dm-integrity.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   4 * Copyright (C) 2016-2017 Milan Broz
   5 * Copyright (C) 2016-2017 Mikulas Patocka
   6 *
   7 * This file is released under the GPL.
   8 */
   9
  10#include "dm-bio-record.h"
  11
  12#include <linux/compiler.h>
  13#include <linux/module.h>
  14#include <linux/device-mapper.h>
  15#include <linux/dm-io.h>
  16#include <linux/vmalloc.h>
  17#include <linux/sort.h>
  18#include <linux/rbtree.h>
  19#include <linux/delay.h>
  20#include <linux/random.h>
  21#include <linux/reboot.h>
  22#include <crypto/hash.h>
  23#include <crypto/skcipher.h>
  24#include <crypto/utils.h>
  25#include <linux/async_tx.h>
  26#include <linux/dm-bufio.h>
  27
  28#include "dm-audit.h"
  29
  30#define DM_MSG_PREFIX "integrity"
  31
  32#define DEFAULT_INTERLEAVE_SECTORS	32768
  33#define DEFAULT_JOURNAL_SIZE_FACTOR	7
  34#define DEFAULT_SECTORS_PER_BITMAP_BIT	32768
  35#define DEFAULT_BUFFER_SECTORS		128
  36#define DEFAULT_JOURNAL_WATERMARK	50
  37#define DEFAULT_SYNC_MSEC		10000
  38#define DEFAULT_MAX_JOURNAL_SECTORS	(IS_ENABLED(CONFIG_64BIT) ? 131072 : 8192)
  39#define MIN_LOG2_INTERLEAVE_SECTORS	3
  40#define MAX_LOG2_INTERLEAVE_SECTORS	31
  41#define METADATA_WORKQUEUE_MAX_ACTIVE	16
  42#define RECALC_SECTORS			(IS_ENABLED(CONFIG_64BIT) ? 32768 : 2048)
  43#define RECALC_WRITE_SUPER		16
  44#define BITMAP_BLOCK_SIZE		4096	/* don't change it */
  45#define BITMAP_FLUSH_INTERVAL		(10 * HZ)
  46#define DISCARD_FILLER			0xf6
  47#define SALT_SIZE			16
  48#define RECHECK_POOL_SIZE		256
  49
  50/*
  51 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  52 * so it should not be enabled in the official kernel
  53 */
  54//#define DEBUG_PRINT
  55//#define INTERNAL_VERIFY
  56
  57/*
  58 * On disk structures
  59 */
  60
  61#define SB_MAGIC			"integrt"
  62#define SB_VERSION_1			1
  63#define SB_VERSION_2			2
  64#define SB_VERSION_3			3
  65#define SB_VERSION_4			4
  66#define SB_VERSION_5			5
  67#define SB_VERSION_6			6
  68#define SB_SECTORS			8
  69#define MAX_SECTORS_PER_BLOCK		8
  70
  71struct superblock {
  72	__u8 magic[8];
  73	__u8 version;
  74	__u8 log2_interleave_sectors;
  75	__le16 integrity_tag_size;
  76	__le32 journal_sections;
  77	__le64 provided_data_sectors;	/* userspace uses this value */
  78	__le32 flags;
  79	__u8 log2_sectors_per_block;
  80	__u8 log2_blocks_per_bitmap_bit;
  81	__u8 pad[2];
  82	__le64 recalc_sector;
  83	__u8 pad2[8];
  84	__u8 salt[SALT_SIZE];
  85};
  86
  87#define SB_FLAG_HAVE_JOURNAL_MAC	0x1
  88#define SB_FLAG_RECALCULATING		0x2
  89#define SB_FLAG_DIRTY_BITMAP		0x4
  90#define SB_FLAG_FIXED_PADDING		0x8
  91#define SB_FLAG_FIXED_HMAC		0x10
  92#define SB_FLAG_INLINE			0x20
  93
  94#define	JOURNAL_ENTRY_ROUNDUP		8
  95
  96typedef __le64 commit_id_t;
  97#define JOURNAL_MAC_PER_SECTOR		8
  98
  99struct journal_entry {
 100	union {
 101		struct {
 102			__le32 sector_lo;
 103			__le32 sector_hi;
 104		} s;
 105		__le64 sector;
 106	} u;
 107	commit_id_t last_bytes[];
 108	/* __u8 tag[0]; */
 109};
 110
 111#define journal_entry_tag(ic, je)		((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 112
 113#if BITS_PER_LONG == 64
 114#define journal_entry_set_sector(je, x)		do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 115#else
 116#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)
 117#endif
 118#define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
 119#define journal_entry_is_unused(je)		((je)->u.s.sector_hi == cpu_to_le32(-1))
 120#define journal_entry_set_unused(je)		((je)->u.s.sector_hi = cpu_to_le32(-1))
 121#define journal_entry_is_inprogress(je)		((je)->u.s.sector_hi == cpu_to_le32(-2))
 122#define journal_entry_set_inprogress(je)	((je)->u.s.sector_hi = cpu_to_le32(-2))
 123
 124#define JOURNAL_BLOCK_SECTORS		8
 125#define JOURNAL_SECTOR_DATA		((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 126#define JOURNAL_MAC_SIZE		(JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 127
 128struct journal_sector {
 129	struct_group(sectors,
 130		__u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 131		__u8 mac[JOURNAL_MAC_PER_SECTOR];
 132	);
 133	commit_id_t commit_id;
 134};
 135
 136#define MAX_TAG_SIZE			255
 137
 138#define METADATA_PADDING_SECTORS	8
 139
 140#define N_COMMIT_IDS			4
 141
 142static unsigned char prev_commit_seq(unsigned char seq)
 143{
 144	return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 145}
 146
 147static unsigned char next_commit_seq(unsigned char seq)
 148{
 149	return (seq + 1) % N_COMMIT_IDS;
 150}
 151
 152/*
 153 * In-memory structures
 154 */
 155
 156struct journal_node {
 157	struct rb_node node;
 158	sector_t sector;
 159};
 160
 161struct alg_spec {
 162	char *alg_string;
 163	char *key_string;
 164	__u8 *key;
 165	unsigned int key_size;
 166};
 167
 168struct dm_integrity_c {
 169	struct dm_dev *dev;
 170	struct dm_dev *meta_dev;
 171	unsigned int tag_size;
 172	__s8 log2_tag_size;
 173	unsigned int tuple_size;
 174	sector_t start;
 175	mempool_t journal_io_mempool;
 176	struct dm_io_client *io;
 177	struct dm_bufio_client *bufio;
 178	struct workqueue_struct *metadata_wq;
 179	struct superblock *sb;
 180	unsigned int journal_pages;
 181	unsigned int n_bitmap_blocks;
 182
 183	struct page_list *journal;
 184	struct page_list *journal_io;
 185	struct page_list *journal_xor;
 186	struct page_list *recalc_bitmap;
 187	struct page_list *may_write_bitmap;
 188	struct bitmap_block_status *bbs;
 189	unsigned int bitmap_flush_interval;
 190	int synchronous_mode;
 191	struct bio_list synchronous_bios;
 192	struct delayed_work bitmap_flush_work;
 193
 194	struct crypto_skcipher *journal_crypt;
 195	struct scatterlist **journal_scatterlist;
 196	struct scatterlist **journal_io_scatterlist;
 197	struct skcipher_request **sk_requests;
 198
 199	struct crypto_shash *journal_mac;
 200
 201	struct journal_node *journal_tree;
 202	struct rb_root journal_tree_root;
 203
 204	sector_t provided_data_sectors;
 205
 206	unsigned short journal_entry_size;
 207	unsigned char journal_entries_per_sector;
 208	unsigned char journal_section_entries;
 209	unsigned short journal_section_sectors;
 210	unsigned int journal_sections;
 211	unsigned int journal_entries;
 212	sector_t data_device_sectors;
 213	sector_t meta_device_sectors;
 214	unsigned int initial_sectors;
 215	unsigned int metadata_run;
 216	__s8 log2_metadata_run;
 217	__u8 log2_buffer_sectors;
 218	__u8 sectors_per_block;
 219	__u8 log2_blocks_per_bitmap_bit;
 220
 221	unsigned char mode;
 222	bool internal_hash;
 223
 224	int failed;
 225
 226	struct crypto_shash *internal_shash;
 227	struct crypto_ahash *internal_ahash;
 228	unsigned int internal_hash_digestsize;
 229
 230	struct dm_target *ti;
 231
 232	/* these variables are locked with endio_wait.lock */
 233	struct rb_root in_progress;
 234	struct list_head wait_list;
 235	wait_queue_head_t endio_wait;
 236	struct workqueue_struct *wait_wq;
 237	struct workqueue_struct *offload_wq;
 238
 239	unsigned char commit_seq;
 240	commit_id_t commit_ids[N_COMMIT_IDS];
 241
 242	unsigned int committed_section;
 243	unsigned int n_committed_sections;
 244
 245	unsigned int uncommitted_section;
 246	unsigned int n_uncommitted_sections;
 247
 248	unsigned int free_section;
 249	unsigned char free_section_entry;
 250	unsigned int free_sectors;
 251
 252	unsigned int free_sectors_threshold;
 253
 254	struct workqueue_struct *commit_wq;
 255	struct work_struct commit_work;
 256
 257	struct workqueue_struct *writer_wq;
 258	struct work_struct writer_work;
 259
 260	struct workqueue_struct *recalc_wq;
 261	struct work_struct recalc_work;
 262
 263	struct bio_list flush_bio_list;
 264
 265	unsigned long autocommit_jiffies;
 266	struct timer_list autocommit_timer;
 267	unsigned int autocommit_msec;
 268
 269	wait_queue_head_t copy_to_journal_wait;
 270
 271	struct completion crypto_backoff;
 272
 273	bool wrote_to_journal;
 274	bool journal_uptodate;
 275	bool just_formatted;
 276	bool recalculate_flag;
 277	bool reset_recalculate_flag;
 278	bool discard;
 279	bool fix_padding;
 280	bool fix_hmac;
 281	bool legacy_recalculate;
 282
 283	mempool_t ahash_req_pool;
 284	struct ahash_request *journal_ahash_req;
 285
 286	struct alg_spec internal_hash_alg;
 287	struct alg_spec journal_crypt_alg;
 288	struct alg_spec journal_mac_alg;
 289
 290	atomic64_t number_of_mismatches;
 291
 292	mempool_t recheck_pool;
 293	struct bio_set recheck_bios;
 294	struct bio_set recalc_bios;
 295
 296	struct notifier_block reboot_notifier;
 297};
 298
 299struct dm_integrity_range {
 300	sector_t logical_sector;
 301	sector_t n_sectors;
 302	bool waiting;
 303	union {
 304		struct rb_node node;
 305		struct {
 306			struct task_struct *task;
 307			struct list_head wait_entry;
 308		};
 309	};
 310};
 311
 312struct dm_integrity_io {
 313	struct work_struct work;
 314
 315	struct dm_integrity_c *ic;
 316	enum req_op op;
 317	bool fua;
 318
 319	struct dm_integrity_range range;
 320
 321	sector_t metadata_block;
 322	unsigned int metadata_offset;
 323
 324	atomic_t in_flight;
 325	blk_status_t bi_status;
 326
 327	struct completion *completion;
 328
 329	struct dm_bio_details bio_details;
 330
 331	char *integrity_payload;
 332	unsigned payload_len;
 333	bool integrity_payload_from_mempool;
 334	bool integrity_range_locked;
 335
 336	struct ahash_request *ahash_req;
 337};
 338
 339struct journal_completion {
 340	struct dm_integrity_c *ic;
 341	atomic_t in_flight;
 342	struct completion comp;
 343};
 344
 345struct journal_io {
 346	struct dm_integrity_range range;
 347	struct journal_completion *comp;
 348};
 349
 350struct bitmap_block_status {
 351	struct work_struct work;
 352	struct dm_integrity_c *ic;
 353	unsigned int idx;
 354	unsigned long *bitmap;
 355	struct bio_list bio_queue;
 356	spinlock_t bio_queue_lock;
 357
 358};
 359
 360static struct kmem_cache *journal_io_cache;
 361
 362#define JOURNAL_IO_MEMPOOL	32
 363#define AHASH_MEMPOOL		32
 364
 365#ifdef DEBUG_PRINT
 366#define DEBUG_print(x, ...)			printk(KERN_DEBUG x, ##__VA_ARGS__)
 367#define DEBUG_bytes(bytes, len, msg, ...)	printk(KERN_DEBUG msg "%s%*ph\n", ##__VA_ARGS__, \
 368						       len ? ": " : "", len, bytes)
 369#else
 370#define DEBUG_print(x, ...)			do { } while (0)
 371#define DEBUG_bytes(bytes, len, msg, ...)	do { } while (0)
 372#endif
 373
 374static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 375static int dm_integrity_map_inline(struct dm_integrity_io *dio, bool from_map);
 376static void integrity_bio_wait(struct work_struct *w);
 377static void dm_integrity_dtr(struct dm_target *ti);
 378
 379static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 380{
 381	if (err == -EILSEQ)
 382		atomic64_inc(&ic->number_of_mismatches);
 383	if (!cmpxchg(&ic->failed, 0, err))
 384		DMERR("Error on %s: %d", msg, err);
 385}
 386
 387static int dm_integrity_failed(struct dm_integrity_c *ic)
 388{
 389	return READ_ONCE(ic->failed);
 390}
 391
 392static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 393{
 394	if (ic->legacy_recalculate)
 395		return false;
 396	if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
 397	    ic->internal_hash_alg.key || ic->journal_mac_alg.key :
 398	    ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
 399		return true;
 400	return false;
 401}
 402
 403static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned int i,
 404					  unsigned int j, unsigned char seq)
 405{
 406	/*
 407	 * Xor the number with section and sector, so that if a piece of
 408	 * journal is written at wrong place, it is detected.
 409	 */
 410	return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 411}
 412
 413static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 414				sector_t *area, sector_t *offset)
 415{
 416	if (!ic->meta_dev) {
 417		__u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 418		*area = data_sector >> log2_interleave_sectors;
 419		*offset = (unsigned int)data_sector & ((1U << log2_interleave_sectors) - 1);
 420	} else {
 421		*area = 0;
 422		*offset = data_sector;
 423	}
 424}
 425
 426#define sector_to_block(ic, n)						\
 427do {									\
 428	BUG_ON((n) & (unsigned int)((ic)->sectors_per_block - 1));		\
 429	(n) >>= (ic)->sb->log2_sectors_per_block;			\
 430} while (0)
 431
 432static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 433					    sector_t offset, unsigned int *metadata_offset)
 434{
 435	__u64 ms;
 436	unsigned int mo;
 437
 438	ms = area << ic->sb->log2_interleave_sectors;
 439	if (likely(ic->log2_metadata_run >= 0))
 440		ms += area << ic->log2_metadata_run;
 441	else
 442		ms += area * ic->metadata_run;
 443	ms >>= ic->log2_buffer_sectors;
 444
 445	sector_to_block(ic, offset);
 446
 447	if (likely(ic->log2_tag_size >= 0)) {
 448		ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 449		mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 450	} else {
 451		ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 452		mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 453	}
 454	*metadata_offset = mo;
 455	return ms;
 456}
 457
 458static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 459{
 460	sector_t result;
 461
 462	if (ic->meta_dev)
 463		return offset;
 464
 465	result = area << ic->sb->log2_interleave_sectors;
 466	if (likely(ic->log2_metadata_run >= 0))
 467		result += (area + 1) << ic->log2_metadata_run;
 468	else
 469		result += (area + 1) * ic->metadata_run;
 470
 471	result += (sector_t)ic->initial_sectors + offset;
 472	result += ic->start;
 473
 474	return result;
 475}
 476
 477static void wraparound_section(struct dm_integrity_c *ic, unsigned int *sec_ptr)
 478{
 479	if (unlikely(*sec_ptr >= ic->journal_sections))
 480		*sec_ptr -= ic->journal_sections;
 481}
 482
 483static void sb_set_version(struct dm_integrity_c *ic)
 484{
 485	if (ic->sb->flags & cpu_to_le32(SB_FLAG_INLINE))
 486		ic->sb->version = SB_VERSION_6;
 487	else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
 488		ic->sb->version = SB_VERSION_5;
 489	else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 490		ic->sb->version = SB_VERSION_4;
 491	else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 492		ic->sb->version = SB_VERSION_3;
 493	else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 494		ic->sb->version = SB_VERSION_2;
 495	else
 496		ic->sb->version = SB_VERSION_1;
 497}
 498
 499static int sb_mac(struct dm_integrity_c *ic, bool wr)
 500{
 501	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 502	int r;
 503	unsigned int mac_size = crypto_shash_digestsize(ic->journal_mac);
 504	__u8 *sb = (__u8 *)ic->sb;
 505	__u8 *mac = sb + (1 << SECTOR_SHIFT) - mac_size;
 506
 507	if (sizeof(struct superblock) + mac_size > 1 << SECTOR_SHIFT ||
 508	    mac_size > HASH_MAX_DIGESTSIZE) {
 509		dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 510		return -EINVAL;
 511	}
 512
 513	desc->tfm = ic->journal_mac;
 514
 515	if (likely(wr)) {
 516		r = crypto_shash_digest(desc, sb, mac - sb, mac);
 517		if (unlikely(r < 0)) {
 518			dm_integrity_io_error(ic, "crypto_shash_digest", r);
 519			return r;
 520		}
 521	} else {
 522		__u8 actual_mac[HASH_MAX_DIGESTSIZE];
 523
 524		r = crypto_shash_digest(desc, sb, mac - sb, actual_mac);
 525		if (unlikely(r < 0)) {
 526			dm_integrity_io_error(ic, "crypto_shash_digest", r);
 527			return r;
 528		}
 529		if (crypto_memneq(mac, actual_mac, mac_size)) {
 530			dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 531			dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
 532			return -EILSEQ;
 533		}
 534	}
 535
 536	return 0;
 537}
 538
 539static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
 540{
 541	struct dm_io_request io_req;
 542	struct dm_io_region io_loc;
 543	const enum req_op op = opf & REQ_OP_MASK;
 544	int r;
 545
 546	io_req.bi_opf = opf;
 547	io_req.mem.type = DM_IO_KMEM;
 548	io_req.mem.ptr.addr = ic->sb;
 549	io_req.notify.fn = NULL;
 550	io_req.client = ic->io;
 551	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 552	io_loc.sector = ic->start;
 553	io_loc.count = SB_SECTORS;
 554
 555	if (op == REQ_OP_WRITE) {
 556		sb_set_version(ic);
 557		if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 558			r = sb_mac(ic, true);
 559			if (unlikely(r))
 560				return r;
 561		}
 562	}
 563
 564	r = dm_io(&io_req, 1, &io_loc, NULL, IOPRIO_DEFAULT);
 565	if (unlikely(r))
 566		return r;
 567
 568	if (op == REQ_OP_READ) {
 569		if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 570			r = sb_mac(ic, false);
 571			if (unlikely(r))
 572				return r;
 573		}
 574	}
 575
 576	return 0;
 577}
 578
 579#define BITMAP_OP_TEST_ALL_SET		0
 580#define BITMAP_OP_TEST_ALL_CLEAR	1
 581#define BITMAP_OP_SET			2
 582#define BITMAP_OP_CLEAR			3
 583
 584static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 585			    sector_t sector, sector_t n_sectors, int mode)
 586{
 587	unsigned long bit, end_bit, this_end_bit, page, end_page;
 588	unsigned long *data;
 589
 590	if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 591		DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 592			sector,
 593			n_sectors,
 594			ic->sb->log2_sectors_per_block,
 595			ic->log2_blocks_per_bitmap_bit,
 596			mode);
 597		BUG();
 598	}
 599
 600	if (unlikely(!n_sectors))
 601		return true;
 602
 603	bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 604	end_bit = (sector + n_sectors - 1) >>
 605		(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 606
 607	page = bit / (PAGE_SIZE * 8);
 608	bit %= PAGE_SIZE * 8;
 609
 610	end_page = end_bit / (PAGE_SIZE * 8);
 611	end_bit %= PAGE_SIZE * 8;
 612
 613repeat:
 614	if (page < end_page)
 615		this_end_bit = PAGE_SIZE * 8 - 1;
 616	else
 617		this_end_bit = end_bit;
 618
 619	data = lowmem_page_address(bitmap[page].page);
 620
 621	if (mode == BITMAP_OP_TEST_ALL_SET) {
 622		while (bit <= this_end_bit) {
 623			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 624				do {
 625					if (data[bit / BITS_PER_LONG] != -1)
 626						return false;
 627					bit += BITS_PER_LONG;
 628				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 629				continue;
 630			}
 631			if (!test_bit(bit, data))
 632				return false;
 633			bit++;
 634		}
 635	} else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 636		while (bit <= this_end_bit) {
 637			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 638				do {
 639					if (data[bit / BITS_PER_LONG] != 0)
 640						return false;
 641					bit += BITS_PER_LONG;
 642				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 643				continue;
 644			}
 645			if (test_bit(bit, data))
 646				return false;
 647			bit++;
 648		}
 649	} else if (mode == BITMAP_OP_SET) {
 650		while (bit <= this_end_bit) {
 651			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 652				do {
 653					data[bit / BITS_PER_LONG] = -1;
 654					bit += BITS_PER_LONG;
 655				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 656				continue;
 657			}
 658			__set_bit(bit, data);
 659			bit++;
 660		}
 661	} else if (mode == BITMAP_OP_CLEAR) {
 662		if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 663			clear_page(data);
 664		else {
 665			while (bit <= this_end_bit) {
 666				if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 667					do {
 668						data[bit / BITS_PER_LONG] = 0;
 669						bit += BITS_PER_LONG;
 670					} while (this_end_bit >= bit + BITS_PER_LONG - 1);
 671					continue;
 672				}
 673				__clear_bit(bit, data);
 674				bit++;
 675			}
 676		}
 677	} else {
 678		BUG();
 679	}
 680
 681	if (unlikely(page < end_page)) {
 682		bit = 0;
 683		page++;
 684		goto repeat;
 685	}
 686
 687	return true;
 688}
 689
 690static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 691{
 692	unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 693	unsigned int i;
 694
 695	for (i = 0; i < n_bitmap_pages; i++) {
 696		unsigned long *dst_data = lowmem_page_address(dst[i].page);
 697		unsigned long *src_data = lowmem_page_address(src[i].page);
 698
 699		copy_page(dst_data, src_data);
 700	}
 701}
 702
 703static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 704{
 705	unsigned int bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 706	unsigned int bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 707
 708	BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 709	return &ic->bbs[bitmap_block];
 710}
 711
 712static void access_journal_check(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 713				 bool e, const char *function)
 714{
 715#if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 716	unsigned int limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 717
 718	if (unlikely(section >= ic->journal_sections) ||
 719	    unlikely(offset >= limit)) {
 720		DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 721		       function, section, offset, ic->journal_sections, limit);
 722		BUG();
 723	}
 724#endif
 725}
 726
 727static void page_list_location(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 728			       unsigned int *pl_index, unsigned int *pl_offset)
 729{
 730	unsigned int sector;
 731
 732	access_journal_check(ic, section, offset, false, "page_list_location");
