_payload *orig_bi_integrity;
 297	struct bvec_iter orig_bi_iter;
 298};
 299
 300struct journal_completion {
 301	struct dm_integrity_c *ic;
 302	atomic_t in_flight;
 303	struct completion comp;
 304};
 305
 306struct journal_io {
 307	struct dm_integrity_range range;
 308	struct journal_completion *comp;
 309};
 310
 311struct bitmap_block_status {
 312	struct work_struct work;
 313	struct dm_integrity_c *ic;
 314	unsigned idx;
 315	unsigned long *bitmap;
 316	struct bio_list bio_queue;
 317	spinlock_t bio_queue_lock;
 318
 319};
 320
 321static struct kmem_cache *journal_io_cache;
 322
 323#define JOURNAL_IO_MEMPOOL	32
 324
 325#ifdef DEBUG_PRINT
 326#define DEBUG_print(x, ...)	printk(KERN_DEBUG x, ##__VA_ARGS__)
 327static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 328{
 329	va_list args;
 330	va_start(args, msg);
 331	vprintk(msg, args);
 332	va_end(args);
 333	if (len)
 334		pr_cont(":");
 335	while (len) {
 336		pr_cont(" %02x", *bytes);
 337		bytes++;
 338		len--;
 339	}
 340	pr_cont("\n");
 341}
 342#define DEBUG_bytes(bytes, len, msg, ...)	__DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 343#else
 344#define DEBUG_print(x, ...)			do { } while (0)
 345#define DEBUG_bytes(bytes, len, msg, ...)	do { } while (0)
 346#endif
 347
 348static void dm_integrity_prepare(struct request *rq)
 349{
 350}
 351
 352static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 353{
 354}
 355
 356/*
 357 * DM Integrity profile, protection is performed layer above (dm-crypt)
 358 */
 359static const struct blk_integrity_profile dm_integrity_profile = {
 360	.name			= "DM-DIF-EXT-TAG",
 361	.generate_fn		= NULL,
 362	.verify_fn		= NULL,
 363	.prepare_fn		= dm_integrity_prepare,
 364	.complete_fn		= dm_integrity_complete,
 365};
 366
 367static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 368static void integrity_bio_wait(struct work_struct *w);
 369static void dm_integrity_dtr(struct dm_target *ti);
 370
 371static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 372{
 373	if (err == -EILSEQ)
 374		atomic64_inc(&ic->number_of_mismatches);
 375	if (!cmpxchg(&ic->failed, 0, err))
 376		DMERR("Error on %s: %d", msg, err);
 377}
 378
 379static int dm_integrity_failed(struct dm_integrity_c *ic)
 380{
 381	return READ_ONCE(ic->failed);
 382}
 383
 384static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 385					  unsigned j, unsigned char seq)
 386{
 387	/*
 388	 * Xor the number with section and sector, so that if a piece of
 389	 * journal is written at wrong place, it is detected.
 390	 */
 391	return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 392}
 393
 394static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 395				sector_t *area, sector_t *offset)
 396{
 397	if (!ic->meta_dev) {
 398		__u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 399		*area = data_sector >> log2_interleave_sectors;
 400		*offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 401	} else {
 402		*area = 0;
 403		*offset = data_sector;
 404	}
 405}
 406
 407#define sector_to_block(ic, n)						\
 408do {									\
 409	BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));		\
 410	(n) >>= (ic)->sb->log2_sectors_per_block;			\
 411} while (0)
 412
 413static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 414					    sector_t offset, unsigned *metadata_offset)
 415{
 416	__u64 ms;
 417	unsigned mo;
 418
 419	ms = area << ic->sb->log2_interleave_sectors;
 420	if (likely(ic->log2_metadata_run >= 0))
 421		ms += area << ic->log2_metadata_run;
 422	else
 423		ms += area * ic->metadata_run;
 424	ms >>= ic->log2_buffer_sectors;
 425
 426	sector_to_block(ic, offset);
 427
 428	if (likely(ic->log2_tag_size >= 0)) {
 429		ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 430		mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 431	} else {
 432		ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 433		mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 434	}
 435	*metadata_offset = mo;
 436	return ms;
 437}
 438
 439static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 440{
 441	sector_t result;
 442
 443	if (ic->meta_dev)
 444		return offset;
 445
 446	result = area << ic->sb->log2_interleave_sectors;
 447	if (likely(ic->log2_metadata_run >= 0))
 448		result += (area + 1) << ic->log2_metadata_run;
 449	else
 450		result += (area + 1) * ic->metadata_run;
 451
 452	result += (sector_t)ic->initial_sectors + offset;
 453	result += ic->start;
 454
 455	return result;
 456}
 457
 458static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 459{
 460	if (unlikely(*sec_ptr >= ic->journal_sections))
 461		*sec_ptr -= ic->journal_sections;
 462}
 463
 464static void sb_set_version(struct dm_integrity_c *ic)
 465{
 466	if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 467		ic->sb->version = SB_VERSION_3;
 468	else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 469		ic->sb->version = SB_VERSION_2;
 470	else
 471		ic->sb->version = SB_VERSION_1;
 472}
 473
 474static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 475{
 476	struct dm_io_request io_req;
 477	struct dm_io_region io_loc;
 478
 479	io_req.bi_op = op;
 480	io_req.bi_op_flags = op_flags;
 481	io_req.mem.type = DM_IO_KMEM;
 482	io_req.mem.ptr.addr = ic->sb;
 483	io_req.notify.fn = NULL;
 484	io_req.client = ic->io;
 485	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 486	io_loc.sector = ic->start;
 487	io_loc.count = SB_SECTORS;
 488
 489	if (op == REQ_OP_WRITE)
 490		sb_set_version(ic);
 491
 492	return dm_io(&io_req, 1, &io_loc, NULL);
 493}
 494
 495#define BITMAP_OP_TEST_ALL_SET		0
 496#define BITMAP_OP_TEST_ALL_CLEAR	1
 497#define BITMAP_OP_SET			2
 498#define BITMAP_OP_CLEAR			3
 499
 500static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 501			    sector_t sector, sector_t n_sectors, int mode)
 502{
 503	unsigned long bit, end_bit, this_end_bit, page, end_page;
 504	unsigned long *data;
 505
 506	if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 507		DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 508			(unsigned long long)sector,
 509			(unsigned long long)n_sectors,
 510			ic->sb->log2_sectors_per_block,
 511			ic->log2_blocks_per_bitmap_bit,
 512			mode);
 513		BUG();
 514	}
 515
 516	if (unlikely(!n_sectors))
 517		return true;
 518
 519	bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 520	end_bit = (sector + n_sectors - 1) >>
 521		(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 522
 523	page = bit / (PAGE_SIZE * 8);
 524	bit %= PAGE_SIZE * 8;
 525
 526	end_page = end_bit / (PAGE_SIZE * 8);
 527	end_bit %= PAGE_SIZE * 8;
 528
 529repeat:
 530	if (page < end_page) {
 531		this_end_bit = PAGE_SIZE * 8 - 1;
 532	} else {
 533		this_end_bit = end_bit;
 534	}
 535
 536	data = lowmem_page_address(bitmap[page].page);
 537
 538	if (mode == BITMAP_OP_TEST_ALL_SET) {
 539		while (bit <= this_end_bit) {
 540			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 541				do {
 542					if (data[bit / BITS_PER_LONG] != -1)
 543						return false;
 544					bit += BITS_PER_LONG;
 545				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 546				continue;
 547			}
 548			if (!test_bit(bit, data))
 549				return false;
 550			bit++;
 551		}
 552	} else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 553		while (bit <= this_end_bit) {
 554			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 555				do {
 556					if (data[bit / BITS_PER_LONG] != 0)
 557						return false;
 558					bit += BITS_PER_LONG;
 559				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 560				continue;
 561			}
 562			if (test_bit(bit, data))
 563				return false;
 564			bit++;
 565		}
 566	} else if (mode == BITMAP_OP_SET) {
 567		while (bit <= this_end_bit) {
 568			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 569				do {
 570					data[bit / BITS_PER_LONG] = -1;
 571					bit += BITS_PER_LONG;
 572				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 573				continue;
 574			}
 575			__set_bit(bit, data);
 576			bit++;
 577		}
 578	} else if (mode == BITMAP_OP_CLEAR) {
 579		if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 580			clear_page(data);
 581		else while (bit <= this_end_bit) {
 582			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 583				do {
 584					data[bit / BITS_PER_LONG] = 0;
 585					bit += BITS_PER_LONG;
 586				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 587				continue;
 588			}
 589			__clear_bit(bit, data);
 590			bit++;
 591		}
 592	} else {
 593		BUG();
 594	}
 595
 596	if (unlikely(page < end_page)) {
 597		bit = 0;
 598		page++;
 599		goto repeat;
 600	}
 601
 602	return true;
 603}
 604
 605static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 606{
 607	unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 608	unsigned i;
 609
 610	for (i = 0; i < n_bitmap_pages; i++) {
 611		unsigned long *dst_data = lowmem_page_address(dst[i].page);
 612		unsigned long *src_data = lowmem_page_address(src[i].page);
 613		copy_page(dst_data, src_data);
 614	}
 615}
 616
 617static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 618{
 619	unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 620	unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 621
 622	BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 623	return &ic->bbs[bitmap_block];
 624}
 625
 626static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 627				 bool e, const char *function)
 628{
 629#if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 630	unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 631
 632	if (unlikely(section >= ic->journal_sections) ||
 633	    unlikely(offset >= limit)) {
 634		DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 635		       function, section, offset, ic->journal_sections, limit);
 636		BUG();
 637	}
 638#endif
 639}
 640
 641static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 642			       unsigned *pl_index, unsigned *pl_offset)
 643{
 644	unsigned sector;
 645
 646	access_journal_check(ic, section, offset, false, "page_list_location");
 647
 648	sector = section * ic->journal_section_sectors + offset;
 649
 650	*pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 651	*pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 652}
 653
 654static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 655					       unsigned section, unsigned offset, unsigned *n_sectors)
 656{
 657	unsigned pl_index, pl_offset;
 658	char *va;
 659
 660	page_list_location(ic, section, offset, &pl_index, &pl_offset);
 661
 662	if (n_sectors)
 663		*n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 664
 665	va = lowmem_page_address(pl[pl_index].page);
 666
 667	return (struct journal_sector *)(va + pl_offset);
 668}
 669
 670static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 671{
 672	return access_page_list(ic, ic->journal, section, offset, NULL);
 673}
 674
 675static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 676{
 677	unsigned rel_sector, offset;
 678	struct journal_sector *js;
 679
 680	access_journal_check(ic, section, n, true, "access_journal_entry");
 681
 682	rel_sector = n % JOURNAL_BLOCK_SECTORS;
 683	offset = n / JOURNAL_BLOCK_SECTORS;
 684
 685	js = access_journal(ic, section, rel_sector);
 686	return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 687}
 688
 689static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 690{
 691	n <<= ic->sb->log2_sectors_per_block;
 692
 693	n += JOURNAL_BLOCK_SECTORS;
 694
 695	access_journal_check(ic, section, n, false, "access_journal_data");
 696
 697	return access_journal(ic, section, n);
 698}
 699
 700static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 701{
 702	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 703	int r;
 704	unsigned j, size;
 705
 706	desc->tfm = ic->journal_mac;
 707
 708	r = crypto_shash_init(desc);
 709	if (unlikely(r)) {
 710		dm_integrity_io_error(ic, "crypto_shash_init", r);
 711		goto err;
 712	}
 713
 714	for (j = 0; j < ic->journal_section_entries; j++) {
 715		struct journal_entry *je = access_journal_entry(ic, section, j);
 716		r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 717		if (unlikely(r)) {
 718			dm_integrity_io_error(ic, "crypto_shash_update", r);
 719			goto err;
 720		}
 721	}
 722
 723	size = crypto_shash_digestsize(ic->journal_mac);
 724
 725	if (likely(size <= JOURNAL_MAC_SIZE)) {
 726		r = crypto_shash_final(desc, result);
 727		if (unlikely(r)) {
 728			dm_integrity_io_error(ic, "crypto_shash_final", r);
 729			goto err;
 730		}
 731		memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 732	} else {
 733		__u8 digest[HASH_MAX_DIGESTSIZE];
 734
 735		if (WARN_ON(size > sizeof(digest))) {
 736			dm_integrity_io_error(ic, "digest_size", -EINVAL);
 737			goto err;
 738		}
 739		r = crypto_shash_final(desc, digest);
 740		if (unlikely(r)) {
 741			dm_integrity_io_error(ic, "crypto_shash_final", r);
 742			goto err;
 743		}
 744		memcpy(result, digest, JOURNAL_MAC_SIZE);
 745	}
 746
 747	return;
 748err:
 749	memset(result, 0, JOURNAL_MAC_SIZE);
 750}
 751
 752static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 753{
 754	__u8 result[JOURNAL_MAC_SIZE];
 755	unsigned j;
 756
 757	if (!ic->journal_mac)
 758		return;
 759
 760	section_mac(ic, section, result);
 761
 762	for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 763		struct journal_sector *js = access_journal(ic, section, j);
 764
 765		if (likely(wr))
 766			memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 767		else {
 768			if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 769				dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 770		}
 771	}
 772}
 773
 774static void complete_journal_op(void *context)
 775{
 776	struct journal_completion *comp = context;
 777	BUG_ON(!atomic_read(&comp->in_flight));
 778	if (likely(atomic_dec_and_test(&comp->in_flight)))
 779		complete(&comp->comp);
 780}
 781
 782static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 783			unsigned n_sections, struct journal_completion *comp)
 784{
 785	struct async_submit_ctl submit;
 786	size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 787	unsigned pl_index, pl_offset, section_index;
 788	struct page_list *source_pl, *target_pl;
 789
 790	if (likely(encrypt)) {
 791		source_pl = ic->journal;
 792		target_pl = ic->journal_io;
 793	} else {
 794		source_pl = ic->journal_io;
 795		target_pl = ic->journal;
 796	}
 797
 798	page_list_location(ic, section, 0, &pl_index, &pl_offset);
 799
 800	atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 801
 802	init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 803
 804	section_index = pl_index;
 805
 806	do {
 807		size_t this_step;
 808		struct page *src_pages[2];
 809		struct page *dst_page;
 810
 811		while (unlikely(pl_index == section_index)) {
 812			unsigned dummy;
 813			if (likely(encrypt))
 814				rw_section_mac(ic, section, true);
 815			section++;
 816			n_sections--;
 817			if (!n_sections)
 818				break;
 819			page_list_location(ic, section, 0, &section_index, &dummy);
 820		}
 821
 822		this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 823		dst_page = target_pl[pl_index].page;
 824		src_pages[0] = source_pl[pl_index].page;
 825		src_pages[1] = ic->journal_xor[pl_index].page;
 826
 827		async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 828
 829		pl_index++;
 830		pl_offset = 0;
 831		n_bytes -= this_step;
 832	} while (n_bytes);
 833
 834	BUG_ON(n_sections);
 835
 836	async_tx_issue_pending_all();
 837}
 838
 839static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 840{
 841	struct journal_completion *comp = req->data;
 842	if (unlikely(err)) {
 843		if (likely(err == -EINPROGRESS)) {
 844			complete(&comp->ic->crypto_backoff);
 845			return;
 846		}
 847		dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 848	}
 849	complete_journal_op(comp);
 850}
 851
 852static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 853{
 854	int r;
 855	skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 856				      complete_journal_encrypt, comp);
 857	if (likely(encrypt))
 858		r = crypto_skcipher_encrypt(req);
 859	else
 860		r = crypto_skcipher_decrypt(req);
 861	if (likely(!r))
 862		return false;
 863	if (likely(r == -EINPROGRESS))
 864		return true;
 865	if (likely(r == -EBUSY)) {
 866		wait_for_completion(&comp->ic->crypto_backoff);
 867		reinit_completion(&comp->ic->crypto_backoff);
 868		return true;
 869	}
 870	dm_integrity_io_error(comp->ic, "encrypt", r);
 871	return false;
 872}
 873
 874static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 875			  unsigned n_sections, struct journal_completion *comp)
