Linux Audio

Check our new training course

Loading...
v6.2
   1/*
   2 * Copyright (C) 2003 Jana Saout <jana@saout.de>
   3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
   4 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
   5 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
   6 *
   7 * This file is released under the GPL.
   8 */
   9
  10#include <linux/completion.h>
  11#include <linux/err.h>
  12#include <linux/module.h>
  13#include <linux/init.h>
  14#include <linux/kernel.h>
  15#include <linux/key.h>
  16#include <linux/bio.h>
  17#include <linux/blkdev.h>
  18#include <linux/blk-integrity.h>
  19#include <linux/mempool.h>
  20#include <linux/slab.h>
  21#include <linux/crypto.h>
  22#include <linux/workqueue.h>
  23#include <linux/kthread.h>
  24#include <linux/backing-dev.h>
  25#include <linux/atomic.h>
  26#include <linux/scatterlist.h>
  27#include <linux/rbtree.h>
  28#include <linux/ctype.h>
  29#include <asm/page.h>
  30#include <asm/unaligned.h>
  31#include <crypto/hash.h>
  32#include <crypto/md5.h>
  33#include <crypto/algapi.h>
  34#include <crypto/skcipher.h>
  35#include <crypto/aead.h>
  36#include <crypto/authenc.h>
  37#include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
  38#include <linux/key-type.h>
  39#include <keys/user-type.h>
  40#include <keys/encrypted-type.h>
  41#include <keys/trusted-type.h>
  42
  43#include <linux/device-mapper.h>
  44
  45#include "dm-audit.h"
  46
  47#define DM_MSG_PREFIX "crypt"
  48
  49/*
  50 * context holding the current state of a multi-part conversion
  51 */
  52struct convert_context {
  53	struct completion restart;
  54	struct bio *bio_in;
  55	struct bio *bio_out;
  56	struct bvec_iter iter_in;
  57	struct bvec_iter iter_out;
  58	u64 cc_sector;
  59	atomic_t cc_pending;
  60	union {
  61		struct skcipher_request *req;
  62		struct aead_request *req_aead;
  63	} r;
  64
  65};
  66
  67/*
  68 * per bio private data
  69 */
  70struct dm_crypt_io {
  71	struct crypt_config *cc;
  72	struct bio *base_bio;
  73	u8 *integrity_metadata;
  74	bool integrity_metadata_from_pool;
  75	struct work_struct work;
  76	struct tasklet_struct tasklet;
  77
  78	struct convert_context ctx;
  79
  80	atomic_t io_pending;
  81	blk_status_t error;
  82	sector_t sector;
  83
  84	struct rb_node rb_node;
  85} CRYPTO_MINALIGN_ATTR;
  86
  87struct dm_crypt_request {
  88	struct convert_context *ctx;
  89	struct scatterlist sg_in[4];
  90	struct scatterlist sg_out[4];
  91	u64 iv_sector;
  92};
  93
  94struct crypt_config;
  95
  96struct crypt_iv_operations {
  97	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
  98		   const char *opts);
  99	void (*dtr)(struct crypt_config *cc);
 100	int (*init)(struct crypt_config *cc);
 101	int (*wipe)(struct crypt_config *cc);
 102	int (*generator)(struct crypt_config *cc, u8 *iv,
 103			 struct dm_crypt_request *dmreq);
 104	int (*post)(struct crypt_config *cc, u8 *iv,
 105		    struct dm_crypt_request *dmreq);
 106};
 107
 108struct iv_benbi_private {
 109	int shift;
 110};
 111
 112#define LMK_SEED_SIZE 64 /* hash + 0 */
 113struct iv_lmk_private {
 114	struct crypto_shash *hash_tfm;
 115	u8 *seed;
 116};
 117
 118#define TCW_WHITENING_SIZE 16
 119struct iv_tcw_private {
 120	struct crypto_shash *crc32_tfm;
 121	u8 *iv_seed;
 122	u8 *whitening;
 123};
 124
 125#define ELEPHANT_MAX_KEY_SIZE 32
 126struct iv_elephant_private {
 127	struct crypto_skcipher *tfm;
 128};
 129
 130/*
 131 * Crypt: maps a linear range of a block device
 132 * and encrypts / decrypts at the same time.
 133 */
 134enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
 135	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
 136	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
 137	     DM_CRYPT_WRITE_INLINE };
 138
 139enum cipher_flags {
 140	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
 141	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
 142	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
 143};
 144
 145/*
 146 * The fields in here must be read only after initialization.
 147 */
 148struct crypt_config {
 149	struct dm_dev *dev;
 150	sector_t start;
 151
 152	struct percpu_counter n_allocated_pages;
 153
 154	struct workqueue_struct *io_queue;
 155	struct workqueue_struct *crypt_queue;
 156
 157	spinlock_t write_thread_lock;
 158	struct task_struct *write_thread;
 159	struct rb_root write_tree;
 160
 161	char *cipher_string;
 162	char *cipher_auth;
 163	char *key_string;
 164
 165	const struct crypt_iv_operations *iv_gen_ops;
 166	union {
 167		struct iv_benbi_private benbi;
 168		struct iv_lmk_private lmk;
 169		struct iv_tcw_private tcw;
 170		struct iv_elephant_private elephant;
 171	} iv_gen_private;
 172	u64 iv_offset;
 173	unsigned int iv_size;
 174	unsigned short int sector_size;
 175	unsigned char sector_shift;
 176
 177	union {
 178		struct crypto_skcipher **tfms;
 179		struct crypto_aead **tfms_aead;
 180	} cipher_tfm;
 181	unsigned tfms_count;
 182	unsigned long cipher_flags;
 183
 184	/*
 185	 * Layout of each crypto request:
 186	 *
 187	 *   struct skcipher_request
 188	 *      context
 189	 *      padding
 190	 *   struct dm_crypt_request
 191	 *      padding
 192	 *   IV
 193	 *
 194	 * The padding is added so that dm_crypt_request and the IV are
 195	 * correctly aligned.
 196	 */
 197	unsigned int dmreq_start;
 198
 199	unsigned int per_bio_data_size;
 200
 201	unsigned long flags;
 202	unsigned int key_size;
 203	unsigned int key_parts;      /* independent parts in key buffer */
 204	unsigned int key_extra_size; /* additional keys length */
 205	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
 206
 207	unsigned int integrity_tag_size;
 208	unsigned int integrity_iv_size;
 209	unsigned int on_disk_tag_size;
 210
 211	/*
 212	 * pool for per bio private data, crypto requests,
 213	 * encryption requeusts/buffer pages and integrity tags
 214	 */
 215	unsigned tag_pool_max_sectors;
 216	mempool_t tag_pool;
 217	mempool_t req_pool;
 218	mempool_t page_pool;
 219
 220	struct bio_set bs;
 221	struct mutex bio_alloc_lock;
 222
 223	u8 *authenc_key; /* space for keys in authenc() format (if used) */
 224	u8 key[];
 225};
 226
 227#define MIN_IOS		64
 228#define MAX_TAG_SIZE	480
 229#define POOL_ENTRY_SIZE	512
 230
 231static DEFINE_SPINLOCK(dm_crypt_clients_lock);
 232static unsigned dm_crypt_clients_n = 0;
 233static volatile unsigned long dm_crypt_pages_per_client;
 234#define DM_CRYPT_MEMORY_PERCENT			2
 235#define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
 236
 237static void crypt_endio(struct bio *clone);
 238static void kcryptd_queue_crypt(struct dm_crypt_io *io);
 239static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
 240					     struct scatterlist *sg);
 241
 242static bool crypt_integrity_aead(struct crypt_config *cc);
 243
 244/*
 245 * Use this to access cipher attributes that are independent of the key.
 246 */
 247static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
 248{
 249	return cc->cipher_tfm.tfms[0];
 250}
 251
 252static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
 253{
 254	return cc->cipher_tfm.tfms_aead[0];
 255}
 256
 257/*
 258 * Different IV generation algorithms:
 259 *
 260 * plain: the initial vector is the 32-bit little-endian version of the sector
 261 *        number, padded with zeros if necessary.
 262 *
 263 * plain64: the initial vector is the 64-bit little-endian version of the sector
 264 *        number, padded with zeros if necessary.
 265 *
 266 * plain64be: the initial vector is the 64-bit big-endian version of the sector
 267 *        number, padded with zeros if necessary.
 268 *
 269 * essiv: "encrypted sector|salt initial vector", the sector number is
 270 *        encrypted with the bulk cipher using a salt as key. The salt
 271 *        should be derived from the bulk cipher's key via hashing.
 272 *
 273 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
 274 *        (needed for LRW-32-AES and possible other narrow block modes)
 275 *
 276 * null: the initial vector is always zero.  Provides compatibility with
 277 *       obsolete loop_fish2 devices.  Do not use for new devices.
 278 *
 279 * lmk:  Compatible implementation of the block chaining mode used
 280 *       by the Loop-AES block device encryption system
 281 *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
 282 *       It operates on full 512 byte sectors and uses CBC
 283 *       with an IV derived from the sector number, the data and
 284 *       optionally extra IV seed.
 285 *       This means that after decryption the first block
 286 *       of sector must be tweaked according to decrypted data.
 287 *       Loop-AES can use three encryption schemes:
 288 *         version 1: is plain aes-cbc mode
 289 *         version 2: uses 64 multikey scheme with lmk IV generator
 290 *         version 3: the same as version 2 with additional IV seed
 291 *                   (it uses 65 keys, last key is used as IV seed)
 292 *
 293 * tcw:  Compatible implementation of the block chaining mode used
 294 *       by the TrueCrypt device encryption system (prior to version 4.1).
 295 *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
 296 *       It operates on full 512 byte sectors and uses CBC
 297 *       with an IV derived from initial key and the sector number.
 298 *       In addition, whitening value is applied on every sector, whitening
 299 *       is calculated from initial key, sector number and mixed using CRC32.
 300 *       Note that this encryption scheme is vulnerable to watermarking attacks
 301 *       and should be used for old compatible containers access only.
 302 *
 303 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
 304 *        The IV is encrypted little-endian byte-offset (with the same key
 305 *        and cipher as the volume).
 306 *
 307 * elephant: The extended version of eboiv with additional Elephant diffuser
 308 *           used with Bitlocker CBC mode.
 309 *           This mode was used in older Windows systems
 310 *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
 311 */
 312
 313static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
 314			      struct dm_crypt_request *dmreq)
 315{
 316	memset(iv, 0, cc->iv_size);
 317	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
 318
 319	return 0;
 320}
 321
 322static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
 323				struct dm_crypt_request *dmreq)
 324{
 325	memset(iv, 0, cc->iv_size);
 326	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 327
 328	return 0;
 329}
 330
 331static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
 332				  struct dm_crypt_request *dmreq)
 333{
 334	memset(iv, 0, cc->iv_size);
 335	/* iv_size is at least of size u64; usually it is 16 bytes */
 336	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
 337
 338	return 0;
 339}
 340
 341static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
 342			      struct dm_crypt_request *dmreq)
 343{
 344	/*
 345	 * ESSIV encryption of the IV is now handled by the crypto API,
 346	 * so just pass the plain sector number here.
 347	 */
 348	memset(iv, 0, cc->iv_size);
 349	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 350
 351	return 0;
 352}
 353
 354static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
 355			      const char *opts)
 356{
 357	unsigned bs;
 358	int log;
 359
 360	if (crypt_integrity_aead(cc))
 361		bs = crypto_aead_blocksize(any_tfm_aead(cc));
 362	else
 363		bs = crypto_skcipher_blocksize(any_tfm(cc));
 364	log = ilog2(bs);
 365
 366	/* we need to calculate how far we must shift the sector count
 367	 * to get the cipher block count, we use this shift in _gen */
 368
 369	if (1 << log != bs) {
 370		ti->error = "cypher blocksize is not a power of 2";
 371		return -EINVAL;
 372	}
 373
 374	if (log > 9) {
 375		ti->error = "cypher blocksize is > 512";
 376		return -EINVAL;
 377	}
 378
 379	cc->iv_gen_private.benbi.shift = 9 - log;
 380
 381	return 0;
 382}
 383
 384static void crypt_iv_benbi_dtr(struct crypt_config *cc)
 385{
 386}
 387
 388static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
 389			      struct dm_crypt_request *dmreq)
 390{
 391	__be64 val;
 392
 393	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
 394
 395	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
 396	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
 397
 398	return 0;
 399}
 400
 401static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
 402			     struct dm_crypt_request *dmreq)
 403{
 404	memset(iv, 0, cc->iv_size);
 405
 406	return 0;
 407}
 408
 409static void crypt_iv_lmk_dtr(struct crypt_config *cc)
 410{
 411	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 412
 413	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
 414		crypto_free_shash(lmk->hash_tfm);
 415	lmk->hash_tfm = NULL;
 416
 417	kfree_sensitive(lmk->seed);
 418	lmk->seed = NULL;
 419}
 420
 421static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
 422			    const char *opts)
 423{
 424	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 425
 426	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 427		ti->error = "Unsupported sector size for LMK";
 428		return -EINVAL;
 429	}
 430
 431	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
 432					   CRYPTO_ALG_ALLOCATES_MEMORY);
 433	if (IS_ERR(lmk->hash_tfm)) {
 434		ti->error = "Error initializing LMK hash";
 435		return PTR_ERR(lmk->hash_tfm);
 436	}
 437
 438	/* No seed in LMK version 2 */
 439	if (cc->key_parts == cc->tfms_count) {
 440		lmk->seed = NULL;
 441		return 0;
 442	}
 443
 444	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
 445	if (!lmk->seed) {
 446		crypt_iv_lmk_dtr(cc);
 447		ti->error = "Error kmallocing seed storage in LMK";
 448		return -ENOMEM;
 449	}
 450
 451	return 0;
 452}
 453
 454static int crypt_iv_lmk_init(struct crypt_config *cc)
 455{
 456	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 457	int subkey_size = cc->key_size / cc->key_parts;
 458
 459	/* LMK seed is on the position of LMK_KEYS + 1 key */
 460	if (lmk->seed)
 461		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
 462		       crypto_shash_digestsize(lmk->hash_tfm));
 463
 464	return 0;
 465}
 466
 467static int crypt_iv_lmk_wipe(struct crypt_config *cc)
 468{
 469	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 470
 471	if (lmk->seed)
 472		memset(lmk->seed, 0, LMK_SEED_SIZE);
 473
 474	return 0;
 475}
 476
 477static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
 478			    struct dm_crypt_request *dmreq,
 479			    u8 *data)
 480{
 481	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 482	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
 483	struct md5_state md5state;
 484	__le32 buf[4];
 485	int i, r;
 486
 487	desc->tfm = lmk->hash_tfm;
 488
 489	r = crypto_shash_init(desc);
 490	if (r)
 491		return r;
 492
 493	if (lmk->seed) {
 494		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
 495		if (r)
 496			return r;
 497	}
 498
 499	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
 500	r = crypto_shash_update(desc, data + 16, 16 * 31);
 501	if (r)
 502		return r;
 503
 504	/* Sector is cropped to 56 bits here */
 505	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
 506	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
 507	buf[2] = cpu_to_le32(4024);
 508	buf[3] = 0;
 509	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
 510	if (r)
 511		return r;
 512
 513	/* No MD5 padding here */
 514	r = crypto_shash_export(desc, &md5state);
 515	if (r)
 516		return r;
 517
 518	for (i = 0; i < MD5_HASH_WORDS; i++)
 519		__cpu_to_le32s(&md5state.hash[i]);
 520	memcpy(iv, &md5state.hash, cc->iv_size);
 521
 522	return 0;
 523}
 524
 525static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
 526			    struct dm_crypt_request *dmreq)
 527{
 528	struct scatterlist *sg;
 529	u8 *src;
 530	int r = 0;
 531
 532	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 533		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 534		src = kmap_atomic(sg_page(sg));
 535		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
 536		kunmap_atomic(src);
 537	} else
 538		memset(iv, 0, cc->iv_size);
 539
 540	return r;
 541}
 542
 543static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
 544			     struct dm_crypt_request *dmreq)
 545{
 546	struct scatterlist *sg;
 547	u8 *dst;
 548	int r;
 549
 550	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
 551		return 0;
 552
 553	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 554	dst = kmap_atomic(sg_page(sg));
 555	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
 556
 557	/* Tweak the first block of plaintext sector */
 558	if (!r)
 559		crypto_xor(dst + sg->offset, iv, cc->iv_size);
 560
 561	kunmap_atomic(dst);
 562	return r;
 563}
 564
 565static void crypt_iv_tcw_dtr(struct crypt_config *cc)
 566{
 567	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 568
 569	kfree_sensitive(tcw->iv_seed);
 570	tcw->iv_seed = NULL;
 571	kfree_sensitive(tcw->whitening);
 572	tcw->whitening = NULL;
 573
 574	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
 575		crypto_free_shash(tcw->crc32_tfm);
 576	tcw->crc32_tfm = NULL;
 577}
 578
 579static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
 580			    const char *opts)
 581{
 582	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 583
 584	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 585		ti->error = "Unsupported sector size for TCW";
 586		return -EINVAL;
 587	}
 588
 589	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
 590		ti->error = "Wrong key size for TCW";
 591		return -EINVAL;
 592	}
 593
 594	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
 595					    CRYPTO_ALG_ALLOCATES_MEMORY);
 596	if (IS_ERR(tcw->crc32_tfm)) {
 597		ti->error = "Error initializing CRC32 in TCW";
 598		return PTR_ERR(tcw->crc32_tfm);
 599	}
 600
 601	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
 602	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
 603	if (!tcw->iv_seed || !tcw->whitening) {
 604		crypt_iv_tcw_dtr(cc);
 605		ti->error = "Error allocating seed storage in TCW";
 606		return -ENOMEM;
 607	}
 608
 609	return 0;
 610}
 611
 612static int crypt_iv_tcw_init(struct crypt_config *cc)
 613{
 614	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 615	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
 616
 617	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
 618	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
 619	       TCW_WHITENING_SIZE);
 620
 621	return 0;
 622}
 623
 624static int crypt_iv_tcw_wipe(struct crypt_config *cc)
 625{
 626	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 627
 628	memset(tcw->iv_seed, 0, cc->iv_size);
 629	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
 630
 631	return 0;
 632}
 633
 634static int crypt_iv_tcw_whitening(struct crypt_config *cc,
 635				  struct dm_crypt_request *dmreq,
 636				  u8 *data)
 637{
 638	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 639	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 640	u8 buf[TCW_WHITENING_SIZE];
 641	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
 642	int i, r;
 643
 644	/* xor whitening with sector number */
 645	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
 646	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
 647
 648	/* calculate crc32 for every 32bit part and xor it */
 649	desc->tfm = tcw->crc32_tfm;
 650	for (i = 0; i < 4; i++) {
 651		r = crypto_shash_init(desc);
 652		if (r)
 653			goto out;
 654		r = crypto_shash_update(desc, &buf[i * 4], 4);
 655		if (r)
 656			goto out;
 657		r = crypto_shash_final(desc, &buf[i * 4]);
 658		if (r)
 659			goto out;
 660	}
 661	crypto_xor(&buf[0], &buf[12], 4);
 662	crypto_xor(&buf[4], &buf[8], 4);
 663
 664	/* apply whitening (8 bytes) to whole sector */
 665	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
 666		crypto_xor(data + i * 8, buf, 8);
 667out:
 668	memzero_explicit(buf, sizeof(buf));
 669	return r;
 670}
 671
 672static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
 673			    struct dm_crypt_request *dmreq)
 674{
 675	struct scatterlist *sg;
 676	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 677	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 678	u8 *src;
 679	int r = 0;
 680
 681	/* Remove whitening from ciphertext */
 682	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
 683		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 684		src = kmap_atomic(sg_page(sg));
 685		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
 686		kunmap_atomic(src);
 687	}
 688
 689	/* Calculate IV */
 690	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
 691	if (cc->iv_size > 8)
 692		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
 693			       cc->iv_size - 8);
 694
 695	return r;
 696}
 697
 698static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
 699			     struct dm_crypt_request *dmreq)
 700{
 701	struct scatterlist *sg;
 702	u8 *dst;
 703	int r;
 704
 705	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
 706		return 0;
 707
 708	/* Apply whitening on ciphertext */
 709	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 710	dst = kmap_atomic(sg_page(sg));
 711	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
 712	kunmap_atomic(dst);
 713
 714	return r;
 715}
 716
 717static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
 718				struct dm_crypt_request *dmreq)
 719{
 720	/* Used only for writes, there must be an additional space to store IV */
 721	get_random_bytes(iv, cc->iv_size);
 722	return 0;
 723}
 724
 725static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
 726			    const char *opts)
 727{
 728	if (crypt_integrity_aead(cc)) {
 729		ti->error = "AEAD transforms not supported for EBOIV";
 730		return -EINVAL;
 731	}
 732
 733	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
 734		ti->error = "Block size of EBOIV cipher does "
 735			    "not match IV size of block cipher";
 736		return -EINVAL;
 737	}
 738
 739	return 0;
 740}
 741
 742static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
 743			    struct dm_crypt_request *dmreq)
 744{
 745	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
 746	struct skcipher_request *req;
 747	struct scatterlist src, dst;
 748	DECLARE_CRYPTO_WAIT(wait);
 749	int err;
 750
 751	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
 752	if (!req)
 753		return -ENOMEM;
 754
 755	memset(buf, 0, cc->iv_size);
 756	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
 757
 758	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
 759	sg_init_one(&dst, iv, cc->iv_size);
 760	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
 761	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
 762	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 763	skcipher_request_free(req);
 764
 765	return err;
 766}
 767
 768static void crypt_iv_elephant_dtr(struct crypt_config *cc)
 769{
 770	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 771
 772	crypto_free_skcipher(elephant->tfm);
 773	elephant->tfm = NULL;
 774}
 775
 776static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
 777			    const char *opts)
 778{
 779	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 780	int r;
 781
 782	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
 783					      CRYPTO_ALG_ALLOCATES_MEMORY);
 784	if (IS_ERR(elephant->tfm)) {
 785		r = PTR_ERR(elephant->tfm);
 786		elephant->tfm = NULL;
 787		return r;
 788	}
 789
 790	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
 791	if (r)
 792		crypt_iv_elephant_dtr(cc);
 793	return r;
 794}
 795
 796static void diffuser_disk_to_cpu(u32 *d, size_t n)
 797{
 798#ifndef __LITTLE_ENDIAN
 799	int i;
 800
 801	for (i = 0; i < n; i++)
 802		d[i] = le32_to_cpu((__le32)d[i]);
 803#endif
 804}
 805
 806static void diffuser_cpu_to_disk(__le32 *d, size_t n)
 807{
 808#ifndef __LITTLE_ENDIAN
 809	int i;
 810
 811	for (i = 0; i < n; i++)
 812		d[i] = cpu_to_le32((u32)d[i]);
 813#endif
 814}
 815
 816static void diffuser_a_decrypt(u32 *d, size_t n)
 817{
 818	int i, i1, i2, i3;
 819
 820	for (i = 0; i < 5; i++) {
 821		i1 = 0;
 822		i2 = n - 2;
 823		i3 = n - 5;
 824
 825		while (i1 < (n - 1)) {
 826			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
 827			i1++; i2++; i3++;
 828
 829			if (i3 >= n)
 830				i3 -= n;
 831
 832			d[i1] += d[i2] ^ d[i3];
 833			i1++; i2++; i3++;
 834
 835			if (i2 >= n)
 836				i2 -= n;
 837
 838			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
 839			i1++; i2++; i3++;
 840
 841			d[i1] += d[i2] ^ d[i3];
 842			i1++; i2++; i3++;
 843		}
 844	}
 845}
 846
 847static void diffuser_a_encrypt(u32 *d, size_t n)
 848{
 849	int i, i1, i2, i3;
 850
 851	for (i = 0; i < 5; i++) {
 852		i1 = n - 1;
 853		i2 = n - 2 - 1;
 854		i3 = n - 5 - 1;
 855
 856		while (i1 > 0) {
 857			d[i1] -= d[i2] ^ d[i3];
 858			i1--; i2--; i3--;
 859
 860			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
 861			i1--; i2--; i3--;
 862
 863			if (i2 < 0)
 864				i2 += n;
 865
 866			d[i1] -= d[i2] ^ d[i3];
 867			i1--; i2--; i3--;
 868
 869			if (i3 < 0)
 870				i3 += n;
 871
 872			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
 873			i1--; i2--; i3--;
 874		}
 875	}
 876}
 877
 878static void diffuser_b_decrypt(u32 *d, size_t n)
 879{
 880	int i, i1, i2, i3;
 881
 882	for (i = 0; i < 3; i++) {
 883		i1 = 0;
 884		i2 = 2;
 885		i3 = 5;
 886
 887		while (i1 < (n - 1)) {
 888			d[i1] += d[i2] ^ d[i3];
 889			i1++; i2++; i3++;
 890
 891			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
 892			i1++; i2++; i3++;
 893
 894			if (i2 >= n)
 895				i2 -= n;
 896
 897			d[i1] += d[i2] ^ d[i3];
 898			i1++; i2++; i3++;
 899
 900			if (i3 >= n)
 901				i3 -= n;
 902
 903			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
 904			i1++; i2++; i3++;
 905		}
 906	}
 907}
 908
 909static void diffuser_b_encrypt(u32 *d, size_t n)
 910{
 911	int i, i1, i2, i3;
 912
 913	for (i = 0; i < 3; i++) {
 914		i1 = n - 1;
 915		i2 = 2 - 1;
 916		i3 = 5 - 1;
 917
 918		while (i1 > 0) {
 919			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
 920			i1--; i2--; i3--;
 921
 922			if (i3 < 0)
 923				i3 += n;
 924
 925			d[i1] -= d[i2] ^ d[i3];
 926			i1--; i2--; i3--;
 927
 928			if (i2 < 0)
 929				i2 += n;
 930
 931			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
 932			i1--; i2--; i3--;
 933
 934			d[i1] -= d[i2] ^ d[i3];
 935			i1--; i2--; i3--;
 936		}
 937	}
 938}
 939
 940static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
 941{
 942	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 943	u8 *es, *ks, *data, *data2, *data_offset;
 944	struct skcipher_request *req;
 945	struct scatterlist *sg, *sg2, src, dst;
 946	DECLARE_CRYPTO_WAIT(wait);
 947	int i, r;
 948
 949	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
 950	es = kzalloc(16, GFP_NOIO); /* Key for AES */
 951	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
 952
 953	if (!req || !es || !ks) {
 954		r = -ENOMEM;
 955		goto out;
 956	}
 957
 958	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
 959
 960	/* E(Ks, e(s)) */
 961	sg_init_one(&src, es, 16);
 962	sg_init_one(&dst, ks, 16);
 963	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
 964	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
 965	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 966	if (r)
 967		goto out;
 968
 969	/* E(Ks, e'(s)) */
 970	es[15] = 0x80;
 971	sg_init_one(&dst, &ks[16], 16);
 972	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 973	if (r)
 974		goto out;
 975
 976	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 977	data = kmap_atomic(sg_page(sg));
 978	data_offset = data + sg->offset;
 979
 980	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
 981	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 982		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
 983		data2 = kmap_atomic(sg_page(sg2));
 984		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
 985		kunmap_atomic(data2);
 986	}
 987
 988	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
 989		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
 990		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 991		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 992		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
 993	}
 994
 995	for (i = 0; i < (cc->sector_size / 32); i++)
 996		crypto_xor(data_offset + i * 32, ks, 32);
 997
 998	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 999		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
1000		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1001		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
1002		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1003	}
1004
1005	kunmap_atomic(data);
1006out:
1007	kfree_sensitive(ks);
1008	kfree_sensitive(es);
1009	skcipher_request_free(req);
1010	return r;
1011}
1012
1013static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1014			    struct dm_crypt_request *dmreq)
1015{
1016	int r;
1017
1018	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1019		r = crypt_iv_elephant(cc, dmreq);
1020		if (r)
1021			return r;
1022	}
1023
1024	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1025}
1026
1027static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1028				  struct dm_crypt_request *dmreq)
1029{
1030	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1031		return crypt_iv_elephant(cc, dmreq);
1032
1033	return 0;
1034}
1035
1036static int crypt_iv_elephant_init(struct crypt_config *cc)
1037{
1038	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1039	int key_offset = cc->key_size - cc->key_extra_size;
1040
1041	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1042}
1043
1044static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1045{
1046	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1047	u8 key[ELEPHANT_MAX_KEY_SIZE];
1048
1049	memset(key, 0, cc->key_extra_size);
1050	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1051}
1052
1053static const struct crypt_iv_operations crypt_iv_plain_ops = {
1054	.generator = crypt_iv_plain_gen
1055};
1056
1057static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1058	.generator = crypt_iv_plain64_gen
1059};
1060
1061static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1062	.generator = crypt_iv_plain64be_gen
1063};
1064
1065static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1066	.generator = crypt_iv_essiv_gen
1067};
1068
1069static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1070	.ctr	   = crypt_iv_benbi_ctr,
1071	.dtr	   = crypt_iv_benbi_dtr,
1072	.generator = crypt_iv_benbi_gen
1073};
1074
1075static const struct crypt_iv_operations crypt_iv_null_ops = {
1076	.generator = crypt_iv_null_gen
1077};
1078
1079static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1080	.ctr	   = crypt_iv_lmk_ctr,
1081	.dtr	   = crypt_iv_lmk_dtr,
1082	.init	   = crypt_iv_lmk_init,
1083	.wipe	   = crypt_iv_lmk_wipe,
1084	.generator = crypt_iv_lmk_gen,
1085	.post	   = crypt_iv_lmk_post
1086};
1087
1088static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1089	.ctr	   = crypt_iv_tcw_ctr,
1090	.dtr	   = crypt_iv_tcw_dtr,
1091	.init	   = crypt_iv_tcw_init,
1092	.wipe	   = crypt_iv_tcw_wipe,
1093	.generator = crypt_iv_tcw_gen,
1094	.post	   = crypt_iv_tcw_post
1095};
1096
1097static const struct crypt_iv_operations crypt_iv_random_ops = {
1098	.generator = crypt_iv_random_gen
1099};
1100
1101static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1102	.ctr	   = crypt_iv_eboiv_ctr,
1103	.generator = crypt_iv_eboiv_gen
1104};
1105
1106static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1107	.ctr	   = crypt_iv_elephant_ctr,
1108	.dtr	   = crypt_iv_elephant_dtr,
1109	.init	   = crypt_iv_elephant_init,
1110	.wipe	   = crypt_iv_elephant_wipe,
1111	.generator = crypt_iv_elephant_gen,
1112	.post	   = crypt_iv_elephant_post
1113};
1114
1115/*
1116 * Integrity extensions
1117 */
1118static bool crypt_integrity_aead(struct crypt_config *cc)
1119{
1120	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1121}
1122
1123static bool crypt_integrity_hmac(struct crypt_config *cc)
1124{
1125	return crypt_integrity_aead(cc) && cc->key_mac_size;
1126}
1127
1128/* Get sg containing data */
1129static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1130					     struct scatterlist *sg)
1131{
1132	if (unlikely(crypt_integrity_aead(cc)))
1133		return &sg[2];
1134
1135	return sg;
1136}
1137
1138static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1139{
1140	struct bio_integrity_payload *bip;
1141	unsigned int tag_len;
1142	int ret;
1143
1144	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1145		return 0;
1146
1147	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1148	if (IS_ERR(bip))
1149		return PTR_ERR(bip);
1150
1151	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1152
1153	bip->bip_iter.bi_size = tag_len;
1154	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1155
1156	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1157				     tag_len, offset_in_page(io->integrity_metadata));
1158	if (unlikely(ret != tag_len))
1159		return -ENOMEM;
1160
1161	return 0;
1162}
1163
1164static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1165{
1166#ifdef CONFIG_BLK_DEV_INTEGRITY
1167	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1168	struct mapped_device *md = dm_table_get_md(ti->table);
1169
1170	/* From now we require underlying device with our integrity profile */
1171	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1172		ti->error = "Integrity profile not supported.";
1173		return -EINVAL;
1174	}
1175
1176	if (bi->tag_size != cc->on_disk_tag_size ||
1177	    bi->tuple_size != cc->on_disk_tag_size) {
1178		ti->error = "Integrity profile tag size mismatch.";
1179		return -EINVAL;
1180	}
1181	if (1 << bi->interval_exp != cc->sector_size) {
1182		ti->error = "Integrity profile sector size mismatch.";
1183		return -EINVAL;
1184	}
1185
1186	if (crypt_integrity_aead(cc)) {
1187		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1188		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1189		       cc->integrity_tag_size, cc->integrity_iv_size);
1190
1191		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1192			ti->error = "Integrity AEAD auth tag size is not supported.";
1193			return -EINVAL;
1194		}
1195	} else if (cc->integrity_iv_size)
1196		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1197		       cc->integrity_iv_size);
1198
1199	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1200		ti->error = "Not enough space for integrity tag in the profile.";
1201		return -EINVAL;
1202	}
1203
1204	return 0;
1205#else
1206	ti->error = "Integrity profile not supported.";
1207	return -EINVAL;
1208#endif
1209}
1210
1211static void crypt_convert_init(struct crypt_config *cc,
1212			       struct convert_context *ctx,
1213			       struct bio *bio_out, struct bio *bio_in,
1214			       sector_t sector)
1215{
1216	ctx->bio_in = bio_in;
1217	ctx->bio_out = bio_out;
1218	if (bio_in)
1219		ctx->iter_in = bio_in->bi_iter;
1220	if (bio_out)
1221		ctx->iter_out = bio_out->bi_iter;
1222	ctx->cc_sector = sector + cc->iv_offset;
1223	init_completion(&ctx->restart);
1224}
1225
1226static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1227					     void *req)
1228{
1229	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1230}
1231
1232static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1233{
1234	return (void *)((char *)dmreq - cc->dmreq_start);
1235}
1236
1237static u8 *iv_of_dmreq(struct crypt_config *cc,
1238		       struct dm_crypt_request *dmreq)
1239{
1240	if (crypt_integrity_aead(cc))
1241		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1242			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1243	else
1244		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1245			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1246}
1247
1248static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1249		       struct dm_crypt_request *dmreq)
1250{
1251	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1252}
1253
1254static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1255		       struct dm_crypt_request *dmreq)
1256{
1257	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1258	return (__le64 *) ptr;
1259}
1260
1261static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1262		       struct dm_crypt_request *dmreq)
1263{
1264	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1265		  cc->iv_size + sizeof(uint64_t);
1266	return (unsigned int*)ptr;
1267}
1268
1269static void *tag_from_dmreq(struct crypt_config *cc,
1270				struct dm_crypt_request *dmreq)
1271{
1272	struct convert_context *ctx = dmreq->ctx;
1273	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1274
1275	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1276		cc->on_disk_tag_size];
1277}
1278
1279static void *iv_tag_from_dmreq(struct crypt_config *cc,
1280			       struct dm_crypt_request *dmreq)
1281{
1282	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1283}
1284
1285static int crypt_convert_block_aead(struct crypt_config *cc,
1286				     struct convert_context *ctx,
1287				     struct aead_request *req,
1288				     unsigned int tag_offset)
1289{
1290	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1291	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1292	struct dm_crypt_request *dmreq;
1293	u8 *iv, *org_iv, *tag_iv, *tag;
1294	__le64 *sector;
1295	int r = 0;
1296
1297	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1298
1299	/* Reject unexpected unaligned bio. */
1300	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1301		return -EIO;
1302
1303	dmreq = dmreq_of_req(cc, req);
1304	dmreq->iv_sector = ctx->cc_sector;
1305	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1306		dmreq->iv_sector >>= cc->sector_shift;
1307	dmreq->ctx = ctx;
1308
1309	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1310
1311	sector = org_sector_of_dmreq(cc, dmreq);
1312	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1313
1314	iv = iv_of_dmreq(cc, dmreq);
1315	org_iv = org_iv_of_dmreq(cc, dmreq);
1316	tag = tag_from_dmreq(cc, dmreq);
1317	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1318
1319	/* AEAD request:
1320	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1321	 *  | (authenticated) | (auth+encryption) |              |
1322	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1323	 */
1324	sg_init_table(dmreq->sg_in, 4);
1325	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1326	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1327	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1328	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1329
1330	sg_init_table(dmreq->sg_out, 4);
1331	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1332	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1333	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1334	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1335
1336	if (cc->iv_gen_ops) {
1337		/* For READs use IV stored in integrity metadata */
1338		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1339			memcpy(org_iv, tag_iv, cc->iv_size);
1340		} else {
1341			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1342			if (r < 0)
1343				return r;
1344			/* Store generated IV in integrity metadata */
1345			if (cc->integrity_iv_size)
1346				memcpy(tag_iv, org_iv, cc->iv_size);
1347		}
1348		/* Working copy of IV, to be modified in crypto API */
1349		memcpy(iv, org_iv, cc->iv_size);
1350	}
1351
1352	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1353	if (bio_data_dir(ctx->bio_in) == WRITE) {
1354		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1355				       cc->sector_size, iv);
1356		r = crypto_aead_encrypt(req);
1357		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1358			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1359			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1360	} else {
1361		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1362				       cc->sector_size + cc->integrity_tag_size, iv);
1363		r = crypto_aead_decrypt(req);
1364	}
1365
1366	if (r == -EBADMSG) {
1367		sector_t s = le64_to_cpu(*sector);
1368
1369		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1370			    ctx->bio_in->bi_bdev, s);
1371		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1372				 ctx->bio_in, s, 0);
1373	}
1374
1375	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1376		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1377
1378	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1379	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1380
1381	return r;
1382}
1383
1384static int crypt_convert_block_skcipher(struct crypt_config *cc,
1385					struct convert_context *ctx,
1386					struct skcipher_request *req,
1387					unsigned int tag_offset)
1388{
1389	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1390	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1391	struct scatterlist *sg_in, *sg_out;
1392	struct dm_crypt_request *dmreq;
1393	u8 *iv, *org_iv, *tag_iv;
1394	__le64 *sector;
1395	int r = 0;
1396
1397	/* Reject unexpected unaligned bio. */
1398	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1399		return -EIO;
1400
1401	dmreq = dmreq_of_req(cc, req);
1402	dmreq->iv_sector = ctx->cc_sector;
1403	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1404		dmreq->iv_sector >>= cc->sector_shift;
1405	dmreq->ctx = ctx;
1406
1407	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1408
1409	iv = iv_of_dmreq(cc, dmreq);
1410	org_iv = org_iv_of_dmreq(cc, dmreq);
1411	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1412
1413	sector = org_sector_of_dmreq(cc, dmreq);
1414	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1415
1416	/* For skcipher we use only the first sg item */
1417	sg_in  = &dmreq->sg_in[0];
1418	sg_out = &dmreq->sg_out[0];
1419
1420	sg_init_table(sg_in, 1);
1421	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1422
1423	sg_init_table(sg_out, 1);
1424	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1425
1426	if (cc->iv_gen_ops) {
1427		/* For READs use IV stored in integrity metadata */
1428		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1429			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1430		} else {
1431			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1432			if (r < 0)
1433				return r;
1434			/* Data can be already preprocessed in generator */
1435			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1436				sg_in = sg_out;
1437			/* Store generated IV in integrity metadata */
1438			if (cc->integrity_iv_size)
1439				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1440		}
1441		/* Working copy of IV, to be modified in crypto API */
1442		memcpy(iv, org_iv, cc->iv_size);
1443	}
1444
1445	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1446
1447	if (bio_data_dir(ctx->bio_in) == WRITE)
1448		r = crypto_skcipher_encrypt(req);
1449	else
1450		r = crypto_skcipher_decrypt(req);
1451
1452	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1453		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1454
1455	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1456	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1457
1458	return r;
1459}
1460
1461static void kcryptd_async_done(struct crypto_async_request *async_req,
1462			       int error);
1463
1464static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1465				     struct convert_context *ctx)
1466{
1467	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1468
1469	if (!ctx->r.req) {
1470		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1471		if (!ctx->r.req)
1472			return -ENOMEM;
1473	}
1474
1475	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1476
1477	/*
1478	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1479	 * requests if driver request queue is full.
1480	 */
1481	skcipher_request_set_callback(ctx->r.req,
1482	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1483	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1484
1485	return 0;
1486}
1487
1488static int crypt_alloc_req_aead(struct crypt_config *cc,
1489				 struct convert_context *ctx)
1490{
1491	if (!ctx->r.req_aead) {
1492		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1493		if (!ctx->r.req_aead)
1494			return -ENOMEM;
1495	}
1496
1497	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1498
1499	/*
1500	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1501	 * requests if driver request queue is full.
1502	 */
1503	aead_request_set_callback(ctx->r.req_aead,
1504	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1505	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1506
1507	return 0;
1508}
1509
1510static int crypt_alloc_req(struct crypt_config *cc,
1511			    struct convert_context *ctx)
1512{
1513	if (crypt_integrity_aead(cc))
1514		return crypt_alloc_req_aead(cc, ctx);
1515	else
1516		return crypt_alloc_req_skcipher(cc, ctx);
1517}
1518
1519static void crypt_free_req_skcipher(struct crypt_config *cc,
1520				    struct skcipher_request *req, struct bio *base_bio)
1521{
1522	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1523
1524	if ((struct skcipher_request *)(io + 1) != req)
1525		mempool_free(req, &cc->req_pool);
1526}
1527
1528static void crypt_free_req_aead(struct crypt_config *cc,
1529				struct aead_request *req, struct bio *base_bio)
1530{
1531	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1532
1533	if ((struct aead_request *)(io + 1) != req)
1534		mempool_free(req, &cc->req_pool);
1535}
1536
1537static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1538{
1539	if (crypt_integrity_aead(cc))
1540		crypt_free_req_aead(cc, req, base_bio);
1541	else
1542		crypt_free_req_skcipher(cc, req, base_bio);
1543}
1544
1545/*
1546 * Encrypt / decrypt data from one bio to another one (can be the same one)
1547 */
1548static blk_status_t crypt_convert(struct crypt_config *cc,
1549			 struct convert_context *ctx, bool atomic, bool reset_pending)
1550{
1551	unsigned int tag_offset = 0;
1552	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1553	int r;
1554
1555	/*
1556	 * if reset_pending is set we are dealing with the bio for the first time,
1557	 * else we're continuing to work on the previous bio, so don't mess with
1558	 * the cc_pending counter
1559	 */
1560	if (reset_pending)
1561		atomic_set(&ctx->cc_pending, 1);
1562
1563	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1564
1565		r = crypt_alloc_req(cc, ctx);
1566		if (r) {
1567			complete(&ctx->restart);
1568			return BLK_STS_DEV_RESOURCE;
1569		}
1570
1571		atomic_inc(&ctx->cc_pending);
1572
1573		if (crypt_integrity_aead(cc))
1574			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1575		else
1576			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1577
1578		switch (r) {
1579		/*
1580		 * The request was queued by a crypto driver
1581		 * but the driver request queue is full, let's wait.
1582		 */
1583		case -EBUSY:
1584			if (in_interrupt()) {
1585				if (try_wait_for_completion(&ctx->restart)) {
1586					/*
1587					 * we don't have to block to wait for completion,
1588					 * so proceed
1589					 */
1590				} else {
1591					/*
1592					 * we can't wait for completion without blocking
1593					 * exit and continue processing in a workqueue
1594					 */
1595					ctx->r.req = NULL;
1596					ctx->cc_sector += sector_step;
1597					tag_offset++;
1598					return BLK_STS_DEV_RESOURCE;
1599				}
1600			} else {
1601				wait_for_completion(&ctx->restart);
1602			}
1603			reinit_completion(&ctx->restart);
1604			fallthrough;
1605		/*
1606		 * The request is queued and processed asynchronously,
1607		 * completion function kcryptd_async_done() will be called.
1608		 */
1609		case -EINPROGRESS:
1610			ctx->r.req = NULL;
1611			ctx->cc_sector += sector_step;
1612			tag_offset++;
1613			continue;
1614		/*
1615		 * The request was already processed (synchronously).
1616		 */
1617		case 0:
1618			atomic_dec(&ctx->cc_pending);
1619			ctx->cc_sector += sector_step;
1620			tag_offset++;
1621			if (!atomic)
1622				cond_resched();
1623			continue;
1624		/*
1625		 * There was a data integrity error.
1626		 */
1627		case -EBADMSG:
1628			atomic_dec(&ctx->cc_pending);
1629			return BLK_STS_PROTECTION;
1630		/*
1631		 * There was an error while processing the request.
1632		 */
1633		default:
1634			atomic_dec(&ctx->cc_pending);
1635			return BLK_STS_IOERR;
1636		}
1637	}
1638
1639	return 0;
1640}
1641
1642static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1643
1644/*
1645 * Generate a new unfragmented bio with the given size
1646 * This should never violate the device limitations (but only because
1647 * max_segment_size is being constrained to PAGE_SIZE).
1648 *
1649 * This function may be called concurrently. If we allocate from the mempool
1650 * concurrently, there is a possibility of deadlock. For example, if we have
1651 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1652 * the mempool concurrently, it may deadlock in a situation where both processes
1653 * have allocated 128 pages and the mempool is exhausted.
1654 *
1655 * In order to avoid this scenario we allocate the pages under a mutex.
1656 *
1657 * In order to not degrade performance with excessive locking, we try
1658 * non-blocking allocations without a mutex first but on failure we fallback
1659 * to blocking allocations with a mutex.
1660 */
1661static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1662{
1663	struct crypt_config *cc = io->cc;
1664	struct bio *clone;
1665	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1666	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1667	unsigned i, len, remaining_size;
1668	struct page *page;
1669
1670retry:
1671	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1672		mutex_lock(&cc->bio_alloc_lock);
1673
1674	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1675				 GFP_NOIO, &cc->bs);
1676	clone->bi_private = io;
1677	clone->bi_end_io = crypt_endio;
 
