File Index
Files and directories
Sign in
FILE

dm-crypt.c

File preview with syntax highlighting.

linux-6.19.14/drivers/md/dm-crypt.c C
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2003 Jana Saout <jana@saout.de>
   4 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org>
   5 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved.
   6 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com>
   7 *
   8 * This file is released under the GPL.
   9 */
  10
  11#include <linux/completion.h>
  12#include <linux/err.h>
  13#include <linux/module.h>
  14#include <linux/init.h>
  15#include <linux/kernel.h>
  16#include <linux/key.h>
  17#include <linux/bio.h>
  18#include <linux/blkdev.h>
  19#include <linux/blk-integrity.h>
  20#include <linux/crc32.h>
  21#include <linux/mempool.h>
  22#include <linux/slab.h>
  23#include <linux/crypto.h>
  24#include <linux/fips.h>
  25#include <linux/workqueue.h>
  26#include <linux/kthread.h>
  27#include <linux/backing-dev.h>
  28#include <linux/atomic.h>
  29#include <linux/scatterlist.h>
  30#include <linux/rbtree.h>
  31#include <linux/ctype.h>
  32#include <asm/page.h>
  33#include <linux/unaligned.h>
  34#include <crypto/hash.h>
  35#include <crypto/md5.h>
  36#include <crypto/skcipher.h>
  37#include <crypto/aead.h>
  38#include <crypto/authenc.h>
  39#include <crypto/utils.h>
  40#include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */
  41#include <linux/key-type.h>
  42#include <keys/user-type.h>
  43#include <keys/encrypted-type.h>
  44#include <keys/trusted-type.h>
  45
  46#include <linux/device-mapper.h>
  47
  48#include "dm-audit.h"
  49
  50#define DM_MSG_PREFIX "crypt"
  51
  52static DEFINE_IDA(workqueue_ida);
  53
  54/*
  55 * context holding the current state of a multi-part conversion
  56 */
  57struct convert_context {
  58	struct completion restart;
  59	struct bio *bio_in;
  60	struct bvec_iter iter_in;
  61	struct bio *bio_out;
  62	struct bvec_iter iter_out;
  63	atomic_t cc_pending;
  64	unsigned int tag_offset;
  65	u64 cc_sector;
  66	union {
  67		struct skcipher_request *req;
  68		struct aead_request *req_aead;
  69	} r;
  70	bool aead_recheck;
  71	bool aead_failed;
  72
  73};
  74
  75/*
  76 * per bio private data
  77 */
  78struct dm_crypt_io {
  79	struct crypt_config *cc;
  80	struct bio *base_bio;
  81	u8 *integrity_metadata;
  82	bool integrity_metadata_from_pool:1;
  83
  84	struct work_struct work;
  85
  86	struct convert_context ctx;
  87
  88	atomic_t io_pending;
  89	blk_status_t error;
  90	sector_t sector;
  91
  92	struct bvec_iter saved_bi_iter;
  93
  94	struct rb_node rb_node;
  95} CRYPTO_MINALIGN_ATTR;
  96
  97struct dm_crypt_request {
  98	struct convert_context *ctx;
  99	struct scatterlist sg_in[4];
 100	struct scatterlist sg_out[4];
 101	u64 iv_sector;
 102};
 103
 104struct crypt_config;
 105
 106struct crypt_iv_operations {
 107	int (*ctr)(struct crypt_config *cc, struct dm_target *ti,
 108		   const char *opts);
 109	void (*dtr)(struct crypt_config *cc);
 110	int (*init)(struct crypt_config *cc);
 111	int (*wipe)(struct crypt_config *cc);
 112	int (*generator)(struct crypt_config *cc, u8 *iv,
 113			 struct dm_crypt_request *dmreq);
 114	int (*post)(struct crypt_config *cc, u8 *iv,
 115		    struct dm_crypt_request *dmreq);
 116};
 117
 118struct iv_benbi_private {
 119	int shift;
 120};
 121
 122#define LMK_SEED_SIZE 64 /* hash + 0 */
 123struct iv_lmk_private {
 124	u8 *seed;
 125};
 126
 127#define TCW_WHITENING_SIZE 16
 128struct iv_tcw_private {
 129	u8 *iv_seed;
 130	u8 *whitening;
 131};
 132
 133#define ELEPHANT_MAX_KEY_SIZE 32
 134struct iv_elephant_private {
 135	struct crypto_skcipher *tfm;
 136};
 137
 138/*
 139 * Crypt: maps a linear range of a block device
 140 * and encrypts / decrypts at the same time.
 141 */
 142enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
 143	     DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY,
 144	     DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE,
 145	     DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE };
 146
 147enum cipher_flags {
 148	CRYPT_MODE_INTEGRITY_AEAD,	/* Use authenticated mode for cipher */
 149	CRYPT_IV_LARGE_SECTORS,		/* Calculate IV from sector_size, not 512B sectors */
 150	CRYPT_ENCRYPT_PREPROCESS,	/* Must preprocess data for encryption (elephant) */
 151	CRYPT_KEY_MAC_SIZE_SET,		/* The integrity_key_size option was used */
 152};
 153
 154/*
 155 * The fields in here must be read only after initialization.
 156 */
 157struct crypt_config {
 158	struct dm_dev *dev;
 159	sector_t start;
 160
 161	struct percpu_counter n_allocated_pages;
 162
 163	struct workqueue_struct *io_queue;
 164	struct workqueue_struct *crypt_queue;
 165
 166	spinlock_t write_thread_lock;
 167	struct task_struct *write_thread;
 168	struct rb_root write_tree;
 169
 170	char *cipher_string;
 171	char *cipher_auth;
 172	char *key_string;
 173
 174	const struct crypt_iv_operations *iv_gen_ops;
 175	union {
 176		struct iv_benbi_private benbi;
 177		struct iv_lmk_private lmk;
 178		struct iv_tcw_private tcw;
 179		struct iv_elephant_private elephant;
 180	} iv_gen_private;
 181	u64 iv_offset;
 182	unsigned int iv_size;
 183	unsigned short sector_size;
 184	unsigned char sector_shift;
 185
 186	union {
 187		struct crypto_skcipher **tfms;
 188		struct crypto_aead **tfms_aead;
 189	} cipher_tfm;
 190	unsigned int tfms_count;
 191	int workqueue_id;
 192	unsigned long cipher_flags;
 193
 194	/*
 195	 * Layout of each crypto request:
 196	 *
 197	 *   struct skcipher_request
 198	 *      context
 199	 *      padding
 200	 *   struct dm_crypt_request
 201	 *      padding
 202	 *   IV
 203	 *
 204	 * The padding is added so that dm_crypt_request and the IV are
 205	 * correctly aligned.
 206	 */
 207	unsigned int dmreq_start;
 208
 209	unsigned int per_bio_data_size;
 210
 211	unsigned long flags;
 212	unsigned int key_size;
 213	unsigned int key_parts;      /* independent parts in key buffer */
 214	unsigned int key_extra_size; /* additional keys length */
 215	unsigned int key_mac_size;   /* MAC key size for authenc(...) */
 216
 217	unsigned int integrity_tag_size;
 218	unsigned int integrity_iv_size;
 219	unsigned int used_tag_size;
 220	unsigned int tuple_size;
 221
 222	/*
 223	 * pool for per bio private data, crypto requests,
 224	 * encryption requeusts/buffer pages and integrity tags
 225	 */
 226	unsigned int tag_pool_max_sectors;
 227	mempool_t tag_pool;
 228	mempool_t req_pool;
 229	mempool_t page_pool;
 230
 231	struct bio_set bs;
 232	struct mutex bio_alloc_lock;
 233
 234	u8 *authenc_key; /* space for keys in authenc() format (if used) */
 235	u8 key[] __counted_by(key_size);
 236};
 237
 238#define MIN_IOS		64
 239#define MAX_TAG_SIZE	480
 240#define POOL_ENTRY_SIZE	512
 241
 242static DEFINE_SPINLOCK(dm_crypt_clients_lock);
 243static unsigned int dm_crypt_clients_n;
 244static volatile unsigned long dm_crypt_pages_per_client;
 245#define DM_CRYPT_MEMORY_PERCENT			2
 246#define DM_CRYPT_MIN_PAGES_PER_CLIENT		(BIO_MAX_VECS * 16)
 247#define DM_CRYPT_DEFAULT_MAX_READ_SIZE		131072
 248#define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE		131072
 249
 250static unsigned int max_read_size = 0;
 251module_param(max_read_size, uint, 0644);
 252MODULE_PARM_DESC(max_read_size, "Maximum size of a read request");
 253static unsigned int max_write_size = 0;
 254module_param(max_write_size, uint, 0644);
 255MODULE_PARM_DESC(max_write_size, "Maximum size of a write request");
 256
 257static unsigned get_max_request_sectors(struct dm_target *ti, struct bio *bio, bool no_split)
 258{
 259	struct crypt_config *cc = ti->private;
 260	unsigned val, sector_align;
 261	bool wrt = op_is_write(bio_op(bio));
 262
 263	if (no_split) {
 264		val = -1;
 265	} else if (wrt) {
 266		val = min_not_zero(READ_ONCE(max_write_size),
 267				   DM_CRYPT_DEFAULT_MAX_WRITE_SIZE);
 268	} else {
 269		val = min_not_zero(READ_ONCE(max_read_size),
 270				   DM_CRYPT_DEFAULT_MAX_READ_SIZE);
 271	}
 272
 273	if (wrt || cc->used_tag_size)
 274		val = min(val, BIO_MAX_VECS << PAGE_SHIFT);
 275
 276	sector_align = max(bdev_logical_block_size(cc->dev->bdev),
 277			   (unsigned)cc->sector_size);
 278	val = round_down(val, sector_align);
 279	if (unlikely(!val))
 280		val = sector_align;
 281	return val >> SECTOR_SHIFT;
 282}
 283
 284static void crypt_endio(struct bio *clone);
 285static void kcryptd_queue_crypt(struct dm_crypt_io *io);
 286static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc,
 287					     struct scatterlist *sg);
 288
 289static bool crypt_integrity_aead(struct crypt_config *cc);
 290
 291/*
 292 * Use this to access cipher attributes that are independent of the key.
 293 */
 294static struct crypto_skcipher *any_tfm(struct crypt_config *cc)
 295{
 296	return cc->cipher_tfm.tfms[0];
 297}
 298
 299static struct crypto_aead *any_tfm_aead(struct crypt_config *cc)
 300{
 301	return cc->cipher_tfm.tfms_aead[0];
 302}
 303
 304/*
 305 * Different IV generation algorithms:
 306 *
 307 * plain: the initial vector is the 32-bit little-endian version of the sector
 308 *        number, padded with zeros if necessary.
 309 *
 310 * plain64: the initial vector is the 64-bit little-endian version of the sector
 311 *        number, padded with zeros if necessary.
 312 *
 313 * plain64be: the initial vector is the 64-bit big-endian version of the sector
 314 *        number, padded with zeros if necessary.
 315 *
 316 * essiv: "encrypted sector|salt initial vector", the sector number is
 317 *        encrypted with the bulk cipher using a salt as key. The salt
 318 *        should be derived from the bulk cipher's key via hashing.
 319 *
 320 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1
 321 *        (needed for LRW-32-AES and possible other narrow block modes)
 322 *
 323 * null: the initial vector is always zero.  Provides compatibility with
 324 *       obsolete loop_fish2 devices.  Do not use for new devices.
 325 *
 326 * lmk:  Compatible implementation of the block chaining mode used
 327 *       by the Loop-AES block device encryption system
 328 *       designed by Jari Ruusu. See http://loop-aes.sourceforge.net/
 329 *       It operates on full 512 byte sectors and uses CBC
 330 *       with an IV derived from the sector number, the data and
 331 *       optionally extra IV seed.
 332 *       This means that after decryption the first block
 333 *       of sector must be tweaked according to decrypted data.
 334 *       Loop-AES can use three encryption schemes:
 335 *         version 1: is plain aes-cbc mode
 336 *         version 2: uses 64 multikey scheme with lmk IV generator
 337 *         version 3: the same as version 2 with additional IV seed
 338 *                   (it uses 65 keys, last key is used as IV seed)
 339 *
 340 * tcw:  Compatible implementation of the block chaining mode used
 341 *       by the TrueCrypt device encryption system (prior to version 4.1).