 876{
 877	struct scatterlist **source_sg;
 878	struct scatterlist **target_sg;
 879
 880	atomic_add(2, &comp->in_flight);
 881
 882	if (likely(encrypt)) {
 883		source_sg = ic->journal_scatterlist;
 884		target_sg = ic->journal_io_scatterlist;
 885	} else {
 886		source_sg = ic->journal_io_scatterlist;
 887		target_sg = ic->journal_scatterlist;
 888	}
 889
 890	do {
 891		struct skcipher_request *req;
 892		unsigned ivsize;
 893		char *iv;
 894
 895		if (likely(encrypt))
 896			rw_section_mac(ic, section, true);
 897
 898		req = ic->sk_requests[section];
 899		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 900		iv = req->iv;
 901
 902		memcpy(iv, iv + ivsize, ivsize);
 903
 904		req->src = source_sg[section];
 905		req->dst = target_sg[section];
 906
 907		if (unlikely(do_crypt(encrypt, req, comp)))
 908			atomic_inc(&comp->in_flight);
 909
 910		section++;
 911		n_sections--;
 912	} while (n_sections);
 913
 914	atomic_dec(&comp->in_flight);
 915	complete_journal_op(comp);
 916}
 917
 918static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 919			    unsigned n_sections, struct journal_completion *comp)
 920{
 921	if (ic->journal_xor)
 922		return xor_journal(ic, encrypt, section, n_sections, comp);
 923	else
 924		return crypt_journal(ic, encrypt, section, n_sections, comp);
 925}
 926
 927static void complete_journal_io(unsigned long error, void *context)
 928{
 929	struct journal_completion *comp = context;
 930	if (unlikely(error != 0))
 931		dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 932	complete_journal_op(comp);
 933}
 934
 935static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 936			       unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 937{
 938	struct dm_io_request io_req;
 939	struct dm_io_region io_loc;
 940	unsigned pl_index, pl_offset;
 941	int r;
 942
 943	if (unlikely(dm_integrity_failed(ic))) {
 944		if (comp)
 945			complete_journal_io(-1UL, comp);
 946		return;
 947	}
 948
 949	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 950	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 951
 952	io_req.bi_op = op;
 953	io_req.bi_op_flags = op_flags;
 954	io_req.mem.type = DM_IO_PAGE_LIST;
 955	if (ic->journal_io)
 956		io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 957	else
 958		io_req.mem.ptr.pl = &ic->journal[pl_index];
 959	io_req.mem.offset = pl_offset;
 960	if (likely(comp != NULL)) {
 961		io_req.notify.fn = complete_journal_io;
 962		io_req.notify.context = comp;
 963	} else {
 964		io_req.notify.fn = NULL;
 965	}
 966	io_req.client = ic->io;
 967	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 968	io_loc.sector = ic->start + SB_SECTORS + sector;
 969	io_loc.count = n_sectors;
 970
 971	r = dm_io(&io_req, 1, &io_loc, NULL);
 972	if (unlikely(r)) {
 973		dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 974		if (comp) {
 975			WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 976			complete_journal_io(-1UL, comp);
 977		}
 978	}
 979}
 980
 981static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 982		       unsigned n_sections, struct journal_completion *comp)
 983{
 984	unsigned sector, n_sectors;
 985
 986	sector = section * ic->journal_section_sectors;
 987	n_sectors = n_sections * ic->journal_section_sectors;
 988
 989	rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
 990}
 991
 992static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 993{
 994	struct journal_completion io_comp;
 995	struct journal_completion crypt_comp_1;
 996	struct journal_completion crypt_comp_2;
 997	unsigned i;
 998
 999	io_comp.ic = ic;
1000	init_completion(&io_comp.comp);
1001
1002	if (commit_start + commit_sections <= ic->journal_sections) {
1003		io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1004		if (ic->journal_io) {
1005			crypt_comp_1.ic = ic;
1006			init_completion(&crypt_comp_1.comp);
1007			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1008			encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1009			wait_for_completion_io(&crypt_comp_1.comp);
1010		} else {
1011			for (i = 0; i < commit_sections; i++)
1012				rw_section_mac(ic, commit_start + i, true);
1013		}
1014		rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1015			   commit_sections, &io_comp);
1016	} else {
1017		unsigned to_end;
1018		io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1019		to_end = ic->journal_sections - commit_start;
1020		if (ic->journal_io) {
1021			crypt_comp_1.ic = ic;
1022			init_completion(&crypt_comp_1.comp);
1023			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1024			encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1025			if (try_wait_for_completion(&crypt_comp_1.comp)) {
1026				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1027				reinit_completion(&crypt_comp_1.comp);
1028				crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1029				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1030				wait_for_completion_io(&crypt_comp_1.comp);
1031			} else {
1032				crypt_comp_2.ic = ic;
1033				init_completion(&crypt_comp_2.comp);
1034				crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1035				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1036				wait_for_completion_io(&crypt_comp_1.comp);
1037				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1038				wait_for_completion_io(&crypt_comp_2.comp);
1039			}
1040		} else {
1041			for (i = 0; i < to_end; i++)
1042				rw_section_mac(ic, commit_start + i, true);
1043			rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1044			for (i = 0; i < commit_sections - to_end; i++)
1045				rw_section_mac(ic, i, true);
1046		}
1047		rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1048	}
1049
1050	wait_for_completion_io(&io_comp.comp);
1051}
1052
1053static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1054			      unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1055{
1056	struct dm_io_request io_req;
1057	struct dm_io_region io_loc;
1058	int r;
1059	unsigned sector, pl_index, pl_offset;
1060
1061	BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1062
1063	if (unlikely(dm_integrity_failed(ic))) {
1064		fn(-1UL, data);
1065		return;
1066	}
1067
1068	sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1069
1070	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1071	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1072
1073	io_req.bi_op = REQ_OP_WRITE;
1074	io_req.bi_op_flags = 0;
1075	io_req.mem.type = DM_IO_PAGE_LIST;
1076	io_req.mem.ptr.pl = &ic->journal[pl_index];
1077	io_req.mem.offset = pl_offset;
1078	io_req.notify.fn = fn;
1079	io_req.notify.context = data;
1080	io_req.client = ic->io;
1081	io_loc.bdev = ic->dev->bdev;
1082	io_loc.sector = target;
1083	io_loc.count = n_sectors;
1084
1085	r = dm_io(&io_req, 1, &io_loc, NULL);
1086	if (unlikely(r)) {
1087		WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1088		fn(-1UL, data);
1089	}
1090}
1091
1092static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1093{
1094	return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1095	       range1->logical_sector + range1->n_sectors > range2->logical_sector;
1096}
1097
1098static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1099{
1100	struct rb_node **n = &ic->in_progress.rb_node;
1101	struct rb_node *parent;
1102
1103	BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1104
1105	if (likely(check_waiting)) {
1106		struct dm_integrity_range *range;
1107		list_for_each_entry(range, &ic->wait_list, wait_entry) {
1108			if (unlikely(ranges_overlap(range, new_range)))
1109				return false;
1110		}
1111	}
1112
1113	parent = NULL;
1114
1115	while (*n) {
1116		struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1117
1118		parent = *n;
1119		if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1120			n = &range->node.rb_left;
1121		} else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1122			n = &range->node.rb_right;
1123		} else {
1124			return false;
1125		}
1126	}
1127
1128	rb_link_node(&new_range->node, parent, n);
1129	rb_insert_color(&new_range->node, &ic->in_progress);
1130
1131	return true;
1132}
1133
1134static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1135{
1136	rb_erase(&range->node, &ic->in_progress);
1137	while (unlikely(!list_empty(&ic->wait_list))) {
1138		struct dm_integrity_range *last_range =
1139			list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1140		struct task_struct *last_range_task;
1141		last_range_task = last_range->task;
1142		list_del(&last_range->wait_entry);
1143		if (!add_new_range(ic, last_range, false)) {
1144			last_range->task = last_range_task;
1145			list_add(&last_range->wait_entry, &ic->wait_list);
1146			break;
1147		}
1148		last_range->waiting = false;
1149		wake_up_process(last_range_task);
1150	}
1151}
1152
1153static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1154{
1155	unsigned long flags;
1156
1157	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1158	remove_range_unlocked(ic, range);
1159	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1160}
1161
1162static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1163{
1164	new_range->waiting = true;
1165	list_add_tail(&new_range->wait_entry, &ic->wait_list);
1166	new_range->task = current;
1167	do {
1168		__set_current_state(TASK_UNINTERRUPTIBLE);
1169		spin_unlock_irq(&ic->endio_wait.lock);
1170		io_schedule();
1171		spin_lock_irq(&ic->endio_wait.lock);
1172	} while (unlikely(new_range->waiting));
1173}
1174
1175static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1176{
1177	if (unlikely(!add_new_range(ic, new_range, true)))
1178		wait_and_add_new_range(ic, new_range);
1179}
1180
1181static void init_journal_node(struct journal_node *node)
1182{
1183	RB_CLEAR_NODE(&node->node);
1184	node->sector = (sector_t)-1;
1185}
1186
1187static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1188{
1189	struct rb_node **link;
1190	struct rb_node *parent;
1191
1192	node->sector = sector;
1193	BUG_ON(!RB_EMPTY_NODE(&node->node));
1194
1195	link = &ic->journal_tree_root.rb_node;
1196	parent = NULL;
1197
1198	while (*link) {
1199		struct journal_node *j;
1200		parent = *link;
1201		j = container_of(parent, struct journal_node, node);
1202		if (sector < j->sector)
1203			link = &j->node.rb_left;
1204		else
1205			link = &j->node.rb_right;
1206	}
1207
1208	rb_link_node(&node->node, parent, link);
1209	rb_insert_color(&node->node, &ic->journal_tree_root);
1210}
1211
1212static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1213{
1214	BUG_ON(RB_EMPTY_NODE(&node->node));
1215	rb_erase(&node->node, &ic->journal_tree_root);
1216	init_journal_node(node);
1217}
1218
1219#define NOT_FOUND	(-1U)
1220
1221static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1222{
1223	struct rb_node *n = ic->journal_tree_root.rb_node;
1224	unsigned found = NOT_FOUND;
1225	*next_sector = (sector_t)-1;
1226	while (n) {
1227		struct journal_node *j = container_of(n, struct journal_node, node);
1228		if (sector == j->sector) {
1229			found = j - ic->journal_tree;
1230		}
1231		if (sector < j->sector) {
1232			*next_sector = j->sector;
1233			n = j->node.rb_left;
1234		} else {
1235			n = j->node.rb_right;
1236		}
1237	}
1238
1239	return found;
1240}
1241
1242static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1243{
1244	struct journal_node *node, *next_node;
1245	struct rb_node *next;
1246
1247	if (unlikely(pos >= ic->journal_entries))
1248		return false;
1249	node = &ic->journal_tree[pos];
1250	if (unlikely(RB_EMPTY_NODE(&node->node)))
1251		return false;
1252	if (unlikely(node->sector != sector))
1253		return false;
1254
1255	next = rb_next(&node->node);
1256	if (unlikely(!next))
1257		return true;
1258
1259	next_node = container_of(next, struct journal_node, node);
1260	return next_node->sector != sector;
1261}
1262
1263static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1264{
1265	struct rb_node *next;
1266	struct journal_node *next_node;
1267	unsigned next_section;
1268
1269	BUG_ON(RB_EMPTY_NODE(&node->node));
1270
1271	next = rb_next(&node->node);
1272	if (unlikely(!next))
1273		return false;
1274
1275	next_node = container_of(next, struct journal_node, node);
1276
1277	if (next_node->sector != node->sector)
1278		return false;
1279
1280	next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1281	if (next_section >= ic->committed_section &&
1282	    next_section < ic->committed_section + ic->n_committed_sections)
1283		return true;
1284	if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1285		return true;
1286
1287	return false;
1288}
1289
1290#define TAG_READ	0
1291#define TAG_WRITE	1
1292#define TAG_CMP		2
1293
1294static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1295			       unsigned *metadata_offset, unsigned total_size, int op)
1296{
1297	do {
1298		unsigned char *data, *dp;
1299		struct dm_buffer *b;
1300		unsigned to_copy;
1301		int r;
1302
1303		r = dm_integrity_failed(ic);
1304		if (unlikely(r))
1305			return r;
1306
1307		data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1308		if (IS_ERR(data))
1309			return PTR_ERR(data);
1310
1311		to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1312		dp = data + *metadata_offset;
1313		if (op == TAG_READ) {
1314			memcpy(tag, dp, to_copy);
1315		} else if (op == TAG_WRITE) {
1316			memcpy(dp, tag, to_copy);
1317			dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1318		} else  {
1319			/* e.g.: op == TAG_CMP */
1320			if (unlikely(memcmp(dp, tag, to_copy))) {
1321				unsigned i;
1322
1323				for (i = 0; i < to_copy; i++) {
1324					if (dp[i] != tag[i])
1325						break;
1326					total_size--;
1327				}
1328				dm_bufio_release(b);
1329				return total_size;
1330			}
1331		}
1332		dm_bufio_release(b);
1333
1334		tag += to_copy;
1335		*metadata_offset += to_copy;
1336		if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1337			(*metadata_block)++;
1338			*metadata_offset = 0;
1339		}
1340		total_size -= to_copy;
1341	} while (unlikely(total_size));
1342
1343	return 0;
1344}
1345
1346static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1347{
1348	int r;
1349	r = dm_bufio_write_dirty_buffers(ic->bufio);
1350	if (unlikely(r))
1351		dm_integrity_io_error(ic, "writing tags", r);
1352}
1353
1354static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1355{
1356	DECLARE_WAITQUEUE(wait, current);
1357	__add_wait_queue(&ic->endio_wait, &wait);
1358	__set_current_state(TASK_UNINTERRUPTIBLE);
1359	spin_unlock_irq(&ic->endio_wait.lock);
1360	io_schedule();
1361	spin_lock_irq(&ic->endio_wait.lock);
1362	__remove_wait_queue(&ic->endio_wait, &wait);
1363}
1364
1365static void autocommit_fn(struct timer_list *t)
1366{
1367	struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1368
1369	if (likely(!dm_integrity_failed(ic)))
1370		queue_work(ic->commit_wq, &ic->commit_work);
1371}
1372
1373static void schedule_autocommit(struct dm_integrity_c *ic)
1374{
1375	if (!timer_pending(&ic->autocommit_timer))
1376		mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1377}
1378
1379static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1380{
1381	struct bio *bio;
1382	unsigned long flags;
1383
1384	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1385	bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1386	bio_list_add(&ic->flush_bio_list, bio);
1387	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1388
1389	queue_work(ic->commit_wq, &ic->commit_work);
1390}
1391
1392static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1393{
1394	int r = dm_integrity_failed(ic);
1395	if (unlikely(r) && !bio->bi_status)
1396		bio->bi_status = errno_to_blk_status(r);
1397	if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1398		unsigned long flags;
1399		spin_lock_irqsave(&ic->endio_wait.lock, flags);
1400		bio_list_add(&ic->synchronous_bios, bio);
1401		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1402		spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1403		return;
1404	}
1405	bio_endio(bio);
1406}
1407
1408static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1409{
1410	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1411
1412	if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1413		submit_flush_bio(ic, dio);
1414	else
1415		do_endio(ic, bio);
1416}
1417
1418static void dec_in_flight(struct dm_integrity_io *dio)
1419{
1420	if (atomic_dec_and_test(&dio->in_flight)) {
1421		struct dm_integrity_c *ic = dio->ic;
1422		struct bio *bio;
1423
1424		remove_range(ic, &dio->range);
1425
1426		if (unlikely(dio->write))
1427			schedule_autocommit(ic);
1428
1429		bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1430