 733
 734	sector = section * ic->journal_section_sectors + offset;
 735
 736	*pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 737	*pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 738}
 739
 740static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 741					       unsigned int section, unsigned int offset, unsigned int *n_sectors)
 742{
 743	unsigned int pl_index, pl_offset;
 744	char *va;
 745
 746	page_list_location(ic, section, offset, &pl_index, &pl_offset);
 747
 748	if (n_sectors)
 749		*n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 750
 751	va = lowmem_page_address(pl[pl_index].page);
 752
 753	return (struct journal_sector *)(va + pl_offset);
 754}
 755
 756static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset)
 757{
 758	return access_page_list(ic, ic->journal, section, offset, NULL);
 759}
 760
 761static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 762{
 763	unsigned int rel_sector, offset;
 764	struct journal_sector *js;
 765
 766	access_journal_check(ic, section, n, true, "access_journal_entry");
 767
 768	rel_sector = n % JOURNAL_BLOCK_SECTORS;
 769	offset = n / JOURNAL_BLOCK_SECTORS;
 770
 771	js = access_journal(ic, section, rel_sector);
 772	return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 773}
 774
 775static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 776{
 777	n <<= ic->sb->log2_sectors_per_block;
 778
 779	n += JOURNAL_BLOCK_SECTORS;
 780
 781	access_journal_check(ic, section, n, false, "access_journal_data");
 782
 783	return access_journal(ic, section, n);
 784}
 785
 786static void section_mac(struct dm_integrity_c *ic, unsigned int section, __u8 result[JOURNAL_MAC_SIZE])
 787{
 788	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 789	int r;
 790	unsigned int j, size;
 791
 792	desc->tfm = ic->journal_mac;
 793
 794	r = crypto_shash_init(desc);
 795	if (unlikely(r < 0)) {
 796		dm_integrity_io_error(ic, "crypto_shash_init", r);
 797		goto err;
 798	}
 799
 800	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 801		__le64 section_le;
 802
 803		r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 804		if (unlikely(r < 0)) {
 805			dm_integrity_io_error(ic, "crypto_shash_update", r);
 806			goto err;
 807		}
 808
 809		section_le = cpu_to_le64(section);
 810		r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof(section_le));
 811		if (unlikely(r < 0)) {
 812			dm_integrity_io_error(ic, "crypto_shash_update", r);
 813			goto err;
 814		}
 815	}
 816
 817	for (j = 0; j < ic->journal_section_entries; j++) {
 818		struct journal_entry *je = access_journal_entry(ic, section, j);
 819
 820		r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof(je->u.sector));
 821		if (unlikely(r < 0)) {
 822			dm_integrity_io_error(ic, "crypto_shash_update", r);
 823			goto err;
 824		}
 825	}
 826
 827	size = crypto_shash_digestsize(ic->journal_mac);
 828
 829	if (likely(size <= JOURNAL_MAC_SIZE)) {
 830		r = crypto_shash_final(desc, result);
 831		if (unlikely(r < 0)) {
 832			dm_integrity_io_error(ic, "crypto_shash_final", r);
 833			goto err;
 834		}
 835		memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 836	} else {
 837		__u8 digest[HASH_MAX_DIGESTSIZE];
 838
 839		if (WARN_ON(size > sizeof(digest))) {
 840			dm_integrity_io_error(ic, "digest_size", -EINVAL);
 841			goto err;
 842		}
 843		r = crypto_shash_final(desc, digest);
 844		if (unlikely(r < 0)) {
 845			dm_integrity_io_error(ic, "crypto_shash_final", r);
 846			goto err;
 847		}
 848		memcpy(result, digest, JOURNAL_MAC_SIZE);
 849	}
 850
 851	return;
 852err:
 853	memset(result, 0, JOURNAL_MAC_SIZE);
 854}
 855
 856static void rw_section_mac(struct dm_integrity_c *ic, unsigned int section, bool wr)
 857{
 858	__u8 result[JOURNAL_MAC_SIZE];
 859	unsigned int j;
 860
 861	if (!ic->journal_mac)
 862		return;
 863
 864	section_mac(ic, section, result);
 865
 866	for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 867		struct journal_sector *js = access_journal(ic, section, j);
 868
 869		if (likely(wr))
 870			memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 871		else {
 872			if (crypto_memneq(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
 873				dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 874				dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
 875			}
 876		}
 877	}
 878}
 879
 880static void complete_journal_op(void *context)
 881{
 882	struct journal_completion *comp = context;
 883
 884	BUG_ON(!atomic_read(&comp->in_flight));
 885	if (likely(atomic_dec_and_test(&comp->in_flight)))
 886		complete(&comp->comp);
 887}
 888
 889static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 890			unsigned int n_sections, struct journal_completion *comp)
 891{
 892	struct async_submit_ctl submit;
 893	size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 894	unsigned int pl_index, pl_offset, section_index;
 895	struct page_list *source_pl, *target_pl;
 896
 897	if (likely(encrypt)) {
 898		source_pl = ic->journal;
 899		target_pl = ic->journal_io;
 900	} else {
 901		source_pl = ic->journal_io;
 902		target_pl = ic->journal;
 903	}
 904
 905	page_list_location(ic, section, 0, &pl_index, &pl_offset);
 906
 907	atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 908
 909	init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 910
 911	section_index = pl_index;
 912
 913	do {
 914		size_t this_step;
 915		struct page *src_pages[2];
 916		struct page *dst_page;
 917
 918		while (unlikely(pl_index == section_index)) {
 919			unsigned int dummy;
 920
 921			if (likely(encrypt))
 922				rw_section_mac(ic, section, true);
 923			section++;
 924			n_sections--;
 925			if (!n_sections)
 926				break;
 927			page_list_location(ic, section, 0, &section_index, &dummy);
 928		}
 929
 930		this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 931		dst_page = target_pl[pl_index].page;
 932		src_pages[0] = source_pl[pl_index].page;
 933		src_pages[1] = ic->journal_xor[pl_index].page;
 934
 935		async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 936
 937		pl_index++;
 938		pl_offset = 0;
 939		n_bytes -= this_step;
 940	} while (n_bytes);
 941
 942	BUG_ON(n_sections);
 943
 944	async_tx_issue_pending_all();
 945}
 946
 947static void complete_journal_encrypt(void *data, int err)
 948{
 949	struct journal_completion *comp = data;
 950
 951	if (unlikely(err)) {
 952		if (likely(err == -EINPROGRESS)) {
 953			complete(&comp->ic->crypto_backoff);
 954			return;
 955		}
 956		dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 957	}
 958	complete_journal_op(comp);
 959}
 960
 961static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 962{
 963	int r;
 964
 965	skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 966				      complete_journal_encrypt, comp);
 967	if (likely(encrypt))
 968		r = crypto_skcipher_encrypt(req);
 969	else
 970		r = crypto_skcipher_decrypt(req);
 971	if (likely(!r))
 972		return false;
 973	if (likely(r == -EINPROGRESS))
 974		return true;
 975	if (likely(r == -EBUSY)) {
 976		wait_for_completion(&comp->ic->crypto_backoff);
 977		reinit_completion(&comp->ic->crypto_backoff);
 978		return true;
 979	}
 980	dm_integrity_io_error(comp->ic, "encrypt", r);
 981	return false;
 982}
 983
 984static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 985			  unsigned int n_sections, struct journal_completion *comp)
 986{
 987	struct scatterlist **source_sg;
 988	struct scatterlist **target_sg;
 989
 990	atomic_add(2, &comp->in_flight);
 991
 992	if (likely(encrypt)) {
 993		source_sg = ic->journal_scatterlist;
 994		target_sg = ic->journal_io_scatterlist;
 995	} else {
 996		source_sg = ic->journal_io_scatterlist;
 997		target_sg = ic->journal_scatterlist;
 998	}
 999
1000	do {
1001		struct skcipher_request *req;
1002		unsigned int ivsize;
1003		char *iv;
1004
1005		if (likely(encrypt))
1006			rw_section_mac(ic, section, true);
1007
1008		req = ic->sk_requests[section];
1009		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1010		iv = req->iv;
1011
1012		memcpy(iv, iv + ivsize, ivsize);
1013
1014		req->src = source_sg[section];
1015		req->dst = target_sg[section];
1016
1017		if (unlikely(do_crypt(encrypt, req, comp)))
1018			atomic_inc(&comp->in_flight);
1019
1020		section++;
1021		n_sections--;
1022	} while (n_sections);
1023
1024	atomic_dec(&comp->in_flight);
1025	complete_journal_op(comp);
1026}
1027
1028static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
1029			    unsigned int n_sections, struct journal_completion *comp)
1030{
1031	if (ic->journal_xor)
1032		return xor_journal(ic, encrypt, section, n_sections, comp);
1033	else
1034		return crypt_journal(ic, encrypt, section, n_sections, comp);
1035}
1036
1037static void complete_journal_io(unsigned long error, void *context)
1038{
1039	struct journal_completion *comp = context;
1040
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, blk_opf_t opf,
1047			       unsigned int sector, unsigned int n_sectors,
1048			       struct journal_completion *comp)
1049{
1050	struct dm_io_request io_req;
1051	struct dm_io_region io_loc;
1052	unsigned int pl_index, pl_offset;
1053	int r;
1054
1055	if (unlikely(dm_integrity_failed(ic))) {
1056		if (comp)
1057			complete_journal_io(-1UL, comp);
1058		return;
1059	}
1060
1061	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1062	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1063
1064	io_req.bi_opf = opf;
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, IOPRIO_DEFAULT);
1083	if (unlikely(r)) {
1084		dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1085				      "reading journal" : "writing journal", r);
1086		if (comp) {
1087			WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1088			complete_journal_io(-1UL, comp);
1089		}
1090	}
1091}
1092
1093static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1094		       unsigned int section, unsigned int n_sections,
1095		       struct journal_completion *comp)
1096{
1097	unsigned int sector, n_sectors;
1098
1099	sector = section * ic->journal_section_sectors;
1100	n_sectors = n_sections * ic->journal_section_sectors;
1101
1102	rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1103}
1104
1105static void write_journal(struct dm_integrity_c *ic, unsigned int commit_start, unsigned int commit_sections)
1106{
1107	struct journal_completion io_comp;
1108	struct journal_completion crypt_comp_1;
1109	struct journal_completion crypt_comp_2;
1110	unsigned int i;
1111
1112	io_comp.ic = ic;
1113	init_completion(&io_comp.comp);
1114
1115	if (commit_start + commit_sections <= ic->journal_sections) {
1116		io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1117		if (ic->journal_io) {
1118			crypt_comp_1.ic = ic;
1119			init_completion(&crypt_comp_1.comp);
1120			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1121			encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1122			wait_for_completion_io(&crypt_comp_1.comp);
1123		} else {
1124			for (i = 0; i < commit_sections; i++)
1125				rw_section_mac(ic, commit_start + i, true);
1126		}
1127		rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1128			   commit_sections, &io_comp);
1129	} else {
1130		unsigned int to_end;
1131
1132		io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1133		to_end = ic->journal_sections - commit_start;
1134		if (ic->journal_io) {
1135			crypt_comp_1.ic = ic;
1136			init_completion(&crypt_comp_1.comp);
1137			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1138			encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1139			if (try_wait_for_completion(&crypt_comp_1.comp)) {
1140				rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1141					   commit_start, to_end, &io_comp);
1142				reinit_completion(&crypt_comp_1.comp);
1143				crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1144				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1145				wait_for_completion_io(&crypt_comp_1.comp);
1146			} else {
1147				crypt_comp_2.ic = ic;
1148				init_completion(&crypt_comp_2.comp);
1149				crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1150				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1151				wait_for_completion_io(&crypt_comp_1.comp);
1152				rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1153				wait_for_completion_io(&crypt_comp_2.comp);
1154			}
1155		} else {
1156			for (i = 0; i < to_end; i++)
1157				rw_section_mac(ic, commit_start + i, true);
1158			rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1159			for (i = 0; i < commit_sections - to_end; i++)
1160				rw_section_mac(ic, i, true);
1161		}
1162		rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1163	}
1164
1165	wait_for_completion_io(&io_comp.comp);
1166}
1167
1168static void copy_from_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
1169			      unsigned int n_sectors, sector_t target, io_notify_fn fn, void *data)
1170{
1171	struct dm_io_request io_req;
1172	struct dm_io_region io_loc;
1173	int r;
1174	unsigned int sector, pl_index, pl_offset;
1175
1176	BUG_ON((target | n_sectors | offset) & (unsigned int)(ic->sectors_per_block - 1));
1177
1178	if (unlikely(dm_integrity_failed(ic))) {
1179		fn(-1UL, data);
1180		return;
1181	}
1182
1183	sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1184
1185	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1186	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1187
1188	io_req.bi_opf = REQ_OP_WRITE;
1189	io_req.mem.type = DM_IO_PAGE_LIST;
1190	io_req.mem.ptr.pl = &ic->journal[pl_index];
1191	io_req.mem.offset = pl_offset;
1192	io_req.notify.fn = fn;
1193	io_req.notify.context = data;
1194	io_req.client = ic->io;
1195	io_loc.bdev = ic->dev->bdev;
1196	io_loc.sector = target;
1197	io_loc.count = n_sectors;
1198
1199	r = dm_io(&io_req, 1, &io_loc, NULL, IOPRIO_DEFAULT);
1200	if (unlikely(r)) {
1201		WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1202		fn(-1UL, data);
1203	}
1204}
1205
1206static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1207{
1208	return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1209	       range1->logical_sector + range1->n_sectors > range2->logical_sector;
1210}
1211
1212static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1213{
1214	struct rb_node **n = &ic->in_progress.rb_node;
1215	struct rb_node *parent;
1216
1217	BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned int)(ic->sectors_per_block - 1));
1218
1219	if (likely(check_waiting)) {
1220		struct dm_integrity_range *range;
1221
1222		list_for_each_entry(range, &ic->wait_list, wait_entry) {
1223			if (unlikely(ranges_overlap(range, new_range)))
1224				return false;
1225		}
1226	}
1227
1228	parent = NULL;
1229
1230	while (*n) {
1231		struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1232
1233		parent = *n;
1234		if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector)
1235			n = &range->node.rb_left;
1236		else if (new_range->logical_sector >= range->logical_sector + range->n_sectors)
1237			n = &range->node.rb_right;
1238		else
1239			return false;
1240	}
1241
1242	rb_link_node(&new_range->node, parent, n);
1243	rb_insert_color(&new_range->node, &ic->in_progress);
1244
1245	return true;
1246}
1247
1248static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1249{
1250	rb_erase(&range->node, &ic->in_progress);
1251	while (unlikely(!list_empty(&ic->wait_list))) {
1252		struct dm_integrity_range *last_range =
1253			list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1254		struct task_struct *last_range_task;
1255
1256		last_range_task = last_range->task;
1257		list_del(&last_range->wait_entry);
1258		if (!add_new_range(ic, last_range, false)) {
1259			last_range->task = last_range_task;
1260			list_add(&last_range->wait_entry, &ic->wait_list);
1261			break;
1262		}
1263		last_range->waiting = false;
1264		wake_up_process(last_range_task);
1265	}
1266}
1267
1268static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1269{
1270	unsigned long flags;
1271
1272	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1273	remove_range_unlocked(ic, range);
1274	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1275}
1276
1277static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1278{
1279	new_range->waiting = true;
1280	list_add_tail(&new_range->wait_entry, &ic->wait_list);
1281	new_range->task = current;
1282	do {
1283		__set_current_state(TASK_UNINTERRUPTIBLE);
1284		spin_unlock_irq(&ic->endio_wait.lock);
1285		io_schedule();
1286		spin_lock_irq(&ic->endio_wait.lock);
1287	} while (unlikely(new_range->waiting));
1288}
1289
1290static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1291{
1292	if (unlikely(!add_new_range(ic, new_range, true)))
1293		wait_and_add_new_range(ic, new_range);
1294}
1295
1296static void init_journal_node(struct journal_node *node)
1297{
1298	RB_CLEAR_NODE(&node->node);
1299	node->sector = (sector_t)-1;
1300}
1301
1302static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1303{
1304	struct rb_node **link;
1305	struct rb_node *parent;
1306
1307	node->sector = sector;
1308	BUG_ON(!RB_EMPTY_NODE(&node->node));
1309
1310	link = &ic->journal_tree_root.rb_node;
1311	parent = NULL;
1312
1313	while (*link) {
1314		struct journal_node *j;
1315
1316		parent = *link;
1317		j = container_of(parent, struct journal_node, node);
1318		if (sector < j->sector)
1319			link = &j->node.rb_left;
1320		else
1321			link = &j->node.rb_right;
1322	}
1323
1324	rb_link_node(&node->node, parent, link);
1325	rb_insert_color(&node->node, &ic->journal_tree_root);
1326}
1327
1328static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1329{
1330	BUG_ON(RB_EMPTY_NODE(&node->node));
1331	rb_erase(&node->node, &ic->journal_tree_root);
1332	init_journal_node(node);
1333}
1334
1335#define NOT_FOUND	(-1U)
1336
1337static unsigned int find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1338{
1339	struct rb_node *n = ic->journal_tree_root.rb_node;
1340	unsigned int found = NOT_FOUND;
1341
1342	*next_sector = (sector_t)-1;
1343	while (n) {
1344		struct journal_node *j = container_of(n, struct journal_node, node);
1345
1346		if (sector == j->sector)
1347			found = j - ic->journal_tree;
1348
1349		if (sector < j->sector) {
1350			*next_sector = j->sector;
1351			n = j->node.rb_left;
1352		} else
1353			n = j->node.rb_right;
1354	}
1355
1356	return found;
1357}
1358
1359static bool test_journal_node(struct dm_integrity_c *ic, unsigned int pos, sector_t sector)
1360{
1361	struct journal_node *node, *next_node;
1362	struct rb_node *next;
1363
1364	if (unlikely(pos >= ic->journal_entries))
1365		return false;
1366	node = &ic->journal_tree[pos];
1367	if (unlikely(RB_EMPTY_NODE(&node->node)))
1368		return false;
1369	if (unlikely(node->sector != sector))
1370		return false;
1371
1372	next = rb_next(&node->node);
1373	if (unlikely(!next))
1374		return true;
1375
1376	next_node = container_of(next, struct journal_node, node);
1377	return next_node->sector != sector;
1378}
1379
1380static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1381{
1382	struct rb_node *next;
1383	struct journal_node *next_node;
1384	unsigned int next_section;
1385
1386	BUG_ON(RB_EMPTY_NODE(&node->node));
1387
1388	next = rb_next(&node->node);
1389	if (unlikely(!next))
1390		return false;
1391
1392	next_node = container_of(next, struct journal_node, node);
1393
1394	if (next_node->sector != node->sector)
1395		return false;
1396
1397	next_section = (unsigned int)(next_node - ic->journal_tree) / ic->journal_section_entries;
1398	if (next_section >= ic->committed_section &&
1399	    next_section < ic->committed_section + ic->n_committed_sections)
1400		return true;
1401	if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1402		return true;
1403
1404	return false;
1405}
1406
1407#define TAG_READ	0
1408#define TAG_WRITE	1
1409#define TAG_CMP		2
1410
1411static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1412			       unsigned int *metadata_offset, unsigned int total_size, int op)
1413{
1414	unsigned int hash_offset = 0;
1415	unsigned char mismatch_hash = 0;
1416	unsigned char mismatch_filler = !ic->discard;
1417
1418	do {
1419		unsigned char *data, *dp;
1420		struct dm_buffer *b;
1421		unsigned int to_copy;
1422		int r;
1423
1424		r = dm_integrity_failed(ic);
1425		if (unlikely(r))
1426			return r;
1427
1428		data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1429		if (IS_ERR(data))
1430			return PTR_ERR(data);
1431
1432		to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1433		dp = data + *metadata_offset;
1434		if (op == TAG_READ) {
1435			memcpy(tag, dp, to_copy);
1436		} else if (op == TAG_WRITE) {
1437			if (crypto_memneq(dp, tag, to_copy)) {
1438				memcpy(dp, tag, to_copy);
1439				dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1440			}
1441		} else {
1442			/* e.g.: op == TAG_CMP */
1443
1444			if (likely(is_power_of_2(ic->tag_size))) {
1445				if (unlikely(crypto_memneq(dp, tag, to_copy)))
1446					goto thorough_test;
1447			} else {
1448				unsigned int i, ts;
1449thorough_test:
1450				ts = total_size;
1451
1452				for (i = 0; i < to_copy; i++, ts--) {
1453					/*
1454					 * Warning: the control flow must not be
1455					 * dependent on match/mismatch of
1456					 * individual bytes.