 342 *       For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat
 343 *       It operates on full 512 byte sectors and uses CBC
 344 *       with an IV derived from initial key and the sector number.
 345 *       In addition, whitening value is applied on every sector, whitening
 346 *       is calculated from initial key, sector number and mixed using CRC32.
 347 *       Note that this encryption scheme is vulnerable to watermarking attacks
 348 *       and should be used for old compatible containers access only.
 349 *
 350 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode)
 351 *        The IV is encrypted little-endian byte-offset (with the same key
 352 *        and cipher as the volume).
 353 *
 354 * elephant: The extended version of eboiv with additional Elephant diffuser
 355 *           used with Bitlocker CBC mode.
 356 *           This mode was used in older Windows systems
 357 *           https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf
 358 */
 359
 360static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv,
 361			      struct dm_crypt_request *dmreq)
 362{
 363	memset(iv, 0, cc->iv_size);
 364	*(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff);
 365
 366	return 0;
 367}
 368
 369static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv,
 370				struct dm_crypt_request *dmreq)
 371{
 372	memset(iv, 0, cc->iv_size);
 373	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 374
 375	return 0;
 376}
 377
 378static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv,
 379				  struct dm_crypt_request *dmreq)
 380{
 381	memset(iv, 0, cc->iv_size);
 382	/* iv_size is at least of size u64; usually it is 16 bytes */
 383	*(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector);
 384
 385	return 0;
 386}
 387
 388static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv,
 389			      struct dm_crypt_request *dmreq)
 390{
 391	/*
 392	 * ESSIV encryption of the IV is now handled by the crypto API,
 393	 * so just pass the plain sector number here.
 394	 */
 395	memset(iv, 0, cc->iv_size);
 396	*(__le64 *)iv = cpu_to_le64(dmreq->iv_sector);
 397
 398	return 0;
 399}
 400
 401static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti,
 402			      const char *opts)
 403{
 404	unsigned int bs;
 405	int log;
 406
 407	if (crypt_integrity_aead(cc))
 408		bs = crypto_aead_blocksize(any_tfm_aead(cc));
 409	else
 410		bs = crypto_skcipher_blocksize(any_tfm(cc));
 411	log = ilog2(bs);
 412
 413	/*
 414	 * We need to calculate how far we must shift the sector count
 415	 * to get the cipher block count, we use this shift in _gen.
 416	 */
 417	if (1 << log != bs) {
 418		ti->error = "cypher blocksize is not a power of 2";
 419		return -EINVAL;
 420	}
 421
 422	if (log > 9) {
 423		ti->error = "cypher blocksize is > 512";
 424		return -EINVAL;
 425	}
 426
 427	cc->iv_gen_private.benbi.shift = 9 - log;
 428
 429	return 0;
 430}
 431
 432static void crypt_iv_benbi_dtr(struct crypt_config *cc)
 433{
 434}
 435
 436static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv,
 437			      struct dm_crypt_request *dmreq)
 438{
 439	__be64 val;
 440
 441	memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */
 442
 443	val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1);
 444	put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64)));
 445
 446	return 0;
 447}
 448
 449static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv,
 450			     struct dm_crypt_request *dmreq)
 451{
 452	memset(iv, 0, cc->iv_size);
 453
 454	return 0;
 455}
 456
 457static void crypt_iv_lmk_dtr(struct crypt_config *cc)
 458{
 459	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 460
 461	kfree_sensitive(lmk->seed);
 462	lmk->seed = NULL;
 463}
 464
 465static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti,
 466			    const char *opts)
 467{
 468	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 469
 470	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 471		ti->error = "Unsupported sector size for LMK";
 472		return -EINVAL;
 473	}
 474
 475	if (fips_enabled) {
 476		ti->error = "LMK support is disabled due to FIPS";
 477		/* ... because it uses MD5. */
 478		return -EINVAL;
 479	}
 480
 481	/* No seed in LMK version 2 */
 482	if (cc->key_parts == cc->tfms_count) {
 483		lmk->seed = NULL;
 484		return 0;
 485	}
 486
 487	lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL);
 488	if (!lmk->seed) {
 489		ti->error = "Error kmallocing seed storage in LMK";
 490		return -ENOMEM;
 491	}
 492
 493	return 0;
 494}
 495
 496static int crypt_iv_lmk_init(struct crypt_config *cc)
 497{
 498	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 499	int subkey_size = cc->key_size / cc->key_parts;
 500
 501	/* LMK seed is on the position of LMK_KEYS + 1 key */
 502	if (lmk->seed)
 503		memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size),
 504		       MD5_DIGEST_SIZE);
 505
 506	return 0;
 507}
 508
 509static int crypt_iv_lmk_wipe(struct crypt_config *cc)
 510{
 511	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 512
 513	if (lmk->seed)
 514		memset(lmk->seed, 0, LMK_SEED_SIZE);
 515
 516	return 0;
 517}
 518
 519static void crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv,
 520			     struct dm_crypt_request *dmreq, u8 *data)
 521{
 522	struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk;
 523	struct md5_ctx ctx;
 524	__le32 buf[4];
 525
 526	md5_init(&ctx);
 527
 528	if (lmk->seed)
 529		md5_update(&ctx, lmk->seed, LMK_SEED_SIZE);
 530
 531	/* Sector is always 512B, block size 16, add data of blocks 1-31 */
 532	md5_update(&ctx, data + 16, 16 * 31);
 533
 534	/* Sector is cropped to 56 bits here */
 535	buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF);
 536	buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000);
 537	buf[2] = cpu_to_le32(4024);
 538	buf[3] = 0;
 539	md5_update(&ctx, (u8 *)buf, sizeof(buf));
 540
 541	/* No MD5 padding here */
 542	cpu_to_le32_array(ctx.state.h, ARRAY_SIZE(ctx.state.h));
 543	memcpy(iv, ctx.state.h, cc->iv_size);
 544}
 545
 546static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv,
 547			    struct dm_crypt_request *dmreq)
 548{
 549	struct scatterlist *sg;
 550	u8 *src;
 551
 552	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) {
 553		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 554		src = kmap_local_page(sg_page(sg));
 555		crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset);
 556		kunmap_local(src);
 557	} else
 558		memset(iv, 0, cc->iv_size);
 559	return 0;
 560}
 561
 562static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv,
 563			     struct dm_crypt_request *dmreq)
 564{
 565	struct scatterlist *sg;
 566	u8 *dst;
 567
 568	if (bio_data_dir(dmreq->ctx->bio_in) == WRITE)
 569		return 0;
 570
 571	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 572	dst = kmap_local_page(sg_page(sg));
 573	crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset);
 574
 575	/* Tweak the first block of plaintext sector */
 576	crypto_xor(dst + sg->offset, iv, cc->iv_size);
 577
 578	kunmap_local(dst);
 579	return 0;
 580}
 581
 582static void crypt_iv_tcw_dtr(struct crypt_config *cc)
 583{
 584	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 585
 586	kfree_sensitive(tcw->iv_seed);
 587	tcw->iv_seed = NULL;
 588	kfree_sensitive(tcw->whitening);
 589	tcw->whitening = NULL;
 590}
 591
 592static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti,
 593			    const char *opts)
 594{
 595	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 596
 597	if (cc->sector_size != (1 << SECTOR_SHIFT)) {
 598		ti->error = "Unsupported sector size for TCW";
 599		return -EINVAL;
 600	}
 601
 602	if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) {
 603		ti->error = "Wrong key size for TCW";
 604		return -EINVAL;
 605	}
 606
 607	tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL);
 608	tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL);
 609	if (!tcw->iv_seed || !tcw->whitening) {
 610		crypt_iv_tcw_dtr(cc);
 611		ti->error = "Error allocating seed storage in TCW";
 612		return -ENOMEM;
 613	}
 614
 615	return 0;
 616}
 617
 618static int crypt_iv_tcw_init(struct crypt_config *cc)
 619{
 620	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 621	int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE;
 622
 623	memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size);
 624	memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size],
 625	       TCW_WHITENING_SIZE);
 626
 627	return 0;
 628}
 629
 630static int crypt_iv_tcw_wipe(struct crypt_config *cc)
 631{
 632	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 633
 634	memset(tcw->iv_seed, 0, cc->iv_size);
 635	memset(tcw->whitening, 0, TCW_WHITENING_SIZE);
 636
 637	return 0;
 638}
 639
 640static void crypt_iv_tcw_whitening(struct crypt_config *cc,
 641				   struct dm_crypt_request *dmreq, u8 *data)
 642{
 643	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 644	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 645	u8 buf[TCW_WHITENING_SIZE];
 646	int i;
 647
 648	/* xor whitening with sector number */
 649	crypto_xor_cpy(buf, tcw->whitening, (u8 *)&sector, 8);
 650	crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)&sector, 8);
 651
 652	/* calculate crc32 for every 32bit part and xor it */
 653	for (i = 0; i < 4; i++)
 654		put_unaligned_le32(crc32(0, &buf[i * 4], 4), &buf[i * 4]);
 655	crypto_xor(&buf[0], &buf[12], 4);
 656	crypto_xor(&buf[4], &buf[8], 4);
 657
 658	/* apply whitening (8 bytes) to whole sector */
 659	for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++)
 660		crypto_xor(data + i * 8, buf, 8);
 661	memzero_explicit(buf, sizeof(buf));
 662}
 663
 664static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv,
 665			    struct dm_crypt_request *dmreq)
 666{
 667	struct scatterlist *sg;
 668	struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw;
 669	__le64 sector = cpu_to_le64(dmreq->iv_sector);
 670	u8 *src;
 671
 672	/* Remove whitening from ciphertext */
 673	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) {
 674		sg = crypt_get_sg_data(cc, dmreq->sg_in);
 675		src = kmap_local_page(sg_page(sg));
 676		crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset);
 677		kunmap_local(src);
 678	}
 679
 680	/* Calculate IV */
 681	crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)&sector, 8);
 682	if (cc->iv_size > 8)
 683		crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)&sector,
 684			       cc->iv_size - 8);
 685
 686	return 0;
 687}
 688
 689static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv,
 690			     struct dm_crypt_request *dmreq)
 691{
 692	struct scatterlist *sg;
 693	u8 *dst;
 694
 695	if (bio_data_dir(dmreq->ctx->bio_in) != WRITE)
 696		return 0;
 697
 698	/* Apply whitening on ciphertext */
 699	sg = crypt_get_sg_data(cc, dmreq->sg_out);
 700	dst = kmap_local_page(sg_page(sg));
 701	crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset);
 702	kunmap_local(dst);
 703
 704	return 0;
 705}
 706
 707static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv,
 708				struct dm_crypt_request *dmreq)
 709{
 710	/* Used only for writes, there must be an additional space to store IV */
 711	get_random_bytes(iv, cc->iv_size);
 712	return 0;
 713}
 714
 715static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti,
 716			    const char *opts)
 717{
 718	if (crypt_integrity_aead(cc)) {
 719		ti->error = "AEAD transforms not supported for EBOIV";
 720		return -EINVAL;
 721	}
 722
 723	if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) {
 724		ti->error = "Block size of EBOIV cipher does not match IV size of block cipher";
 725		return -EINVAL;
 726	}
 727
 728	return 0;
 729}
 730
 731static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv,
 732			    struct dm_crypt_request *dmreq)
 733{
 734	struct crypto_skcipher *tfm = any_tfm(cc);
 735	struct skcipher_request *req;
 736	struct scatterlist src, dst;
 737	DECLARE_CRYPTO_WAIT(wait);
 738	unsigned int reqsize;
 739	int err;
 740	u8 *buf;
 741
 742	reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm);
 743	reqsize = ALIGN(reqsize, __alignof__(__le64));
 744
 745	req = kmalloc(reqsize + cc->iv_size, GFP_NOIO);
 746	if (!req)
 747		return -ENOMEM;
 748
 749	skcipher_request_set_tfm(req, tfm);
 750
 751	buf = (u8 *)req + reqsize;
 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	kfree_sensitive(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_local_page(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_local_page(sg_page(sg2));
 981		memcpy(data_offset, data2 + sg2->offset, cc->sector_size);
 982		kunmap_local(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_local(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->tuple_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->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift);
1149
1150	bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
1151
1152	ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata),