1431		if (unlikely(dio->bi_status) && !bio->bi_status)
1432			bio->bi_status = dio->bi_status;
1433		if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1434			dio->range.logical_sector += dio->range.n_sectors;
1435			bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1436			INIT_WORK(&dio->work, integrity_bio_wait);
1437			queue_work(ic->wait_wq, &dio->work);
1438			return;
1439		}
1440		do_endio_flush(ic, dio);
1441	}
1442}
1443
1444static void integrity_end_io(struct bio *bio)
1445{
1446	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1447
1448	bio->bi_iter = dio->orig_bi_iter;
1449	bio->bi_disk = dio->orig_bi_disk;
1450	bio->bi_partno = dio->orig_bi_partno;
1451	if (dio->orig_bi_integrity) {
1452		bio->bi_integrity = dio->orig_bi_integrity;
1453		bio->bi_opf |= REQ_INTEGRITY;
1454	}
1455	bio->bi_end_io = dio->orig_bi_end_io;
1456
1457	if (dio->completion)
1458		complete(dio->completion);
1459
1460	dec_in_flight(dio);
1461}
1462
1463static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1464				      const char *data, char *result)
1465{
1466	__u64 sector_le = cpu_to_le64(sector);
1467	SHASH_DESC_ON_STACK(req, ic->internal_hash);
1468	int r;
1469	unsigned digest_size;
1470
1471	req->tfm = ic->internal_hash;
1472
1473	r = crypto_shash_init(req);
1474	if (unlikely(r < 0)) {
1475		dm_integrity_io_error(ic, "crypto_shash_init", r);
1476		goto failed;
1477	}
1478
1479	r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1480	if (unlikely(r < 0)) {
1481		dm_integrity_io_error(ic, "crypto_shash_update", r);
1482		goto failed;
1483	}
1484
1485	r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1486	if (unlikely(r < 0)) {
1487		dm_integrity_io_error(ic, "crypto_shash_update", r);
1488		goto failed;
1489	}
1490
1491	r = crypto_shash_final(req, result);
1492	if (unlikely(r < 0)) {
1493		dm_integrity_io_error(ic, "crypto_shash_final", r);
1494		goto failed;
1495	}
1496
1497	digest_size = crypto_shash_digestsize(ic->internal_hash);
1498	if (unlikely(digest_size < ic->tag_size))
1499		memset(result + digest_size, 0, ic->tag_size - digest_size);
1500
1501	return;
1502
1503failed:
1504	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1505	get_random_bytes(result, ic->tag_size);
1506}
1507
1508static void integrity_metadata(struct work_struct *w)
1509{
1510	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1511	struct dm_integrity_c *ic = dio->ic;
1512
1513	int r;
1514
1515	if (ic->internal_hash) {
1516		struct bvec_iter iter;
1517		struct bio_vec bv;
1518		unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1519		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1520		char *checksums;
1521		unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1522		char checksums_onstack[HASH_MAX_DIGESTSIZE];
1523		unsigned sectors_to_process = dio->range.n_sectors;
1524		sector_t sector = dio->range.logical_sector;
1525
1526		if (unlikely(ic->mode == 'R'))
1527			goto skip_io;
1528
1529		checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1530				    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1531		if (!checksums) {
1532			checksums = checksums_onstack;
1533			if (WARN_ON(extra_space &&
1534				    digest_size > sizeof(checksums_onstack))) {
1535				r = -EINVAL;
1536				goto error;
1537			}
1538		}
1539
1540		__bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1541			unsigned pos;
1542			char *mem, *checksums_ptr;
1543
1544again:
1545			mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1546			pos = 0;
1547			checksums_ptr = checksums;
1548			do {
1549				integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1550				checksums_ptr += ic->tag_size;
1551				sectors_to_process -= ic->sectors_per_block;
1552				pos += ic->sectors_per_block << SECTOR_SHIFT;
1553				sector += ic->sectors_per_block;
1554			} while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1555			kunmap_atomic(mem);
1556
1557			r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1558						checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1559			if (unlikely(r)) {
1560				if (r > 0) {
1561					DMERR_LIMIT("Checksum failed at sector 0x%llx",
1562						    (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1563					r = -EILSEQ;
1564					atomic64_inc(&ic->number_of_mismatches);
1565				}
1566				if (likely(checksums != checksums_onstack))
1567					kfree(checksums);
1568				goto error;
1569			}
1570
1571			if (!sectors_to_process)
1572				break;
1573
1574			if (unlikely(pos < bv.bv_len)) {
1575				bv.bv_offset += pos;
1576				bv.bv_len -= pos;
1577				goto again;
1578			}
1579		}
1580
1581		if (likely(checksums != checksums_onstack))
1582			kfree(checksums);
1583	} else {
1584		struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1585
1586		if (bip) {
1587			struct bio_vec biv;
1588			struct bvec_iter iter;
1589			unsigned data_to_process = dio->range.n_sectors;
1590			sector_to_block(ic, data_to_process);
1591			data_to_process *= ic->tag_size;
1592
1593			bip_for_each_vec(biv, bip, iter) {
1594				unsigned char *tag;
1595				unsigned this_len;
1596
1597				BUG_ON(PageHighMem(biv.bv_page));
1598				tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1599				this_len = min(biv.bv_len, data_to_process);
1600				r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1601							this_len, !dio->write ? TAG_READ : TAG_WRITE);
1602				if (unlikely(r))
1603					goto error;
1604				data_to_process -= this_len;
1605				if (!data_to_process)
1606					break;
1607			}
1608		}
1609	}
1610skip_io:
1611	dec_in_flight(dio);
1612	return;
1613error:
1614	dio->bi_status = errno_to_blk_status(r);
1615	dec_in_flight(dio);
1616}
1617
1618static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1619{
1620	struct dm_integrity_c *ic = ti->private;
1621	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1622	struct bio_integrity_payload *bip;
1623
1624	sector_t area, offset;
1625
1626	dio->ic = ic;
1627	dio->bi_status = 0;
1628
1629	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1630		submit_flush_bio(ic, dio);
1631		return DM_MAPIO_SUBMITTED;
1632	}
1633
1634	dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1635	dio->write = bio_op(bio) == REQ_OP_WRITE;
1636	dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1637	if (unlikely(dio->fua)) {
1638		/*
1639		 * Don't pass down the FUA flag because we have to flush
1640		 * disk cache anyway.
1641		 */
1642		bio->bi_opf &= ~REQ_FUA;
1643	}
1644	if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1645		DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1646		      (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1647		      (unsigned long long)ic->provided_data_sectors);
1648		return DM_MAPIO_KILL;
1649	}
1650	if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1651		DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1652		      ic->sectors_per_block,
1653		      (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1654		return DM_MAPIO_KILL;
1655	}
1656
1657	if (ic->sectors_per_block > 1) {
1658		struct bvec_iter iter;
1659		struct bio_vec bv;
1660		bio_for_each_segment(bv, bio, iter) {
1661			if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1662				DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1663					bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1664				return DM_MAPIO_KILL;
1665			}
1666		}
1667	}
1668
1669	bip = bio_integrity(bio);
1670	if (!ic->internal_hash) {
1671		if (bip) {
1672			unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1673			if (ic->log2_tag_size >= 0)
1674				wanted_tag_size <<= ic->log2_tag_size;
1675			else
1676				wanted_tag_size *= ic->tag_size;
1677			if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1678				DMERR("Invalid integrity data size %u, expected %u",
1679				      bip->bip_iter.bi_size, wanted_tag_size);
1680				return DM_MAPIO_KILL;
1681			}
1682		}
1683	} else {
1684		if (unlikely(bip != NULL)) {
1685			DMERR("Unexpected integrity data when using internal hash");
1686			return DM_MAPIO_KILL;
1687		}
1688	}
1689
1690	if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1691		return DM_MAPIO_KILL;
1692
1693	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1694	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1695	bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1696
1697	dm_integrity_map_continue(dio, true);
1698	return DM_MAPIO_SUBMITTED;
1699}
1700
1701static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1702				 unsigned journal_section, unsigned journal_entry)
1703{
1704	struct dm_integrity_c *ic = dio->ic;
1705	sector_t logical_sector;
1706	unsigned n_sectors;
1707
1708	logical_sector = dio->range.logical_sector;
1709	n_sectors = dio->range.n_sectors;
1710	do {
1711		struct bio_vec bv = bio_iovec(bio);
1712		char *mem;
1713
1714		if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1715			bv.bv_len = n_sectors << SECTOR_SHIFT;
1716		n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1717		bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1718retry_kmap:
1719		mem = kmap_atomic(bv.bv_page);
1720		if (likely(dio->write))
1721			flush_dcache_page(bv.bv_page);
1722
1723		do {
1724			struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1725
1726			if (unlikely(!dio->write)) {
1727				struct journal_sector *js;
1728				char *mem_ptr;
1729				unsigned s;
1730
1731				if (unlikely(journal_entry_is_inprogress(je))) {
1732					flush_dcache_page(bv.bv_page);
1733					kunmap_atomic(mem);
1734
1735					__io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1736					goto retry_kmap;
1737				}
1738				smp_rmb();
1739				BUG_ON(journal_entry_get_sector(je) != logical_sector);
1740				js = access_journal_data(ic, journal_section, journal_entry);
1741				mem_ptr = mem + bv.bv_offset;
1742				s = 0;
1743				do {
1744					memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1745					*(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1746					js++;
1747					mem_ptr += 1 << SECTOR_SHIFT;
1748				} while (++s < ic->sectors_per_block);
1749#ifdef INTERNAL_VERIFY
1750				if (ic->internal_hash) {
1751					char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1752
1753					integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1754					if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1755						DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1756							    (unsigned long long)logical_sector);
1757					}
1758				}
1759#endif
1760			}
1761
1762			if (!ic->internal_hash) {
1763				struct bio_integrity_payload *bip = bio_integrity(bio);
1764				unsigned tag_todo = ic->tag_size;
1765				char *tag_ptr = journal_entry_tag(ic, je);
1766
1767				if (bip) do {
1768					struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1769					unsigned tag_now = min(biv.bv_len, tag_todo);
1770					char *tag_addr;
1771					BUG_ON(PageHighMem(biv.bv_page));
1772					tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1773					if (likely(dio->write))
1774						memcpy(tag_ptr, tag_addr, tag_now);
1775					else
1776						memcpy(tag_addr, tag_ptr, tag_now);
1777					bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1778					tag_ptr += tag_now;
1779					tag_todo -= tag_now;
1780				} while (unlikely(tag_todo)); else {
1781					if (likely(dio->write))
1782						memset(tag_ptr, 0, tag_todo);
1783				}
1784			}
1785
1786			if (likely(dio->write)) {
1787				struct journal_sector *js;
1788				unsigned s;
1789
1790				js = access_journal_data(ic, journal_section, journal_entry);
1791				memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1792
1793				s = 0;
1794				do {
1795					je->last_bytes[s] = js[s].commit_id;
1796				} while (++s < ic->sectors_per_block);
1797
1798				if (ic->internal_hash) {
1799					unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1800					if (unlikely(digest_size > ic->tag_size)) {
1801						char checksums_onstack[HASH_MAX_DIGESTSIZE];
1802						integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1803						memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1804					} else
1805						integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1806				}
1807
1808				journal_entry_set_sector(je, logical_sector);
1809			}
1810			logical_sector += ic->sectors_per_block;
1811
1812			journal_entry++;
1813			if (unlikely(journal_entry == ic->journal_section_entries)) {
1814				journal_entry = 0;
1815				journal_section++;
1816				wraparound_section(ic, &journal_section);
1817			}
1818
1819			bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1820		} while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1821
1822		if (unlikely(!dio->write))
1823			flush_dcache_page(bv.bv_page);
1824		kunmap_atomic(mem);
1825	} while (n_sectors);
1826
1827	if (likely(dio->write)) {
1828		smp_mb();
1829		if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1830			wake_up(&ic->copy_to_journal_wait);
1831		if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1832			queue_work(ic->commit_wq, &ic->commit_work);
1833		} else {
1834			schedule_autocommit(ic);
1835		}
1836	} else {
1837		remove_range(ic, &dio->range);
1838	}
1839
1840	if (unlikely(bio->bi_iter.bi_size)) {
1841		sector_t area, offset;
1842
1843		dio->range.logical_sector = logical_sector;
1844		get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1845		dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1846		return true;
1847	}
1848
1849	return false;
1850}
1851
1852static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1853{
1854	struct dm_integrity_c *ic = dio->ic;
1855	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1856	unsigned journal_section, journal_entry;
1857	unsigned journal_read_pos;
1858	struct completion read_comp;
1859	bool need_sync_io = ic->internal_hash && !dio->write;
1860
1861	if (need_sync_io && from_map) {
1862		INIT_WORK(&dio->work, integrity_bio_wait);
1863		queue_work(ic->metadata_wq, &dio->work);
1864		return;
1865	}
1866
1867lock_retry:
1868	spin_lock_irq(&ic->endio_wait.lock);
1869retry:
1870	if (unlikely(dm_integrity_failed(ic))) {
1871		spin_unlock_irq(&ic->endio_wait.lock);
1872		do_endio(ic, bio);
1873		return;
1874	}
1875	dio->range.n_sectors = bio_sectors(bio);
1876	journal_read_pos = NOT_FOUND;
1877	if (likely(ic->mode == 'J')) {
1878		if (dio->write) {
1879			unsigned next_entry, i, pos;
1880			unsigned ws, we, range_sectors;
1881
1882			dio->range.n_sectors = min(dio->range.n_sectors,
1883						   (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1884			if (unlikely(!dio->range.n_sectors)) {
1885				if (from_map)
1886					goto offload_to_thread;
1887				sleep_on_endio_wait(ic);
1888				goto retry;
1889			}
1890			range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1891			ic->free_sectors -= range_sectors;
1892			journal_section = ic->free_section;
1893			journal_entry = ic->free_section_entry;
1894
1895			next_entry = ic->free_section_entry + range_sectors;
1896			ic->free_section_entry = next_entry % ic->journal_section_entries;
1897			ic->free_section += next_entry / ic->journal_section_entries;
1898			ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1899			wraparound_section(ic, &ic->free_section);
1900
1901			pos = journal_section * ic->journal_section_entries + journal_entry;
1902			ws = journal_section;
1903			we = journal_entry;
1904			i = 0;
1905			do {
1906				struct journal_entry *je;
1907
1908				add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1909				pos++;
1910				if (unlikely(pos >= ic->journal_entries))
1911					pos = 0;
1912
1913				je = access_journal_entry(ic, ws, we);
1914				BUG_ON(!journal_entry_is_unused(je));
1915				journal_entry_set_inprogress(je);
1916				we++;
1917				if (unlikely(we == ic->journal_section_entries)) {
1918					we = 0;
1919					ws++;
1920					wraparound_section(ic, &ws);
1921				}
1922			} while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1923
1924			spin_unlock_irq(&ic->endio_wait.lock);
1925			goto journal_read_write;
1926		} else {
1927			sector_t next_sector;
1928			journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1929			if (likely(journal_read_pos == NOT_FOUND)) {
1930				if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1931					dio->range.n_sectors = next_sector - dio->range.logical_sector;
1932			} else {
1933				unsigned i;
1934				unsigned jp = journal_read_pos + 1;
1935				for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1936					if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1937						break;
1938				}
1939				dio->range.n_sectors = i;
1940			}
1941		}
1942	}
1943	if (unlikely(!add_new_range(ic, &dio->range, true))) {
1944		/*
1945		 * We must not sleep in the request routine because it could
1946		 * stall bios on current->bio_list.
