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 "dm-bio-record.h"
  10
  11#include <linux/compiler.h>
  12#include <linux/module.h>
  13#include <linux/device-mapper.h>
  14#include <linux/dm-io.h>
  15#include <linux/vmalloc.h>
  16#include <linux/sort.h>
  17#include <linux/rbtree.h>
  18#include <linux/delay.h>
  19#include <linux/random.h>
  20#include <linux/reboot.h>
  21#include <crypto/hash.h>
  22#include <crypto/skcipher.h>
  23#include <linux/async_tx.h>
  24#include <linux/dm-bufio.h>
  25
  26#define DM_MSG_PREFIX "integrity"
  27
  28#define DEFAULT_INTERLEAVE_SECTORS	32768
  29#define DEFAULT_JOURNAL_SIZE_FACTOR	7
  30#define DEFAULT_SECTORS_PER_BITMAP_BIT	32768
  31#define DEFAULT_BUFFER_SECTORS		128
  32#define DEFAULT_JOURNAL_WATERMARK	50
  33#define DEFAULT_SYNC_MSEC		10000
  34#define DEFAULT_MAX_JOURNAL_SECTORS	131072
  35#define MIN_LOG2_INTERLEAVE_SECTORS	3
  36#define MAX_LOG2_INTERLEAVE_SECTORS	31
  37#define METADATA_WORKQUEUE_MAX_ACTIVE	16
  38#define RECALC_SECTORS			32768
  39#define RECALC_WRITE_SUPER		16
  40#define BITMAP_BLOCK_SIZE		4096	/* don't change it */
  41#define BITMAP_FLUSH_INTERVAL		(10 * HZ)
  42#define DISCARD_FILLER			0xf6
  43#define SALT_SIZE			16
  44
  45/*
  46 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  47 * so it should not be enabled in the official kernel
  48 */
  49//#define DEBUG_PRINT
  50//#define INTERNAL_VERIFY
  51
  52/*
  53 * On disk structures
  54 */
  55
  56#define SB_MAGIC			"integrt"
  57#define SB_VERSION_1			1
  58#define SB_VERSION_2			2
  59#define SB_VERSION_3			3
  60#define SB_VERSION_4			4
  61#define SB_VERSION_5			5
  62#define SB_SECTORS			8
  63#define MAX_SECTORS_PER_BLOCK		8
  64
  65struct superblock {
  66	__u8 magic[8];
  67	__u8 version;
  68	__u8 log2_interleave_sectors;
  69	__le16 integrity_tag_size;
  70	__le32 journal_sections;
  71	__le64 provided_data_sectors;	/* userspace uses this value */
  72	__le32 flags;
  73	__u8 log2_sectors_per_block;
  74	__u8 log2_blocks_per_bitmap_bit;
  75	__u8 pad[2];
  76	__le64 recalc_sector;
  77	__u8 pad2[8];
  78	__u8 salt[SALT_SIZE];
  79};
  80
  81#define SB_FLAG_HAVE_JOURNAL_MAC	0x1
  82#define SB_FLAG_RECALCULATING		0x2
  83#define SB_FLAG_DIRTY_BITMAP		0x4
  84#define SB_FLAG_FIXED_PADDING		0x8
  85#define SB_FLAG_FIXED_HMAC		0x10
  86
  87#define	JOURNAL_ENTRY_ROUNDUP		8
  88
  89typedef __le64 commit_id_t;
  90#define JOURNAL_MAC_PER_SECTOR		8
  91
  92struct journal_entry {
  93	union {
  94		struct {
  95			__le32 sector_lo;
  96			__le32 sector_hi;
  97		} s;
  98		__le64 sector;
  99	} u;
 100	commit_id_t last_bytes[];
 101	/* __u8 tag[0]; */
 102};
 103
 104#define journal_entry_tag(ic, je)		((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 105
 106#if BITS_PER_LONG == 64
 107#define journal_entry_set_sector(je, x)		do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 108#else
 109#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)
 110#endif
 111#define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
 112#define journal_entry_is_unused(je)		((je)->u.s.sector_hi == cpu_to_le32(-1))
 113#define journal_entry_set_unused(je)		do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 114#define journal_entry_is_inprogress(je)		((je)->u.s.sector_hi == cpu_to_le32(-2))
 115#define journal_entry_set_inprogress(je)	do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 116
 117#define JOURNAL_BLOCK_SECTORS		8
 118#define JOURNAL_SECTOR_DATA		((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 119#define JOURNAL_MAC_SIZE		(JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 120
 121struct journal_sector {
 122	__u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 123	__u8 mac[JOURNAL_MAC_PER_SECTOR];
 124	commit_id_t commit_id;
 125};
 126
 127#define MAX_TAG_SIZE			(JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 128
 129#define METADATA_PADDING_SECTORS	8
 130
 131#define N_COMMIT_IDS			4
 132
 133static unsigned char prev_commit_seq(unsigned char seq)
 134{
 135	return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 136}
 137
 138static unsigned char next_commit_seq(unsigned char seq)
 139{
 140	return (seq + 1) % N_COMMIT_IDS;
 141}
 142
 143/*
 144 * In-memory structures
 145 */
 146
 147struct journal_node {
 148	struct rb_node node;
 149	sector_t sector;
 150};
 151
 152struct alg_spec {
 153	char *alg_string;
 154	char *key_string;
 155	__u8 *key;
 156	unsigned key_size;
 157};
 158
 159struct dm_integrity_c {
 160	struct dm_dev *dev;
 161	struct dm_dev *meta_dev;
 162	unsigned tag_size;
 163	__s8 log2_tag_size;
 164	sector_t start;
 165	mempool_t journal_io_mempool;
 166	struct dm_io_client *io;
 167	struct dm_bufio_client *bufio;
 168	struct workqueue_struct *metadata_wq;
 169	struct superblock *sb;
 170	unsigned journal_pages;
 171	unsigned n_bitmap_blocks;
 172
 173	struct page_list *journal;
 174	struct page_list *journal_io;
 175	struct page_list *journal_xor;
 176	struct page_list *recalc_bitmap;
 177	struct page_list *may_write_bitmap;
 178	struct bitmap_block_status *bbs;
 179	unsigned bitmap_flush_interval;
 180	int synchronous_mode;
 181	struct bio_list synchronous_bios;
 182	struct delayed_work bitmap_flush_work;
 183
 184	struct crypto_skcipher *journal_crypt;
 185	struct scatterlist **journal_scatterlist;
 186	struct scatterlist **journal_io_scatterlist;
 187	struct skcipher_request **sk_requests;
 188
 189	struct crypto_shash *journal_mac;
 190
 191	struct journal_node *journal_tree;
 192	struct rb_root journal_tree_root;
 193
 194	sector_t provided_data_sectors;
 195
 196	unsigned short journal_entry_size;
 197	unsigned char journal_entries_per_sector;
 198	unsigned char journal_section_entries;
 199	unsigned short journal_section_sectors;
 200	unsigned journal_sections;
 201	unsigned journal_entries;
 202	sector_t data_device_sectors;
 203	sector_t meta_device_sectors;
 204	unsigned initial_sectors;
 205	unsigned metadata_run;
 206	__s8 log2_metadata_run;
 207	__u8 log2_buffer_sectors;
 208	__u8 sectors_per_block;
 209	__u8 log2_blocks_per_bitmap_bit;
 210
 211	unsigned char mode;
 212
 213	int failed;
 214
 215	struct crypto_shash *internal_hash;
 216
 217	struct dm_target *ti;
 218
 219	/* these variables are locked with endio_wait.lock */
 220	struct rb_root in_progress;
 221	struct list_head wait_list;
 222	wait_queue_head_t endio_wait;
 223	struct workqueue_struct *wait_wq;
 224	struct workqueue_struct *offload_wq;
 225
 226	unsigned char commit_seq;
 227	commit_id_t commit_ids[N_COMMIT_IDS];
 228
 229	unsigned committed_section;
 230	unsigned n_committed_sections;
 231
 232	unsigned uncommitted_section;
 233	unsigned n_uncommitted_sections;
 234
 235	unsigned free_section;
 236	unsigned char free_section_entry;
 237	unsigned free_sectors;
 238
 239	unsigned free_sectors_threshold;
 240
 241	struct workqueue_struct *commit_wq;
 242	struct work_struct commit_work;
 243
 244	struct workqueue_struct *writer_wq;
 245	struct work_struct writer_work;
 246
 247	struct workqueue_struct *recalc_wq;
 248	struct work_struct recalc_work;
 249	u8 *recalc_buffer;
 250	u8 *recalc_tags;
 251
 252	struct bio_list flush_bio_list;
 253
 254	unsigned long autocommit_jiffies;
 255	struct timer_list autocommit_timer;
 256	unsigned autocommit_msec;
 257
 258	wait_queue_head_t copy_to_journal_wait;
 259
 260	struct completion crypto_backoff;
 261
 262	bool journal_uptodate;
 263	bool just_formatted;
 264	bool recalculate_flag;
 265	bool reset_recalculate_flag;
 266	bool discard;
 267	bool fix_padding;
 268	bool fix_hmac;
 269	bool legacy_recalculate;
 270
 271	struct alg_spec internal_hash_alg;
 272	struct alg_spec journal_crypt_alg;
 273	struct alg_spec journal_mac_alg;
 274
 275	atomic64_t number_of_mismatches;
 276
 277	struct notifier_block reboot_notifier;
 278};
 279
 280struct dm_integrity_range {
 281	sector_t logical_sector;
 282	sector_t n_sectors;
 283	bool waiting;
 284	union {
 285		struct rb_node node;
 286		struct {
 287			struct task_struct *task;
 288			struct list_head wait_entry;
 289		};
 290	};
 291};
 292
 293struct dm_integrity_io {
 294	struct work_struct work;
 295
 296	struct dm_integrity_c *ic;
 297	enum req_opf op;
 298	bool fua;
 299
 300	struct dm_integrity_range range;
 301
 302	sector_t metadata_block;
 303	unsigned metadata_offset;
 304
 305	atomic_t in_flight;
 306	blk_status_t bi_status;
 307
 308	struct completion *completion;
 309
 310	struct dm_bio_details bio_details;
 311};
 312
 313struct journal_completion {
 314	struct dm_integrity_c *ic;
 315	atomic_t in_flight;
 316	struct completion comp;
 317};
 318
 319struct journal_io {
 320	struct dm_integrity_range range;
 321	struct journal_completion *comp;
 322};
 323
 324struct bitmap_block_status {
 325	struct work_struct work;
 326	struct dm_integrity_c *ic;
 327	unsigned idx;
 328	unsigned long *bitmap;
 329	struct bio_list bio_queue;
 330	spinlock_t bio_queue_lock;
 331
 332};
 333
 334static struct kmem_cache *journal_io_cache;
 335
 336#define JOURNAL_IO_MEMPOOL	32
 337
 338#ifdef DEBUG_PRINT
 339#define DEBUG_print(x, ...)	printk(KERN_DEBUG x, ##__VA_ARGS__)
 340static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 341{
 342	va_list args;
 343	va_start(args, msg);
 344	vprintk(msg, args);
 345	va_end(args);
 346	if (len)
 347		pr_cont(":");
 348	while (len) {
 349		pr_cont(" %02x", *bytes);
 350		bytes++;
 351		len--;
 352	}
 353	pr_cont("\n");
 354}
 355#define DEBUG_bytes(bytes, len, msg, ...)	__DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 356#else
 357#define DEBUG_print(x, ...)			do { } while (0)
 358#define DEBUG_bytes(bytes, len, msg, ...)	do { } while (0)
 359#endif
 360
 361static void dm_integrity_prepare(struct request *rq)
 362{
 363}
 364
 365static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 366{
 367}
 368
 369/*
 370 * DM Integrity profile, protection is performed layer above (dm-crypt)
 371 */
 372static const struct blk_integrity_profile dm_integrity_profile = {
 373	.name			= "DM-DIF-EXT-TAG",
 374	.generate_fn		= NULL,
 375	.verify_fn		= NULL,
 376	.prepare_fn		= dm_integrity_prepare,
 377	.complete_fn		= dm_integrity_complete,
 378};
 379
 380static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 381static void integrity_bio_wait(struct work_struct *w);
 382static void dm_integrity_dtr(struct dm_target *ti);
 383
 384static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 385{
 386	if (err == -EILSEQ)
 387		atomic64_inc(&ic->number_of_mismatches);
 388	if (!cmpxchg(&ic->failed, 0, err))
 389		DMERR("Error on %s: %d", msg, err);
 390}
 391
 392static int dm_integrity_failed(struct dm_integrity_c *ic)
 393{
 394	return READ_ONCE(ic->failed);
 395}
 396
 397static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 398{
 399	if (ic->legacy_recalculate)
 400		return false;
 401	if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
 402	    ic->internal_hash_alg.key || ic->journal_mac_alg.key :
 403	    ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
 404		return true;
 405	return false;
 406}
 407
 408static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 409					  unsigned j, unsigned char seq)
 410{
 411	/*
 412	 * Xor the number with section and sector, so that if a piece of
 413	 * journal is written at wrong place, it is detected.