1153				     tag_len, offset_in_page(io->integrity_metadata));
1154	if (unlikely(ret != tag_len))
1155		return -ENOMEM;
1156
1157	return 0;
1158}
1159
1160static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti)
1161{
1162#ifdef CONFIG_BLK_DEV_INTEGRITY
1163	struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk);
1164	struct mapped_device *md = dm_table_get_md(ti->table);
1165
1166	/* We require an underlying device with non-PI metadata */
1167	if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) {
1168		ti->error = "Integrity profile not supported.";
1169		return -EINVAL;
1170	}
1171
1172	if (bi->metadata_size < cc->used_tag_size) {
1173		ti->error = "Integrity profile tag size mismatch.";
1174		return -EINVAL;
1175	}
1176	cc->tuple_size = bi->metadata_size;
1177	if (1 << bi->interval_exp != cc->sector_size) {
1178		ti->error = "Integrity profile sector size mismatch.";
1179		return -EINVAL;
1180	}
1181
1182	if (crypt_integrity_aead(cc)) {
1183		cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size;
1184		DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md),
1185		       cc->integrity_tag_size, cc->integrity_iv_size);
1186
1187		if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) {
1188			ti->error = "Integrity AEAD auth tag size is not supported.";
1189			return -EINVAL;
1190		}
1191	} else if (cc->integrity_iv_size)
1192		DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md),
1193		       cc->integrity_iv_size);
1194
1195	if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) {
1196		ti->error = "Not enough space for integrity tag in the profile.";
1197		return -EINVAL;
1198	}
1199
1200	return 0;
1201#else
1202	ti->error = "Integrity profile not supported.";
1203	return -EINVAL;
1204#endif
1205}
1206
1207static void crypt_convert_init(struct crypt_config *cc,
1208			       struct convert_context *ctx,
1209			       struct bio *bio_out, struct bio *bio_in,
1210			       sector_t sector)
1211{
1212	ctx->bio_in = bio_in;
1213	ctx->bio_out = bio_out;
1214	if (bio_in)
1215		ctx->iter_in = bio_in->bi_iter;
1216	if (bio_out)
1217		ctx->iter_out = bio_out->bi_iter;
1218	ctx->cc_sector = sector + cc->iv_offset;
1219	ctx->tag_offset = 0;
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
1256	return (__le64 *) ptr;
1257}
1258
1259static unsigned int *org_tag_of_dmreq(struct crypt_config *cc,
1260		       struct dm_crypt_request *dmreq)
1261{
1262	u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size +
1263		  cc->iv_size + sizeof(uint64_t);
1264
1265	return (unsigned int *)ptr;
1266}
1267
1268static void *tag_from_dmreq(struct crypt_config *cc,
1269				struct dm_crypt_request *dmreq)
1270{
1271	struct convert_context *ctx = dmreq->ctx;
1272	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
1273
1274	return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) *
1275		cc->tuple_size];
1276}
1277
1278static void *iv_tag_from_dmreq(struct crypt_config *cc,
1279			       struct dm_crypt_request *dmreq)
1280{
1281	return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size;
1282}
1283
1284static int crypt_convert_block_aead(struct crypt_config *cc,
1285				     struct convert_context *ctx,
1286				     struct aead_request *req,
1287				     unsigned int tag_offset)
1288{
1289	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1290	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1291	struct dm_crypt_request *dmreq;
1292	u8 *iv, *org_iv, *tag_iv, *tag;
1293	__le64 *sector;
1294	int r = 0;
1295
1296	BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size);
1297
1298	/* Reject unexpected unaligned bio. */
1299	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1300		return -EIO;
1301
1302	dmreq = dmreq_of_req(cc, req);
1303	dmreq->iv_sector = ctx->cc_sector;
1304	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1305		dmreq->iv_sector >>= cc->sector_shift;
1306	dmreq->ctx = ctx;
1307
1308	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1309
1310	sector = org_sector_of_dmreq(cc, dmreq);
1311	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1312
1313	iv = iv_of_dmreq(cc, dmreq);
1314	org_iv = org_iv_of_dmreq(cc, dmreq);
1315	tag = tag_from_dmreq(cc, dmreq);
1316	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1317
1318	/* AEAD request:
1319	 *  |----- AAD -------|------ DATA -------|-- AUTH TAG --|
1320	 *  | (authenticated) | (auth+encryption) |              |
1321	 *  | sector_LE |  IV |  sector in/out    |  tag in/out  |
1322	 */
1323	sg_init_table(dmreq->sg_in, 4);
1324	sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t));
1325	sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size);
1326	sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1327	sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size);
1328
1329	sg_init_table(dmreq->sg_out, 4);
1330	sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t));
1331	sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size);
1332	sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1333	sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size);
1334
1335	if (cc->iv_gen_ops) {
1336		/* For READs use IV stored in integrity metadata */
1337		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1338			memcpy(org_iv, tag_iv, cc->iv_size);
1339		} else {
1340			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1341			if (r < 0)
1342				return r;
1343			/* Store generated IV in integrity metadata */
1344			if (cc->integrity_iv_size)
1345				memcpy(tag_iv, org_iv, cc->iv_size);
1346		}
1347		/* Working copy of IV, to be modified in crypto API */
1348		memcpy(iv, org_iv, cc->iv_size);
1349	}
1350
1351	aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size);
1352	if (bio_data_dir(ctx->bio_in) == WRITE) {
1353		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1354				       cc->sector_size, iv);
1355		r = crypto_aead_encrypt(req);
1356		if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size)
1357			memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0,
1358			       cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size));
1359	} else {
1360		aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out,
1361				       cc->sector_size + cc->integrity_tag_size, iv);
1362		r = crypto_aead_decrypt(req);
1363	}
1364
1365	if (r == -EBADMSG) {
1366		sector_t s = le64_to_cpu(*sector);
1367
1368		ctx->aead_failed = true;
1369		if (ctx->aead_recheck) {
1370			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
1371				    ctx->bio_in->bi_bdev, s);
1372			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
1373					 ctx->bio_in, s, 0);
1374		}
1375	}
1376
1377	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1378		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1379
1380	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1381	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1382
1383	return r;
1384}
1385
1386static int crypt_convert_block_skcipher(struct crypt_config *cc,
1387					struct convert_context *ctx,
1388					struct skcipher_request *req,
1389					unsigned int tag_offset)
1390{
1391	struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in);
1392	struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out);
1393	struct scatterlist *sg_in, *sg_out;
1394	struct dm_crypt_request *dmreq;
1395	u8 *iv, *org_iv, *tag_iv;
1396	__le64 *sector;
1397	int r = 0;
1398
1399	/* Reject unexpected unaligned bio. */
1400	if (unlikely(bv_in.bv_len & (cc->sector_size - 1)))
1401		return -EIO;
1402
1403	dmreq = dmreq_of_req(cc, req);
1404	dmreq->iv_sector = ctx->cc_sector;
1405	if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
1406		dmreq->iv_sector >>= cc->sector_shift;
1407	dmreq->ctx = ctx;
1408
1409	*org_tag_of_dmreq(cc, dmreq) = tag_offset;
1410
1411	iv = iv_of_dmreq(cc, dmreq);
1412	org_iv = org_iv_of_dmreq(cc, dmreq);
1413	tag_iv = iv_tag_from_dmreq(cc, dmreq);
1414
1415	sector = org_sector_of_dmreq(cc, dmreq);
1416	*sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset);
1417
1418	/* For skcipher we use only the first sg item */
1419	sg_in  = &dmreq->sg_in[0];
1420	sg_out = &dmreq->sg_out[0];
1421
1422	sg_init_table(sg_in, 1);
1423	sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset);
1424
1425	sg_init_table(sg_out, 1);
1426	sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset);
1427
1428	if (cc->iv_gen_ops) {
1429		/* For READs use IV stored in integrity metadata */
1430		if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) {
1431			memcpy(org_iv, tag_iv, cc->integrity_iv_size);
1432		} else {
1433			r = cc->iv_gen_ops->generator(cc, org_iv, dmreq);
1434			if (r < 0)
1435				return r;
1436			/* Data can be already preprocessed in generator */
1437			if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags))
1438				sg_in = sg_out;
1439			/* Store generated IV in integrity metadata */
1440			if (cc->integrity_iv_size)
1441				memcpy(tag_iv, org_iv, cc->integrity_iv_size);
1442		}
1443		/* Working copy of IV, to be modified in crypto API */
1444		memcpy(iv, org_iv, cc->iv_size);
1445	}
1446
1447	skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv);
1448
1449	if (bio_data_dir(ctx->bio_in) == WRITE)
1450		r = crypto_skcipher_encrypt(req);
1451	else
1452		r = crypto_skcipher_decrypt(req);
1453
1454	if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post)
1455		r = cc->iv_gen_ops->post(cc, org_iv, dmreq);
1456
1457	bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size);
1458	bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size);
1459
1460	return r;
1461}
1462
1463static void kcryptd_async_done(void *async_req, int error);
1464
1465static int crypt_alloc_req_skcipher(struct crypt_config *cc,
1466				     struct convert_context *ctx)
1467{
1468	unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1);
1469
1470	if (!ctx->r.req) {
1471		ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1472		if (!ctx->r.req)
1473			return -ENOMEM;
1474	}
1475
1476	skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
1477
1478	/*
1479	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1480	 * requests if driver request queue is full.
1481	 */
1482	skcipher_request_set_callback(ctx->r.req,
1483	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1484	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
1485
1486	return 0;
1487}
1488
1489static int crypt_alloc_req_aead(struct crypt_config *cc,
1490				 struct convert_context *ctx)
1491{
1492	if (!ctx->r.req_aead) {
1493		ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO);
1494		if (!ctx->r.req_aead)
1495			return -ENOMEM;
1496	}
1497
1498	aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]);
1499
1500	/*
1501	 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
1502	 * requests if driver request queue is full.
1503	 */
1504	aead_request_set_callback(ctx->r.req_aead,
1505	    CRYPTO_TFM_REQ_MAY_BACKLOG,
1506	    kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead));
1507
1508	return 0;
1509}
1510
1511static int crypt_alloc_req(struct crypt_config *cc,
1512			    struct convert_context *ctx)
1513{
1514	if (crypt_integrity_aead(cc))
1515		return crypt_alloc_req_aead(cc, ctx);
1516	else
1517		return crypt_alloc_req_skcipher(cc, ctx);
1518}
1519
1520static void crypt_free_req_skcipher(struct crypt_config *cc,
1521				    struct skcipher_request *req, struct bio *base_bio)
1522{
1523	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1524
1525	if ((struct skcipher_request *)(io + 1) != req)
1526		mempool_free(req, &cc->req_pool);
1527}
1528
1529static void crypt_free_req_aead(struct crypt_config *cc,
1530				struct aead_request *req, struct bio *base_bio)
1531{
1532	struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size);
1533
1534	if ((struct aead_request *)(io + 1) != req)
1535		mempool_free(req, &cc->req_pool);
1536}
1537
1538static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio)
1539{
1540	if (crypt_integrity_aead(cc))
1541		crypt_free_req_aead(cc, req, base_bio);
1542	else
1543		crypt_free_req_skcipher(cc, req, base_bio);
1544}
1545
1546/*
1547 * Encrypt / decrypt data from one bio to another one (can be the same one)
1548 */
1549static blk_status_t crypt_convert(struct crypt_config *cc,
1550			 struct convert_context *ctx, bool atomic, bool reset_pending)
1551{
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, ctx->tag_offset);
1575		else
1576			r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->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->tag_offset++;
1597					ctx->cc_sector += sector_step;
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->tag_offset++;
1612			ctx->cc_sector += sector_step;
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			ctx->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 if it did then block
1647 * core should split the bio as needed).
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 *
1661 * In order to reduce allocation overhead, we try to allocate compound pages in
1662 * the first pass. If they are not available, we fall back to the mempool.