1947		 * So, we offload the bio to a workqueue if we have to sleep.
1948		 */
1949		if (from_map) {
1950offload_to_thread:
1951			spin_unlock_irq(&ic->endio_wait.lock);
1952			INIT_WORK(&dio->work, integrity_bio_wait);
1953			queue_work(ic->wait_wq, &dio->work);
1954			return;
1955		}
1956		if (journal_read_pos != NOT_FOUND)
1957			dio->range.n_sectors = ic->sectors_per_block;
1958		wait_and_add_new_range(ic, &dio->range);
1959		/*
1960		 * wait_and_add_new_range drops the spinlock, so the journal
1961		 * may have been changed arbitrarily. We need to recheck.
1962		 * To simplify the code, we restrict I/O size to just one block.
1963		 */
1964		if (journal_read_pos != NOT_FOUND) {
1965			sector_t next_sector;
1966			unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1967			if (unlikely(new_pos != journal_read_pos)) {
1968				remove_range_unlocked(ic, &dio->range);
1969				goto retry;
1970			}
1971		}
1972	}
1973	spin_unlock_irq(&ic->endio_wait.lock);
1974
1975	if (unlikely(journal_read_pos != NOT_FOUND)) {
1976		journal_section = journal_read_pos / ic->journal_section_entries;
1977		journal_entry = journal_read_pos % ic->journal_section_entries;
1978		goto journal_read_write;
1979	}
1980
1981	if (ic->mode == 'B' && dio->write) {
1982		if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
1983				     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
1984			struct bitmap_block_status *bbs;
1985
1986			bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
1987			spin_lock(&bbs->bio_queue_lock);
1988			bio_list_add(&bbs->bio_queue, bio);
1989			spin_unlock(&bbs->bio_queue_lock);
1990			queue_work(ic->writer_wq, &bbs->work);
1991			return;
1992		}
1993	}
1994
1995	dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1996
1997	if (need_sync_io) {
1998		init_completion(&read_comp);
1999		dio->completion = &read_comp;
2000	} else
2001		dio->completion = NULL;
2002
2003	dio->orig_bi_iter = bio->bi_iter;
2004
2005	dio->orig_bi_disk = bio->bi_disk;
2006	dio->orig_bi_partno = bio->bi_partno;
2007	bio_set_dev(bio, ic->dev->bdev);
2008
2009	dio->orig_bi_integrity = bio_integrity(bio);
2010	bio->bi_integrity = NULL;
2011	bio->bi_opf &= ~REQ_INTEGRITY;
2012
2013	dio->orig_bi_end_io = bio->bi_end_io;
2014	bio->bi_end_io = integrity_end_io;
2015
2016	bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2017	generic_make_request(bio);
2018
2019	if (need_sync_io) {
2020		wait_for_completion_io(&read_comp);
2021		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2022		    dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2023			goto skip_check;
2024		if (ic->mode == 'B') {
2025			if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2026					     dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2027				goto skip_check;
2028		}
2029
2030		if (likely(!bio->bi_status))
2031			integrity_metadata(&dio->work);
2032		else
2033skip_check:
2034			dec_in_flight(dio);
2035
2036	} else {
2037		INIT_WORK(&dio->work, integrity_metadata);
2038		queue_work(ic->metadata_wq, &dio->work);
2039	}
2040
2041	return;
2042
2043journal_read_write:
2044	if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2045		goto lock_retry;
2046
2047	do_endio_flush(ic, dio);
2048}
2049
2050
2051static void integrity_bio_wait(struct work_struct *w)
2052{
2053	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2054
2055	dm_integrity_map_continue(dio, false);
2056}
2057
2058static void pad_uncommitted(struct dm_integrity_c *ic)
2059{
2060	if (ic->free_section_entry) {
2061		ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2062		ic->free_section_entry = 0;
2063		ic->free_section++;
2064		wraparound_section(ic, &ic->free_section);
2065		ic->n_uncommitted_sections++;
2066	}
2067	if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2068		    (ic->n_uncommitted_sections + ic->n_committed_sections) *
2069		    ic->journal_section_entries + ic->free_sectors)) {
2070		DMCRIT("journal_sections %u, journal_section_entries %u, "
2071		       "n_uncommitted_sections %u, n_committed_sections %u, "
2072		       "journal_section_entries %u, free_sectors %u",
2073		       ic->journal_sections, ic->journal_section_entries,
2074		       ic->n_uncommitted_sections, ic->n_committed_sections,
2075		       ic->journal_section_entries, ic->free_sectors);
2076	}
2077}
2078
2079static void integrity_commit(struct work_struct *w)
2080{
2081	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2082	unsigned commit_start, commit_sections;
2083	unsigned i, j, n;
2084	struct bio *flushes;
2085
2086	del_timer(&ic->autocommit_timer);
2087
2088	spin_lock_irq(&ic->endio_wait.lock);
2089	flushes = bio_list_get(&ic->flush_bio_list);
2090	if (unlikely(ic->mode != 'J')) {
2091		spin_unlock_irq(&ic->endio_wait.lock);
2092		dm_integrity_flush_buffers(ic);
2093		goto release_flush_bios;
2094	}
2095
2096	pad_uncommitted(ic);
2097	commit_start = ic->uncommitted_section;
2098	commit_sections = ic->n_uncommitted_sections;
2099	spin_unlock_irq(&ic->endio_wait.lock);
2100
2101	if (!commit_sections)
2102		goto release_flush_bios;
2103
2104	i = commit_start;
2105	for (n = 0; n < commit_sections; n++) {
2106		for (j = 0; j < ic->journal_section_entries; j++) {
2107			struct journal_entry *je;
2108			je = access_journal_entry(ic, i, j);
2109			io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2110		}
2111		for (j = 0; j < ic->journal_section_sectors; j++) {
2112			struct journal_sector *js;
2113			js = access_journal(ic, i, j);
2114			js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2115		}
2116		i++;
2117		if (unlikely(i >= ic->journal_sections))
2118			ic->commit_seq = next_commit_seq(ic->commit_seq);
2119		wraparound_section(ic, &i);
2120	}
2121	smp_rmb();
2122
2123	write_journal(ic, commit_start, commit_sections);
2124
2125	spin_lock_irq(&ic->endio_wait.lock);
2126	ic->uncommitted_section += commit_sections;
2127	wraparound_section(ic, &ic->uncommitted_section);
2128	ic->n_uncommitted_sections -= commit_sections;
2129	ic->n_committed_sections += commit_sections;
2130	spin_unlock_irq(&ic->endio_wait.lock);
2131
2132	if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2133		queue_work(ic->writer_wq, &ic->writer_work);
2134
2135release_flush_bios:
2136	while (flushes) {
2137		struct bio *next = flushes->bi_next;
2138		flushes->bi_next = NULL;
2139		do_endio(ic, flushes);
2140		flushes = next;
2141	}
2142}
2143
2144static void complete_copy_from_journal(unsigned long error, void *context)
2145{
2146	struct journal_io *io = context;
2147	struct journal_completion *comp = io->comp;
2148	struct dm_integrity_c *ic = comp->ic;
2149	remove_range(ic, &io->range);
2150	mempool_free(io, &ic->journal_io_mempool);
2151	if (unlikely(error != 0))
2152		dm_integrity_io_error(ic, "copying from journal", -EIO);
2153	complete_journal_op(comp);
2154}
2155
2156static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2157			       struct journal_entry *je)
2158{
2159	unsigned s = 0;
2160	do {
2161		js->commit_id = je->last_bytes[s];
2162		js++;
2163	} while (++s < ic->sectors_per_block);
2164}
2165
2166static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2167			     unsigned write_sections, bool from_replay)
2168{
2169	unsigned i, j, n;
2170	struct journal_completion comp;
2171	struct blk_plug plug;
2172
2173	blk_start_plug(&plug);
2174
2175	comp.ic = ic;
2176	comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2177	init_completion(&comp.comp);
2178
2179	i = write_start;
2180	for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2181#ifndef INTERNAL_VERIFY
2182		if (unlikely(from_replay))
2183#endif
2184			rw_section_mac(ic, i, false);
2185		for (j = 0; j < ic->journal_section_entries; j++) {
2186			struct journal_entry *je = access_journal_entry(ic, i, j);
2187			sector_t sec, area, offset;
2188			unsigned k, l, next_loop;
2189			sector_t metadata_block;
2190			unsigned metadata_offset;
2191			struct journal_io *io;
2192
2193			if (journal_entry_is_unused(je))
2194				continue;
2195			BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2196			sec = journal_entry_get_sector(je);
2197			if (unlikely(from_replay)) {
2198				if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2199					dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2200					sec &= ~(sector_t)(ic->sectors_per_block - 1);
2201				}
2202			}
2203			get_area_and_offset(ic, sec, &area, &offset);
2204			restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2205			for (k = j + 1; k < ic->journal_section_entries; k++) {
2206				struct journal_entry *je2 = access_journal_entry(ic, i, k);
2207				sector_t sec2, area2, offset2;
2208				if (journal_entry_is_unused(je2))
2209					break;
2210				BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2211				sec2 = journal_entry_get_sector(je2);
2212				get_area_and_offset(ic, sec2, &area2, &offset2);
2213				if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2214					break;
2215				restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2216			}
2217			next_loop = k - 1;
2218
2219			io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2220			io->comp = &comp;
2221			io->range.logical_sector = sec;
2222			io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2223
2224			spin_lock_irq(&ic->endio_wait.lock);
2225			add_new_range_and_wait(ic, &io->range);
2226
2227			if (likely(!from_replay)) {
2228				struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2229
2230				/* don't write if there is newer committed sector */
2231				while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2232					struct journal_entry *je2 = access_journal_entry(ic, i, j);
2233
2234					journal_entry_set_unused(je2);
2235					remove_journal_node(ic, &section_node[j]);
2236					j++;
2237					sec += ic->sectors_per_block;
2238					offset += ic->sectors_per_block;
2239				}
2240				while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2241					struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2242
2243					journal_entry_set_unused(je2);
2244					remove_journal_node(ic, &section_node[k - 1]);
2245					k--;
2246				}
2247				if (j == k) {
2248					remove_range_unlocked(ic, &io->range);
2249					spin_unlock_irq(&ic->endio_wait.lock);
2250					mempool_free(io, &ic->journal_io_mempool);
2251					goto skip_io;
2252				}
2253				for (l = j; l < k; l++) {
2254					remove_journal_node(ic, &section_node[l]);
2255				}
2256			}
2257			spin_unlock_irq(&ic->endio_wait.lock);
2258
2259			metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2260			for (l = j; l < k; l++) {
2261				int r;
2262				struct journal_entry *je2 = access_journal_entry(ic, i, l);
2263
2264				if (
2265#ifndef INTERNAL_VERIFY
2266				    unlikely(from_replay) &&
2267#endif
2268				    ic->internal_hash) {
2269					char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2270
2271					integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2272								  (char *)access_journal_data(ic, i, l), test_tag);
2273					if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2274						dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2275				}
2276
2277				journal_entry_set_unused(je2);
2278				r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2279							ic->tag_size, TAG_WRITE);
2280				if (unlikely(r)) {
2281					dm_integrity_io_error(ic, "reading tags", r);
2282				}
2283			}
2284
2285			atomic_inc(&comp.in_flight);
2286			copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2287					  (k - j) << ic->sb->log2_sectors_per_block,
2288					  get_data_sector(ic, area, offset),
2289					  complete_copy_from_journal, io);
2290skip_io:
2291			j = next_loop;
2292		}
2293	}
2294
2295	dm_bufio_write_dirty_buffers_async(ic->bufio);
2296
2297	blk_finish_plug(&plug);
2298
2299	complete_journal_op(&comp);
2300	wait_for_completion_io(&comp.comp);
2301
2302	dm_integrity_flush_buffers(ic);
2303}
2304
2305static void integrity_writer(struct work_struct *w)
2306{
2307	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2308	unsigned write_start, write_sections;
2309
2310	unsigned prev_free_sectors;
2311
2312	/* the following test is not needed, but it tests the replay code */
2313	if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2314		return;
2315
2316	spin_lock_irq(&ic->endio_wait.lock);
2317	write_start = ic->committed_section;