 414	 */
 415	return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 416}
 417
 418static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 419				sector_t *area, sector_t *offset)
 420{
 421	if (!ic->meta_dev) {
 422		__u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 423		*area = data_sector >> log2_interleave_sectors;
 424		*offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 425	} else {
 426		*area = 0;
 427		*offset = data_sector;
 428	}
 429}
 430
 431#define sector_to_block(ic, n)						\
 432do {									\
 433	BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));		\
 434	(n) >>= (ic)->sb->log2_sectors_per_block;			\
 435} while (0)
 436
 437static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 438					    sector_t offset, unsigned *metadata_offset)
 439{
 440	__u64 ms;
 441	unsigned mo;
 442
 443	ms = area << ic->sb->log2_interleave_sectors;
 444	if (likely(ic->log2_metadata_run >= 0))
 445		ms += area << ic->log2_metadata_run;
 446	else
 447		ms += area * ic->metadata_run;
 448	ms >>= ic->log2_buffer_sectors;
 449
 450	sector_to_block(ic, offset);
 451
 452	if (likely(ic->log2_tag_size >= 0)) {
 453		ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 454		mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 455	} else {
 456		ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 457		mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 458	}
 459	*metadata_offset = mo;
 460	return ms;
 461}
 462
 463static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 464{
 465	sector_t result;
 466
 467	if (ic->meta_dev)
 468		return offset;
 469
 470	result = area << ic->sb->log2_interleave_sectors;
 471	if (likely(ic->log2_metadata_run >= 0))
 472		result += (area + 1) << ic->log2_metadata_run;
 473	else
 474		result += (area + 1) * ic->metadata_run;
 475
 476	result += (sector_t)ic->initial_sectors + offset;
 477	result += ic->start;
 478
 479	return result;
 480}
 481
 482static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 483{
 484	if (unlikely(*sec_ptr >= ic->journal_sections))
 485		*sec_ptr -= ic->journal_sections;
 486}
 487
 488static void sb_set_version(struct dm_integrity_c *ic)
 489{
 490	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
 491		ic->sb->version = SB_VERSION_5;
 492	else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 493		ic->sb->version = SB_VERSION_4;
 494	else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 495		ic->sb->version = SB_VERSION_3;
 496	else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 497		ic->sb->version = SB_VERSION_2;
 498	else
 499		ic->sb->version = SB_VERSION_1;
 500}
 501
 502static int sb_mac(struct dm_integrity_c *ic, bool wr)
 503{
 504	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 505	int r;
 506	unsigned size = crypto_shash_digestsize(ic->journal_mac);
 507
 508	if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
 509		dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 510		return -EINVAL;
 511	}
 512
 513	desc->tfm = ic->journal_mac;
 514
 515	r = crypto_shash_init(desc);
 516	if (unlikely(r < 0)) {
 517		dm_integrity_io_error(ic, "crypto_shash_init", r);
 518		return r;
 519	}
 520
 521	r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
 522	if (unlikely(r < 0)) {
 523		dm_integrity_io_error(ic, "crypto_shash_update", r);
 524		return r;
 525	}
 526
 527	if (likely(wr)) {
 528		r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
 529		if (unlikely(r < 0)) {
 530			dm_integrity_io_error(ic, "crypto_shash_final", r);
 531			return r;
 532		}
 533	} else {
 534		__u8 result[HASH_MAX_DIGESTSIZE];
 535		r = crypto_shash_final(desc, result);
 536		if (unlikely(r < 0)) {
 537			dm_integrity_io_error(ic, "crypto_shash_final", r);
 538			return r;
 539		}
 540		if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
 541			dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 542			return -EILSEQ;
 543		}
 544	}
 545
 546	return 0;
 547}
 548
 549static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 550{
 551	struct dm_io_request io_req;
 552	struct dm_io_region io_loc;
 553	int r;
 554
 555	io_req.bi_op = op;
 556	io_req.bi_op_flags = op_flags;
 557	io_req.mem.type = DM_IO_KMEM;
 558	io_req.mem.ptr.addr = ic->sb;
 559	io_req.notify.fn = NULL;
 560	io_req.client = ic->io;
 561	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 562	io_loc.sector = ic->start;
 563	io_loc.count = SB_SECTORS;
 564
 565	if (op == REQ_OP_WRITE) {
 566		sb_set_version(ic);
 567		if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 568			r = sb_mac(ic, true);
 569			if (unlikely(r))
 570				return r;
 571		}
 572	}
 573
 574	r = dm_io(&io_req, 1, &io_loc, NULL);
 575	if (unlikely(r))
 576		return r;
 577
 578	if (op == REQ_OP_READ) {
 579		if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 580			r = sb_mac(ic, false);
 581			if (unlikely(r))
 582				return r;
 583		}
 584	}
 585
 586	return 0;
 587}
 588
 589#define BITMAP_OP_TEST_ALL_SET		0
 590#define BITMAP_OP_TEST_ALL_CLEAR	1
 591#define BITMAP_OP_SET			2
 592#define BITMAP_OP_CLEAR			3
 593
 594static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 595			    sector_t sector, sector_t n_sectors, int mode)
 596{
 597	unsigned long bit, end_bit, this_end_bit, page, end_page;
 598	unsigned long *data;
 599
 600	if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 601		DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 602			sector,
 603			n_sectors,
 604			ic->sb->log2_sectors_per_block,
 605			ic->log2_blocks_per_bitmap_bit,
 606			mode);
 607		BUG();
 608	}
 609
 610	if (unlikely(!n_sectors))
 611		return true;
 612
 613	bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 614	end_bit = (sector + n_sectors - 1) >>
 615		(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 616
 617	page = bit / (PAGE_SIZE * 8);
 618	bit %= PAGE_SIZE * 8;
 619
 620	end_page = end_bit / (PAGE_SIZE * 8);
 621	end_bit %= PAGE_SIZE * 8;
 622
 623repeat:
 624	if (page < end_page) {
 625		this_end_bit = PAGE_SIZE * 8 - 1;
 626	} else {
 627		this_end_bit = end_bit;
 628	}
 629
 630	data = lowmem_page_address(bitmap[page].page);
 631
 632	if (mode == BITMAP_OP_TEST_ALL_SET) {
 633		while (bit <= this_end_bit) {
 634			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 635				do {
 636					if (data[bit / BITS_PER_LONG] != -1)
 637						return false;
 638					bit += BITS_PER_LONG;
 639				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 640				continue;
 641			}
 642			if (!test_bit(bit, data))
 643				return false;
 644			bit++;
 645		}
 646	} else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 647		while (bit <= this_end_bit) {
 648			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 649				do {
 650					if (data[bit / BITS_PER_LONG] != 0)
 651						return false;
 652					bit += BITS_PER_LONG;
 653				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 654				continue;
 655			}
 656			if (test_bit(bit, data))
 657				return false;
 658			bit++;
 659		}
 660	} else if (mode == BITMAP_OP_SET) {
 661		while (bit <= this_end_bit) {
 662			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 663				do {
 664					data[bit / BITS_PER_LONG] = -1;
 665					bit += BITS_PER_LONG;
 666				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 667				continue;
 668			}
 669			__set_bit(bit, data);
 670			bit++;
 671		}
 672	} else if (mode == BITMAP_OP_CLEAR) {
 673		if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 674			clear_page(data);
 675		else while (bit <= this_end_bit) {
 676			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 677				do {
 678					data[bit / BITS_PER_LONG] = 0;
 679					bit += BITS_PER_LONG;
 680				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 681				continue;
 682			}
 683			__clear_bit(bit, data);
 684			bit++;
 685		}
 686	} else {
 687		BUG();
 688	}
 689
 690	if (unlikely(page < end_page)) {
 691		bit = 0;
 692		page++;
 693		goto repeat;
 694	}
 695
 696	return true;
 697}
 698
 699static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 700{
 701	unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 702	unsigned i;
 703
 704	for (i = 0; i < n_bitmap_pages; i++) {
 705		unsigned long *dst_data = lowmem_page_address(dst[i].page);
 706		unsigned long *src_data = lowmem_page_address(src[i].page);
 707		copy_page(dst_data, src_data);
 708	}
 709}
 710
 711static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 712{
 713	unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 714	unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 715
 716	BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 717	return &ic->bbs[bitmap_block];
 718}
 719
 720static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 721				 bool e, const char *function)
 722{
 723#if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 724	unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 725
 726	if (unlikely(section >= ic->journal_sections) ||
 727	    unlikely(offset >= limit)) {
 728		DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 729		       function, section, offset, ic->journal_sections, limit);
 730		BUG();
 731	}
 732#endif
 733}
 734
 735static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 736			       unsigned *pl_index, unsigned *pl_offset)
 737{
 738	unsigned sector;
 739
 740	access_journal_check(ic, section, offset, false, "page_list_location");
 741
 742	sector = section * ic->journal_section_sectors + offset;
 743
 744	*pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 745	*pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 746}
 747
 748static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 749					       unsigned section, unsigned offset, unsigned *n_sectors)
 750{
 751	unsigned pl_index, pl_offset;
 752	char *va;
 753
 754	page_list_location(ic, section, offset, &pl_index, &pl_offset);
 755
 756	if (n_sectors)
 757		*n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 758
 759	va = lowmem_page_address(pl[pl_index].page);
 760
 761	return (struct journal_sector *)(va + pl_offset);
 762}
 763
 764static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 765{
 766	return access_page_list(ic, ic->journal, section, offset, NULL);
 767}
 768
 769static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 770{
 771	unsigned rel_sector, offset;
 772	struct journal_sector *js;
 773
 774	access_journal_check(ic, section, n, true, "access_journal_entry");
 775
 776	rel_sector = n % JOURNAL_BLOCK_SECTORS;
 777	offset = n / JOURNAL_BLOCK_SECTORS;
 778
 779	js = access_journal(ic, section, rel_sector);
 780	return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 781}
 782
 783static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 784{
 785	n <<= ic->sb->log2_sectors_per_block;
 786
 787	n += JOURNAL_BLOCK_SECTORS;
 788
 789	access_journal_check(ic, section, n, false, "access_journal_data");
 790
 791	return access_journal(ic, section, n);
 792}
 793
 794static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 795{
 796	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 797	int r;
 798	unsigned j, size;
 799
 800	desc->tfm = ic->journal_mac;
 801
 802	r = crypto_shash_init(desc);
 803	if (unlikely(r < 0)) {
 804		dm_integrity_io_error(ic, "crypto_shash_init", r);
 805		goto err;
 806	}
 807
 808	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 809		__le64 section_le;
 810
 811		r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 812		if (unlikely(r < 0)) {
 813			dm_integrity_io_error(ic, "crypto_shash_update", r);
 814			goto err;
 815		}
 816
 817		section_le = cpu_to_le64(section);
 818		r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof section_le);
 819		if (unlikely(r < 0)) {
 820			dm_integrity_io_error(ic, "crypto_shash_update", r);
 821			goto err;
 822		}
 823	}
 824
 825	for (j = 0; j < ic->journal_section_entries; j++) {
 826		struct journal_entry *je = access_journal_entry(ic, section, j);
 827		r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 828		if (unlikely(r < 0)) {
 829			dm_integrity_io_error(ic, "crypto_shash_update", r);
 830			goto err;
 831		}
 832	}
 833
 834	size = crypto_shash_digestsize(ic->journal_mac);
 835
 836	if (likely(size <= JOURNAL_MAC_SIZE)) {
 837		r = crypto_shash_final(desc, result);
 838		if (unlikely(r < 0)) {
 839			dm_integrity_io_error(ic, "crypto_shash_final", r);
 840			goto err;
 841		}
 842		memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 843	} else {
 844		__u8 digest[HASH_MAX_DIGESTSIZE];
 845
 846		if (WARN_ON(size > sizeof(digest))) {
 847			dm_integrity_io_error(ic, "digest_size", -EINVAL);
 848			goto err;
 849		}
 850		r = crypto_shash_final(desc, digest);
 851		if (unlikely(r < 0)) {
 852			dm_integrity_io_error(ic, "crypto_shash_final", r);
 853			goto err;
 854		}
 855		memcpy(result, digest, JOURNAL_MAC_SIZE);
 856	}
 857
 858	return;
 859err:
 860	memset(result, 0, JOURNAL_MAC_SIZE);
 861}
 862
 863static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 864{
 865	__u8 result[JOURNAL_MAC_SIZE];
 866	unsigned j;
 867
 868	if (!ic->journal_mac)
 869		return;
 870
 871	section_mac(ic, section, result);
 872
 873	for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 874		struct journal_sector *js = access_journal(ic, section, j);
 875
 876		if (likely(wr))
 877			memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 878		else {
 879			if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 880				dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 881		}
 882	}
 883}
 884
 885static void complete_journal_op(void *context)
 886{
 887	struct journal_completion *comp = context;
 888	BUG_ON(!atomic_read(&comp->in_flight));
 889	if (likely(atomic_dec_and_test(&comp->in_flight)))
 890		complete(&comp->comp);
 891}
 892
 893static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 894			unsigned n_sections, struct journal_completion *comp)
 895{
 896	struct async_submit_ctl submit;
 897	size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 898	unsigned pl_index, pl_offset, section_index;
 899	struct page_list *source_pl, *target_pl;
 900
 901	if (likely(encrypt)) {
 902		source_pl = ic->journal;
 903		target_pl = ic->journal_io;
 904	} else {
 905		source_pl = ic->journal_io;
 906		target_pl = ic->journal;
 907	}
 908
 909	page_list_location(ic, section, 0, &pl_index, &pl_offset);
 910
 911	atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 912
 913	init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 914
 915	section_index = pl_index;
 916
 917	do {
 918		size_t this_step;
 919		struct page *src_pages[2];
 920		struct page *dst_page;
 921
 922		while (unlikely(pl_index == section_index)) {
 923			unsigned dummy;
 924			if (likely(encrypt))
 925				rw_section_mac(ic, section, true);
 926			section++;
 927			n_sections--;
 928			if (!n_sections)
 929				break;
 930			page_list_location(ic, section, 0, &section_index, &dummy);
 931		}
 932
 933		this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 934		dst_page = target_pl[pl_index].page;
 935		src_pages[0] = source_pl[pl_index].page;
 936		src_pages[1] = ic->journal_xor[pl_index].page;
 937
 938		async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 939
 940		pl_index++;
 941		pl_offset = 0;
 942		n_bytes -= this_step;
 943	} while (n_bytes);
 944
 945	BUG_ON(n_sections);
 946
 947	async_tx_issue_pending_all();
 948}
 949
 950static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 951{
 952	struct journal_completion *comp = req->data;
 953	if (unlikely(err)) {
 954		if (likely(err == -EINPROGRESS)) {
 955			complete(&comp->ic->crypto_backoff);
 956			return;
 957		}
 958		dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 959	}
 960	complete_journal_op(comp);
 961}
 962
 963static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 964{
 965	int r;
 966	skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 967				      complete_journal_encrypt, comp);
 968	if (likely(encrypt))
 969		r = crypto_skcipher_encrypt(req);
 970	else
 971		r = crypto_skcipher_decrypt(req);
 972	if (likely(!r))
 973		return false;
 974	if (likely(r == -EINPROGRESS))
 975		return true;
 976	if (likely(r == -EBUSY)) {
 977		wait_for_completion(&comp->ic->crypto_backoff);
 978		reinit_completion(&comp->ic->crypto_backoff);
 979		return true;
 980	}
 981	dm_integrity_io_error(comp->ic, "encrypt", r);
 982	return false;
 983}
 984
 985static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 986			  unsigned n_sections, struct journal_completion *comp)
 987{
 988	struct scatterlist **source_sg;
 989	struct scatterlist **target_sg;
 990
 991	atomic_add(2, &comp->in_flight);
 992
 993	if (likely(encrypt)) {
 994		source_sg = ic->journal_scatterlist;
 995		target_sg = ic->journal_io_scatterlist;
 996	} else {
 997		source_sg = ic->journal_io_scatterlist;
 998		target_sg = ic->journal_scatterlist;
 999	}
1000
1001	do {
1002		struct skcipher_request *req;
1003		unsigned ivsize;
1004		char *iv;
1005
1006		if (likely(encrypt))
1007			rw_section_mac(ic, section, true);
1008
1009		req = ic->sk_requests[section];
1010		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1011		iv = req->iv;
1012
1013		memcpy(iv, iv + ivsize, ivsize);
1014
1015		req->src = source_sg[section];
1016		req->dst = target_sg[section];
1017
1018		if (unlikely(do_crypt(encrypt, req, comp)))
1019			atomic_inc(&comp->in_flight);
1020
1021		section++;
1022		n_sections--;
1023	} while (n_sections);
1024
1025	atomic_dec(&comp->in_flight);
1026	complete_journal_op(comp);
1027}
1028
1029static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1030			    unsigned n_sections, struct journal_completion *comp)
1031{
1032	if (ic->journal_xor)
1033		return xor_journal(ic, encrypt, section, n_sections, comp);
1034	else
1035		return crypt_journal(ic, encrypt, section, n_sections, comp);
1036}
1037
1038static void complete_journal_io(unsigned long error, void *context)
1039{
1040	struct journal_completion *comp = context;
1041	if (unlikely(error != 0))
1042		dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1043	complete_journal_op(comp);
1044}
1045
1046static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
1047			       unsigned sector, unsigned n_sectors, struct journal_completion *comp)
1048{
1049	struct dm_io_request io_req;
1050	struct dm_io_region io_loc;
1051	unsigned pl_index, pl_offset;
1052	int r;
1053
1054	if (unlikely(dm_integrity_failed(ic))) {
1055		if (comp)
1056			complete_journal_io(-1UL, comp);
1057		return;
1058	}
1059
1060	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1061	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1062
1063	io_req.bi_op = op;
1064	io_req.bi_op_flags = op_flags;
1065	io_req.mem.type = DM_IO_PAGE_LIST;
1066	if (ic->journal_io)
1067		io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1068	else
1069		io_req.mem.ptr.pl = &ic->journal[pl_index];
1070	io_req.mem.offset = pl_offset;
1071	if (likely(comp != NULL)) {
1072		io_req.notify.fn = complete_journal_io;
1073		io_req.notify.context = comp;
1074	} else {
1075		io_req.notify.fn = NULL;
1076	}
1077	io_req.client = ic->io;
1078	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1079	io_loc.sector = ic->start + SB_SECTORS + sector;
1080	io_loc.count = n_sectors;
1081
1082	r = dm_io(&io_req, 1, &io_loc, NULL);
1083	if (unlikely(r)) {
1084		dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
1085		if (comp) {
1086			WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1087			complete_journal_io(-1UL, comp);
1088		}
1089	}
1090}
1091
1092static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
1093		       unsigned n_sections, struct journal_completion *comp)
1094{
1095	unsigned sector, n_sectors;
1096
1097	sector = section * ic->journal_section_sectors;
1098	n_sectors = n_sections * ic->journal_section_sectors;
1099
1100	rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1101}
1102
1103static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1104{
1105	struct journal_completion io_comp;
1106	struct journal_completion crypt_comp_1;
1107	struct journal_completion crypt_comp_2;
1108	unsigned i;
1109
1110	io_comp.ic = ic;
1111	init_completion(&io_comp.comp);
1112
1113	if (commit_start + commit_sections <= ic->journal_sections) {
1114		io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1115		if (ic->journal_io) {
1116			crypt_comp_1.ic = ic;
1117			init_completion(&crypt_comp_1.comp);
1118			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1119			encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1120			wait_for_completion_io(&crypt_comp_1.comp);
1121		} else {
1122			for (i = 0; i < commit_sections; i++)
1123				rw_section_mac(ic, commit_start + i, true);
1124		}
1125		rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1126			   commit_sections, &io_comp);
1127	} else {
1128		unsigned to_end;
1129		io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1130		to_end = ic->journal_sections - commit_start;
1131		if (ic->journal_io) {
1132			crypt_comp_1.ic = ic;
1133			init_completion(&crypt_comp_1.comp);
1134			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1135			encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1136			if (try_wait_for_completion(&crypt_comp_1.comp)) {
1137				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1138				reinit_completion(&crypt_comp_1.comp);
1139				crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1140				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1141				wait_for_completion_io(&crypt_comp_1.comp);
1142			} else {
1143				crypt_comp_2.ic = ic;
1144				init_completion(&crypt_comp_2.comp);
1145				crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1146				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1147				wait_for_completion_io(&crypt_comp_1.comp);