1678
1679	remaining_size = size;
1680
1681	for (i = 0; i < nr_iovecs; i++) {
1682		page = mempool_alloc(&cc->page_pool, gfp_mask);
1683		if (!page) {
1684			crypt_free_buffer_pages(cc, clone);
1685			bio_put(clone);
1686			gfp_mask |= __GFP_DIRECT_RECLAIM;
1687			goto retry;
1688		}
1689
1690		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1691
1692		bio_add_page(clone, page, len, 0);
1693
1694		remaining_size -= len;
1695	}
1696
1697	/* Allocate space for integrity tags */
1698	if (dm_crypt_integrity_io_alloc(io, clone)) {
1699		crypt_free_buffer_pages(cc, clone);
1700		bio_put(clone);
1701		clone = NULL;
1702	}
1703
1704	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1705		mutex_unlock(&cc->bio_alloc_lock);
1706
1707	return clone;
1708}
1709
1710static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1711{
1712	struct bio_vec *bv;
1713	struct bvec_iter_all iter_all;
1714
1715	bio_for_each_segment_all(bv, clone, iter_all) {
1716		BUG_ON(!bv->bv_page);
1717		mempool_free(bv->bv_page, &cc->page_pool);
1718	}
1719}
1720
1721static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1722			  struct bio *bio, sector_t sector)
1723{
1724	io->cc = cc;
1725	io->base_bio = bio;
1726	io->sector = sector;
1727	io->error = 0;
1728	io->ctx.r.req = NULL;
1729	io->integrity_metadata = NULL;
1730	io->integrity_metadata_from_pool = false;
1731	atomic_set(&io->io_pending, 0);
1732}
1733
1734static void crypt_inc_pending(struct dm_crypt_io *io)
1735{
1736	atomic_inc(&io->io_pending);
1737}
1738
1739static void kcryptd_io_bio_endio(struct work_struct *work)
1740{
1741	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1742	bio_endio(io->base_bio);
1743}
1744
1745/*
1746 * One of the bios was finished. Check for completion of
1747 * the whole request and correctly clean up the buffer.
1748 */
1749static void crypt_dec_pending(struct dm_crypt_io *io)
1750{
1751	struct crypt_config *cc = io->cc;
1752	struct bio *base_bio = io->base_bio;
1753	blk_status_t error = io->error;
1754
1755	if (!atomic_dec_and_test(&io->io_pending))
1756		return;
1757
1758	if (io->ctx.r.req)
1759		crypt_free_req(cc, io->ctx.r.req, base_bio);
1760
1761	if (unlikely(io->integrity_metadata_from_pool))
1762		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1763	else
1764		kfree(io->integrity_metadata);
1765
1766	base_bio->bi_status = error;
1767
1768	/*
1769	 * If we are running this function from our tasklet,
1770	 * we can't call bio_endio() here, because it will call
1771	 * clone_endio() from dm.c, which in turn will
1772	 * free the current struct dm_crypt_io structure with
1773	 * our tasklet. In this case we need to delay bio_endio()
1774	 * execution to after the tasklet is done and dequeued.
1775	 */
1776	if (tasklet_trylock(&io->tasklet)) {
1777		tasklet_unlock(&io->tasklet);
1778		bio_endio(base_bio);
1779		return;
1780	}
1781
1782	INIT_WORK(&io->work, kcryptd_io_bio_endio);
1783	queue_work(cc->io_queue, &io->work);
1784}
1785
1786/*
1787 * kcryptd/kcryptd_io:
1788 *
1789 * Needed because it would be very unwise to do decryption in an
1790 * interrupt context.
1791 *
1792 * kcryptd performs the actual encryption or decryption.
1793 *
1794 * kcryptd_io performs the IO submission.
1795 *
1796 * They must be separated as otherwise the final stages could be
1797 * starved by new requests which can block in the first stages due
1798 * to memory allocation.
1799 *
1800 * The work is done per CPU global for all dm-crypt instances.
1801 * They should not depend on each other and do not block.
1802 */
1803static void crypt_endio(struct bio *clone)
1804{
1805	struct dm_crypt_io *io = clone->bi_private;
1806	struct crypt_config *cc = io->cc;
1807	unsigned rw = bio_data_dir(clone);
1808	blk_status_t error;
1809
1810	/*
1811	 * free the processed pages
1812	 */
1813	if (rw == WRITE)
1814		crypt_free_buffer_pages(cc, clone);
1815
1816	error = clone->bi_status;
1817	bio_put(clone);
1818
1819	if (rw == READ && !error) {
1820		kcryptd_queue_crypt(io);
1821		return;
1822	}
1823
1824	if (unlikely(error))
1825		io->error = error;
1826
1827	crypt_dec_pending(io);
1828}
1829
1830#define CRYPT_MAP_READ_GFP GFP_NOWAIT
 
 
 
 
 
 
 
 
1831
1832static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1833{
1834	struct crypt_config *cc = io->cc;
1835	struct bio *clone;
1836
1837	/*
1838	 * We need the original biovec array in order to decrypt the whole bio
1839	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1840	 * worry about the block layer modifying the biovec array; so leverage
1841	 * bio_alloc_clone().
1842	 */
1843	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1844	if (!clone)
1845		return 1;
1846	clone->bi_private = io;
1847	clone->bi_end_io = crypt_endio;
1848
1849	crypt_inc_pending(io);
1850
 