1663 */
1664static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size)
1665{
1666	struct crypt_config *cc = io->cc;
1667	struct bio *clone;
1668	unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1669	gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM;
1670	unsigned int remaining_size;
1671	unsigned int order = MAX_PAGE_ORDER;
1672
1673retry:
1674	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1675		mutex_lock(&cc->bio_alloc_lock);
1676
1677	clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf,
1678				 GFP_NOIO, &cc->bs);
1679	clone->bi_private = io;
1680	clone->bi_end_io = crypt_endio;
1681	clone->bi_ioprio = io->base_bio->bi_ioprio;
1682	clone->bi_iter.bi_sector = cc->start + io->sector;
1683
1684	remaining_size = size;
1685
1686	while (remaining_size) {
1687		struct page *pages;
1688		unsigned size_to_add;
1689		unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT);
1690		order = min(order, remaining_order);
1691
1692		while (order > 0) {
1693			if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) +
1694					(1 << order) > dm_crypt_pages_per_client))
1695				goto decrease_order;
1696			pages = alloc_pages(gfp_mask
1697				| __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP,
1698				order);
1699			if (likely(pages != NULL)) {
1700				percpu_counter_add(&cc->n_allocated_pages, 1 << order);
1701				goto have_pages;
1702			}
1703decrease_order:
1704			order--;
1705		}
1706
1707		pages = mempool_alloc(&cc->page_pool, gfp_mask);
1708		if (!pages) {
1709			crypt_free_buffer_pages(cc, clone);
1710			bio_put(clone);
1711			gfp_mask |= __GFP_DIRECT_RECLAIM;
1712			order = 0;
1713			goto retry;
1714		}
1715
1716have_pages:
1717		size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size);
1718		__bio_add_page(clone, pages, size_to_add, 0);
1719		remaining_size -= size_to_add;
1720	}
1721
1722	/* Allocate space for integrity tags */
1723	if (dm_crypt_integrity_io_alloc(io, clone)) {
1724		crypt_free_buffer_pages(cc, clone);
1725		bio_put(clone);
1726		clone = NULL;
1727	}
1728
1729	if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM))
1730		mutex_unlock(&cc->bio_alloc_lock);
1731
1732	return clone;
1733}
1734
1735static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone)
1736{
1737	struct folio_iter fi;
1738
1739	if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */
1740		bio_for_each_folio_all(fi, clone) {
1741			if (folio_test_large(fi.folio)) {
1742				percpu_counter_sub(&cc->n_allocated_pages,
1743						folio_nr_pages(fi.folio));
1744				folio_put(fi.folio);
1745			} else {
1746				mempool_free(&fi.folio->page, &cc->page_pool);
1747			}
1748		}
1749	}
1750}
1751
1752static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc,
1753			  struct bio *bio, sector_t sector)
1754{
1755	io->cc = cc;
1756	io->base_bio = bio;
1757	io->sector = sector;
1758	io->error = 0;
1759	io->ctx.aead_recheck = false;
1760	io->ctx.aead_failed = false;
1761	io->ctx.r.req = NULL;
1762	io->integrity_metadata = NULL;
1763	io->integrity_metadata_from_pool = false;
1764	atomic_set(&io->io_pending, 0);
1765}
1766
1767static void crypt_inc_pending(struct dm_crypt_io *io)
1768{
1769	atomic_inc(&io->io_pending);
1770}
1771
1772static void kcryptd_queue_read(struct dm_crypt_io *io);
1773
1774/*
1775 * One of the bios was finished. Check for completion of
1776 * the whole request and correctly clean up the buffer.
1777 */
1778static void crypt_dec_pending(struct dm_crypt_io *io)
1779{
1780	struct crypt_config *cc = io->cc;
1781	struct bio *base_bio = io->base_bio;
1782	blk_status_t error = io->error;
1783
1784	if (!atomic_dec_and_test(&io->io_pending))
1785		return;
1786
1787	if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) &&
1788	    cc->used_tag_size && bio_data_dir(base_bio) == READ) {
1789		io->ctx.aead_recheck = true;
1790		io->ctx.aead_failed = false;
1791		io->error = 0;
1792		kcryptd_queue_read(io);
1793		return;
1794	}
1795
1796	if (io->ctx.r.req)
1797		crypt_free_req(cc, io->ctx.r.req, base_bio);
1798
1799	if (unlikely(io->integrity_metadata_from_pool))
1800		mempool_free(io->integrity_metadata, &io->cc->tag_pool);
1801	else
1802		kfree(io->integrity_metadata);
1803
1804	base_bio->bi_status = error;
1805
1806	bio_endio(base_bio);
1807}
1808
1809/*
1810 * kcryptd/kcryptd_io:
1811 *
1812 * Needed because it would be very unwise to do decryption in an
1813 * interrupt context.
1814 *
1815 * kcryptd performs the actual encryption or decryption.
1816 *
1817 * kcryptd_io performs the IO submission.
1818 *
1819 * They must be separated as otherwise the final stages could be
1820 * starved by new requests which can block in the first stages due
1821 * to memory allocation.
1822 *
1823 * The work is done per CPU global for all dm-crypt instances.
1824 * They should not depend on each other and do not block.
1825 */
1826static void crypt_endio(struct bio *clone)
1827{
1828	struct dm_crypt_io *io = clone->bi_private;
1829	struct crypt_config *cc = io->cc;
1830	unsigned int rw = bio_data_dir(clone);
1831	blk_status_t error = clone->bi_status;
1832
1833	if (io->ctx.aead_recheck && !error) {
1834		kcryptd_queue_crypt(io);
1835		return;
1836	}
1837
1838	/*
1839	 * free the processed pages
1840	 */
1841	if (rw == WRITE || io->ctx.aead_recheck)
1842		crypt_free_buffer_pages(cc, clone);
1843
1844	bio_put(clone);
1845
1846	if (rw == READ && !error) {
1847		kcryptd_queue_crypt(io);
1848		return;
1849	}
1850
1851	if (unlikely(error))
1852		io->error = error;
1853
1854	crypt_dec_pending(io);
1855}
1856
1857#define CRYPT_MAP_READ_GFP GFP_NOWAIT
1858
1859static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp)
1860{
1861	struct crypt_config *cc = io->cc;
1862	struct bio *clone;
1863
1864	if (io->ctx.aead_recheck) {
1865		if (!(gfp & __GFP_DIRECT_RECLAIM))
1866			return 1;
1867		crypt_inc_pending(io);
1868		clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
1869		if (unlikely(!clone)) {
1870			crypt_dec_pending(io);
1871			return 1;
1872		}
1873		crypt_convert_init(cc, &io->ctx, clone, clone, io->sector);
1874		io->saved_bi_iter = clone->bi_iter;
1875		dm_submit_bio_remap(io->base_bio, clone);
1876		return 0;
1877	}
1878
1879	/*
1880	 * We need the original biovec array in order to decrypt the whole bio
1881	 * data *afterwards* -- thanks to immutable biovecs we don't need to
1882	 * worry about the block layer modifying the biovec array; so leverage
1883	 * bio_alloc_clone().
1884	 */
1885	clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs);
1886	if (!clone)
1887		return 1;
1888
1889	clone->bi_iter.bi_sector = cc->start + io->sector;
1890	clone->bi_private = io;
1891	clone->bi_end_io = crypt_endio;
1892
1893	crypt_inc_pending(io);
1894
1895	if (dm_crypt_integrity_io_alloc(io, clone)) {
1896		crypt_dec_pending(io);
1897		bio_put(clone);
1898		return 1;
1899	}
1900
1901	dm_submit_bio_remap(io->base_bio, clone);
1902	return 0;
1903}
1904
1905static void kcryptd_io_read_work(struct work_struct *work)
1906{
1907	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
1908
1909	crypt_inc_pending(io);
1910	if (kcryptd_io_read(io, GFP_NOIO))
1911		io->error = BLK_STS_RESOURCE;
1912	crypt_dec_pending(io);
1913}
1914
1915static void kcryptd_queue_read(struct dm_crypt_io *io)
1916{
1917	struct crypt_config *cc = io->cc;
1918
1919	INIT_WORK(&io->work, kcryptd_io_read_work);
1920	queue_work(cc->io_queue, &io->work);
1921}
1922
1923static void kcryptd_io_write(struct dm_crypt_io *io)
1924{
1925	struct bio *clone = io->ctx.bio_out;
1926
1927	dm_submit_bio_remap(io->base_bio, clone);
1928}
1929
1930#define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node)
1931
1932static int dmcrypt_write(void *data)
1933{
1934	struct crypt_config *cc = data;
1935	struct dm_crypt_io *io;
1936
1937	while (1) {
1938		struct rb_root write_tree;
1939		struct blk_plug plug;
1940
1941		spin_lock_irq(&cc->write_thread_lock);
1942continue_locked:
1943
1944		if (!RB_EMPTY_ROOT(&cc->write_tree))
1945			goto pop_from_list;
1946
1947		set_current_state(TASK_INTERRUPTIBLE);
1948
1949		spin_unlock_irq(&cc->write_thread_lock);
1950
1951		if (unlikely(kthread_should_stop())) {
1952			set_current_state(TASK_RUNNING);
1953			break;
1954		}
1955
1956		schedule();
1957
1958		spin_lock_irq(&cc->write_thread_lock);
1959		goto continue_locked;
1960
1961pop_from_list:
1962		write_tree = cc->write_tree;
1963		cc->write_tree = RB_ROOT;
1964		spin_unlock_irq(&cc->write_thread_lock);
1965
1966		BUG_ON(rb_parent(write_tree.rb_node));
1967
1968		/*
1969		 * Note: we cannot walk the tree here with rb_next because
1970		 * the structures may be freed when kcryptd_io_write is called.
1971		 */
1972		blk_start_plug(&plug);
1973		do {
1974			io = crypt_io_from_node(rb_first(&write_tree));
1975			rb_erase(&io->rb_node, &write_tree);
1976			kcryptd_io_write(io);
1977			cond_resched();
1978		} while (!RB_EMPTY_ROOT(&write_tree));
1979		blk_finish_plug(&plug);
1980	}
1981	return 0;
1982}
1983
1984static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async)
1985{
1986	struct bio *clone = io->ctx.bio_out;
1987	struct crypt_config *cc = io->cc;
1988	unsigned long flags;
1989	sector_t sector;
1990	struct rb_node **rbp, *parent;
1991
1992	if (unlikely(io->error)) {
1993		crypt_free_buffer_pages(cc, clone);
1994		bio_put(clone);
1995		crypt_dec_pending(io);
1996		return;
1997	}
1998
1999	/* crypt_convert should have filled the clone bio */
2000	BUG_ON(io->ctx.iter_out.bi_size);
2001
2002	if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) ||
2003	    test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) {
2004		dm_submit_bio_remap(io->base_bio, clone);
2005		return;
2006	}
2007
2008	spin_lock_irqsave(&cc->write_thread_lock, flags);
2009	if (RB_EMPTY_ROOT(&cc->write_tree))
2010		wake_up_process(cc->write_thread);
2011	rbp = &cc->write_tree.rb_node;
2012	parent = NULL;
2013	sector = io->sector;
2014	while (*rbp) {
2015		parent = *rbp;
2016		if (sector < crypt_io_from_node(parent)->sector)
2017			rbp = &(*rbp)->rb_left;
2018		else
2019			rbp = &(*rbp)->rb_right;
2020	}
2021	rb_link_node(&io->rb_node, parent, rbp);
2022	rb_insert_color(&io->rb_node, &cc->write_tree);
2023	spin_unlock_irqrestore(&cc->write_thread_lock, flags);
2024}
2025
2026static bool kcryptd_crypt_write_inline(struct crypt_config *cc,
2027				       struct convert_context *ctx)
2028
2029{
2030	if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags))
2031		return false;
2032
2033	/*
2034	 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering
2035	 * constraints so they do not need to be issued inline by
2036	 * kcryptd_crypt_write_convert().
2037	 */
2038	switch (bio_op(ctx->bio_in)) {
2039	case REQ_OP_WRITE:
2040	case REQ_OP_WRITE_ZEROES:
2041		return true;
2042	default:
2043		return false;
2044	}
2045}
2046
2047static void kcryptd_crypt_write_continue(struct work_struct *work)
2048{
2049	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2050	struct crypt_config *cc = io->cc;
2051	struct convert_context *ctx = &io->ctx;
2052	int crypt_finished;
2053	blk_status_t r;
2054
2055	wait_for_completion(&ctx->restart);
2056	reinit_completion(&ctx->restart);
2057
2058	r = crypt_convert(cc, &io->ctx, false, false);
2059	if (r)
2060		io->error = r;
2061	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2062	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2063		/* Wait for completion signaled by kcryptd_async_done() */
2064		wait_for_completion(&ctx->restart);
2065		crypt_finished = 1;
2066	}
2067
2068	/* Encryption was already finished, submit io now */
2069	if (crypt_finished)
2070		kcryptd_crypt_write_io_submit(io, 0);
2071
2072	crypt_dec_pending(io);
2073}
2074
2075static void kcryptd_crypt_write_convert(struct dm_crypt_io *io)
2076{
2077	struct crypt_config *cc = io->cc;
2078	struct convert_context *ctx = &io->ctx;
2079	struct bio *clone;
2080	int crypt_finished;
2081	blk_status_t r;
2082
2083	/*
2084	 * Prevent io from disappearing until this function completes.