1148				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1149				wait_for_completion_io(&crypt_comp_2.comp);
1150			}
1151		} else {
1152			for (i = 0; i < to_end; i++)
1153				rw_section_mac(ic, commit_start + i, true);
1154			rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1155			for (i = 0; i < commit_sections - to_end; i++)
1156				rw_section_mac(ic, i, true);
1157		}
1158		rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1159	}
1160
1161	wait_for_completion_io(&io_comp.comp);
1162}
1163
1164static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1165			      unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1166{
1167	struct dm_io_request io_req;
1168	struct dm_io_region io_loc;
1169	int r;
1170	unsigned sector, pl_index, pl_offset;
1171
1172	BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1173
1174	if (unlikely(dm_integrity_failed(ic))) {
1175		fn(-1UL, data);
1176		return;
1177	}
1178
1179	sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1180
1181	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1182	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1183
1184	io_req.bi_op = REQ_OP_WRITE;
1185	io_req.bi_op_flags = 0;
1186	io_req.mem.type = DM_IO_PAGE_LIST;
1187	io_req.mem.ptr.pl = &ic->journal[pl_index];
1188	io_req.mem.offset = pl_offset;
1189	io_req.notify.fn = fn;
1190	io_req.notify.context = data;
1191	io_req.client = ic->io;
1192	io_loc.bdev = ic->dev->bdev;
1193	io_loc.sector = target;
1194	io_loc.count = n_sectors;
1195
1196	r = dm_io(&io_req, 1, &io_loc, NULL);
1197	if (unlikely(r)) {
1198		WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1199		fn(-1UL, data);
1200	}
1201}
1202
1203static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1204{
1205	return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1206	       range1->logical_sector + range1->n_sectors > range2->logical_sector;
1207}
1208
1209static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1210{
1211	struct rb_node **n = &ic->in_progress.rb_node;
1212	struct rb_node *parent;
1213
1214	BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1215
1216	if (likely(check_waiting)) {
1217		struct dm_integrity_range *range;
1218		list_for_each_entry(range, &ic->wait_list, wait_entry) {
1219			if (unlikely(ranges_overlap(range, new_range)))
1220				return false;
1221		}
1222	}
1223
1224	parent = NULL;
1225
1226	while (*n) {
1227		struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1228
1229		parent = *n;
1230		if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1231			n = &range->node.rb_left;
1232		} else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1233			n = &range->node.rb_right;
1234		} else {
1235			return false;
1236		}
1237	}
1238
1239	rb_link_node(&new_range->node, parent, n);
1240	rb_insert_color(&new_range->node, &ic->in_progress);
1241
1242	return true;
1243}
1244
1245static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1246{
1247	rb_erase(&range->node, &ic->in_progress);
1248	while (unlikely(!list_empty(&ic->wait_list))) {
1249		struct dm_integrity_range *last_range =
1250			list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1251		struct task_struct *last_range_task;
1252		last_range_task = last_range->task;
1253		list_del(&last_range->wait_entry);
1254		if (!add_new_range(ic, last_range, false)) {
1255			last_range->task = last_range_task;
1256			list_add(&last_range->wait_entry, &ic->wait_list);
1257			break;
1258		}
1259		last_range->waiting = false;
1260		wake_up_process(last_range_task);
1261	}
1262}
1263
1264static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1265{
1266	unsigned long flags;
1267
1268	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1269	remove_range_unlocked(ic, range);
1270	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1271}
1272
1273static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1274{
1275	new_range->waiting = true;
1276	list_add_tail(&new_range->wait_entry, &ic->wait_list);
1277	new_range->task = current;
1278	do {
1279		__set_current_state(TASK_UNINTERRUPTIBLE);
1280		spin_unlock_irq(&ic->endio_wait.lock);
1281		io_schedule();
1282		spin_lock_irq(&ic->endio_wait.lock);
1283	} while (unlikely(new_range->waiting));
1284}
1285
1286static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1287{
1288	if (unlikely(!add_new_range(ic, new_range, true)))
1289		wait_and_add_new_range(ic, new_range);
1290}
1291
1292static void init_journal_node(struct journal_node *node)
1293{
1294	RB_CLEAR_NODE(&node->node);
1295	node->sector = (sector_t)-1;
1296}
1297
1298static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1299{
1300	struct rb_node **link;
1301	struct rb_node *parent;
1302
1303	node->sector = sector;
1304	BUG_ON(!RB_EMPTY_NODE(&node->node));
1305
1306	link = &ic->journal_tree_root.rb_node;
1307	parent = NULL;
1308
1309	while (*link) {
1310		struct journal_node *j;
1311		parent = *link;
1312		j = container_of(parent, struct journal_node, node);
1313		if (sector < j->sector)
1314			link = &j->node.rb_left;
1315		else
1316			link = &j->node.rb_right;
1317	}
1318
1319	rb_link_node(&node->node, parent, link);
1320	rb_insert_color(&node->node, &ic->journal_tree_root);
1321}
1322
1323static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1324{
1325	BUG_ON(RB_EMPTY_NODE(&node->node));
1326	rb_erase(&node->node, &ic->journal_tree_root);
1327	init_journal_node(node);
1328}
1329
1330#define NOT_FOUND	(-1U)
1331
1332static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1333{
1334	struct rb_node *n = ic->journal_tree_root.rb_node;
1335	unsigned found = NOT_FOUND;
1336	*next_sector = (sector_t)-1;
1337	while (n) {
1338		struct journal_node *j = container_of(n, struct journal_node, node);
1339		if (sector == j->sector) {
1340			found = j - ic->journal_tree;
1341		}
1342		if (sector < j->sector) {
1343			*next_sector = j->sector;
1344			n = j->node.rb_left;
1345		} else {
1346			n = j->node.rb_right;
1347		}
1348	}
1349
1350	return found;
1351}
1352
1353static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1354{
1355	struct journal_node *node, *next_node;
1356	struct rb_node *next;
1357
1358	if (unlikely(pos >= ic->journal_entries))
1359		return false;
1360	node = &ic->journal_tree[pos];
1361	if (unlikely(RB_EMPTY_NODE(&node->node)))
1362		return false;
1363	if (unlikely(node->sector != sector))
1364		return false;
1365
1366	next = rb_next(&node->node);
1367	if (unlikely(!next))
1368		return true;
1369
1370	next_node = container_of(next, struct journal_node, node);
1371	return next_node->sector != sector;
1372}
1373
1374static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1375{
1376	struct rb_node *next;
1377	struct journal_node *next_node;
1378	unsigned next_section;
1379
1380	BUG_ON(RB_EMPTY_NODE(&node->node));
1381
1382	next = rb_next(&node->node);
1383	if (unlikely(!next))
1384		return false;
1385
1386	next_node = container_of(next, struct journal_node, node);
1387
1388	if (next_node->sector != node->sector)
1389		return false;
1390
1391	next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1392	if (next_section >= ic->committed_section &&
1393	    next_section < ic->committed_section + ic->n_committed_sections)
1394		return true;
1395	if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1396		return true;
1397
1398	return false;
1399}
1400
1401#define TAG_READ	0
1402#define TAG_WRITE	1
1403#define TAG_CMP		2
1404
1405static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1406			       unsigned *metadata_offset, unsigned total_size, int op)
1407{
1408#define MAY_BE_FILLER		1
1409#define MAY_BE_HASH		2
1410	unsigned hash_offset = 0;
1411	unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1412
1413	do {
1414		unsigned char *data, *dp;
1415		struct dm_buffer *b;
1416		unsigned to_copy;
1417		int r;
1418
1419		r = dm_integrity_failed(ic);
1420		if (unlikely(r))
1421			return r;
1422
1423		data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1424		if (IS_ERR(data))
1425			return PTR_ERR(data);
1426
1427		to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1428		dp = data + *metadata_offset;
1429		if (op == TAG_READ) {
1430			memcpy(tag, dp, to_copy);
1431		} else if (op == TAG_WRITE) {
1432			if (memcmp(dp, tag, to_copy)) {
1433				memcpy(dp, tag, to_copy);
1434				dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1435			}
1436		} else {
1437			/* e.g.: op == TAG_CMP */
1438
1439			if (likely(is_power_of_2(ic->tag_size))) {
1440				if (unlikely(memcmp(dp, tag, to_copy)))
1441					if (unlikely(!ic->discard) ||
1442					    unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1443						goto thorough_test;
1444				}
1445			} else {
1446				unsigned i, ts;
1447thorough_test:
1448				ts = total_size;
1449
1450				for (i = 0; i < to_copy; i++, ts--) {
1451					if (unlikely(dp[i] != tag[i]))
1452						may_be &= ~MAY_BE_HASH;
1453					if (likely(dp[i] != DISCARD_FILLER))
1454						may_be &= ~MAY_BE_FILLER;
1455					hash_offset++;
1456					if (unlikely(hash_offset == ic->tag_size)) {
1457						if (unlikely(!may_be)) {
1458							dm_bufio_release(b);
1459							return ts;
1460						}
1461						hash_offset = 0;
1462						may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1463					}
1464				}
1465			}
1466		}
1467		dm_bufio_release(b);
1468
1469		tag += to_copy;
1470		*metadata_offset += to_copy;
1471		if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1472			(*metadata_block)++;
1473			*metadata_offset = 0;
1474		}
1475
1476		if (unlikely(!is_power_of_2(ic->tag_size))) {
1477			hash_offset = (hash_offset + to_copy) % ic->tag_size;
1478		}
1479
1480		total_size -= to_copy;
1481	} while (unlikely(total_size));
1482
1483	return 0;
1484#undef MAY_BE_FILLER
1485#undef MAY_BE_HASH
1486}
1487
1488struct flush_request {
1489	struct dm_io_request io_req;
1490	struct dm_io_region io_reg;
1491	struct dm_integrity_c *ic;
1492	struct completion comp;
1493};
1494
1495static void flush_notify(unsigned long error, void *fr_)
1496{
1497	struct flush_request *fr = fr_;
1498	if (unlikely(error != 0))
1499		dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1500	complete(&fr->comp);
1501}
1502
1503static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1504{
1505	int r;
1506
1507	struct flush_request fr;
1508
1509	if (!ic->meta_dev)
1510		flush_data = false;
1511	if (flush_data) {
1512		fr.io_req.bi_op = REQ_OP_WRITE,
1513		fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1514		fr.io_req.mem.type = DM_IO_KMEM,
1515		fr.io_req.mem.ptr.addr = NULL,
1516		fr.io_req.notify.fn = flush_notify,
1517		fr.io_req.notify.context = &fr;
1518		fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1519		fr.io_reg.bdev = ic->dev->bdev,
1520		fr.io_reg.sector = 0,
1521		fr.io_reg.count = 0,
1522		fr.ic = ic;
1523		init_completion(&fr.comp);
1524		r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1525		BUG_ON(r);
1526	}
1527
1528	r = dm_bufio_write_dirty_buffers(ic->bufio);
1529	if (unlikely(r))
1530		dm_integrity_io_error(ic, "writing tags", r);
1531
1532	if (flush_data)
1533		wait_for_completion(&fr.comp);
1534}
1535
1536static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1537{
1538	DECLARE_WAITQUEUE(wait, current);
1539	__add_wait_queue(&ic->endio_wait, &wait);
1540	__set_current_state(TASK_UNINTERRUPTIBLE);
1541	spin_unlock_irq(&ic->endio_wait.lock);
1542	io_schedule();
1543	spin_lock_irq(&ic->endio_wait.lock);
1544	__remove_wait_queue(&ic->endio_wait, &wait);
1545}
1546
1547static void autocommit_fn(struct timer_list *t)
1548{
1549	struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1550
1551	if (likely(!dm_integrity_failed(ic)))
1552		queue_work(ic->commit_wq, &ic->commit_work);
1553}
1554
1555static void schedule_autocommit(struct dm_integrity_c *ic)
1556{
1557	if (!timer_pending(&ic->autocommit_timer))
1558		mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1559}
1560
1561static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1562{
1563	struct bio *bio;
1564	unsigned long flags;
1565
1566	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1567	bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1568	bio_list_add(&ic->flush_bio_list, bio);
1569	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1570
1571	queue_work(ic->commit_wq, &ic->commit_work);
1572}
1573
1574static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1575{
1576	int r = dm_integrity_failed(ic);
1577	if (unlikely(r) && !bio->bi_status)
1578		bio->bi_status = errno_to_blk_status(r);
1579	if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1580		unsigned long flags;
1581		spin_lock_irqsave(&ic->endio_wait.lock, flags);
1582		bio_list_add(&ic->synchronous_bios, bio);
1583		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1584		spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1585		return;
1586	}
1587	bio_endio(bio);
1588}
1589
1590static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1591{
1592	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1593
1594	if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1595		submit_flush_bio(ic, dio);
1596	else
1597		do_endio(ic, bio);
1598}
1599
1600static void dec_in_flight(struct dm_integrity_io *dio)
1601{
1602	if (atomic_dec_and_test(&dio->in_flight)) {
1603		struct dm_integrity_c *ic = dio->ic;
1604		struct bio *bio;
1605
1606		remove_range(ic, &dio->range);
1607
1608		if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1609			schedule_autocommit(ic);
1610
1611		bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1612
1613		if (unlikely(dio->bi_status) && !bio->bi_status)
1614			bio->bi_status = dio->bi_status;
1615		if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1616			dio->range.logical_sector += dio->range.n_sectors;
1617			bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1618			INIT_WORK(&dio->work, integrity_bio_wait);
1619			queue_work(ic->offload_wq, &dio->work);
1620			return;
1621		}
1622		do_endio_flush(ic, dio);
1623	}
1624}
1625
1626static void integrity_end_io(struct bio *bio)
1627{
1628	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1629
1630	dm_bio_restore(&dio->bio_details, bio);
1631	if (bio->bi_integrity)
1632		bio->bi_opf |= REQ_INTEGRITY;
1633
1634	if (dio->completion)
1635		complete(dio->completion);
1636
1637	dec_in_flight(dio);
1638}
1639
1640static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1641				      const char *data, char *result)
1642{
1643	__le64 sector_le = cpu_to_le64(sector);
1644	SHASH_DESC_ON_STACK(req, ic->internal_hash);
1645	int r;
1646	unsigned digest_size;
1647
1648	req->tfm = ic->internal_hash;
1649
1650	r = crypto_shash_init(req);
1651	if (unlikely(r < 0)) {
1652		dm_integrity_io_error(ic, "crypto_shash_init", r);
1653		goto failed;
1654	}
1655
1656	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1657		r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1658		if (unlikely(r < 0)) {
1659			dm_integrity_io_error(ic, "crypto_shash_update", r);
1660			goto failed;
1661		}
1662	}
1663
1664	r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1665	if (unlikely(r < 0)) {
1666		dm_integrity_io_error(ic, "crypto_shash_update", r);
1667		goto failed;
1668	}
1669
1670	r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1671	if (unlikely(r < 0)) {
1672		dm_integrity_io_error(ic, "crypto_shash_update", r);
1673		goto failed;
1674	}
1675
1676	r = crypto_shash_final(req, result);
1677	if (unlikely(r < 0)) {
1678		dm_integrity_io_error(ic, "crypto_shash_final", r);
1679		goto failed;
1680	}
1681
1682	digest_size = crypto_shash_digestsize(ic->internal_hash);
1683	if (unlikely(digest_size < ic->tag_size))
1684		memset(result + digest_size, 0, ic->tag_size - digest_size);
1685
1686	return;
1687
1688failed:
1689	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1690	get_random_bytes(result, ic->tag_size);
1691}
1692
1693static void integrity_metadata(struct work_struct *w)
1694{
1695	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1696	struct dm_integrity_c *ic = dio->ic;
1697
1698	int r;
1699
1700	if (ic->internal_hash) {
1701		struct bvec_iter iter;
1702		struct bio_vec bv;
1703		unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1704		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1705		char *checksums;
1706		unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1707		char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1708		sector_t sector;
1709		unsigned sectors_to_process;
1710
1711		if (unlikely(ic->mode == 'R'))
1712			goto skip_io;
1713
1714		if (likely(dio->op != REQ_OP_DISCARD))
1715			checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1716					    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1717		else
1718			checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1719		if (!checksums) {
1720			checksums = checksums_onstack;
1721			if (WARN_ON(extra_space &&
1722				    digest_size > sizeof(checksums_onstack))) {
1723				r = -EINVAL;
1724				goto error;
1725			}
1726		}
1727
1728		if (unlikely(dio->op == REQ_OP_DISCARD)) {
1729			sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1730			unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1731			unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1732			unsigned max_blocks = max_size / ic->tag_size;
1733			memset(checksums, DISCARD_FILLER, max_size);
1734
1735			while (bi_size) {
1736				unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1737				this_step_blocks = min(this_step_blocks, max_blocks);
1738				r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1739							this_step_blocks * ic->tag_size, TAG_WRITE);
1740				if (unlikely(r)) {
1741					if (likely(checksums != checksums_onstack))
1742						kfree(checksums);
1743					goto error;
1744				}
1745
1746				/*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1747					printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1748					printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1749					BUG();
1750				}*/
1751				bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1752				bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1753			}
1754
1755			if (likely(checksums != checksums_onstack))
1756				kfree(checksums);
1757			goto skip_io;
1758		}
1759
1760		sector = dio->range.logical_sector;
1761		sectors_to_process = dio->range.n_sectors;
1762
1763		__bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1764			unsigned pos;
1765			char *mem, *checksums_ptr;
1766
1767again:
1768			mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1769			pos = 0;
1770			checksums_ptr = checksums;
1771			do {
1772				integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1773				checksums_ptr += ic->tag_size;
1774				sectors_to_process -= ic->sectors_per_block;
1775				pos += ic->sectors_per_block << SECTOR_SHIFT;
1776				sector += ic->sectors_per_block;
1777			} while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1778			kunmap_atomic(mem);
1779
1780			r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1781						checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1782			if (unlikely(r)) {
1783				if (r > 0) {
1784					char b[BDEVNAME_SIZE];
1785					DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1786						    (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1787					r = -EILSEQ;
1788					atomic64_inc(&ic->number_of_mismatches);
1789				}
1790				if (likely(checksums != checksums_onstack))
1791					kfree(checksums);
1792				goto error;
1793			}
1794
1795			if (!sectors_to_process)
1796				break;
1797
1798			if (unlikely(pos < bv.bv_len)) {
1799				bv.bv_offset += pos;
1800				bv.bv_len -= pos;
1801				goto again;
1802			}
1803		}
1804
1805		if (likely(checksums != checksums_onstack))
1806			kfree(checksums);
1807	} else {
1808		struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1809
1810		if (bip) {
1811			struct bio_vec biv;
1812			struct bvec_iter iter;
1813			unsigned data_to_process = dio->range.n_sectors;
1814			sector_to_block(ic, data_to_process);
1815			data_to_process *= ic->tag_size;
1816
1817			bip_for_each_vec(biv, bip, iter) {
1818				unsigned char *tag;
1819				unsigned this_len;
1820
1821				BUG_ON(PageHighMem(biv.bv_page));
1822				tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1823				this_len = min(biv.bv_len, data_to_process);
1824				r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1825							this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1826				if (unlikely(r))
1827					goto error;
1828				data_to_process -= this_len;
1829				if (!data_to_process)
1830					break;
1831			}
1832		}
1833	}
1834skip_io:
1835	dec_in_flight(dio);
1836	return;
1837error:
1838	dio->bi_status = errno_to_blk_status(r);
1839	dec_in_flight(dio);
1840}
1841
1842static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1843{
1844	struct dm_integrity_c *ic = ti->private;
1845	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1846	struct bio_integrity_payload *bip;
1847
1848	sector_t area, offset;
1849
1850	dio->ic = ic;
1851	dio->bi_status = 0;
1852	dio->op = bio_op(bio);
1853
1854	if (unlikely(dio->op == REQ_OP_DISCARD)) {
1855		if (ti->max_io_len) {
1856			sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1857			unsigned log2_max_io_len = __fls(ti->max_io_len);
1858			sector_t start_boundary = sec >> log2_max_io_len;
1859			sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1860			if (start_boundary < end_boundary) {
1861				sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1862				dm_accept_partial_bio(bio, len);
1863			}
1864		}
1865	}
1866
1867	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1868		submit_flush_bio(ic, dio);
1869		return DM_MAPIO_SUBMITTED;
1870	}
1871
1872	dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1873	dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1874	if (unlikely(dio->fua)) {
1875		/*
1876		 * Don't pass down the FUA flag because we have to flush
1877		 * disk cache anyway.