1851	clone->bi_iter.bi_sector = cc->start + io->sector;
1852
1853	if (dm_crypt_integrity_io_alloc(io, clone)) {
1854		crypt_dec_pending(io);
1855		bio_put(clone);
1856		return 1;
1857	}
1858
1859	dm_submit_bio_remap(io->base_bio, clone);
1860	return 0;
1861}
1862
1863static void kcryptd_io_read_work(struct work_struct *work)
1864{
1865	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1866
1867	crypt_inc_pending(io);
1868	if (kcryptd_io_read(io, GFP_NOIO))
1869		io->error = BLK_STS_RESOURCE;
1870	crypt_dec_pending(io);
1871}
1872
1873static void kcryptd_queue_read(struct dm_crypt_io *io)
1874{
1875	struct crypt_config *cc = io->cc;
1876
1877	INIT_WORK(&io->work, kcryptd_io_read_work);
1878	queue_work(cc->io_queue, &io->work);
1879}
1880
1881static void kcryptd_io_write(struct dm_crypt_io *io)
1882{
1883	struct bio *clone = io->ctx.bio_out;
1884
1885	dm_submit_bio_remap(io->base_bio, clone);
1886}
1887
1888#define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1889
1890static int dmcrypt_write(void *data)
1891{
1892	struct crypt_config *cc = data;
1893	struct dm_crypt_io *io;
1894
1895	while (1) {
1896		struct rb_root write_tree;
1897		struct blk_plug plug;
1898
1899		spin_lock_irq(&cc->write_thread_lock);
1900continue_locked:
1901
1902		if (!RB_EMPTY_ROOT(&cc->write_tree))
1903			goto pop_from_list;
1904
1905		set_current_state(TASK_INTERRUPTIBLE);
1906
1907		spin_unlock_irq(&cc->write_thread_lock);
1908
1909		if (unlikely(kthread_should_stop())) {
1910			set_current_state(TASK_RUNNING);
1911			break;
1912		}
1913
1914		schedule();
1915
1916		set_current_state(TASK_RUNNING);
1917		spin_lock_irq(&cc->write_thread_lock);
1918		goto continue_locked;
1919
1920pop_from_list:
1921		write_tree = cc->write_tree;
1922		cc->write_tree = RB_ROOT;
1923		spin_unlock_irq(&cc->write_thread_lock);
1924
1925		BUG_ON(rb_parent(write_tree.rb_node));
1926
1927		/*
1928		 * Note: we cannot walk the tree here with rb_next because
1929		 * the structures may be freed when kcryptd_io_write is called.
1930		 */
1931		blk_start_plug(&plug);
1932		do {
1933			io = crypt_io_from_node(rb_first(&write_tree));
1934			rb_erase(&io->rb_node, &write_tree);
1935			kcryptd_io_write(io);
1936		} while (!RB_EMPTY_ROOT(&write_tree));
1937		blk_finish_plug(&plug);
1938	}
1939	return 0;
1940}
1941
1942static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1943{
1944	struct bio *clone = io->ctx.bio_out;
1945	struct crypt_config *cc = io->cc;
1946	unsigned long flags;
1947	sector_t sector;
1948	struct rb_node **rbp, *parent;
1949
1950	if (unlikely(io->error)) {
1951		crypt_free_buffer_pages(cc, clone);
1952		bio_put(clone);
1953		crypt_dec_pending(io);
1954		return;
1955	}
1956
1957	/* crypt_convert should have filled the clone bio */
1958	BUG_ON(io->ctx.iter_out.bi_size);
1959
1960	clone->bi_iter.bi_sector = cc->start + io->sector;
1961
1962	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1963	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1964		dm_submit_bio_remap(io->base_bio, clone);
1965		return;
1966	}
1967
1968	spin_lock_irqsave(&cc->write_thread_lock, flags);
1969	if (RB_EMPTY_ROOT(&cc->write_tree))
1970		wake_up_process(cc->write_thread);
1971	rbp = &cc->write_tree.rb_node;
1972	parent = NULL;
1973	sector = io->sector;
1974	while (*rbp) {
1975		parent = *rbp;
1976		if (sector < crypt_io_from_node(parent)->sector)
1977			rbp = &(*rbp)->rb_left;
1978		else
1979			rbp = &(*rbp)->rb_right;
1980	}
1981	rb_link_node(&io->rb_node, parent, rbp);
1982	rb_insert_color(&io->rb_node, &cc->write_tree);
1983	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1984}
1985
1986static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1987				       struct convert_context *ctx)
1988
1989{
1990	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1991		return false;
1992
1993	/*
1994	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1995	 * constraints so they do not need to be issued inline by
1996	 * kcryptd_crypt_write_convert().
1997	 */
1998	switch (bio_op(ctx->bio_in)) {
1999	case REQ_OP_WRITE:
 
2000	case REQ_OP_WRITE_ZEROES:
2001		return true;
2002	default:
2003		return false;
2004	}
2005}
2006
2007static void kcryptd_crypt_write_continue(struct work_struct *work)
2008{
2009	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2010	struct crypt_config *cc = io->cc;
2011	struct convert_context *ctx = &io->ctx;
2012	int crypt_finished;
2013	sector_t sector = io->sector;
2014	blk_status_t r;
2015
2016	wait_for_completion(&ctx->restart);
2017	reinit_completion(&ctx->restart);
2018
2019	r = crypt_convert(cc, &io->ctx, true, false);
2020	if (r)
2021		io->error = r;
2022	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2023	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2024		/* Wait for completion signaled by kcryptd_async_done() */
2025		wait_for_completion(&ctx->restart);
2026		crypt_finished = 1;
2027	}
2028
2029	/* Encryption was already finished, submit io now */
2030	if (crypt_finished) {
2031		kcryptd_crypt_write_io_submit(io, 0);
2032		io->sector = sector;
2033	}
2034
2035	crypt_dec_pending(io);
2036}
2037
2038static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2039{
2040	struct crypt_config *cc = io->cc;
2041	struct convert_context *ctx = &io->ctx;
2042	struct bio *clone;
2043	int crypt_finished;
2044	sector_t sector = io->sector;
2045	blk_status_t r;
2046
2047	/*
2048	 * Prevent io from disappearing until this function completes.
2049	 */
2050	crypt_inc_pending(io);
2051	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2052
2053	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2054	if (unlikely(!clone)) {
2055		io->error = BLK_STS_IOERR;
2056		goto dec;
2057	}
2058
2059	io->ctx.bio_out = clone;
2060	io->ctx.iter_out = clone->bi_iter;
2061
2062	sector += bio_sectors(clone);
2063
2064	crypt_inc_pending(io);
2065	r = crypt_convert(cc, ctx,
2066			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2067	/*
2068	 * Crypto API backlogged the request, because its queue was full
2069	 * and we're in softirq context, so continue from a workqueue
2070	 * (TODO: is it actually possible to be in softirq in the write path?)
2071	 */
2072	if (r == BLK_STS_DEV_RESOURCE) {
2073		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2074		queue_work(cc->crypt_queue, &io->work);
2075		return;
2076	}
2077	if (r)
2078		io->error = r;
2079	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2080	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2081		/* Wait for completion signaled by kcryptd_async_done() */
2082		wait_for_completion(&ctx->restart);
2083		crypt_finished = 1;
2084	}
2085
2086	/* Encryption was already finished, submit io now */
2087	if (crypt_finished) {
2088		kcryptd_crypt_write_io_submit(io, 0);
2089		io->sector = sector;
2090	}
2091
2092dec:
2093	crypt_dec_pending(io);
2094}
2095
2096static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2097{
2098	crypt_dec_pending(io);
2099}
2100
2101static void kcryptd_crypt_read_continue(struct work_struct *work)
2102{
2103	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2104	struct crypt_config *cc = io->cc;
2105	blk_status_t r;
2106
2107	wait_for_completion(&io->ctx.restart);
2108	reinit_completion(&io->ctx.restart);
2109
2110	r = crypt_convert(cc, &io->ctx, true, false);
2111	if (r)
2112		io->error = r;
2113
2114	if (atomic_dec_and_test(&io->ctx.cc_pending))
2115		kcryptd_crypt_read_done(io);
2116
2117	crypt_dec_pending(io);
2118}
2119
2120static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2121{
2122	struct crypt_config *cc = io->cc;
2123	blk_status_t r;
2124
2125	crypt_inc_pending(io);
2126
2127	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2128			   io->sector);
2129
2130	r = crypt_convert(cc, &io->ctx,
2131			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2132	/*
2133	 * Crypto API backlogged the request, because its queue was full
2134	 * and we're in softirq context, so continue from a workqueue
2135	 */
2136	if (r == BLK_STS_DEV_RESOURCE) {
2137		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2138		queue_work(cc->crypt_queue, &io->work);
2139		return;
2140	}
2141	if (r)
2142		io->error = r;
2143
2144	if (atomic_dec_and_test(&io->ctx.cc_pending))
2145		kcryptd_crypt_read_done(io);
2146
2147	crypt_dec_pending(io);
2148}
2149
2150static void kcryptd_async_done(struct crypto_async_request *async_req,
2151			       int error)
2152{
2153	struct dm_crypt_request *dmreq = async_req->data;
2154	struct convert_context *ctx = dmreq->ctx;
2155	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2156	struct crypt_config *cc = io->cc;
2157
2158	/*
2159	 * A request from crypto driver backlog is going to be processed now,
2160	 * finish the completion and continue in crypt_convert().
2161	 * (Callback will be called for the second time for this request.)
2162	 */
2163	if (error == -EINPROGRESS) {
2164		complete(&ctx->restart);
2165		return;
2166	}
2167
2168	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2169		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2170
2171	if (error == -EBADMSG) {
2172		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2173
2174		DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2175			    ctx->bio_in->bi_bdev, s);
2176		dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2177				 ctx->bio_in, s, 0);
2178		io->error = BLK_STS_PROTECTION;
2179	} else if (error < 0)
2180		io->error = BLK_STS_IOERR;
2181
2182	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2183
2184	if (!atomic_dec_and_test(&ctx->cc_pending))
2185		return;
2186
2187	/*
2188	 * The request is fully completed: for inline writes, let
2189	 * kcryptd_crypt_write_convert() do the IO submission.
2190	 */
2191	if (bio_data_dir(io->base_bio) == READ) {
2192		kcryptd_crypt_read_done(io);
2193		return;
2194	}
2195
2196	if (kcryptd_crypt_write_inline(cc, ctx)) {
2197		complete(&ctx->restart);
2198		return;
2199	}
2200
2201	kcryptd_crypt_write_io_submit(io, 1);
2202}
2203
2204static void kcryptd_crypt(struct work_struct *work)
2205{
2206	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2207
2208	if (bio_data_dir(io->base_bio) == READ)
2209		kcryptd_crypt_read_convert(io);
2210	else
2211		kcryptd_crypt_write_convert(io);
2212}
2213
2214static void kcryptd_crypt_tasklet(unsigned long work)
2215{
2216	kcryptd_crypt((struct work_struct *)work);
2217}
2218
2219static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2220{
2221	struct crypt_config *cc = io->cc;
2222
2223	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2224	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2225		/*
2226		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2227		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2228		 * it is being executed with irqs disabled.
2229		 */
2230		if (in_hardirq() || irqs_disabled()) {
2231			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2232			tasklet_schedule(&io->tasklet);
2233			return;
2234		}
2235
2236		kcryptd_crypt(&io->work);
2237		return;
2238	}
2239
2240	INIT_WORK(&io->work, kcryptd_crypt);
2241	queue_work(cc->crypt_queue, &io->work);
2242}
2243
2244static void crypt_free_tfms_aead(struct crypt_config *cc)
2245{
2246	if (!cc->cipher_tfm.tfms_aead)
2247		return;
2248
2249	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2250		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2251		cc->cipher_tfm.tfms_aead[0] = NULL;
2252	}
2253
2254	kfree(cc->cipher_tfm.tfms_aead);
2255	cc->cipher_tfm.tfms_aead = NULL;
2256}
2257
2258static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2259{
2260	unsigned i;
2261
2262	if (!cc->cipher_tfm.tfms)
2263		return;
2264
2265	for (i = 0; i < cc->tfms_count; i++)
2266		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2267			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2268			cc->cipher_tfm.tfms[i] = NULL;
2269		}
2270
2271	kfree(cc->cipher_tfm.tfms);
2272	cc->cipher_tfm.tfms = NULL;
2273}
2274
2275static void crypt_free_tfms(struct crypt_config *cc)
2276{
2277	if (crypt_integrity_aead(cc))
2278		crypt_free_tfms_aead(cc);
2279	else
2280		crypt_free_tfms_skcipher(cc);
2281}
2282
2283static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2284{
2285	unsigned i;
2286	int err;
2287
2288	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2289				      sizeof(struct crypto_skcipher *),
2290				      GFP_KERNEL);
2291	if (!cc->cipher_tfm.tfms)
2292		return -ENOMEM;
2293
2294	for (i = 0; i < cc->tfms_count; i++) {
2295		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2296						CRYPTO_ALG_ALLOCATES_MEMORY);
2297		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2298			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2299			crypt_free_tfms(cc);
2300			return err;
2301		}
2302	}
2303
2304	/*
2305	 * dm-crypt performance can vary greatly depending on which crypto
2306	 * algorithm implementation is used.  Help people debug performance
2307	 * problems by logging the ->cra_driver_name.
2308	 */
2309	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2310	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2311	return 0;
2312}
2313
2314static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2315{
2316	int err;
2317
2318	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2319	if (!cc->cipher_tfm.tfms)
2320		return -ENOMEM;
2321
2322	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2323						CRYPTO_ALG_ALLOCATES_MEMORY);
2324	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2325		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2326		crypt_free_tfms(cc);
2327		return err;
2328	}
2329
2330	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2331	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2332	return 0;
2333}
2334
2335static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2336{
2337	if (crypt_integrity_aead(cc))
2338		return crypt_alloc_tfms_aead(cc, ciphermode);
2339	else
2340		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2341}
2342
2343static unsigned crypt_subkey_size(struct crypt_config *cc)
2344{
2345	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2346}
2347
2348static unsigned crypt_authenckey_size(struct crypt_config *cc)
2349{
2350	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2351}
2352
2353/*
2354 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2355 * the key must be for some reason in special format.
2356 * This funcion converts cc->key to this special format.
2357 */
2358static void crypt_copy_authenckey(char *p, const void *key,
2359				  unsigned enckeylen, unsigned authkeylen)
2360{
2361	struct crypto_authenc_key_param *param;
2362	struct rtattr *rta;
2363
2364	rta = (struct rtattr *)p;
2365	param = RTA_DATA(rta);
2366	param->enckeylen = cpu_to_be32(enckeylen);
2367	rta->rta_len = RTA_LENGTH(sizeof(*param));
2368	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2369	p += RTA_SPACE(sizeof(*param));
2370	memcpy(p, key + enckeylen, authkeylen);
2371	p += authkeylen;
2372	memcpy(p, key, enckeylen);
2373}
2374
2375static int crypt_setkey(struct crypt_config *cc)
2376{
2377	unsigned subkey_size;
2378	int err = 0, i, r;
2379
2380	/* Ignore extra keys (which are used for IV etc) */
2381	subkey_size = crypt_subkey_size(cc);
2382
2383	if (crypt_integrity_hmac(cc)) {
2384		if (subkey_size < cc->key_mac_size)
2385			return -EINVAL;
2386
2387		crypt_copy_authenckey(cc->authenc_key, cc->key,
2388				      subkey_size - cc->key_mac_size,
2389				      cc->key_mac_size);
2390	}
2391
2392	for (i = 0; i < cc->tfms_count; i++) {
2393		if (crypt_integrity_hmac(cc))
2394			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2395				cc->authenc_key, crypt_authenckey_size(cc));
2396		else if (crypt_integrity_aead(cc))
2397			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2398					       cc->key + (i * subkey_size),
2399					       subkey_size);
2400		else
2401			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2402						   cc->key + (i * subkey_size),
2403						   subkey_size);
2404		if (r)
2405			err = r;
2406	}
2407
2408	if (crypt_integrity_hmac(cc))
2409		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2410
2411	return err;
2412}
2413
2414#ifdef CONFIG_KEYS
2415
2416static bool contains_whitespace(const char *str)
2417{
2418	while (*str)
2419		if (isspace(*str++))
2420			return true;
2421	return false;
2422}
2423
2424static int set_key_user(struct crypt_config *cc, struct key *key)
2425{
2426	const struct user_key_payload *ukp;
2427
2428	ukp = user_key_payload_locked(key);
2429	if (!ukp)
2430		return -EKEYREVOKED;
2431
2432	if (cc->key_size != ukp->datalen)
2433		return -EINVAL;
2434
2435	memcpy(cc->key, ukp->data, cc->key_size);
2436
2437	return 0;
2438}
2439
2440static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2441{
2442	const struct encrypted_key_payload *ekp;
2443
2444	ekp = key->payload.data[0];
2445	if (!ekp)
2446		return -EKEYREVOKED;
2447
2448	if (cc->key_size != ekp->decrypted_datalen)
2449		return -EINVAL;
2450
2451	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2452
2453	return 0;
2454}
2455
2456static int set_key_trusted(struct crypt_config *cc, struct key *key)
2457{
2458	const struct trusted_key_payload *tkp;
2459
2460	tkp = key->payload.data[0];
2461	if (!tkp)
2462		return -EKEYREVOKED;
2463
2464	if (cc->key_size != tkp->key_len)
2465		return -EINVAL;
2466
2467	memcpy(cc->key, tkp->key, cc->key_size);
2468
2469	return 0;
2470}
2471
2472static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2473{
2474	char *new_key_string, *key_desc;
2475	int ret;
2476	struct key_type *type;
2477	struct key *key;
2478	int (*set_key)(struct crypt_config *cc, struct key *key);
2479
2480	/*
2481	 * Reject key_string with whitespace. dm core currently lacks code for
2482	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2483	 */
2484	if (contains_whitespace(key_string)) {
2485		DMERR("whitespace chars not allowed in key string");
2486		return -EINVAL;
2487	}
2488
2489	/* look for next ':' separating key_type from key_description */
2490	key_desc = strpbrk(key_string, ":");
2491	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2492		return -EINVAL;
2493
2494	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2495		type = &key_type_logon;
2496		set_key = set_key_user;
2497	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2498		type = &key_type_user;
2499		set_key = set_key_user;
2500	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2501		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2502		type = &key_type_encrypted;
2503		set_key = set_key_encrypted;
2504	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2505	           !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2506		type = &key_type_trusted;
2507		set_key = set_key_trusted;
2508	} else {
2509		return -EINVAL;
2510	}
2511
2512	new_key_string = kstrdup(key_string, GFP_KERNEL);
2513	if (!new_key_string)
2514		return -ENOMEM;
2515
2516	key = request_key(type, key_desc + 1, NULL);
2517	if (IS_ERR(key)) {
2518		kfree_sensitive(new_key_string);
2519		return PTR_ERR(key);
2520	}
2521
2522	down_read(&key->sem);
2523
2524	ret = set_key(cc, key);
2525	if (ret < 0) {
2526		up_read(&key->sem);
2527		key_put(key);
2528		kfree_sensitive(new_key_string);
2529		return ret;
2530	}
2531
2532	up_read(&key->sem);
2533	key_put(key);
2534
2535	/* clear the flag since following operations may invalidate previously valid key */
2536	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2537
2538	ret = crypt_setkey(cc);
2539
2540	if (!ret) {
2541		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2542		kfree_sensitive(cc->key_string);
2543		cc->key_string = new_key_string;
2544	} else
2545		kfree_sensitive(new_key_string);
2546
2547	return ret;
2548}
2549
2550static int get_key_size(char **key_string)
2551{
2552	char *colon, dummy;
2553	int ret;
2554
2555	if (*key_string[0] != ':')
2556		return strlen(*key_string) >> 1;
2557
2558	/* look for next ':' in key string */
2559	colon = strpbrk(*key_string + 1, ":");
2560	if (!colon)
2561		return -EINVAL;
2562
2563	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2564		return -EINVAL;
2565
2566	*key_string = colon;
2567
2568	/* remaining key string should be :<logon|user>:<key_desc> */
2569
2570	return ret;
2571}
2572
2573#else
2574
2575static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2576{
2577	return -EINVAL;
2578}
2579
2580static int get_key_size(char **key_string)
2581{
2582	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2583}
2584
2585#endif /* CONFIG_KEYS */
2586
2587static int crypt_set_key(struct crypt_config *cc, char *key)
2588{
2589	int r = -EINVAL;
2590	int key_string_len = strlen(key);
2591
2592	/* Hyphen (which gives a key_size of zero) means there is no key. */
2593	if (!cc->key_size && strcmp(key, "-"))
2594		goto out;
2595
2596	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2597	if (key[0] == ':') {
2598		r = crypt_set_keyring_key(cc, key + 1);
2599		goto out;
2600	}
2601
2602	/* clear the flag since following operations may invalidate previously valid key */
2603	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2604
2605	/* wipe references to any kernel keyring key */
2606	kfree_sensitive(cc->key_string);
2607	cc->key_string = NULL;
2608
2609	/* Decode key from its hex representation. */
2610	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2611		goto out;
2612
2613	r = crypt_setkey(cc);
2614	if (!r)
2615		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2616
2617out:
2618	/* Hex key string not needed after here, so wipe it. */
2619	memset(key, '0', key_string_len);
2620
2621	return r;
2622}
2623
2624static int crypt_wipe_key(struct crypt_config *cc)
2625{
2626	int r;
2627
2628	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2629	get_random_bytes(&cc->key, cc->key_size);
2630
2631	/* Wipe IV private keys */
2632	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2633		r = cc->iv_gen_ops->wipe(cc);
2634		if (r)
2635			return r;
2636	}
2637
2638	kfree_sensitive(cc->key_string);
2639	cc->key_string = NULL;
2640	r = crypt_setkey(cc);
2641	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2642
2643	return r;
2644}
2645
2646static void crypt_calculate_pages_per_client(void)
2647{
2648	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2649
2650	if (!dm_crypt_clients_n)
2651		return;
2652
2653	pages /= dm_crypt_clients_n;
2654	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2655		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2656	dm_crypt_pages_per_client = pages;
2657}
2658
2659static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2660{
2661	struct crypt_config *cc = pool_data;
2662	struct page *page;
2663
2664	/*
2665	 * Note, percpu_counter_read_positive() may over (and under) estimate
2666	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2667	 * but avoids potential spinlock contention of an exact result.
2668	 */
2669	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2670	    likely(gfp_mask & __GFP_NORETRY))
2671		return NULL;
2672
2673	page = alloc_page(gfp_mask);
2674	if (likely(page != NULL))
2675		percpu_counter_add(&cc->n_allocated_pages, 1);
2676
2677	return page;
2678}
2679
2680static void crypt_page_free(void *page, void *pool_data)
2681{
2682	struct crypt_config *cc = pool_data;
2683
2684	__free_page(page);
2685	percpu_counter_sub(&cc->n_allocated_pages, 1);
2686}
2687
2688static void crypt_dtr(struct dm_target *ti)
2689{
2690	struct crypt_config *cc = ti->private;
2691
2692	ti->private = NULL;
2693
2694	if (!cc)
2695		return;
2696
2697	if (cc->write_thread)
2698		kthread_stop(cc->write_thread);
2699
2700	if (cc->io_queue)
2701		destroy_workqueue(cc->io_queue);
2702	if (cc->crypt_queue)
2703		destroy_workqueue(cc->crypt_queue);
2704
2705	crypt_free_tfms(cc);
2706
2707	bioset_exit(&cc->bs);
2708
2709	mempool_exit(&cc->page_pool);
2710	mempool_exit(&cc->req_pool);
2711	mempool_exit(&cc->tag_pool);
2712
2713	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2714	percpu_counter_destroy(&cc->n_allocated_pages);
2715
2716	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2717		cc->iv_gen_ops->dtr(cc);
2718
2719	if (cc->dev)
2720		dm_put_device(ti, cc->dev);
2721
2722	kfree_sensitive(cc->cipher_string);
2723	kfree_sensitive(cc->key_string);
2724	kfree_sensitive(cc->cipher_auth);
2725	kfree_sensitive(cc->authenc_key);
2726
2727	mutex_destroy(&cc->bio_alloc_lock);
2728
2729	/* Must zero key material before freeing */
2730	kfree_sensitive(cc);
2731
2732	spin_lock(&dm_crypt_clients_lock);
2733	WARN_ON(!dm_crypt_clients_n);
2734	dm_crypt_clients_n--;
2735	crypt_calculate_pages_per_client();
2736	spin_unlock(&dm_crypt_clients_lock);
2737
2738	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2739}
2740
2741static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2742{
2743	struct crypt_config *cc = ti->private;
2744
2745	if (crypt_integrity_aead(cc))
2746		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2747	else
2748		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2749
2750	if (cc->iv_size)
2751		/* at least a 64 bit sector number should fit in our buffer */
2752		cc->iv_size = max(cc->iv_size,
2753				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2754	else if (ivmode) {
2755		DMWARN("Selected cipher does not support IVs");
2756		ivmode = NULL;
2757	}
2758
2759	/* Choose ivmode, see comments at iv code. */
2760	if (ivmode == NULL)
2761		cc->iv_gen_ops = NULL;
2762	else if (strcmp(ivmode, "plain") == 0)
2763		cc->iv_gen_ops = &crypt_iv_plain_ops;
2764	else if (strcmp(ivmode, "plain64") == 0)
2765		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2766	else if (strcmp(ivmode, "plain64be") == 0)
2767		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2768	else if (strcmp(ivmode, "essiv") == 0)
2769		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2770	else if (strcmp(ivmode, "benbi") == 0)
2771		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2772	else if (strcmp(ivmode, "null") == 0)
2773		cc->iv_gen_ops = &crypt_iv_null_ops;
2774	else if (strcmp(ivmode, "eboiv") == 0)
2775		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2776	else if (strcmp(ivmode, "elephant") == 0) {
2777		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2778		cc->key_parts = 2;
2779		cc->key_extra_size = cc->key_size / 2;
2780		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2781			return -EINVAL;
2782		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2783	} else if (strcmp(ivmode, "lmk") == 0) {
2784		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2785		/*
2786		 * Version 2 and 3 is recognised according
2787		 * to length of provided multi-key string.
2788		 * If present (version 3), last key is used as IV seed.
2789		 * All keys (including IV seed) are always the same size.
2790		 */
2791		if (cc->key_size % cc->key_parts) {
2792			cc->key_parts++;
2793			cc->key_extra_size = cc->key_size / cc->key_parts;
2794		}
2795	} else if (strcmp(ivmode, "tcw") == 0) {
2796		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2797		cc->key_parts += 2; /* IV + whitening */
2798		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2799	} else if (strcmp(ivmode, "random") == 0) {
2800		cc->iv_gen_ops = &crypt_iv_random_ops;
2801		/* Need storage space in integrity fields. */
2802		cc->integrity_iv_size = cc->iv_size;
2803	} else {
2804		ti->error = "Invalid IV mode";
2805		return -EINVAL;
2806	}
2807
2808	return 0;
2809}
2810
2811/*
2812 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2813 * The HMAC is needed to calculate tag size (HMAC digest size).
2814 * This should be probably done by crypto-api calls (once available...)
2815 */
2816static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2817{
2818	char *start, *end, *mac_alg = NULL;
2819	struct crypto_ahash *mac;
2820
2821	if (!strstarts(cipher_api, "authenc("))
2822		return 0;
2823
2824	start = strchr(cipher_api, '(');
2825	end = strchr(cipher_api, ',');
2826	if (!start || !end || ++start > end)
2827		return -EINVAL;
2828
2829	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2830	if (!mac_alg)
2831		return -ENOMEM;
2832	strncpy(mac_alg, start, end - start);
2833
2834	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2835	kfree(mac_alg);
2836
2837	if (IS_ERR(mac))
2838		return PTR_ERR(mac);
2839
2840	cc->key_mac_size = crypto_ahash_digestsize(mac);
2841	crypto_free_ahash(mac);
2842
2843	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2844	if (!cc->authenc_key)
2845		return -ENOMEM;
2846
2847	return 0;
2848}
2849
2850static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2851				char **ivmode, char **ivopts)
2852{
2853	struct crypt_config *cc = ti->private;
2854	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2855	int ret = -EINVAL;
2856
2857	cc->tfms_count = 1;
2858
2859	/*
2860	 * New format (capi: prefix)
2861	 * capi:cipher_api_spec-iv:ivopts
2862	 */
2863	tmp = &cipher_in[strlen("capi:")];
2864
2865	/* Separate IV options if present, it can contain another '-' in hash name */
2866	*ivopts = strrchr(tmp, ':');
2867	if (*ivopts) {
2868		**ivopts = '\0';
2869		(*ivopts)++;
2870	}
2871	/* Parse IV mode */
2872	*ivmode = strrchr(tmp, '-');
2873	if (*ivmode) {
2874		**ivmode = '\0';
2875		(*ivmode)++;
2876	}
2877	/* The rest is crypto API spec */
2878	cipher_api = tmp;
2879
2880	/* Alloc AEAD, can be used only in new format. */
2881	if (crypt_integrity_aead(cc)) {
2882		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2883		if (ret < 0) {
2884			ti->error = "Invalid AEAD cipher spec";
2885			return -ENOMEM;
2886		}
2887	}
2888
2889	if (*ivmode && !strcmp(*ivmode, "lmk"))
2890		cc->tfms_count = 64;
2891
2892	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2893		if (!*ivopts) {
2894			ti->error = "Digest algorithm missing for ESSIV mode";
2895			return -EINVAL;
2896		}
2897		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2898			       cipher_api, *ivopts);
2899		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2900			ti->error = "Cannot allocate cipher string";
2901			return -ENOMEM;
2902		}
2903		cipher_api = buf;
2904	}
2905
2906	cc->key_parts = cc->tfms_count;
2907
2908	/* Allocate cipher */
2909	ret = crypt_alloc_tfms(cc, cipher_api);
2910	if (ret < 0) {
2911		ti->error = "Error allocating crypto tfm";
2912		return ret;
2913	}
2914
2915	if (crypt_integrity_aead(cc))
2916		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2917	else
2918		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2919
2920	return 0;
2921}
2922
2923static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2924				char **ivmode, char **ivopts)
2925{
2926	struct crypt_config *cc = ti->private;
2927	char *tmp, *cipher, *chainmode, *keycount;
2928	char *cipher_api = NULL;
2929	int ret = -EINVAL;
2930	char dummy;
2931
2932	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2933		ti->error = "Bad cipher specification";
2934		return -EINVAL;
2935	}
2936
2937	/*
2938	 * Legacy dm-crypt cipher specification
2939	 * cipher[:keycount]-mode-iv:ivopts
2940	 */
2941	tmp = cipher_in;
2942	keycount = strsep(&tmp, "-");
2943	cipher = strsep(&keycount, ":");
2944
2945	if (!keycount)
2946		cc->tfms_count = 1;
2947	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2948		 !is_power_of_2(cc->tfms_count)) {
2949		ti->error = "Bad cipher key count specification";
2950		return -EINVAL;
2951	}
2952	cc->key_parts = cc->tfms_count;
2953
2954	chainmode = strsep(&tmp, "-");
2955	*ivmode = strsep(&tmp, ":");
2956	*ivopts = tmp;
2957
2958	/*
2959	 * For compatibility with the original dm-crypt mapping format, if
2960	 * only the cipher name is supplied, use cbc-plain.
2961	 */
2962	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2963		chainmode = "cbc";
2964		*ivmode = "plain";
2965	}
2966
2967	if (strcmp(chainmode, "ecb") && !*ivmode) {
2968		ti->error = "IV mechanism required";
2969		return -EINVAL;
2970	}
2971
2972	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2973	if (!cipher_api)
2974		goto bad_mem;
2975
2976	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2977		if (!*ivopts) {
2978			ti->error = "Digest algorithm missing for ESSIV mode";
2979			kfree(cipher_api);
2980			return -EINVAL;
2981		}
2982		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2983			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2984	} else {
2985		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2986			       "%s(%s)", chainmode, cipher);
2987	}
2988	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2989		kfree(cipher_api);
2990		goto bad_mem;
2991	}
2992
2993	/* Allocate cipher */
2994	ret = crypt_alloc_tfms(cc, cipher_api);
2995	if (ret < 0) {
2996		ti->error = "Error allocating crypto tfm";
2997		kfree(cipher_api);
2998		return ret;
2999	}
3000	kfree(cipher_api);
3001
3002	return 0;
3003bad_mem:
3004	ti->error = "Cannot allocate cipher strings";
3005	return -ENOMEM;
3006}
3007
3008static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3009{
3010	struct crypt_config *cc = ti->private;
3011	char *ivmode = NULL, *ivopts = NULL;
3012	int ret;
3013
3014	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3015	if (!cc->cipher_string) {
3016		ti->error = "Cannot allocate cipher strings";
3017		return -ENOMEM;
3018	}
3019
3020	if (strstarts(cipher_in, "capi:"))
3021		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3022	else
3023		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3024	if (ret)
3025		return ret;
3026
3027	/* Initialize IV */
3028	ret = crypt_ctr_ivmode(ti, ivmode);
3029	if (ret < 0)
3030		return ret;
3031
3032	/* Initialize and set key */
3033	ret = crypt_set_key(cc, key);
3034	if (ret < 0) {
3035		ti->error = "Error decoding and setting key";
3036		return ret;
3037	}
3038
3039	/* Allocate IV */
3040	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3041		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3042		if (ret < 0) {
3043			ti->error = "Error creating IV";
3044			return ret;
3045		}
3046	}
3047
3048	/* Initialize IV (set keys for ESSIV etc) */
3049	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3050		ret = cc->iv_gen_ops->init(cc);
3051		if (ret < 0) {
3052			ti->error = "Error initialising IV";
3053			return ret;
3054		}
3055	}
3056
3057	/* wipe the kernel key payload copy */
3058	if (cc->key_string)
3059		memset(cc->key, 0, cc->key_size * sizeof(u8));
3060
3061	return ret;
3062}
3063
3064static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3065{
3066	struct crypt_config *cc = ti->private;
3067	struct dm_arg_set as;
3068	static const struct dm_arg _args[] = {
3069		{0, 8, "Invalid number of feature args"},
3070	};
3071	unsigned int opt_params, val;
3072	const char *opt_string, *sval;
3073	char dummy;
3074	int ret;
3075
3076	/* Optional parameters */
3077	as.argc = argc;
3078	as.argv = argv;
3079
3080	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3081	if (ret)
3082		return ret;
3083
3084	while (opt_params--) {
3085		opt_string = dm_shift_arg(&as);
3086		if (!opt_string) {
3087			ti->error = "Not enough feature arguments";
3088			return -EINVAL;
3089		}
3090
3091		if (!strcasecmp(opt_string, "allow_discards"))
3092			ti->num_discard_bios = 1;
3093
3094		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3095			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3096
3097		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3098			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3099		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3100			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3101		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3102			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3103		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3104			if (val == 0 || val > MAX_TAG_SIZE) {
3105				ti->error = "Invalid integrity arguments";
3106				return -EINVAL;
3107			}
3108			cc->on_disk_tag_size = val;
3109			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3110			if (!strcasecmp(sval, "aead")) {
3111				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3112			} else  if (strcasecmp(sval, "none")) {
3113				ti->error = "Unknown integrity profile";
3114				return -EINVAL;
3115			}
3116
3117			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3118			if (!cc->cipher_auth)
3119				return -ENOMEM;
3120		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3121			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3122			    cc->sector_size > 4096 ||
3123			    (cc->sector_size & (cc->sector_size - 1))) {
3124				ti->error = "Invalid feature value for sector_size";
3125				return -EINVAL;
3126			}
3127			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3128				ti->error = "Device size is not multiple of sector_size feature";
3129				return -EINVAL;
3130			}
3131			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3132		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3133			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3134		else {
3135			ti->error = "Invalid feature arguments";
3136			return -EINVAL;
3137		}
3138	}
3139
3140	return 0;
3141}
3142
3143#ifdef CONFIG_BLK_DEV_ZONED
3144static int crypt_report_zones(struct dm_target *ti,
3145		struct dm_report_zones_args *args, unsigned int nr_zones)
3146{
3147	struct crypt_config *cc = ti->private;
3148
3149	return dm_report_zones(cc->dev->bdev, cc->start,
3150			cc->start + dm_target_offset(ti, args->next_sector),
3151			args, nr_zones);
3152}
3153#else
3154#define crypt_report_zones NULL
3155#endif
3156
3157/*
3158 * Construct an encryption mapping:
3159 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3160 */
3161static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3162{
3163	struct crypt_config *cc;
3164	const char *devname = dm_table_device_name(ti->table);
3165	int key_size;
3166	unsigned int align_mask;
3167	unsigned long long tmpll;
3168	int ret;
3169	size_t iv_size_padding, additional_req_size;
3170	char dummy;
3171
3172	if (argc < 5) {
3173		ti->error = "Not enough arguments";
3174		return -EINVAL;
3175	}
3176
3177	key_size = get_key_size(&argv[1]);
3178	if (key_size < 0) {
3179		ti->error = "Cannot parse key size";
3180		return -EINVAL;
3181	}
3182
3183	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3184	if (!cc) {
3185		ti->error = "Cannot allocate encryption context";
3186		return -ENOMEM;
3187	}
3188	cc->key_size = key_size;
3189	cc->sector_size = (1 << SECTOR_SHIFT);
3190	cc->sector_shift = 0;
3191
3192	ti->private = cc;
3193
3194	spin_lock(&dm_crypt_clients_lock);
3195	dm_crypt_clients_n++;
3196	crypt_calculate_pages_per_client();
3197	spin_unlock(&dm_crypt_clients_lock);
3198
3199	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3200	if (ret < 0)
3201		goto bad;
3202
3203	/* Optional parameters need to be read before cipher constructor */
3204	if (argc > 5) {
3205		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3206		if (ret)
3207			goto bad;
3208	}
3209
3210	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3211	if (ret < 0)
3212		goto bad;
3213
3214	if (crypt_integrity_aead(cc)) {
3215		cc->dmreq_start = sizeof(struct aead_request);
3216		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3217		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3218	} else {
3219		cc->dmreq_start = sizeof(struct skcipher_request);
3220		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3221		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3222	}
3223	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3224
3225	if (align_mask < CRYPTO_MINALIGN) {
3226		/* Allocate the padding exactly */
3227		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3228				& align_mask;
3229	} else {
3230		/*
3231		 * If the cipher requires greater alignment than kmalloc
3232		 * alignment, we don't know the exact position of the
3233		 * initialization vector. We must assume worst case.
3234		 */
3235		iv_size_padding = align_mask;
3236	}
3237
3238	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3239	additional_req_size = sizeof(struct dm_crypt_request) +
3240		iv_size_padding + cc->iv_size +
3241		cc->iv_size +
3242		sizeof(uint64_t) +
3243		sizeof(unsigned int);
3244
3245	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3246	if (ret) {
3247		ti->error = "Cannot allocate crypt request mempool";
3248		goto bad;
3249	}
3250
3251	cc->per_bio_data_size = ti->per_io_data_size =
3252		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3253		      ARCH_KMALLOC_MINALIGN);
3254
3255	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3256	if (ret) {
3257		ti->error = "Cannot allocate page mempool";
3258		goto bad;
3259	}
3260
3261	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3262	if (ret) {
3263		ti->error = "Cannot allocate crypt bioset";
3264		goto bad;
3265	}
3266
3267	mutex_init(&cc->bio_alloc_lock);
3268
3269	ret = -EINVAL;
3270	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3271	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3272		ti->error = "Invalid iv_offset sector";
3273		goto bad;
3274	}
3275	cc->iv_offset = tmpll;
3276
3277	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3278	if (ret) {
3279		ti->error = "Device lookup failed";
3280		goto bad;
3281	}
3282
3283	ret = -EINVAL;
3284	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3285		ti->error = "Invalid device sector";
3286		goto bad;
3287	}
3288	cc->start = tmpll;
3289
3290	if (bdev_is_zoned(cc->dev->bdev)) {
3291		/*
3292		 * For zoned block devices, we need to preserve the issuer write
3293		 * ordering. To do so, disable write workqueues and force inline
3294		 * encryption completion.
3295		 */
3296		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3297		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3298
3299		/*
3300		 * All zone append writes to a zone of a zoned block device will
3301		 * have the same BIO sector, the start of the zone. When the
3302		 * cypher IV mode uses sector values, all data targeting a
3303		 * zone will be encrypted using the first sector numbers of the
3304		 * zone. This will not result in write errors but will
3305		 * cause most reads to fail as reads will use the sector values
3306		 * for the actual data locations, resulting in IV mismatch.
3307		 * To avoid this problem, ask DM core to emulate zone append
3308		 * operations with regular writes.
3309		 */
3310		DMDEBUG("Zone append operations will be emulated");
3311		ti->emulate_zone_append = true;
3312	}
3313
3314	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3315		ret = crypt_integrity_ctr(cc, ti);
3316		if (ret)
3317			goto bad;
3318
3319		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3320		if (!cc->tag_pool_max_sectors)
3321			cc->tag_pool_max_sectors = 1;
3322
3323		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3324			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3325		if (ret) {
3326			ti->error = "Cannot allocate integrity tags mempool";
3327			goto bad;
3328		}
3329
3330		cc->tag_pool_max_sectors <<= cc->sector_shift;
3331	}
3332
3333	ret = -ENOMEM;
3334	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3335	if (!cc->io_queue) {
3336		ti->error = "Couldn't create kcryptd io queue";
3337		goto bad;
3338	}
3339
3340	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3341		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3342						  1, devname);
3343	else
3344		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3345						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3346						  num_online_cpus(), devname);
3347	if (!cc->crypt_queue) {
3348		ti->error = "Couldn't create kcryptd queue";
3349		goto bad;
3350	}
3351
3352	spin_lock_init(&cc->write_thread_lock);
3353	cc->write_tree = RB_ROOT;
3354
3355	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3356	if (IS_ERR(cc->write_thread)) {
3357		ret = PTR_ERR(cc->write_thread);
3358		cc->write_thread = NULL;
3359		ti->error = "Couldn't spawn write thread";
3360		goto bad;
3361	}
 