2085	 */
2086	crypt_inc_pending(io);
2087	crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector);
2088
2089	clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size);
2090	if (unlikely(!clone)) {
2091		io->error = BLK_STS_IOERR;
2092		goto dec;
2093	}
2094
2095	io->ctx.bio_out = clone;
2096	io->ctx.iter_out = clone->bi_iter;
2097
2098	if (crypt_integrity_aead(cc)) {
2099		bio_copy_data(clone, io->base_bio);
2100		io->ctx.bio_in = clone;
2101		io->ctx.iter_in = clone->bi_iter;
2102	}
2103
2104	crypt_inc_pending(io);
2105	r = crypt_convert(cc, ctx,
2106			  test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true);
2107	/*
2108	 * Crypto API backlogged the request, because its queue was full
2109	 * and we're in softirq context, so continue from a workqueue
2110	 * (TODO: is it actually possible to be in softirq in the write path?)
2111	 */
2112	if (r == BLK_STS_DEV_RESOURCE) {
2113		INIT_WORK(&io->work, kcryptd_crypt_write_continue);
2114		queue_work(cc->crypt_queue, &io->work);
2115		return;
2116	}
2117	if (r)
2118		io->error = r;
2119	crypt_finished = atomic_dec_and_test(&ctx->cc_pending);
2120	if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) {
2121		/* Wait for completion signaled by kcryptd_async_done() */
2122		wait_for_completion(&ctx->restart);
2123		crypt_finished = 1;
2124	}
2125
2126	/* Encryption was already finished, submit io now */
2127	if (crypt_finished)
2128		kcryptd_crypt_write_io_submit(io, 0);
2129
2130dec:
2131	crypt_dec_pending(io);
2132}
2133
2134static void kcryptd_crypt_read_done(struct dm_crypt_io *io)
2135{
2136	if (io->ctx.aead_recheck) {
2137		if (!io->error) {
2138			io->ctx.bio_in->bi_iter = io->saved_bi_iter;
2139			bio_copy_data(io->base_bio, io->ctx.bio_in);
2140		}
2141		crypt_free_buffer_pages(io->cc, io->ctx.bio_in);
2142		bio_put(io->ctx.bio_in);
2143	}
2144	crypt_dec_pending(io);
2145}
2146
2147static void kcryptd_crypt_read_continue(struct work_struct *work)
2148{
2149	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2150	struct crypt_config *cc = io->cc;
2151	blk_status_t r;
2152
2153	wait_for_completion(&io->ctx.restart);
2154	reinit_completion(&io->ctx.restart);
2155
2156	r = crypt_convert(cc, &io->ctx, false, false);
2157	if (r)
2158		io->error = r;
2159
2160	if (atomic_dec_and_test(&io->ctx.cc_pending))
2161		kcryptd_crypt_read_done(io);
2162
2163	crypt_dec_pending(io);
2164}
2165
2166static void kcryptd_crypt_read_convert(struct dm_crypt_io *io)
2167{
2168	struct crypt_config *cc = io->cc;
2169	blk_status_t r;
2170
2171	crypt_inc_pending(io);
2172
2173	if (io->ctx.aead_recheck) {
2174		r = crypt_convert(cc, &io->ctx,
2175				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2176	} else {
2177		crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio,
2178				   io->sector);
2179
2180		r = crypt_convert(cc, &io->ctx,
2181				  test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true);
2182	}
2183	/*
2184	 * Crypto API backlogged the request, because its queue was full
2185	 * and we're in softirq context, so continue from a workqueue
2186	 */
2187	if (r == BLK_STS_DEV_RESOURCE) {
2188		INIT_WORK(&io->work, kcryptd_crypt_read_continue);
2189		queue_work(cc->crypt_queue, &io->work);
2190		return;
2191	}
2192	if (r)
2193		io->error = r;
2194
2195	if (atomic_dec_and_test(&io->ctx.cc_pending))
2196		kcryptd_crypt_read_done(io);
2197
2198	crypt_dec_pending(io);
2199}
2200
2201static void kcryptd_async_done(void *data, int error)
2202{
2203	struct dm_crypt_request *dmreq = data;
2204	struct convert_context *ctx = dmreq->ctx;
2205	struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx);
2206	struct crypt_config *cc = io->cc;
2207
2208	/*
2209	 * A request from crypto driver backlog is going to be processed now,
2210	 * finish the completion and continue in crypt_convert().
2211	 * (Callback will be called for the second time for this request.)
2212	 */
2213	if (error == -EINPROGRESS) {
2214		complete(&ctx->restart);
2215		return;
2216	}
2217
2218	if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post)
2219		error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq);
2220
2221	if (error == -EBADMSG) {
2222		sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq));
2223
2224		ctx->aead_failed = true;
2225		if (ctx->aead_recheck) {
2226			DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu",
2227				    ctx->bio_in->bi_bdev, s);
2228			dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead",
2229					 ctx->bio_in, s, 0);
2230		}
2231		io->error = BLK_STS_PROTECTION;
2232	} else if (error < 0)
2233		io->error = BLK_STS_IOERR;
2234
2235	crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio);
2236
2237	if (!atomic_dec_and_test(&ctx->cc_pending))
2238		return;
2239
2240	/*
2241	 * The request is fully completed: for inline writes, let
2242	 * kcryptd_crypt_write_convert() do the IO submission.
2243	 */
2244	if (bio_data_dir(io->base_bio) == READ) {
2245		kcryptd_crypt_read_done(io);
2246		return;
2247	}
2248
2249	if (kcryptd_crypt_write_inline(cc, ctx)) {
2250		complete(&ctx->restart);
2251		return;
2252	}
2253
2254	kcryptd_crypt_write_io_submit(io, 1);
2255}
2256
2257static void kcryptd_crypt(struct work_struct *work)
2258{
2259	struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work);
2260
2261	if (bio_data_dir(io->base_bio) == READ)
2262		kcryptd_crypt_read_convert(io);
2263	else
2264		kcryptd_crypt_write_convert(io);
2265}
2266
2267static void kcryptd_queue_crypt(struct dm_crypt_io *io)
2268{
2269	struct crypt_config *cc = io->cc;
2270
2271	if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) ||
2272	    (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) {
2273		/*
2274		 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context.
2275		 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but
2276		 * it is being executed with irqs disabled.
2277		 */
2278		if (in_hardirq() || irqs_disabled()) {
2279			INIT_WORK(&io->work, kcryptd_crypt);
2280			queue_work(system_bh_wq, &io->work);
2281			return;
2282		} else {
2283			kcryptd_crypt(&io->work);
2284			return;
2285		}
2286	}
2287
2288	INIT_WORK(&io->work, kcryptd_crypt);
2289	queue_work(cc->crypt_queue, &io->work);
2290}
2291
2292static void crypt_free_tfms_aead(struct crypt_config *cc)
2293{
2294	if (!cc->cipher_tfm.tfms_aead)
2295		return;
2296
2297	if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2298		crypto_free_aead(cc->cipher_tfm.tfms_aead[0]);
2299		cc->cipher_tfm.tfms_aead[0] = NULL;
2300	}
2301
2302	kfree(cc->cipher_tfm.tfms_aead);
2303	cc->cipher_tfm.tfms_aead = NULL;
2304}
2305
2306static void crypt_free_tfms_skcipher(struct crypt_config *cc)
2307{
2308	unsigned int i;
2309
2310	if (!cc->cipher_tfm.tfms)
2311		return;
2312
2313	for (i = 0; i < cc->tfms_count; i++)
2314		if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) {
2315			crypto_free_skcipher(cc->cipher_tfm.tfms[i]);
2316			cc->cipher_tfm.tfms[i] = NULL;
2317		}
2318
2319	kfree(cc->cipher_tfm.tfms);
2320	cc->cipher_tfm.tfms = NULL;
2321}
2322
2323static void crypt_free_tfms(struct crypt_config *cc)
2324{
2325	if (crypt_integrity_aead(cc))
2326		crypt_free_tfms_aead(cc);
2327	else
2328		crypt_free_tfms_skcipher(cc);
2329}
2330
2331static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode)
2332{
2333	unsigned int i;
2334	int err;
2335
2336	cc->cipher_tfm.tfms = kcalloc(cc->tfms_count,
2337				      sizeof(struct crypto_skcipher *),
2338				      GFP_KERNEL);
2339	if (!cc->cipher_tfm.tfms)
2340		return -ENOMEM;
2341
2342	for (i = 0; i < cc->tfms_count; i++) {
2343		cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0,
2344						CRYPTO_ALG_ALLOCATES_MEMORY);
2345		if (IS_ERR(cc->cipher_tfm.tfms[i])) {
2346			err = PTR_ERR(cc->cipher_tfm.tfms[i]);
2347			crypt_free_tfms(cc);
2348			return err;
2349		}
2350	}
2351
2352	/*
2353	 * dm-crypt performance can vary greatly depending on which crypto
2354	 * algorithm implementation is used.  Help people debug performance
2355	 * problems by logging the ->cra_driver_name.
2356	 */
2357	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2358	       crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name);
2359	return 0;
2360}
2361
2362static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode)
2363{
2364	int err;
2365
2366	cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL);
2367	if (!cc->cipher_tfm.tfms)
2368		return -ENOMEM;
2369
2370	cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0,
2371						CRYPTO_ALG_ALLOCATES_MEMORY);
2372	if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) {
2373		err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]);
2374		crypt_free_tfms(cc);
2375		return err;
2376	}
2377
2378	DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode,
2379	       crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name);
2380	return 0;
2381}
2382
2383static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode)
2384{
2385	if (crypt_integrity_aead(cc))
2386		return crypt_alloc_tfms_aead(cc, ciphermode);
2387	else
2388		return crypt_alloc_tfms_skcipher(cc, ciphermode);
2389}
2390
2391static unsigned int crypt_subkey_size(struct crypt_config *cc)
2392{
2393	return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count);
2394}
2395
2396static unsigned int crypt_authenckey_size(struct crypt_config *cc)
2397{
2398	return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param));
2399}
2400
2401/*
2402 * If AEAD is composed like authenc(hmac(sha256),xts(aes)),
2403 * the key must be for some reason in special format.
2404 * This funcion converts cc->key to this special format.