1878		 */
1879		bio->bi_opf &= ~REQ_FUA;
1880	}
1881	if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1882		DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1883		      dio->range.logical_sector, bio_sectors(bio),
1884		      ic->provided_data_sectors);
1885		return DM_MAPIO_KILL;
1886	}
1887	if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1888		DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1889		      ic->sectors_per_block,
1890		      dio->range.logical_sector, bio_sectors(bio));
1891		return DM_MAPIO_KILL;
1892	}
1893
1894	if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1895		struct bvec_iter iter;
1896		struct bio_vec bv;
1897		bio_for_each_segment(bv, bio, iter) {
1898			if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1899				DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1900					bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1901				return DM_MAPIO_KILL;
1902			}
1903		}
1904	}
1905
1906	bip = bio_integrity(bio);
1907	if (!ic->internal_hash) {
1908		if (bip) {
1909			unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1910			if (ic->log2_tag_size >= 0)
1911				wanted_tag_size <<= ic->log2_tag_size;
1912			else
1913				wanted_tag_size *= ic->tag_size;
1914			if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1915				DMERR("Invalid integrity data size %u, expected %u",
1916				      bip->bip_iter.bi_size, wanted_tag_size);
1917				return DM_MAPIO_KILL;
1918			}
1919		}
1920	} else {
1921		if (unlikely(bip != NULL)) {
1922			DMERR("Unexpected integrity data when using internal hash");
1923			return DM_MAPIO_KILL;
1924		}
1925	}
1926
1927	if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1928		return DM_MAPIO_KILL;
1929
1930	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1931	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1932	bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1933
1934	dm_integrity_map_continue(dio, true);
1935	return DM_MAPIO_SUBMITTED;
1936}
1937
1938static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1939				 unsigned journal_section, unsigned journal_entry)
1940{
1941	struct dm_integrity_c *ic = dio->ic;
1942	sector_t logical_sector;
1943	unsigned n_sectors;
1944
1945	logical_sector = dio->range.logical_sector;
1946	n_sectors = dio->range.n_sectors;
1947	do {
1948		struct bio_vec bv = bio_iovec(bio);
1949		char *mem;
1950
1951		if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1952			bv.bv_len = n_sectors << SECTOR_SHIFT;
1953		n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1954		bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1955retry_kmap:
1956		mem = kmap_atomic(bv.bv_page);
1957		if (likely(dio->op == REQ_OP_WRITE))
1958			flush_dcache_page(bv.bv_page);
1959
1960		do {
1961			struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1962
1963			if (unlikely(dio->op == REQ_OP_READ)) {
1964				struct journal_sector *js;
1965				char *mem_ptr;
1966				unsigned s;
1967
1968				if (unlikely(journal_entry_is_inprogress(je))) {
1969					flush_dcache_page(bv.bv_page);
1970					kunmap_atomic(mem);
1971
1972					__io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1973					goto retry_kmap;
1974				}
1975				smp_rmb();
1976				BUG_ON(journal_entry_get_sector(je) != logical_sector);
1977				js = access_journal_data(ic, journal_section, journal_entry);
1978				mem_ptr = mem + bv.bv_offset;
1979				s = 0;
1980				do {
1981					memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1982					*(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1983					js++;
1984					mem_ptr += 1 << SECTOR_SHIFT;
1985				} while (++s < ic->sectors_per_block);
1986#ifdef INTERNAL_VERIFY
1987				if (ic->internal_hash) {
1988					char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1989
1990					integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1991					if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1992						DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1993							    logical_sector);
1994					}
1995				}
1996#endif
1997			}
1998
1999			if (!ic->internal_hash) {
2000				struct bio_integrity_payload *bip = bio_integrity(bio);
2001				unsigned tag_todo = ic->tag_size;
2002				char *tag_ptr = journal_entry_tag(ic, je);
2003
2004				if (bip) do {
2005					struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2006					unsigned tag_now = min(biv.bv_len, tag_todo);
2007					char *tag_addr;
2008					BUG_ON(PageHighMem(biv.bv_page));
2009					tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
2010					if (likely(dio->op == REQ_OP_WRITE))
2011						memcpy(tag_ptr, tag_addr, tag_now);
2012					else
2013						memcpy(tag_addr, tag_ptr, tag_now);
2014					bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2015					tag_ptr += tag_now;
2016					tag_todo -= tag_now;
2017				} while (unlikely(tag_todo)); else {
2018					if (likely(dio->op == REQ_OP_WRITE))
2019						memset(tag_ptr, 0, tag_todo);
2020				}
2021			}
2022
2023			if (likely(dio->op == REQ_OP_WRITE)) {
2024				struct journal_sector *js;
2025				unsigned s;
2026
2027				js = access_journal_data(ic, journal_section, journal_entry);
2028				memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2029
2030				s = 0;
2031				do {
2032					je->last_bytes[s] = js[s].commit_id;
2033				} while (++s < ic->sectors_per_block);
2034
2035				if (ic->internal_hash) {
2036					unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2037					if (unlikely(digest_size > ic->tag_size)) {
2038						char checksums_onstack[HASH_MAX_DIGESTSIZE];
2039						integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2040						memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2041					} else
2042						integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2043				}
2044
2045				journal_entry_set_sector(je, logical_sector);
2046			}
2047			logical_sector += ic->sectors_per_block;
2048
2049			journal_entry++;
2050			if (unlikely(journal_entry == ic->journal_section_entries)) {
2051				journal_entry = 0;
2052				journal_section++;
2053				wraparound_section(ic, &journal_section);
2054			}
2055
2056			bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2057		} while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2058
2059		if (unlikely(dio->op == REQ_OP_READ))
2060			flush_dcache_page(bv.bv_page);
2061		kunmap_atomic(mem);
2062	} while (n_sectors);
2063
2064	if (likely(dio->op == REQ_OP_WRITE)) {
2065		smp_mb();
2066		if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2067			wake_up(&ic->copy_to_journal_wait);
2068		if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2069			queue_work(ic->commit_wq, &ic->commit_work);
2070		} else {
2071			schedule_autocommit(ic);
2072		}
2073	} else {
2074		remove_range(ic, &dio->range);
2075	}
2076
2077	if (unlikely(bio->bi_iter.bi_size)) {
2078		sector_t area, offset;
2079
2080		dio->range.logical_sector = logical_sector;
2081		get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2082		dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2083		return true;
2084	}
2085
2086	return false;
2087}
2088
2089static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2090{
2091	struct dm_integrity_c *ic = dio->ic;
2092	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2093	unsigned journal_section, journal_entry;
2094	unsigned journal_read_pos;
2095	struct completion read_comp;
2096	bool discard_retried = false;
2097	bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2098	if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2099		need_sync_io = true;
2100
2101	if (need_sync_io && from_map) {
2102		INIT_WORK(&dio->work, integrity_bio_wait);
2103		queue_work(ic->offload_wq, &dio->work);
2104		return;
2105	}
2106
2107lock_retry:
2108	spin_lock_irq(&ic->endio_wait.lock);
2109retry:
2110	if (unlikely(dm_integrity_failed(ic))) {
2111		spin_unlock_irq(&ic->endio_wait.lock);
2112		do_endio(ic, bio);
2113		return;
2114	}
2115	dio->range.n_sectors = bio_sectors(bio);
2116	journal_read_pos = NOT_FOUND;
2117	if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2118		if (dio->op == REQ_OP_WRITE) {
2119			unsigned next_entry, i, pos;
2120			unsigned ws, we, range_sectors;
2121
2122			dio->range.n_sectors = min(dio->range.n_sectors,
2123						   (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2124			if (unlikely(!dio->range.n_sectors)) {
2125				if (from_map)
2126					goto offload_to_thread;
2127				sleep_on_endio_wait(ic);
2128				goto retry;
2129			}
2130			range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2131			ic->free_sectors -= range_sectors;
2132			journal_section = ic->free_section;
2133			journal_entry = ic->free_section_entry;
2134
2135			next_entry = ic->free_section_entry + range_sectors;
2136			ic->free_section_entry = next_entry % ic->journal_section_entries;
2137			ic->free_section += next_entry / ic->journal_section_entries;
2138			ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2139			wraparound_section(ic, &ic->free_section);
2140
2141			pos = journal_section * ic->journal_section_entries + journal_entry;
2142			ws = journal_section;
2143			we = journal_entry;
2144			i = 0;
2145			do {
2146				struct journal_entry *je;
2147
2148				add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2149				pos++;
2150				if (unlikely(pos >= ic->journal_entries))
2151					pos = 0;
2152
2153				je = access_journal_entry(ic, ws, we);
2154				BUG_ON(!journal_entry_is_unused(je));
2155				journal_entry_set_inprogress(je);
2156				we++;
2157				if (unlikely(we == ic->journal_section_entries)) {
2158					we = 0;
2159					ws++;
2160					wraparound_section(ic, &ws);
2161				}
2162			} while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2163
2164			spin_unlock_irq(&ic->endio_wait.lock);
2165			goto journal_read_write;
2166		} else {
2167			sector_t next_sector;
2168			journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2169			if (likely(journal_read_pos == NOT_FOUND)) {
2170				if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2171					dio->range.n_sectors = next_sector - dio->range.logical_sector;
2172			} else {
2173				unsigned i;
2174				unsigned jp = journal_read_pos + 1;
2175				for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2176					if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2177						break;
2178				}
2179				dio->range.n_sectors = i;
2180			}
2181		}
2182	}
2183	if (unlikely(!add_new_range(ic, &dio->range, true))) {
2184		/*
2185		 * We must not sleep in the request routine because it could
2186		 * stall bios on current->bio_list.
2187		 * So, we offload the bio to a workqueue if we have to sleep.
2188		 */
2189		if (from_map) {
2190offload_to_thread:
2191			spin_unlock_irq(&ic->endio_wait.lock);
2192			INIT_WORK(&dio->work, integrity_bio_wait);
2193			queue_work(ic->wait_wq, &dio->work);
2194			return;
2195		}
2196		if (journal_read_pos != NOT_FOUND)
2197			dio->range.n_sectors = ic->sectors_per_block;
2198		wait_and_add_new_range(ic, &dio->range);
2199		/*
2200		 * wait_and_add_new_range drops the spinlock, so the journal
2201		 * may have been changed arbitrarily. We need to recheck.
2202		 * To simplify the code, we restrict I/O size to just one block.