3362
3363	ti->num_flush_bios = 1;
3364	ti->limit_swap_bios = true;
3365	ti->accounts_remapped_io = true;
3366
3367	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3368	return 0;
3369
3370bad:
3371	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3372	crypt_dtr(ti);
3373	return ret;
3374}
3375
3376static int crypt_map(struct dm_target *ti, struct bio *bio)
3377{
3378	struct dm_crypt_io *io;
3379	struct crypt_config *cc = ti->private;
3380
3381	/*
3382	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3383	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3384	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3385	 */
3386	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3387	    bio_op(bio) == REQ_OP_DISCARD)) {
3388		bio_set_dev(bio, cc->dev->bdev);
3389		if (bio_sectors(bio))
3390			bio->bi_iter.bi_sector = cc->start +
3391				dm_target_offset(ti, bio->bi_iter.bi_sector);
3392		return DM_MAPIO_REMAPPED;
3393	}
3394
3395	/*
3396	 * Check if bio is too large, split as needed.
3397	 */
3398	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3399	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3400		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3401
3402	/*
3403	 * Ensure that bio is a multiple of internal sector encryption size
3404	 * and is aligned to this size as defined in IO hints.
3405	 */
3406	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3407		return DM_MAPIO_KILL;
3408
3409	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3410		return DM_MAPIO_KILL;
3411
3412	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3413	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3414
3415	if (cc->on_disk_tag_size) {
3416		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3417
3418		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3419		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3420				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3421			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3422				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3423			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3424			io->integrity_metadata_from_pool = true;
3425		}
3426	}
3427
3428	if (crypt_integrity_aead(cc))
3429		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3430	else
3431		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3432
3433	if (bio_data_dir(io->base_bio) == READ) {
3434		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3435			kcryptd_queue_read(io);
3436	} else
3437		kcryptd_queue_crypt(io);
3438
3439	return DM_MAPIO_SUBMITTED;
3440}
3441
3442static char hex2asc(unsigned char c)
3443{
3444	return c + '0' + ((unsigned)(9 - c) >> 4 & 0x27);
3445}
3446
3447static void crypt_status(struct dm_target *ti, status_type_t type,
3448			 unsigned status_flags, char *result, unsigned maxlen)
3449{
3450	struct crypt_config *cc = ti->private;
3451	unsigned i, sz = 0;
3452	int num_feature_args = 0;
3453
3454	switch (type) {
3455	case STATUSTYPE_INFO:
3456		result[0] = '\0';
3457		break;
3458
3459	case STATUSTYPE_TABLE:
3460		DMEMIT("%s ", cc->cipher_string);
3461
3462		if (cc->key_size > 0) {
3463			if (cc->key_string)
3464				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3465			else {
3466				for (i = 0; i < cc->key_size; i++) {
3467					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3468					       hex2asc(cc->key[i] & 0xf));
3469				}
3470			}
3471		} else
3472			DMEMIT("-");
3473
3474		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3475				cc->dev->name, (unsigned long long)cc->start);
3476
3477		num_feature_args += !!ti->num_discard_bios;
3478		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3479		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3480		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3481		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3482		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3483		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3484		if (cc->on_disk_tag_size)
3485			num_feature_args++;
3486		if (num_feature_args) {
3487			DMEMIT(" %d", num_feature_args);
3488			if (ti->num_discard_bios)
3489				DMEMIT(" allow_discards");
3490			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3491				DMEMIT(" same_cpu_crypt");
3492			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3493				DMEMIT(" submit_from_crypt_cpus");
3494			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3495				DMEMIT(" no_read_workqueue");
3496			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3497				DMEMIT(" no_write_workqueue");
3498			if (cc->on_disk_tag_size)
3499				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3500			if (cc->sector_size != (1 << SECTOR_SHIFT))
3501				DMEMIT(" sector_size:%d", cc->sector_size);
3502			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3503				DMEMIT(" iv_large_sectors");
3504		}
3505		break;
3506
3507	case STATUSTYPE_IMA:
3508		DMEMIT_TARGET_NAME_VERSION(ti->type);
3509		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3510		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3511		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3512		       'y' : 'n');
3513		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3514		       'y' : 'n');
3515		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3516		       'y' : 'n');
3517		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3518		       'y' : 'n');
3519
3520		if (cc->on_disk_tag_size)
3521			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3522			       cc->on_disk_tag_size, cc->cipher_auth);
3523		if (cc->sector_size != (1 << SECTOR_SHIFT))
3524			DMEMIT(",sector_size=%d", cc->sector_size);
3525		if (cc->cipher_string)
3526			DMEMIT(",cipher_string=%s", cc->cipher_string);
3527
3528		DMEMIT(",key_size=%u", cc->key_size);
3529		DMEMIT(",key_parts=%u", cc->key_parts);
3530		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3531		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3532		DMEMIT(";");
3533		break;
3534	}
3535}
3536
3537static void crypt_postsuspend(struct dm_target *ti)
3538{
3539	struct crypt_config *cc = ti->private;
3540
3541	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3542}
3543
3544static int crypt_preresume(struct dm_target *ti)
3545{
3546	struct crypt_config *cc = ti->private;
3547
3548	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3549		DMERR("aborting resume - crypt key is not set.");
3550		return -EAGAIN;
3551	}
3552
3553	return 0;
3554}
3555
3556static void crypt_resume(struct dm_target *ti)
3557{
3558	struct crypt_config *cc = ti->private;
3559
3560	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3561}
3562
3563/* Message interface
3564 *	key set <key>
3565 *	key wipe
3566 */
3567static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3568			 char *result, unsigned maxlen)
3569{
3570	struct crypt_config *cc = ti->private;
3571	int key_size, ret = -EINVAL;
3572
3573	if (argc < 2)
3574		goto error;
3575
3576	if (!strcasecmp(argv[0], "key")) {
3577		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3578			DMWARN("not suspended during key manipulation.");
3579			return -EINVAL;
3580		}
3581		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3582			/* The key size may not be changed. */
3583			key_size = get_key_size(&argv[2]);
3584			if (key_size < 0 || cc->key_size != key_size) {
3585				memset(argv[2], '0', strlen(argv[2]));
3586				return -EINVAL;
3587			}
3588
3589			ret = crypt_set_key(cc, argv[2]);
3590			if (ret)
3591				return ret;
3592			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3593				ret = cc->iv_gen_ops->init(cc);
3594			/* wipe the kernel key payload copy */
3595			if (cc->key_string)
3596				memset(cc->key, 0, cc->key_size * sizeof(u8));
3597			return ret;
3598		}
3599		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3600			return crypt_wipe_key(cc);
3601	}
3602
3603error:
3604	DMWARN("unrecognised message received.");
3605	return -EINVAL;
3606}
3607
3608static int crypt_iterate_devices(struct dm_target *ti,
3609				 iterate_devices_callout_fn fn, void *data)
3610{
3611	struct crypt_config *cc = ti->private;
3612
3613	return fn(ti, cc->dev, cc->start, ti->len, data);
3614}
3615
3616static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3617{
3618	struct crypt_config *cc = ti->private;
3619
3620	/*
3621	 * Unfortunate constraint that is required to avoid the potential
3622	 * for exceeding underlying device's max_segments limits -- due to
3623	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3624	 * bio that are not as physically contiguous as the original bio.
3625	 */
3626	limits->max_segment_size = PAGE_SIZE;
3627
3628	limits->logical_block_size =
3629		max_t(unsigned, limits->logical_block_size, cc->sector_size);
3630	limits->physical_block_size =
3631		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3632	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
3633	limits->dma_alignment = limits->logical_block_size - 1;
3634}
3635
3636static struct target_type crypt_target = {
3637	.name   = "crypt",
3638	.version = {1, 24, 0},
3639	.module = THIS_MODULE,
3640	.ctr    = crypt_ctr,
3641	.dtr    = crypt_dtr,
3642	.features = DM_TARGET_ZONED_HM,
3643	.report_zones = crypt_report_zones,
3644	.map    = crypt_map,
3645	.status = crypt_status,
3646	.postsuspend = crypt_postsuspend,
3647	.preresume = crypt_preresume,
3648	.resume = crypt_resume,
3649	.message = crypt_message,
3650	.iterate_devices = crypt_iterate_devices,
3651	.io_hints = crypt_io_hints,
3652};
3653
3654static int __init dm_crypt_init(void)
3655{
3656	int r;
3657
3658	r = dm_register_target(&crypt_target);
3659	if (r < 0)
3660		DMERR("register failed %d", r);
3661
3662	return r;
3663}
3664
3665static void __exit dm_crypt_exit(void)
3666{
3667	dm_unregister_target(&crypt_target);
3668}
3669
3670module_init(dm_crypt_init);
3671module_exit(dm_crypt_exit);
3672
3673MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3674MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3675MODULE_LICENSE("GPL");
v5.14.15
   1/*
   2 * Copyright (C) 2003 Jana Saout <jana@saout.de>
   3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
   4 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
   5 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
   6 *
   7 * This file is released under the GPL.
   8 */
   9
  10#include <linux/completion.h>
  11#include <linux/err.h>
  12#include <linux/module.h>
  13#include <linux/init.h>
  14#include <linux/kernel.h>
  15#include <linux/key.h>
  16#include <linux/bio.h>
  17#include <linux/blkdev.h>
 
  18#include <linux/mempool.h>
  19#include <linux/slab.h>
  20#include <linux/crypto.h>
  21#include <linux/workqueue.h>
  22#include <linux/kthread.h>
  23#include <linux/backing-dev.h>
  24#include <linux/atomic.h>
  25#include <linux/scatterlist.h>
  26#include <linux/rbtree.h>
  27#include <linux/ctype.h>
  28#include <asm/page.h>
  29#include <asm/unaligned.h>
  30#include <crypto/hash.h>
  31#include <crypto/md5.h>
  32#include <crypto/algapi.h>
  33#include <crypto/skcipher.h>
  34#include <crypto/aead.h>
  35#include <crypto/authenc.h>
  36#include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
  37#include <linux/key-type.h>
  38#include <keys/user-type.h>
  39#include <keys/encrypted-type.h>
  40#include <keys/trusted-type.h>
  41
  42#include <linux/device-mapper.h>
  43
 