1457					 */
1458					mismatch_hash |= dp[i] ^ tag[i];
1459					mismatch_filler |= dp[i] ^ DISCARD_FILLER;
1460					hash_offset++;
1461					if (unlikely(hash_offset == ic->tag_size)) {
1462						if (unlikely(mismatch_hash) && unlikely(mismatch_filler)) {
1463							dm_bufio_release(b);
1464							return ts;
1465						}
1466						hash_offset = 0;
1467						mismatch_hash = 0;
1468						mismatch_filler = !ic->discard;
1469					}
1470				}
1471			}
1472		}
1473		dm_bufio_release(b);
1474
1475		tag += to_copy;
1476		*metadata_offset += to_copy;
1477		if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1478			(*metadata_block)++;
1479			*metadata_offset = 0;
1480		}
1481
1482		if (unlikely(!is_power_of_2(ic->tag_size)))
1483			hash_offset = (hash_offset + to_copy) % ic->tag_size;
1484
1485		total_size -= to_copy;
1486	} while (unlikely(total_size));
1487
1488	return 0;
1489}
1490
1491struct flush_request {
1492	struct dm_io_request io_req;
1493	struct dm_io_region io_reg;
1494	struct dm_integrity_c *ic;
1495	struct completion comp;
1496};
1497
1498static void flush_notify(unsigned long error, void *fr_)
1499{
1500	struct flush_request *fr = fr_;
1501
1502	if (unlikely(error != 0))
1503		dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1504	complete(&fr->comp);
1505}
1506
1507static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1508{
1509	int r;
1510	struct flush_request fr;
1511
1512	if (!ic->meta_dev)
1513		flush_data = false;
1514	if (flush_data) {
1515		fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1516		fr.io_req.mem.type = DM_IO_KMEM;
1517		fr.io_req.mem.ptr.addr = NULL;
1518		fr.io_req.notify.fn = flush_notify;
1519		fr.io_req.notify.context = &fr;
1520		fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio);
1521		fr.io_reg.bdev = ic->dev->bdev;
1522		fr.io_reg.sector = 0;
1523		fr.io_reg.count = 0;
1524		fr.ic = ic;
1525		init_completion(&fr.comp);
1526		r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL, IOPRIO_DEFAULT);
1527		BUG_ON(r);
1528	}
1529
1530	r = dm_bufio_write_dirty_buffers(ic->bufio);
1531	if (unlikely(r))
1532		dm_integrity_io_error(ic, "writing tags", r);
1533
1534	if (flush_data)
1535		wait_for_completion(&fr.comp);
1536}
1537
1538static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1539{
1540	DECLARE_WAITQUEUE(wait, current);
1541
1542	__add_wait_queue(&ic->endio_wait, &wait);
1543	__set_current_state(TASK_UNINTERRUPTIBLE);
1544	spin_unlock_irq(&ic->endio_wait.lock);
1545	io_schedule();
1546	spin_lock_irq(&ic->endio_wait.lock);
1547	__remove_wait_queue(&ic->endio_wait, &wait);
1548}
1549
1550static void autocommit_fn(struct timer_list *t)
1551{
1552	struct dm_integrity_c *ic = timer_container_of(ic, t,
1553						       autocommit_timer);
1554
1555	if (likely(!dm_integrity_failed(ic)))
1556		queue_work(ic->commit_wq, &ic->commit_work);
1557}
1558
1559static void schedule_autocommit(struct dm_integrity_c *ic)
1560{
1561	if (!timer_pending(&ic->autocommit_timer))
1562		mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1563}
1564
1565static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1566{
1567	struct bio *bio;
1568	unsigned long flags;
1569
1570	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1571	bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1572	bio_list_add(&ic->flush_bio_list, bio);
1573	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1574
1575	queue_work(ic->commit_wq, &ic->commit_work);
1576}
1577
1578static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1579{
1580	int r;
1581
1582	r = dm_integrity_failed(ic);
1583	if (unlikely(r) && !bio->bi_status)
1584		bio->bi_status = errno_to_blk_status(r);
1585	if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1586		unsigned long flags;
1587
1588		spin_lock_irqsave(&ic->endio_wait.lock, flags);
1589		bio_list_add(&ic->synchronous_bios, bio);
1590		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1591		spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1592		return;
1593	}
1594	bio_endio(bio);
1595}
1596
1597static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1598{
1599	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1600
1601	if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1602		submit_flush_bio(ic, dio);
1603	else
1604		do_endio(ic, bio);
1605}
1606
1607static void dec_in_flight(struct dm_integrity_io *dio)
1608{
1609	if (atomic_dec_and_test(&dio->in_flight)) {
1610		struct dm_integrity_c *ic = dio->ic;
1611		struct bio *bio;
1612
1613		remove_range(ic, &dio->range);
1614
1615		if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1616			schedule_autocommit(ic);
1617
1618		bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1619		if (unlikely(dio->bi_status) && !bio->bi_status)
1620			bio->bi_status = dio->bi_status;
1621		if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1622			dio->range.logical_sector += dio->range.n_sectors;
1623			bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1624			INIT_WORK(&dio->work, integrity_bio_wait);
1625			queue_work(ic->offload_wq, &dio->work);
1626			return;
1627		}
1628		do_endio_flush(ic, dio);
1629	}
1630}
1631
1632static void integrity_end_io(struct bio *bio)
1633{
1634	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1635
1636	dm_bio_restore(&dio->bio_details, bio);
1637	if (bio->bi_integrity)
1638		bio->bi_opf |= REQ_INTEGRITY;
1639
1640	if (dio->completion)
1641		complete(dio->completion);
1642
1643	dec_in_flight(dio);
1644}
1645
1646static void integrity_sector_checksum_shash(struct dm_integrity_c *ic, sector_t sector,
1647					    const char *data, unsigned offset, char *result)
1648{
1649	__le64 sector_le = cpu_to_le64(sector);
1650	SHASH_DESC_ON_STACK(req, ic->internal_shash);
1651	int r;
1652	unsigned int digest_size;
1653
1654	req->tfm = ic->internal_shash;
1655
1656	r = crypto_shash_init(req);
1657	if (unlikely(r < 0)) {
1658		dm_integrity_io_error(ic, "crypto_shash_init", r);
1659		goto failed;
1660	}
1661
1662	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1663		r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1664		if (unlikely(r < 0)) {
1665			dm_integrity_io_error(ic, "crypto_shash_update", r);
1666			goto failed;
1667		}
1668	}
1669
1670	r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof(sector_le));
1671	if (unlikely(r < 0)) {
1672		dm_integrity_io_error(ic, "crypto_shash_update", r);
1673		goto failed;
1674	}
1675
1676	r = crypto_shash_update(req, data + offset, ic->sectors_per_block << SECTOR_SHIFT);
1677	if (unlikely(r < 0)) {
1678		dm_integrity_io_error(ic, "crypto_shash_update", r);
1679		goto failed;
1680	}
1681
1682	r = crypto_shash_final(req, result);
1683	if (unlikely(r < 0)) {
1684		dm_integrity_io_error(ic, "crypto_shash_final", r);
1685		goto failed;
1686	}
1687
1688	digest_size = ic->internal_hash_digestsize;
1689	if (unlikely(digest_size < ic->tag_size))
1690		memset(result + digest_size, 0, ic->tag_size - digest_size);
1691
1692	return;
1693
1694failed:
1695	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1696	get_random_bytes(result, ic->tag_size);
1697}
1698
1699static void integrity_sector_checksum_ahash(struct dm_integrity_c *ic, struct ahash_request **ahash_req,
1700					    sector_t sector, struct page *page, unsigned offset, char *result)
1701{
1702	__le64 sector_le = cpu_to_le64(sector);
1703	struct ahash_request *req;
1704	DECLARE_CRYPTO_WAIT(wait);
1705	struct scatterlist sg[3], *s = sg;
1706	int r;
1707	unsigned int digest_size;
1708	unsigned int nbytes = 0;
1709
1710	might_sleep();
1711
1712	req = *ahash_req;
1713	if (unlikely(!req)) {
1714		req = mempool_alloc(&ic->ahash_req_pool, GFP_NOIO);
1715		*ahash_req = req;
1716	}
1717
1718	ahash_request_set_tfm(req, ic->internal_ahash);
1719	ahash_request_set_callback(req, CRYPTO_TFM_REQ_MAY_SLEEP, crypto_req_done, &wait);
1720
1721	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1722		sg_init_table(sg, 3);
1723		sg_set_buf(s, (const __u8 *)&ic->sb->salt, SALT_SIZE);
1724		nbytes += SALT_SIZE;
1725		s++;
1726	} else {
1727		sg_init_table(sg, 2);
1728	}
1729
1730	if (likely(!is_vmalloc_addr(&sector_le))) {
1731		sg_set_buf(s, &sector_le, sizeof(sector_le));
1732	} else {
1733		struct page *sec_page = vmalloc_to_page(&sector_le);
1734		unsigned int sec_off = offset_in_page(&sector_le);
1735		sg_set_page(s, sec_page, sizeof(sector_le), sec_off);
1736	}
1737	nbytes += sizeof(sector_le);
1738	s++;
1739
1740	sg_set_page(s, page, ic->sectors_per_block << SECTOR_SHIFT, offset);
1741	nbytes += ic->sectors_per_block << SECTOR_SHIFT;
1742
1743	ahash_request_set_crypt(req, sg, result, nbytes);
1744
1745	r = crypto_wait_req(crypto_ahash_digest(req), &wait);
1746	if (unlikely(r)) {
1747		dm_integrity_io_error(ic, "crypto_ahash_digest", r);
1748		goto failed;
1749	}
1750
1751	digest_size = ic->internal_hash_digestsize;
1752	if (unlikely(digest_size < ic->tag_size))
1753		memset(result + digest_size, 0, ic->tag_size - digest_size);
1754
1755	return;
1756
1757failed:
1758	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1759	get_random_bytes(result, ic->tag_size);
1760}
1761
1762static void integrity_sector_checksum(struct dm_integrity_c *ic, struct ahash_request **ahash_req,
1763				      sector_t sector, const char *data, unsigned offset, char *result)
1764{
1765	if (likely(ic->internal_shash != NULL))
1766		integrity_sector_checksum_shash(ic, sector, data, offset, result);
1767	else
1768		integrity_sector_checksum_ahash(ic, ahash_req, sector, (struct page *)data, offset, result);
1769}
1770
1771static void *integrity_kmap(struct dm_integrity_c *ic, struct page *p)
1772{
1773	if (likely(ic->internal_shash != NULL))
1774		return kmap_local_page(p);
1775	else
1776		return p;
1777}
1778
1779static void integrity_kunmap(struct dm_integrity_c *ic, const void *ptr)
1780{
1781	if (likely(ic->internal_shash != NULL))
1782		kunmap_local(ptr);
1783}
1784
1785static void *integrity_identity(struct dm_integrity_c *ic, void *data)
1786{
1787#ifdef CONFIG_DEBUG_SG
1788	BUG_ON(offset_in_page(data));
1789	BUG_ON(!virt_addr_valid(data));
1790#endif
1791	if (likely(ic->internal_shash != NULL))
1792		return data;
1793	else
1794		return virt_to_page(data);
1795}
1796
1797static noinline void integrity_recheck(struct dm_integrity_io *dio, char *checksum)
1798{
1799	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1800	struct dm_integrity_c *ic = dio->ic;
1801	struct bvec_iter iter;
1802	struct bio_vec bv;
1803	sector_t sector, logical_sector, area, offset;
1804	struct page *page;
1805
1806	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1807	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset,
1808							     &dio->metadata_offset);
1809	sector = get_data_sector(ic, area, offset);
1810	logical_sector = dio->range.logical_sector;
1811
1812	page = mempool_alloc(&ic->recheck_pool, GFP_NOIO);
1813
1814	__bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1815		unsigned pos = 0;
1816
1817		do {
1818			sector_t alignment;
1819			char *mem;
1820			char *buffer = page_to_virt(page);
1821			unsigned int buffer_offset;
1822			int r;
1823			struct dm_io_request io_req;
1824			struct dm_io_region io_loc;
1825			io_req.bi_opf = REQ_OP_READ;
1826			io_req.mem.type = DM_IO_KMEM;
1827			io_req.mem.ptr.addr = buffer;
1828			io_req.notify.fn = NULL;
1829			io_req.client = ic->io;
1830			io_loc.bdev = ic->dev->bdev;
1831			io_loc.sector = sector;
1832			io_loc.count = ic->sectors_per_block;
1833
1834			/* Align the bio to logical block size */
1835			alignment = dio->range.logical_sector | bio_sectors(bio) | (PAGE_SIZE >> SECTOR_SHIFT);
1836			alignment &= -alignment;
1837			io_loc.sector = round_down(io_loc.sector, alignment);
1838			io_loc.count += sector - io_loc.sector;
1839			buffer_offset = (sector - io_loc.sector) << SECTOR_SHIFT;
1840			io_loc.count = round_up(io_loc.count, alignment);
1841
1842			r = dm_io(&io_req, 1, &io_loc, NULL, IOPRIO_DEFAULT);
1843			if (unlikely(r)) {
1844				dio->bi_status = errno_to_blk_status(r);
1845				goto free_ret;
1846			}
1847
1848			integrity_sector_checksum(ic, &dio->ahash_req, logical_sector, integrity_identity(ic, buffer), buffer_offset, checksum);
1849			r = dm_integrity_rw_tag(ic, checksum, &dio->metadata_block,
1850						&dio->metadata_offset, ic->tag_size, TAG_CMP);
1851			if (r) {
1852				if (r > 0) {
1853					DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1854						    bio->bi_bdev, logical_sector);
1855					atomic64_inc(&ic->number_of_mismatches);
1856					dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1857							 bio, logical_sector, 0);
1858					r = -EILSEQ;
1859				}
1860				dio->bi_status = errno_to_blk_status(r);
1861				goto free_ret;
1862			}
1863
1864			mem = bvec_kmap_local(&bv);
1865			memcpy(mem + pos, buffer + buffer_offset, ic->sectors_per_block << SECTOR_SHIFT);
1866			kunmap_local(mem);
1867
1868			pos += ic->sectors_per_block << SECTOR_SHIFT;
1869			sector += ic->sectors_per_block;
1870			logical_sector += ic->sectors_per_block;
1871		} while (pos < bv.bv_len);
1872	}
1873free_ret:
1874	mempool_free(page, &ic->recheck_pool);
1875}
1876
1877static void integrity_metadata(struct work_struct *w)
1878{
1879	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1880	struct dm_integrity_c *ic = dio->ic;
1881
1882	int r;
1883
1884	if (ic->internal_hash) {
1885		struct bvec_iter iter;
1886		struct bio_vec bv;
1887		unsigned int digest_size = ic->internal_hash_digestsize;
1888		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1889		char *checksums;
1890		unsigned int extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1891		char checksums_onstack[MAX_T(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1892		sector_t sector;
1893		unsigned int sectors_to_process;
1894
1895		if (unlikely(ic->mode == 'R'))
1896			goto skip_io;
1897
1898		if (likely(dio->op != REQ_OP_DISCARD))
1899			checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1900					    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1901		else
1902			checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1903		if (!checksums) {
1904			checksums = checksums_onstack;
1905			if (WARN_ON(extra_space &&
1906				    digest_size > sizeof(checksums_onstack))) {
1907				r = -EINVAL;
1908				goto error;
1909			}
1910		}
1911
1912		if (unlikely(dio->op == REQ_OP_DISCARD)) {
1913			unsigned int bi_size = dio->bio_details.bi_iter.bi_size;
1914			unsigned int max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1915			unsigned int max_blocks = max_size / ic->tag_size;
1916
1917			memset(checksums, DISCARD_FILLER, max_size);
1918
1919			while (bi_size) {
1920				unsigned int this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1921
1922				this_step_blocks = min(this_step_blocks, max_blocks);
1923				r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1924							this_step_blocks * ic->tag_size, TAG_WRITE);
1925				if (unlikely(r)) {
1926					if (likely(checksums != checksums_onstack))
1927						kfree(checksums);
1928					goto error;
1929				}
1930
1931				bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1932			}
1933
1934			if (likely(checksums != checksums_onstack))
1935				kfree(checksums);
1936			goto skip_io;
1937		}
1938
1939		sector = dio->range.logical_sector;
1940		sectors_to_process = dio->range.n_sectors;
1941
1942		__bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1943			struct bio_vec bv_copy = bv;
1944			unsigned int pos;
1945			char *mem, *checksums_ptr;
1946
1947again:
1948			mem = integrity_kmap(ic, bv_copy.bv_page);
1949			pos = 0;
1950			checksums_ptr = checksums;
1951			do {
1952				integrity_sector_checksum(ic, &dio->ahash_req, sector, mem, bv_copy.bv_offset + pos, checksums_ptr);
1953				checksums_ptr += ic->tag_size;
1954				sectors_to_process -= ic->sectors_per_block;
1955				pos += ic->sectors_per_block << SECTOR_SHIFT;
1956				sector += ic->sectors_per_block;
1957			} while (pos < bv_copy.bv_len && sectors_to_process && checksums != checksums_onstack);
1958			integrity_kunmap(ic, mem);
1959
1960			r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1961						checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1962			if (unlikely(r)) {
1963				if (likely(checksums != checksums_onstack))
1964					kfree(checksums);
1965				if (r > 0) {
1966					integrity_recheck(dio, checksums_onstack);
1967					goto skip_io;
1968				}
1969				goto error;
1970			}
1971
1972			if (!sectors_to_process)
1973				break;
1974
1975			if (unlikely(pos < bv_copy.bv_len)) {
1976				bv_copy.bv_offset += pos;
1977				bv_copy.bv_len -= pos;
1978				goto again;
1979			}
1980		}
1981
1982		if (likely(checksums != checksums_onstack))
1983			kfree(checksums);
1984	} else {
1985		struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1986
1987		if (bip) {
1988			struct bio_vec biv;
1989			struct bvec_iter iter;
1990			unsigned int data_to_process = dio->range.n_sectors;
1991
1992			sector_to_block(ic, data_to_process);
1993			data_to_process *= ic->tag_size;
1994
1995			bip_for_each_vec(biv, bip, iter) {
1996				unsigned char *tag;
1997				unsigned int this_len;
1998
1999				BUG_ON(PageHighMem(biv.bv_page));
2000				tag = bvec_virt(&biv);
2001				this_len = min(biv.bv_len, data_to_process);
2002				r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
2003							this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
2004				if (unlikely(r))
2005					goto error;
2006				data_to_process -= this_len;
2007				if (!data_to_process)
2008					break;
2009			}
2010		}
2011	}
2012skip_io:
2013	dec_in_flight(dio);
2014	return;
2015error:
2016	dio->bi_status = errno_to_blk_status(r);
2017	dec_in_flight(dio);
2018}
2019
2020static inline bool dm_integrity_check_limits(struct dm_integrity_c *ic, sector_t logical_sector, struct bio *bio)
2021{
2022	if (unlikely(logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
2023		DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
2024		      logical_sector, bio_sectors(bio),
2025		      ic->provided_data_sectors);
2026		return false;
2027	}
2028	if (unlikely((logical_sector | bio_sectors(bio)) & (unsigned int)(ic->sectors_per_block - 1))) {
2029		DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
2030		      ic->sectors_per_block,
2031		      logical_sector, bio_sectors(bio));
2032		return false;
2033	}
2034	if (ic->sectors_per_block > 1 && likely(bio_op(bio) != REQ_OP_DISCARD)) {
2035		struct bvec_iter iter;
2036		struct bio_vec bv;
2037
2038		bio_for_each_segment(bv, bio, iter) {
2039			if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
2040				DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
2041					bv.bv_offset, bv.bv_len, ic->sectors_per_block);
2042				return false;
2043			}
2044		}
2045	}
2046	return true;
2047}
2048
2049static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
2050{
2051	struct dm_integrity_c *ic = ti->private;
2052	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2053	struct bio_integrity_payload *bip;
2054
2055	sector_t area, offset;
2056
2057	dio->ic = ic;
2058	dio->bi_status = 0;
2059	dio->op = bio_op(bio);
2060	dio->ahash_req = NULL;
2061
2062	if (ic->mode == 'I') {
2063		bio->bi_iter.bi_sector = dm_target_offset(ic->ti, bio->bi_iter.bi_sector);
2064		dio->integrity_payload = NULL;
2065		dio->integrity_payload_from_mempool = false;
2066		dio->integrity_range_locked = false;
2067		return dm_integrity_map_inline(dio, true);
2068	}
2069
2070	if (unlikely(dio->op == REQ_OP_DISCARD)) {
2071		if (ti->max_io_len) {
2072			sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
2073			unsigned int log2_max_io_len = __fls(ti->max_io_len);
2074			sector_t start_boundary = sec >> log2_max_io_len;
2075			sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
2076
2077			if (start_boundary < end_boundary) {
2078				sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
2079
2080				dm_accept_partial_bio(bio, len);
2081			}
2082		}
2083	}
2084
2085	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
2086		submit_flush_bio(ic, dio);
2087		return DM_MAPIO_SUBMITTED;
2088	}
2089
2090	dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
2091	dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
2092	if (unlikely(dio->fua)) {
2093		/*
2094		 * Don't pass down the FUA flag because we have to flush
2095		 * disk cache anyway.