2405 */
2406static void crypt_copy_authenckey(char *p, const void *key,
2407				  unsigned int enckeylen, unsigned int authkeylen)
2408{
2409	struct crypto_authenc_key_param *param;
2410	struct rtattr *rta;
2411
2412	rta = (struct rtattr *)p;
2413	param = RTA_DATA(rta);
2414	param->enckeylen = cpu_to_be32(enckeylen);
2415	rta->rta_len = RTA_LENGTH(sizeof(*param));
2416	rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM;
2417	p += RTA_SPACE(sizeof(*param));
2418	memcpy(p, key + enckeylen, authkeylen);
2419	p += authkeylen;
2420	memcpy(p, key, enckeylen);
2421}
2422
2423static int crypt_setkey(struct crypt_config *cc)
2424{
2425	unsigned int subkey_size;
2426	int err = 0, i, r;
2427
2428	/* Ignore extra keys (which are used for IV etc) */
2429	subkey_size = crypt_subkey_size(cc);
2430
2431	if (crypt_integrity_hmac(cc)) {
2432		if (subkey_size < cc->key_mac_size)
2433			return -EINVAL;
2434
2435		crypt_copy_authenckey(cc->authenc_key, cc->key,
2436				      subkey_size - cc->key_mac_size,
2437				      cc->key_mac_size);
2438	}
2439
2440	for (i = 0; i < cc->tfms_count; i++) {
2441		if (crypt_integrity_hmac(cc))
2442			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2443				cc->authenc_key, crypt_authenckey_size(cc));
2444		else if (crypt_integrity_aead(cc))
2445			r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i],
2446					       cc->key + (i * subkey_size),
2447					       subkey_size);
2448		else
2449			r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i],
2450						   cc->key + (i * subkey_size),
2451						   subkey_size);
2452		if (r)
2453			err = r;
2454	}
2455
2456	if (crypt_integrity_hmac(cc))
2457		memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc));
2458
2459	return err;
2460}
2461
2462#ifdef CONFIG_KEYS
2463
2464static bool contains_whitespace(const char *str)
2465{
2466	while (*str)
2467		if (isspace(*str++))
2468			return true;
2469	return false;
2470}
2471
2472static int set_key_user(struct crypt_config *cc, struct key *key)
2473{
2474	const struct user_key_payload *ukp;
2475
2476	ukp = user_key_payload_locked(key);
2477	if (!ukp)
2478		return -EKEYREVOKED;
2479
2480	if (cc->key_size != ukp->datalen)
2481		return -EINVAL;
2482
2483	memcpy(cc->key, ukp->data, cc->key_size);
2484
2485	return 0;
2486}
2487
2488static int set_key_encrypted(struct crypt_config *cc, struct key *key)
2489{
2490	const struct encrypted_key_payload *ekp;
2491
2492	ekp = key->payload.data[0];
2493	if (!ekp)
2494		return -EKEYREVOKED;
2495
2496	if (cc->key_size != ekp->decrypted_datalen)
2497		return -EINVAL;
2498
2499	memcpy(cc->key, ekp->decrypted_data, cc->key_size);
2500
2501	return 0;
2502}
2503
2504static int set_key_trusted(struct crypt_config *cc, struct key *key)
2505{
2506	const struct trusted_key_payload *tkp;
2507
2508	tkp = key->payload.data[0];
2509	if (!tkp)
2510		return -EKEYREVOKED;
2511
2512	if (cc->key_size != tkp->key_len)
2513		return -EINVAL;
2514
2515	memcpy(cc->key, tkp->key, cc->key_size);
2516
2517	return 0;
2518}
2519
2520static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2521{
2522	char *new_key_string, *key_desc;
2523	int ret;
2524	struct key_type *type;
2525	struct key *key;
2526	int (*set_key)(struct crypt_config *cc, struct key *key);
2527
2528	/*
2529	 * Reject key_string with whitespace. dm core currently lacks code for
2530	 * proper whitespace escaping in arguments on DM_TABLE_STATUS path.
2531	 */
2532	if (contains_whitespace(key_string)) {
2533		DMERR("whitespace chars not allowed in key string");
2534		return -EINVAL;
2535	}
2536
2537	/* look for next ':' separating key_type from key_description */
2538	key_desc = strchr(key_string, ':');
2539	if (!key_desc || key_desc == key_string || !strlen(key_desc + 1))
2540		return -EINVAL;
2541
2542	if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) {
2543		type = &key_type_logon;
2544		set_key = set_key_user;
2545	} else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) {
2546		type = &key_type_user;
2547		set_key = set_key_user;
2548	} else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) &&
2549		   !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) {
2550		type = &key_type_encrypted;
2551		set_key = set_key_encrypted;
2552	} else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) &&
2553		   !strncmp(key_string, "trusted:", key_desc - key_string + 1)) {
2554		type = &key_type_trusted;
2555		set_key = set_key_trusted;
2556	} else {
2557		return -EINVAL;
2558	}
2559
2560	new_key_string = kstrdup(key_string, GFP_KERNEL);
2561	if (!new_key_string)
2562		return -ENOMEM;
2563
2564	key = request_key(type, key_desc + 1, NULL);
2565	if (IS_ERR(key)) {
2566		ret = PTR_ERR(key);
2567		goto free_new_key_string;
2568	}
2569
2570	down_read(&key->sem);
2571	ret = set_key(cc, key);
2572	up_read(&key->sem);
2573	key_put(key);
2574	if (ret < 0)
2575		goto free_new_key_string;
2576
2577	/* clear the flag since following operations may invalidate previously valid key */
2578	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2579
2580	ret = crypt_setkey(cc);
2581	if (ret)
2582		goto free_new_key_string;
2583
2584	set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2585	kfree_sensitive(cc->key_string);
2586	cc->key_string = new_key_string;
2587	return 0;
2588
2589free_new_key_string:
2590	kfree_sensitive(new_key_string);
2591	return ret;
2592}
2593
2594static int get_key_size(char **key_string)
2595{
2596	char *colon, dummy;
2597	int ret;
2598
2599	if (*key_string[0] != ':')
2600		return strlen(*key_string) >> 1;
2601
2602	/* look for next ':' in key string */
2603	colon = strpbrk(*key_string + 1, ":");
2604	if (!colon)
2605		return -EINVAL;
2606
2607	if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':')
2608		return -EINVAL;
2609
2610	*key_string = colon;
2611
2612	/* remaining key string should be :<logon|user>:<key_desc> */
2613
2614	return ret;
2615}
2616
2617#else
2618
2619static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string)
2620{
2621	return -EINVAL;
2622}
2623
2624static int get_key_size(char **key_string)
2625{
2626	return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1);
2627}
2628
2629#endif /* CONFIG_KEYS */
2630
2631static int crypt_set_key(struct crypt_config *cc, char *key)
2632{
2633	int r = -EINVAL;
2634	int key_string_len = strlen(key);
2635
2636	/* Hyphen (which gives a key_size of zero) means there is no key. */
2637	if (!cc->key_size && strcmp(key, "-"))
2638		goto out;
2639
2640	/* ':' means the key is in kernel keyring, short-circuit normal key processing */
2641	if (key[0] == ':') {
2642		r = crypt_set_keyring_key(cc, key + 1);
2643		goto out;
2644	}
2645
2646	/* clear the flag since following operations may invalidate previously valid key */
2647	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2648
2649	/* wipe references to any kernel keyring key */
2650	kfree_sensitive(cc->key_string);
2651	cc->key_string = NULL;
2652
2653	/* Decode key from its hex representation. */
2654	if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0)
2655		goto out;
2656
2657	r = crypt_setkey(cc);
2658	if (!r)
2659		set_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2660
2661out:
2662	/* Hex key string not needed after here, so wipe it. */
2663	memset(key, '0', key_string_len);
2664
2665	return r;
2666}
2667
2668static int crypt_wipe_key(struct crypt_config *cc)
2669{
2670	int r;
2671
2672	clear_bit(DM_CRYPT_KEY_VALID, &cc->flags);
2673	get_random_bytes(&cc->key, cc->key_size);
2674
2675	/* Wipe IV private keys */
2676	if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) {
2677		r = cc->iv_gen_ops->wipe(cc);
2678		if (r)
2679			return r;
2680	}
2681
2682	kfree_sensitive(cc->key_string);
2683	cc->key_string = NULL;
2684	r = crypt_setkey(cc);
2685	memset(&cc->key, 0, cc->key_size * sizeof(u8));
2686
2687	return r;
2688}
2689
2690static void crypt_calculate_pages_per_client(void)
2691{
2692	unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100;
2693
2694	if (!dm_crypt_clients_n)
2695		return;
2696
2697	pages /= dm_crypt_clients_n;
2698	if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT)
2699		pages = DM_CRYPT_MIN_PAGES_PER_CLIENT;
2700	dm_crypt_pages_per_client = pages;
2701}
2702
2703static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data)
2704{
2705	struct crypt_config *cc = pool_data;
2706	struct page *page;
2707
2708	/*
2709	 * Note, percpu_counter_read_positive() may over (and under) estimate
2710	 * the current usage by at most (batch - 1) * num_online_cpus() pages,
2711	 * but avoids potential spinlock contention of an exact result.
2712	 */
2713	if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) &&
2714	    likely(gfp_mask & __GFP_NORETRY))
2715		return NULL;
2716
2717	page = alloc_page(gfp_mask);
2718	if (likely(page != NULL))
2719		percpu_counter_add(&cc->n_allocated_pages, 1);
2720
2721	return page;
2722}
2723
2724static void crypt_page_free(void *page, void *pool_data)
2725{
2726	struct crypt_config *cc = pool_data;
2727
2728	__free_page(page);
2729	percpu_counter_sub(&cc->n_allocated_pages, 1);
2730}
2731
2732static void crypt_dtr(struct dm_target *ti)
2733{
2734	struct crypt_config *cc = ti->private;
2735
2736	ti->private = NULL;
2737
2738	if (!cc)
2739		return;
2740
2741	if (cc->write_thread)
2742		kthread_stop(cc->write_thread);
2743
2744	if (cc->io_queue)
2745		destroy_workqueue(cc->io_queue);
2746	if (cc->crypt_queue)
2747		destroy_workqueue(cc->crypt_queue);
2748
2749	if (cc->workqueue_id)
2750		ida_free(&workqueue_ida, cc->workqueue_id);
2751
2752	crypt_free_tfms(cc);
2753
2754	bioset_exit(&cc->bs);
2755
2756	mempool_exit(&cc->page_pool);
2757	mempool_exit(&cc->req_pool);
2758	mempool_exit(&cc->tag_pool);
2759
2760	WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0);
2761	percpu_counter_destroy(&cc->n_allocated_pages);
2762
2763	if (cc->iv_gen_ops && cc->iv_gen_ops->dtr)
2764		cc->iv_gen_ops->dtr(cc);
2765
2766	if (cc->dev)
2767		dm_put_device(ti, cc->dev);
2768
2769	kfree_sensitive(cc->cipher_string);
2770	kfree_sensitive(cc->key_string);
2771	kfree_sensitive(cc->cipher_auth);
2772	kfree_sensitive(cc->authenc_key);
2773
2774	mutex_destroy(&cc->bio_alloc_lock);
2775
2776	/* Must zero key material before freeing */
2777	kfree_sensitive(cc);
2778
2779	spin_lock(&dm_crypt_clients_lock);
2780	WARN_ON(!dm_crypt_clients_n);
2781	dm_crypt_clients_n--;
2782	crypt_calculate_pages_per_client();
2783	spin_unlock(&dm_crypt_clients_lock);
2784
2785	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
2786}
2787
2788static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode)
2789{
2790	struct crypt_config *cc = ti->private;
2791
2792	if (crypt_integrity_aead(cc))
2793		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2794	else
2795		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2796
2797	if (cc->iv_size)
2798		/* at least a 64 bit sector number should fit in our buffer */
2799		cc->iv_size = max(cc->iv_size,
2800				  (unsigned int)(sizeof(u64) / sizeof(u8)));
2801	else if (ivmode) {
2802		DMWARN("Selected cipher does not support IVs");
2803		ivmode = NULL;
2804	}
2805
2806	/* Choose ivmode, see comments at iv code. */
2807	if (ivmode == NULL)
2808		cc->iv_gen_ops = NULL;
2809	else if (strcmp(ivmode, "plain") == 0)
2810		cc->iv_gen_ops = &crypt_iv_plain_ops;
2811	else if (strcmp(ivmode, "plain64") == 0)
2812		cc->iv_gen_ops = &crypt_iv_plain64_ops;
2813	else if (strcmp(ivmode, "plain64be") == 0)
2814		cc->iv_gen_ops = &crypt_iv_plain64be_ops;
2815	else if (strcmp(ivmode, "essiv") == 0)
2816		cc->iv_gen_ops = &crypt_iv_essiv_ops;
2817	else if (strcmp(ivmode, "benbi") == 0)
2818		cc->iv_gen_ops = &crypt_iv_benbi_ops;
2819	else if (strcmp(ivmode, "null") == 0)
2820		cc->iv_gen_ops = &crypt_iv_null_ops;
2821	else if (strcmp(ivmode, "eboiv") == 0)
2822		cc->iv_gen_ops = &crypt_iv_eboiv_ops;
2823	else if (strcmp(ivmode, "elephant") == 0) {
2824		cc->iv_gen_ops = &crypt_iv_elephant_ops;
2825		cc->key_parts = 2;
2826		cc->key_extra_size = cc->key_size / 2;
2827		if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE)
2828			return -EINVAL;
2829		set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags);
2830	} else if (strcmp(ivmode, "lmk") == 0) {
2831		cc->iv_gen_ops = &crypt_iv_lmk_ops;
2832		/*
2833		 * Version 2 and 3 is recognised according
2834		 * to length of provided multi-key string.
2835		 * If present (version 3), last key is used as IV seed.
2836		 * All keys (including IV seed) are always the same size.