2203		 */
2204		if (journal_read_pos != NOT_FOUND) {
2205			sector_t next_sector;
2206			unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2207			if (unlikely(new_pos != journal_read_pos)) {
2208				remove_range_unlocked(ic, &dio->range);
2209				goto retry;
2210			}
2211		}
2212	}
2213	if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2214		sector_t next_sector;
2215		unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2216		if (unlikely(new_pos != NOT_FOUND) ||
2217		    unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2218			remove_range_unlocked(ic, &dio->range);
2219			spin_unlock_irq(&ic->endio_wait.lock);
2220			queue_work(ic->commit_wq, &ic->commit_work);
2221			flush_workqueue(ic->commit_wq);
2222			queue_work(ic->writer_wq, &ic->writer_work);
2223			flush_workqueue(ic->writer_wq);
2224			discard_retried = true;
2225			goto lock_retry;
2226		}
2227	}
2228	spin_unlock_irq(&ic->endio_wait.lock);
2229
2230	if (unlikely(journal_read_pos != NOT_FOUND)) {
2231		journal_section = journal_read_pos / ic->journal_section_entries;
2232		journal_entry = journal_read_pos % ic->journal_section_entries;
2233		goto journal_read_write;
2234	}
2235
2236	if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2237		if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2238				     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2239			struct bitmap_block_status *bbs;
2240
2241			bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2242			spin_lock(&bbs->bio_queue_lock);
2243			bio_list_add(&bbs->bio_queue, bio);
2244			spin_unlock(&bbs->bio_queue_lock);
2245			queue_work(ic->writer_wq, &bbs->work);
2246			return;
2247		}
2248	}
2249
2250	dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2251
2252	if (need_sync_io) {
2253		init_completion(&read_comp);
2254		dio->completion = &read_comp;
2255	} else
2256		dio->completion = NULL;
2257
2258	dm_bio_record(&dio->bio_details, bio);
2259	bio_set_dev(bio, ic->dev->bdev);
2260	bio->bi_integrity = NULL;
2261	bio->bi_opf &= ~REQ_INTEGRITY;
2262	bio->bi_end_io = integrity_end_io;
2263	bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2264
2265	if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2266		integrity_metadata(&dio->work);
2267		dm_integrity_flush_buffers(ic, false);
2268
2269		dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2270		dio->completion = NULL;
2271
2272		submit_bio_noacct(bio);
2273
2274		return;
2275	}
2276
2277	submit_bio_noacct(bio);
2278
2279	if (need_sync_io) {
2280		wait_for_completion_io(&read_comp);
2281		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2282		    dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2283			goto skip_check;
2284		if (ic->mode == 'B') {
2285			if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2286					     dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2287				goto skip_check;
2288		}
2289
2290		if (likely(!bio->bi_status))
2291			integrity_metadata(&dio->work);
2292		else
2293skip_check:
2294			dec_in_flight(dio);
2295
2296	} else {
2297		INIT_WORK(&dio->work, integrity_metadata);
2298		queue_work(ic->metadata_wq, &dio->work);
2299	}
2300
2301	return;
2302
2303journal_read_write:
2304	if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2305		goto lock_retry;
2306
2307	do_endio_flush(ic, dio);
2308}
2309
2310
2311static void integrity_bio_wait(struct work_struct *w)
2312{
2313	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2314
2315	dm_integrity_map_continue(dio, false);
2316}
2317
2318static void pad_uncommitted(struct dm_integrity_c *ic)
2319{
2320	if (ic->free_section_entry) {
2321		ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2322		ic->free_section_entry = 0;
2323		ic->free_section++;
2324		wraparound_section(ic, &ic->free_section);
2325		ic->n_uncommitted_sections++;
2326	}
2327	if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2328		    (ic->n_uncommitted_sections + ic->n_committed_sections) *
2329		    ic->journal_section_entries + ic->free_sectors)) {
2330		DMCRIT("journal_sections %u, journal_section_entries %u, "
2331		       "n_uncommitted_sections %u, n_committed_sections %u, "
2332		       "journal_section_entries %u, free_sectors %u",
2333		       ic->journal_sections, ic->journal_section_entries,
2334		       ic->n_uncommitted_sections, ic->n_committed_sections,
2335		       ic->journal_section_entries, ic->free_sectors);
2336	}
2337}
2338
2339static void integrity_commit(struct work_struct *w)
2340{
2341	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2342	unsigned commit_start, commit_sections;
2343	unsigned i, j, n;
2344	struct bio *flushes;
2345
2346	del_timer(&ic->autocommit_timer);
2347
2348	spin_lock_irq(&ic->endio_wait.lock);
2349	flushes = bio_list_get(&ic->flush_bio_list);
2350	if (unlikely(ic->mode != 'J')) {
2351		spin_unlock_irq(&ic->endio_wait.lock);
2352		dm_integrity_flush_buffers(ic, true);
2353		goto release_flush_bios;
2354	}
2355
2356	pad_uncommitted(ic);
2357	commit_start = ic->uncommitted_section;
2358	commit_sections = ic->n_uncommitted_sections;
2359	spin_unlock_irq(&ic->endio_wait.lock);
2360
2361	if (!commit_sections)
2362		goto release_flush_bios;
2363
2364	i = commit_start;
2365	for (n = 0; n < commit_sections; n++) {
2366		for (j = 0; j < ic->journal_section_entries; j++) {
2367			struct journal_entry *je;
2368			je = access_journal_entry(ic, i, j);
2369			io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2370		}
2371		for (j = 0; j < ic->journal_section_sectors; j++) {
2372			struct journal_sector *js;
2373			js = access_journal(ic, i, j);
2374			js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2375		}
2376		i++;
2377		if (unlikely(i >= ic->journal_sections))
2378			ic->commit_seq = next_commit_seq(ic->commit_seq);
2379		wraparound_section(ic, &i);
2380	}
2381	smp_rmb();
2382
2383	write_journal(ic, commit_start, commit_sections);
2384
2385	spin_lock_irq(&ic->endio_wait.lock);
2386	ic->uncommitted_section += commit_sections;
2387	wraparound_section(ic, &ic->uncommitted_section);
2388	ic->n_uncommitted_sections -= commit_sections;
2389	ic->n_committed_sections += commit_sections;
2390	spin_unlock_irq(&ic->endio_wait.lock);
2391
2392	if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2393		queue_work(ic->writer_wq, &ic->writer_work);
2394
2395release_flush_bios:
2396	while (flushes) {
2397		struct bio *next = flushes->bi_next;
2398		flushes->bi_next = NULL;
2399		do_endio(ic, flushes);
2400		flushes = next;
2401	}
2402}
2403
2404static void complete_copy_from_journal(unsigned long error, void *context)
2405{
2406	struct journal_io *io = context;
2407	struct journal_completion *comp = io->comp;
2408	struct dm_integrity_c *ic = comp->ic;
2409	remove_range(ic, &io->range);
2410	mempool_free(io, &ic->journal_io_mempool);
2411	if (unlikely(error != 0))
2412		dm_integrity_io_error(ic, "copying from journal", -EIO);
2413	complete_journal_op(comp);
2414}
2415
2416static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2417			       struct journal_entry *je)
2418{
2419	unsigned s = 0;
2420	do {
2421		js->commit_id = je->last_bytes[s];
2422		js++;
2423	} while (++s < ic->sectors_per_block);
2424}
2425
2426static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2427			     unsigned write_sections, bool from_replay)
2428{
2429	unsigned i, j, n;
2430	struct journal_completion comp;
2431	struct blk_plug plug;
2432
2433	blk_start_plug(&plug);
2434
2435	comp.ic = ic;
2436	comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2437	init_completion(&comp.comp);
2438
2439	i = write_start;
2440	for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2441#ifndef INTERNAL_VERIFY
2442		if (unlikely(from_replay))
2443#endif
2444			rw_section_mac(ic, i, false);
2445		for (j = 0; j < ic->journal_section_entries; j++) {
2446			struct journal_entry *je = access_journal_entry(ic, i, j);
2447			sector_t sec, area, offset;
2448			unsigned k, l, next_loop;
2449			sector_t metadata_block;
2450			unsigned metadata_offset;
2451			struct journal_io *io;
2452
2453			if (journal_entry_is_unused(je))
2454				continue;
2455			BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2456			sec = journal_entry_get_sector(je);
2457			if (unlikely(from_replay)) {
2458				if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2459					dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2460					sec &= ~(sector_t)(ic->sectors_per_block - 1);
2461				}
2462			}
2463			if (unlikely(sec >= ic->provided_data_sectors))
2464				continue;
2465			get_area_and_offset(ic, sec, &area, &offset);
2466			restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2467			for (k = j + 1; k < ic->journal_section_entries; k++) {
2468				struct journal_entry *je2 = access_journal_entry(ic, i, k);
2469				sector_t sec2, area2, offset2;
2470				if (journal_entry_is_unused(je2))
2471					break;
2472				BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2473				sec2 = journal_entry_get_sector(je2);
2474				if (unlikely(sec2 >= ic->provided_data_sectors))
2475					break;
2476				get_area_and_offset(ic, sec2, &area2, &offset2);
2477				if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2478					break;
2479				restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2480			}
2481			next_loop = k - 1;
2482
2483			io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2484			io->comp = &comp;
2485			io->range.logical_sector = sec;
2486			io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2487
2488			spin_lock_irq(&ic->endio_wait.lock);
2489			add_new_range_and_wait(ic, &io->range);
2490
2491			if (likely(!from_replay)) {
2492				struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2493
2494				/* don't write if there is newer committed sector */
2495				while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2496					struct journal_entry *je2 = access_journal_entry(ic, i, j);
2497
2498					journal_entry_set_unused(je2);
2499					remove_journal_node(ic, &section_node[j]);
2500					j++;
2501					sec += ic->sectors_per_block;
2502					offset += ic->sectors_per_block;
2503				}
2504				while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2505					struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2506
2507					journal_entry_set_unused(je2);
2508					remove_journal_node(ic, &section_node[k - 1]);
2509					k--;
2510				}
2511				if (j == k) {
2512					remove_range_unlocked(ic, &io->range);
2513					spin_unlock_irq(&ic->endio_wait.lock);
2514					mempool_free(io, &ic->journal_io_mempool);
2515					goto skip_io;
2516				}
2517				for (l = j; l < k; l++) {
2518					remove_journal_node(ic, &section_node[l]);
2519				}
2520			}
2521			spin_unlock_irq(&ic->endio_wait.lock);
2522
2523			metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2524			for (l = j; l < k; l++) {
2525				int r;
2526				struct journal_entry *je2 = access_journal_entry(ic, i, l);
2527
2528				if (
2529#ifndef INTERNAL_VERIFY
2530				    unlikely(from_replay) &&
2531#endif
2532				    ic->internal_hash) {
2533					char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2534
2535					integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2536								  (char *)access_journal_data(ic, i, l), test_tag);
2537					if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2538						dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2539				}
2540
2541				journal_entry_set_unused(je2);
2542				r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2543							ic->tag_size, TAG_WRITE);
2544				if (unlikely(r)) {
2545					dm_integrity_io_error(ic, "reading tags", r);
2546				}
2547			}
2548
2549			atomic_inc(&comp.in_flight);
2550			copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2551					  (k - j) << ic->sb->log2_sectors_per_block,
2552					  get_data_sector(ic, area, offset),
2553					  complete_copy_from_journal, io);
2554skip_io:
2555			j = next_loop;
2556		}
2557	}
2558
2559	dm_bufio_write_dirty_buffers_async(ic->bufio);
2560
2561	blk_finish_plug(&plug);
2562
2563	complete_journal_op(&comp);
2564	wait_for_completion_io(&comp.comp);
2565
2566	dm_integrity_flush_buffers(ic, true);
2567}
2568
2569static void integrity_writer(struct work_struct *w)
2570{
2571	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2572	unsigned write_start, write_sections;
2573
2574	unsigned prev_free_sectors;
2575
2576	/* the following test is not needed, but it tests the replay code */
2577	if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2578		return;
2579
2580	spin_lock_irq(&ic->endio_wait.lock);
2581	write_start = ic->committed_section;
2582	write_sections = ic->n_committed_sections;
2583	spin_unlock_irq(&ic->endio_wait.lock);
2584
2585	if (!write_sections)
2586		return;
2587
2588	do_journal_write(ic, write_start, write_sections, false);
2589
2590	spin_lock_irq(&ic->endio_wait.lock);
2591
2592	ic->committed_section += write_sections;
2593	wraparound_section(ic, &ic->committed_section);
2594	ic->n_committed_sections -= write_sections;
2595
2596	prev_free_sectors = ic->free_sectors;
2597	ic->free_sectors += write_sections * ic->journal_section_entries;
2598	if (unlikely(!prev_free_sectors))
2599		wake_up_locked(&ic->endio_wait);
2600
2601	spin_unlock_irq(&ic->endio_wait.lock);
2602}
2603
2604static void recalc_write_super(struct dm_integrity_c *ic)
2605{
2606	int r;
2607
2608	dm_integrity_flush_buffers(ic, false);
2609	if (dm_integrity_failed(ic))
2610		return;
2611
2612	r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2613	if (unlikely(r))
2614		dm_integrity_io_error(ic, "writing superblock", r);
2615}
2616
2617static void integrity_recalc(struct work_struct *w)
2618{
2619	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2620	struct dm_integrity_range range;
2621	struct dm_io_request io_req;
2622	struct dm_io_region io_loc;
2623	sector_t area, offset;
2624	sector_t metadata_block;
2625	unsigned metadata_offset;
2626	sector_t logical_sector, n_sectors;
2627	__u8 *t;
2628	unsigned i;
2629	int r;
2630	unsigned super_counter = 0;
2631
2632	DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2633
2634	spin_lock_irq(&ic->endio_wait.lock);
2635
2636next_chunk:
2637
2638	if (unlikely(dm_post_suspending(ic->ti)))
2639		goto unlock_ret;
2640
2641	range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2642	if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2643		if (ic->mode == 'B') {
2644			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2645			DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2646			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2647		}
2648		goto unlock_ret;
2649	}
2650
2651	get_area_and_offset(ic, range.logical_sector, &area, &offset);
2652	range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2653	if (!ic->meta_dev)
2654		range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2655
2656	add_new_range_and_wait(ic, &range);
2657	spin_unlock_irq(&ic->endio_wait.lock);
2658	logical_sector = range.logical_sector;
2659	n_sectors = range.n_sectors;
2660
2661	if (ic->mode == 'B') {
2662		if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2663			goto advance_and_next;
2664		}
2665		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2666				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2667			logical_sector += ic->sectors_per_block;
2668			n_sectors -= ic->sectors_per_block;
2669			cond_resched();
2670		}
2671		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2672				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2673			n_sectors -= ic->sectors_per_block;
2674			cond_resched();
2675		}
2676		get_area_and_offset(ic, logical_sector, &area, &offset);
2677	}
2678
2679	DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2680
2681	if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2682		recalc_write_super(ic);
2683		if (ic->mode == 'B') {
2684			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2685		}
2686		super_counter = 0;
2687	}
2688
2689	if (unlikely(dm_integrity_failed(ic)))
2690		goto err;
2691
2692	io_req.bi_op = REQ_OP_READ;
2693	io_req.bi_op_flags = 0;
2694	io_req.mem.type = DM_IO_VMA;
2695	io_req.mem.ptr.addr = ic->recalc_buffer;
2696	io_req.notify.fn = NULL;
2697	io_req.client = ic->io;
2698	io_loc.bdev = ic->dev->bdev;
2699	io_loc.sector = get_data_sector(ic, area, offset);
2700	io_loc.count = n_sectors;
2701
2702	r = dm_io(&io_req, 1, &io_loc, NULL);
2703	if (unlikely(r)) {
2704		dm_integrity_io_error(ic, "reading data", r);
2705		goto err;
2706	}
2707
2708	t = ic->recalc_tags;
2709	for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2710		integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2711		t += ic->tag_size;
2712	}
2713
2714	metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2715
2716	r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2717	if (unlikely(r)) {
2718		dm_integrity_io_error(ic, "writing tags", r);
2719		goto err;
2720	}
2721
2722	if (ic->mode == 'B') {
2723		sector_t start, end;
2724		start = (range.logical_sector >>
2725			 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2726			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2727		end = ((range.logical_sector + range.n_sectors) >>
2728		       (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2729			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2730		block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2731	}
2732
2733advance_and_next:
2734	cond_resched();
2735
2736	spin_lock_irq(&ic->endio_wait.lock);
2737	remove_range_unlocked(ic, &range);
2738	ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2739	goto next_chunk;
2740
2741err:
2742	remove_range(ic, &range);
2743	return;
2744
2745unlock_ret:
2746	spin_unlock_irq(&ic->endio_wait.lock);
2747
2748	recalc_write_super(ic);
2749}
2750
2751static void bitmap_block_work(struct work_struct *w)
2752{
2753	struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2754	struct dm_integrity_c *ic = bbs->ic;
2755	struct bio *bio;
2756	struct bio_list bio_queue;
2757	struct bio_list waiting;
2758
2759	bio_list_init(&waiting);
2760
2761	spin_lock(&bbs->bio_queue_lock);
2762	bio_queue = bbs->bio_queue;
2763	bio_list_init(&bbs->bio_queue);
2764	spin_unlock(&bbs->bio_queue_lock);
2765
2766	while ((bio = bio_list_pop(&bio_queue))) {
2767		struct dm_integrity_io *dio;
2768
2769		dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2770
2771		if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2772				    dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2773			remove_range(ic, &dio->range);
2774			INIT_WORK(&dio->work, integrity_bio_wait);
2775			queue_work(ic->offload_wq, &dio->work);
2776		} else {
2777			block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2778					dio->range.n_sectors, BITMAP_OP_SET);
2779			bio_list_add(&waiting, bio);
2780		}
2781	}
2782
2783	if (bio_list_empty(&waiting))
2784		return;
2785
2786	rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2787			   bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2788			   BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2789
2790	while ((bio = bio_list_pop(&waiting))) {
2791		struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2792
2793		block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2794				dio->range.n_sectors, BITMAP_OP_SET);
2795
2796		remove_range(ic, &dio->range);
2797		INIT_WORK(&dio->work, integrity_bio_wait);
2798		queue_work(ic->offload_wq, &dio->work);
2799	}
2800
2801	queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2802}
2803
2804static void bitmap_flush_work(struct work_struct *work)
2805{
2806	struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2807	struct dm_integrity_range range;
2808	unsigned long limit;
2809	struct bio *bio;
2810
2811	dm_integrity_flush_buffers(ic, false);
2812
2813	range.logical_sector = 0;
2814	range.n_sectors = ic->provided_data_sectors;
2815
2816	spin_lock_irq(&ic->endio_wait.lock);
2817	add_new_range_and_wait(ic, &range);
2818	spin_unlock_irq(&ic->endio_wait.lock);
2819
2820	dm_integrity_flush_buffers(ic, true);
2821
2822	limit = ic->provided_data_sectors;
2823	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2824		limit = le64_to_cpu(ic->sb->recalc_sector)
2825			>> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2826			<< (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2827	}
2828	/*DEBUG_print("zeroing journal\n");*/
2829	block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2830	block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2831
2832	rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2833			   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2834
2835	spin_lock_irq(&ic->endio_wait.lock);
2836	remove_range_unlocked(ic, &range);
2837	while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2838		bio_endio(bio);
2839		spin_unlock_irq(&ic->endio_wait.lock);
2840		spin_lock_irq(&ic->endio_wait.lock);
2841	}
2842	spin_unlock_irq(&ic->endio_wait.lock);
2843}
2844
2845
2846static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2847			 unsigned n_sections, unsigned char commit_seq)
2848{
2849	unsigned i, j, n;
2850
2851	if (!n_sections)
2852		return;
2853
2854	for (n = 0; n < n_sections; n++) {
2855		i = start_section + n;
2856		wraparound_section(ic, &i);
2857		for (j = 0; j < ic->journal_section_sectors; j++) {
2858			struct journal_sector *js = access_journal(ic, i, j);
2859			memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2860			js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2861		}
2862		for (j = 0; j < ic->journal_section_entries; j++) {
2863			struct journal_entry *je = access_journal_entry(ic, i, j);
2864			journal_entry_set_unused(je);
2865		}
2866	}
2867
2868	write_journal(ic, start_section, n_sections);
2869}
2870
2871static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2872{
2873	unsigned char k;
2874	for (k = 0; k < N_COMMIT_IDS; k++) {
2875		if (dm_integrity_commit_id(ic, i, j, k) == id)
2876			return k;
2877	}
2878	dm_integrity_io_error(ic, "journal commit id", -EIO);
2879	return -EIO;
2880}
2881
2882static void replay_journal(struct dm_integrity_c *ic)
2883{
2884	unsigned i, j;
2885	bool used_commit_ids[N_COMMIT_IDS];
2886	unsigned max_commit_id_sections[N_COMMIT_IDS];
2887	unsigned write_start, write_sections;
2888	unsigned continue_section;
2889	bool journal_empty;
2890	unsigned char unused, last_used, want_commit_seq;
2891
2892	if (ic->mode == 'R')
2893		return;
2894
2895	if (ic->journal_uptodate)
2896		return;
2897
2898	last_used = 0;
2899	write_start = 0;
2900
2901	if (!ic->just_formatted) {
2902		DEBUG_print("reading journal\n");
2903		rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2904		if (ic->journal_io)
2905			DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2906		if (ic->journal_io) {
2907			struct journal_completion crypt_comp;
2908			crypt_comp.ic = ic;
2909			init_completion(&crypt_comp.comp);
2910			crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2911			encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2912			wait_for_completion(&crypt_comp.comp);
2913		}
2914		DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2915	}
2916
2917	if (dm_integrity_failed(ic))
2918		goto clear_journal;
2919
2920	journal_empty = true;
2921	memset(used_commit_ids, 0, sizeof used_commit_ids);
2922	memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2923	for (i = 0; i < ic->journal_sections; i++) {
2924		for (j = 0; j < ic->journal_section_sectors; j++) {
2925			int k;
2926			struct journal_sector *js = access_journal(ic, i, j);
2927			k = find_commit_seq(ic, i, j, js->commit_id);
2928			if (k < 0)
2929				goto clear_journal;
2930			used_commit_ids[k] = true;
2931			max_commit_id_sections[k] = i;
2932		}
2933		if (journal_empty) {
2934			for (j = 0; j < ic->journal_section_entries; j++) {
2935				struct journal_entry *je = access_journal_entry(ic, i, j);
2936				if (!journal_entry_is_unused(je)) {
2937					journal_empty = false;
2938					break;
2939				}
2940			}
2941		}
2942	}
2943
2944	if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2945		unused = N_COMMIT_IDS - 1;
2946		while (unused && !used_commit_ids[unused - 1])
2947			unused--;
2948	} else {
2949		for (unused = 0; unused < N_COMMIT_IDS; unused++)
2950			if (!used_commit_ids[unused])
2951				break;
2952		if (unused == N_COMMIT_IDS) {
2953			dm_integrity_io_error(ic, "journal commit ids", -EIO);
2954			goto clear_journal;
2955		}
2956	}
2957	DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2958		    unused, used_commit_ids[0], used_commit_ids[1],
2959		    used_commit_ids[2], used_commit_ids[3]);
2960
2961	last_used = prev_commit_seq(unused);
2962	want_commit_seq = prev_commit_seq(last_used);
2963
2964	if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2965		journal_empty = true;
2966
2967	write_start = max_commit_id_sections[last_used] + 1;
2968	if (unlikely(write_start >= ic->journal_sections))
2969		want_commit_seq = next_commit_seq(want_commit_seq);
2970	wraparound_section(ic, &write_start);
2971
2972	i = write_start;
2973	for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2974		for (j = 0; j < ic->journal_section_sectors; j++) {
2975			struct journal_sector *js = access_journal(ic, i, j);
2976
2977			if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2978				/*
2979				 * This could be caused by crash during writing.