2096		 */
2097		bio->bi_opf &= ~REQ_FUA;
2098	}
2099	if (unlikely(!dm_integrity_check_limits(ic, dio->range.logical_sector, bio)))
2100		return DM_MAPIO_KILL;
2101
2102	bip = bio_integrity(bio);
2103	if (!ic->internal_hash) {
2104		if (bip) {
2105			unsigned int wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
2106
2107			if (ic->log2_tag_size >= 0)
2108				wanted_tag_size <<= ic->log2_tag_size;
2109			else
2110				wanted_tag_size *= ic->tag_size;
2111			if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
2112				DMERR("Invalid integrity data size %u, expected %u",
2113				      bip->bip_iter.bi_size, wanted_tag_size);
2114				return DM_MAPIO_KILL;
2115			}
2116		}
2117	} else {
2118		if (unlikely(bip != NULL)) {
2119			DMERR("Unexpected integrity data when using internal hash");
2120			return DM_MAPIO_KILL;
2121		}
2122	}
2123
2124	if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
2125		return DM_MAPIO_KILL;
2126
2127	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2128	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2129	bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
2130
2131	dm_integrity_map_continue(dio, true);
2132	return DM_MAPIO_SUBMITTED;
2133}
2134
2135static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
2136				 unsigned int journal_section, unsigned int journal_entry)
2137{
2138	struct dm_integrity_c *ic = dio->ic;
2139	sector_t logical_sector;
2140	unsigned int n_sectors;
2141
2142	logical_sector = dio->range.logical_sector;
2143	n_sectors = dio->range.n_sectors;
2144	do {
2145		struct bio_vec bv = bio_iovec(bio);
2146		char *mem;
2147
2148		if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
2149			bv.bv_len = n_sectors << SECTOR_SHIFT;
2150		n_sectors -= bv.bv_len >> SECTOR_SHIFT;
2151		bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
2152retry_kmap:
2153		mem = kmap_local_page(bv.bv_page);
2154		if (likely(dio->op == REQ_OP_WRITE))
2155			flush_dcache_page(bv.bv_page);
2156
2157		do {
2158			struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
2159
2160			if (unlikely(dio->op == REQ_OP_READ)) {
2161				struct journal_sector *js;
2162				char *mem_ptr;
2163				unsigned int s;
2164
2165				if (unlikely(journal_entry_is_inprogress(je))) {
2166					flush_dcache_page(bv.bv_page);
2167					kunmap_local(mem);
2168
2169					__io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2170					goto retry_kmap;
2171				}
2172				smp_rmb();
2173				BUG_ON(journal_entry_get_sector(je) != logical_sector);
2174				js = access_journal_data(ic, journal_section, journal_entry);
2175				mem_ptr = mem + bv.bv_offset;
2176				s = 0;
2177				do {
2178					memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
2179					*(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
2180					js++;
2181					mem_ptr += 1 << SECTOR_SHIFT;
2182				} while (++s < ic->sectors_per_block);
2183			}
2184
2185			if (!ic->internal_hash) {
2186				struct bio_integrity_payload *bip = bio_integrity(bio);
2187				unsigned int tag_todo = ic->tag_size;
2188				char *tag_ptr = journal_entry_tag(ic, je);
2189
2190				if (bip) {
2191					do {
2192						struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2193						unsigned int tag_now = min(biv.bv_len, tag_todo);
2194						char *tag_addr;
2195
2196						BUG_ON(PageHighMem(biv.bv_page));
2197						tag_addr = bvec_virt(&biv);
2198						if (likely(dio->op == REQ_OP_WRITE))
2199							memcpy(tag_ptr, tag_addr, tag_now);
2200						else
2201							memcpy(tag_addr, tag_ptr, tag_now);
2202						bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2203						tag_ptr += tag_now;
2204						tag_todo -= tag_now;
2205					} while (unlikely(tag_todo));
2206				} else if (likely(dio->op == REQ_OP_WRITE))
2207					memset(tag_ptr, 0, tag_todo);
2208			}
2209
2210			if (likely(dio->op == REQ_OP_WRITE)) {
2211				struct journal_sector *js;
2212				unsigned int s;
2213
2214				js = access_journal_data(ic, journal_section, journal_entry);
2215				memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2216
2217				s = 0;
2218				do {
2219					je->last_bytes[s] = js[s].commit_id;
2220				} while (++s < ic->sectors_per_block);
2221
2222				if (ic->internal_hash) {
2223					unsigned int digest_size = ic->internal_hash_digestsize;
2224					void *js_page = integrity_identity(ic, (char *)js - offset_in_page(js));
2225					unsigned js_offset = offset_in_page(js);
2226
2227					if (unlikely(digest_size > ic->tag_size)) {
2228						char checksums_onstack[HASH_MAX_DIGESTSIZE];
2229
2230						integrity_sector_checksum(ic, &dio->ahash_req, logical_sector, js_page, js_offset, checksums_onstack);
2231						memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2232					} else
2233						integrity_sector_checksum(ic, &dio->ahash_req, logical_sector, js_page, js_offset, journal_entry_tag(ic, je));
2234				}
2235
2236				journal_entry_set_sector(je, logical_sector);
2237			}
2238			logical_sector += ic->sectors_per_block;
2239
2240			journal_entry++;
2241			if (unlikely(journal_entry == ic->journal_section_entries)) {
2242				journal_entry = 0;
2243				journal_section++;
2244				wraparound_section(ic, &journal_section);
2245			}
2246
2247			bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2248		} while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2249
2250		if (unlikely(dio->op == REQ_OP_READ))
2251			flush_dcache_page(bv.bv_page);
2252		kunmap_local(mem);
2253	} while (n_sectors);
2254
2255	if (likely(dio->op == REQ_OP_WRITE)) {
2256		smp_mb();
2257		if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2258			wake_up(&ic->copy_to_journal_wait);
2259		if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2260			queue_work(ic->commit_wq, &ic->commit_work);
2261		else
2262			schedule_autocommit(ic);
2263	} else
2264		remove_range(ic, &dio->range);
2265
2266	if (unlikely(bio->bi_iter.bi_size)) {
2267		sector_t area, offset;
2268
2269		dio->range.logical_sector = logical_sector;
2270		get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2271		dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2272		return true;
2273	}
2274
2275	return false;
2276}
2277
2278static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2279{
2280	struct dm_integrity_c *ic = dio->ic;
2281	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2282	unsigned int journal_section, journal_entry;
2283	unsigned int journal_read_pos;
2284	sector_t recalc_sector;
2285	struct completion read_comp;
2286	bool discard_retried = false;
2287	bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2288
2289	if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2290		need_sync_io = true;
2291
2292	if (need_sync_io && from_map) {
2293		INIT_WORK(&dio->work, integrity_bio_wait);
2294		queue_work(ic->offload_wq, &dio->work);
2295		return;
2296	}
2297
2298lock_retry:
2299	spin_lock_irq(&ic->endio_wait.lock);
2300retry:
2301	if (unlikely(dm_integrity_failed(ic))) {
2302		spin_unlock_irq(&ic->endio_wait.lock);
2303		do_endio(ic, bio);
2304		return;
2305	}
2306	dio->range.n_sectors = bio_sectors(bio);
2307	journal_read_pos = NOT_FOUND;
2308	if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2309		if (dio->op == REQ_OP_WRITE) {
2310			unsigned int next_entry, i, pos;
2311			unsigned int ws, we, range_sectors;
2312
2313			dio->range.n_sectors = min(dio->range.n_sectors,
2314						   (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2315			if (unlikely(!dio->range.n_sectors)) {
2316				if (from_map)
2317					goto offload_to_thread;
2318				sleep_on_endio_wait(ic);
2319				goto retry;
2320			}
2321			range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2322			ic->free_sectors -= range_sectors;
2323			journal_section = ic->free_section;
2324			journal_entry = ic->free_section_entry;
2325
2326			next_entry = ic->free_section_entry + range_sectors;
2327			ic->free_section_entry = next_entry % ic->journal_section_entries;
2328			ic->free_section += next_entry / ic->journal_section_entries;
2329			ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2330			wraparound_section(ic, &ic->free_section);
2331
2332			pos = journal_section * ic->journal_section_entries + journal_entry;
2333			ws = journal_section;
2334			we = journal_entry;
2335			i = 0;
2336			do {
2337				struct journal_entry *je;
2338
2339				add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2340				pos++;
2341				if (unlikely(pos >= ic->journal_entries))
2342					pos = 0;
2343
2344				je = access_journal_entry(ic, ws, we);
2345				BUG_ON(!journal_entry_is_unused(je));
2346				journal_entry_set_inprogress(je);
2347				we++;
2348				if (unlikely(we == ic->journal_section_entries)) {
2349					we = 0;
2350					ws++;
2351					wraparound_section(ic, &ws);
2352				}
2353			} while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2354
2355			spin_unlock_irq(&ic->endio_wait.lock);
2356			goto journal_read_write;
2357		} else {
2358			sector_t next_sector;
2359
2360			journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2361			if (likely(journal_read_pos == NOT_FOUND)) {
2362				if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2363					dio->range.n_sectors = next_sector - dio->range.logical_sector;
2364			} else {
2365				unsigned int i;
2366				unsigned int jp = journal_read_pos + 1;
2367
2368				for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2369					if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2370						break;
2371				}
2372				dio->range.n_sectors = i;
2373			}
2374		}
2375	}
2376	if (unlikely(!add_new_range(ic, &dio->range, true))) {
2377		/*
2378		 * We must not sleep in the request routine because it could
2379		 * stall bios on current->bio_list.
2380		 * So, we offload the bio to a workqueue if we have to sleep.
2381		 */
2382		if (from_map) {
2383offload_to_thread:
2384			spin_unlock_irq(&ic->endio_wait.lock);
2385			INIT_WORK(&dio->work, integrity_bio_wait);
2386			queue_work(ic->wait_wq, &dio->work);
2387			return;
2388		}
2389		if (journal_read_pos != NOT_FOUND)
2390			dio->range.n_sectors = ic->sectors_per_block;
2391		wait_and_add_new_range(ic, &dio->range);
2392		/*
2393		 * wait_and_add_new_range drops the spinlock, so the journal
2394		 * may have been changed arbitrarily. We need to recheck.
2395		 * To simplify the code, we restrict I/O size to just one block.
2396		 */
2397		if (journal_read_pos != NOT_FOUND) {
2398			sector_t next_sector;
2399			unsigned int new_pos;
2400
2401			new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2402			if (unlikely(new_pos != journal_read_pos)) {
2403				remove_range_unlocked(ic, &dio->range);
2404				goto retry;
2405			}
2406		}
2407	}
2408	if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2409		sector_t next_sector;
2410		unsigned int new_pos;
2411
2412		new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2413		if (unlikely(new_pos != NOT_FOUND) ||
2414		    unlikely(next_sector < dio->range.logical_sector + dio->range.n_sectors)) {
2415			remove_range_unlocked(ic, &dio->range);
2416			spin_unlock_irq(&ic->endio_wait.lock);
2417			queue_work(ic->commit_wq, &ic->commit_work);
2418			flush_workqueue(ic->commit_wq);
2419			queue_work(ic->writer_wq, &ic->writer_work);
2420			flush_workqueue(ic->writer_wq);
2421			discard_retried = true;
2422			goto lock_retry;
2423		}
2424	}
2425	recalc_sector = le64_to_cpu(ic->sb->recalc_sector);
2426	spin_unlock_irq(&ic->endio_wait.lock);
2427
2428	if (unlikely(journal_read_pos != NOT_FOUND)) {
2429		journal_section = journal_read_pos / ic->journal_section_entries;
2430		journal_entry = journal_read_pos % ic->journal_section_entries;
2431		goto journal_read_write;
2432	}
2433
2434	if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2435		if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2436				     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2437			struct bitmap_block_status *bbs;
2438
2439			bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2440			spin_lock(&bbs->bio_queue_lock);
2441			bio_list_add(&bbs->bio_queue, bio);
2442			spin_unlock(&bbs->bio_queue_lock);
2443			queue_work(ic->writer_wq, &bbs->work);
2444			return;
2445		}
2446	}
2447
2448	dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2449
2450	if (need_sync_io) {
2451		init_completion(&read_comp);
2452		dio->completion = &read_comp;
2453	} else
2454		dio->completion = NULL;
2455
2456	dm_bio_record(&dio->bio_details, bio);
2457	bio_set_dev(bio, ic->dev->bdev);
2458	bio->bi_integrity = NULL;
2459	bio->bi_opf &= ~REQ_INTEGRITY;
2460	bio->bi_end_io = integrity_end_io;
2461	bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2462
2463	if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2464		integrity_metadata(&dio->work);
2465		dm_integrity_flush_buffers(ic, false);
2466
2467		dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2468		dio->completion = NULL;
2469
2470		submit_bio_noacct(bio);
2471
2472		return;
2473	}
2474
2475	submit_bio_noacct(bio);
2476
2477	if (need_sync_io) {
2478		wait_for_completion_io(&read_comp);
2479		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2480		    dio->range.logical_sector + dio->range.n_sectors > recalc_sector)
2481			goto skip_check;
2482		if (ic->mode == 'B') {
2483			if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2484					     dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2485				goto skip_check;
2486		}
2487
2488		if (likely(!bio->bi_status))
2489			integrity_metadata(&dio->work);
2490		else
2491skip_check:
2492			dec_in_flight(dio);
2493	} else {
2494		INIT_WORK(&dio->work, integrity_metadata);
2495		queue_work(ic->metadata_wq, &dio->work);
2496	}
2497
2498	return;
2499
2500journal_read_write:
2501	if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2502		goto lock_retry;
2503
2504	do_endio_flush(ic, dio);
2505}
2506
2507static int dm_integrity_map_inline(struct dm_integrity_io *dio, bool from_map)
2508{
2509	struct dm_integrity_c *ic = dio->ic;
2510	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2511	struct bio_integrity_payload *bip;
2512	unsigned ret;
2513	sector_t recalc_sector;
2514
2515	if (unlikely(bio_integrity(bio))) {
2516		bio->bi_status = BLK_STS_NOTSUPP;
2517		bio_endio(bio);
2518		return DM_MAPIO_SUBMITTED;
2519	}
2520
2521	bio_set_dev(bio, ic->dev->bdev);
2522	if (unlikely((bio->bi_opf & REQ_PREFLUSH) != 0))
2523		return DM_MAPIO_REMAPPED;
2524
2525retry:
2526	if (!dio->integrity_payload) {
2527		unsigned digest_size, extra_size;
2528		dio->payload_len = ic->tuple_size * (bio_sectors(bio) >> ic->sb->log2_sectors_per_block);
2529		digest_size = ic->internal_hash_digestsize;
2530		extra_size = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
2531		dio->payload_len += extra_size;
2532		dio->integrity_payload = kmalloc(dio->payload_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
2533		if (unlikely(!dio->integrity_payload)) {
2534			const unsigned x_size = PAGE_SIZE << 1;
2535			if (dio->payload_len > x_size) {
2536				unsigned sectors = ((x_size - extra_size) / ic->tuple_size) << ic->sb->log2_sectors_per_block;
2537				if (WARN_ON(!sectors || sectors >= bio_sectors(bio))) {
2538					bio->bi_status = BLK_STS_NOTSUPP;
2539					bio_endio(bio);
2540					return DM_MAPIO_SUBMITTED;
2541				}
2542				dm_accept_partial_bio(bio, sectors);
2543				goto retry;
2544			}
2545		}
2546	}
2547
2548	dio->range.logical_sector = bio->bi_iter.bi_sector;
2549	dio->range.n_sectors = bio_sectors(bio);
2550
2551	if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)))
2552		goto skip_spinlock;
2553#ifdef CONFIG_64BIT
2554	/*
2555	 * On 64-bit CPUs we can optimize the lock away (so that it won't cause
2556	 * cache line bouncing) and use acquire/release barriers instead.
2557	 *
2558	 * Paired with smp_store_release in integrity_recalc_inline.
2559	 */
2560	recalc_sector = le64_to_cpu(smp_load_acquire(&ic->sb->recalc_sector));
2561	if (likely(dio->range.logical_sector + dio->range.n_sectors <= recalc_sector))
2562		goto skip_spinlock;
2563#endif
2564	spin_lock_irq(&ic->endio_wait.lock);
2565	recalc_sector = le64_to_cpu(ic->sb->recalc_sector);
2566	if (dio->range.logical_sector + dio->range.n_sectors <= recalc_sector)
2567		goto skip_unlock;
2568	if (unlikely(!add_new_range(ic, &dio->range, true))) {
2569		if (from_map) {
2570			spin_unlock_irq(&ic->endio_wait.lock);
2571			INIT_WORK(&dio->work, integrity_bio_wait);
2572			queue_work(ic->wait_wq, &dio->work);
2573			return DM_MAPIO_SUBMITTED;
2574		}
2575		wait_and_add_new_range(ic, &dio->range);
2576	}
2577	dio->integrity_range_locked = true;
2578skip_unlock:
2579	spin_unlock_irq(&ic->endio_wait.lock);
2580skip_spinlock:
2581
2582	if (unlikely(!dio->integrity_payload)) {
2583		dio->integrity_payload = page_to_virt((struct page *)mempool_alloc(&ic->recheck_pool, GFP_NOIO));
2584		dio->integrity_payload_from_mempool = true;
2585	}
2586
2587	dio->bio_details.bi_iter = bio->bi_iter;
2588
2589	if (unlikely(!dm_integrity_check_limits(ic, bio->bi_iter.bi_sector, bio))) {
2590		return DM_MAPIO_KILL;
2591	}
2592
2593	bio->bi_iter.bi_sector += ic->start + SB_SECTORS;
2594
2595	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
2596	if (IS_ERR(bip)) {
2597		bio->bi_status = errno_to_blk_status(PTR_ERR(bip));
2598		bio_endio(bio);
2599		return DM_MAPIO_SUBMITTED;
2600	}
2601
2602	if (dio->op == REQ_OP_WRITE) {
2603		unsigned pos = 0;
2604		while (dio->bio_details.bi_iter.bi_size) {
2605			struct bio_vec bv = bio_iter_iovec(bio, dio->bio_details.bi_iter);
2606			const char *mem = integrity_kmap(ic, bv.bv_page);
2607			if (ic->tag_size < ic->tuple_size)
2608				memset(dio->integrity_payload + pos + ic->tag_size, 0, ic->tuple_size - ic->tuple_size);
2609			integrity_sector_checksum(ic, &dio->ahash_req, dio->bio_details.bi_iter.bi_sector, mem, bv.bv_offset, dio->integrity_payload + pos);
2610			integrity_kunmap(ic, mem);
2611			pos += ic->tuple_size;
2612			bio_advance_iter_single(bio, &dio->bio_details.bi_iter, ic->sectors_per_block << SECTOR_SHIFT);
2613		}
2614	}
2615
2616	ret = bio_integrity_add_page(bio, virt_to_page(dio->integrity_payload),
2617					dio->payload_len, offset_in_page(dio->integrity_payload));
2618	if (unlikely(ret != dio->payload_len)) {
2619		bio->bi_status = BLK_STS_RESOURCE;
2620		bio_endio(bio);
2621		return DM_MAPIO_SUBMITTED;
2622	}
2623
2624	return DM_MAPIO_REMAPPED;
2625}
2626
2627static inline void dm_integrity_free_payload(struct dm_integrity_io *dio)
2628{
2629	struct dm_integrity_c *ic = dio->ic;
2630	if (unlikely(dio->integrity_payload_from_mempool))
2631		mempool_free(virt_to_page(dio->integrity_payload), &ic->recheck_pool);
2632	else
2633		kfree(dio->integrity_payload);
2634	dio->integrity_payload = NULL;
2635	dio->integrity_payload_from_mempool = false;
2636}
2637
2638static void dm_integrity_inline_recheck(struct work_struct *w)
2639{
2640	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2641	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2642	struct dm_integrity_c *ic = dio->ic;
2643	struct bio *outgoing_bio;
2644	void *outgoing_data;
2645
2646	dio->integrity_payload = page_to_virt((struct page *)mempool_alloc(&ic->recheck_pool, GFP_NOIO));
2647	dio->integrity_payload_from_mempool = true;
2648
2649	outgoing_data = dio->integrity_payload + PAGE_SIZE;
2650
2651	while (dio->bio_details.bi_iter.bi_size) {
2652		char digest[HASH_MAX_DIGESTSIZE];
2653		int r;
2654		struct bio_integrity_payload *bip;
2655		struct bio_vec bv;
2656		char *mem;
2657
2658		outgoing_bio = bio_alloc_bioset(ic->dev->bdev, 1, REQ_OP_READ, GFP_NOIO, &ic->recheck_bios);
2659		bio_add_virt_nofail(outgoing_bio, outgoing_data,
2660				ic->sectors_per_block << SECTOR_SHIFT);
2661
2662		bip = bio_integrity_alloc(outgoing_bio, GFP_NOIO, 1);
2663		if (IS_ERR(bip)) {
2664			bio_put(outgoing_bio);
2665			bio->bi_status = errno_to_blk_status(PTR_ERR(bip));
2666			bio_endio(bio);
2667			return;
2668		}
2669
2670		r = bio_integrity_add_page(outgoing_bio, virt_to_page(dio->integrity_payload), ic->tuple_size, 0);
2671		if (unlikely(r != ic->tuple_size)) {
2672			bio_put(outgoing_bio);
2673			bio->bi_status = BLK_STS_RESOURCE;
2674			bio_endio(bio);
2675			return;
2676		}
2677
2678		outgoing_bio->bi_iter.bi_sector = dio->bio_details.bi_iter.bi_sector + ic->start + SB_SECTORS;
2679
2680		r = submit_bio_wait(outgoing_bio);
2681		if (unlikely(r != 0)) {
2682			bio_put(outgoing_bio);
2683			bio->bi_status = errno_to_blk_status(r);
2684			bio_endio(bio);
2685			return;
2686		}
2687		bio_put(outgoing_bio);
2688
2689		integrity_sector_checksum(ic, &dio->ahash_req, dio->bio_details.bi_iter.bi_sector, integrity_identity(ic, outgoing_data), 0, digest);
2690		if (unlikely(crypto_memneq(digest, dio->integrity_payload, min(ic->internal_hash_digestsize, ic->tag_size)))) {
2691			DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
2692				ic->dev->bdev, dio->bio_details.bi_iter.bi_sector);
2693			atomic64_inc(&ic->number_of_mismatches);
2694			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
2695				bio, dio->bio_details.bi_iter.bi_sector, 0);
2696
2697			bio->bi_status = BLK_STS_PROTECTION;
2698			bio_endio(bio);
2699			return;
2700		}
2701
2702		bv = bio_iter_iovec(bio, dio->bio_details.bi_iter);
2703		mem = bvec_kmap_local(&bv);
2704		memcpy(mem, outgoing_data, ic->sectors_per_block << SECTOR_SHIFT);
2705		kunmap_local(mem);
2706
2707		bio_advance_iter_single(bio, &dio->bio_details.bi_iter, ic->sectors_per_block << SECTOR_SHIFT);
2708	}
2709
2710	bio_endio(bio);
2711}
2712
2713static inline bool dm_integrity_check(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
2714{
2715	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2716	unsigned pos = 0;
2717
2718	while (dio->bio_details.bi_iter.bi_size) {
2719		char digest[HASH_MAX_DIGESTSIZE];
2720		struct bio_vec bv = bio_iter_iovec(bio, dio->bio_details.bi_iter);
2721		char *mem = integrity_kmap(ic, bv.bv_page);
2722		integrity_sector_checksum(ic, &dio->ahash_req, dio->bio_details.bi_iter.bi_sector, mem, bv.bv_offset, digest);
2723		if (unlikely(crypto_memneq(digest, dio->integrity_payload + pos,
2724				min(ic->internal_hash_digestsize, ic->tag_size)))) {
2725			integrity_kunmap(ic, mem);
2726			dm_integrity_free_payload(dio);
2727			INIT_WORK(&dio->work, dm_integrity_inline_recheck);
2728			queue_work(ic->offload_wq, &dio->work);
2729			return false;
2730		}
2731		integrity_kunmap(ic, mem);
2732		pos += ic->tuple_size;
2733		bio_advance_iter_single(bio, &dio->bio_details.bi_iter, ic->sectors_per_block << SECTOR_SHIFT);
2734	}
2735
2736	return true;
2737}
2738
2739static void dm_integrity_inline_async_check(struct work_struct *w)