 
  44#define DM_MSG_PREFIX "crypt"
  45
  46/*
  47 * context holding the current state of a multi-part conversion
  48 */
  49struct convert_context {
  50	struct completion restart;
  51	struct bio *bio_in;
  52	struct bio *bio_out;
  53	struct bvec_iter iter_in;
  54	struct bvec_iter iter_out;
  55	u64 cc_sector;
  56	atomic_t cc_pending;
  57	union {
  58		struct skcipher_request *req;
  59		struct aead_request *req_aead;
  60	} r;
  61
  62};
  63
  64/*
  65 * per bio private data
  66 */
  67struct dm_crypt_io {
  68	struct crypt_config *cc;
  69	struct bio *base_bio;
  70	u8 *integrity_metadata;
  71	bool integrity_metadata_from_pool;
  72	struct work_struct work;
  73	struct tasklet_struct tasklet;
  74
  75	struct convert_context ctx;
  76
  77	atomic_t io_pending;
  78	blk_status_t error;
  79	sector_t sector;
  80
  81	struct rb_node rb_node;
  82} CRYPTO_MINALIGN_ATTR;
  83
  84struct dm_crypt_request {
  85	struct convert_context *ctx;
  86	struct scatterlist sg_in[4];
  87	struct scatterlist sg_out[4];
  88	u64 iv_sector;
  89};
  90
  91struct crypt_config;
  92
  93struct crypt_iv_operations {
  94	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
  95		   const char *opts);
  96	void (*dtr)(struct crypt_config *cc);
  97	int (*init)(struct crypt_config *cc);
  98	int (*wipe)(struct crypt_config *cc);
  99	int (*generator)(struct crypt_config *cc, u8 *iv,
 100			 struct dm_crypt_request *dmreq);
 101	int (*post)(struct crypt_config *cc, u8 *iv,
 102		    struct dm_crypt_request *dmreq);
 103};
 104
 105struct iv_benbi_private {
 106	int shift;
 107};
 108
 109#define LMK_SEED_SIZE 64 /* hash + 0 */
 110struct iv_lmk_private {
 111	struct crypto_shash *hash_tfm;
 112	u8 *seed;
 113};
 114
 115#define TCW_WHITENING_SIZE 16
 116struct iv_tcw_private {
 117	struct crypto_shash *crc32_tfm;
 118	u8 *iv_seed;
 119	u8 *whitening;
 120};
 121
 122#define ELEPHANT_MAX_KEY_SIZE 32
 123struct iv_elephant_private {
 124	struct crypto_skcipher *tfm;
 125};
 126
 127/*
 128 * Crypt: maps a linear range of a block device
 129 * and encrypts / decrypts at the same time.
 130 */
 131enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
 132	     DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD,
 133	     DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE,
 134	     DM_CRYPT_WRITE_INLINE };
 135
 136enum cipher_flags {
 137	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
 138	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
 139	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
 140};
 141
 142/*
 143 * The fields in here must be read only after initialization.
 144 */
 145struct crypt_config {
 146	struct dm_dev *dev;
 147	sector_t start;
 148
 149	struct percpu_counter n_allocated_pages;
 150
 151	struct workqueue_struct *io_queue;
 152	struct workqueue_struct *crypt_queue;
 153
 154	spinlock_t write_thread_lock;
 155	struct task_struct *write_thread;
 156	struct rb_root write_tree;
 157
 158	char *cipher_string;
 159	char *cipher_auth;
 160	char *key_string;
 161
 162	const struct crypt_iv_operations *iv_gen_ops;
 163	union {
 164		struct iv_benbi_private benbi;
 165		struct iv_lmk_private lmk;
 166		struct iv_tcw_private tcw;
 167		struct iv_elephant_private elephant;
 168	} iv_gen_private;
 169	u64 iv_offset;
 170	unsigned int iv_size;
 171	unsigned short int sector_size;
 172	unsigned char sector_shift;
 173
 174	union {
 175		struct crypto_skcipher **tfms;
 176		struct crypto_aead **tfms_aead;
 177	} cipher_tfm;
 178	unsigned tfms_count;
 179	unsigned long cipher_flags;
 180
 181	/*
 182	 * Layout of each crypto request:
 183	 *
 184	 *   struct skcipher_request
 185	 *      context
 186	 *      padding
 187	 *   struct dm_crypt_request
 188	 *      padding
 189	 *   IV
 190	 *
 191	 * The padding is added so that dm_crypt_request and the IV are
 192	 * correctly aligned.
 193	 */
 194	unsigned int dmreq_start;
 195
 196	unsigned int per_bio_data_size;
 197
 198	unsigned long flags;
 199	unsigned int key_size;
 200	unsigned int key_parts;      /* independent parts in key buffer */
 201	unsigned int key_extra_size; /* additional keys length */
 202	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
 203
 204	unsigned int integrity_tag_size;
 205	unsigned int integrity_iv_size;
 206	unsigned int on_disk_tag_size;
 207
 208	/*
 209	 * pool for per bio private data, crypto requests,
 210	 * encryption requeusts/buffer pages and integrity tags
 211	 */
 212	unsigned tag_pool_max_sectors;
 213	mempool_t tag_pool;
 214	mempool_t req_pool;
 215	mempool_t page_pool;
 216
 217	struct bio_set bs;
 218	struct mutex bio_alloc_lock;
 219
 220	u8 *authenc_key; /* space for keys in authenc() format (if used) */
 221	u8 key[];
 222};
 223
 224#define MIN_IOS		64
 225#define MAX_TAG_SIZE	480
 226#define POOL_ENTRY_SIZE	512
 227
 228static DEFINE_SPINLOCK(dm_crypt_clients_lock);
 229static unsigned dm_crypt_clients_n = 0;
 230static volatile unsigned long dm_crypt_pages_per_client;
 231#define DM_CRYPT_MEMORY_PERCENT			2
 232#define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
 233
 234static void clone_init(struct dm_crypt_io *, struct bio *);
 235static void kcryptd_queue_crypt(struct dm_crypt_io *io);
 236static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
 237					     struct scatterlist *sg);
 238
 239static bool crypt_integrity_aead(struct crypt_config *cc);
 240
 241/*
 242 * Use this to access cipher attributes that are independent of the key.
 243 */
 244static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
 245{
 246	return cc->cipher_tfm.tfms[0];
 247}
 248
 249static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
 250{
 251	return cc->cipher_tfm.tfms_aead[0];
 252}
 253
 254/*
 255 * Different IV generation algorithms:
 256 *
 257 * plain: the initial vector is the 32-bit little-endian version of the sector
 258 *        number, padded with zeros if necessary.
 259 *
 260 * plain64: the initial vector is the 64-bit little-endian version of the sector
 261 *        number, padded with zeros if necessary.
 262 *
 263 * plain64be: the initial vector is the 64-bit big-endian version of the sector
 264 *        number, padded with zeros if necessary.
 265 *
 266 * essiv: "encrypted sector|salt initial vector", the sector number is
 267 *        encrypted with the bulk cipher using a salt as key. The salt
 268 *        should be derived from the bulk cipher's key via hashing.
 269 *
 270 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
 271 *        (needed for LRW-32-AES and possible other narrow block modes)
 272 *
 273 * null: the initial vector is always zero.  Provides compatibility with
 274 *       obsolete loop_fish2 devices.  Do not use for new devices.
 275 *
 276 * lmk:  Compatible implementation of the block chaining mode used
 277 *       by the Loop-AES block device encryption system
 278 *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
 279 *       It operates on full 512 byte sectors and uses CBC
 280 *       with an IV derived from the sector number, the data and
 281 *       optionally extra IV seed.
 282 *       This means that after decryption the first block
 283 *       of sector must be tweaked according to decrypted data.
 284 *       Loop-AES can use three encryption schemes:
 285 *         version 1: is plain aes-cbc mode
 286 *         version 2: uses 64 multikey scheme with lmk IV generator
 287 *         version 3: the same as version 2 with additional IV seed
 288 *                   (it uses 65 keys, last key is used as IV seed)
 289 *
 290 * tcw:  Compatible implementation of the block chaining mode used
 291 *       by the TrueCrypt device encryption system (prior to version 4.1).
 292 *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
 293 *       It operates on full 512 byte sectors and uses CBC
 294 *       with an IV derived from initial key and the sector number.
 295 *       In addition, whitening value is applied on every sector, whitening
 296 *       is calculated from initial key, sector number and mixed using CRC32.
 297 *       Note that this encryption scheme is vulnerable to watermarking attacks
 298 *       and should be used for old compatible containers access only.
 299 *
 300 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
 301 *        The IV is encrypted little-endian byte-offset (with the same key
 302 *        and cipher as the volume).
 303 *
 304 * elephant: The extended version of eboiv with additional Elephant diffuser
 305 *           used with Bitlocker CBC mode.
 306 *           This mode was used in older Windows systems
 307 *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
 308 */
 309
 310static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
 311			      struct dm_crypt_request *dmreq)
 312{
 313	memset(iv, 0, cc->iv_size);
 314	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
 315
 316	return 0;
 317}
 318
 319static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
 320				struct dm_crypt_request *dmreq)
 321{
 322	memset(iv, 0, cc->iv_size);
 323	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 324
 325	return 0;
 326}
 327
 328static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
 329				  struct dm_crypt_request *dmreq)
 330{
 331	memset(iv, 0, cc->iv_size);
 332	/* iv_size is at least of size u64; usually it is 16 bytes */
 333	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
 334
 335	return 0;
 336}
 337
 338static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
 339			      struct dm_crypt_request *dmreq)
 340{
 341	/*
 342	 * ESSIV encryption of the IV is now handled by the crypto API,
 343	 * so just pass the plain sector number here.
 344	 */
 345	memset(iv, 0, cc->iv_size);
 346	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 347
 348	return 0;
 349}
 350
 351static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
 352			      const char *opts)
 353{
 354	unsigned bs;
 355	int log;
 356
 357	if (crypt_integrity_aead(cc))
 358		bs = crypto_aead_blocksize(any_tfm_aead(cc));
 359	else
 360		bs = crypto_skcipher_blocksize(any_tfm(cc));
 361	log = ilog2(bs);
 362
 363	/* we need to calculate how far we must shift the sector count
 364	 * to get the cipher block count, we use this shift in _gen */
 365
 366	if (1 << log != bs) {
 367		ti->error = "cypher blocksize is not a power of 2";
 368		return -EINVAL;
 369	}
 370
 371	if (log > 9) {
 372		ti->error = "cypher blocksize is > 512";
 373		return -EINVAL;
 374	}
 375
 376	cc->iv_gen_private.benbi.shift = 9 - log;
 377
 378	return 0;
 379}
 380
 381static void crypt_iv_benbi_dtr(struct crypt_config *cc)
 382{
 383}
 384
 385static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
 386			      struct dm_crypt_request *dmreq)
 387{
 388	__be64 val;
 389
 390	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
 391
 392	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
 393	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
 394
 395	return 0;
 396}
 397
 398static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
 399			     struct dm_crypt_request *dmreq)
 400{
 401	memset(iv, 0, cc->iv_size);
 402
 403	return 0;
 404}
 405
 406static void crypt_iv_lmk_dtr(struct crypt_config *cc)
 407{
 408	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 409
 410	if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm))
 411		crypto_free_shash(lmk->hash_tfm);
 412	lmk->hash_tfm = NULL;
 413
 414	kfree_sensitive(lmk->seed);
 415	lmk->seed = NULL;
 416}
 417
 418static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
 419			    const char *opts)
 420{
 421	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 422
 423	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 424		ti->error = "Unsupported sector size for LMK";
 425		return -EINVAL;
 426	}
 427
 428	lmk->hash_tfm = crypto_alloc_shash("md5", 0,
 429					   CRYPTO_ALG_ALLOCATES_MEMORY);
 430	if (IS_ERR(lmk->hash_tfm)) {
 431		ti->error = "Error initializing LMK hash";
 432		return PTR_ERR(lmk->hash_tfm);
 433	}
 434
 435	/* No seed in LMK version 2 */
 436	if (cc->key_parts == cc->tfms_count) {
 437		lmk->seed = NULL;
 438		return 0;
 439	}
 440
 441	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
 442	if (!lmk->seed) {
 443		crypt_iv_lmk_dtr(cc);
 444		ti->error = "Error kmallocing seed storage in LMK";
 445		return -ENOMEM;
 446	}
 447
 448	return 0;
 449}
 450
 451static int crypt_iv_lmk_init(struct crypt_config *cc)
 452{
 453	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 454	int subkey_size = cc->key_size / cc->key_parts;
 455
 456	/* LMK seed is on the position of LMK_KEYS + 1 key */
 457	if (lmk->seed)
 458		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
 459		       crypto_shash_digestsize(lmk->hash_tfm));
 460
 461	return 0;
 462}
 463
 464static int crypt_iv_lmk_wipe(struct crypt_config *cc)
 465{
 466	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 467
 468	if (lmk->seed)
 469		memset(lmk->seed, 0, LMK_SEED_SIZE);
 470
 471	return 0;
 472}
 473
 474static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
 475			    struct dm_crypt_request *dmreq,
 476			    u8 *data)
 477{
 478	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 479	SHASH_DESC_ON_STACK(desc, lmk->hash_tfm);
 480	struct md5_state md5state;
 481	__le32 buf[4];
 482	int i, r;
 483
 484	desc->tfm = lmk->hash_tfm;
 485
 486	r = crypto_shash_init(desc);
 487	if (r)
 488		return r;
 489
 490	if (lmk->seed) {
 491		r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE);
 492		if (r)
 493			return r;
 494	}
 495
 496	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
 497	r = crypto_shash_update(desc, data + 16, 16 * 31);
 498	if (r)
 499		return r;
 500
 501	/* Sector is cropped to 56 bits here */
 502	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
 503	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
 504	buf[2] = cpu_to_le32(4024);
 505	buf[3] = 0;
 506	r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf));
 507	if (r)
 508		return r;
 509
 510	/* No MD5 padding here */
 511	r = crypto_shash_export(desc, &md5state);
 512	if (r)
 513		return r;
 514
 515	for (i = 0; i < MD5_HASH_WORDS; i++)
 516		__cpu_to_le32s(&md5state.hash[i]);
 517	memcpy(iv, &md5state.hash, cc->iv_size);
 518
 519	return 0;
 520}
 521
 522static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
 523			    struct dm_crypt_request *dmreq)
 524{
 525	struct scatterlist *sg;
 526	u8 *src;
 527	int r = 0;
 528
 529	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 530		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 531		src = kmap_atomic(sg_page(sg));
 532		r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
 533		kunmap_atomic(src);
 534	} else
 535		memset(iv, 0, cc->iv_size);
 536
 537	return r;
 538}
 539
 540static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
 541			     struct dm_crypt_request *dmreq)
 542{
 543	struct scatterlist *sg;
 544	u8 *dst;
 545	int r;
 546
 547	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
 548		return 0;
 549
 550	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 551	dst = kmap_atomic(sg_page(sg));
 552	r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
 553
 554	/* Tweak the first block of plaintext sector */
 555	if (!r)
 556		crypto_xor(dst + sg->offset, iv, cc->iv_size);
 557
 558	kunmap_atomic(dst);
 559	return r;
 560}
 561
 562static void crypt_iv_tcw_dtr(struct crypt_config *cc)
 563{
 564	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 565
 566	kfree_sensitive(tcw->iv_seed);
 567	tcw->iv_seed = NULL;
 568	kfree_sensitive(tcw->whitening);
 569	tcw->whitening = NULL;
 570
 571	if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm))
 572		crypto_free_shash(tcw->crc32_tfm);
 573	tcw->crc32_tfm = NULL;
 574}
 575
 576static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
 577			    const char *opts)
 578{
 579	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 580
 581	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 582		ti->error = "Unsupported sector size for TCW";
 583		return -EINVAL;
 584	}
 585
 586	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
 587		ti->error = "Wrong key size for TCW";
 588		return -EINVAL;
 589	}
 590
 591	tcw->crc32_tfm = crypto_alloc_shash("crc32", 0,
 592					    CRYPTO_ALG_ALLOCATES_MEMORY);
 593	if (IS_ERR(tcw->crc32_tfm)) {
 594		ti->error = "Error initializing CRC32 in TCW";
 595		return PTR_ERR(tcw->crc32_tfm);
 596	}
 597
 598	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
 599	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
 600	if (!tcw->iv_seed || !tcw->whitening) {
 601		crypt_iv_tcw_dtr(cc);
 602		ti->error = "Error allocating seed storage in TCW";
 603		return -ENOMEM;
 604	}
 605
 606	return 0;
 607}
 608
 609static int crypt_iv_tcw_init(struct crypt_config *cc)
 610{
 611	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 612	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
 613
 614	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
 615	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
 616	       TCW_WHITENING_SIZE);
 617
 618	return 0;
 619}
 620
 621static int crypt_iv_tcw_wipe(struct crypt_config *cc)
 622{
 623	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 624
 625	memset(tcw->iv_seed, 0, cc->iv_size);
 626	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
 627
 628	return 0;
 629}
 630
 631static int crypt_iv_tcw_whitening(struct crypt_config *cc,
 632				  struct dm_crypt_request *dmreq,
 633				  u8 *data)
 634{
 635	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 636	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 637	u8 buf[TCW_WHITENING_SIZE];
 638	SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm);
 639	int i, r;
 640
 641	/* xor whitening with sector number */
 642	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
 643	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
 644
 645	/* calculate crc32 for every 32bit part and xor it */
 646	desc->tfm = tcw->crc32_tfm;
 647	for (i = 0; i < 4; i++) {
 648		r = crypto_shash_init(desc);
 649		if (r)
 650			goto out;
 651		r = crypto_shash_update(desc, &buf[i * 4], 4);
 652		if (r)
 653			goto out;
 654		r = crypto_shash_final(desc, &buf[i * 4]);
 655		if (r)
 656			goto out;
 657	}
 658	crypto_xor(&buf[0], &buf[12], 4);
 659	crypto_xor(&buf[4], &buf[8], 4);
 660
 661	/* apply whitening (8 bytes) to whole sector */
 662	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
 663		crypto_xor(data + i * 8, buf, 8);
 664out:
 665	memzero_explicit(buf, sizeof(buf));
 666	return r;
 667}
 668
 669static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
 670			    struct dm_crypt_request *dmreq)
 671{
 672	struct scatterlist *sg;
 673	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 674	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 675	u8 *src;
 676	int r = 0;
 677
 678	/* Remove whitening from ciphertext */
 679	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
 680		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 681		src = kmap_atomic(sg_page(sg));
 682		r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
 683		kunmap_atomic(src);
 684	}
 685
 686	/* Calculate IV */
 687	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
 688	if (cc->iv_size > 8)
 689		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
 690			       cc->iv_size - 8);
 691
 692	return r;
 693}
 694
 695static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
 696			     struct dm_crypt_request *dmreq)
 697{
 698	struct scatterlist *sg;
 699	u8 *dst;
 700	int r;
 701
 702	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
 703		return 0;
 704
 705	/* Apply whitening on ciphertext */
 706	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 707	dst = kmap_atomic(sg_page(sg));
 708	r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
 709	kunmap_atomic(dst);
 710
 711	return r;
 712}
 713
 714static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
 715				struct dm_crypt_request *dmreq)
 716{
 717	/* Used only for writes, there must be an additional space to store IV */
 718	get_random_bytes(iv, cc->iv_size);
 719	return 0;
 720}
 721
 722static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
 723			    const char *opts)
 724{
 725	if (crypt_integrity_aead(cc)) {
 726		ti->error = "AEAD transforms not supported for EBOIV";
 727		return -EINVAL;
 728	}
 729
 730	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
 731		ti->error = "Block size of EBOIV cipher does "
 732			    "not match IV size of block cipher";
 733		return -EINVAL;
 734	}
 735
 736	return 0;
 737}
 738
 739static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
 740			    struct dm_crypt_request *dmreq)
 741{
 742	u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64));
 743	struct skcipher_request *req;
 744	struct scatterlist src, dst;
 745	DECLARE_CRYPTO_WAIT(wait);
 746	int err;
 747
 748	req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO);
 749	if (!req)
 750		return -ENOMEM;
 751
 752	memset(buf, 0, cc->iv_size);
 753	*(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
 754
 755	sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size);
 756	sg_init_one(&dst, iv, cc->iv_size);
 757	skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf);
 758	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
 759	err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 760	skcipher_request_free(req);
 761
 762	return err;
 763}
 764
 765static void crypt_iv_elephant_dtr(struct crypt_config *cc)
 766{
 767	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 768
 769	crypto_free_skcipher(elephant->tfm);
 770	elephant->tfm = NULL;
 771}
 772
 773static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti,
 774			    const char *opts)
 775{
 776	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 777	int r;
 778
 779	elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0,
 780					      CRYPTO_ALG_ALLOCATES_MEMORY);
 781	if (IS_ERR(elephant->tfm)) {
 782		r = PTR_ERR(elephant->tfm);
 783		elephant->tfm = NULL;
 784		return r;
 785	}
 786
 787	r = crypt_iv_eboiv_ctr(cc, ti, NULL);
 788	if (r)
 789		crypt_iv_elephant_dtr(cc);
 790	return r;
 791}
 792
 793static void diffuser_disk_to_cpu(u32 *d, size_t n)
 794{
 795#ifndef __LITTLE_ENDIAN
 796	int i;
 797
 798	for (i = 0; i < n; i++)
 799		d[i] = le32_to_cpu((__le32)d[i]);
 800#endif
 801}
 802
 803static void diffuser_cpu_to_disk(__le32 *d, size_t n)
 804{
 805#ifndef __LITTLE_ENDIAN
 806	int i;
 807
 808	for (i = 0; i < n; i++)
 809		d[i] = cpu_to_le32((u32)d[i]);
 810#endif
 811}
 812
 813static void diffuser_a_decrypt(u32 *d, size_t n)
 814{
 815	int i, i1, i2, i3;
 816
 817	for (i = 0; i < 5; i++) {
 818		i1 = 0;
 819		i2 = n - 2;
 820		i3 = n - 5;
 821
 822		while (i1 < (n - 1)) {
 823			d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
 824			i1++; i2++; i3++;
 825
 826			if (i3 >= n)
 827				i3 -= n;
 828
 829			d[i1] += d[i2] ^ d[i3];
 830			i1++; i2++; i3++;
 831
 832			if (i2 >= n)
 833				i2 -= n;
 834
 835			d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
 836			i1++; i2++; i3++;
 837
 838			d[i1] += d[i2] ^ d[i3];
 839			i1++; i2++; i3++;
 840		}
 841	}
 842}
 843
 844static void diffuser_a_encrypt(u32 *d, size_t n)
 845{
 846	int i, i1, i2, i3;
 847
 848	for (i = 0; i < 5; i++) {
 849		i1 = n - 1;
 850		i2 = n - 2 - 1;
 851		i3 = n - 5 - 1;
 852
 853		while (i1 > 0) {
 854			d[i1] -= d[i2] ^ d[i3];
 855			i1--; i2--; i3--;
 856
 857			d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19);
 858			i1--; i2--; i3--;
 859
 860			if (i2 < 0)
 861				i2 += n;
 862
 863			d[i1] -= d[i2] ^ d[i3];
 864			i1--; i2--; i3--;
 865
 866			if (i3 < 0)
 867				i3 += n;
 868
 869			d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23);
 870			i1--; i2--; i3--;
 871		}
 872	}
 873}
 874
 875static void diffuser_b_decrypt(u32 *d, size_t n)
 876{
 877	int i, i1, i2, i3;
 878
 879	for (i = 0; i < 3; i++) {
 880		i1 = 0;
 881		i2 = 2;
 882		i3 = 5;
 883
 884		while (i1 < (n - 1)) {
 885			d[i1] += d[i2] ^ d[i3];
 886			i1++; i2++; i3++;
 887
 888			d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
 889			i1++; i2++; i3++;
 890
 891			if (i2 >= n)
 892				i2 -= n;
 893
 894			d[i1] += d[i2] ^ d[i3];
 895			i1++; i2++; i3++;
 896
 897			if (i3 >= n)
 898				i3 -= n;
 899
 900			d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
 901			i1++; i2++; i3++;
 902		}
 903	}
 904}
 905
 906static void diffuser_b_encrypt(u32 *d, size_t n)
 907{
 908	int i, i1, i2, i3;
 909
 910	for (i = 0; i < 3; i++) {
 911		i1 = n - 1;
 912		i2 = 2 - 1;
 913		i3 = 5 - 1;
 914
 915		while (i1 > 0) {
 916			d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7);
 917			i1--; i2--; i3--;
 918
 919			if (i3 < 0)
 920				i3 += n;
 921
 922			d[i1] -= d[i2] ^ d[i3];
 923			i1--; i2--; i3--;
 924
 925			if (i2 < 0)
 926				i2 += n;
 927
 928			d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22);
 929			i1--; i2--; i3--;
 930
 931			d[i1] -= d[i2] ^ d[i3];
 932			i1--; i2--; i3--;
 933		}
 934	}
 935}
 936
 937static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq)
 938{
 939	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
 940	u8 *es, *ks, *data, *data2, *data_offset;
 941	struct skcipher_request *req;
 942	struct scatterlist *sg, *sg2, src, dst;
 943	DECLARE_CRYPTO_WAIT(wait);
 944	int i, r;
 945
 946	req = skcipher_request_alloc(elephant->tfm, GFP_NOIO);
 947	es = kzalloc(16, GFP_NOIO); /* Key for AES */
 948	ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */
 949
 950	if (!req || !es || !ks) {
 951		r = -ENOMEM;
 952		goto out;
 953	}
 954
 955	*(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size);
 956
 957	/* E(Ks, e(s)) */
 958	sg_init_one(&src, es, 16);
 959	sg_init_one(&dst, ks, 16);
 960	skcipher_request_set_crypt(req, &src, &dst, 16, NULL);
 961	skcipher_request_set_callback(req, 0, crypto_req_done, &wait);
 962	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 963	if (r)
 964		goto out;
 965
 966	/* E(Ks, e'(s)) */
 967	es[15] = 0x80;
 968	sg_init_one(&dst, &ks[16], 16);
 969	r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
 970	if (r)
 971		goto out;
 972
 973	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 974	data = kmap_atomic(sg_page(sg));
 975	data_offset = data + sg->offset;
 976
 977	/* Cannot modify original bio, copy to sg_out and apply Elephant to it */
 978	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 979		sg2 = crypt_get_sg_data(cc, dmreq->sg_in);
 980		data2 = kmap_atomic(sg_page(sg2));
 981		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
 982		kunmap_atomic(data2);
 983	}
 984
 985	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
 986		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
 987		diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 988		diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 989		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
 990	}
 991
 992	for (i = 0; i < (cc->sector_size / 32); i++)
 993		crypto_xor(data_offset + i * 32, ks, 32);
 994
 995	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 996		diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32));
 997		diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 998		diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32));
 999		diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32));
1000	}
1001
1002	kunmap_atomic(data);
1003out:
1004	kfree_sensitive(ks);
1005	kfree_sensitive(es);
1006	skcipher_request_free(req);
1007	return r;
1008}
1009
1010static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv,
1011			    struct dm_crypt_request *dmreq)
1012{
1013	int r;
1014
1015	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
1016		r = crypt_iv_elephant(cc, dmreq);
1017		if (r)
1018			return r;
1019	}
1020
1021	return crypt_iv_eboiv_gen(cc, iv, dmreq);
1022}
1023
1024static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv,
1025				  struct dm_crypt_request *dmreq)
1026{
1027	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
1028		return crypt_iv_elephant(cc, dmreq);
1029
1030	return 0;
1031}
1032
1033static int crypt_iv_elephant_init(struct crypt_config *cc)
1034{
1035	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1036	int key_offset = cc->key_size - cc->key_extra_size;
1037
1038	return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size);
1039}
1040
1041static int crypt_iv_elephant_wipe(struct crypt_config *cc)
1042{
1043	struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant;
1044	u8 key[ELEPHANT_MAX_KEY_SIZE];
1045
1046	memset(key, 0, cc->key_extra_size);
1047	return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size);
1048}
1049
1050static const struct crypt_iv_operations crypt_iv_plain_ops = {
1051	.generator = crypt_iv_plain_gen
1052};
1053
1054static const struct crypt_iv_operations crypt_iv_plain64_ops = {
1055	.generator = crypt_iv_plain64_gen
1056};
1057
1058static const struct crypt_iv_operations crypt_iv_plain64be_ops = {
1059	.generator = crypt_iv_plain64be_gen
1060};
1061
1062static const struct crypt_iv_operations crypt_iv_essiv_ops = {
1063	.generator = crypt_iv_essiv_gen
1064};
1065
1066static const struct crypt_iv_operations crypt_iv_benbi_ops = {
1067	.ctr	   = crypt_iv_benbi_ctr,
1068	.dtr	   = crypt_iv_benbi_dtr,
1069	.generator = crypt_iv_benbi_gen
1070};
1071
1072static const struct crypt_iv_operations crypt_iv_null_ops = {
1073	.generator = crypt_iv_null_gen
1074};
1075
1076static const struct crypt_iv_operations crypt_iv_lmk_ops = {
1077	.ctr	   = crypt_iv_lmk_ctr,
1078	.dtr	   = crypt_iv_lmk_dtr,
1079	.init	   = crypt_iv_lmk_init,
1080	.wipe	   = crypt_iv_lmk_wipe,
1081	.generator = crypt_iv_lmk_gen,
1082	.post	   = crypt_iv_lmk_post
1083};
1084
1085static const struct crypt_iv_operations crypt_iv_tcw_ops = {
1086	.ctr	   = crypt_iv_tcw_ctr,
1087	.dtr	   = crypt_iv_tcw_dtr,
1088	.init	   = crypt_iv_tcw_init,
1089	.wipe	   = crypt_iv_tcw_wipe,
1090	.generator = crypt_iv_tcw_gen,
1091	.post	   = crypt_iv_tcw_post
1092};
1093
1094static const struct crypt_iv_operations crypt_iv_random_ops = {
1095	.generator = crypt_iv_random_gen
1096};
1097
1098static const struct crypt_iv_operations crypt_iv_eboiv_ops = {
1099	.ctr	   = crypt_iv_eboiv_ctr,
1100	.generator = crypt_iv_eboiv_gen
1101};
1102
1103static const struct crypt_iv_operations crypt_iv_elephant_ops = {
1104	.ctr	   = crypt_iv_elephant_ctr,
1105	.dtr	   = crypt_iv_elephant_dtr,
1106	.init	   = crypt_iv_elephant_init,
1107	.wipe	   = crypt_iv_elephant_wipe,
1108	.generator = crypt_iv_elephant_gen,
1109	.post	   = crypt_iv_elephant_post
1110};
1111
1112/*
1113 * Integrity extensions
1114 */
1115static bool crypt_integrity_aead(struct crypt_config *cc)
1116{
1117	return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
1118}
1119
1120static bool crypt_integrity_hmac(struct crypt_config *cc)
1121{
1122	return crypt_integrity_aead(cc) && cc->key_mac_size;
1123}
1124
1125/* Get sg containing data */
1126static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
1127					     struct scatterlist *sg)
1128{
1129	if (unlikely(crypt_integrity_aead(cc)))
1130		return &sg[2];
1131
1132	return sg;
1133}
1134
1135static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio)
1136{
1137	struct bio_integrity_payload *bip;
1138	unsigned int tag_len;
1139	int ret;
1140
1141	if (!bio_sectors(bio) || !io->cc->on_disk_tag_size)
1142		return 0;
1143
1144	bip = bio_integrity_alloc(bio, GFP_NOIO, 1);
1145	if (IS_ERR(bip))
1146		return PTR_ERR(bip);
1147
1148	tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift);
1149
1150	bip->bip_iter.bi_size = tag_len;
1151	bip->bip_iter.bi_sector = io->cc->start + io->sector;
1152
1153	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1154				     tag_len, offset_in_page(io->integrity_metadata));
1155	if (unlikely(ret != tag_len))
1156		return -ENOMEM;
1157
1158	return 0;
1159}
1160
1161static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1162{
1163#ifdef CONFIG_BLK_DEV_INTEGRITY
1164	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1165	struct mapped_device *md = dm_table_get_md(ti->table);
1166
1167	/* From now we require underlying device with our integrity profile */
1168	if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) {
1169		ti->error = "Integrity profile not supported.";
1170		return -EINVAL;
1171	}
1172
1173	if (bi->tag_size != cc->on_disk_tag_size ||
1174	    bi->tuple_size != cc->on_disk_tag_size) {
1175		ti->error = "Integrity profile tag size mismatch.";
1176		return -EINVAL;
1177	}
1178	if (1 << bi->interval_exp != cc->sector_size) {
1179		ti->error = "Integrity profile sector size mismatch.";
1180		return -EINVAL;
1181	}
1182
1183	if (crypt_integrity_aead(cc)) {
1184		cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size;
1185		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1186		       cc->integrity_tag_size, cc->integrity_iv_size);
1187
1188		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1189			ti->error = "Integrity AEAD auth tag size is not supported.";
1190			return -EINVAL;
1191		}
1192	} else if (cc->integrity_iv_size)
1193		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1194		       cc->integrity_iv_size);
1195
1196	if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) {
1197		ti->error = "Not enough space for integrity tag in the profile.";
1198		return -EINVAL;
1199	}
1200
1201	return 0;
1202#else
1203	ti->error = "Integrity profile not supported.";
1204	return -EINVAL;
1205#endif
1206}
1207
1208static void crypt_convert_init(struct crypt_config *cc,
1209			       struct convert_context *ctx,
1210			       struct bio *bio_out, struct bio *bio_in,
1211			       sector_t sector)
1212{
1213	ctx->bio_in = bio_in;
1214	ctx->bio_out = bio_out;
1215	if (bio_in)
1216		ctx->iter_in = bio_in->bi_iter;
1217	if (bio_out)
1218		ctx->iter_out = bio_out->bi_iter;
1219	ctx->cc_sector = sector + cc->iv_offset;
1220	init_completion(&ctx->restart);
1221}
1222
1223static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc,
1224					     void *req)
1225{
1226	return (struct dm_crypt_request *)((char *)req + cc->dmreq_start);
1227}
1228
1229static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq)
1230{
1231	return (void *)((char *)dmreq - cc->dmreq_start);
1232}
1233
1234static u8 *iv_of_dmreq(struct crypt_config *cc,
1235		       struct dm_crypt_request *dmreq)
1236{
1237	if (crypt_integrity_aead(cc))
1238		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1239			crypto_aead_alignmask(any_tfm_aead(cc)) + 1);
1240	else
1241		return (u8 *)ALIGN((unsigned long)(dmreq + 1),
1242			crypto_skcipher_alignmask(any_tfm(cc)) + 1);
1243}
1244
1245static u8 *org_iv_of_dmreq(struct crypt_config *cc,
1246		       struct dm_crypt_request *dmreq)
1247{
1248	return iv_of_dmreq(cc, dmreq) + cc->iv_size;
1249}
1250
1251static __le64 *org_sector_of_dmreq(struct crypt_config *cc,
1252		       struct dm_crypt_request *dmreq)
1253{
1254	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size;
1255	return (__le64 *) ptr;
1256}
1257
1258static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1259		       struct dm_crypt_request *dmreq)
1260{
1261	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1262		  cc->iv_size + sizeof(uint64_t);
1263	return (unsigned int*)ptr;
1264}
1265
1266static void *tag_from_dmreq(struct crypt_config *cc,
1267				struct dm_crypt_request *dmreq)
1268{
1269	struct convert_context *ctx = dmreq->ctx;
1270	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1271
1272	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1273		cc->on_disk_tag_size];
1274}
1275
1276static void *iv_tag_from_dmreq(struct crypt_config *cc,
1277			       struct dm_crypt_request *dmreq)
1278{
1279	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1280}
1281
1282static int crypt_convert_block_aead(struct crypt_config *cc,
1283				     struct convert_context *ctx,
1284				     struct aead_request *req,
1285				     unsigned int tag_offset)
1286{
1287	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1288	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1289	struct dm_crypt_request *dmreq;
1290	u8 *iv, *org_iv, *tag_iv, *tag;
1291	__le64 *sector;
1292	int r = 0;
1293
1294	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1295
1296	/* Reject unexpected unaligned bio. */
1297	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1298		return -EIO;
1299
1300	dmreq = dmreq_of_req(cc, req);
1301	dmreq->iv_sector = ctx->cc_sector;
1302	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1303		dmreq->iv_sector >>= cc->sector_shift;
1304	dmreq->ctx = ctx;
1305
1306	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1307
1308	sector = org_sector_of_dmreq(cc, dmreq);
1309	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1310
1311	iv = iv_of_dmreq(cc, dmreq);
1312	org_iv = org_iv_of_dmreq(cc, dmreq);
1313	tag = tag_from_dmreq(cc, dmreq);
1314	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1315
1316	/* AEAD request:
1317	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1318	 *  | (authenticated) | (auth+encryption) |              |
1319	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1320	 */
1321	sg_init_table(dmreq->sg_in, 4);
1322	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1323	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1324	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1325	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1326
1327	sg_init_table(dmreq->sg_out, 4);
1328	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1329	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1330	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1331	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1332
1333	if (cc->iv_gen_ops) {
1334		/* For READs use IV stored in integrity metadata */
1335		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1336			memcpy(org_iv, tag_iv, cc->iv_size);
1337		} else {
1338			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1339			if (r < 0)
1340				return r;
1341			/* Store generated IV in integrity metadata */
1342			if (cc->integrity_iv_size)
1343				memcpy(tag_iv, org_iv, cc->iv_size);
1344		}
1345		/* Working copy of IV, to be modified in crypto API */
1346		memcpy(iv, org_iv, cc->iv_size);
1347	}
1348
1349	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1350	if (bio_data_dir(ctx->bio_in) == WRITE) {
1351		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1352				       cc->sector_size, iv);
1353		r = crypto_aead_encrypt(req);
1354		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size)
1355			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1356			       cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1357	} else {
1358		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1359				       cc->sector_size + cc->integrity_tag_size, iv);
1360		r = crypto_aead_decrypt(req);
1361	}
1362
1363	if (r == -EBADMSG) {
1364		char b[BDEVNAME_SIZE];
1365		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
1366			    (unsigned long long)le64_to_cpu(*sector));
 