2318	write_sections = ic->n_committed_sections;
2319	spin_unlock_irq(&ic->endio_wait.lock);
2320
2321	if (!write_sections)
2322		return;
2323
2324	do_journal_write(ic, write_start, write_sections, false);
2325
2326	spin_lock_irq(&ic->endio_wait.lock);
2327
2328	ic->committed_section += write_sections;
2329	wraparound_section(ic, &ic->committed_section);
2330	ic->n_committed_sections -= write_sections;
2331
2332	prev_free_sectors = ic->free_sectors;
2333	ic->free_sectors += write_sections * ic->journal_section_entries;
2334	if (unlikely(!prev_free_sectors))
2335		wake_up_locked(&ic->endio_wait);
2336
2337	spin_unlock_irq(&ic->endio_wait.lock);
2338}
2339
2340static void recalc_write_super(struct dm_integrity_c *ic)
2341{
2342	int r;
2343
2344	dm_integrity_flush_buffers(ic);
2345	if (dm_integrity_failed(ic))
2346		return;
2347
2348	r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2349	if (unlikely(r))
2350		dm_integrity_io_error(ic, "writing superblock", r);
2351}
2352
2353static void integrity_recalc(struct work_struct *w)
2354{
2355	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2356	struct dm_integrity_range range;
2357	struct dm_io_request io_req;
2358	struct dm_io_region io_loc;
2359	sector_t area, offset;
2360	sector_t metadata_block;
2361	unsigned metadata_offset;
2362	sector_t logical_sector, n_sectors;
2363	__u8 *t;
2364	unsigned i;
2365	int r;
2366	unsigned super_counter = 0;
2367
2368	DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2369
2370	spin_lock_irq(&ic->endio_wait.lock);
2371
2372next_chunk:
2373
2374	if (unlikely(READ_ONCE(ic->suspending)))
2375		goto unlock_ret;
2376
2377	range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2378	if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2379		if (ic->mode == 'B') {
2380			DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2381			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2382		}
2383		goto unlock_ret;
2384	}
2385
2386	get_area_and_offset(ic, range.logical_sector, &area, &offset);
2387	range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2388	if (!ic->meta_dev)
2389		range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2390
2391	add_new_range_and_wait(ic, &range);
2392	spin_unlock_irq(&ic->endio_wait.lock);
2393	logical_sector = range.logical_sector;
2394	n_sectors = range.n_sectors;
2395
2396	if (ic->mode == 'B') {
2397		if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2398			goto advance_and_next;
2399		}
2400		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2401				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2402			logical_sector += ic->sectors_per_block;
2403			n_sectors -= ic->sectors_per_block;
2404			cond_resched();
2405		}
2406		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2407				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2408			n_sectors -= ic->sectors_per_block;
2409			cond_resched();
2410		}
2411		get_area_and_offset(ic, logical_sector, &area, &offset);
2412	}
2413
2414	DEBUG_print("recalculating: %lx, %lx\n", logical_sector, n_sectors);
2415
2416	if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2417		recalc_write_super(ic);
2418		if (ic->mode == 'B') {
2419			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2420		}
2421		super_counter = 0;
2422	}
2423
2424	if (unlikely(dm_integrity_failed(ic)))
2425		goto err;
2426
2427	io_req.bi_op = REQ_OP_READ;
2428	io_req.bi_op_flags = 0;
2429	io_req.mem.type = DM_IO_VMA;
2430	io_req.mem.ptr.addr = ic->recalc_buffer;
2431	io_req.notify.fn = NULL;
2432	io_req.client = ic->io;
2433	io_loc.bdev = ic->dev->bdev;
2434	io_loc.sector = get_data_sector(ic, area, offset);
2435	io_loc.count = n_sectors;
2436
2437	r = dm_io(&io_req, 1, &io_loc, NULL);
2438	if (unlikely(r)) {
2439		dm_integrity_io_error(ic, "reading data", r);
2440		goto err;
2441	}
2442
2443	t = ic->recalc_tags;
2444	for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2445		integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2446		t += ic->tag_size;
2447	}
2448
2449	metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2450
2451	r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2452	if (unlikely(r)) {
2453		dm_integrity_io_error(ic, "writing tags", r);
2454		goto err;
2455	}
2456
2457advance_and_next:
2458	cond_resched();
2459
2460	spin_lock_irq(&ic->endio_wait.lock);
2461	remove_range_unlocked(ic, &range);
2462	ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2463	goto next_chunk;
2464
2465err:
2466	remove_range(ic, &range);
2467	return;
2468
2469unlock_ret:
2470	spin_unlock_irq(&ic->endio_wait.lock);
2471
2472	recalc_write_super(ic);
2473}
2474
2475static void bitmap_block_work(struct work_struct *w)
2476{
2477	struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2478	struct dm_integrity_c *ic = bbs->ic;
2479	struct bio *bio;
2480	struct bio_list bio_queue;
2481	struct bio_list waiting;
2482
2483	bio_list_init(&waiting);
2484
2485	spin_lock(&bbs->bio_queue_lock);
2486	bio_queue = bbs->bio_queue;
2487	bio_list_init(&bbs->bio_queue);
2488	spin_unlock(&bbs->bio_queue_lock);
2489
2490	while ((bio = bio_list_pop(&bio_queue))) {
2491		struct dm_integrity_io *dio;
2492
2493		dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2494
2495		if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2496				    dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2497			remove_range(ic, &dio->range);
2498			INIT_WORK(&dio->work, integrity_bio_wait);
2499			queue_work(ic->wait_wq, &dio->work);
2500		} else {
2501			block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2502					dio->range.n_sectors, BITMAP_OP_SET);
2503			bio_list_add(&waiting, bio);
2504		}
2505	}
2506
2507	if (bio_list_empty(&waiting))
2508		return;
2509
2510	rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2511			   bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2512			   BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2513
2514	while ((bio = bio_list_pop(&waiting))) {
2515		struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2516
2517		block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2518				dio->range.n_sectors, BITMAP_OP_SET);
2519
2520		remove_range(ic, &dio->range);
2521		INIT_WORK(&dio->work, integrity_bio_wait);
2522		queue_work(ic->wait_wq, &dio->work);
2523	}
2524
2525	queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2526}
2527
2528static void bitmap_flush_work(struct work_struct *work)
2529{
2530	struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2531	struct dm_integrity_range range;
2532	unsigned long limit;
2533	struct bio *bio;
2534
2535	dm_integrity_flush_buffers(ic);
2536
2537	range.logical_sector = 0;
2538	range.n_sectors = ic->provided_data_sectors;
2539
2540	spin_lock_irq(&ic->endio_wait.lock);
2541	add_new_range_and_wait(ic, &range);
2542	spin_unlock_irq(&ic->endio_wait.lock);
2543
2544	dm_integrity_flush_buffers(ic);
2545	if (ic->meta_dev)
2546		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2547
2548	limit = ic->provided_data_sectors;
2549	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2550		limit = le64_to_cpu(ic->sb->recalc_sector)
2551			>> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2552			<< (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2553	}
2554	/*DEBUG_print("zeroing journal\n");*/
2555	block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2556	block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2557
2558	rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2559			   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2560
2561	spin_lock_irq(&ic->endio_wait.lock);
2562	remove_range_unlocked(ic, &range);
2563	while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2564		bio_endio(bio);
2565		spin_unlock_irq(&ic->endio_wait.lock);
2566		spin_lock_irq(&ic->endio_wait.lock);
2567	}
2568	spin_unlock_irq(&ic->endio_wait.lock);
2569}
2570
2571
2572static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2573			 unsigned n_sections, unsigned char commit_seq)
2574{
2575	unsigned i, j, n;
2576
2577	if (!n_sections)
2578		return;
2579
2580	for (n = 0; n < n_sections; n++) {
2581		i = start_section + n;
2582		wraparound_section(ic, &i);
2583		for (j = 0; j < ic->journal_section_sectors; j++) {
2584			struct journal_sector *js = access_journal(ic, i, j);
2585			memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2586			js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2587		}
2588		for (j = 0; j < ic->journal_section_entries; j++) {
2589			struct journal_entry *je = access_journal_entry(ic, i, j);
2590			journal_entry_set_unused(je);
2591		}
2592	}
2593
2594	write_journal(ic, start_section, n_sections);
2595}
2596
2597static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2598{
2599	unsigned char k;
2600	for (k = 0; k < N_COMMIT_IDS; k++) {
2601		if (dm_integrity_commit_id(ic, i, j, k) == id)
2602			return k;
2603	}
2604	dm_integrity_io_error(ic, "journal commit id", -EIO);
2605	return -EIO;
2606}
2607
2608static void replay_journal(struct dm_integrity_c *ic)
2609{
2610	unsigned i, j;
2611	bool used_commit_ids[N_COMMIT_IDS];
2612	unsigned max_commit_id_sections[N_COMMIT_IDS];
2613	unsigned write_start, write_sections;
2614	unsigned continue_section;
2615	bool journal_empty;
2616	unsigned char unused, last_used, want_commit_seq;
2617
2618	if (ic->mode == 'R')
2619		return;
2620
2621	if (ic->journal_uptodate)
2622		return;
2623
2624	last_used = 0;
2625	write_start = 0;
2626
2627	if (!ic->just_formatted) {
2628		DEBUG_print("reading journal\n");
2629		rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2630		if (ic->journal_io)
2631			DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2632		if (ic->journal_io) {
2633			struct journal_completion crypt_comp;
2634			crypt_comp.ic = ic;
2635			init_completion(&crypt_comp.comp);
2636			crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2637			encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2638			wait_for_completion(&crypt_comp.comp);
2639		}
2640		DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2641	}
2642
2643	if (dm_integrity_failed(ic))
2644		goto clear_journal;
2645
2646	journal_empty = true;
2647	memset(used_commit_ids, 0, sizeof used_commit_ids);
2648	memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2649	for (i = 0; i < ic->journal_sections; i++) {
2650		for (j = 0; j < ic->journal_section_sectors; j++) {
2651			int k;
2652			struct journal_sector *js = access_journal(ic, i, j);
2653			k = find_commit_seq(ic, i, j, js->commit_id);
2654			if (k < 0)
2655				goto clear_journal;
2656			used_commit_ids[k] = true;
2657			max_commit_id_sections[k] = i;
2658		}
2659		if (journal_empty) {
2660			for (j = 0; j < ic->journal_section_entries; j++) {
2661				struct journal_entry *je = access_journal_entry(ic, i, j);
2662				if (!journal_entry_is_unused(je)) {
2663					journal_empty = false;
2664					break;
2665				}
2666			}
2667		}
2668	}
2669
2670	if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2671		unused = N_COMMIT_IDS - 1;
2672		while (unused && !used_commit_ids[unused - 1])
2673			unused--;
2674	} else {
2675		for (unused = 0; unused < N_COMMIT_IDS; unused++)
2676			if (!used_commit_ids[unused])
2677				break;
2678		if (unused == N_COMMIT_IDS) {
2679			dm_integrity_io_error(ic, "journal commit ids", -EIO);
2680			goto clear_journal;
2681		}
2682	}
2683	DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2684		    unused, used_commit_ids[0], used_commit_ids[1],
2685		    used_commit_ids[2], used_commit_ids[3]);
2686
2687	last_used = prev_commit_seq(unused);
2688	want_commit_seq = prev_commit_seq(last_used);
2689
2690	if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2691		journal_empty = true;
2692
2693	write_start = max_commit_id_sections[last_used] + 1;
2694	if (unlikely(write_start >= ic->journal_sections))
2695		want_commit_seq = next_commit_seq(want_commit_seq);
2696	wraparound_section(ic, &write_start);
2697
2698	i = write_start;
2699	for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2700		for (j = 0; j < ic->journal_section_sectors; j++) {
2701			struct journal_sector *js = access_journal(ic, i, j);
2702
2703			if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2704				/*
2705				 * This could be caused by crash during writing.
2706				 * We won't replay the inconsistent part of the
2707				 * journal.