2740{
2741	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2742	struct dm_integrity_c *ic = dio->ic;
2743	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2744
2745	if (likely(dm_integrity_check(ic, dio)))
2746		bio_endio(bio);
2747}
2748
2749static int dm_integrity_end_io(struct dm_target *ti, struct bio *bio, blk_status_t *status)
2750{
2751	struct dm_integrity_c *ic = ti->private;
2752	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2753	if (ic->mode == 'I') {
2754		if (dio->op == REQ_OP_READ && likely(*status == BLK_STS_OK) && likely(dio->bio_details.bi_iter.bi_size != 0)) {
2755			if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2756			    unlikely(dio->integrity_range_locked))
2757			    	goto skip_check;
2758			if (likely(ic->internal_shash != NULL)) {
2759				if (unlikely(!dm_integrity_check(ic, dio)))
2760					return DM_ENDIO_INCOMPLETE;
2761			} else {
2762				INIT_WORK(&dio->work, dm_integrity_inline_async_check);
2763				queue_work(ic->offload_wq, &dio->work);
2764				return DM_ENDIO_INCOMPLETE;
2765			}
2766		}
2767skip_check:
2768		dm_integrity_free_payload(dio);
2769		if (unlikely(dio->integrity_range_locked))
2770			remove_range(ic, &dio->range);
2771	}
2772	if (unlikely(dio->ahash_req))
2773		mempool_free(dio->ahash_req, &ic->ahash_req_pool);
2774	return DM_ENDIO_DONE;
2775}
2776
2777static void integrity_bio_wait(struct work_struct *w)
2778{
2779	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2780	struct dm_integrity_c *ic = dio->ic;
2781
2782	if (ic->mode == 'I') {
2783		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2784		int r = dm_integrity_map_inline(dio, false);
2785		switch (r) {
2786			case DM_MAPIO_KILL:
2787				bio->bi_status = BLK_STS_IOERR;
2788				fallthrough;
2789			case DM_MAPIO_REMAPPED:
2790				submit_bio_noacct(bio);
2791				fallthrough;
2792			case DM_MAPIO_SUBMITTED:
2793				return;
2794			default:
2795				BUG();
2796		}
2797	} else {
2798		dm_integrity_map_continue(dio, false);
2799	}
2800}
2801
2802static void pad_uncommitted(struct dm_integrity_c *ic)
2803{
2804	if (ic->free_section_entry) {
2805		ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2806		ic->free_section_entry = 0;
2807		ic->free_section++;
2808		wraparound_section(ic, &ic->free_section);
2809		ic->n_uncommitted_sections++;
2810	}
2811	if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2812		    (ic->n_uncommitted_sections + ic->n_committed_sections) *
2813		    ic->journal_section_entries + ic->free_sectors)) {
2814		DMCRIT("journal_sections %u, journal_section_entries %u, "
2815		       "n_uncommitted_sections %u, n_committed_sections %u, "
2816		       "journal_section_entries %u, free_sectors %u",
2817		       ic->journal_sections, ic->journal_section_entries,
2818		       ic->n_uncommitted_sections, ic->n_committed_sections,
2819		       ic->journal_section_entries, ic->free_sectors);
2820	}
2821}
2822
2823static void integrity_commit(struct work_struct *w)
2824{
2825	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2826	unsigned int commit_start, commit_sections;
2827	unsigned int i, j, n;
2828	struct bio *flushes;
2829
2830	timer_delete(&ic->autocommit_timer);
2831
2832	if (ic->mode == 'I')
2833		return;
2834
2835	spin_lock_irq(&ic->endio_wait.lock);
2836	flushes = bio_list_get(&ic->flush_bio_list);
2837	if (unlikely(ic->mode != 'J')) {
2838		spin_unlock_irq(&ic->endio_wait.lock);
2839		dm_integrity_flush_buffers(ic, true);
2840		goto release_flush_bios;
2841	}
2842
2843	pad_uncommitted(ic);
2844	commit_start = ic->uncommitted_section;
2845	commit_sections = ic->n_uncommitted_sections;
2846	spin_unlock_irq(&ic->endio_wait.lock);
2847
2848	if (!commit_sections)
2849		goto release_flush_bios;
2850
2851	ic->wrote_to_journal = true;
2852
2853	i = commit_start;
2854	for (n = 0; n < commit_sections; n++) {
2855		for (j = 0; j < ic->journal_section_entries; j++) {
2856			struct journal_entry *je;
2857
2858			je = access_journal_entry(ic, i, j);
2859			io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2860		}
2861		for (j = 0; j < ic->journal_section_sectors; j++) {
2862			struct journal_sector *js;
2863
2864			js = access_journal(ic, i, j);
2865			js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2866		}
2867		i++;
2868		if (unlikely(i >= ic->journal_sections))
2869			ic->commit_seq = next_commit_seq(ic->commit_seq);
2870		wraparound_section(ic, &i);
2871	}
2872	smp_rmb();
2873
2874	write_journal(ic, commit_start, commit_sections);
2875
2876	spin_lock_irq(&ic->endio_wait.lock);
2877	ic->uncommitted_section += commit_sections;
2878	wraparound_section(ic, &ic->uncommitted_section);
2879	ic->n_uncommitted_sections -= commit_sections;
2880	ic->n_committed_sections += commit_sections;
2881	spin_unlock_irq(&ic->endio_wait.lock);
2882
2883	if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2884		queue_work(ic->writer_wq, &ic->writer_work);
2885
2886release_flush_bios:
2887	while (flushes) {
2888		struct bio *next = flushes->bi_next;
2889
2890		flushes->bi_next = NULL;
2891		do_endio(ic, flushes);
2892		flushes = next;
2893	}
2894}
2895
2896static void complete_copy_from_journal(unsigned long error, void *context)
2897{
2898	struct journal_io *io = context;
2899	struct journal_completion *comp = io->comp;
2900	struct dm_integrity_c *ic = comp->ic;
2901
2902	remove_range(ic, &io->range);
2903	mempool_free(io, &ic->journal_io_mempool);
2904	if (unlikely(error != 0))
2905		dm_integrity_io_error(ic, "copying from journal", -EIO);
2906	complete_journal_op(comp);
2907}
2908
2909static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2910			       struct journal_entry *je)
2911{
2912	unsigned int s = 0;
2913
2914	do {
2915		js->commit_id = je->last_bytes[s];
2916		js++;
2917	} while (++s < ic->sectors_per_block);
2918}
2919
2920static void do_journal_write(struct dm_integrity_c *ic, unsigned int write_start,
2921			     unsigned int write_sections, bool from_replay)
2922{
2923	unsigned int i, j, n;
2924	struct journal_completion comp;
2925	struct blk_plug plug;
2926
2927	blk_start_plug(&plug);
2928
2929	comp.ic = ic;
2930	comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2931	init_completion(&comp.comp);
2932
2933	i = write_start;
2934	for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2935#ifndef INTERNAL_VERIFY
2936		if (unlikely(from_replay))
2937#endif
2938			rw_section_mac(ic, i, false);
2939		for (j = 0; j < ic->journal_section_entries; j++) {
2940			struct journal_entry *je = access_journal_entry(ic, i, j);
2941			sector_t sec, area, offset;
2942			unsigned int k, l, next_loop;
2943			sector_t metadata_block;
2944			unsigned int metadata_offset;
2945			struct journal_io *io;
2946
2947			if (journal_entry_is_unused(je))
2948				continue;
2949			BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2950			sec = journal_entry_get_sector(je);
2951			if (unlikely(from_replay)) {
2952				if (unlikely(sec & (unsigned int)(ic->sectors_per_block - 1))) {
2953					dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2954					sec &= ~(sector_t)(ic->sectors_per_block - 1);
2955				}
2956				if (unlikely(sec >= ic->provided_data_sectors)) {
2957					journal_entry_set_unused(je);
2958					continue;
2959				}
2960			}
2961			get_area_and_offset(ic, sec, &area, &offset);
2962			restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2963			for (k = j + 1; k < ic->journal_section_entries; k++) {
2964				struct journal_entry *je2 = access_journal_entry(ic, i, k);
2965				sector_t sec2, area2, offset2;
2966
2967				if (journal_entry_is_unused(je2))
2968					break;
2969				BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2970				sec2 = journal_entry_get_sector(je2);
2971				if (unlikely(sec2 >= ic->provided_data_sectors))
2972					break;
2973				get_area_and_offset(ic, sec2, &area2, &offset2);
2974				if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2975					break;
2976				restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2977			}
2978			next_loop = k - 1;
2979
2980			io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2981			io->comp = &comp;
2982			io->range.logical_sector = sec;
2983			io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2984
2985			spin_lock_irq(&ic->endio_wait.lock);
2986			add_new_range_and_wait(ic, &io->range);
2987
2988			if (likely(!from_replay)) {
2989				struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2990
2991				/* don't write if there is newer committed sector */
2992				while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2993					struct journal_entry *je2 = access_journal_entry(ic, i, j);
2994
2995					journal_entry_set_unused(je2);
2996					remove_journal_node(ic, &section_node[j]);
2997					j++;
2998					sec += ic->sectors_per_block;
2999					offset += ic->sectors_per_block;
3000				}
3001				while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
3002					struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
3003
3004					journal_entry_set_unused(je2);
3005					remove_journal_node(ic, &section_node[k - 1]);
3006					k--;
3007				}
3008				if (j == k) {
3009					remove_range_unlocked(ic, &io->range);
3010					spin_unlock_irq(&ic->endio_wait.lock);
3011					mempool_free(io, &ic->journal_io_mempool);
3012					goto skip_io;
3013				}
3014				for (l = j; l < k; l++)
3015					remove_journal_node(ic, &section_node[l]);
3016			}
3017			spin_unlock_irq(&ic->endio_wait.lock);
3018
3019			metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
3020			for (l = j; l < k; l++) {
3021				int r;
3022				struct journal_entry *je2 = access_journal_entry(ic, i, l);
3023
3024				if (
3025#ifndef INTERNAL_VERIFY
3026				    unlikely(from_replay) &&
3027#endif
3028				    ic->internal_hash) {
3029					char test_tag[MAX_T(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
3030					struct journal_sector *js = access_journal_data(ic, i, l);
3031					void *js_page = integrity_identity(ic, (char *)js - offset_in_page(js));
3032					unsigned js_offset = offset_in_page(js);
3033
3034					integrity_sector_checksum(ic, &ic->journal_ahash_req, sec + ((l - j) << ic->sb->log2_sectors_per_block),
3035								  js_page, js_offset, test_tag);
3036					if (unlikely(crypto_memneq(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
3037						dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
3038						dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
3039					}
3040				}
3041
3042				journal_entry_set_unused(je2);
3043				r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
3044							ic->tag_size, TAG_WRITE);
3045				if (unlikely(r))
3046					dm_integrity_io_error(ic, "reading tags", r);
3047			}
3048
3049			atomic_inc(&comp.in_flight);
3050			copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
3051					  (k - j) << ic->sb->log2_sectors_per_block,
3052					  get_data_sector(ic, area, offset),
3053					  complete_copy_from_journal, io);
3054skip_io:
3055			j = next_loop;
3056		}
3057	}
3058
3059	dm_bufio_write_dirty_buffers_async(ic->bufio);
3060
3061	blk_finish_plug(&plug);
3062
3063	complete_journal_op(&comp);
3064	wait_for_completion_io(&comp.comp);
3065
3066	dm_integrity_flush_buffers(ic, true);
3067}
3068
3069static void integrity_writer(struct work_struct *w)
3070{
3071	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
3072	unsigned int write_start, write_sections;
3073	unsigned int prev_free_sectors;
3074
3075	spin_lock_irq(&ic->endio_wait.lock);
3076	write_start = ic->committed_section;
3077	write_sections = ic->n_committed_sections;
3078	spin_unlock_irq(&ic->endio_wait.lock);
3079
3080	if (!write_sections)
3081		return;
3082
3083	do_journal_write(ic, write_start, write_sections, false);
3084
3085	spin_lock_irq(&ic->endio_wait.lock);
3086
3087	ic->committed_section += write_sections;
3088	wraparound_section(ic, &ic->committed_section);
3089	ic->n_committed_sections -= write_sections;
3090
3091	prev_free_sectors = ic->free_sectors;
3092	ic->free_sectors += write_sections * ic->journal_section_entries;
3093	if (unlikely(!prev_free_sectors))
3094		wake_up_locked(&ic->endio_wait);
3095
3096	spin_unlock_irq(&ic->endio_wait.lock);
3097}
3098
3099static void recalc_write_super(struct dm_integrity_c *ic)
3100{
3101	int r;
3102
3103	dm_integrity_flush_buffers(ic, false);
3104	if (dm_integrity_failed(ic))
3105		return;
3106
3107	r = sync_rw_sb(ic, REQ_OP_WRITE);
3108	if (unlikely(r))
3109		dm_integrity_io_error(ic, "writing superblock", r);
3110}
3111
3112static void integrity_recalc(struct work_struct *w)
3113{
3114	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
3115	size_t recalc_tags_size;
3116	u8 *recalc_buffer = NULL;
3117	u8 *recalc_tags = NULL;
3118	struct ahash_request *ahash_req = NULL;
3119	struct dm_integrity_range range;
3120	struct dm_io_request io_req;
3121	struct dm_io_region io_loc;
3122	sector_t area, offset;
3123	sector_t metadata_block;
3124	unsigned int metadata_offset;
3125	sector_t logical_sector, n_sectors;
3126	__u8 *t;
3127	unsigned int i;
3128	int r;
3129	unsigned int super_counter = 0;
3130	unsigned recalc_sectors = RECALC_SECTORS;
3131
3132retry:
3133	recalc_buffer = kmalloc(recalc_sectors << SECTOR_SHIFT, GFP_NOIO | __GFP_NOWARN);
3134	if (!recalc_buffer) {
3135oom:
3136		recalc_sectors >>= 1;
3137		if (recalc_sectors >= 1U << ic->sb->log2_sectors_per_block)
3138			goto retry;
3139		DMCRIT("out of memory for recalculate buffer - recalculation disabled");
3140		goto free_ret;
3141	}
3142	recalc_tags_size = (recalc_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3143	if (ic->internal_hash_digestsize > ic->tag_size)
3144		recalc_tags_size += ic->internal_hash_digestsize - ic->tag_size;
3145	recalc_tags = kvmalloc(recalc_tags_size, GFP_NOIO);
3146	if (!recalc_tags) {
3147		kfree(recalc_buffer);
3148		recalc_buffer = NULL;
3149		goto oom;
3150	}
3151
3152	DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
3153
3154	spin_lock_irq(&ic->endio_wait.lock);
3155
3156next_chunk:
3157
3158	if (unlikely(dm_post_suspending(ic->ti)))
3159		goto unlock_ret;
3160
3161	range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
3162	if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
3163		if (ic->mode == 'B') {
3164			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3165			DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
3166			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3167		}
3168		goto unlock_ret;
3169	}
3170
3171	get_area_and_offset(ic, range.logical_sector, &area, &offset);
3172	range.n_sectors = min((sector_t)recalc_sectors, ic->provided_data_sectors - range.logical_sector);
3173	if (!ic->meta_dev)
3174		range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned int)offset);
3175
3176	add_new_range_and_wait(ic, &range);
3177	spin_unlock_irq(&ic->endio_wait.lock);
3178	logical_sector = range.logical_sector;
3179	n_sectors = range.n_sectors;
3180
3181	if (ic->mode == 'B') {
3182		if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
3183			goto advance_and_next;
3184
3185		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
3186				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
3187			logical_sector += ic->sectors_per_block;
3188			n_sectors -= ic->sectors_per_block;
3189			cond_resched();
3190		}
3191		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
3192				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
3193			n_sectors -= ic->sectors_per_block;
3194			cond_resched();
3195		}
3196		get_area_and_offset(ic, logical_sector, &area, &offset);
3197	}
3198
3199	DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
3200
3201	if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
3202		recalc_write_super(ic);
3203		if (ic->mode == 'B')
3204			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
3205
3206		super_counter = 0;
3207	}
3208
3209	if (unlikely(dm_integrity_failed(ic)))
3210		goto err;
3211
3212	io_req.bi_opf = REQ_OP_READ;
3213	io_req.mem.type = DM_IO_KMEM;
3214	io_req.mem.ptr.addr = recalc_buffer;
3215	io_req.notify.fn = NULL;
3216	io_req.client = ic->io;
3217	io_loc.bdev = ic->dev->bdev;
3218	io_loc.sector = get_data_sector(ic, area, offset);
3219	io_loc.count = n_sectors;
3220
3221	r = dm_io(&io_req, 1, &io_loc, NULL, IOPRIO_DEFAULT);
3222	if (unlikely(r)) {
3223		dm_integrity_io_error(ic, "reading data", r);
3224		goto err;
3225	}
3226
3227	t = recalc_tags;
3228	for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
3229		void *ptr = recalc_buffer + (i << SECTOR_SHIFT);
3230		void *ptr_page = integrity_identity(ic, (char *)ptr - offset_in_page(ptr));
3231		unsigned ptr_offset = offset_in_page(ptr);
3232		integrity_sector_checksum(ic, &ahash_req, logical_sector + i, ptr_page, ptr_offset, t);
3233		t += ic->tag_size;
3234	}
3235
3236	metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
3237
3238	r = dm_integrity_rw_tag(ic, recalc_tags, &metadata_block, &metadata_offset, t - recalc_tags, TAG_WRITE);
3239	if (unlikely(r)) {
3240		dm_integrity_io_error(ic, "writing tags", r);
3241		goto err;
3242	}
3243
3244	if (ic->mode == 'B') {
3245		sector_t start, end;
3246
3247		start = (range.logical_sector >>
3248			 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
3249			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
3250		end = ((range.logical_sector + range.n_sectors) >>
3251		       (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
3252			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
3253		block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
3254	}
3255
3256advance_and_next:
3257	cond_resched();
3258
3259	spin_lock_irq(&ic->endio_wait.lock);
3260	remove_range_unlocked(ic, &range);
3261	ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
3262	goto next_chunk;
3263
3264err:
3265	remove_range(ic, &range);
3266	goto free_ret;
3267
3268unlock_ret:
3269	spin_unlock_irq(&ic->endio_wait.lock);
3270
3271	recalc_write_super(ic);
3272
3273free_ret:
3274	kfree(recalc_buffer);
3275	kvfree(recalc_tags);
3276	mempool_free(ahash_req, &ic->ahash_req_pool);
3277}
3278
3279static void integrity_recalc_inline(struct work_struct *w)
3280{
3281	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
3282	size_t recalc_tags_size;
3283	u8 *recalc_buffer = NULL;
3284	u8 *recalc_tags = NULL;
3285	struct ahash_request *ahash_req = NULL;
3286	struct dm_integrity_range range;
3287	struct bio *bio;
3288	struct bio_integrity_payload *bip;
3289	__u8 *t;
3290	unsigned int i;
3291	int r;
3292	unsigned ret;
3293	unsigned int super_counter = 0;
3294	unsigned recalc_sectors = RECALC_SECTORS;
3295
3296retry:
3297	recalc_buffer = kmalloc(recalc_sectors << SECTOR_SHIFT, GFP_NOIO | __GFP_NOWARN);
3298	if (!recalc_buffer) {
3299oom:
3300		recalc_sectors >>= 1;
3301		if (recalc_sectors >= 1U << ic->sb->log2_sectors_per_block)
3302			goto retry;
3303		DMCRIT("out of memory for recalculate buffer - recalculation disabled");
3304		goto free_ret;
3305	}
3306
3307	recalc_tags_size = (recalc_sectors >> ic->sb->log2_sectors_per_block) * ic->tuple_size;
3308	if (ic->internal_hash_digestsize > ic->tuple_size)
3309		recalc_tags_size += ic->internal_hash_digestsize - ic->tuple_size;
3310	recalc_tags = kmalloc(recalc_tags_size, GFP_NOIO | __GFP_NOWARN);
3311	if (!recalc_tags) {
3312		kfree(recalc_buffer);
3313		recalc_buffer = NULL;
3314		goto oom;
3315	}
3316
3317	spin_lock_irq(&ic->endio_wait.lock);
3318
3319next_chunk:
3320	if (unlikely(dm_post_suspending(ic->ti)))
3321		goto unlock_ret;
3322
3323	range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
3324	if (unlikely(range.logical_sector >= ic->provided_data_sectors))
3325		goto unlock_ret;
3326	range.n_sectors = min((sector_t)recalc_sectors, ic->provided_data_sectors - range.logical_sector);
3327
3328	add_new_range_and_wait(ic, &range);
3329	spin_unlock_irq(&ic->endio_wait.lock);
3330
3331	if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
3332		recalc_write_super(ic);
3333		super_counter = 0;
3334	}
3335
3336	if (unlikely(dm_integrity_failed(ic)))
3337		goto err;
3338
3339	DEBUG_print("recalculating: %llx - %llx\n", range.logical_sector, range.n_sectors);
3340
3341	bio = bio_alloc_bioset(ic->dev->bdev, 1, REQ_OP_READ, GFP_NOIO, &ic->recalc_bios);
3342	bio->bi_iter.bi_sector = ic->start + SB_SECTORS + range.logical_sector;
3343	bio_add_virt_nofail(bio, recalc_buffer,
3344			range.n_sectors << SECTOR_SHIFT);
3345	r = submit_bio_wait(bio);
3346	bio_put(bio);
3347	if (unlikely(r)) {
3348		dm_integrity_io_error(ic, "reading data", r);
3349		goto err;
3350	}
3351
3352	t = recalc_tags;
3353	for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
3354		void *ptr = recalc_buffer + (i << SECTOR_SHIFT);
3355		void *ptr_page = integrity_identity(ic, (char *)ptr - offset_in_page(ptr));
3356		unsigned ptr_offset = offset_in_page(ptr);
3357		memset(t, 0, ic->tuple_size);
3358		integrity_sector_checksum(ic, &ahash_req, range.logical_sector + i, ptr_page, ptr_offset, t);
3359		t += ic->tuple_size;
3360	}
3361
3362	bio = bio_alloc_bioset(ic->dev->bdev, 1, REQ_OP_WRITE, GFP_NOIO, &ic->recalc_bios);
3363	bio->bi_iter.bi_sector = ic->start + SB_SECTORS + range.logical_sector;
3364	bio_add_virt_nofail(bio, recalc_buffer,
3365			range.n_sectors << SECTOR_SHIFT);
3366
3367	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
3368	if (unlikely(IS_ERR(bip))) {
3369		bio_put(bio);
3370		DMCRIT("out of memory for bio integrity payload - recalculation disabled");
3371		goto err;
3372	}
3373	ret = bio_integrity_add_page(bio, virt_to_page(recalc_tags), t - recalc_tags, offset_in_page(recalc_tags));
3374	if (unlikely(ret != t - recalc_tags)) {
3375		bio_put(bio);
3376		dm_integrity_io_error(ic, "attaching integrity tags", -ENOMEM);
3377		goto err;
3378	}
3379
3380	r = submit_bio_wait(bio);
3381	bio_put(bio);
3382	if (unlikely(r)) {
3383		dm_integrity_io_error(ic, "writing data", r);
3384		goto err;
3385	}
3386
3387	cond_resched();
3388	spin_lock_irq(&ic->endio_wait.lock);
3389	remove_range_unlocked(ic, &range);
3390#ifdef CONFIG_64BIT
3391	/* Paired with smp_load_acquire in dm_integrity_map_inline. */
3392	smp_store_release(&ic->sb->recalc_sector, cpu_to_le64(range.logical_sector + range.n_sectors));
3393#else
3394	ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
3395#endif
3396	goto next_chunk;
3397
3398err:
3399	remove_range(ic, &range);
3400	goto free_ret;
3401
3402unlock_ret:
3403	spin_unlock_irq(&ic->endio_wait.lock);
3404
3405	recalc_write_super(ic);
3406
3407free_ret:
3408	kfree(recalc_buffer);
3409	kfree(recalc_tags);
3410	mempool_free(ahash_req, &ic->ahash_req_pool);
3411}
3412
3413static void bitmap_block_work(struct work_struct *w)