2980				 * We won't replay the inconsistent part of the
2981				 * journal.
2982				 */
2983				DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2984					    i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2985				goto brk;
2986			}
2987		}
2988		i++;
2989		if (unlikely(i >= ic->journal_sections))
2990			want_commit_seq = next_commit_seq(want_commit_seq);
2991		wraparound_section(ic, &i);
2992	}
2993brk:
2994
2995	if (!journal_empty) {
2996		DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2997			    write_sections, write_start, want_commit_seq);
2998		do_journal_write(ic, write_start, write_sections, true);
2999	}
3000
3001	if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3002		continue_section = write_start;
3003		ic->commit_seq = want_commit_seq;
3004		DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3005	} else {
3006		unsigned s;
3007		unsigned char erase_seq;
3008clear_journal:
3009		DEBUG_print("clearing journal\n");
3010
3011		erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3012		s = write_start;
3013		init_journal(ic, s, 1, erase_seq);
3014		s++;
3015		wraparound_section(ic, &s);
3016		if (ic->journal_sections >= 2) {
3017			init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3018			s += ic->journal_sections - 2;
3019			wraparound_section(ic, &s);
3020			init_journal(ic, s, 1, erase_seq);
3021		}
3022
3023		continue_section = 0;
3024		ic->commit_seq = next_commit_seq(erase_seq);
3025	}
3026
3027	ic->committed_section = continue_section;
3028	ic->n_committed_sections = 0;
3029
3030	ic->uncommitted_section = continue_section;
3031	ic->n_uncommitted_sections = 0;
3032
3033	ic->free_section = continue_section;
3034	ic->free_section_entry = 0;
3035	ic->free_sectors = ic->journal_entries;
3036
3037	ic->journal_tree_root = RB_ROOT;
3038	for (i = 0; i < ic->journal_entries; i++)
3039		init_journal_node(&ic->journal_tree[i]);
3040}
3041
3042static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3043{
3044	DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3045
3046	if (ic->mode == 'B') {
3047		ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3048		ic->synchronous_mode = 1;
3049
3050		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3051		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3052		flush_workqueue(ic->commit_wq);
3053	}
3054}
3055
3056static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3057{
3058	struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3059
3060	DEBUG_print("dm_integrity_reboot\n");
3061
3062	dm_integrity_enter_synchronous_mode(ic);
3063
3064	return NOTIFY_DONE;
3065}
3066
3067static void dm_integrity_postsuspend(struct dm_target *ti)
3068{
3069	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3070	int r;
3071
3072	WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3073
3074	del_timer_sync(&ic->autocommit_timer);
3075
3076	if (ic->recalc_wq)
3077		drain_workqueue(ic->recalc_wq);
3078
3079	if (ic->mode == 'B')
3080		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3081
3082	queue_work(ic->commit_wq, &ic->commit_work);
3083	drain_workqueue(ic->commit_wq);
3084
3085	if (ic->mode == 'J') {
3086		if (ic->meta_dev)
3087			queue_work(ic->writer_wq, &ic->writer_work);
3088		drain_workqueue(ic->writer_wq);
3089		dm_integrity_flush_buffers(ic, true);
3090	}
3091
3092	if (ic->mode == 'B') {
3093		dm_integrity_flush_buffers(ic, true);
3094#if 1
3095		/* set to 0 to test bitmap replay code */
3096		init_journal(ic, 0, ic->journal_sections, 0);
3097		ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3098		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3099		if (unlikely(r))
3100			dm_integrity_io_error(ic, "writing superblock", r);
3101#endif
3102	}
3103
3104	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3105
3106	ic->journal_uptodate = true;
3107}
3108
3109static void dm_integrity_resume(struct dm_target *ti)
3110{
3111	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3112	__u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3113	int r;
3114
3115	DEBUG_print("resume\n");
3116
3117	if (ic->provided_data_sectors != old_provided_data_sectors) {
3118		if (ic->provided_data_sectors > old_provided_data_sectors &&
3119		    ic->mode == 'B' &&
3120		    ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3121			rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3122					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3123			block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3124					ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3125			rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3126					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3127		}
3128
3129		ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3130		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3131		if (unlikely(r))
3132			dm_integrity_io_error(ic, "writing superblock", r);
3133	}
3134
3135	if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3136		DEBUG_print("resume dirty_bitmap\n");
3137		rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3138				   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3139		if (ic->mode == 'B') {
3140			if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3141			    !ic->reset_recalculate_flag) {
3142				block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3143				block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3144				if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3145						     BITMAP_OP_TEST_ALL_CLEAR)) {
3146					ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3147					ic->sb->recalc_sector = cpu_to_le64(0);
3148				}
3149			} else {
3150				DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3151					    ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3152				ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3153				block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3154				block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3155				block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3156				rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3157						   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3158				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3159				ic->sb->recalc_sector = cpu_to_le64(0);
3160			}
3161		} else {
3162			if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3163			      block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3164			    ic->reset_recalculate_flag) {
3165				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3166				ic->sb->recalc_sector = cpu_to_le64(0);
3167			}
3168			init_journal(ic, 0, ic->journal_sections, 0);
3169			replay_journal(ic);
3170			ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3171		}
3172		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3173		if (unlikely(r))
3174			dm_integrity_io_error(ic, "writing superblock", r);
3175	} else {
3176		replay_journal(ic);
3177		if (ic->reset_recalculate_flag) {
3178			ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3179			ic->sb->recalc_sector = cpu_to_le64(0);
3180		}
3181		if (ic->mode == 'B') {
3182			ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3183			ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3184			r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3185			if (unlikely(r))
3186				dm_integrity_io_error(ic, "writing superblock", r);
3187
3188			block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3189			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3190			block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3191			if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3192			    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3193				block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3194						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3195				block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3196						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3197				block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3198						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3199			}
3200			rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3201					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3202		}
3203	}
3204
3205	DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3206	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3207		__u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3208		DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3209		if (recalc_pos < ic->provided_data_sectors) {
3210			queue_work(ic->recalc_wq, &ic->recalc_work);
3211		} else if (recalc_pos > ic->provided_data_sectors) {
3212			ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3213			recalc_write_super(ic);
3214		}
3215	}
3216
3217	ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3218	ic->reboot_notifier.next = NULL;
3219	ic->reboot_notifier.priority = INT_MAX - 1;	/* be notified after md and before hardware drivers */
3220	WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3221
3222#if 0
3223	/* set to 1 to stress test synchronous mode */
3224	dm_integrity_enter_synchronous_mode(ic);
3225#endif
3226}
3227
3228static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3229				unsigned status_flags, char *result, unsigned maxlen)
3230{
3231	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3232	unsigned arg_count;
3233	size_t sz = 0;
3234
3235	switch (type) {
3236	case STATUSTYPE_INFO:
3237		DMEMIT("%llu %llu",
3238			(unsigned long long)atomic64_read(&ic->number_of_mismatches),
3239			ic->provided_data_sectors);
3240		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3241			DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3242		else
3243			DMEMIT(" -");
3244		break;
3245
3246	case STATUSTYPE_TABLE: {
3247		__u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3248		watermark_percentage += ic->journal_entries / 2;
3249		do_div(watermark_percentage, ic->journal_entries);
3250		arg_count = 3;
3251		arg_count += !!ic->meta_dev;
3252		arg_count += ic->sectors_per_block != 1;
3253		arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3254		arg_count += ic->reset_recalculate_flag;
3255		arg_count += ic->discard;
3256		arg_count += ic->mode == 'J';
3257		arg_count += ic->mode == 'J';
3258		arg_count += ic->mode == 'B';
3259		arg_count += ic->mode == 'B';
3260		arg_count += !!ic->internal_hash_alg.alg_string;
3261		arg_count += !!ic->journal_crypt_alg.alg_string;
3262		arg_count += !!ic->journal_mac_alg.alg_string;
3263		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3264		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3265		arg_count += ic->legacy_recalculate;
3266		DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3267		       ic->tag_size, ic->mode, arg_count);
3268		if (ic->meta_dev)
3269			DMEMIT(" meta_device:%s", ic->meta_dev->name);
3270		if (ic->sectors_per_block != 1)
3271			DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3272		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3273			DMEMIT(" recalculate");
3274		if (ic->reset_recalculate_flag)
3275			DMEMIT(" reset_recalculate");
3276		if (ic->discard)
3277			DMEMIT(" allow_discards");
3278		DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3279		DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3280		DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3281		if (ic->mode == 'J') {
3282			DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3283			DMEMIT(" commit_time:%u", ic->autocommit_msec);
3284		}
3285		if (ic->mode == 'B') {
3286			DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3287			DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3288		}
3289		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3290			DMEMIT(" fix_padding");
3291		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3292			DMEMIT(" fix_hmac");
3293		if (ic->legacy_recalculate)
3294			DMEMIT(" legacy_recalculate");
3295
3296#define EMIT_ALG(a, n)							\
3297		do {							\
3298			if (ic->a.alg_string) {				\
3299				DMEMIT(" %s:%s", n, ic->a.alg_string);	\
3300				if (ic->a.key_string)			\
3301					DMEMIT(":%s", ic->a.key_string);\
3302			}						\
3303		} while (0)
3304		EMIT_ALG(internal_hash_alg, "internal_hash");
3305		EMIT_ALG(journal_crypt_alg, "journal_crypt");
3306		EMIT_ALG(journal_mac_alg, "journal_mac");
3307		break;
3308	}
3309	}
3310}
3311
3312static int dm_integrity_iterate_devices(struct dm_target *ti,
3313					iterate_devices_callout_fn fn, void *data)
3314{
3315	struct dm_integrity_c *ic = ti->private;
3316
3317	if (!ic->meta_dev)
3318		return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3319	else
3320		return fn(ti, ic->dev, 0, ti->len, data);
3321}
3322
3323static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3324{
3325	struct dm_integrity_c *ic = ti->private;
3326
3327	if (ic->sectors_per_block > 1) {
3328		limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3329		limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3330		blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3331	}
3332}
3333
3334static void calculate_journal_section_size(struct dm_integrity_c *ic)
3335{
3336	unsigned sector_space = JOURNAL_SECTOR_DATA;
3337
3338	ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3339	ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3340					 JOURNAL_ENTRY_ROUNDUP);
3341
3342	if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3343		sector_space -= JOURNAL_MAC_PER_SECTOR;
3344	ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3345	ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3346	ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3347	ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3348}
3349
3350static int calculate_device_limits(struct dm_integrity_c *ic)
3351{
3352	__u64 initial_sectors;
3353
3354	calculate_journal_section_size(ic);
3355	initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3356	if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3357		return -EINVAL;
3358	ic->initial_sectors = initial_sectors;
3359
3360	if (!ic->meta_dev) {
3361		sector_t last_sector, last_area, last_offset;
3362
3363		/* we have to maintain excessive padding for compatibility with existing volumes */
3364		__u64 metadata_run_padding =
3365			ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3366			(__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3367			(__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3368
3369		ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3370					    metadata_run_padding) >> SECTOR_SHIFT;
3371		if (!(ic->metadata_run & (ic->metadata_run - 1)))
3372			ic->log2_metadata_run = __ffs(ic->metadata_run);
3373		else
3374			ic->log2_metadata_run = -1;
3375
3376		get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3377		last_sector = get_data_sector(ic, last_area, last_offset);
3378		if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3379			return -EINVAL;
3380	} else {
3381		__u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3382		meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3383				>> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3384		meta_size <<= ic->log2_buffer_sectors;
3385		if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3386		    ic->initial_sectors + meta_size > ic->meta_device_sectors)
3387			return -EINVAL;
3388		ic->metadata_run = 1;
3389		ic->log2_metadata_run = 0;
3390	}
3391
3392	return 0;
3393}
3394
3395static void get_provided_data_sectors(struct dm_integrity_c *ic)
3396{
3397	if (!ic->meta_dev) {
3398		int test_bit;
3399		ic->provided_data_sectors = 0;
3400		for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3401			__u64 prev_data_sectors = ic->provided_data_sectors;
3402
3403			ic->provided_data_sectors |= (sector_t)1 << test_bit;
3404			if (calculate_device_limits(ic))
3405				ic->provided_data_sectors = prev_data_sectors;
3406		}
3407	} else {
3408		ic->provided_data_sectors = ic->data_device_sectors;
3409		ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3410	}
3411}
3412
3413static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3414{
3415	unsigned journal_sections;
3416	int test_bit;
3417
3418	memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3419	memcpy(ic->sb->magic, SB_MAGIC, 8);
3420	ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3421	ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3422	if (ic->journal_mac_alg.alg_string)
3423		ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3424
3425	calculate_journal_section_size(ic);
3426	journal_sections = journal_sectors / ic->journal_section_sectors;
3427	if (!journal_sections)
3428		journal_sections = 1;
3429
3430	if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3431		ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3432		get_random_bytes(ic->sb->salt, SALT_SIZE);
3433	}
3434
3435	if (!ic->meta_dev) {
3436		if (ic->fix_padding)
3437			ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3438		ic->sb->journal_sections = cpu_to_le32(journal_sections);
3439		if (!interleave_sectors)
3440			interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3441		ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3442		ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3443		ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3444
3445		get_provided_data_sectors(ic);
3446		if (!ic->provided_data_sectors)
3447			return -EINVAL;
3448	} else {
3449		ic->sb->log2_interleave_sectors = 0;
3450
3451		get_provided_data_sectors(ic);
3452		if (!ic->provided_data_sectors)
3453			return -EINVAL;
3454
3455try_smaller_buffer:
3456		ic->sb->journal_sections = cpu_to_le32(0);
3457		for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3458			__u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3459			__u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3460			if (test_journal_sections > journal_sections)
3461				continue;
3462			ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3463			if (calculate_device_limits(ic))
3464				ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3465
3466		}
3467		if (!le32_to_cpu(ic->sb->journal_sections)) {
3468			if (ic->log2_buffer_sectors > 3) {
3469				ic->log2_buffer_sectors--;
3470				goto try_smaller_buffer;
3471			}
3472			return -EINVAL;
3473		}
3474	}
3475
3476	ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3477
3478	sb_set_version(ic);
3479
3480	return 0;
3481}
3482
3483static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3484{
3485	struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3486	struct blk_integrity bi;
3487
3488	memset(&bi, 0, sizeof(bi));
3489	bi.profile = &dm_integrity_profile;
3490	bi.tuple_size = ic->tag_size;
3491	bi.tag_size = bi.tuple_size;
3492	bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3493
3494	blk_integrity_register(disk, &bi);
3495	blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3496}
3497
3498static void dm_integrity_free_page_list(struct page_list *pl)
3499{
3500	unsigned i;
3501
3502	if (!pl)
3503		return;
3504	for (i = 0; pl[i].page; i++)
3505		__free_page(pl[i].page);
3506	kvfree(pl);
3507}
3508
3509static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3510{
3511	struct page_list *pl;
3512	unsigned i;
3513
3514	pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3515	if (!pl)
3516		return NULL;
3517
3518	for (i = 0; i < n_pages; i++) {
3519		pl[i].page = alloc_page(GFP_KERNEL);
3520		if (!pl[i].page) {
3521			dm_integrity_free_page_list(pl);
3522			return NULL;
3523		}
3524		if (i)
3525			pl[i - 1].next = &pl[i];
3526	}
3527	pl[i].page = NULL;
3528	pl[i].next = NULL;
3529