2708				 */
2709				DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2710					    i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2711				goto brk;
2712			}
2713		}
2714		i++;
2715		if (unlikely(i >= ic->journal_sections))
2716			want_commit_seq = next_commit_seq(want_commit_seq);
2717		wraparound_section(ic, &i);
2718	}
2719brk:
2720
2721	if (!journal_empty) {
2722		DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2723			    write_sections, write_start, want_commit_seq);
2724		do_journal_write(ic, write_start, write_sections, true);
2725	}
2726
2727	if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2728		continue_section = write_start;
2729		ic->commit_seq = want_commit_seq;
2730		DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2731	} else {
2732		unsigned s;
2733		unsigned char erase_seq;
2734clear_journal:
2735		DEBUG_print("clearing journal\n");
2736
2737		erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2738		s = write_start;
2739		init_journal(ic, s, 1, erase_seq);
2740		s++;
2741		wraparound_section(ic, &s);
2742		if (ic->journal_sections >= 2) {
2743			init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2744			s += ic->journal_sections - 2;
2745			wraparound_section(ic, &s);
2746			init_journal(ic, s, 1, erase_seq);
2747		}
2748
2749		continue_section = 0;
2750		ic->commit_seq = next_commit_seq(erase_seq);
2751	}
2752
2753	ic->committed_section = continue_section;
2754	ic->n_committed_sections = 0;
2755
2756	ic->uncommitted_section = continue_section;
2757	ic->n_uncommitted_sections = 0;
2758
2759	ic->free_section = continue_section;
2760	ic->free_section_entry = 0;
2761	ic->free_sectors = ic->journal_entries;
2762
2763	ic->journal_tree_root = RB_ROOT;
2764	for (i = 0; i < ic->journal_entries; i++)
2765		init_journal_node(&ic->journal_tree[i]);
2766}
2767
2768static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2769{
2770	DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2771
2772	if (ic->mode == 'B') {
2773		ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2774		ic->synchronous_mode = 1;
2775
2776		cancel_delayed_work_sync(&ic->bitmap_flush_work);
2777		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2778		flush_workqueue(ic->commit_wq);
2779	}
2780}
2781
2782static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2783{
2784	struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2785
2786	DEBUG_print("dm_integrity_reboot\n");
2787
2788	dm_integrity_enter_synchronous_mode(ic);
2789
2790	return NOTIFY_DONE;
2791}
2792
2793static void dm_integrity_postsuspend(struct dm_target *ti)
2794{
2795	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2796	int r;
2797
2798	WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2799
2800	del_timer_sync(&ic->autocommit_timer);
2801
2802	WRITE_ONCE(ic->suspending, 1);
2803
2804	if (ic->recalc_wq)
2805		drain_workqueue(ic->recalc_wq);
2806
2807	if (ic->mode == 'B')
2808		cancel_delayed_work_sync(&ic->bitmap_flush_work);
2809
2810	queue_work(ic->commit_wq, &ic->commit_work);
2811	drain_workqueue(ic->commit_wq);
2812
2813	if (ic->mode == 'J') {
2814		if (ic->meta_dev)
2815			queue_work(ic->writer_wq, &ic->writer_work);
2816		drain_workqueue(ic->writer_wq);
2817		dm_integrity_flush_buffers(ic);
2818	}
2819
2820	if (ic->mode == 'B') {
2821		dm_integrity_flush_buffers(ic);
2822#if 1
2823		/* set to 0 to test bitmap replay code */
2824		init_journal(ic, 0, ic->journal_sections, 0);
2825		ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2826		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2827		if (unlikely(r))
2828			dm_integrity_io_error(ic, "writing superblock", r);
2829#endif
2830	}
2831
2832	WRITE_ONCE(ic->suspending, 0);
2833
2834	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2835
2836	ic->journal_uptodate = true;
2837}
2838
2839static void dm_integrity_resume(struct dm_target *ti)
2840{
2841	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2842	int r;
2843	DEBUG_print("resume\n");
2844
2845	if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2846		DEBUG_print("resume dirty_bitmap\n");
2847		rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2848				   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2849		if (ic->mode == 'B') {
2850			if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2851				block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2852				block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2853				if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
2854						     BITMAP_OP_TEST_ALL_CLEAR)) {
2855					ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2856					ic->sb->recalc_sector = cpu_to_le64(0);
2857				}
2858			} else {
2859				DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
2860					    ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2861				ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2862				block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2863				block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2864				block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2865				rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2866						   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2867				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2868				ic->sb->recalc_sector = cpu_to_le64(0);
2869			}
2870		} else {
2871			if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
2872			      block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
2873				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2874				ic->sb->recalc_sector = cpu_to_le64(0);
2875			}
2876			init_journal(ic, 0, ic->journal_sections, 0);
2877			replay_journal(ic);
2878			ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2879		}
2880		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2881		if (unlikely(r))
2882			dm_integrity_io_error(ic, "writing superblock", r);
2883	} else {
2884		replay_journal(ic);
2885		if (ic->mode == 'B') {
2886			int mode;
2887			ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2888			ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2889			r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2890			if (unlikely(r))
2891				dm_integrity_io_error(ic, "writing superblock", r);
2892
2893			mode = ic->recalculate_flag ? BITMAP_OP_SET : BITMAP_OP_CLEAR;
2894			block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, mode);
2895			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, mode);
2896			block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, mode);
2897			rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2898					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2899		}
2900	}
2901
2902	DEBUG_print("testing recalc: %x\n", ic->sb->flags);
2903	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2904		__u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2905		DEBUG_print("recalc pos: %lx / %lx\n", (long)recalc_pos, ic->provided_data_sectors);
2906		if (recalc_pos < ic->provided_data_sectors) {
2907			queue_work(ic->recalc_wq, &ic->recalc_work);
2908		} else if (recalc_pos > ic->provided_data_sectors) {
2909			ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2910			recalc_write_super(ic);
2911		}
2912	}
2913
2914	ic->reboot_notifier.notifier_call = dm_integrity_reboot;
2915	ic->reboot_notifier.next = NULL;
2916	ic->reboot_notifier.priority = INT_MAX - 1;	/* be notified after md and before hardware drivers */
2917	WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
2918
2919#if 0
2920	/* set to 1 to stress test synchronous mode */
2921	dm_integrity_enter_synchronous_mode(ic);
2922#endif
2923}
2924
2925static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2926				unsigned status_flags, char *result, unsigned maxlen)
2927{
2928	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2929	unsigned arg_count;
2930	size_t sz = 0;
2931
2932	switch (type) {
2933	case STATUSTYPE_INFO:
2934		DMEMIT("%llu %llu",
2935			(unsigned long long)atomic64_read(&ic->number_of_mismatches),
2936			(unsigned long long)ic->provided_data_sectors);
2937		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2938			DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2939		else
2940			DMEMIT(" -");
2941		break;
2942
2943	case STATUSTYPE_TABLE: {
2944		__u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2945		watermark_percentage += ic->journal_entries / 2;
2946		do_div(watermark_percentage, ic->journal_entries);
2947		arg_count = 3;
2948		arg_count += !!ic->meta_dev;
2949		arg_count += ic->sectors_per_block != 1;
2950		arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2951		arg_count += ic->mode == 'J';
2952		arg_count += ic->mode == 'J';
2953		arg_count += ic->mode == 'B';
2954		arg_count += ic->mode == 'B';
2955		arg_count += !!ic->internal_hash_alg.alg_string;
2956		arg_count += !!ic->journal_crypt_alg.alg_string;
2957		arg_count += !!ic->journal_mac_alg.alg_string;
2958		DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2959		       ic->tag_size, ic->mode, arg_count);
2960		if (ic->meta_dev)
2961			DMEMIT(" meta_device:%s", ic->meta_dev->name);
2962		if (ic->sectors_per_block != 1)
2963			DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2964		if (ic->recalculate_flag)
2965			DMEMIT(" recalculate");
2966		DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2967		DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2968		DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2969		if (ic->mode == 'J') {
2970			DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2971			DMEMIT(" commit_time:%u", ic->autocommit_msec);
2972		}
2973		if (ic->mode == 'B') {
2974			DMEMIT(" sectors_per_bit:%llu", (unsigned long long)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
2975			DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
2976		}
2977
2978#define EMIT_ALG(a, n)							\
2979		do {							\
2980			if (ic->a.alg_string) {				\
2981				DMEMIT(" %s:%s", n, ic->a.alg_string);	\
2982				if (ic->a.key_string)			\
2983					DMEMIT(":%s", ic->a.key_string);\
2984			}						\
2985		} while (0)
2986		EMIT_ALG(internal_hash_alg, "internal_hash");
2987		EMIT_ALG(journal_crypt_alg, "journal_crypt");
2988		EMIT_ALG(journal_mac_alg, "journal_mac");
2989		break;
2990	}
2991	}
2992}
2993
2994static int dm_integrity_iterate_devices(struct dm_target *ti,
2995					iterate_devices_callout_fn fn, void *data)
2996{
2997	struct dm_integrity_c *ic = ti->private;
2998
2999	if (!ic->meta_dev)
3000		return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3001	else
3002		return fn(ti, ic->dev, 0, ti->len, data);
3003}
3004
3005static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3006{
3007	struct dm_integrity_c *ic = ti->private;
3008
3009	if (ic->sectors_per_block > 1) {
3010		limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3011		limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3012		blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3013	}
3014}
3015
3016static void calculate_journal_section_size(struct dm_integrity_c *ic)
3017{
3018	unsigned sector_space = JOURNAL_SECTOR_DATA;
3019
3020	ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3021	ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3022					 JOURNAL_ENTRY_ROUNDUP);
3023
3024	if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3025		sector_space -= JOURNAL_MAC_PER_SECTOR;
3026	ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3027	ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3028	ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3029	ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3030}
3031
3032static int calculate_device_limits(struct dm_integrity_c *ic)
3033{
3034	__u64 initial_sectors;
3035
3036	calculate_journal_section_size(ic);
3037	initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3038	if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3039		return -EINVAL;
3040	ic->initial_sectors = initial_sectors;
3041
3042	if (!ic->meta_dev) {
3043		sector_t last_sector, last_area, last_offset;
3044
3045		ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3046					   (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
3047		if (!(ic->metadata_run & (ic->metadata_run - 1)))
3048			ic->log2_metadata_run = __ffs(ic->metadata_run);
3049		else
3050			ic->log2_metadata_run = -1;
3051
3052		get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3053		last_sector = get_data_sector(ic, last_area, last_offset);
3054		if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3055			return -EINVAL;
3056	} else {
3057		__u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3058		meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3059				>> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3060		meta_size <<= ic->log2_buffer_sectors;
3061		if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3062		    ic->initial_sectors + meta_size > ic->meta_device_sectors)
3063			return -EINVAL;
3064		ic->metadata_run = 1;
3065		ic->log2_metadata_run = 0;
3066	}
3067
3068	return 0;
3069}
3070
3071static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3072{
3073	unsigned journal_sections;
3074	int test_bit;
3075
3076	memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3077	memcpy(ic->sb->magic, SB_MAGIC, 8);
3078	ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3079	ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3080	if (ic->journal_mac_alg.alg_string)
3081		ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3082
3083	calculate_journal_section_size(ic);
3084	journal_sections = journal_sectors / ic->journal_section_sectors;
3085	if (!journal_sections)
3086		journal_sections = 1;
3087
3088	if (!ic->meta_dev) {
3089		ic->sb->journal_sections = cpu_to_le32(journal_sections);
3090		if (!interleave_sectors)
3091			interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3092		ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3093		ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3094		ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3095
3096		ic->provided_data_sectors = 0;
3097		for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3098			__u64 prev_data_sectors = ic->provided_data_sectors;
3099
3100			ic->provided_data_sectors |= (sector_t)1 << test_bit;
3101			if (calculate_device_limits(ic))
3102				ic->provided_data_sectors = prev_data_sectors;
3103		}
3104		if (!ic->provided_data_sectors)
3105			return -EINVAL;
3106	} else {
3107		ic->sb->log2_interleave_sectors = 0;
3108		ic->provided_data_sectors = ic->data_device_sectors;
3109		ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3110
3111try_smaller_buffer:
3112		ic->sb->journal_sections = cpu_to_le32(0);
3113		for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3114			__u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3115			__u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3116			if (test_journal_sections > journal_sections)
3117				continue;
3118			ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3119			if (calculate_device_limits(ic))
3120				ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3121
3122		}
3123		if (!le32_to_cpu(ic->sb->journal_sections)) {
3124			if (ic->log2_buffer_sectors > 3) {
3125				ic->log2_buffer_sectors--;
3126				goto try_smaller_buffer;
3127			}
3128			return -EINVAL;
3129		}
3130	}
3131
3132	ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3133
3134	sb_set_version(ic);
3135
3136	return 0;
3137}
3138
3139static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3140{
3141	struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3142	struct blk_integrity bi;
3143
3144	memset(&bi, 0, sizeof(bi));
3145	bi.profile = &dm_integrity_profile;
3146	bi.tuple_size = ic->tag_size;
3147	bi.tag_size = bi.tuple_size;
3148	bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3149
3150	blk_integrity_register(disk, &bi);
3151	blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3152}
3153
3154static void dm_integrity_free_page_list(struct page_list *pl)
3155{
3156	unsigned i;
3157
3158	if (!pl)
3159		return;
3160	for (i = 0; pl[i].page; i++)
3161		__free_page(pl[i].page);
3162	kvfree(pl);
3163}
3164
3165static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3166{
3167	struct page_list *pl;
3168	unsigned i;
3169
3170	pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3171	if (!pl)
3172		return NULL;
3173
3174	for (i = 0; i < n_pages; i++) {
3175		pl[i].page = alloc_page(GFP_KERNEL);
3176		if (!pl[i].page) {
3177			dm_integrity_free_page_list(pl);
3178			return NULL;
3179		}
3180		if (i)
3181			pl[i - 1].next = &pl[i];
3182	}
3183	pl[i].page = NULL;
3184	pl[i].next = NULL;
3185
3186	return pl;
3187}
3188
3189static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3190{
3191	unsigned i;
3192	for (i = 0; i < ic->journal_sections; i++)
3193		kvfree(sl[i]);
3194	kvfree(sl);
3195}
3196
3197static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3198								   struct page_list *pl)
3199{
3200	struct scatterlist **sl;
3201	unsigned i;
3202
3203	sl = kvmalloc_array(ic->journal_sections,
3204			    sizeof(struct scatterlist *),
3205			    GFP_KERNEL | __GFP_ZERO);
3206	if (!sl)
3207		return NULL;
3208
3209	for (i = 0; i < ic->journal_sections; i++) {
3210		struct scatterlist *s;
3211		unsigned start_index, start_offset;
3212		unsigned end_index, end_offset;
3213		unsigned n_pages;
3214		unsigned idx;
3215
3216		page_list_location(ic, i, 0, &start_index, &start_offset);
3217		page_list_location(ic, i, ic->journal_section_sectors - 1,
3218				   &end_index, &end_offset);
3219
3220		n_pages = (end_index - start_index + 1);
3221
3222		s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3223				   GFP_KERNEL);
3224		if (!s) {
3225			dm_integrity_free_journal_scatterlist(ic, sl);