 
 
1367	}
1368
1369	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1370		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1371
1372	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1373	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1374
1375	return r;
1376}
1377
1378static int crypt_convert_block_skcipher(struct crypt_config *cc,
1379					struct convert_context *ctx,
1380					struct skcipher_request *req,
1381					unsigned int tag_offset)
1382{
1383	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1384	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1385	struct scatterlist *sg_in, *sg_out;
1386	struct dm_crypt_request *dmreq;
1387	u8 *iv, *org_iv, *tag_iv;
1388	__le64 *sector;
1389	int r = 0;
1390
1391	/* Reject unexpected unaligned bio. */
1392	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1393		return -EIO;
1394
1395	dmreq = dmreq_of_req(cc, req);
1396	dmreq->iv_sector = ctx->cc_sector;
1397	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1398		dmreq->iv_sector >>= cc->sector_shift;
1399	dmreq->ctx = ctx;
1400
1401	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1402
1403	iv = iv_of_dmreq(cc, dmreq);
1404	org_iv = org_iv_of_dmreq(cc, dmreq);
1405	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1406
1407	sector = org_sector_of_dmreq(cc, dmreq);
1408	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1409
1410	/* For skcipher we use only the first sg item */
1411	sg_in  = &dmreq->sg_in[0];
1412	sg_out = &dmreq->sg_out[0];
1413
1414	sg_init_table(sg_in, 1);
1415	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1416
1417	sg_init_table(sg_out, 1);
1418	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1419
1420	if (cc->iv_gen_ops) {
1421		/* For READs use IV stored in integrity metadata */
1422		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1423			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1424		} else {
1425			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1426			if (r < 0)
1427				return r;
1428			/* Data can be already preprocessed in generator */
1429			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1430				sg_in = sg_out;
1431			/* Store generated IV in integrity metadata */
1432			if (cc->integrity_iv_size)
1433				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1434		}
1435		/* Working copy of IV, to be modified in crypto API */
1436		memcpy(iv, org_iv, cc->iv_size);
1437	}
1438
1439	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1440
1441	if (bio_data_dir(ctx->bio_in) == WRITE)
1442		r = crypto_skcipher_encrypt(req);
1443	else
1444		r = crypto_skcipher_decrypt(req);
1445
1446	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1447		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1448
1449	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1450	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1451
1452	return r;
1453}
1454
1455static void kcryptd_async_done(struct crypto_async_request *async_req,
1456			       int error);
1457
1458static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1459				     struct convert_context *ctx)
1460{
1461	unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1);
1462
1463	if (!ctx->r.req) {
1464		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1465		if (!ctx->r.req)
1466			return -ENOMEM;
1467	}
1468
1469	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1470
1471	/*
1472	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1473	 * requests if driver request queue is full.
1474	 */
1475	skcipher_request_set_callback(ctx->r.req,
1476	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1477	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1478
1479	return 0;
1480}
1481
1482static int crypt_alloc_req_aead(struct crypt_config *cc,
1483				 struct convert_context *ctx)
1484{
1485	if (!ctx->r.req_aead) {
1486		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1487		if (!ctx->r.req_aead)
1488			return -ENOMEM;
1489	}
1490
1491	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1492
1493	/*
1494	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1495	 * requests if driver request queue is full.
1496	 */
1497	aead_request_set_callback(ctx->r.req_aead,
1498	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1499	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1500
1501	return 0;
1502}
1503
1504static int crypt_alloc_req(struct crypt_config *cc,
1505			    struct convert_context *ctx)
1506{
1507	if (crypt_integrity_aead(cc))
1508		return crypt_alloc_req_aead(cc, ctx);
1509	else
1510		return crypt_alloc_req_skcipher(cc, ctx);
1511}
1512
1513static void crypt_free_req_skcipher(struct crypt_config *cc,
1514				    struct skcipher_request *req, struct bio *base_bio)
1515{
1516	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1517
1518	if ((struct skcipher_request *)(io + 1) != req)
1519		mempool_free(req, &cc->req_pool);
1520}
1521
1522static void crypt_free_req_aead(struct crypt_config *cc,
1523				struct aead_request *req, struct bio *base_bio)
1524{
1525	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1526
1527	if ((struct aead_request *)(io + 1) != req)
1528		mempool_free(req, &cc->req_pool);
1529}
1530
1531static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1532{
1533	if (crypt_integrity_aead(cc))
1534		crypt_free_req_aead(cc, req, base_bio);
1535	else
1536		crypt_free_req_skcipher(cc, req, base_bio);
1537}
1538
1539/*
1540 * Encrypt / decrypt data from one bio to another one (can be the same one)
1541 */
1542static blk_status_t crypt_convert(struct crypt_config *cc,
1543			 struct convert_context *ctx, bool atomic, bool reset_pending)
1544{
1545	unsigned int tag_offset = 0;
1546	unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT;
1547	int r;
1548
1549	/*
1550	 * if reset_pending is set we are dealing with the bio for the first time,
1551	 * else we're continuing to work on the previous bio, so don't mess with
1552	 * the cc_pending counter
1553	 */
1554	if (reset_pending)
1555		atomic_set(&ctx->cc_pending, 1);
1556
1557	while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) {
1558
1559		r = crypt_alloc_req(cc, ctx);
1560		if (r) {
1561			complete(&ctx->restart);
1562			return BLK_STS_DEV_RESOURCE;
1563		}
1564
1565		atomic_inc(&ctx->cc_pending);
1566
1567		if (crypt_integrity_aead(cc))
1568			r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset);
1569		else
1570			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset);
1571
1572		switch (r) {
1573		/*
1574		 * The request was queued by a crypto driver
1575		 * but the driver request queue is full, let's wait.
1576		 */
1577		case -EBUSY:
1578			if (in_interrupt()) {
1579				if (try_wait_for_completion(&ctx->restart)) {
1580					/*
1581					 * we don't have to block to wait for completion,
1582					 * so proceed
1583					 */
1584				} else {
1585					/*
1586					 * we can't wait for completion without blocking
1587					 * exit and continue processing in a workqueue
1588					 */
1589					ctx->r.req = NULL;
1590					ctx->cc_sector += sector_step;
1591					tag_offset++;
1592					return BLK_STS_DEV_RESOURCE;
1593				}
1594			} else {
1595				wait_for_completion(&ctx->restart);
1596			}
1597			reinit_completion(&ctx->restart);
1598			fallthrough;
1599		/*
1600		 * The request is queued and processed asynchronously,
1601		 * completion function kcryptd_async_done() will be called.
1602		 */
1603		case -EINPROGRESS:
1604			ctx->r.req = NULL;
1605			ctx->cc_sector += sector_step;
1606			tag_offset++;
1607			continue;
1608		/*
1609		 * The request was already processed (synchronously).
1610		 */
1611		case 0:
1612			atomic_dec(&ctx->cc_pending);
1613			ctx->cc_sector += sector_step;
1614			tag_offset++;
1615			if (!atomic)
1616				cond_resched();
1617			continue;
1618		/*
1619		 * There was a data integrity error.
1620		 */
1621		case -EBADMSG:
1622			atomic_dec(&ctx->cc_pending);
1623			return BLK_STS_PROTECTION;
1624		/*
1625		 * There was an error while processing the request.
1626		 */
1627		default:
1628			atomic_dec(&ctx->cc_pending);
1629			return BLK_STS_IOERR;
1630		}
1631	}
1632
1633	return 0;
1634}
1635
1636static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone);
1637
1638/*
1639 * Generate a new unfragmented bio with the given size
1640 * This should never violate the device limitations (but only because
1641 * max_segment_size is being constrained to PAGE_SIZE).
1642 *
1643 * This function may be called concurrently. If we allocate from the mempool
1644 * concurrently, there is a possibility of deadlock. For example, if we have
1645 * mempool of 256 pages, two processes, each wanting 256, pages allocate from
1646 * the mempool concurrently, it may deadlock in a situation where both processes
1647 * have allocated 128 pages and the mempool is exhausted.
1648 *
1649 * In order to avoid this scenario we allocate the pages under a mutex.
1650 *
1651 * In order to not degrade performance with excessive locking, we try
1652 * non-blocking allocations without a mutex first but on failure we fallback
1653 * to blocking allocations with a mutex.
1654 */
1655static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size)
1656{
1657	struct crypt_config *cc = io->cc;
1658	struct bio *clone;
1659	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1660	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1661	unsigned i, len, remaining_size;
1662	struct page *page;
1663
1664retry:
1665	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1666		mutex_lock(&cc->bio_alloc_lock);
1667
1668	clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs);
1669	if (!clone)
1670		goto out;
1671
1672	clone_init(io, clone);
1673
1674	remaining_size = size;
1675
1676	for (i = 0; i < nr_iovecs; i++) {
1677		page = mempool_alloc(&cc->page_pool, gfp_mask);
1678		if (!page) {
1679			crypt_free_buffer_pages(cc, clone);
1680			bio_put(clone);
1681			gfp_mask |= __GFP_DIRECT_RECLAIM;
1682			goto retry;
1683		}
1684
1685		len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size;
1686
1687		bio_add_page(clone, page, len, 0);
1688
1689		remaining_size -= len;
1690	}
1691
1692	/* Allocate space for integrity tags */
1693	if (dm_crypt_integrity_io_alloc(io, clone)) {
1694		crypt_free_buffer_pages(cc, clone);
1695		bio_put(clone);
1696		clone = NULL;
1697	}
1698out:
1699	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1700		mutex_unlock(&cc->bio_alloc_lock);
1701
1702	return clone;
1703}
1704
1705static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1706{
1707	struct bio_vec *bv;
1708	struct bvec_iter_all iter_all;
1709
1710	bio_for_each_segment_all(bv, clone, iter_all) {
1711		BUG_ON(!bv->bv_page);
1712		mempool_free(bv->bv_page, &cc->page_pool);
1713	}
1714}
1715
1716static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1717			  struct bio *bio, sector_t sector)
1718{
1719	io->cc = cc;
1720	io->base_bio = bio;
1721	io->sector = sector;
1722	io->error = 0;
1723	io->ctx.r.req = NULL;
1724	io->integrity_metadata = NULL;
1725	io->integrity_metadata_from_pool = false;
1726	atomic_set(&io->io_pending, 0);
1727}
1728
1729static void crypt_inc_pending(struct dm_crypt_io *io)
1730{
1731	atomic_inc(&io->io_pending);
1732}
1733
1734static void kcryptd_io_bio_endio(struct work_struct *work)
1735{
1736	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1737	bio_endio(io->base_bio);
1738}
1739
1740/*
1741 * One of the bios was finished. Check for completion of
1742 * the whole request and correctly clean up the buffer.
1743 */
1744static void crypt_dec_pending(struct dm_crypt_io *io)
1745{
1746	struct crypt_config *cc = io->cc;
1747	struct bio *base_bio = io->base_bio;
1748	blk_status_t error = io->error;
1749
1750	if (!atomic_dec_and_test(&io->io_pending))
1751		return;
1752
1753	if (io->ctx.r.req)
1754		crypt_free_req(cc, io->ctx.r.req, base_bio);
1755
1756	if (unlikely(io->integrity_metadata_from_pool))
1757		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1758	else
1759		kfree(io->integrity_metadata);
1760
1761	base_bio->bi_status = error;
1762
1763	/*
1764	 * If we are running this function from our tasklet,
1765	 * we can't call bio_endio() here, because it will call
1766	 * clone_endio() from dm.c, which in turn will
1767	 * free the current struct dm_crypt_io structure with
1768	 * our tasklet. In this case we need to delay bio_endio()
1769	 * execution to after the tasklet is done and dequeued.
1770	 */
1771	if (tasklet_trylock(&io->tasklet)) {
1772		tasklet_unlock(&io->tasklet);
1773		bio_endio(base_bio);
1774		return;
1775	}
1776
1777	INIT_WORK(&io->work, kcryptd_io_bio_endio);
1778	queue_work(cc->io_queue, &io->work);
1779}
1780
1781/*
1782 * kcryptd/kcryptd_io:
1783 *
1784 * Needed because it would be very unwise to do decryption in an
1785 * interrupt context.
1786 *
1787 * kcryptd performs the actual encryption or decryption.
1788 *
1789 * kcryptd_io performs the IO submission.
1790 *
1791 * They must be separated as otherwise the final stages could be
1792 * starved by new requests which can block in the first stages due
1793 * to memory allocation.
1794 *
1795 * The work is done per CPU global for all dm-crypt instances.
1796 * They should not depend on each other and do not block.
1797 */
1798static void crypt_endio(struct bio *clone)
1799{
1800	struct dm_crypt_io *io = clone->bi_private;
1801	struct crypt_config *cc = io->cc;
1802	unsigned rw = bio_data_dir(clone);
1803	blk_status_t error;
1804
1805	/*
1806	 * free the processed pages
1807	 */
1808	if (rw == WRITE)
1809		crypt_free_buffer_pages(cc, clone);
1810
1811	error = clone->bi_status;
1812	bio_put(clone);
1813
1814	if (rw == READ && !error) {
1815		kcryptd_queue_crypt(io);
1816		return;
1817	}
1818
1819	if (unlikely(error))
1820		io->error = error;
1821
1822	crypt_dec_pending(io);
1823}
1824
1825static void clone_init(struct dm_crypt_io *io, struct bio *clone)
1826{
1827	struct crypt_config *cc = io->cc;
1828
1829	clone->bi_private = io;
1830	clone->bi_end_io  = crypt_endio;
1831	bio_set_dev(clone, cc->dev->bdev);
1832	clone->bi_opf	  = io->base_bio->bi_opf;
1833}
1834
1835static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1836{
1837	struct crypt_config *cc = io->cc;
1838	struct bio *clone;
1839
1840	/*
1841	 * We need the original biovec array in order to decrypt
1842	 * the whole bio data *afterwards* -- thanks to immutable
1843	 * biovecs we don't need to worry about the block layer
1844	 * modifying the biovec array; so leverage bio_clone_fast().
1845	 */
1846	clone = bio_clone_fast(io->base_bio, gfp, &cc->bs);
1847	if (!clone)
1848		return 1;
 