3414{
3415	struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
3416	struct dm_integrity_c *ic = bbs->ic;
3417	struct bio *bio;
3418	struct bio_list bio_queue;
3419	struct bio_list waiting;
3420
3421	bio_list_init(&waiting);
3422
3423	spin_lock(&bbs->bio_queue_lock);
3424	bio_queue = bbs->bio_queue;
3425	bio_list_init(&bbs->bio_queue);
3426	spin_unlock(&bbs->bio_queue_lock);
3427
3428	while ((bio = bio_list_pop(&bio_queue))) {
3429		struct dm_integrity_io *dio;
3430
3431		dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
3432
3433		if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
3434				    dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
3435			remove_range(ic, &dio->range);
3436			INIT_WORK(&dio->work, integrity_bio_wait);
3437			queue_work(ic->offload_wq, &dio->work);
3438		} else {
3439			block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
3440					dio->range.n_sectors, BITMAP_OP_SET);
3441			bio_list_add(&waiting, bio);
3442		}
3443	}
3444
3445	if (bio_list_empty(&waiting))
3446		return;
3447
3448	rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
3449			   bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
3450			   BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
3451
3452	while ((bio = bio_list_pop(&waiting))) {
3453		struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
3454
3455		block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
3456				dio->range.n_sectors, BITMAP_OP_SET);
3457
3458		remove_range(ic, &dio->range);
3459		INIT_WORK(&dio->work, integrity_bio_wait);
3460		queue_work(ic->offload_wq, &dio->work);
3461	}
3462
3463	queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
3464}
3465
3466static void bitmap_flush_work(struct work_struct *work)
3467{
3468	struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
3469	struct dm_integrity_range range;
3470	unsigned long limit;
3471	struct bio *bio;
3472
3473	dm_integrity_flush_buffers(ic, false);
3474
3475	range.logical_sector = 0;
3476	range.n_sectors = ic->provided_data_sectors;
3477
3478	spin_lock_irq(&ic->endio_wait.lock);
3479	add_new_range_and_wait(ic, &range);
3480	spin_unlock_irq(&ic->endio_wait.lock);
3481
3482	dm_integrity_flush_buffers(ic, true);
3483
3484	limit = ic->provided_data_sectors;
3485	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3486		limit = le64_to_cpu(ic->sb->recalc_sector)
3487			>> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
3488			<< (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
3489	}
3490	/*DEBUG_print("zeroing journal\n");*/
3491	block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
3492	block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
3493
3494	rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3495			   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3496
3497	spin_lock_irq(&ic->endio_wait.lock);
3498	remove_range_unlocked(ic, &range);
3499	while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
3500		bio_endio(bio);
3501		spin_unlock_irq(&ic->endio_wait.lock);
3502		spin_lock_irq(&ic->endio_wait.lock);
3503	}
3504	spin_unlock_irq(&ic->endio_wait.lock);
3505}
3506
3507
3508static void init_journal(struct dm_integrity_c *ic, unsigned int start_section,
3509			 unsigned int n_sections, unsigned char commit_seq)
3510{
3511	unsigned int i, j, n;
3512
3513	if (!n_sections)
3514		return;
3515
3516	for (n = 0; n < n_sections; n++) {
3517		i = start_section + n;
3518		wraparound_section(ic, &i);
3519		for (j = 0; j < ic->journal_section_sectors; j++) {
3520			struct journal_sector *js = access_journal(ic, i, j);
3521
3522			BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
3523			memset(&js->sectors, 0, sizeof(js->sectors));
3524			js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
3525		}
3526		for (j = 0; j < ic->journal_section_entries; j++) {
3527			struct journal_entry *je = access_journal_entry(ic, i, j);
3528
3529			journal_entry_set_unused(je);
3530		}
3531	}
3532
3533	write_journal(ic, start_section, n_sections);
3534}
3535
3536static int find_commit_seq(struct dm_integrity_c *ic, unsigned int i, unsigned int j, commit_id_t id)
3537{
3538	unsigned char k;
3539
3540	for (k = 0; k < N_COMMIT_IDS; k++) {
3541		if (dm_integrity_commit_id(ic, i, j, k) == id)
3542			return k;
3543	}
3544	dm_integrity_io_error(ic, "journal commit id", -EIO);
3545	return -EIO;
3546}
3547
3548static void replay_journal(struct dm_integrity_c *ic)
3549{
3550	unsigned int i, j;
3551	bool used_commit_ids[N_COMMIT_IDS];
3552	unsigned int max_commit_id_sections[N_COMMIT_IDS];
3553	unsigned int write_start, write_sections;
3554	unsigned int continue_section;
3555	bool journal_empty;
3556	unsigned char unused, last_used, want_commit_seq;
3557
3558	if (ic->mode == 'R')
3559		return;
3560
3561	if (ic->journal_uptodate)
3562		return;
3563
3564	last_used = 0;
3565	write_start = 0;
3566
3567	if (!ic->just_formatted) {
3568		DEBUG_print("reading journal\n");
3569		rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
3570		if (ic->journal_io)
3571			DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
3572		if (ic->journal_io) {
3573			struct journal_completion crypt_comp;
3574
3575			crypt_comp.ic = ic;
3576			init_completion(&crypt_comp.comp);
3577			crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
3578			encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
3579			wait_for_completion(&crypt_comp.comp);
3580		}
3581		DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
3582	}
3583
3584	if (dm_integrity_failed(ic))
3585		goto clear_journal;
3586
3587	journal_empty = true;
3588	memset(used_commit_ids, 0, sizeof(used_commit_ids));
3589	memset(max_commit_id_sections, 0, sizeof(max_commit_id_sections));
3590	for (i = 0; i < ic->journal_sections; i++) {
3591		for (j = 0; j < ic->journal_section_sectors; j++) {
3592			int k;
3593			struct journal_sector *js = access_journal(ic, i, j);
3594
3595			k = find_commit_seq(ic, i, j, js->commit_id);
3596			if (k < 0)
3597				goto clear_journal;
3598			used_commit_ids[k] = true;
3599			max_commit_id_sections[k] = i;
3600		}
3601		if (journal_empty) {
3602			for (j = 0; j < ic->journal_section_entries; j++) {
3603				struct journal_entry *je = access_journal_entry(ic, i, j);
3604
3605				if (!journal_entry_is_unused(je)) {
3606					journal_empty = false;
3607					break;
3608				}
3609			}
3610		}
3611	}
3612
3613	if (!used_commit_ids[N_COMMIT_IDS - 1]) {
3614		unused = N_COMMIT_IDS - 1;
3615		while (unused && !used_commit_ids[unused - 1])
3616			unused--;
3617	} else {
3618		for (unused = 0; unused < N_COMMIT_IDS; unused++)
3619			if (!used_commit_ids[unused])
3620				break;
3621		if (unused == N_COMMIT_IDS) {
3622			dm_integrity_io_error(ic, "journal commit ids", -EIO);
3623			goto clear_journal;
3624		}
3625	}
3626	DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
3627		    unused, used_commit_ids[0], used_commit_ids[1],
3628		    used_commit_ids[2], used_commit_ids[3]);
3629
3630	last_used = prev_commit_seq(unused);
3631	want_commit_seq = prev_commit_seq(last_used);
3632
3633	if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
3634		journal_empty = true;
3635
3636	write_start = max_commit_id_sections[last_used] + 1;
3637	if (unlikely(write_start >= ic->journal_sections))
3638		want_commit_seq = next_commit_seq(want_commit_seq);
3639	wraparound_section(ic, &write_start);
3640
3641	i = write_start;
3642	for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
3643		for (j = 0; j < ic->journal_section_sectors; j++) {
3644			struct journal_sector *js = access_journal(ic, i, j);
3645
3646			if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
3647				/*
3648				 * This could be caused by crash during writing.
3649				 * We won't replay the inconsistent part of the
3650				 * journal.
3651				 */
3652				DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3653					    i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3654				goto brk;
3655			}
3656		}
3657		i++;
3658		if (unlikely(i >= ic->journal_sections))
3659			want_commit_seq = next_commit_seq(want_commit_seq);
3660		wraparound_section(ic, &i);
3661	}
3662brk:
3663
3664	if (!journal_empty) {
3665		DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3666			    write_sections, write_start, want_commit_seq);
3667		do_journal_write(ic, write_start, write_sections, true);
3668	}
3669
3670	if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3671		continue_section = write_start;
3672		ic->commit_seq = want_commit_seq;
3673		DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3674	} else {
3675		unsigned int s;
3676		unsigned char erase_seq;
3677
3678clear_journal:
3679		DEBUG_print("clearing journal\n");
3680
3681		erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3682		s = write_start;
3683		init_journal(ic, s, 1, erase_seq);
3684		s++;
3685		wraparound_section(ic, &s);
3686		if (ic->journal_sections >= 2) {
3687			init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3688			s += ic->journal_sections - 2;
3689			wraparound_section(ic, &s);
3690			init_journal(ic, s, 1, erase_seq);
3691		}
3692
3693		continue_section = 0;
3694		ic->commit_seq = next_commit_seq(erase_seq);
3695	}
3696
3697	ic->committed_section = continue_section;
3698	ic->n_committed_sections = 0;
3699
3700	ic->uncommitted_section = continue_section;
3701	ic->n_uncommitted_sections = 0;
3702
3703	ic->free_section = continue_section;
3704	ic->free_section_entry = 0;
3705	ic->free_sectors = ic->journal_entries;
3706
3707	ic->journal_tree_root = RB_ROOT;
3708	for (i = 0; i < ic->journal_entries; i++)
3709		init_journal_node(&ic->journal_tree[i]);
3710}
3711
3712static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3713{
3714	DEBUG_print("%s\n", __func__);
3715
3716	if (ic->mode == 'B') {
3717		ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3718		ic->synchronous_mode = 1;
3719
3720		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3721		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3722		flush_workqueue(ic->commit_wq);
3723	}
3724}
3725
3726static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3727{
3728	struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3729
3730	DEBUG_print("%s\n", __func__);
3731
3732	dm_integrity_enter_synchronous_mode(ic);
3733
3734	return NOTIFY_DONE;
3735}
3736
3737static void dm_integrity_postsuspend(struct dm_target *ti)
3738{
3739	struct dm_integrity_c *ic = ti->private;
3740	int r;
3741
3742	WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3743
3744	timer_delete_sync(&ic->autocommit_timer);
3745
3746	if (ic->recalc_wq)
3747		drain_workqueue(ic->recalc_wq);
3748
3749	if (ic->mode == 'B')
3750		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3751
3752	queue_work(ic->commit_wq, &ic->commit_work);
3753	drain_workqueue(ic->commit_wq);
3754
3755	if (ic->mode == 'J') {
3756		queue_work(ic->writer_wq, &ic->writer_work);
3757		drain_workqueue(ic->writer_wq);
3758		dm_integrity_flush_buffers(ic, true);
3759		if (ic->wrote_to_journal) {
3760			init_journal(ic, ic->free_section,
3761				     ic->journal_sections - ic->free_section, ic->commit_seq);
3762			if (ic->free_section) {
3763				init_journal(ic, 0, ic->free_section,
3764					     next_commit_seq(ic->commit_seq));
3765			}
3766		}
3767	}
3768
3769	if (ic->mode == 'B') {
3770		dm_integrity_flush_buffers(ic, true);
3771#if 1
3772		/* set to 0 to test bitmap replay code */
3773		init_journal(ic, 0, ic->journal_sections, 0);
3774		ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3775		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3776		if (unlikely(r))
3777			dm_integrity_io_error(ic, "writing superblock", r);
3778#endif
3779	}
3780
3781	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3782
3783	ic->journal_uptodate = true;
3784}
3785
3786static void dm_integrity_resume(struct dm_target *ti)
3787{
3788	struct dm_integrity_c *ic = ti->private;
3789	__u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3790	int r;
3791	__le32 flags;
3792
3793	DEBUG_print("resume\n");
3794
3795	ic->wrote_to_journal = false;
3796
3797	flags = ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING);
3798	r = sync_rw_sb(ic, REQ_OP_READ);
3799	if (r)
3800		dm_integrity_io_error(ic, "reading superblock", r);
3801	if ((ic->sb->flags & flags) != flags) {
3802		ic->sb->flags |= flags;
3803		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3804		if (unlikely(r))
3805			dm_integrity_io_error(ic, "writing superblock", r);
3806	}
3807
3808	if (ic->provided_data_sectors != old_provided_data_sectors) {
3809		if (ic->provided_data_sectors > old_provided_data_sectors &&
3810		    ic->mode == 'B' &&
3811		    ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP) &&
3812		    ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3813			rw_journal_sectors(ic, REQ_OP_READ, 0,
3814					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3815			block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3816					ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3817			rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3818					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3819		}
3820
3821		ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3822		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3823		if (unlikely(r))
3824			dm_integrity_io_error(ic, "writing superblock", r);
3825	}
3826
3827	if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3828		DEBUG_print("resume dirty_bitmap\n");
3829		rw_journal_sectors(ic, REQ_OP_READ, 0,
3830				   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3831		if (ic->mode == 'B') {
3832			if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3833			    !ic->reset_recalculate_flag) {
3834				block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3835				block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3836				if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3837						     BITMAP_OP_TEST_ALL_CLEAR)) {
3838					ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3839					ic->sb->recalc_sector = cpu_to_le64(0);
3840				}
3841			} else {
3842				DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3843					    ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3844				ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3845				block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3846				block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3847				block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3848				rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3849						   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3850				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3851				ic->sb->recalc_sector = cpu_to_le64(0);
3852			}
3853		} else {
3854			if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3855			      block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3856			    ic->reset_recalculate_flag) {
3857				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3858				ic->sb->recalc_sector = cpu_to_le64(0);
3859			}
3860			init_journal(ic, 0, ic->journal_sections, 0);
3861			replay_journal(ic);
3862			ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3863		}
3864		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3865		if (unlikely(r))
3866			dm_integrity_io_error(ic, "writing superblock", r);
3867	} else {
3868		replay_journal(ic);
3869		if (ic->reset_recalculate_flag) {
3870			ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3871			ic->sb->recalc_sector = cpu_to_le64(0);
3872		}
3873		if (ic->mode == 'B') {
3874			ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3875			ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3876			r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3877			if (unlikely(r))
3878				dm_integrity_io_error(ic, "writing superblock", r);
3879
3880			block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3881			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3882			block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3883			if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3884			    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3885				block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3886						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3887				block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3888						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3889				block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3890						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3891			}
3892			rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3893					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3894		}
3895	}
3896
3897	DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3898	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3899		__u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3900
3901		DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3902		if (recalc_pos < ic->provided_data_sectors) {
3903			queue_work(ic->recalc_wq, &ic->recalc_work);
3904		} else if (recalc_pos > ic->provided_data_sectors) {
3905			ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3906			recalc_write_super(ic);
3907		}
3908	}
3909
3910	ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3911	ic->reboot_notifier.next = NULL;
3912	ic->reboot_notifier.priority = INT_MAX - 1;	/* be notified after md and before hardware drivers */
3913	WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3914
3915#if 0
3916	/* set to 1 to stress test synchronous mode */
3917	dm_integrity_enter_synchronous_mode(ic);
3918#endif
3919}
3920
3921static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3922				unsigned int status_flags, char *result, unsigned int maxlen)
3923{
3924	struct dm_integrity_c *ic = ti->private;
3925	unsigned int arg_count;
3926	size_t sz = 0;
3927
3928	switch (type) {
3929	case STATUSTYPE_INFO:
3930		DMEMIT("%llu %llu",
3931			(unsigned long long)atomic64_read(&ic->number_of_mismatches),
3932			ic->provided_data_sectors);
3933		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3934			DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3935		else
3936			DMEMIT(" -");
3937		break;
3938
3939	case STATUSTYPE_TABLE: {
3940		arg_count = 1; /* buffer_sectors */
3941		arg_count += !!ic->meta_dev;
3942		arg_count += ic->sectors_per_block != 1;
3943		arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3944		arg_count += ic->reset_recalculate_flag;
3945		arg_count += ic->discard;
3946		arg_count += ic->mode != 'I'; /* interleave_sectors */
3947		arg_count += ic->mode == 'J'; /* journal_sectors */
3948		arg_count += ic->mode == 'J'; /* journal_watermark */
3949		arg_count += ic->mode == 'J'; /* commit_time */
3950		arg_count += ic->mode == 'B'; /* sectors_per_bit */
3951		arg_count += ic->mode == 'B'; /* bitmap_flush_interval */
3952		arg_count += !!ic->internal_hash_alg.alg_string;
3953		arg_count += !!ic->journal_crypt_alg.alg_string;
3954		arg_count += !!ic->journal_mac_alg.alg_string;
3955		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3956		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3957		arg_count += ic->legacy_recalculate;
3958		DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3959		       ic->tag_size, ic->mode, arg_count);
3960		if (ic->meta_dev)
3961			DMEMIT(" meta_device:%s", ic->meta_dev->name);
3962		if (ic->sectors_per_block != 1)
3963			DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3964		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3965			DMEMIT(" recalculate");
3966		if (ic->reset_recalculate_flag)
3967			DMEMIT(" reset_recalculate");
3968		if (ic->discard)
3969			DMEMIT(" allow_discards");
3970		if (ic->mode != 'I')
3971			DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3972		DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3973		if (ic->mode == 'J') {
3974			__u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3975
3976			watermark_percentage += ic->journal_entries / 2;
3977			do_div(watermark_percentage, ic->journal_entries);
3978			DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3979			DMEMIT(" journal_watermark:%u", (unsigned int)watermark_percentage);
3980			DMEMIT(" commit_time:%u", ic->autocommit_msec);
3981		}
3982		if (ic->mode == 'B') {
3983			DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3984			DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3985		}
3986		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3987			DMEMIT(" fix_padding");
3988		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3989			DMEMIT(" fix_hmac");
3990		if (ic->legacy_recalculate)
3991			DMEMIT(" legacy_recalculate");
3992
3993#define EMIT_ALG(a, n)							\
3994		do {							\
3995			if (ic->a.alg_string) {				\
3996				DMEMIT(" %s:%s", n, ic->a.alg_string);	\
3997				if (ic->a.key_string)			\
3998					DMEMIT(":%s", ic->a.key_string);\
3999			}						\
4000		} while (0)
4001		EMIT_ALG(internal_hash_alg, "internal_hash");
4002		EMIT_ALG(journal_crypt_alg, "journal_crypt");
4003		EMIT_ALG(journal_mac_alg, "journal_mac");
4004		break;
4005	}
4006	case STATUSTYPE_IMA:
4007		DMEMIT_TARGET_NAME_VERSION(ti->type);
4008		DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
4009			ic->dev->name, ic->start, ic->tag_size, ic->mode);
4010
4011		if (ic->meta_dev)
4012			DMEMIT(",meta_device=%s", ic->meta_dev->name);
4013		if (ic->sectors_per_block != 1)
4014			DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
4015
4016		DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
4017		       'y' : 'n');
4018		DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
4019		DMEMIT(",fix_padding=%c",
4020		       ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
4021		DMEMIT(",fix_hmac=%c",
4022		       ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
4023		DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
4024
4025		DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
4026		DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
4027		DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
4028		DMEMIT(";");
4029		break;
4030	}
4031}
4032
4033static int dm_integrity_iterate_devices(struct dm_target *ti,
4034					iterate_devices_callout_fn fn, void *data)
4035{
4036	struct dm_integrity_c *ic = ti->private;
4037
4038	if (!ic->meta_dev)
4039		return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
4040	else
4041		return fn(ti, ic->dev, 0, ti->len, data);
4042}
4043
4044static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
4045{
4046	struct dm_integrity_c *ic = ti->private;