3226			return NULL;
3227		}
3228
3229		sg_init_table(s, n_pages);
3230		for (idx = start_index; idx <= end_index; idx++) {
3231			char *va = lowmem_page_address(pl[idx].page);
3232			unsigned start = 0, end = PAGE_SIZE;
3233			if (idx == start_index)
3234				start = start_offset;
3235			if (idx == end_index)
3236				end = end_offset + (1 << SECTOR_SHIFT);
3237			sg_set_buf(&s[idx - start_index], va + start, end - start);
3238		}
3239
3240		sl[i] = s;
3241	}
3242
3243	return sl;
3244}
3245
3246static void free_alg(struct alg_spec *a)
3247{
3248	kzfree(a->alg_string);
3249	kzfree(a->key);
3250	memset(a, 0, sizeof *a);
3251}
3252
3253static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3254{
3255	char *k;
3256
3257	free_alg(a);
3258
3259	a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3260	if (!a->alg_string)
3261		goto nomem;
3262
3263	k = strchr(a->alg_string, ':');
3264	if (k) {
3265		*k = 0;
3266		a->key_string = k + 1;
3267		if (strlen(a->key_string) & 1)
3268			goto inval;
3269
3270		a->key_size = strlen(a->key_string) / 2;
3271		a->key = kmalloc(a->key_size, GFP_KERNEL);
3272		if (!a->key)
3273			goto nomem;
3274		if (hex2bin(a->key, a->key_string, a->key_size))
3275			goto inval;
3276	}
3277
3278	return 0;
3279inval:
3280	*error = error_inval;
3281	return -EINVAL;
3282nomem:
3283	*error = "Out of memory for an argument";
3284	return -ENOMEM;
3285}
3286
3287static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3288		   char *error_alg, char *error_key)
3289{
3290	int r;
3291
3292	if (a->alg_string) {
3293		*hash = crypto_alloc_shash(a->alg_string, 0, 0);
3294		if (IS_ERR(*hash)) {
3295			*error = error_alg;
3296			r = PTR_ERR(*hash);
3297			*hash = NULL;
3298			return r;
3299		}
3300
3301		if (a->key) {
3302			r = crypto_shash_setkey(*hash, a->key, a->key_size);
3303			if (r) {
3304				*error = error_key;
3305				return r;
3306			}
3307		} else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3308			*error = error_key;
3309			return -ENOKEY;
3310		}
3311	}
3312
3313	return 0;
3314}
3315
3316static int create_journal(struct dm_integrity_c *ic, char **error)
3317{
3318	int r = 0;
3319	unsigned i;
3320	__u64 journal_pages, journal_desc_size, journal_tree_size;
3321	unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3322	struct skcipher_request *req = NULL;
3323
3324	ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3325	ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3326	ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3327	ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3328
3329	journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3330				PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3331	journal_desc_size = journal_pages * sizeof(struct page_list);
3332	if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3333		*error = "Journal doesn't fit into memory";
3334		r = -ENOMEM;
3335		goto bad;
3336	}
3337	ic->journal_pages = journal_pages;
3338
3339	ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3340	if (!ic->journal) {
3341		*error = "Could not allocate memory for journal";
3342		r = -ENOMEM;
3343		goto bad;
3344	}
3345	if (ic->journal_crypt_alg.alg_string) {
3346		unsigned ivsize, blocksize;
3347		struct journal_completion comp;
3348
3349		comp.ic = ic;
3350		ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3351		if (IS_ERR(ic->journal_crypt)) {
3352			*error = "Invalid journal cipher";
3353			r = PTR_ERR(ic->journal_crypt);
3354			ic->journal_crypt = NULL;
3355			goto bad;
3356		}
3357		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3358		blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3359
3360		if (ic->journal_crypt_alg.key) {
3361			r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3362						   ic->journal_crypt_alg.key_size);
3363			if (r) {
3364				*error = "Error setting encryption key";
3365				goto bad;
3366			}
3367		}
3368		DEBUG_print("cipher %s, block size %u iv size %u\n",
3369			    ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3370
3371		ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3372		if (!ic->journal_io) {
3373			*error = "Could not allocate memory for journal io";
3374			r = -ENOMEM;
3375			goto bad;
3376		}
3377
3378		if (blocksize == 1) {
3379			struct scatterlist *sg;
3380
3381			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3382			if (!req) {
3383				*error = "Could not allocate crypt request";
3384				r = -ENOMEM;
3385				goto bad;
3386			}
3387
3388			crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3389			if (!crypt_iv) {
3390				*error = "Could not allocate iv";
3391				r = -ENOMEM;
3392				goto bad;
3393			}
3394
3395			ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3396			if (!ic->journal_xor) {
3397				*error = "Could not allocate memory for journal xor";
3398				r = -ENOMEM;
3399				goto bad;
3400			}
3401
3402			sg = kvmalloc_array(ic->journal_pages + 1,
3403					    sizeof(struct scatterlist),
3404					    GFP_KERNEL);
3405			if (!sg) {
3406				*error = "Unable to allocate sg list";
3407				r = -ENOMEM;
3408				goto bad;
3409			}
3410			sg_init_table(sg, ic->journal_pages + 1);
3411			for (i = 0; i < ic->journal_pages; i++) {
3412				char *va = lowmem_page_address(ic->journal_xor[i].page);
3413				clear_page(va);
3414				sg_set_buf(&sg[i], va, PAGE_SIZE);
3415			}
3416			sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3417
3418			skcipher_request_set_crypt(req, sg, sg,
3419						   PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3420			init_completion(&comp.comp);
3421			comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3422			if (do_crypt(true, req, &comp))
3423				wait_for_completion(&comp.comp);
3424			kvfree(sg);
3425			r = dm_integrity_failed(ic);
3426			if (r) {
3427				*error = "Unable to encrypt journal";
3428				goto bad;
3429			}
3430			DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3431
3432			crypto_free_skcipher(ic->journal_crypt);
3433			ic->journal_crypt = NULL;
3434		} else {
3435			unsigned crypt_len = roundup(ivsize, blocksize);
3436
3437			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3438			if (!req) {
3439				*error = "Could not allocate crypt request";
3440				r = -ENOMEM;
3441				goto bad;
3442			}
3443
3444			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3445			if (!crypt_iv) {
3446				*error = "Could not allocate iv";
3447				r = -ENOMEM;
3448				goto bad;
3449			}
3450
3451			crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3452			if (!crypt_data) {
3453				*error = "Unable to allocate crypt data";
3454				r = -ENOMEM;
3455				goto bad;
3456			}
3457
3458			ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3459			if (!ic->journal_scatterlist) {
3460				*error = "Unable to allocate sg list";
3461				r = -ENOMEM;
3462				goto bad;
3463			}
3464			ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3465			if (!ic->journal_io_scatterlist) {
3466				*error = "Unable to allocate sg list";
3467				r = -ENOMEM;
3468				goto bad;
3469			}
3470			ic->sk_requests = kvmalloc_array(ic->journal_sections,
3471							 sizeof(struct skcipher_request *),
3472							 GFP_KERNEL | __GFP_ZERO);
3473			if (!ic->sk_requests) {
3474				*error = "Unable to allocate sk requests";
3475				r = -ENOMEM;
3476				goto bad;
3477			}
3478			for (i = 0; i < ic->journal_sections; i++) {
3479				struct scatterlist sg;
3480				struct skcipher_request *section_req;
3481				__u32 section_le = cpu_to_le32(i);
3482
3483				memset(crypt_iv, 0x00, ivsize);
3484				memset(crypt_data, 0x00, crypt_len);
3485				memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3486
3487				sg_init_one(&sg, crypt_data, crypt_len);
3488				skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3489				init_completion(&comp.comp);
3490				comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3491				if (do_crypt(true, req, &comp))
3492					wait_for_completion(&comp.comp);
3493
3494				r = dm_integrity_failed(ic);
3495				if (r) {
3496					*error = "Unable to generate iv";
3497					goto bad;
3498				}
3499
3500				section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3501				if (!section_req) {
3502					*error = "Unable to allocate crypt request";
3503					r = -ENOMEM;
3504					goto bad;
3505				}
3506				section_req->iv = kmalloc_array(ivsize, 2,
3507								GFP_KERNEL);
3508				if (!section_req->iv) {
3509					skcipher_request_free(section_req);
3510					*error = "Unable to allocate iv";
3511					r = -ENOMEM;
3512					goto bad;
3513				}
3514				memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3515				section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3516				ic->sk_requests[i] = section_req;
3517				DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3518			}
3519		}
3520	}
3521
3522	for (i = 0; i < N_COMMIT_IDS; i++) {
3523		unsigned j;
3524retest_commit_id:
3525		for (j = 0; j < i; j++) {
3526			if (ic->commit_ids[j] == ic->commit_ids[i]) {
3527				ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3528				goto retest_commit_id;
3529			}
3530		}
3531		DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3532	}
3533
3534	journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3535	if (journal_tree_size > ULONG_MAX) {
3536		*error = "Journal doesn't fit into memory";
3537		r = -ENOMEM;
3538		goto bad;
3539	}
3540	ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3541	if (!ic->journal_tree) {
3542		*error = "Could not allocate memory for journal tree";
3543		r = -ENOMEM;
3544	}
3545bad:
3546	kfree(crypt_data);
3547	kfree(crypt_iv);
3548	skcipher_request_free(req);
3549
3550	return r;
3551}
3552
3553/*
3554 * Construct a integrity mapping
3555 *
3556 * Arguments:
3557 *	device
3558 *	offset from the start of the device
3559 *	tag size
3560 *	D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3561 *	number of optional arguments
3562 *	optional arguments:
3563 *		journal_sectors
3564 *		interleave_sectors
3565 *		buffer_sectors
3566 *		journal_watermark
3567 *		commit_time
3568 *		meta_device
3569 *		block_size
3570 *		sectors_per_bit
3571 *		bitmap_flush_interval
3572 *		internal_hash
3573 *		journal_crypt
3574 *		journal_mac
3575 *		recalculate
3576 */
3577static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3578{
3579	struct dm_integrity_c *ic;
3580	char dummy;
3581	int r;
3582	unsigned extra_args;
3583	struct dm_arg_set as;
3584	static const struct dm_arg _args[] = {
3585		{0, 9, "Invalid number of feature args"},
3586	};
3587	unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3588	bool should_write_sb;
3589	__u64 threshold;
3590	unsigned long long start;
3591	__s8 log2_sectors_per_bitmap_bit = -1;
3592	__s8 log2_blocks_per_bitmap_bit;
3593	__u64 bits_in_journal;
3594	__u64 n_bitmap_bits;
3595
3596#define DIRECT_ARGUMENTS	4
3597
3598	if (argc <= DIRECT_ARGUMENTS) {
3599		ti->error = "Invalid argument count";
3600		return -EINVAL;
3601	}
3602
3603	ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3604	if (!ic) {
3605		ti->error = "Cannot allocate integrity context";
3606		return -ENOMEM;
3607	}
3608	ti->private = ic;
3609	ti->per_io_data_size = sizeof(struct dm_integrity_io);
3610
3611	ic->in_progress = RB_ROOT;
3612	INIT_LIST_HEAD(&ic->wait_list);
3613	init_waitqueue_head(&ic->endio_wait);
3614	bio_list_init(&ic->flush_bio_list);
3615	init_waitqueue_head(&ic->copy_to_journal_wait);
3616	init_completion(&ic->crypto_backoff);
3617	atomic64_set(&ic->number_of_mismatches, 0);
3618	ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3619
3620	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3621	if (r) {
3622		ti->error = "Device lookup failed";
3623		goto bad;
3624	}
3625
3626	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3627		ti->error = "Invalid starting offset";
3628		r = -EINVAL;
3629		goto bad;
3630	}
3631	ic->start = start;
3632
3633	if (strcmp(argv[2], "-")) {
3634		if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3635			ti->error = "Invalid tag size";
3636			r = -EINVAL;
3637			goto bad;
3638		}
3639	}
3640
3641	if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3642	    !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3643		ic->mode = argv[3][0];
3644	} else {
3645		ti->error = "Invalid mode (expecting J, B, D, R)";
3646		r = -EINVAL;
3647		goto bad;
3648	}
3649
3650	journal_sectors = 0;
3651	interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3652	buffer_sectors = DEFAULT_BUFFER_SECTORS;
3653	journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3654	sync_msec = DEFAULT_SYNC_MSEC;
3655	ic->sectors_per_block = 1;
3656
3657	as.argc = argc - DIRECT_ARGUMENTS;
3658	as.argv = argv + DIRECT_ARGUMENTS;
3659	r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3660	if (r)
3661		goto bad;
3662
3663	while (extra_args--) {
3664		const char *opt_string;
3665		unsigned val;
3666		unsigned long long llval;
3667		opt_string = dm_shift_arg(&as);
3668		if (!opt_string) {
3669			r = -EINVAL;
3670			ti->error = "Not enough feature arguments";
3671			goto bad;
3672		}
3673		if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3674			journal_sectors = val ? val : 1;
3675		else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3676			interleave_sectors = val;
3677		else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3678			buffer_sectors = val;
3679		else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3680			journal_watermark = val;
3681		else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3682			sync_msec = val;
3683		else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3684			if (ic->meta_dev) {
3685				dm_put_device(ti, ic->meta_dev);
3686				ic->meta_dev = NULL;
3687			}
3688			r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3689					  dm_table_get_mode(ti->table), &ic->meta_dev);
3690			if (r) {
3691				ti->error = "Device lookup failed";
3692				goto bad;
3693			}
3694		} else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3695			if (val < 1 << SECTOR_SHIFT ||
3696			    val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3697			    (val & (val -1))) {
3698				r = -EINVAL;
3699				ti->error = "Invalid block_size argument";
3700				goto bad;
3701			}
3702			ic->sectors_per_block = val >> SECTOR_SHIFT;
3703		} else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3704			log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3705		} else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3706			if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3707				r = -EINVAL;
3708				ti->error = "Invalid bitmap_flush_interval argument";
3709			}
3710			ic->bitmap_flush_interval = msecs_to_jiffies(val);
3711		} else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3712			r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3713					    "Invalid internal_hash argument");
3714			if (r)
3715				goto bad;
3716		} else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3717			r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3718					    "Invalid journal_crypt argument");
3719			if (r)
3720				goto bad;
3721		} else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3722			r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3723					    "Invalid journal_mac argument");
3724			if (r)
3725				goto bad;
3726		} else if (!strcmp(opt_string, "recalculate")) {
3727			ic->recalculate_flag = true;
3728		} else {
3729			r = -EINVAL;
3730			ti->error = "Invalid argument";
3731			goto bad;
3732		}
3733	}
3734
3735	ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3736	if (!ic->meta_dev)
3737		ic->meta_device_sectors = ic->data_device_sectors;
3738	else
3739		ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3740
3741	if (!journal_sectors) {
3742		journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3743				      ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3744	}
3745
3746	if (!buffer_sectors)
3747		buffer_sectors = 1;
3748	ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3749
3750	r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3751		    "Invalid internal hash", "Error setting internal hash key");
3752	if (r)
3753		goto bad;
3754
3755	r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3756		    "Invalid journal mac", "Error setting journal mac key");
3757	if (r)
3758		goto bad;
3759
3760	if (!ic->tag_size) {
3761		if (!ic->internal_hash) {
3762			ti->error = "Unknown tag size";
3763			r = -EINVAL;
3764			goto bad;
3765		}
3766		ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3767	}
3768	if (ic->tag_size > MAX_TAG_SIZE) {
3769		ti->error = "Too big tag size";
3770		r = -EINVAL;
3771		goto bad;
3772	}
3773	if (!(ic->tag_size & (ic->tag_size - 1)))
3774		ic->log2_tag_size = __ffs(ic->tag_size);
3775	else
3776		ic->log2_tag_size = -1;
3777
3778	if (ic->mode == 'B' && !ic->internal_hash) {
3779		r = -EINVAL;
3780		ti->error = "Bitmap mode can be only used with internal hash";
3781		goto bad;
3782	}
3783
3784	ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3785	ic->autocommit_msec = sync_msec;
3786	timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3787
3788	ic->io = dm_io_client_create();
3789	if (IS_ERR(ic->io)) {
3790		r = PTR_ERR(ic->io);
3791		ic->io = NULL;
3792		ti->error = "Cannot allocate dm io";
3793		goto bad;
3794	}
3795
3796	r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3797	if (r) {
3798		ti->error = "Cannot allocate mempool";
3799		goto bad;
3800	}
3801
3802	ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3803					  WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3804	if (!ic->metadata_wq) {
3805		ti->error = "Cannot allocate workqueue";
3806		r = -ENOMEM;
3807		goto bad;
3808	}
3809
3810	/*
3811	 * If this workqueue were percpu, it would cause bio reordering
3812	 * and reduced performance.