 
1849
1850	crypt_inc_pending(io);
1851
1852	clone_init(io, clone);
1853	clone->bi_iter.bi_sector = cc->start + io->sector;
1854
1855	if (dm_crypt_integrity_io_alloc(io, clone)) {
1856		crypt_dec_pending(io);
1857		bio_put(clone);
1858		return 1;
1859	}
1860
1861	submit_bio_noacct(clone);
1862	return 0;
1863}
1864
1865static void kcryptd_io_read_work(struct work_struct *work)
1866{
1867	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1868
1869	crypt_inc_pending(io);
1870	if (kcryptd_io_read(io, GFP_NOIO))
1871		io->error = BLK_STS_RESOURCE;
1872	crypt_dec_pending(io);
1873}
1874
1875static void kcryptd_queue_read(struct dm_crypt_io *io)
1876{
1877	struct crypt_config *cc = io->cc;
1878
1879	INIT_WORK(&io->work, kcryptd_io_read_work);
1880	queue_work(cc->io_queue, &io->work);
1881}
1882
1883static void kcryptd_io_write(struct dm_crypt_io *io)
1884{
1885	struct bio *clone = io->ctx.bio_out;
1886
1887	submit_bio_noacct(clone);
1888}
1889
1890#define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1891
1892static int dmcrypt_write(void *data)
1893{
1894	struct crypt_config *cc = data;
1895	struct dm_crypt_io *io;
1896
1897	while (1) {
1898		struct rb_root write_tree;
1899		struct blk_plug plug;
1900
1901		spin_lock_irq(&cc->write_thread_lock);
1902continue_locked:
1903
1904		if (!RB_EMPTY_ROOT(&cc->write_tree))
1905			goto pop_from_list;
1906
1907		set_current_state(TASK_INTERRUPTIBLE);
1908
1909		spin_unlock_irq(&cc->write_thread_lock);
1910
1911		if (unlikely(kthread_should_stop())) {
1912			set_current_state(TASK_RUNNING);
1913			break;
1914		}
1915
1916		schedule();
1917
1918		set_current_state(TASK_RUNNING);
1919		spin_lock_irq(&cc->write_thread_lock);
1920		goto continue_locked;
1921
1922pop_from_list:
1923		write_tree = cc->write_tree;
1924		cc->write_tree = RB_ROOT;
1925		spin_unlock_irq(&cc->write_thread_lock);
1926
1927		BUG_ON(rb_parent(write_tree.rb_node));
1928
1929		/*
1930		 * Note: we cannot walk the tree here with rb_next because
1931		 * the structures may be freed when kcryptd_io_write is called.
1932		 */
1933		blk_start_plug(&plug);
1934		do {
1935			io = crypt_io_from_node(rb_first(&write_tree));
1936			rb_erase(&io->rb_node, &write_tree);
1937			kcryptd_io_write(io);
1938		} while (!RB_EMPTY_ROOT(&write_tree));
1939		blk_finish_plug(&plug);
1940	}
1941	return 0;
1942}
1943
1944static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1945{
1946	struct bio *clone = io->ctx.bio_out;
1947	struct crypt_config *cc = io->cc;
1948	unsigned long flags;
1949	sector_t sector;
1950	struct rb_node **rbp, *parent;
1951
1952	if (unlikely(io->error)) {
1953		crypt_free_buffer_pages(cc, clone);
1954		bio_put(clone);
1955		crypt_dec_pending(io);
1956		return;
1957	}
1958
1959	/* crypt_convert should have filled the clone bio */
1960	BUG_ON(io->ctx.iter_out.bi_size);
1961
1962	clone->bi_iter.bi_sector = cc->start + io->sector;
1963
1964	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
1965	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
1966		submit_bio_noacct(clone);
1967		return;
1968	}
1969
1970	spin_lock_irqsave(&cc->write_thread_lock, flags);
1971	if (RB_EMPTY_ROOT(&cc->write_tree))
1972		wake_up_process(cc->write_thread);
1973	rbp = &cc->write_tree.rb_node;
1974	parent = NULL;
1975	sector = io->sector;
1976	while (*rbp) {
1977		parent = *rbp;
1978		if (sector < crypt_io_from_node(parent)->sector)
1979			rbp = &(*rbp)->rb_left;
1980		else
1981			rbp = &(*rbp)->rb_right;
1982	}
1983	rb_link_node(&io->rb_node, parent, rbp);
1984	rb_insert_color(&io->rb_node, &cc->write_tree);
1985	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
1986}
1987
1988static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
1989				       struct convert_context *ctx)
1990
1991{
1992	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
1993		return false;
1994
1995	/*
1996	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
1997	 * constraints so they do not need to be issued inline by
1998	 * kcryptd_crypt_write_convert().
1999	 */
2000	switch (bio_op(ctx->bio_in)) {
2001	case REQ_OP_WRITE:
2002	case REQ_OP_WRITE_SAME:
2003	case REQ_OP_WRITE_ZEROES:
2004		return true;
2005	default:
2006		return false;
2007	}
2008}
2009
2010static void kcryptd_crypt_write_continue(struct work_struct *work)
2011{
2012	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2013	struct crypt_config *cc = io->cc;
2014	struct convert_context *ctx = &io->ctx;
2015	int crypt_finished;
2016	sector_t sector = io->sector;
2017	blk_status_t r;
2018
2019	wait_for_completion(&ctx->restart);
2020	reinit_completion(&ctx->restart);
2021
2022	r = crypt_convert(cc, &io->ctx, true, false);
2023	if (r)
2024		io->error = r;
2025	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2026	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2027		/* Wait for completion signaled by kcryptd_async_done() */
2028		wait_for_completion(&ctx->restart);
2029		crypt_finished = 1;
2030	}
2031
2032	/* Encryption was already finished, submit io now */
2033	if (crypt_finished) {
2034		kcryptd_crypt_write_io_submit(io, 0);
2035		io->sector = sector;
2036	}
2037
2038	crypt_dec_pending(io);
2039}
2040
2041static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2042{
2043	struct crypt_config *cc = io->cc;
2044	struct convert_context *ctx = &io->ctx;
2045	struct bio *clone;
2046	int crypt_finished;
2047	sector_t sector = io->sector;
2048	blk_status_t r;
2049
2050	/*
2051	 * Prevent io from disappearing until this function completes.
2052	 */
2053	crypt_inc_pending(io);
2054	crypt_convert_init(cc, ctx, NULL, io->base_bio, sector);
2055
2056	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2057	if (unlikely(!clone)) {
2058		io->error = BLK_STS_IOERR;
2059		goto dec;
2060	}
2061
2062	io->ctx.bio_out = clone;
2063	io->ctx.iter_out = clone->bi_iter;
2064
2065	sector += bio_sectors(clone);
2066
2067	crypt_inc_pending(io);
2068	r = crypt_convert(cc, ctx,
2069			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2070	/*
2071	 * Crypto API backlogged the request, because its queue was full
2072	 * and we're in softirq context, so continue from a workqueue
2073	 * (TODO: is it actually possible to be in softirq in the write path?)
2074	 */
2075	if (r == BLK_STS_DEV_RESOURCE) {
2076		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2077		queue_work(cc->crypt_queue, &io->work);
2078		return;
2079	}
2080	if (r)
2081		io->error = r;
2082	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2083	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2084		/* Wait for completion signaled by kcryptd_async_done() */
2085		wait_for_completion(&ctx->restart);
2086		crypt_finished = 1;
2087	}
2088
2089	/* Encryption was already finished, submit io now */
2090	if (crypt_finished) {
2091		kcryptd_crypt_write_io_submit(io, 0);
2092		io->sector = sector;
2093	}
2094
2095dec:
2096	crypt_dec_pending(io);
2097}
2098
2099static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2100{
2101	crypt_dec_pending(io);
2102}
2103
2104static void kcryptd_crypt_read_continue(struct work_struct *work)
2105{
2106	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2107	struct crypt_config *cc = io->cc;
2108	blk_status_t r;
2109
2110	wait_for_completion(&io->ctx.restart);
2111	reinit_completion(&io->ctx.restart);
2112
2113	r = crypt_convert(cc, &io->ctx, true, false);
2114	if (r)
2115		io->error = r;
2116
2117	if (atomic_dec_and_test(&io->ctx.cc_pending))
2118		kcryptd_crypt_read_done(io);
2119
2120	crypt_dec_pending(io);
2121}
2122
2123static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2124{
2125	struct crypt_config *cc = io->cc;
2126	blk_status_t r;
2127
2128	crypt_inc_pending(io);
2129
2130	crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2131			   io->sector);
2132
2133	r = crypt_convert(cc, &io->ctx,
2134			  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2135	/*
2136	 * Crypto API backlogged the request, because its queue was full
2137	 * and we're in softirq context, so continue from a workqueue
2138	 */
2139	if (r == BLK_STS_DEV_RESOURCE) {
2140		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2141		queue_work(cc->crypt_queue, &io->work);
2142		return;
2143	}
2144	if (r)
2145		io->error = r;
2146
2147	if (atomic_dec_and_test(&io->ctx.cc_pending))
2148		kcryptd_crypt_read_done(io);
2149
2150	crypt_dec_pending(io);
2151}
2152
2153static void kcryptd_async_done(struct crypto_async_request *async_req,
2154			       int error)
2155{
2156	struct dm_crypt_request *dmreq = async_req->data;
2157	struct convert_context *ctx = dmreq->ctx;
2158	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2159	struct crypt_config *cc = io->cc;
2160
2161	/*
2162	 * A request from crypto driver backlog is going to be processed now,
2163	 * finish the completion and continue in crypt_convert().
2164	 * (Callback will be called for the second time for this request.)
2165	 */
2166	if (error == -EINPROGRESS) {
2167		complete(&ctx->restart);
2168		return;
2169	}
2170
2171	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2172		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2173
2174	if (error == -EBADMSG) {
2175		char b[BDEVNAME_SIZE];
2176		DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", bio_devname(ctx->bio_in, b),
2177			    (unsigned long long)le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)));
 
 
 
2178		io->error = BLK_STS_PROTECTION;
2179	} else if (error < 0)
2180		io->error = BLK_STS_IOERR;
2181
2182	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2183
2184	if (!atomic_dec_and_test(&ctx->cc_pending))
2185		return;
2186
2187	/*
2188	 * The request is fully completed: for inline writes, let
2189	 * kcryptd_crypt_write_convert() do the IO submission.
2190	 */
2191	if (bio_data_dir(io->base_bio) == READ) {
2192		kcryptd_crypt_read_done(io);
2193		return;
2194	}
2195
2196	if (kcryptd_crypt_write_inline(cc, ctx)) {
2197		complete(&ctx->restart);
2198		return;
2199	}
2200
2201	kcryptd_crypt_write_io_submit(io, 1);
2202}
2203
2204static void kcryptd_crypt(struct work_struct *work)
2205{
2206	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2207
2208	if (bio_data_dir(io->base_bio) == READ)
2209		kcryptd_crypt_read_convert(io);
2210	else
2211		kcryptd_crypt_write_convert(io);
2212}
2213
2214static void kcryptd_crypt_tasklet(unsigned long work)
2215{
2216	kcryptd_crypt((struct work_struct *)work);
2217}
2218
2219static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2220{
2221	struct crypt_config *cc = io->cc;
2222
2223	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2224	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2225		/*
2226		 * in_irq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2227		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2228		 * it is being executed with irqs disabled.
2229		 */
2230		if (in_irq() || irqs_disabled()) {
2231			tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work);
2232			tasklet_schedule(&io->tasklet);
2233			return;
2234		}
2235
2236		kcryptd_crypt(&io->work);
2237		return;
2238	}
2239
2240	INIT_WORK(&io->work, kcryptd_crypt);
2241	queue_work(cc->crypt_queue, &io->work);
2242}
2243
2244static void crypt_free_tfms_aead(struct crypt_config *cc)
2245{
2246	if (!cc->cipher_tfm.tfms_aead)
2247		return;
2248
2249	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2250		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2251		cc->cipher_tfm.tfms_aead[0] = NULL;
2252	}
2253
2254	kfree(cc->cipher_tfm.tfms_aead);
2255	cc->cipher_tfm.tfms_aead = NULL;
2256}
2257
2258static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2259{
2260	unsigned i;
2261
2262	if (!cc->cipher_tfm.tfms)
2263		return;
2264
2265	for (i = 0; i < cc->tfms_count; i++)
2266		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2267			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2268			cc->cipher_tfm.tfms[i] = NULL;
2269		}
2270
2271	kfree(cc->cipher_tfm.tfms);
2272	cc->cipher_tfm.tfms = NULL;
2273}
2274
2275static void crypt_free_tfms(struct crypt_config *cc)
2276{
2277	if (crypt_integrity_aead(cc))
2278		crypt_free_tfms_aead(cc);
2279	else
2280		crypt_free_tfms_skcipher(cc);
2281}
2282
2283static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2284{
2285	unsigned i;
2286	int err;
2287
2288	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2289				      sizeof(struct crypto_skcipher *),
2290				      GFP_KERNEL);
2291	if (!cc->cipher_tfm.tfms)
2292		return -ENOMEM;
2293
2294	for (i = 0; i < cc->tfms_count; i++) {
2295		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2296						CRYPTO_ALG_ALLOCATES_MEMORY);
2297		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2298			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2299			crypt_free_tfms(cc);
2300			return err;
2301		}
2302	}
2303
2304	/*
2305	 * dm-crypt performance can vary greatly depending on which crypto
2306	 * algorithm implementation is used.  Help people debug performance
2307	 * problems by logging the ->cra_driver_name.
2308	 */
2309	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2310	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2311	return 0;
2312}
2313
2314static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2315{
2316	int err;
2317
2318	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2319	if (!cc->cipher_tfm.tfms)
2320		return -ENOMEM;
2321
2322	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2323						CRYPTO_ALG_ALLOCATES_MEMORY);
2324	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2325		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2326		crypt_free_tfms(cc);
2327		return err;
2328	}
2329
2330	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2331	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2332	return 0;
2333}
2334
2335static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2336{
2337	if (crypt_integrity_aead(cc))
2338		return crypt_alloc_tfms_aead(cc, ciphermode);
2339	else
2340		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2341}
2342
2343static unsigned crypt_subkey_size(struct crypt_config *cc)
2344{
2345	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2346}
2347
2348static unsigned crypt_authenckey_size(struct crypt_config *cc)
2349{
2350	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2351}
2352
2353/*
2354 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2355 * the key must be for some reason in special format.
2356 * This funcion converts cc->key to this special format.
2357 */
2358static void crypt_copy_authenckey(char *p, const void *key,
2359				  unsigned enckeylen, unsigned authkeylen)
2360{
2361	struct crypto_authenc_key_param *param;
2362	struct rtattr *rta;
2363
2364	rta = (struct rtattr *)p;
2365	param = RTA_DATA(rta);
2366	param->enckeylen = cpu_to_be32(enckeylen);
2367	rta->rta_len = RTA_LENGTH(sizeof(*param));
2368	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2369	p += RTA_SPACE(sizeof(*param));
2370	memcpy(p, key + enckeylen, authkeylen);
2371	p += authkeylen;
2372	memcpy(p, key, enckeylen);
2373}
2374
2375static int crypt_setkey(struct crypt_config *cc)
2376{
2377	unsigned subkey_size;
2378	int err = 0, i, r;
2379
2380	/* Ignore extra keys (which are used for IV etc) */
2381	subkey_size = crypt_subkey_size(cc);
2382
2383	if (crypt_integrity_hmac(cc)) {
2384		if (subkey_size < cc->key_mac_size)
2385			return -EINVAL;
2386
2387		crypt_copy_authenckey(cc->authenc_key, cc->key,
2388				      subkey_size - cc->key_mac_size,
2389				      cc->key_mac_size);
2390	}
2391
2392	for (i = 0; i < cc->tfms_count; i++) {
2393		if (crypt_integrity_hmac(cc))
2394			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2395				cc->authenc_key, crypt_authenckey_size(cc));
2396		else if (crypt_integrity_aead(cc))
2397			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2398					       cc->key + (i * subkey_size),
2399					       subkey_size);
2400		else
2401			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2402						   cc->key + (i * subkey_size),
2403						   subkey_size);
2404		if (r)
2405			err = r;
2406	}
2407
2408	if (crypt_integrity_hmac(cc))
2409		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2410
2411	return err;
2412}
2413
2414#ifdef CONFIG_KEYS
2415
2416static bool contains_whitespace(const char *str)
2417{
2418	while (*str)
2419		if (isspace(*str++))
2420			return true;
2421	return false;
2422}
2423
2424static int set_key_user(struct crypt_config *cc, struct key *key)
2425{
2426	const struct user_key_payload *ukp;
2427
2428	ukp = user_key_payload_locked(key);
2429	if (!ukp)
2430		return -EKEYREVOKED;
2431
2432	if (cc->key_size != ukp->datalen)
2433		return -EINVAL;
2434
2435	memcpy(cc->key, ukp->data, cc->key_size);
2436
2437	return 0;
2438}
2439
2440static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2441{
2442	const struct encrypted_key_payload *ekp;
2443
2444	ekp = key->payload.data[0];
2445	if (!ekp)
2446		return -EKEYREVOKED;
2447
2448	if (cc->key_size != ekp->decrypted_datalen)
2449		return -EINVAL;
2450
2451	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2452
2453	return 0;
2454}
2455
2456static int set_key_trusted(struct crypt_config *cc, struct key *key)
2457{
2458	const struct trusted_key_payload *tkp;
2459
2460	tkp = key->payload.data[0];
2461	if (!tkp)
2462		return -EKEYREVOKED;
2463
2464	if (cc->key_size != tkp->key_len)
2465		return -EINVAL;
2466
2467	memcpy(cc->key, tkp->key, cc->key_size);
2468
2469	return 0;
2470}
2471
2472static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2473{
2474	char *new_key_string, *key_desc;
2475	int ret;
2476	struct key_type *type;
2477	struct key *key;
2478	int (*set_key)(struct crypt_config *cc, struct key *key);
2479
2480	/*
2481	 * Reject key_string with whitespace. dm core currently lacks code for
2482	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2483	 */
2484	if (contains_whitespace(key_string)) {
2485		DMERR("whitespace chars not allowed in key string");
2486		return -EINVAL;
2487	}
2488
2489	/* look for next ':' separating key_type from key_description */
2490	key_desc = strpbrk(key_string, ":");
2491	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2492		return -EINVAL;
2493
2494	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2495		type = &key_type_logon;
2496		set_key = set_key_user;
2497	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2498		type = &key_type_user;
2499		set_key = set_key_user;
2500	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2501		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2502		type = &key_type_encrypted;
2503		set_key = set_key_encrypted;
2504	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2505	           !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2506		type = &key_type_trusted;
2507		set_key = set_key_trusted;
2508	} else {
2509		return -EINVAL;
2510	}
2511
2512	new_key_string = kstrdup(key_string, GFP_KERNEL);
2513	if (!new_key_string)
2514		return -ENOMEM;
2515
2516	key = request_key(type, key_desc + 1, NULL);
2517	if (IS_ERR(key)) {
2518		kfree_sensitive(new_key_string);
2519		return PTR_ERR(key);
2520	}
2521
2522	down_read(&key->sem);
2523
2524	ret = set_key(cc, key);
2525	if (ret < 0) {
2526		up_read(&key->sem);
2527		key_put(key);
2528		kfree_sensitive(new_key_string);
2529		return ret;
2530	}
2531
2532	up_read(&key->sem);
2533	key_put(key);
2534
2535	/* clear the flag since following operations may invalidate previously valid key */
2536	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2537
2538	ret = crypt_setkey(cc);
2539
2540	if (!ret) {
2541		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2542		kfree_sensitive(cc->key_string);
2543		cc->key_string = new_key_string;
2544	} else
2545		kfree_sensitive(new_key_string);
2546
2547	return ret;
2548}
2549
2550static int get_key_size(char **key_string)
2551{
2552	char *colon, dummy;
2553	int ret;
2554
2555	if (*key_string[0] != ':')
2556		return strlen(*key_string) >> 1;
2557
2558	/* look for next ':' in key string */
2559	colon = strpbrk(*key_string + 1, ":");
2560	if (!colon)
2561		return -EINVAL;
2562
2563	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2564		return -EINVAL;
2565
2566	*key_string = colon;
2567
2568	/* remaining key string should be :<logon|user>:<key_desc> */
2569
2570	return ret;
2571}
2572
2573#else
2574
2575static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2576{
2577	return -EINVAL;
2578}
2579
2580static int get_key_size(char **key_string)
2581{
2582	return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1;
2583}
2584
2585#endif /* CONFIG_KEYS */
2586
2587static int crypt_set_key(struct crypt_config *cc, char *key)
2588{
2589	int r = -EINVAL;
2590	int key_string_len = strlen(key);
2591
2592	/* Hyphen (which gives a key_size of zero) means there is no key. */
2593	if (!cc->key_size && strcmp(key, "-"))
2594		goto out;
2595
2596	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2597	if (key[0] == ':') {
2598		r = crypt_set_keyring_key(cc, key + 1);
2599		goto out;
2600	}
2601
2602	/* clear the flag since following operations may invalidate previously valid key */
2603	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2604
2605	/* wipe references to any kernel keyring key */
2606	kfree_sensitive(cc->key_string);
2607	cc->key_string = NULL;
2608
2609	/* Decode key from its hex representation. */
2610	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2611		goto out;
2612
2613	r = crypt_setkey(cc);
2614	if (!r)
2615		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2616
2617out:
2618	/* Hex key string not needed after here, so wipe it. */
2619	memset(key, '0', key_string_len);
2620
2621	return r;
2622}
2623
2624static int crypt_wipe_key(struct crypt_config *cc)
2625{
2626	int r;
2627
2628	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2629	get_random_bytes(&cc->key, cc->key_size);
2630
2631	/* Wipe IV private keys */
2632	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2633		r = cc->iv_gen_ops->wipe(cc);
2634		if (r)
2635			return r;
2636	}
2637
2638	kfree_sensitive(cc->key_string);
2639	cc->key_string = NULL;
2640	r = crypt_setkey(cc);
2641	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2642
2643	return r;
2644}
2645
2646static void crypt_calculate_pages_per_client(void)
2647{
2648	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2649
2650	if (!dm_crypt_clients_n)
2651		return;
2652
2653	pages /= dm_crypt_clients_n;
2654	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2655		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2656	dm_crypt_pages_per_client = pages;
2657}
2658
2659static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2660{
2661	struct crypt_config *cc = pool_data;
2662	struct page *page;
2663
2664	/*
2665	 * Note, percpu_counter_read_positive() may over (and under) estimate
2666	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2667	 * but avoids potential spinlock contention of an exact result.
2668	 */
2669	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2670	    likely(gfp_mask & __GFP_NORETRY))
2671		return NULL;
2672
2673	page = alloc_page(gfp_mask);
2674	if (likely(page != NULL))
2675		percpu_counter_add(&cc->n_allocated_pages, 1);
2676
2677	return page;
2678}
2679
2680static void crypt_page_free(void *page, void *pool_data)
2681{
2682	struct crypt_config *cc = pool_data;
2683
2684	__free_page(page);
2685	percpu_counter_sub(&cc->n_allocated_pages, 1);
2686}
2687
2688static void crypt_dtr(struct dm_target *ti)
2689{
2690	struct crypt_config *cc = ti->private;
2691
2692	ti->private = NULL;
2693
2694	if (!cc)
2695		return;
2696
2697	if (cc->write_thread)
2698		kthread_stop(cc->write_thread);
2699
2700	if (cc->io_queue)
2701		destroy_workqueue(cc->io_queue);
2702	if (cc->crypt_queue)
2703		destroy_workqueue(cc->crypt_queue);
2704
2705	crypt_free_tfms(cc);
2706
2707	bioset_exit(&cc->bs);
2708
2709	mempool_exit(&cc->page_pool);
2710	mempool_exit(&cc->req_pool);
2711	mempool_exit(&cc->tag_pool);
2712
2713	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2714	percpu_counter_destroy(&cc->n_allocated_pages);
2715
2716	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2717		cc->iv_gen_ops->dtr(cc);
2718
2719	if (cc->dev)
2720		dm_put_device(ti, cc->dev);
2721
2722	kfree_sensitive(cc->cipher_string);
2723	kfree_sensitive(cc->key_string);
2724	kfree_sensitive(cc->cipher_auth);
2725	kfree_sensitive(cc->authenc_key);
2726
2727	mutex_destroy(&cc->bio_alloc_lock);
2728
2729	/* Must zero key material before freeing */
2730	kfree_sensitive(cc);
2731
2732	spin_lock(&dm_crypt_clients_lock);
2733	WARN_ON(!dm_crypt_clients_n);
2734	dm_crypt_clients_n--;
2735	crypt_calculate_pages_per_client();
2736	spin_unlock(&dm_crypt_clients_lock);
 