4047
4048	if (ic->sectors_per_block > 1) {
4049		limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
4050		limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
4051		limits->io_min = ic->sectors_per_block << SECTOR_SHIFT;
4052		limits->dma_alignment = limits->logical_block_size - 1;
4053		limits->discard_granularity = ic->sectors_per_block << SECTOR_SHIFT;
4054	}
4055
4056	if (!ic->internal_hash) {
4057		struct blk_integrity *bi = &limits->integrity;
4058
4059		memset(bi, 0, sizeof(*bi));
4060		bi->metadata_size = ic->tag_size;
4061		bi->tag_size = bi->metadata_size;
4062		bi->interval_exp =
4063			ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
4064	}
4065
4066	limits->max_integrity_segments = USHRT_MAX;
4067}
4068
4069static void calculate_journal_section_size(struct dm_integrity_c *ic)
4070{
4071	unsigned int sector_space = JOURNAL_SECTOR_DATA;
4072
4073	ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
4074	ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
4075					 JOURNAL_ENTRY_ROUNDUP);
4076
4077	if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
4078		sector_space -= JOURNAL_MAC_PER_SECTOR;
4079	ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
4080	ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
4081	ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
4082	ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
4083}
4084
4085static int calculate_device_limits(struct dm_integrity_c *ic)
4086{
4087	__u64 initial_sectors;
4088
4089	calculate_journal_section_size(ic);
4090	initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
4091	if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
4092		return -EINVAL;
4093	ic->initial_sectors = initial_sectors;
4094
4095	if (ic->mode == 'I') {
4096		if (ic->initial_sectors + ic->provided_data_sectors > ic->meta_device_sectors)
4097			return -EINVAL;
4098	} else if (!ic->meta_dev) {
4099		sector_t last_sector, last_area, last_offset;
4100
4101		/* we have to maintain excessive padding for compatibility with existing volumes */
4102		__u64 metadata_run_padding =
4103			ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
4104			(__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
4105			(__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
4106
4107		ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
4108					    metadata_run_padding) >> SECTOR_SHIFT;
4109		if (!(ic->metadata_run & (ic->metadata_run - 1)))
4110			ic->log2_metadata_run = __ffs(ic->metadata_run);
4111		else
4112			ic->log2_metadata_run = -1;
4113
4114		get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
4115		last_sector = get_data_sector(ic, last_area, last_offset);
4116		if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
4117			return -EINVAL;
4118	} else {
4119		__u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4120
4121		meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
4122				>> (ic->log2_buffer_sectors + SECTOR_SHIFT);
4123		meta_size <<= ic->log2_buffer_sectors;
4124		if (ic->initial_sectors + meta_size < ic->initial_sectors ||
4125		    ic->initial_sectors + meta_size > ic->meta_device_sectors)
4126			return -EINVAL;
4127		ic->metadata_run = 1;
4128		ic->log2_metadata_run = 0;
4129	}
4130
4131	return 0;
4132}
4133
4134static void get_provided_data_sectors(struct dm_integrity_c *ic)
4135{
4136	if (!ic->meta_dev) {
4137		int test_bit;
4138
4139		ic->provided_data_sectors = 0;
4140		for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
4141			__u64 prev_data_sectors = ic->provided_data_sectors;
4142
4143			ic->provided_data_sectors |= (sector_t)1 << test_bit;
4144			if (calculate_device_limits(ic))
4145				ic->provided_data_sectors = prev_data_sectors;
4146		}
4147	} else {
4148		ic->provided_data_sectors = ic->data_device_sectors;
4149		ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
4150	}
4151}
4152
4153static int initialize_superblock(struct dm_integrity_c *ic,
4154				 unsigned int journal_sectors, unsigned int interleave_sectors)
4155{
4156	unsigned int journal_sections;
4157	int test_bit;
4158
4159	memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
4160	memcpy(ic->sb->magic, SB_MAGIC, 8);
4161	if (ic->mode == 'I')
4162		ic->sb->flags |= cpu_to_le32(SB_FLAG_INLINE);
4163	ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
4164	ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
4165	if (ic->journal_mac_alg.alg_string)
4166		ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
4167
4168	calculate_journal_section_size(ic);
4169	journal_sections = journal_sectors / ic->journal_section_sectors;
4170	if (!journal_sections)
4171		journal_sections = 1;
4172	if (ic->mode == 'I')
4173		journal_sections = 0;
4174
4175	if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
4176		ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
4177		get_random_bytes(ic->sb->salt, SALT_SIZE);
4178	}
4179
4180	if (!ic->meta_dev) {
4181		if (ic->fix_padding)
4182			ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
4183		ic->sb->journal_sections = cpu_to_le32(journal_sections);
4184		if (!interleave_sectors)
4185			interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4186		ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
4187		ic->sb->log2_interleave_sectors = max_t(__u8, MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
4188		ic->sb->log2_interleave_sectors = min_t(__u8, MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
4189
4190		get_provided_data_sectors(ic);
4191		if (!ic->provided_data_sectors)
4192			return -EINVAL;
4193	} else {
4194		ic->sb->log2_interleave_sectors = 0;
4195
4196		get_provided_data_sectors(ic);
4197		if (!ic->provided_data_sectors)
4198			return -EINVAL;
4199
4200try_smaller_buffer:
4201		ic->sb->journal_sections = cpu_to_le32(0);
4202		for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
4203			__u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
4204			__u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
4205
4206			if (test_journal_sections > journal_sections)
4207				continue;
4208			ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
4209			if (calculate_device_limits(ic))
4210				ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
4211
4212		}
4213		if (!le32_to_cpu(ic->sb->journal_sections)) {
4214			if (ic->log2_buffer_sectors > 3) {
4215				ic->log2_buffer_sectors--;
4216				goto try_smaller_buffer;
4217			}
4218			return -EINVAL;
4219		}
4220	}
4221
4222	ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
4223
4224	sb_set_version(ic);
4225
4226	return 0;
4227}
4228
4229static void dm_integrity_free_page_list(struct page_list *pl)
4230{
4231	unsigned int i;
4232
4233	if (!pl)
4234		return;
4235	for (i = 0; pl[i].page; i++)
4236		__free_page(pl[i].page);
4237	kvfree(pl);
4238}
4239
4240static struct page_list *dm_integrity_alloc_page_list(unsigned int n_pages)
4241{
4242	struct page_list *pl;
4243	unsigned int i;
4244
4245	pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
4246	if (!pl)
4247		return NULL;
4248
4249	for (i = 0; i < n_pages; i++) {
4250		pl[i].page = alloc_page(GFP_KERNEL);
4251		if (!pl[i].page) {
4252			dm_integrity_free_page_list(pl);
4253			return NULL;
4254		}
4255		if (i)
4256			pl[i - 1].next = &pl[i];
4257	}
4258	pl[i].page = NULL;
4259	pl[i].next = NULL;
4260
4261	return pl;
4262}
4263
4264static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
4265{
4266	unsigned int i;
4267
4268	for (i = 0; i < ic->journal_sections; i++)
4269		kvfree(sl[i]);
4270	kvfree(sl);
4271}
4272
4273static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
4274								   struct page_list *pl)
4275{
4276	struct scatterlist **sl;
4277	unsigned int i;
4278
4279	sl = kvmalloc_array(ic->journal_sections,
4280			    sizeof(struct scatterlist *),
4281			    GFP_KERNEL | __GFP_ZERO);
4282	if (!sl)
4283		return NULL;
4284
4285	for (i = 0; i < ic->journal_sections; i++) {
4286		struct scatterlist *s;
4287		unsigned int start_index, start_offset;
4288		unsigned int end_index, end_offset;
4289		unsigned int n_pages;
4290		unsigned int idx;
4291
4292		page_list_location(ic, i, 0, &start_index, &start_offset);
4293		page_list_location(ic, i, ic->journal_section_sectors - 1,
4294				   &end_index, &end_offset);
4295
4296		n_pages = (end_index - start_index + 1);
4297
4298		s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
4299				   GFP_KERNEL);
4300		if (!s) {
4301			dm_integrity_free_journal_scatterlist(ic, sl);
4302			return NULL;
4303		}
4304
4305		sg_init_table(s, n_pages);
4306		for (idx = start_index; idx <= end_index; idx++) {
4307			char *va = lowmem_page_address(pl[idx].page);
4308			unsigned int start = 0, end = PAGE_SIZE;
4309
4310			if (idx == start_index)
4311				start = start_offset;
4312			if (idx == end_index)
4313				end = end_offset + (1 << SECTOR_SHIFT);
4314			sg_set_buf(&s[idx - start_index], va + start, end - start);
4315		}
4316
4317		sl[i] = s;
4318	}
4319
4320	return sl;
4321}
4322
4323static void free_alg(struct alg_spec *a)
4324{
4325	kfree_sensitive(a->alg_string);
4326	kfree_sensitive(a->key);
4327	memset(a, 0, sizeof(*a));
4328}
4329
4330static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
4331{
4332	char *k;
4333
4334	free_alg(a);
4335
4336	a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
4337	if (!a->alg_string)
4338		goto nomem;
4339
4340	k = strchr(a->alg_string, ':');
4341	if (k) {
4342		*k = 0;
4343		a->key_string = k + 1;
4344		if (strlen(a->key_string) & 1)
4345			goto inval;
4346
4347		a->key_size = strlen(a->key_string) / 2;
4348		a->key = kmalloc(a->key_size, GFP_KERNEL);
4349		if (!a->key)
4350			goto nomem;
4351		if (hex2bin(a->key, a->key_string, a->key_size))
4352			goto inval;
4353	}
4354
4355	return 0;
4356inval:
4357	*error = error_inval;
4358	return -EINVAL;
4359nomem:
4360	*error = "Out of memory for an argument";
4361	return -ENOMEM;
4362}
4363
4364static int get_mac(struct crypto_shash **shash, struct crypto_ahash **ahash,
4365		   struct alg_spec *a, char **error, char *error_alg, char *error_key)
4366{
4367	int r;
4368
4369	if (a->alg_string) {
4370		if (shash) {
4371			*shash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
4372			if (IS_ERR(*shash)) {
4373				*shash = NULL;
4374				goto try_ahash;
4375			}
4376			if (a->key) {
4377				r = crypto_shash_setkey(*shash, a->key, a->key_size);
4378				if (r) {
4379					*error = error_key;
4380					return r;
4381				}
4382			} else if (crypto_shash_get_flags(*shash) & CRYPTO_TFM_NEED_KEY) {
4383				*error = error_key;
4384				return -ENOKEY;
4385			}
4386			return 0;
4387		}
4388try_ahash:
4389		if (ahash) {
4390			*ahash = crypto_alloc_ahash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
4391			if (IS_ERR(*ahash)) {
4392				*error = error_alg;
4393				r = PTR_ERR(*ahash);
4394				*ahash = NULL;
4395				return r;
4396			}
4397			if (a->key) {
4398				r = crypto_ahash_setkey(*ahash, a->key, a->key_size);
4399				if (r) {
4400					*error = error_key;
4401					return r;
4402				}
4403			} else if (crypto_ahash_get_flags(*ahash) & CRYPTO_TFM_NEED_KEY) {
4404				*error = error_key;
4405				return -ENOKEY;
4406			}
4407			return 0;
4408		}
4409		*error = error_alg;
4410		return -ENOENT;
4411	}
4412
4413	return 0;
4414}
4415
4416static int create_journal(struct dm_integrity_c *ic, char **error)
4417{
4418	int r = 0;
4419	unsigned int i;
4420	__u64 journal_pages, journal_desc_size, journal_tree_size;
4421	unsigned char *crypt_data = NULL, *crypt_iv = NULL;
4422	struct skcipher_request *req = NULL;
4423
4424	ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
4425	ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
4426	ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
4427	ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
4428
4429	journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
4430				PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
4431	journal_desc_size = journal_pages * sizeof(struct page_list);
4432	if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
4433		*error = "Journal doesn't fit into memory";
4434		r = -ENOMEM;
4435		goto bad;
4436	}
4437	ic->journal_pages = journal_pages;
4438
4439	ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
4440	if (!ic->journal) {
4441		*error = "Could not allocate memory for journal";
4442		r = -ENOMEM;
4443		goto bad;
4444	}
4445	if (ic->journal_crypt_alg.alg_string) {
4446		unsigned int ivsize, blocksize;
4447		struct journal_completion comp;
4448
4449		comp.ic = ic;
4450		ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
4451		if (IS_ERR(ic->journal_crypt)) {
4452			*error = "Invalid journal cipher";
4453			r = PTR_ERR(ic->journal_crypt);
4454			ic->journal_crypt = NULL;
4455			goto bad;
4456		}
4457		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
4458		blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
4459
4460		if (ic->journal_crypt_alg.key) {
4461			r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
4462						   ic->journal_crypt_alg.key_size);
4463			if (r) {
4464				*error = "Error setting encryption key";
4465				goto bad;
4466			}
4467		}
4468		DEBUG_print("cipher %s, block size %u iv size %u\n",
4469			    ic->journal_crypt_alg.alg_string, blocksize, ivsize);
4470
4471		ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
4472		if (!ic->journal_io) {
4473			*error = "Could not allocate memory for journal io";
4474			r = -ENOMEM;
4475			goto bad;
4476		}
4477
4478		if (blocksize == 1) {
4479			struct scatterlist *sg;
4480
4481			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
4482			if (!req) {
4483				*error = "Could not allocate crypt request";
4484				r = -ENOMEM;
4485				goto bad;
4486			}
4487
4488			crypt_iv = kzalloc(ivsize, GFP_KERNEL);
4489			if (!crypt_iv) {
4490				*error = "Could not allocate iv";
4491				r = -ENOMEM;
4492				goto bad;
4493			}
4494
4495			ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
4496			if (!ic->journal_xor) {
4497				*error = "Could not allocate memory for journal xor";
4498				r = -ENOMEM;
4499				goto bad;
4500			}
4501
4502			sg = kvmalloc_array(ic->journal_pages + 1,
4503					    sizeof(struct scatterlist),
4504					    GFP_KERNEL);
4505			if (!sg) {
4506				*error = "Unable to allocate sg list";
4507				r = -ENOMEM;
4508				goto bad;
4509			}
4510			sg_init_table(sg, ic->journal_pages + 1);
4511			for (i = 0; i < ic->journal_pages; i++) {
4512				char *va = lowmem_page_address(ic->journal_xor[i].page);
4513
4514				clear_page(va);
4515				sg_set_buf(&sg[i], va, PAGE_SIZE);
4516			}
4517			sg_set_buf(&sg[i], &ic->commit_ids, sizeof(ic->commit_ids));
4518
4519			skcipher_request_set_crypt(req, sg, sg,
4520						   PAGE_SIZE * ic->journal_pages + sizeof(ic->commit_ids), crypt_iv);
4521			init_completion(&comp.comp);
4522			comp.in_flight = (atomic_t)ATOMIC_INIT(1);
4523			if (do_crypt(true, req, &comp))
4524				wait_for_completion(&comp.comp);
4525			kvfree(sg);
4526			r = dm_integrity_failed(ic);
4527			if (r) {
4528				*error = "Unable to encrypt journal";
4529				goto bad;
4530			}
4531			DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
4532
4533			crypto_free_skcipher(ic->journal_crypt);
4534			ic->journal_crypt = NULL;
4535		} else {
4536			unsigned int crypt_len = roundup(ivsize, blocksize);
4537
4538			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
4539			if (!req) {
4540				*error = "Could not allocate crypt request";
4541				r = -ENOMEM;
4542				goto bad;
4543			}
4544
4545			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
4546			if (!crypt_iv) {
4547				*error = "Could not allocate iv";
4548				r = -ENOMEM;
4549				goto bad;
4550			}
4551
4552			crypt_data = kmalloc(crypt_len, GFP_KERNEL);
4553			if (!crypt_data) {
4554				*error = "Unable to allocate crypt data";
4555				r = -ENOMEM;
4556				goto bad;
4557			}
4558
4559			ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
4560			if (!ic->journal_scatterlist) {
4561				*error = "Unable to allocate sg list";
4562				r = -ENOMEM;
4563				goto bad;
4564			}
4565			ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
4566			if (!ic->journal_io_scatterlist) {
4567				*error = "Unable to allocate sg list";
4568				r = -ENOMEM;
4569				goto bad;
4570			}
4571			ic->sk_requests = kvmalloc_array(ic->journal_sections,
4572							 sizeof(struct skcipher_request *),
4573							 GFP_KERNEL | __GFP_ZERO);
4574			if (!ic->sk_requests) {
4575				*error = "Unable to allocate sk requests";
4576				r = -ENOMEM;
4577				goto bad;
4578			}
4579			for (i = 0; i < ic->journal_sections; i++) {
4580				struct scatterlist sg;
4581				struct skcipher_request *section_req;
4582				__le32 section_le = cpu_to_le32(i);
4583
4584				memset(crypt_iv, 0x00, ivsize);
4585				memset(crypt_data, 0x00, crypt_len);
4586				memcpy(crypt_data, &section_le, min_t(size_t, crypt_len, sizeof(section_le)));
4587
4588				sg_init_one(&sg, crypt_data, crypt_len);
4589				skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
4590				init_completion(&comp.comp);
4591				comp.in_flight = (atomic_t)ATOMIC_INIT(1);
4592				if (do_crypt(true, req, &comp))
4593					wait_for_completion(&comp.comp);
4594
4595				r = dm_integrity_failed(ic);
4596				if (r) {
4597					*error = "Unable to generate iv";
4598					goto bad;
4599				}
4600
4601				section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
4602				if (!section_req) {
4603					*error = "Unable to allocate crypt request";
4604					r = -ENOMEM;
4605					goto bad;
4606				}
4607				section_req->iv = kmalloc_array(ivsize, 2,
4608								GFP_KERNEL);
4609				if (!section_req->iv) {
4610					skcipher_request_free(section_req);
4611					*error = "Unable to allocate iv";
4612					r = -ENOMEM;
4613					goto bad;
4614				}
4615				memcpy(section_req->iv + ivsize, crypt_data, ivsize);
4616				section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
4617				ic->sk_requests[i] = section_req;
4618				DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
4619			}
4620		}
4621	}
4622
4623	for (i = 0; i < N_COMMIT_IDS; i++) {
4624		unsigned int j;
4625
4626retest_commit_id:
4627		for (j = 0; j < i; j++) {
4628			if (ic->commit_ids[j] == ic->commit_ids[i]) {
4629				ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
4630				goto retest_commit_id;
4631			}
4632		}
4633		DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
4634	}
4635
4636	journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
4637	if (journal_tree_size > ULONG_MAX) {
4638		*error = "Journal doesn't fit into memory";
4639		r = -ENOMEM;
4640		goto bad;
4641	}
4642	ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
4643	if (!ic->journal_tree) {
4644		*error = "Could not allocate memory for journal tree";
4645		r = -ENOMEM;
4646	}
4647bad:
4648	kfree(crypt_data);
4649	kfree(crypt_iv);
4650	skcipher_request_free(req);
4651
4652	return r;
4653}
4654
4655/*
4656 * Construct a integrity mapping
4657 *
4658 * Arguments:
4659 *	device
4660 *	offset from the start of the device
4661 *	tag size
4662 *	D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
4663 *	number of optional arguments
4664 *	optional arguments:
4665 *		journal_sectors
4666 *		interleave_sectors
4667 *		buffer_sectors
4668 *		journal_watermark
4669 *		commit_time
4670 *		meta_device
4671 *		block_size
4672 *		sectors_per_bit
4673 *		bitmap_flush_interval
4674 *		internal_hash
4675 *		journal_crypt
4676 *		journal_mac
4677 *		recalculate
4678 */
4679static int dm_integrity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4680{
4681	struct dm_integrity_c *ic;
4682	char dummy;
4683	int r;
4684	unsigned int extra_args;
4685	struct dm_arg_set as;
4686	static const struct dm_arg _args[] = {
4687		{0, 18, "Invalid number of feature args"},
4688	};
4689	unsigned int journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
4690	bool should_write_sb;
4691	__u64 threshold;
4692	unsigned long long start;
4693	__s8 log2_sectors_per_bitmap_bit = -1;
4694	__s8 log2_blocks_per_bitmap_bit;
4695	__u64 bits_in_journal;
4696	__u64 n_bitmap_bits;
4697
4698#define DIRECT_ARGUMENTS	4
4699
4700	if (argc <= DIRECT_ARGUMENTS) {
4701		ti->error = "Invalid argument count";
4702		return -EINVAL;
4703	}
4704
4705	ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
4706	if (!ic) {
4707		ti->error = "Cannot allocate integrity context";
4708		return -ENOMEM;
4709	}
4710	ti->private = ic;
4711	ti->per_io_data_size = sizeof(struct dm_integrity_io);
4712	ic->ti = ti;
4713
4714	ic->in_progress = RB_ROOT;
4715	INIT_LIST_HEAD(&ic->wait_list);
4716	init_waitqueue_head(&ic->endio_wait);
4717	bio_list_init(&ic->flush_bio_list);
4718	init_waitqueue_head(&ic->copy_to_journal_wait);
4719	init_completion(&ic->crypto_backoff);
4720	atomic64_set(&ic->number_of_mismatches, 0);
4721	ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4722
4723	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4724	if (r) {
4725		ti->error = "Device lookup failed";
4726		goto bad;
4727	}
4728
4729	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4730		ti->error = "Invalid starting offset";
4731		r = -EINVAL;
4732		goto bad;
4733	}
4734	ic->start = start;
4735
4736	if (strcmp(argv[2], "-")) {
4737		if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4738			ti->error = "Invalid tag size";
4739			r = -EINVAL;
4740			goto bad;
4741		}
4742	}
4743
4744	if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4745	    !strcmp(argv[3], "D") || !strcmp(argv[3], "R") ||
4746	    !strcmp(argv[3], "I")) {
4747		ic->mode = argv[3][0];
4748	} else {
4749		ti->error = "Invalid mode (expecting J, B, D, R, I)";
4750		r = -EINVAL;
4751		goto bad;
4752	}
4753
4754	journal_sectors = 0;
4755	interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4756	buffer_sectors = DEFAULT_BUFFER_SECTORS;
4757	journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4758	sync_msec = DEFAULT_SYNC_MSEC;
4759	ic->sectors_per_block = 1;
4760
4761	as.argc = argc - DIRECT_ARGUMENTS;
4762	as.argv = argv + DIRECT_ARGUMENTS;
4763	r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4764	if (r)
4765		goto bad;
4766
4767	while (extra_args--) {
4768		const char *opt_string;
4769		unsigned int val;
4770		unsigned long long llval;
4771
4772		opt_string = dm_shift_arg(&as);
4773		if (!opt_string) {
4774			r = -EINVAL;
4775			ti->error = "Not enough feature arguments";
4776			goto bad;
4777		}
4778		if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4779			journal_sectors = val ? val : 1;
4780		else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4781			interleave_sectors = val;