3813	 */
3814	ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3815	if (!ic->wait_wq) {
3816		ti->error = "Cannot allocate workqueue";
3817		r = -ENOMEM;
3818		goto bad;
3819	}
3820
3821	ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3822	if (!ic->commit_wq) {
3823		ti->error = "Cannot allocate workqueue";
3824		r = -ENOMEM;
3825		goto bad;
3826	}
3827	INIT_WORK(&ic->commit_work, integrity_commit);
3828
3829	if (ic->mode == 'J' || ic->mode == 'B') {
3830		ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3831		if (!ic->writer_wq) {
3832			ti->error = "Cannot allocate workqueue";
3833			r = -ENOMEM;
3834			goto bad;
3835		}
3836		INIT_WORK(&ic->writer_work, integrity_writer);
3837	}
3838
3839	ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3840	if (!ic->sb) {
3841		r = -ENOMEM;
3842		ti->error = "Cannot allocate superblock area";
3843		goto bad;
3844	}
3845
3846	r = sync_rw_sb(ic, REQ_OP_READ, 0);
3847	if (r) {
3848		ti->error = "Error reading superblock";
3849		goto bad;
3850	}
3851	should_write_sb = false;
3852	if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3853		if (ic->mode != 'R') {
3854			if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3855				r = -EINVAL;
3856				ti->error = "The device is not initialized";
3857				goto bad;
3858			}
3859		}
3860
3861		r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3862		if (r) {
3863			ti->error = "Could not initialize superblock";
3864			goto bad;
3865		}
3866		if (ic->mode != 'R')
3867			should_write_sb = true;
3868	}
3869
3870	if (!ic->sb->version || ic->sb->version > SB_VERSION_3) {
3871		r = -EINVAL;
3872		ti->error = "Unknown version";
3873		goto bad;
3874	}
3875	if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3876		r = -EINVAL;
3877		ti->error = "Tag size doesn't match the information in superblock";
3878		goto bad;
3879	}
3880	if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3881		r = -EINVAL;
3882		ti->error = "Block size doesn't match the information in superblock";
3883		goto bad;
3884	}
3885	if (!le32_to_cpu(ic->sb->journal_sections)) {
3886		r = -EINVAL;
3887		ti->error = "Corrupted superblock, journal_sections is 0";
3888		goto bad;
3889	}
3890	/* make sure that ti->max_io_len doesn't overflow */
3891	if (!ic->meta_dev) {
3892		if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3893		    ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3894			r = -EINVAL;
3895			ti->error = "Invalid interleave_sectors in the superblock";
3896			goto bad;
3897		}
3898	} else {
3899		if (ic->sb->log2_interleave_sectors) {
3900			r = -EINVAL;
3901			ti->error = "Invalid interleave_sectors in the superblock";
3902			goto bad;
3903		}
3904	}
3905	ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3906	if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3907		/* test for overflow */
3908		r = -EINVAL;
3909		ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3910		goto bad;
3911	}
3912	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3913		r = -EINVAL;
3914		ti->error = "Journal mac mismatch";
3915		goto bad;
3916	}
3917
3918try_smaller_buffer:
3919	r = calculate_device_limits(ic);
3920	if (r) {
3921		if (ic->meta_dev) {
3922			if (ic->log2_buffer_sectors > 3) {
3923				ic->log2_buffer_sectors--;
3924				goto try_smaller_buffer;
3925			}
3926		}
3927		ti->error = "The device is too small";
3928		goto bad;
3929	}
3930
3931	if (log2_sectors_per_bitmap_bit < 0)
3932		log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
3933	if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
3934		log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
3935
3936	bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
3937	if (bits_in_journal > UINT_MAX)
3938		bits_in_journal = UINT_MAX;
3939	while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
3940		log2_sectors_per_bitmap_bit++;
3941
3942	log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
3943	ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3944	if (should_write_sb) {
3945		ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3946	}
3947	n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
3948				+ (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
3949	ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
3950
3951	if (!ic->meta_dev)
3952		ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3953
3954	if (ti->len > ic->provided_data_sectors) {
3955		r = -EINVAL;
3956		ti->error = "Not enough provided sectors for requested mapping size";
3957		goto bad;
3958	}
3959
3960
3961	threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3962	threshold += 50;
3963	do_div(threshold, 100);
3964	ic->free_sectors_threshold = threshold;
3965
3966	DEBUG_print("initialized:\n");
3967	DEBUG_print("	integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3968	DEBUG_print("	journal_entry_size %u\n", ic->journal_entry_size);
3969	DEBUG_print("	journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3970	DEBUG_print("	journal_section_entries %u\n", ic->journal_section_entries);
3971	DEBUG_print("	journal_section_sectors %u\n", ic->journal_section_sectors);
3972	DEBUG_print("	journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3973	DEBUG_print("	journal_entries %u\n", ic->journal_entries);
3974	DEBUG_print("	log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3975	DEBUG_print("	data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
3976	DEBUG_print("	initial_sectors 0x%x\n", ic->initial_sectors);
3977	DEBUG_print("	metadata_run 0x%x\n", ic->metadata_run);
3978	DEBUG_print("	log2_metadata_run %d\n", ic->log2_metadata_run);
3979	DEBUG_print("	provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3980		    (unsigned long long)ic->provided_data_sectors);
3981	DEBUG_print("	log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3982	DEBUG_print("	bits_in_journal %llu\n", (unsigned long long)bits_in_journal);
3983
3984	if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3985		ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3986		ic->sb->recalc_sector = cpu_to_le64(0);
3987	}
3988
3989	if (ic->internal_hash) {
3990		ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
3991		if (!ic->recalc_wq ) {
3992			ti->error = "Cannot allocate workqueue";
3993			r = -ENOMEM;
3994			goto bad;
3995		}
3996		INIT_WORK(&ic->recalc_work, integrity_recalc);
3997		ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3998		if (!ic->recalc_buffer) {
3999			ti->error = "Cannot allocate buffer for recalculating";
4000			r = -ENOMEM;
4001			goto bad;
4002		}
4003		ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4004						 ic->tag_size, GFP_KERNEL);
4005		if (!ic->recalc_tags) {
4006			ti->error = "Cannot allocate tags for recalculating";
4007			r = -ENOMEM;
4008			goto bad;
4009		}
4010	}
4011
4012	ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4013			1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4014	if (IS_ERR(ic->bufio)) {
4015		r = PTR_ERR(ic->bufio);
4016		ti->error = "Cannot initialize dm-bufio";
4017		ic->bufio = NULL;
4018		goto bad;
4019	}
4020	dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4021
4022	if (ic->mode != 'R') {
4023		r = create_journal(ic, &ti->error);
4024		if (r)
4025			goto bad;
4026
4027	}
4028
4029	if (ic->mode == 'B') {
4030		unsigned i;
4031		unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4032
4033		ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4034		if (!ic->recalc_bitmap) {
4035			r = -ENOMEM;
4036			goto bad;
4037		}
4038		ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4039		if (!ic->may_write_bitmap) {
4040			r = -ENOMEM;
4041			goto bad;
4042		}
4043		ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4044		if (!ic->bbs) {
4045			r = -ENOMEM;
4046			goto bad;
4047		}
4048		INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4049		for (i = 0; i < ic->n_bitmap_blocks; i++) {
4050			struct bitmap_block_status *bbs = &ic->bbs[i];
4051			unsigned sector, pl_index, pl_offset;
4052
4053			INIT_WORK(&bbs->work, bitmap_block_work);
4054			bbs->ic = ic;
4055			bbs->idx = i;
4056			bio_list_init(&bbs->bio_queue);
4057			spin_lock_init(&bbs->bio_queue_lock);
4058
4059			sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4060			pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4061			pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4062
4063			bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4064		}
4065	}
4066
4067	if (should_write_sb) {
4068		int r;
4069
4070		init_journal(ic, 0, ic->journal_sections, 0);
4071		r = dm_integrity_failed(ic);
4072		if (unlikely(r)) {
4073			ti->error = "Error initializing journal";
4074			goto bad;
4075		}
4076		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4077		if (r) {
4078			ti->error = "Error initializing superblock";
4079			goto bad;
4080		}
4081		ic->just_formatted = true;
4082	}
4083
4084	if (!ic->meta_dev) {
4085		r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4086		if (r)
4087			goto bad;
4088	}
4089	if (ic->mode == 'B') {
4090		unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4091		if (!max_io_len)
4092			max_io_len = 1U << 31;
4093		DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4094		if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4095			r = dm_set_target_max_io_len(ti, max_io_len);
4096			if (r)
4097				goto bad;
4098		}
4099	}
4100
4101	if (!ic->internal_hash)
4102		dm_integrity_set(ti, ic);
4103
4104	ti->num_flush_bios = 1;
4105	ti->flush_supported = true;
4106
4107	return 0;
4108
4109bad:
4110	dm_integrity_dtr(ti);
4111	return r;
4112}
4113
4114static void dm_integrity_dtr(struct dm_target *ti)
4115{
4116	struct dm_integrity_c *ic = ti->private;
4117
4118	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4119	BUG_ON(!list_empty(&ic->wait_list));
4120
4121	if (ic->metadata_wq)
4122		destroy_workqueue(ic->metadata_wq);
4123	if (ic->wait_wq)
4124		destroy_workqueue(ic->wait_wq);
4125	if (ic->commit_wq)
4126		destroy_workqueue(ic->commit_wq);
4127	if (ic->writer_wq)
4128		destroy_workqueue(ic->writer_wq);
4129	if (ic->recalc_wq)
4130		destroy_workqueue(ic->recalc_wq);
4131	vfree(ic->recalc_buffer);
4132	kvfree(ic->recalc_tags);
4133	kvfree(ic->bbs);
4134	if (ic->bufio)
4135		dm_bufio_client_destroy(ic->bufio);
4136	mempool_exit(&ic->journal_io_mempool);
4137	if (ic->io)
4138		dm_io_client_destroy(ic->io);
4139	if (ic->dev)
4140		dm_put_device(ti, ic->dev);
4141	if (ic->meta_dev)
4142		dm_put_device(ti, ic->meta_dev);
4143	dm_integrity_free_page_list(ic->journal);
4144	dm_integrity_free_page_list(ic->journal_io);
4145	dm_integrity_free_page_list(ic->journal_xor);
4146	dm_integrity_free_page_list(ic->recalc_bitmap);
4147	dm_integrity_free_page_list(ic->may_write_bitmap);
4148	if (ic->journal_scatterlist)
4149		dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4150	if (ic->journal_io_scatterlist)
4151		dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4152	if (ic->sk_requests) {
4153		unsigned i;
4154
4155		for (i = 0; i < ic->journal_sections; i++) {
4156			struct skcipher_request *req = ic->sk_requests[i];
4157			if (req) {
4158				kzfree(req->iv);
4159				skcipher_request_free(req);
4160			}
4161		}
4162		kvfree(ic->sk_requests);
4163	}
4164	kvfree(ic->journal_tree);
4165	if (ic->sb)
4166		free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4167
4168	if (ic->internal_hash)
4169		crypto_free_shash(ic->internal_hash);
4170	free_alg(&ic->internal_hash_alg);
4171
4172	if (ic->journal_crypt)
4173		crypto_free_skcipher(ic->journal_crypt);
4174	free_alg(&ic->journal_crypt_alg);
4175
4176	if (ic->journal_mac)
4177		crypto_free_shash(ic->journal_mac);
4178	free_alg(&ic->journal_mac_alg);
4179
4180	kfree(ic);
4181}
4182
4183static struct target_type integrity_target = {
4184	.name			= "integrity",
4185	.version		= {1, 3, 0},
4186	.module			= THIS_MODULE,
4187	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4188	.ctr			= dm_integrity_ctr,
4189	.dtr			= dm_integrity_dtr,
4190	.map			= dm_integrity_map,
4191	.postsuspend		= dm_integrity_postsuspend,
4192	.resume			= dm_integrity_resume,
4193	.status			= dm_integrity_status,
4194	.iterate_devices	= dm_integrity_iterate_devices,
4195	.io_hints		= dm_integrity_io_hints,
4196};
4197
4198static int __init dm_integrity_init(void)
4199{
4200	int r;
4201
4202	journal_io_cache = kmem_cache_create("integrity_journal_io",
4203					     sizeof(struct journal_io), 0, 0, NULL);
4204	if (!journal_io_cache) {
4205		DMERR("can't allocate journal io cache");
4206		return -ENOMEM;
4207	}
4208
4209	r = dm_register_target(&integrity_target);
4210
4211	if (r < 0)
4212		DMERR("register failed %d", r);
4213
4214	return r;
4215}
4216
4217static void __exit dm_integrity_exit(void)
4218{
4219	dm_unregister_target(&integrity_target);
4220	kmem_cache_destroy(journal_io_cache);
4221}
4222
4223module_init(dm_integrity_init);
4224module_exit(dm_integrity_exit);
4225
4226MODULE_AUTHOR("Milan Broz");
4227MODULE_AUTHOR("Mikulas Patocka");
4228MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4229MODULE_LICENSE("GPL");