2837		 */
2838		if (cc->key_size % cc->key_parts) {
2839			cc->key_parts++;
2840			cc->key_extra_size = cc->key_size / cc->key_parts;
2841		}
2842	} else if (strcmp(ivmode, "tcw") == 0) {
2843		cc->iv_gen_ops = &crypt_iv_tcw_ops;
2844		cc->key_parts += 2; /* IV + whitening */
2845		cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE;
2846	} else if (strcmp(ivmode, "random") == 0) {
2847		cc->iv_gen_ops = &crypt_iv_random_ops;
2848		/* Need storage space in integrity fields. */
2849		cc->integrity_iv_size = cc->iv_size;
2850	} else {
2851		ti->error = "Invalid IV mode";
2852		return -EINVAL;
2853	}
2854
2855	return 0;
2856}
2857
2858/*
2859 * Workaround to parse HMAC algorithm from AEAD crypto API spec.
2860 * The HMAC is needed to calculate tag size (HMAC digest size).
2861 * This should be probably done by crypto-api calls (once available...)
2862 */
2863static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api)
2864{
2865	char *start, *end, *mac_alg = NULL;
2866	struct crypto_ahash *mac;
2867
2868	if (!strstarts(cipher_api, "authenc("))
2869		return 0;
2870
2871	start = strchr(cipher_api, '(');
2872	end = strchr(cipher_api, ',');
2873	if (!start || !end || ++start > end)
2874		return -EINVAL;
2875
2876	mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL);
2877	if (!mac_alg)
2878		return -ENOMEM;
2879
2880	mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
2881	kfree(mac_alg);
2882
2883	if (IS_ERR(mac))
2884		return PTR_ERR(mac);
2885
2886	if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
2887		cc->key_mac_size = crypto_ahash_digestsize(mac);
2888	crypto_free_ahash(mac);
2889
2890	cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL);
2891	if (!cc->authenc_key)
2892		return -ENOMEM;
2893
2894	return 0;
2895}
2896
2897static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key,
2898				char **ivmode, char **ivopts)
2899{
2900	struct crypt_config *cc = ti->private;
2901	char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME];
2902	int ret = -EINVAL;
2903
2904	cc->tfms_count = 1;
2905
2906	/*
2907	 * New format (capi: prefix)
2908	 * capi:cipher_api_spec-iv:ivopts
2909	 */
2910	tmp = &cipher_in[strlen("capi:")];
2911
2912	/* Separate IV options if present, it can contain another '-' in hash name */
2913	*ivopts = strrchr(tmp, ':');
2914	if (*ivopts) {
2915		**ivopts = '\0';
2916		(*ivopts)++;
2917	}
2918	/* Parse IV mode */
2919	*ivmode = strrchr(tmp, '-');
2920	if (*ivmode) {
2921		**ivmode = '\0';
2922		(*ivmode)++;
2923	}
2924	/* The rest is crypto API spec */
2925	cipher_api = tmp;
2926
2927	/* Alloc AEAD, can be used only in new format. */
2928	if (crypt_integrity_aead(cc)) {
2929		ret = crypt_ctr_auth_cipher(cc, cipher_api);
2930		if (ret < 0) {
2931			ti->error = "Invalid AEAD cipher spec";
2932			return ret;
2933		}
2934	}
2935
2936	if (*ivmode && !strcmp(*ivmode, "lmk"))
2937		cc->tfms_count = 64;
2938
2939	if (*ivmode && !strcmp(*ivmode, "essiv")) {
2940		if (!*ivopts) {
2941			ti->error = "Digest algorithm missing for ESSIV mode";
2942			return -EINVAL;
2943		}
2944		ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)",
2945			       cipher_api, *ivopts);
2946		if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
2947			ti->error = "Cannot allocate cipher string";
2948			return -ENOMEM;
2949		}
2950		cipher_api = buf;
2951	}
2952
2953	cc->key_parts = cc->tfms_count;
2954
2955	/* Allocate cipher */
2956	ret = crypt_alloc_tfms(cc, cipher_api);
2957	if (ret < 0) {
2958		ti->error = "Error allocating crypto tfm";
2959		return ret;
2960	}
2961
2962	if (crypt_integrity_aead(cc))
2963		cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc));
2964	else
2965		cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc));
2966
2967	return 0;
2968}
2969
2970static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key,
2971				char **ivmode, char **ivopts)
2972{
2973	struct crypt_config *cc = ti->private;
2974	char *tmp, *cipher, *chainmode, *keycount;
2975	char *cipher_api = NULL;
2976	int ret = -EINVAL;
2977	char dummy;
2978
2979	if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) {
2980		ti->error = "Bad cipher specification";
2981		return -EINVAL;
2982	}
2983
2984	/*
2985	 * Legacy dm-crypt cipher specification
2986	 * cipher[:keycount]-mode-iv:ivopts
2987	 */
2988	tmp = cipher_in;
2989	keycount = strsep(&tmp, "-");
2990	cipher = strsep(&keycount, ":");
2991
2992	if (!keycount)
2993		cc->tfms_count = 1;
2994	else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 ||
2995		 !is_power_of_2(cc->tfms_count)) {
2996		ti->error = "Bad cipher key count specification";
2997		return -EINVAL;
2998	}
2999	cc->key_parts = cc->tfms_count;
3000
3001	chainmode = strsep(&tmp, "-");
3002	*ivmode = strsep(&tmp, ":");
3003	*ivopts = tmp;
3004
3005	/*
3006	 * For compatibility with the original dm-crypt mapping format, if
3007	 * only the cipher name is supplied, use cbc-plain.
3008	 */
3009	if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) {
3010		chainmode = "cbc";
3011		*ivmode = "plain";
3012	}
3013
3014	if (strcmp(chainmode, "ecb") && !*ivmode) {
3015		ti->error = "IV mechanism required";
3016		return -EINVAL;
3017	}
3018
3019	cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL);
3020	if (!cipher_api)
3021		goto bad_mem;
3022
3023	if (*ivmode && !strcmp(*ivmode, "essiv")) {
3024		if (!*ivopts) {
3025			ti->error = "Digest algorithm missing for ESSIV mode";
3026			kfree(cipher_api);
3027			return -EINVAL;
3028		}
3029		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3030			       "essiv(%s(%s),%s)", chainmode, cipher, *ivopts);
3031	} else {
3032		ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME,
3033			       "%s(%s)", chainmode, cipher);
3034	}
3035	if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) {
3036		kfree(cipher_api);
3037		goto bad_mem;
3038	}
3039
3040	/* Allocate cipher */
3041	ret = crypt_alloc_tfms(cc, cipher_api);
3042	if (ret < 0) {
3043		ti->error = "Error allocating crypto tfm";
3044		kfree(cipher_api);
3045		return ret;
3046	}
3047	kfree(cipher_api);
3048
3049	return 0;
3050bad_mem:
3051	ti->error = "Cannot allocate cipher strings";
3052	return -ENOMEM;
3053}
3054
3055static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key)
3056{
3057	struct crypt_config *cc = ti->private;
3058	char *ivmode = NULL, *ivopts = NULL;
3059	int ret;
3060
3061	cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL);
3062	if (!cc->cipher_string) {
3063		ti->error = "Cannot allocate cipher strings";
3064		return -ENOMEM;
3065	}
3066
3067	if (strstarts(cipher_in, "capi:"))
3068		ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts);
3069	else
3070		ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts);
3071	if (ret)
3072		return ret;
3073
3074	/* Initialize IV */
3075	ret = crypt_ctr_ivmode(ti, ivmode);
3076	if (ret < 0)
3077		return ret;
3078
3079	/* Initialize and set key */
3080	ret = crypt_set_key(cc, key);
3081	if (ret < 0) {
3082		ti->error = "Error decoding and setting key";
3083		return ret;
3084	}
3085
3086	/* Allocate IV */
3087	if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) {
3088		ret = cc->iv_gen_ops->ctr(cc, ti, ivopts);
3089		if (ret < 0) {
3090			ti->error = "Error creating IV";
3091			return ret;
3092		}
3093	}
3094
3095	/* Initialize IV (set keys for ESSIV etc) */
3096	if (cc->iv_gen_ops && cc->iv_gen_ops->init) {
3097		ret = cc->iv_gen_ops->init(cc);
3098		if (ret < 0) {
3099			ti->error = "Error initialising IV";
3100			return ret;
3101		}
3102	}
3103
3104	/* wipe the kernel key payload copy */
3105	if (cc->key_string)
3106		memset(cc->key, 0, cc->key_size * sizeof(u8));
3107
3108	return ret;
3109}
3110
3111static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv)
3112{
3113	struct crypt_config *cc = ti->private;
3114	struct dm_arg_set as;
3115	static const struct dm_arg _args[] = {
3116		{0, 9, "Invalid number of feature args"},
3117	};
3118	unsigned int opt_params, val;
3119	const char *opt_string, *sval;
3120	char dummy;
3121	int ret;
3122
3123	/* Optional parameters */
3124	as.argc = argc;
3125	as.argv = argv;
3126
3127	ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error);
3128	if (ret)
3129		return ret;
3130
3131	while (opt_params--) {
3132		opt_string = dm_shift_arg(&as);
3133		if (!opt_string) {
3134			ti->error = "Not enough feature arguments";
3135			return -EINVAL;
3136		}
3137
3138		if (!strcasecmp(opt_string, "allow_discards"))
3139			ti->num_discard_bios = 1;
3140
3141		else if (!strcasecmp(opt_string, "same_cpu_crypt"))
3142			set_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3143		else if (!strcasecmp(opt_string, "high_priority"))
3144			set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3145
3146		else if (!strcasecmp(opt_string, "submit_from_crypt_cpus"))
3147			set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3148		else if (!strcasecmp(opt_string, "no_read_workqueue"))
3149			set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3150		else if (!strcasecmp(opt_string, "no_write_workqueue"))
3151			set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3152		else if (sscanf(opt_string, "integrity:%u:", &val) == 1) {
3153			if (val == 0 || val > MAX_TAG_SIZE) {
3154				ti->error = "Invalid integrity arguments";
3155				return -EINVAL;
3156			}
3157			cc->used_tag_size = val;
3158			sval = strchr(opt_string + strlen("integrity:"), ':') + 1;
3159			if (!strcasecmp(sval, "aead")) {
3160				set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags);
3161			} else if (strcasecmp(sval, "none")) {
3162				ti->error = "Unknown integrity profile";
3163				return -EINVAL;
3164			}
3165
3166			cc->cipher_auth = kstrdup(sval, GFP_KERNEL);
3167			if (!cc->cipher_auth)
3168				return -ENOMEM;
3169		} else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) {
3170			if (!val) {
3171				ti->error = "Invalid integrity_key_size argument";
3172				return -EINVAL;
3173			}
3174			cc->key_mac_size = val;
3175			set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3176		} else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) {
3177			if (cc->sector_size < (1 << SECTOR_SHIFT) ||
3178			    cc->sector_size > 4096 ||
3179			    (cc->sector_size & (cc->sector_size - 1))) {
3180				ti->error = "Invalid feature value for sector_size";
3181				return -EINVAL;
3182			}
3183			if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) {
3184				ti->error = "Device size is not multiple of sector_size feature";
3185				return -EINVAL;
3186			}
3187			cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT;
3188		} else if (!strcasecmp(opt_string, "iv_large_sectors"))
3189			set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3190		else {
3191			ti->error = "Invalid feature arguments";
3192			return -EINVAL;
3193		}
3194	}
3195
3196	return 0;
3197}
3198
3199#ifdef CONFIG_BLK_DEV_ZONED
3200static int crypt_report_zones(struct dm_target *ti,
3201		struct dm_report_zones_args *args, unsigned int nr_zones)
3202{
3203	struct crypt_config *cc = ti->private;
3204
3205	return dm_report_zones(cc->dev->bdev, cc->start,
3206			cc->start + dm_target_offset(ti, args->next_sector),
3207			args, nr_zones);
3208}
3209#else
3210#define crypt_report_zones NULL
3211#endif
3212
3213/*
3214 * Construct an encryption mapping:
3215 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start>
3216 */
3217static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3218{
3219	struct crypt_config *cc;
3220	const char *devname = dm_table_device_name(ti->table);
3221	int key_size, wq_id;
3222	unsigned int align_mask;
3223	unsigned int common_wq_flags;
3224	unsigned long long tmpll;
3225	int ret;
3226	size_t iv_size_padding, additional_req_size;
3227	char dummy;
3228
3229	if (argc < 5) {
3230		ti->error = "Not enough arguments";
3231		return -EINVAL;
3232	}
3233
3234	key_size = get_key_size(&argv[1]);
3235	if (key_size < 0) {
3236		ti->error = "Cannot parse key size";
3237		return -EINVAL;
3238	}
3239
3240	cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL);
3241	if (!cc) {
3242		ti->error = "Cannot allocate encryption context";
3243		return -ENOMEM;
3244	}
3245	cc->key_size = key_size;
3246	cc->sector_size = (1 << SECTOR_SHIFT);
3247	cc->sector_shift = 0;
3248
3249	ti->private = cc;
3250
3251	spin_lock(&dm_crypt_clients_lock);
3252	dm_crypt_clients_n++;
3253	crypt_calculate_pages_per_client();
3254	spin_unlock(&dm_crypt_clients_lock);
3255
3256	ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL);
3257	if (ret < 0)
3258		goto bad;
3259
3260	/* Optional parameters need to be read before cipher constructor */
3261	if (argc > 5) {
3262		ret = crypt_ctr_optional(ti, argc - 5, &argv[5]);
3263		if (ret)
3264			goto bad;
3265	}
3266
3267	ret = crypt_ctr_cipher(ti, argv[0], argv[1]);
3268	if (ret < 0)
3269		goto bad;
3270
3271	if (crypt_integrity_aead(cc)) {
3272		cc->dmreq_start = sizeof(struct aead_request);
3273		cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc));
3274		align_mask = crypto_aead_alignmask(any_tfm_aead(cc));
3275	} else {
3276		cc->dmreq_start = sizeof(struct skcipher_request);
3277		cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc));
3278		align_mask = crypto_skcipher_alignmask(any_tfm(cc));
3279	}
3280	cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request));
3281
3282	if (align_mask < CRYPTO_MINALIGN) {
3283		/* Allocate the padding exactly */
3284		iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request))
3285				& align_mask;
3286	} else {
3287		/*
3288		 * If the cipher requires greater alignment than kmalloc
3289		 * alignment, we don't know the exact position of the
3290		 * initialization vector. We must assume worst case.