4782		else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4783			buffer_sectors = val;
4784		else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4785			journal_watermark = val;
4786		else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4787			sync_msec = val;
4788		else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4789			if (ic->meta_dev) {
4790				dm_put_device(ti, ic->meta_dev);
4791				ic->meta_dev = NULL;
4792			}
4793			r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4794					  dm_table_get_mode(ti->table), &ic->meta_dev);
4795			if (r) {
4796				ti->error = "Device lookup failed";
4797				goto bad;
4798			}
4799		} else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4800			if (val < 1 << SECTOR_SHIFT ||
4801			    val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4802			    (val & (val - 1))) {
4803				r = -EINVAL;
4804				ti->error = "Invalid block_size argument";
4805				goto bad;
4806			}
4807			ic->sectors_per_block = val >> SECTOR_SHIFT;
4808		} else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4809			log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4810		} else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4811			if ((uint64_t)val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4812				r = -EINVAL;
4813				ti->error = "Invalid bitmap_flush_interval argument";
4814				goto bad;
4815			}
4816			ic->bitmap_flush_interval = msecs_to_jiffies(val);
4817		} else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4818			r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4819					    "Invalid internal_hash argument");
4820			if (r)
4821				goto bad;
4822		} else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4823			r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4824					    "Invalid journal_crypt argument");
4825			if (r)
4826				goto bad;
4827		} else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4828			r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4829					    "Invalid journal_mac argument");
4830			if (r)
4831				goto bad;
4832		} else if (!strcmp(opt_string, "recalculate")) {
4833			ic->recalculate_flag = true;
4834		} else if (!strcmp(opt_string, "reset_recalculate")) {
4835			ic->recalculate_flag = true;
4836			ic->reset_recalculate_flag = true;
4837		} else if (!strcmp(opt_string, "allow_discards")) {
4838			ic->discard = true;
4839		} else if (!strcmp(opt_string, "fix_padding")) {
4840			ic->fix_padding = true;
4841		} else if (!strcmp(opt_string, "fix_hmac")) {
4842			ic->fix_hmac = true;
4843		} else if (!strcmp(opt_string, "legacy_recalculate")) {
4844			ic->legacy_recalculate = true;
4845		} else {
4846			r = -EINVAL;
4847			ti->error = "Invalid argument";
4848			goto bad;
4849		}
4850	}
4851
4852	ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4853	if (!ic->meta_dev)
4854		ic->meta_device_sectors = ic->data_device_sectors;
4855	else
4856		ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4857
4858	if (!journal_sectors) {
4859		journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4860				      ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4861	}
4862
4863	if (!buffer_sectors)
4864		buffer_sectors = 1;
4865	ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4866
4867	r = get_mac(&ic->internal_shash, &ic->internal_ahash, &ic->internal_hash_alg, &ti->error,
4868		    "Invalid internal hash", "Error setting internal hash key");
4869	if (r)
4870		goto bad;
4871	if (ic->internal_shash) {
4872		ic->internal_hash = true;
4873		ic->internal_hash_digestsize = crypto_shash_digestsize(ic->internal_shash);
4874	}
4875	if (ic->internal_ahash) {
4876		ic->internal_hash = true;
4877		ic->internal_hash_digestsize = crypto_ahash_digestsize(ic->internal_ahash);
4878		r = mempool_init_kmalloc_pool(&ic->ahash_req_pool, AHASH_MEMPOOL,
4879					      sizeof(struct ahash_request) + crypto_ahash_reqsize(ic->internal_ahash));
4880		if (r) {
4881			ti->error = "Cannot allocate mempool";
4882			goto bad;
4883		}
4884	}
4885
4886	r = get_mac(&ic->journal_mac, NULL, &ic->journal_mac_alg, &ti->error,
4887		    "Invalid journal mac", "Error setting journal mac key");
4888	if (r)
4889		goto bad;
4890
4891	if (!ic->tag_size) {
4892		if (!ic->internal_hash) {
4893			ti->error = "Unknown tag size";
4894			r = -EINVAL;
4895			goto bad;
4896		}
4897		ic->tag_size = ic->internal_hash_digestsize;
4898	}
4899	if (ic->tag_size > MAX_TAG_SIZE) {
4900		ti->error = "Too big tag size";
4901		r = -EINVAL;
4902		goto bad;
4903	}
4904	if (!(ic->tag_size & (ic->tag_size - 1)))
4905		ic->log2_tag_size = __ffs(ic->tag_size);
4906	else
4907		ic->log2_tag_size = -1;
4908
4909	if (ic->mode == 'I') {
4910		struct blk_integrity *bi;
4911		if (ic->meta_dev) {
4912			r = -EINVAL;
4913			ti->error = "Metadata device not supported in inline mode";
4914			goto bad;
4915		}
4916		if (!ic->internal_hash_alg.alg_string) {
4917			r = -EINVAL;
4918			ti->error = "Internal hash not set in inline mode";
4919			goto bad;
4920		}
4921		if (ic->journal_crypt_alg.alg_string || ic->journal_mac_alg.alg_string) {
4922			r = -EINVAL;
4923			ti->error = "Journal crypt not supported in inline mode";
4924			goto bad;
4925		}
4926		if (ic->discard) {
4927			r = -EINVAL;
4928			ti->error = "Discards not supported in inline mode";
4929			goto bad;
4930		}
4931		bi = blk_get_integrity(ic->dev->bdev->bd_disk);
4932		if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
4933			r = -EINVAL;
4934			ti->error = "Integrity profile not supported";
4935			goto bad;
4936		}
4937		/*printk("tag_size: %u, metadata_size: %u\n", bi->tag_size, bi->metadata_size);*/
4938		if (bi->metadata_size < ic->tag_size) {
4939			r = -EINVAL;
4940			ti->error = "The integrity profile is smaller than tag size";
4941			goto bad;
4942		}
4943		if ((unsigned long)bi->metadata_size > PAGE_SIZE / 2) {
4944			r = -EINVAL;
4945			ti->error = "Too big tuple size";
4946			goto bad;
4947		}
4948		ic->tuple_size = bi->metadata_size;
4949		if (1 << bi->interval_exp != ic->sectors_per_block << SECTOR_SHIFT) {
4950			r = -EINVAL;
4951			ti->error = "Integrity profile sector size mismatch";
4952			goto bad;
4953		}
4954	}
4955
4956	if (ic->mode == 'B' && !ic->internal_hash) {
4957		r = -EINVAL;
4958		ti->error = "Bitmap mode can be only used with internal hash";
4959		goto bad;
4960	}
4961
4962	if (ic->discard && !ic->internal_hash) {
4963		r = -EINVAL;
4964		ti->error = "Discard can be only used with internal hash";
4965		goto bad;
4966	}
4967
4968	ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4969	ic->autocommit_msec = sync_msec;
4970	timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4971
4972	ic->io = dm_io_client_create();
4973	if (IS_ERR(ic->io)) {
4974		r = PTR_ERR(ic->io);
4975		ic->io = NULL;
4976		ti->error = "Cannot allocate dm io";
4977		goto bad;
4978	}
4979
4980	r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4981	if (r) {
4982		ti->error = "Cannot allocate mempool";
4983		goto bad;
4984	}
4985
4986	r = mempool_init_page_pool(&ic->recheck_pool, 1, ic->mode == 'I' ? 1 : 0);
4987	if (r) {
4988		ti->error = "Cannot allocate mempool";
4989		goto bad;
4990	}
4991
4992	if (ic->mode == 'I') {
4993		r = bioset_init(&ic->recheck_bios, RECHECK_POOL_SIZE, 0, BIOSET_NEED_BVECS);
4994		if (r) {
4995			ti->error = "Cannot allocate bio set";
4996			goto bad;
4997		}
4998		r = bioset_init(&ic->recalc_bios, 1, 0, BIOSET_NEED_BVECS);
4999		if (r) {
5000			ti->error = "Cannot allocate bio set";
5001			goto bad;
5002		}
5003	}
5004
5005	ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
5006					  WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
5007	if (!ic->metadata_wq) {
5008		ti->error = "Cannot allocate workqueue";
5009		r = -ENOMEM;
5010		goto bad;
5011	}
5012
5013	/*
5014	 * If this workqueue weren't ordered, it would cause bio reordering
5015	 * and reduced performance.
5016	 */
5017	ic->wait_wq = alloc_ordered_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM);
5018	if (!ic->wait_wq) {
5019		ti->error = "Cannot allocate workqueue";
5020		r = -ENOMEM;
5021		goto bad;
5022	}
5023
5024	ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
5025					  METADATA_WORKQUEUE_MAX_ACTIVE);
5026	if (!ic->offload_wq) {
5027		ti->error = "Cannot allocate workqueue";
5028		r = -ENOMEM;
5029		goto bad;
5030	}
5031
5032	ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
5033	if (!ic->commit_wq) {
5034		ti->error = "Cannot allocate workqueue";
5035		r = -ENOMEM;
5036		goto bad;
5037	}
5038	INIT_WORK(&ic->commit_work, integrity_commit);
5039
5040	if (ic->mode == 'J' || ic->mode == 'B') {
5041		ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
5042		if (!ic->writer_wq) {
5043			ti->error = "Cannot allocate workqueue";
5044			r = -ENOMEM;
5045			goto bad;
5046		}
5047		INIT_WORK(&ic->writer_work, integrity_writer);
5048	}
5049
5050	ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
5051	if (!ic->sb) {
5052		r = -ENOMEM;
5053		ti->error = "Cannot allocate superblock area";
5054		goto bad;
5055	}
5056
5057	r = sync_rw_sb(ic, REQ_OP_READ);
5058	if (r) {
5059		ti->error = "Error reading superblock";
5060		goto bad;
5061	}
5062	should_write_sb = false;
5063	if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
5064		if (ic->mode != 'R') {
5065			if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
5066				r = -EINVAL;
5067				ti->error = "The device is not initialized";
5068				goto bad;
5069			}
5070		}
5071
5072		r = initialize_superblock(ic, journal_sectors, interleave_sectors);
5073		if (r) {
5074			ti->error = "Could not initialize superblock";
5075			goto bad;
5076		}
5077		if (ic->mode != 'R')
5078			should_write_sb = true;
5079	}
5080
5081	if (!ic->sb->version || ic->sb->version > SB_VERSION_6) {
5082		r = -EINVAL;
5083		ti->error = "Unknown version";
5084		goto bad;
5085	}
5086	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_INLINE)) != (ic->mode == 'I')) {
5087		r = -EINVAL;
5088		ti->error = "Inline flag mismatch";
5089		goto bad;
5090	}
5091	if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
5092		r = -EINVAL;
5093		ti->error = "Tag size doesn't match the information in superblock";
5094		goto bad;
5095	}
5096	if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
5097		r = -EINVAL;
5098		ti->error = "Block size doesn't match the information in superblock";
5099		goto bad;
5100	}
5101	if (ic->mode != 'I') {
5102		if (!le32_to_cpu(ic->sb->journal_sections)) {
5103			r = -EINVAL;
5104			ti->error = "Corrupted superblock, journal_sections is 0";
5105			goto bad;
5106		}
5107	} else {
5108		if (le32_to_cpu(ic->sb->journal_sections)) {
5109			r = -EINVAL;
5110			ti->error = "Corrupted superblock, journal_sections is not 0";
5111			goto bad;
5112		}
5113	}
5114	/* make sure that ti->max_io_len doesn't overflow */
5115	if (!ic->meta_dev) {
5116		if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
5117		    ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
5118			r = -EINVAL;
5119			ti->error = "Invalid interleave_sectors in the superblock";
5120			goto bad;
5121		}
5122	} else {
5123		if (ic->sb->log2_interleave_sectors) {
5124			r = -EINVAL;
5125			ti->error = "Invalid interleave_sectors in the superblock";
5126			goto bad;
5127		}
5128	}
5129	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
5130		r = -EINVAL;
5131		ti->error = "Journal mac mismatch";
5132		goto bad;
5133	}
5134
5135	get_provided_data_sectors(ic);
5136	if (!ic->provided_data_sectors) {
5137		r = -EINVAL;
5138		ti->error = "The device is too small";
5139		goto bad;
5140	}
5141
5142try_smaller_buffer:
5143	r = calculate_device_limits(ic);
5144	if (r) {
5145		if (ic->meta_dev) {
5146			if (ic->log2_buffer_sectors > 3) {
5147				ic->log2_buffer_sectors--;
5148				goto try_smaller_buffer;
5149			}
5150		}
5151		ti->error = "The device is too small";
5152		goto bad;
5153	}
5154
5155	if (log2_sectors_per_bitmap_bit < 0)
5156		log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
5157	if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
5158		log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
5159
5160	bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
5161	if (bits_in_journal > UINT_MAX)
5162		bits_in_journal = UINT_MAX;
5163	if (bits_in_journal)
5164		while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
5165			log2_sectors_per_bitmap_bit++;
5166
5167	log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
5168	ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
5169	if (should_write_sb)
5170		ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
5171
5172	n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
5173				+ (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
5174	ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
5175
5176	if (!ic->meta_dev)
5177		ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
5178
5179	if (ti->len > ic->provided_data_sectors) {
5180		r = -EINVAL;
5181		ti->error = "Not enough provided sectors for requested mapping size";
5182		goto bad;
5183	}
5184
5185	threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
5186	threshold += 50;
5187	do_div(threshold, 100);
5188	ic->free_sectors_threshold = threshold;
5189
5190	DEBUG_print("initialized:\n");
5191	DEBUG_print("	integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
5192	DEBUG_print("	journal_entry_size %u\n", ic->journal_entry_size);
5193	DEBUG_print("	journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
5194	DEBUG_print("	journal_section_entries %u\n", ic->journal_section_entries);
5195	DEBUG_print("	journal_section_sectors %u\n", ic->journal_section_sectors);
5196	DEBUG_print("	journal_sections %u\n", (unsigned int)le32_to_cpu(ic->sb->journal_sections));
5197	DEBUG_print("	journal_entries %u\n", ic->journal_entries);
5198	DEBUG_print("	log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
5199	DEBUG_print("	data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
5200	DEBUG_print("	initial_sectors 0x%x\n", ic->initial_sectors);
5201	DEBUG_print("	metadata_run 0x%x\n", ic->metadata_run);
5202	DEBUG_print("	log2_metadata_run %d\n", ic->log2_metadata_run);
5203	DEBUG_print("	provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
5204	DEBUG_print("	log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
5205	DEBUG_print("	bits_in_journal %llu\n", bits_in_journal);
5206
5207	if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
5208		ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
5209		ic->sb->recalc_sector = cpu_to_le64(0);
5210	}
5211
5212	if (ic->internal_hash) {
5213		ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
5214		if (!ic->recalc_wq) {
5215			ti->error = "Cannot allocate workqueue";
5216			r = -ENOMEM;
5217			goto bad;
5218		}
5219		INIT_WORK(&ic->recalc_work, ic->mode == 'I' ? integrity_recalc_inline : integrity_recalc);
5220	} else {
5221		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
5222			ti->error = "Recalculate can only be specified with internal_hash";
5223			r = -EINVAL;
5224			goto bad;
5225		}
5226	}
5227
5228	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
5229	    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
5230	    dm_integrity_disable_recalculate(ic)) {
5231		ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
5232		r = -EOPNOTSUPP;
5233		goto bad;
5234	}
5235
5236	ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
5237			1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
5238	if (IS_ERR(ic->bufio)) {
5239		r = PTR_ERR(ic->bufio);
5240		ti->error = "Cannot initialize dm-bufio";
5241		ic->bufio = NULL;
5242		goto bad;
5243	}
5244	dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
5245
5246	if (ic->mode != 'R' && ic->mode != 'I') {
5247		r = create_journal(ic, &ti->error);
5248		if (r)
5249			goto bad;
5250
5251	}
5252
5253	if (ic->mode == 'B') {
5254		unsigned int i;
5255		unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
5256
5257		ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
5258		if (!ic->recalc_bitmap) {
5259			ti->error = "Could not allocate memory for bitmap";
5260			r = -ENOMEM;
5261			goto bad;
5262		}
5263		ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
5264		if (!ic->may_write_bitmap) {
5265			ti->error = "Could not allocate memory for bitmap";
5266			r = -ENOMEM;
5267			goto bad;
5268		}
5269		ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
5270		if (!ic->bbs) {
5271			ti->error = "Could not allocate memory for bitmap";
5272			r = -ENOMEM;
5273			goto bad;
5274		}
5275		INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
5276		for (i = 0; i < ic->n_bitmap_blocks; i++) {
5277			struct bitmap_block_status *bbs = &ic->bbs[i];
5278			unsigned int sector, pl_index, pl_offset;
5279
5280			INIT_WORK(&bbs->work, bitmap_block_work);
5281			bbs->ic = ic;
5282			bbs->idx = i;
5283			bio_list_init(&bbs->bio_queue);
5284			spin_lock_init(&bbs->bio_queue_lock);
5285
5286			sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
5287			pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
5288			pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
5289
5290			bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
5291		}
5292	}
5293
5294	if (should_write_sb) {
5295		init_journal(ic, 0, ic->journal_sections, 0);
5296		r = dm_integrity_failed(ic);
5297		if (unlikely(r)) {
5298			ti->error = "Error initializing journal";
5299			goto bad;
5300		}
5301		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
5302		if (r) {
5303			ti->error = "Error initializing superblock";
5304			goto bad;
5305		}
5306		ic->just_formatted = true;
5307	}
5308
5309	if (!ic->meta_dev && ic->mode != 'I') {
5310		r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
5311		if (r)
5312			goto bad;
5313	}
5314	if (ic->mode == 'B') {
5315		unsigned int max_io_len;
5316
5317		max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
5318		if (!max_io_len)
5319			max_io_len = 1U << 31;
5320		DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
5321		if (!ti->max_io_len || ti->max_io_len > max_io_len) {
5322			r = dm_set_target_max_io_len(ti, max_io_len);
5323			if (r)
5324				goto bad;
5325		}
5326	}
5327
5328	ti->num_flush_bios = 1;
5329	ti->flush_supported = true;
5330	if (ic->discard)
5331		ti->num_discard_bios = 1;
5332
5333	if (ic->mode == 'I')
5334		ti->mempool_needs_integrity = true;
5335
5336	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
5337	return 0;
5338
5339bad:
5340	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
5341	dm_integrity_dtr(ti);
5342	return r;
5343}
5344
5345static void dm_integrity_dtr(struct dm_target *ti)
5346{
5347	struct dm_integrity_c *ic = ti->private;
5348
5349	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
5350	BUG_ON(!list_empty(&ic->wait_list));
5351
5352	if (ic->mode == 'B' && ic->bitmap_flush_work.work.func)
5353		cancel_delayed_work_sync(&ic->bitmap_flush_work);
5354	if (ic->metadata_wq)
5355		destroy_workqueue(ic->metadata_wq);
5356	if (ic->wait_wq)
5357		destroy_workqueue(ic->wait_wq);
5358	if (ic->offload_wq)
5359		destroy_workqueue(ic->offload_wq);
5360	if (ic->commit_wq)
5361		destroy_workqueue(ic->commit_wq);
5362	if (ic->writer_wq)
5363		destroy_workqueue(ic->writer_wq);
5364	if (ic->recalc_wq)
5365		destroy_workqueue(ic->recalc_wq);
5366	kvfree(ic->bbs);
5367	if (ic->bufio)
5368		dm_bufio_client_destroy(ic->bufio);
5369	mempool_free(ic->journal_ahash_req, &ic->ahash_req_pool);
5370	mempool_exit(&ic->ahash_req_pool);
5371	bioset_exit(&ic->recalc_bios);
5372	bioset_exit(&ic->recheck_bios);
5373	mempool_exit(&ic->recheck_pool);
5374	mempool_exit(&ic->journal_io_mempool);
5375	if (ic->io)
5376		dm_io_client_destroy(ic->io);
5377	if (ic->dev)
5378		dm_put_device(ti, ic->dev);
5379	if (ic->meta_dev)
5380		dm_put_device(ti, ic->meta_dev);
5381	dm_integrity_free_page_list(ic->journal);
5382	dm_integrity_free_page_list(ic->journal_io);
5383	dm_integrity_free_page_list(ic->journal_xor);
5384	dm_integrity_free_page_list(ic->recalc_bitmap);
5385	dm_integrity_free_page_list(ic->may_write_bitmap);
5386	if (ic->journal_scatterlist)
5387		dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
5388	if (ic->journal_io_scatterlist)
5389		dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
5390	if (ic->sk_requests) {
5391		unsigned int i;
5392
5393		for (i = 0; i < ic->journal_sections; i++) {
5394			struct skcipher_request *req;
5395
5396			req = ic->sk_requests[i];
5397			if (req) {
5398				kfree_sensitive(req->iv);
5399				skcipher_request_free(req);
5400			}
5401		}
5402		kvfree(ic->sk_requests);
5403	}
5404	kvfree(ic->journal_tree);
5405	if (ic->sb)
5406		free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
5407
5408	if (ic->internal_shash)
5409		crypto_free_shash(ic->internal_shash);
5410	if (ic->internal_ahash)
5411		crypto_free_ahash(ic->internal_ahash);
5412	free_alg(&ic->internal_hash_alg);
5413
5414	if (ic->journal_crypt)
5415		crypto_free_skcipher(ic->journal_crypt);
5416	free_alg(&ic->journal_crypt_alg);
5417
5418	if (ic->journal_mac)
5419		crypto_free_shash(ic->journal_mac);
5420	free_alg(&ic->journal_mac_alg);
5421
5422	kfree(ic);
5423	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
5424}
5425
5426static struct target_type integrity_target = {
5427	.name			= "integrity",
5428	.version		= {1, 14, 0},
5429	.module			= THIS_MODULE,
5430	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
5431	.ctr			= dm_integrity_ctr,
5432	.dtr			= dm_integrity_dtr,
5433	.map			= dm_integrity_map,
5434	.end_io			= dm_integrity_end_io,
5435	.postsuspend		= dm_integrity_postsuspend,
5436	.resume			= dm_integrity_resume,
5437	.status			= dm_integrity_status,
5438	.iterate_devices	= dm_integrity_iterate_devices,
5439	.io_hints		= dm_integrity_io_hints,
5440};
5441
5442static int __init dm_integrity_init(void)
5443{
5444	int r;
5445
5446	journal_io_cache = kmem_cache_create("integrity_journal_io",
5447					     sizeof(struct journal_io), 0, 0, NULL);
5448	if (!journal_io_cache) {
5449		DMERR("can't allocate journal io cache");
5450		return -ENOMEM;
5451	}
5452
5453	r = dm_register_target(&integrity_target);
5454	if (r < 0) {
5455		kmem_cache_destroy(journal_io_cache);
5456		return r;
5457	}
5458
5459	return 0;
5460}
5461
5462static void __exit dm_integrity_exit(void)
5463{
5464	dm_unregister_target(&integrity_target);
5465	kmem_cache_destroy(journal_io_cache);
5466}
5467
5468module_init(dm_integrity_init);
5469module_exit(dm_integrity_exit);
5470
5471MODULE_AUTHOR("Milan Broz");
5472MODULE_AUTHOR("Mikulas Patocka");
5473MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
5474MODULE_LICENSE("GPL");