 
2737}
2738
2739static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2740{
2741	struct crypt_config *cc = ti->private;
2742
2743	if (crypt_integrity_aead(cc))
2744		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2745	else
2746		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2747
2748	if (cc->iv_size)
2749		/* at least a 64 bit sector number should fit in our buffer */
2750		cc->iv_size = max(cc->iv_size,
2751				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2752	else if (ivmode) {
2753		DMWARN("Selected cipher does not support IVs");
2754		ivmode = NULL;
2755	}
2756
2757	/* Choose ivmode, see comments at iv code. */
2758	if (ivmode == NULL)
2759		cc->iv_gen_ops = NULL;
2760	else if (strcmp(ivmode, "plain") == 0)
2761		cc->iv_gen_ops = &crypt_iv_plain_ops;
2762	else if (strcmp(ivmode, "plain64") == 0)
2763		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2764	else if (strcmp(ivmode, "plain64be") == 0)
2765		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2766	else if (strcmp(ivmode, "essiv") == 0)
2767		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2768	else if (strcmp(ivmode, "benbi") == 0)
2769		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2770	else if (strcmp(ivmode, "null") == 0)
2771		cc->iv_gen_ops = &crypt_iv_null_ops;
2772	else if (strcmp(ivmode, "eboiv") == 0)
2773		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2774	else if (strcmp(ivmode, "elephant") == 0) {
2775		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2776		cc->key_parts = 2;
2777		cc->key_extra_size = cc->key_size / 2;
2778		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2779			return -EINVAL;
2780		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2781	} else if (strcmp(ivmode, "lmk") == 0) {
2782		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2783		/*
2784		 * Version 2 and 3 is recognised according
2785		 * to length of provided multi-key string.
2786		 * If present (version 3), last key is used as IV seed.
2787		 * All keys (including IV seed) are always the same size.
2788		 */
2789		if (cc->key_size % cc->key_parts) {
2790			cc->key_parts++;
2791			cc->key_extra_size = cc->key_size / cc->key_parts;
2792		}
2793	} else if (strcmp(ivmode, "tcw") == 0) {
2794		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2795		cc->key_parts += 2; /* IV + whitening */
2796		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2797	} else if (strcmp(ivmode, "random") == 0) {
2798		cc->iv_gen_ops = &crypt_iv_random_ops;
2799		/* Need storage space in integrity fields. */
2800		cc->integrity_iv_size = cc->iv_size;
2801	} else {
2802		ti->error = "Invalid IV mode";
2803		return -EINVAL;
2804	}
2805
2806	return 0;
2807}
2808
2809/*
2810 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2811 * The HMAC is needed to calculate tag size (HMAC digest size).
2812 * This should be probably done by crypto-api calls (once available...)
2813 */
2814static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2815{
2816	char *start, *end, *mac_alg = NULL;
2817	struct crypto_ahash *mac;
2818
2819	if (!strstarts(cipher_api, "authenc("))
2820		return 0;
2821
2822	start = strchr(cipher_api, '(');
2823	end = strchr(cipher_api, ',');
2824	if (!start || !end || ++start > end)
2825		return -EINVAL;
2826
2827	mac_alg = kzalloc(end - start + 1, GFP_KERNEL);
2828	if (!mac_alg)
2829		return -ENOMEM;
2830	strncpy(mac_alg, start, end - start);
2831
2832	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2833	kfree(mac_alg);
2834
2835	if (IS_ERR(mac))
2836		return PTR_ERR(mac);
2837
2838	cc->key_mac_size = crypto_ahash_digestsize(mac);
2839	crypto_free_ahash(mac);
2840
2841	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2842	if (!cc->authenc_key)
2843		return -ENOMEM;
2844
2845	return 0;
2846}
2847
2848static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2849				char **ivmode, char **ivopts)
2850{
2851	struct crypt_config *cc = ti->private;
2852	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2853	int ret = -EINVAL;
2854
2855	cc->tfms_count = 1;
2856
2857	/*
2858	 * New format (capi: prefix)
2859	 * capi:cipher_api_spec-iv:ivopts
2860	 */
2861	tmp = &cipher_in[strlen("capi:")];
2862
2863	/* Separate IV options if present, it can contain another '-' in hash name */
2864	*ivopts = strrchr(tmp, ':');
2865	if (*ivopts) {
2866		**ivopts = '\0';
2867		(*ivopts)++;
2868	}
2869	/* Parse IV mode */
2870	*ivmode = strrchr(tmp, '-');
2871	if (*ivmode) {
2872		**ivmode = '\0';
2873		(*ivmode)++;
2874	}
2875	/* The rest is crypto API spec */
2876	cipher_api = tmp;
2877
2878	/* Alloc AEAD, can be used only in new format. */
2879	if (crypt_integrity_aead(cc)) {
2880		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2881		if (ret < 0) {
2882			ti->error = "Invalid AEAD cipher spec";
2883			return -ENOMEM;
2884		}
2885	}
2886
2887	if (*ivmode && !strcmp(*ivmode, "lmk"))
2888		cc->tfms_count = 64;
2889
2890	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2891		if (!*ivopts) {
2892			ti->error = "Digest algorithm missing for ESSIV mode";
2893			return -EINVAL;
2894		}
2895		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2896			       cipher_api, *ivopts);
2897		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2898			ti->error = "Cannot allocate cipher string";
2899			return -ENOMEM;
2900		}
2901		cipher_api = buf;
2902	}
2903
2904	cc->key_parts = cc->tfms_count;
2905
2906	/* Allocate cipher */
2907	ret = crypt_alloc_tfms(cc, cipher_api);
2908	if (ret < 0) {
2909		ti->error = "Error allocating crypto tfm";
2910		return ret;
2911	}
2912
2913	if (crypt_integrity_aead(cc))
2914		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2915	else
2916		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2917
2918	return 0;
2919}
2920
2921static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2922				char **ivmode, char **ivopts)
2923{
2924	struct crypt_config *cc = ti->private;
2925	char *tmp, *cipher, *chainmode, *keycount;
2926	char *cipher_api = NULL;
2927	int ret = -EINVAL;
2928	char dummy;
2929
2930	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2931		ti->error = "Bad cipher specification";
2932		return -EINVAL;
2933	}
2934
2935	/*
2936	 * Legacy dm-crypt cipher specification
2937	 * cipher[:keycount]-mode-iv:ivopts
2938	 */
2939	tmp = cipher_in;
2940	keycount = strsep(&tmp, "-");
2941	cipher = strsep(&keycount, ":");
2942
2943	if (!keycount)
2944		cc->tfms_count = 1;
2945	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2946		 !is_power_of_2(cc->tfms_count)) {
2947		ti->error = "Bad cipher key count specification";
2948		return -EINVAL;
2949	}
2950	cc->key_parts = cc->tfms_count;
2951
2952	chainmode = strsep(&tmp, "-");
2953	*ivmode = strsep(&tmp, ":");
2954	*ivopts = tmp;
2955
2956	/*
2957	 * For compatibility with the original dm-crypt mapping format, if
2958	 * only the cipher name is supplied, use cbc-plain.
2959	 */
2960	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
2961		chainmode = "cbc";
2962		*ivmode = "plain";
2963	}
2964
2965	if (strcmp(chainmode, "ecb") && !*ivmode) {
2966		ti->error = "IV mechanism required";
2967		return -EINVAL;
2968	}
2969
2970	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
2971	if (!cipher_api)
2972		goto bad_mem;
2973
2974	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2975		if (!*ivopts) {
2976			ti->error = "Digest algorithm missing for ESSIV mode";
2977			kfree(cipher_api);
2978			return -EINVAL;
2979		}
2980		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2981			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
2982	} else {
2983		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
2984			       "%s(%s)", chainmode, cipher);
2985	}
2986	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2987		kfree(cipher_api);
2988		goto bad_mem;
2989	}
2990
2991	/* Allocate cipher */
2992	ret = crypt_alloc_tfms(cc, cipher_api);
2993	if (ret < 0) {
2994		ti->error = "Error allocating crypto tfm";
2995		kfree(cipher_api);
2996		return ret;
2997	}
2998	kfree(cipher_api);
2999
3000	return 0;
3001bad_mem:
3002	ti->error = "Cannot allocate cipher strings";
3003	return -ENOMEM;
3004}
3005
3006static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3007{
3008	struct crypt_config *cc = ti->private;
3009	char *ivmode = NULL, *ivopts = NULL;
3010	int ret;
3011
3012	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3013	if (!cc->cipher_string) {
3014		ti->error = "Cannot allocate cipher strings";
3015		return -ENOMEM;
3016	}
3017
3018	if (strstarts(cipher_in, "capi:"))
3019		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3020	else
3021		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3022	if (ret)
3023		return ret;
3024
3025	/* Initialize IV */
3026	ret = crypt_ctr_ivmode(ti, ivmode);
3027	if (ret < 0)
3028		return ret;
3029
3030	/* Initialize and set key */
3031	ret = crypt_set_key(cc, key);
3032	if (ret < 0) {
3033		ti->error = "Error decoding and setting key";
3034		return ret;
3035	}
3036
3037	/* Allocate IV */
3038	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3039		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3040		if (ret < 0) {
3041			ti->error = "Error creating IV";
3042			return ret;
3043		}
3044	}
3045
3046	/* Initialize IV (set keys for ESSIV etc) */
3047	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3048		ret = cc->iv_gen_ops->init(cc);
3049		if (ret < 0) {
3050			ti->error = "Error initialising IV";
3051			return ret;
3052		}
3053	}
3054
3055	/* wipe the kernel key payload copy */
3056	if (cc->key_string)
3057		memset(cc->key, 0, cc->key_size * sizeof(u8));
3058
3059	return ret;
3060}
3061
3062static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3063{
3064	struct crypt_config *cc = ti->private;
3065	struct dm_arg_set as;
3066	static const struct dm_arg _args[] = {
3067		{0, 8, "Invalid number of feature args"},
3068	};
3069	unsigned int opt_params, val;
3070	const char *opt_string, *sval;
3071	char dummy;
3072	int ret;
3073
3074	/* Optional parameters */
3075	as.argc = argc;
3076	as.argv = argv;
3077
3078	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3079	if (ret)
3080		return ret;
3081
3082	while (opt_params--) {
3083		opt_string = dm_shift_arg(&as);
3084		if (!opt_string) {
3085			ti->error = "Not enough feature arguments";
3086			return -EINVAL;
3087		}
3088
3089		if (!strcasecmp(opt_string, "allow_discards"))
3090			ti->num_discard_bios = 1;
3091
3092		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3093			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3094
3095		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3096			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3097		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3098			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3099		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3100			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3101		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3102			if (val == 0 || val > MAX_TAG_SIZE) {
3103				ti->error = "Invalid integrity arguments";
3104				return -EINVAL;
3105			}
3106			cc->on_disk_tag_size = val;
3107			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3108			if (!strcasecmp(sval, "aead")) {
3109				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3110			} else  if (strcasecmp(sval, "none")) {
3111				ti->error = "Unknown integrity profile";
3112				return -EINVAL;
3113			}
3114
3115			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3116			if (!cc->cipher_auth)
3117				return -ENOMEM;
3118		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3119			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3120			    cc->sector_size > 4096 ||
3121			    (cc->sector_size & (cc->sector_size - 1))) {
3122				ti->error = "Invalid feature value for sector_size";
3123				return -EINVAL;
3124			}
3125			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3126				ti->error = "Device size is not multiple of sector_size feature";
3127				return -EINVAL;
3128			}
3129			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3130		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3131			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3132		else {
3133			ti->error = "Invalid feature arguments";
3134			return -EINVAL;
3135		}
3136	}
3137
3138	return 0;
3139}
3140
3141#ifdef CONFIG_BLK_DEV_ZONED
3142static int crypt_report_zones(struct dm_target *ti,
3143		struct dm_report_zones_args *args, unsigned int nr_zones)
3144{
3145	struct crypt_config *cc = ti->private;
3146
3147	return dm_report_zones(cc->dev->bdev, cc->start,
3148			cc->start + dm_target_offset(ti, args->next_sector),
3149			args, nr_zones);
3150}
3151#else
3152#define crypt_report_zones NULL
3153#endif
3154
3155/*
3156 * Construct an encryption mapping:
3157 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3158 */
3159static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3160{
3161	struct crypt_config *cc;
3162	const char *devname = dm_table_device_name(ti->table);
3163	int key_size;
3164	unsigned int align_mask;
3165	unsigned long long tmpll;
3166	int ret;
3167	size_t iv_size_padding, additional_req_size;
3168	char dummy;
3169
3170	if (argc < 5) {
3171		ti->error = "Not enough arguments";
3172		return -EINVAL;
3173	}
3174
3175	key_size = get_key_size(&argv[1]);
3176	if (key_size < 0) {
3177		ti->error = "Cannot parse key size";
3178		return -EINVAL;
3179	}
3180
3181	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3182	if (!cc) {
3183		ti->error = "Cannot allocate encryption context";
3184		return -ENOMEM;
3185	}
3186	cc->key_size = key_size;
3187	cc->sector_size = (1 << SECTOR_SHIFT);
3188	cc->sector_shift = 0;
3189
3190	ti->private = cc;
3191
3192	spin_lock(&dm_crypt_clients_lock);
3193	dm_crypt_clients_n++;
3194	crypt_calculate_pages_per_client();
3195	spin_unlock(&dm_crypt_clients_lock);
3196
3197	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3198	if (ret < 0)
3199		goto bad;
3200
3201	/* Optional parameters need to be read before cipher constructor */
3202	if (argc > 5) {
3203		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3204		if (ret)
3205			goto bad;
3206	}
3207
3208	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3209	if (ret < 0)
3210		goto bad;
3211
3212	if (crypt_integrity_aead(cc)) {
3213		cc->dmreq_start = sizeof(struct aead_request);
3214		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3215		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3216	} else {
3217		cc->dmreq_start = sizeof(struct skcipher_request);
3218		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3219		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3220	}
3221	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3222
3223	if (align_mask < CRYPTO_MINALIGN) {
3224		/* Allocate the padding exactly */
3225		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3226				& align_mask;
3227	} else {
3228		/*
3229		 * If the cipher requires greater alignment than kmalloc
3230		 * alignment, we don't know the exact position of the
3231		 * initialization vector. We must assume worst case.
3232		 */
3233		iv_size_padding = align_mask;
3234	}
3235
3236	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3237	additional_req_size = sizeof(struct dm_crypt_request) +
3238		iv_size_padding + cc->iv_size +
3239		cc->iv_size +
3240		sizeof(uint64_t) +
3241		sizeof(unsigned int);
3242
3243	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3244	if (ret) {
3245		ti->error = "Cannot allocate crypt request mempool";
3246		goto bad;
3247	}
3248
3249	cc->per_bio_data_size = ti->per_io_data_size =
3250		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3251		      ARCH_KMALLOC_MINALIGN);
3252
3253	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3254	if (ret) {
3255		ti->error = "Cannot allocate page mempool";
3256		goto bad;
3257	}
3258
3259	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3260	if (ret) {
3261		ti->error = "Cannot allocate crypt bioset";
3262		goto bad;
3263	}
3264
3265	mutex_init(&cc->bio_alloc_lock);
3266
3267	ret = -EINVAL;
3268	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3269	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3270		ti->error = "Invalid iv_offset sector";
3271		goto bad;
3272	}
3273	cc->iv_offset = tmpll;
3274
3275	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3276	if (ret) {
3277		ti->error = "Device lookup failed";
3278		goto bad;
3279	}
3280
3281	ret = -EINVAL;
3282	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3283		ti->error = "Invalid device sector";
3284		goto bad;
3285	}
3286	cc->start = tmpll;
3287
3288	if (bdev_is_zoned(cc->dev->bdev)) {
3289		/*
3290		 * For zoned block devices, we need to preserve the issuer write
3291		 * ordering. To do so, disable write workqueues and force inline
3292		 * encryption completion.
3293		 */
3294		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3295		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3296
3297		/*
3298		 * All zone append writes to a zone of a zoned block device will
3299		 * have the same BIO sector, the start of the zone. When the
3300		 * cypher IV mode uses sector values, all data targeting a
3301		 * zone will be encrypted using the first sector numbers of the
3302		 * zone. This will not result in write errors but will
3303		 * cause most reads to fail as reads will use the sector values
3304		 * for the actual data locations, resulting in IV mismatch.
3305		 * To avoid this problem, ask DM core to emulate zone append
3306		 * operations with regular writes.
3307		 */
3308		DMDEBUG("Zone append operations will be emulated");
3309		ti->emulate_zone_append = true;
3310	}
3311
3312	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3313		ret = crypt_integrity_ctr(cc, ti);
3314		if (ret)
3315			goto bad;
3316
3317		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size;
3318		if (!cc->tag_pool_max_sectors)
3319			cc->tag_pool_max_sectors = 1;
3320
3321		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3322			cc->tag_pool_max_sectors * cc->on_disk_tag_size);
3323		if (ret) {
3324			ti->error = "Cannot allocate integrity tags mempool";
3325			goto bad;
3326		}
3327
3328		cc->tag_pool_max_sectors <<= cc->sector_shift;
3329	}
3330
3331	ret = -ENOMEM;
3332	cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname);
3333	if (!cc->io_queue) {
3334		ti->error = "Couldn't create kcryptd io queue";
3335		goto bad;
3336	}
3337
3338	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3339		cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM,
3340						  1, devname);
3341	else
3342		cc->crypt_queue = alloc_workqueue("kcryptd/%s",
3343						  WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND,
3344						  num_online_cpus(), devname);
3345	if (!cc->crypt_queue) {
3346		ti->error = "Couldn't create kcryptd queue";
3347		goto bad;
3348	}
3349
3350	spin_lock_init(&cc->write_thread_lock);
3351	cc->write_tree = RB_ROOT;
3352
3353	cc->write_thread = kthread_create(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3354	if (IS_ERR(cc->write_thread)) {
3355		ret = PTR_ERR(cc->write_thread);
3356		cc->write_thread = NULL;
3357		ti->error = "Couldn't spawn write thread";
3358		goto bad;
3359	}
3360	wake_up_process(cc->write_thread);
3361
3362	ti->num_flush_bios = 1;
3363	ti->limit_swap_bios = true;
 
3364
 
3365	return 0;
3366
3367bad:
 
3368	crypt_dtr(ti);
3369	return ret;
3370}
3371
3372static int crypt_map(struct dm_target *ti, struct bio *bio)
3373{
3374	struct dm_crypt_io *io;
3375	struct crypt_config *cc = ti->private;
3376
3377	/*
3378	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3379	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3380	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3381	 */
3382	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3383	    bio_op(bio) == REQ_OP_DISCARD)) {
3384		bio_set_dev(bio, cc->dev->bdev);
3385		if (bio_sectors(bio))
3386			bio->bi_iter.bi_sector = cc->start +
3387				dm_target_offset(ti, bio->bi_iter.bi_sector);
3388		return DM_MAPIO_REMAPPED;
3389	}
3390
3391	/*
3392	 * Check if bio is too large, split as needed.
3393	 */
3394	if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) &&
3395	    (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size))
3396		dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT));
3397
3398	/*
3399	 * Ensure that bio is a multiple of internal sector encryption size
3400	 * and is aligned to this size as defined in IO hints.
3401	 */
3402	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3403		return DM_MAPIO_KILL;
3404
3405	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3406		return DM_MAPIO_KILL;
3407
3408	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3409	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3410
3411	if (cc->on_disk_tag_size) {
3412		unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift);
3413
3414		if (unlikely(tag_len > KMALLOC_MAX_SIZE) ||
3415		    unlikely(!(io->integrity_metadata = kmalloc(tag_len,
3416				GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) {
3417			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3418				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3419			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3420			io->integrity_metadata_from_pool = true;
3421		}
3422	}
3423
3424	if (crypt_integrity_aead(cc))
3425		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3426	else
3427		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3428
3429	if (bio_data_dir(io->base_bio) == READ) {
3430		if (kcryptd_io_read(io, GFP_NOWAIT))
3431			kcryptd_queue_read(io);
3432	} else
3433		kcryptd_queue_crypt(io);
3434
3435	return DM_MAPIO_SUBMITTED;
3436}
3437
 
 
 
 
 
3438static void crypt_status(struct dm_target *ti, status_type_t type,
3439			 unsigned status_flags, char *result, unsigned maxlen)
3440{
3441	struct crypt_config *cc = ti->private;
3442	unsigned i, sz = 0;
3443	int num_feature_args = 0;
3444
3445	switch (type) {
3446	case STATUSTYPE_INFO:
3447		result[0] = '\0';
3448		break;
3449
3450	case STATUSTYPE_TABLE:
3451		DMEMIT("%s ", cc->cipher_string);
3452
3453		if (cc->key_size > 0) {
3454			if (cc->key_string)
3455				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3456			else
3457				for (i = 0; i < cc->key_size; i++)
3458					DMEMIT("%02x", cc->key[i]);
 
 
 
3459		} else
3460			DMEMIT("-");
3461
3462		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3463				cc->dev->name, (unsigned long long)cc->start);
3464
3465		num_feature_args += !!ti->num_discard_bios;
3466		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3467		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3468		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3469		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3470		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3471		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3472		if (cc->on_disk_tag_size)
3473			num_feature_args++;
3474		if (num_feature_args) {
3475			DMEMIT(" %d", num_feature_args);
3476			if (ti->num_discard_bios)
3477				DMEMIT(" allow_discards");
3478			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3479				DMEMIT(" same_cpu_crypt");
3480			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3481				DMEMIT(" submit_from_crypt_cpus");
3482			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3483				DMEMIT(" no_read_workqueue");
3484			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3485				DMEMIT(" no_write_workqueue");
3486			if (cc->on_disk_tag_size)
3487				DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth);
3488			if (cc->sector_size != (1 << SECTOR_SHIFT))
3489				DMEMIT(" sector_size:%d", cc->sector_size);
3490			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3491				DMEMIT(" iv_large_sectors");
3492		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3493
 
 
 
 
 
 
 
 
 
 
 
 
 
3494		break;
3495	}
3496}
3497
3498static void crypt_postsuspend(struct dm_target *ti)
3499{
3500	struct crypt_config *cc = ti->private;
3501
3502	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3503}
3504
3505static int crypt_preresume(struct dm_target *ti)
3506{
3507	struct crypt_config *cc = ti->private;
3508
3509	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3510		DMERR("aborting resume - crypt key is not set.");
3511		return -EAGAIN;
3512	}
3513
3514	return 0;
3515}
3516
3517static void crypt_resume(struct dm_target *ti)
3518{
3519	struct crypt_config *cc = ti->private;
3520
3521	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3522}
3523
3524/* Message interface
3525 *	key set <key>
3526 *	key wipe
3527 */
3528static int crypt_message(struct dm_target *ti, unsigned argc, char **argv,
3529			 char *result, unsigned maxlen)
3530{
3531	struct crypt_config *cc = ti->private;
3532	int key_size, ret = -EINVAL;
3533
3534	if (argc < 2)
3535		goto error;
3536
3537	if (!strcasecmp(argv[0], "key")) {
3538		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3539			DMWARN("not suspended during key manipulation.");
3540			return -EINVAL;
3541		}
3542		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3543			/* The key size may not be changed. */
3544			key_size = get_key_size(&argv[2]);
3545			if (key_size < 0 || cc->key_size != key_size) {
3546				memset(argv[2], '0', strlen(argv[2]));
3547				return -EINVAL;
3548			}
3549
3550			ret = crypt_set_key(cc, argv[2]);
3551			if (ret)
3552				return ret;
3553			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3554				ret = cc->iv_gen_ops->init(cc);
3555			/* wipe the kernel key payload copy */
3556			if (cc->key_string)
3557				memset(cc->key, 0, cc->key_size * sizeof(u8));
3558			return ret;
3559		}
3560		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3561			return crypt_wipe_key(cc);
3562	}
3563
3564error:
3565	DMWARN("unrecognised message received.");
3566	return -EINVAL;
3567}
3568
3569static int crypt_iterate_devices(struct dm_target *ti,
3570				 iterate_devices_callout_fn fn, void *data)
3571{
3572	struct crypt_config *cc = ti->private;
3573
3574	return fn(ti, cc->dev, cc->start, ti->len, data);
3575}
3576
3577static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3578{
3579	struct crypt_config *cc = ti->private;
3580
3581	/*
3582	 * Unfortunate constraint that is required to avoid the potential
3583	 * for exceeding underlying device's max_segments limits -- due to
3584	 * crypt_alloc_buffer() possibly allocating pages for the encryption
3585	 * bio that are not as physically contiguous as the original bio.
3586	 */
3587	limits->max_segment_size = PAGE_SIZE;
3588
3589	limits->logical_block_size =
3590		max_t(unsigned, limits->logical_block_size, cc->sector_size);
3591	limits->physical_block_size =
3592		max_t(unsigned, limits->physical_block_size, cc->sector_size);
3593	limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size);
 
3594}
3595
3596static struct target_type crypt_target = {
3597	.name   = "crypt",
3598	.version = {1, 23, 0},
3599	.module = THIS_MODULE,
3600	.ctr    = crypt_ctr,
3601	.dtr    = crypt_dtr,
3602	.features = DM_TARGET_ZONED_HM,
3603	.report_zones = crypt_report_zones,
3604	.map    = crypt_map,
3605	.status = crypt_status,
3606	.postsuspend = crypt_postsuspend,
3607	.preresume = crypt_preresume,
3608	.resume = crypt_resume,
3609	.message = crypt_message,
3610	.iterate_devices = crypt_iterate_devices,
3611	.io_hints = crypt_io_hints,
3612};
3613
3614static int __init dm_crypt_init(void)
3615{
3616	int r;
3617
3618	r = dm_register_target(&crypt_target);
3619	if (r < 0)
3620		DMERR("register failed %d", r);
3621
3622	return r;
3623}
3624
3625static void __exit dm_crypt_exit(void)
3626{
3627	dm_unregister_target(&crypt_target);
3628}
3629
3630module_init(dm_crypt_init);
3631module_exit(dm_crypt_exit);
3632
3633MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3634MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3635MODULE_LICENSE("GPL");