3530	return pl;
3531}
3532
3533static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3534{
3535	unsigned i;
3536	for (i = 0; i < ic->journal_sections; i++)
3537		kvfree(sl[i]);
3538	kvfree(sl);
3539}
3540
3541static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3542								   struct page_list *pl)
3543{
3544	struct scatterlist **sl;
3545	unsigned i;
3546
3547	sl = kvmalloc_array(ic->journal_sections,
3548			    sizeof(struct scatterlist *),
3549			    GFP_KERNEL | __GFP_ZERO);
3550	if (!sl)
3551		return NULL;
3552
3553	for (i = 0; i < ic->journal_sections; i++) {
3554		struct scatterlist *s;
3555		unsigned start_index, start_offset;
3556		unsigned end_index, end_offset;
3557		unsigned n_pages;
3558		unsigned idx;
3559
3560		page_list_location(ic, i, 0, &start_index, &start_offset);
3561		page_list_location(ic, i, ic->journal_section_sectors - 1,
3562				   &end_index, &end_offset);
3563
3564		n_pages = (end_index - start_index + 1);
3565
3566		s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3567				   GFP_KERNEL);
3568		if (!s) {
3569			dm_integrity_free_journal_scatterlist(ic, sl);
3570			return NULL;
3571		}
3572
3573		sg_init_table(s, n_pages);
3574		for (idx = start_index; idx <= end_index; idx++) {
3575			char *va = lowmem_page_address(pl[idx].page);
3576			unsigned start = 0, end = PAGE_SIZE;
3577			if (idx == start_index)
3578				start = start_offset;
3579			if (idx == end_index)
3580				end = end_offset + (1 << SECTOR_SHIFT);
3581			sg_set_buf(&s[idx - start_index], va + start, end - start);
3582		}
3583
3584		sl[i] = s;
3585	}
3586
3587	return sl;
3588}
3589
3590static void free_alg(struct alg_spec *a)
3591{
3592	kfree_sensitive(a->alg_string);
3593	kfree_sensitive(a->key);
3594	memset(a, 0, sizeof *a);
3595}
3596
3597static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3598{
3599	char *k;
3600
3601	free_alg(a);
3602
3603	a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3604	if (!a->alg_string)
3605		goto nomem;
3606
3607	k = strchr(a->alg_string, ':');
3608	if (k) {
3609		*k = 0;
3610		a->key_string = k + 1;
3611		if (strlen(a->key_string) & 1)
3612			goto inval;
3613
3614		a->key_size = strlen(a->key_string) / 2;
3615		a->key = kmalloc(a->key_size, GFP_KERNEL);
3616		if (!a->key)
3617			goto nomem;
3618		if (hex2bin(a->key, a->key_string, a->key_size))
3619			goto inval;
3620	}
3621
3622	return 0;
3623inval:
3624	*error = error_inval;
3625	return -EINVAL;
3626nomem:
3627	*error = "Out of memory for an argument";
3628	return -ENOMEM;
3629}
3630
3631static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3632		   char *error_alg, char *error_key)
3633{
3634	int r;
3635
3636	if (a->alg_string) {
3637		*hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3638		if (IS_ERR(*hash)) {
3639			*error = error_alg;
3640			r = PTR_ERR(*hash);
3641			*hash = NULL;
3642			return r;
3643		}
3644
3645		if (a->key) {
3646			r = crypto_shash_setkey(*hash, a->key, a->key_size);
3647			if (r) {
3648				*error = error_key;
3649				return r;
3650			}
3651		} else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3652			*error = error_key;
3653			return -ENOKEY;
3654		}
3655	}
3656
3657	return 0;
3658}
3659
3660static int create_journal(struct dm_integrity_c *ic, char **error)
3661{
3662	int r = 0;
3663	unsigned i;
3664	__u64 journal_pages, journal_desc_size, journal_tree_size;
3665	unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3666	struct skcipher_request *req = NULL;
3667
3668	ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3669	ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3670	ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3671	ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3672
3673	journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3674				PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3675	journal_desc_size = journal_pages * sizeof(struct page_list);
3676	if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3677		*error = "Journal doesn't fit into memory";
3678		r = -ENOMEM;
3679		goto bad;
3680	}
3681	ic->journal_pages = journal_pages;
3682
3683	ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3684	if (!ic->journal) {
3685		*error = "Could not allocate memory for journal";
3686		r = -ENOMEM;
3687		goto bad;
3688	}
3689	if (ic->journal_crypt_alg.alg_string) {
3690		unsigned ivsize, blocksize;
3691		struct journal_completion comp;
3692
3693		comp.ic = ic;
3694		ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3695		if (IS_ERR(ic->journal_crypt)) {
3696			*error = "Invalid journal cipher";
3697			r = PTR_ERR(ic->journal_crypt);
3698			ic->journal_crypt = NULL;
3699			goto bad;
3700		}
3701		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3702		blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3703
3704		if (ic->journal_crypt_alg.key) {
3705			r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3706						   ic->journal_crypt_alg.key_size);
3707			if (r) {
3708				*error = "Error setting encryption key";
3709				goto bad;
3710			}
3711		}
3712		DEBUG_print("cipher %s, block size %u iv size %u\n",
3713			    ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3714
3715		ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3716		if (!ic->journal_io) {
3717			*error = "Could not allocate memory for journal io";
3718			r = -ENOMEM;
3719			goto bad;
3720		}
3721
3722		if (blocksize == 1) {
3723			struct scatterlist *sg;
3724
3725			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3726			if (!req) {
3727				*error = "Could not allocate crypt request";
3728				r = -ENOMEM;
3729				goto bad;
3730			}
3731
3732			crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3733			if (!crypt_iv) {
3734				*error = "Could not allocate iv";
3735				r = -ENOMEM;
3736				goto bad;
3737			}
3738
3739			ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3740			if (!ic->journal_xor) {
3741				*error = "Could not allocate memory for journal xor";
3742				r = -ENOMEM;
3743				goto bad;
3744			}
3745
3746			sg = kvmalloc_array(ic->journal_pages + 1,
3747					    sizeof(struct scatterlist),
3748					    GFP_KERNEL);
3749			if (!sg) {
3750				*error = "Unable to allocate sg list";
3751				r = -ENOMEM;
3752				goto bad;
3753			}
3754			sg_init_table(sg, ic->journal_pages + 1);
3755			for (i = 0; i < ic->journal_pages; i++) {
3756				char *va = lowmem_page_address(ic->journal_xor[i].page);
3757				clear_page(va);
3758				sg_set_buf(&sg[i], va, PAGE_SIZE);
3759			}
3760			sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3761
3762			skcipher_request_set_crypt(req, sg, sg,
3763						   PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3764			init_completion(&comp.comp);
3765			comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3766			if (do_crypt(true, req, &comp))
3767				wait_for_completion(&comp.comp);
3768			kvfree(sg);
3769			r = dm_integrity_failed(ic);
3770			if (r) {
3771				*error = "Unable to encrypt journal";
3772				goto bad;
3773			}
3774			DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3775
3776			crypto_free_skcipher(ic->journal_crypt);
3777			ic->journal_crypt = NULL;
3778		} else {
3779			unsigned crypt_len = roundup(ivsize, blocksize);
3780
3781			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3782			if (!req) {
3783				*error = "Could not allocate crypt request";
3784				r = -ENOMEM;
3785				goto bad;
3786			}
3787
3788			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3789			if (!crypt_iv) {
3790				*error = "Could not allocate iv";
3791				r = -ENOMEM;
3792				goto bad;
3793			}
3794
3795			crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3796			if (!crypt_data) {
3797				*error = "Unable to allocate crypt data";
3798				r = -ENOMEM;
3799				goto bad;
3800			}
3801
3802			ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3803			if (!ic->journal_scatterlist) {
3804				*error = "Unable to allocate sg list";
3805				r = -ENOMEM;
3806				goto bad;
3807			}
3808			ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3809			if (!ic->journal_io_scatterlist) {
3810				*error = "Unable to allocate sg list";
3811				r = -ENOMEM;
3812				goto bad;
3813			}
3814			ic->sk_requests = kvmalloc_array(ic->journal_sections,
3815							 sizeof(struct skcipher_request *),
3816							 GFP_KERNEL | __GFP_ZERO);
3817			if (!ic->sk_requests) {
3818				*error = "Unable to allocate sk requests";
3819				r = -ENOMEM;
3820				goto bad;
3821			}
3822			for (i = 0; i < ic->journal_sections; i++) {
3823				struct scatterlist sg;
3824				struct skcipher_request *section_req;
3825				__le32 section_le = cpu_to_le32(i);
3826
3827				memset(crypt_iv, 0x00, ivsize);
3828				memset(crypt_data, 0x00, crypt_len);
3829				memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3830
3831				sg_init_one(&sg, crypt_data, crypt_len);
3832				skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3833				init_completion(&comp.comp);
3834				comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3835				if (do_crypt(true, req, &comp))
3836					wait_for_completion(&comp.comp);
3837
3838				r = dm_integrity_failed(ic);
3839				if (r) {
3840					*error = "Unable to generate iv";
3841					goto bad;
3842				}
3843
3844				section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3845				if (!section_req) {
3846					*error = "Unable to allocate crypt request";
3847					r = -ENOMEM;
3848					goto bad;
3849				}
3850				section_req->iv = kmalloc_array(ivsize, 2,
3851								GFP_KERNEL);
3852				if (!section_req->iv) {
3853					skcipher_request_free(section_req);
3854					*error = "Unable to allocate iv";
3855					r = -ENOMEM;
3856					goto bad;
3857				}
3858				memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3859				section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3860				ic->sk_requests[i] = section_req;
3861				DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3862			}
3863		}
3864	}
3865
3866	for (i = 0; i < N_COMMIT_IDS; i++) {
3867		unsigned j;
3868retest_commit_id:
3869		for (j = 0; j < i; j++) {
3870			if (ic->commit_ids[j] == ic->commit_ids[i]) {
3871				ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3872				goto retest_commit_id;
3873			}
3874		}
3875		DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3876	}
3877
3878	journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3879	if (journal_tree_size > ULONG_MAX) {
3880		*error = "Journal doesn't fit into memory";
3881		r = -ENOMEM;
3882		goto bad;
3883	}
3884	ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3885	if (!ic->journal_tree) {
3886		*error = "Could not allocate memory for journal tree";
3887		r = -ENOMEM;
3888	}
3889bad:
3890	kfree(crypt_data);
3891	kfree(crypt_iv);
3892	skcipher_request_free(req);
3893
3894	return r;
3895}
3896
3897/*
3898 * Construct a integrity mapping
3899 *
3900 * Arguments:
3901 *	device
3902 *	offset from the start of the device
3903 *	tag size
3904 *	D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3905 *	number of optional arguments
3906 *	optional arguments:
3907 *		journal_sectors
3908 *		interleave_sectors
3909 *		buffer_sectors
3910 *		journal_watermark
3911 *		commit_time
3912 *		meta_device
3913 *		block_size
3914 *		sectors_per_bit
3915 *		bitmap_flush_interval
3916 *		internal_hash
3917 *		journal_crypt
3918 *		journal_mac
3919 *		recalculate
3920 */
3921static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3922{
3923	struct dm_integrity_c *ic;
3924	char dummy;
3925	int r;
3926	unsigned extra_args;
3927	struct dm_arg_set as;
3928	static const struct dm_arg _args[] = {
3929		{0, 18, "Invalid number of feature args"},
3930	};
3931	unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3932	bool should_write_sb;
3933	__u64 threshold;
3934	unsigned long long start;
3935	__s8 log2_sectors_per_bitmap_bit = -1;
3936	__s8 log2_blocks_per_bitmap_bit;
3937	__u64 bits_in_journal;
3938	__u64 n_bitmap_bits;
3939
3940#define DIRECT_ARGUMENTS	4
3941
3942	if (argc <= DIRECT_ARGUMENTS) {
3943		ti->error = "Invalid argument count";
3944		return -EINVAL;
3945	}
3946
3947	ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3948	if (!ic) {
3949		ti->error = "Cannot allocate integrity context";
3950		return -ENOMEM;
3951	}
3952	ti->private = ic;
3953	ti->per_io_data_size = sizeof(struct dm_integrity_io);
3954	ic->ti = ti;
3955
3956	ic->in_progress = RB_ROOT;
3957	INIT_LIST_HEAD(&ic->wait_list);
3958	init_waitqueue_head(&ic->endio_wait);
3959	bio_list_init(&ic->flush_bio_list);
3960	init_waitqueue_head(&ic->copy_to_journal_wait);
3961	init_completion(&ic->crypto_backoff);
3962	atomic64_set(&ic->number_of_mismatches, 0);
3963	ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3964
3965	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3966	if (r) {
3967		ti->error = "Device lookup failed";
3968		goto bad;
3969	}
3970
3971	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3972		ti->error = "Invalid starting offset";
3973		r = -EINVAL;
3974		goto bad;
3975	}
3976	ic->start = start;
3977
3978	if (strcmp(argv[2], "-")) {
3979		if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3980			ti->error = "Invalid tag size";
3981			r = -EINVAL;
3982			goto bad;
3983		}
3984	}
3985
3986	if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3987	    !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3988		ic->mode = argv[3][0];
3989	} else {
3990		ti->error = "Invalid mode (expecting J, B, D, R)";
3991		r = -EINVAL;
3992		goto bad;
3993	}
3994
3995	journal_sectors = 0;
3996	interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3997	buffer_sectors = DEFAULT_BUFFER_SECTORS;
3998	journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3999	sync_msec = DEFAULT_SYNC_MSEC;
4000	ic->sectors_per_block = 1;
4001
4002	as.argc = argc - DIRECT_ARGUMENTS;
4003	as.argv = argv + DIRECT_ARGUMENTS;
4004	r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4005	if (r)
4006		goto bad;
4007
4008	while (extra_args--) {
4009		const char *opt_string;
4010		unsigned val;
4011		unsigned long long llval;
4012		opt_string = dm_shift_arg(&as);
4013		if (!opt_string) {
4014			r = -EINVAL;
4015			ti->error = "Not enough feature arguments";
4016			goto bad;
4017		}
4018		if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4019			journal_sectors = val ? val : 1;
4020		else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4021			interleave_sectors = val;
4022		else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4023			buffer_sectors = val;
4024		else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4025			journal_watermark = val;
4026		else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4027			sync_msec = val;
4028		else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4029			if (ic->meta_dev) {
4030				dm_put_device(ti, ic->meta_dev);
4031				ic->meta_dev = NULL;
4032			}
4033			r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4034					  dm_table_get_mode(ti->table), &ic->meta_dev);
4035			if (r) {
4036				ti->error = "Device lookup failed";
4037				goto bad;
4038			}
4039		} else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4040			if (val < 1 << SECTOR_SHIFT ||
4041			    val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4042			    (val & (val -1))) {
4043				r = -EINVAL;
4044				ti->error = "Invalid block_size argument";
4045				goto bad;
4046			}
4047			ic->sectors_per_block = val >> SECTOR_SHIFT;
4048		} else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4049			log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4050		} else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4051			if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4052				r = -EINVAL;
4053				ti->error = "Invalid bitmap_flush_interval argument";
4054				goto bad;
4055			}
4056			ic->bitmap_flush_interval = msecs_to_jiffies(val);
4057		} else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4058			r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4059					    "Invalid internal_hash argument");
4060			if (r)
4061				goto bad;
4062		} else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4063			r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4064					    "Invalid journal_crypt argument");
4065			if (r)
4066				goto bad;
4067		} else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4068			r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4069					    "Invalid journal_mac argument");
4070			if (r)
4071				goto bad;
4072		} else if (!strcmp(opt_string, "recalculate")) {
4073			ic->recalculate_flag = true;
4074		} else if (!strcmp(opt_string, "reset_recalculate")) {
4075			ic->recalculate_flag = true;
4076			ic->reset_recalculate_flag = true;
4077		} else if (!strcmp(opt_string, "allow_discards")) {
4078			ic->discard = true;
4079		} else if (!strcmp(opt_string, "fix_padding")) {
4080			ic->fix_padding = true;
4081		} else if (!strcmp(opt_string, "fix_hmac")) {
4082			ic->fix_hmac = true;
4083		} else if (!strcmp(opt_string, "legacy_recalculate")) {
4084			ic->legacy_recalculate = true;
4085		} else {
4086			r = -EINVAL;
4087			ti->error = "Invalid argument";
4088			goto bad;
4089		}
4090	}
4091
4092	ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
4093	if (!ic->meta_dev)
4094		ic->meta_device_sectors = ic->data_device_sectors;
4095	else
4096		ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
4097
4098	if (!journal_sectors) {
4099		journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4100				      ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4101	}
4102
4103	if (!buffer_sectors)
4104		buffer_sectors = 1;
4105	ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4106
4107	r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4108		    "Invalid internal hash", "Error setting internal hash key");
4109	if (r)
4110		goto bad;
4111
4112	r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4113		    "Invalid journal mac", "Error setting journal mac key");
4114	if (r)
4115		goto bad;
4116
4117	if (!ic->tag_size) {
4118		if (!ic->internal_hash) {
4119			ti->error = "Unknown tag size";
4120			r = -EINVAL;
4121			goto bad;
4122		}
4123		ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4124	}
4125	if (ic->tag_size > MAX_TAG_SIZE) {
4126		ti->error = "Too big tag size";
4127		r = -EINVAL;
4128		goto bad;
4129	}
4130	if (!(ic->tag_size & (ic->tag_size - 1)))
4131		ic->log2_tag_size = __ffs(ic->tag_size);
4132	else
4133		ic->log2_tag_size = -1;
4134
4135	if (ic->mode == 'B' && !ic->internal_hash) {
4136		r = -EINVAL;
4137		ti->error = "Bitmap mode can be only used with internal hash";
4138		goto bad;
4139	}
4140
4141	if (ic->discard && !ic->internal_hash) {
4142		r = -EINVAL;
4143		ti->error = "Discard can be only used with internal hash";
4144		goto bad;
4145	}
4146
4147	ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4148	ic->autocommit_msec = sync_msec;
4149	timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4150
4151	ic->io = dm_io_client_create();
4152	if (IS_ERR(ic->io)) {
4153		r = PTR_ERR(ic->io);
4154		ic->io = NULL;
4155		ti->error = "Cannot allocate dm io";
4156		goto bad;
4157	}
4158
4159	r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4160	if (r) {
4161		ti->error = "Cannot allocate mempool";
4162		goto bad;
4163	}
4164
4165	ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4166					  WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4167	if (!ic->metadata_wq) {
4168		ti->error = "Cannot allocate workqueue";
4169		r = -ENOMEM;
4170		goto bad;
4171	}
4172
4173	/*
4174	 * If this workqueue were percpu, it would cause bio reordering
4175	 * and reduced performance.
4176	 */
4177	ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4178	if (!ic->wait_wq) {
4179		ti->error = "Cannot allocate workqueue";
4180		r = -ENOMEM;
4181		goto bad;
4182	}
4183
4184	ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4185					  METADATA_WORKQUEUE_MAX_ACTIVE);
4186	if (!ic->offload_wq) {
4187		ti->error = "Cannot allocate workqueue";
4188		r = -ENOMEM;
4189		goto bad;
4190	}
4191
4192	ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4193	if (!ic->commit_wq) {
4194		ti->error = "Cannot allocate workqueue";
4195		r = -ENOMEM;
4196		goto bad;
4197	}
4198	INIT_WORK(&ic->commit_work, integrity_commit);
4199
4200	if (ic->mode == 'J' || ic->mode == 'B') {
4201		ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4202		if (!ic->writer_wq) {
4203			ti->error = "Cannot allocate workqueue";
4204			r = -ENOMEM;
4205			goto bad;
4206		}
4207		INIT_WORK(&ic->writer_work, integrity_writer);
4208	}
4209
4210	ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4211	if (!ic->sb) {
4212		r = -ENOMEM;
4213		ti->error = "Cannot allocate superblock area";
4214		goto bad;
4215	}
4216
4217	r = sync_rw_sb(ic, REQ_OP_READ, 0);
4218	if (r) {
4219		ti->error = "Error reading superblock";
4220		goto bad;
4221	}
4222	should_write_sb = false;
4223	if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4224		if (ic->mode != 'R') {
4225			if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4226				r = -EINVAL;
4227				ti->error = "The device is not initialized";
4228				goto bad;
4229			}
4230		}
4231
4232		r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4233		if (r) {
4234			ti->error = "Could not initialize superblock";
4235			goto bad;
4236		}
4237		if (ic->mode != 'R')
4238			should_write_sb = true;
4239	}
4240
4241	if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4242		r = -EINVAL;
4243		ti->error = "Unknown version";
4244		goto bad;
4245	}
4246	if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4247		r = -EINVAL;
4248		ti->error = "Tag size doesn't match the information in superblock";
4249		goto bad;
4250	}
4251	if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4252		r = -EINVAL;
4253		ti->error = "Block size doesn't match the information in superblock";
4254		goto bad;
4255	}
4256	if (!le32_to_cpu(ic->sb->journal_sections)) {
4257		r = -EINVAL;
4258		ti->error = "Corrupted superblock, journal_sections is 0";
4259		goto bad;
4260	}
4261	/* make sure that ti->max_io_len doesn't overflow */
4262	if (!ic->meta_dev) {
4263		if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4264		    ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4265			r = -EINVAL;
4266			ti->error = "Invalid interleave_sectors in the superblock";
4267			goto bad;
4268		}
4269	} else {
4270		if (ic->sb->log2_interleave_sectors) {
4271			r = -EINVAL;
4272			ti->error = "Invalid interleave_sectors in the superblock";
4273			goto bad;
4274		}
4275	}
4276	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4277		r = -EINVAL;
4278		ti->error = "Journal mac mismatch";
4279		goto bad;
4280	}
4281
4282	get_provided_data_sectors(ic);
4283	if (!ic->provided_data_sectors) {
4284		r = -EINVAL;
4285		ti->error = "The device is too small";
4286		goto bad;
4287	}
4288
4289try_smaller_buffer:
4290	r = calculate_device_limits(ic);
4291	if (r) {
4292		if (ic->meta_dev) {
4293			if (ic->log2_buffer_sectors > 3) {
4294				ic->log2_buffer_sectors--;
4295				goto try_smaller_buffer;
4296			}
4297		}
4298		ti->error = "The device is too small";
4299		goto bad;
4300	}
4301
4302	if (log2_sectors_per_bitmap_bit < 0)
4303		log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4304	if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4305		log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4306
4307	bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4308	if (bits_in_journal > UINT_MAX)
4309		bits_in_journal = UINT_MAX;
4310	while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4311		log2_sectors_per_bitmap_bit++;
4312
4313	log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4314	ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4315	if (should_write_sb) {
4316		ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4317	}
4318	n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4319				+ (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4320	ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4321
4322	if (!ic->meta_dev)
4323		ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4324
4325	if (ti->len > ic->provided_data_sectors) {
4326		r = -EINVAL;
4327		ti->error = "Not enough provided sectors for requested mapping size";
4328		goto bad;
4329	}
4330
4331
4332	threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4333	threshold += 50;
4334	do_div(threshold, 100);
4335	ic->free_sectors_threshold = threshold;
4336
4337	DEBUG_print("initialized:\n");
4338	DEBUG_print("	integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4339	DEBUG_print("	journal_entry_size %u\n", ic->journal_entry_size);
4340	DEBUG_print("	journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4341	DEBUG_print("	journal_section_entries %u\n", ic->journal_section_entries);
4342	DEBUG_print("	journal_section_sectors %u\n", ic->journal_section_sectors);
4343	DEBUG_print("	journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4344	DEBUG_print("	journal_entries %u\n", ic->journal_entries);
4345	DEBUG_print("	log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4346	DEBUG_print("	data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4347	DEBUG_print("	initial_sectors 0x%x\n", ic->initial_sectors);
4348	DEBUG_print("	metadata_run 0x%x\n", ic->metadata_run);
4349	DEBUG_print("	log2_metadata_run %d\n", ic->log2_metadata_run);
4350	DEBUG_print("	provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4351	DEBUG_print("	log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4352	DEBUG_print("	bits_in_journal %llu\n", bits_in_journal);
4353
4354	if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4355		ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4356		ic->sb->recalc_sector = cpu_to_le64(0);
4357	}
4358
4359	if (ic->internal_hash) {
4360		ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4361		if (!ic->recalc_wq ) {
4362			ti->error = "Cannot allocate workqueue";
4363			r = -ENOMEM;
4364			goto bad;
4365		}
4366		INIT_WORK(&ic->recalc_work, integrity_recalc);
4367		ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4368		if (!ic->recalc_buffer) {
4369			ti->error = "Cannot allocate buffer for recalculating";
4370			r = -ENOMEM;
4371			goto bad;
4372		}
4373		ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4374						 ic->tag_size, GFP_KERNEL);
4375		if (!ic->recalc_tags) {
4376			ti->error = "Cannot allocate tags for recalculating";
4377			r = -ENOMEM;
4378			goto bad;
4379		}
4380	} else {
4381		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4382			ti->error = "Recalculate can only be specified with internal_hash";
4383			r = -EINVAL;
4384			goto bad;
4385		}
4386	}
4387
4388	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4389	    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4390	    dm_integrity_disable_recalculate(ic)) {
4391		ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4392		r = -EOPNOTSUPP;
4393		goto bad;
4394	}
4395
4396	ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4397			1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4398	if (IS_ERR(ic->bufio)) {
4399		r = PTR_ERR(ic->bufio);
4400		ti->error = "Cannot initialize dm-bufio";
4401		ic->bufio = NULL;
4402		goto bad;
4403	}
4404	dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4405
4406	if (ic->mode != 'R') {
4407		r = create_journal(ic, &ti->error);
4408		if (r)
4409			goto bad;
4410
4411	}
4412
4413	if (ic->mode == 'B') {
4414		unsigned i;
4415		unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4416
4417		ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4418		if (!ic->recalc_bitmap) {
4419			r = -ENOMEM;
4420			goto bad;
4421		}
4422		ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4423		if (!ic->may_write_bitmap) {
4424			r = -ENOMEM;
4425			goto bad;
4426		}
4427		ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4428		if (!ic->bbs) {
4429			r = -ENOMEM;
4430			goto bad;
4431		}
4432		INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4433		for (i = 0; i < ic->n_bitmap_blocks; i++) {
4434			struct bitmap_block_status *bbs = &ic->bbs[i];
4435			unsigned sector, pl_index, pl_offset;
4436
4437			INIT_WORK(&bbs->work, bitmap_block_work);
4438			bbs->ic = ic;
4439			bbs->idx = i;
4440			bio_list_init(&bbs->bio_queue);
4441			spin_lock_init(&bbs->bio_queue_lock);
4442
4443			sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4444			pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4445			pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4446
4447			bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4448		}
4449	}
4450
4451	if (should_write_sb) {
4452		int r;
4453
4454		init_journal(ic, 0, ic->journal_sections, 0);
4455		r = dm_integrity_failed(ic);
4456		if (unlikely(r)) {
4457			ti->error = "Error initializing journal";
4458			goto bad;
4459		}
4460		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4461		if (r) {
4462			ti->error = "Error initializing superblock";
4463			goto bad;
4464		}
4465		ic->just_formatted = true;
4466	}
4467
4468	if (!ic->meta_dev) {
4469		r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4470		if (r)
4471			goto bad;
4472	}
4473	if (ic->mode == 'B') {
4474		unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4475		if (!max_io_len)
4476			max_io_len = 1U << 31;
4477		DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4478		if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4479			r = dm_set_target_max_io_len(ti, max_io_len);
4480			if (r)
4481				goto bad;
4482		}
4483	}
4484
4485	if (!ic->internal_hash)
4486		dm_integrity_set(ti, ic);
4487
4488	ti->num_flush_bios = 1;
4489	ti->flush_supported = true;
4490	if (ic->discard)
4491		ti->num_discard_bios = 1;
4492
4493	return 0;
4494
4495bad:
4496	dm_integrity_dtr(ti);
4497	return r;
4498}
4499
4500static void dm_integrity_dtr(struct dm_target *ti)
4501{
4502	struct dm_integrity_c *ic = ti->private;
4503
4504	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4505	BUG_ON(!list_empty(&ic->wait_list));
4506
4507	if (ic->metadata_wq)
4508		destroy_workqueue(ic->metadata_wq);
4509	if (ic->wait_wq)
4510		destroy_workqueue(ic->wait_wq);
4511	if (ic->offload_wq)
4512		destroy_workqueue(ic->offload_wq);
4513	if (ic->commit_wq)
4514		destroy_workqueue(ic->commit_wq);
4515	if (ic->writer_wq)
4516		destroy_workqueue(ic->writer_wq);
4517	if (ic->recalc_wq)
4518		destroy_workqueue(ic->recalc_wq);
4519	vfree(ic->recalc_buffer);
4520	kvfree(ic->recalc_tags);
4521	kvfree(ic->bbs);
4522	if (ic->bufio)
4523		dm_bufio_client_destroy(ic->bufio);
4524	mempool_exit(&ic->journal_io_mempool);
4525	if (ic->io)
4526		dm_io_client_destroy(ic->io);
4527	if (ic->dev)
4528		dm_put_device(ti, ic->dev);
4529	if (ic->meta_dev)
4530		dm_put_device(ti, ic->meta_dev);
4531	dm_integrity_free_page_list(ic->journal);
4532	dm_integrity_free_page_list(ic->journal_io);
4533	dm_integrity_free_page_list(ic->journal_xor);
4534	dm_integrity_free_page_list(ic->recalc_bitmap);
4535	dm_integrity_free_page_list(ic->may_write_bitmap);
4536	if (ic->journal_scatterlist)
4537		dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4538	if (ic->journal_io_scatterlist)
4539		dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4540	if (ic->sk_requests) {
4541		unsigned i;
4542
4543		for (i = 0; i < ic->journal_sections; i++) {
4544			struct skcipher_request *req = ic->sk_requests[i];
4545			if (req) {
4546				kfree_sensitive(req->iv);
4547				skcipher_request_free(req);
4548			}
4549		}
4550		kvfree(ic->sk_requests);
4551	}
4552	kvfree(ic->journal_tree);
4553	if (ic->sb)
4554		free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4555
4556	if (ic->internal_hash)
4557		crypto_free_shash(ic->internal_hash);
4558	free_alg(&ic->internal_hash_alg);
4559
4560	if (ic->journal_crypt)
4561		crypto_free_skcipher(ic->journal_crypt);
4562	free_alg(&ic->journal_crypt_alg);
4563
4564	if (ic->journal_mac)
4565		crypto_free_shash(ic->journal_mac);
4566	free_alg(&ic->journal_mac_alg);
4567
4568	kfree(ic);
4569}
4570
4571static struct target_type integrity_target = {
4572	.name			= "integrity",
4573	.version		= {1, 10, 0},
4574	.module			= THIS_MODULE,
4575	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4576	.ctr			= dm_integrity_ctr,
4577	.dtr			= dm_integrity_dtr,
4578	.map			= dm_integrity_map,
4579	.postsuspend		= dm_integrity_postsuspend,
4580	.resume			= dm_integrity_resume,
4581	.status			= dm_integrity_status,
4582	.iterate_devices	= dm_integrity_iterate_devices,
4583	.io_hints		= dm_integrity_io_hints,
4584};
4585
4586static int __init dm_integrity_init(void)
4587{
4588	int r;
4589
4590	journal_io_cache = kmem_cache_create("integrity_journal_io",
4591					     sizeof(struct journal_io), 0, 0, NULL);
4592	if (!journal_io_cache) {
4593		DMERR("can't allocate journal io cache");
4594		return -ENOMEM;
4595	}
4596
4597	r = dm_register_target(&integrity_target);
4598
4599	if (r < 0)
4600		DMERR("register failed %d", r);
4601
4602	return r;
4603}
4604
4605static void __exit dm_integrity_exit(void)
4606{
4607	dm_unregister_target(&integrity_target);
4608	kmem_cache_destroy(journal_io_cache);
4609}
4610
4611module_init(dm_integrity_init);
4612module_exit(dm_integrity_exit);
4613
4614MODULE_AUTHOR("Milan Broz");
4615MODULE_AUTHOR("Mikulas Patocka");
4616MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4617MODULE_LICENSE("GPL");