3291		 */
3292		iv_size_padding = align_mask;
3293	}
3294
3295	/*  ...| IV + padding | original IV | original sec. number | bio tag offset | */
3296	additional_req_size = sizeof(struct dm_crypt_request) +
3297		iv_size_padding + cc->iv_size +
3298		cc->iv_size +
3299		sizeof(uint64_t) +
3300		sizeof(unsigned int);
3301
3302	ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size);
3303	if (ret) {
3304		ti->error = "Cannot allocate crypt request mempool";
3305		goto bad;
3306	}
3307
3308	cc->per_bio_data_size = ti->per_io_data_size =
3309		ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size,
3310		      ARCH_DMA_MINALIGN);
3311
3312	ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc);
3313	if (ret) {
3314		ti->error = "Cannot allocate page mempool";
3315		goto bad;
3316	}
3317
3318	ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS);
3319	if (ret) {
3320		ti->error = "Cannot allocate crypt bioset";
3321		goto bad;
3322	}
3323
3324	mutex_init(&cc->bio_alloc_lock);
3325
3326	ret = -EINVAL;
3327	if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) ||
3328	    (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) {
3329		ti->error = "Invalid iv_offset sector";
3330		goto bad;
3331	}
3332	cc->iv_offset = tmpll;
3333
3334	ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev);
3335	if (ret) {
3336		ti->error = "Device lookup failed";
3337		goto bad;
3338	}
3339
3340	ret = -EINVAL;
3341	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
3342		ti->error = "Invalid device sector";
3343		goto bad;
3344	}
3345	cc->start = tmpll;
3346
3347	if (bdev_is_zoned(cc->dev->bdev)) {
3348		/*
3349		 * For zoned block devices, we need to preserve the issuer write
3350		 * ordering. To do so, disable write workqueues and force inline
3351		 * encryption completion.
3352		 */
3353		set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3354		set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags);
3355
3356		/*
3357		 * All zone append writes to a zone of a zoned block device will
3358		 * have the same BIO sector, the start of the zone. When the
3359		 * cypher IV mode uses sector values, all data targeting a
3360		 * zone will be encrypted using the first sector numbers of the
3361		 * zone. This will not result in write errors but will
3362		 * cause most reads to fail as reads will use the sector values
3363		 * for the actual data locations, resulting in IV mismatch.
3364		 * To avoid this problem, ask DM core to emulate zone append
3365		 * operations with regular writes.
3366		 */
3367		DMDEBUG("Zone append operations will be emulated");
3368		ti->emulate_zone_append = true;
3369	}
3370
3371	if (crypt_integrity_aead(cc) || cc->integrity_iv_size) {
3372		ret = crypt_integrity_ctr(cc, ti);
3373		if (ret)
3374			goto bad;
3375
3376		cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size;
3377		if (!cc->tag_pool_max_sectors)
3378			cc->tag_pool_max_sectors = 1;
3379
3380		ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS,
3381			cc->tag_pool_max_sectors * cc->tuple_size);
3382		if (ret) {
3383			ti->error = "Cannot allocate integrity tags mempool";
3384			goto bad;
3385		}
3386
3387		cc->tag_pool_max_sectors <<= cc->sector_shift;
3388	}
3389
3390	wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL);
3391	if (wq_id < 0) {
3392		ti->error = "Couldn't get workqueue id";
3393		ret = wq_id;
3394		goto bad;
3395	}
3396	cc->workqueue_id = wq_id;
3397
3398	ret = -ENOMEM;
3399	common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS;
3400	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3401		common_wq_flags |= WQ_HIGHPRI;
3402
3403	cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", common_wq_flags, 1, devname, wq_id);
3404	if (!cc->io_queue) {
3405		ti->error = "Couldn't create kcryptd io queue";
3406		goto bad;
3407	}
3408
3409	if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) {
3410		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3411						  common_wq_flags | WQ_CPU_INTENSIVE,
3412						  1, devname, wq_id);
3413	} else {
3414		/*
3415		 * While crypt_queue is certainly CPU intensive, the use of
3416		 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND.
3417		 */
3418		cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d",
3419						  common_wq_flags | WQ_UNBOUND,
3420						  num_online_cpus(), devname, wq_id);
3421	}
3422	if (!cc->crypt_queue) {
3423		ti->error = "Couldn't create kcryptd queue";
3424		goto bad;
3425	}
3426
3427	spin_lock_init(&cc->write_thread_lock);
3428	cc->write_tree = RB_ROOT;
3429
3430	cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname);
3431	if (IS_ERR(cc->write_thread)) {
3432		ret = PTR_ERR(cc->write_thread);
3433		cc->write_thread = NULL;
3434		ti->error = "Couldn't spawn write thread";
3435		goto bad;
3436	}
3437	if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3438		set_user_nice(cc->write_thread, MIN_NICE);
3439
3440	ti->num_flush_bios = 1;
3441	ti->limit_swap_bios = true;
3442	ti->accounts_remapped_io = true;
3443
3444	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
3445	return 0;
3446
3447bad:
3448	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
3449	crypt_dtr(ti);
3450	return ret;
3451}
3452
3453static int crypt_map(struct dm_target *ti, struct bio *bio)
3454{
3455	struct dm_crypt_io *io;
3456	struct crypt_config *cc = ti->private;
3457	unsigned max_sectors;
3458	bool no_split;
3459
3460	/*
3461	 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues.
3462	 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight
3463	 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters
3464	 */
3465	if (unlikely(bio->bi_opf & REQ_PREFLUSH ||
3466	    bio_op(bio) == REQ_OP_DISCARD)) {
3467		bio_set_dev(bio, cc->dev->bdev);
3468		if (bio_sectors(bio))
3469			bio->bi_iter.bi_sector = cc->start +
3470				dm_target_offset(ti, bio->bi_iter.bi_sector);
3471		return DM_MAPIO_REMAPPED;
3472	}
3473
3474	/*
3475	 * Check if bio is too large, split as needed.
3476	 *
3477	 * For zoned devices, splitting write operations creates the
3478	 * risk of deadlocking queue freeze operations with zone write
3479	 * plugging BIO work when the reminder of a split BIO is
3480	 * issued. So always allow the entire BIO to proceed.
3481	 */
3482	no_split = (ti->emulate_zone_append && op_is_write(bio_op(bio))) ||
3483		   (bio->bi_opf & REQ_ATOMIC);
3484	max_sectors = get_max_request_sectors(ti, bio, no_split);
3485	if (unlikely(bio_sectors(bio) > max_sectors)) {
3486		if (unlikely(no_split))
3487			return DM_MAPIO_KILL;
3488		dm_accept_partial_bio(bio, max_sectors);
3489	}
3490
3491	/*
3492	 * Ensure that bio is a multiple of internal sector encryption size
3493	 * and is aligned to this size as defined in IO hints.
3494	 */
3495	if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0))
3496		return DM_MAPIO_KILL;
3497
3498	if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1)))
3499		return DM_MAPIO_KILL;
3500
3501	io = dm_per_bio_data(bio, cc->per_bio_data_size);
3502	crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector));
3503
3504	if (cc->tuple_size) {
3505		unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift);
3506
3507		if (unlikely(tag_len > KMALLOC_MAX_SIZE))
3508			io->integrity_metadata = NULL;
3509		else
3510			io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
3511
3512		if (unlikely(!io->integrity_metadata)) {
3513			if (bio_sectors(bio) > cc->tag_pool_max_sectors)
3514				dm_accept_partial_bio(bio, cc->tag_pool_max_sectors);
3515			io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO);
3516			io->integrity_metadata_from_pool = true;
3517		}
3518	}
3519
3520	if (crypt_integrity_aead(cc))
3521		io->ctx.r.req_aead = (struct aead_request *)(io + 1);
3522	else
3523		io->ctx.r.req = (struct skcipher_request *)(io + 1);
3524
3525	if (bio_data_dir(io->base_bio) == READ) {
3526		if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP))
3527			kcryptd_queue_read(io);
3528	} else
3529		kcryptd_queue_crypt(io);
3530
3531	return DM_MAPIO_SUBMITTED;
3532}
3533
3534static char hex2asc(unsigned char c)
3535{
3536	return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27);
3537}
3538
3539static void crypt_status(struct dm_target *ti, status_type_t type,
3540			 unsigned int status_flags, char *result, unsigned int maxlen)
3541{
3542	struct crypt_config *cc = ti->private;
3543	unsigned int i, sz = 0;
3544	int num_feature_args = 0;
3545
3546	switch (type) {
3547	case STATUSTYPE_INFO:
3548		result[0] = '\0';
3549		break;
3550
3551	case STATUSTYPE_TABLE:
3552		DMEMIT("%s ", cc->cipher_string);
3553
3554		if (cc->key_size > 0) {
3555			if (cc->key_string)
3556				DMEMIT(":%u:%s", cc->key_size, cc->key_string);
3557			else {
3558				for (i = 0; i < cc->key_size; i++) {
3559					DMEMIT("%c%c", hex2asc(cc->key[i] >> 4),
3560					       hex2asc(cc->key[i] & 0xf));
3561				}
3562			}
3563		} else
3564			DMEMIT("-");
3565
3566		DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset,
3567				cc->dev->name, (unsigned long long)cc->start);
3568
3569		num_feature_args += !!ti->num_discard_bios;
3570		num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags);
3571		num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags);
3572		num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags);
3573		num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags);
3574		num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags);
3575		num_feature_args += !!cc->used_tag_size;
3576		num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT);
3577		num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags);
3578		num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags);
3579		if (num_feature_args) {
3580			DMEMIT(" %d", num_feature_args);
3581			if (ti->num_discard_bios)
3582				DMEMIT(" allow_discards");
3583			if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags))
3584				DMEMIT(" same_cpu_crypt");
3585			if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags))
3586				DMEMIT(" high_priority");
3587			if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags))
3588				DMEMIT(" submit_from_crypt_cpus");
3589			if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags))
3590				DMEMIT(" no_read_workqueue");
3591			if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))
3592				DMEMIT(" no_write_workqueue");
3593			if (cc->used_tag_size)
3594				DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth);
3595			if (cc->sector_size != (1 << SECTOR_SHIFT))
3596				DMEMIT(" sector_size:%d", cc->sector_size);
3597			if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags))
3598				DMEMIT(" iv_large_sectors");
3599			if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags))
3600				DMEMIT(" integrity_key_size:%u", cc->key_mac_size);
3601		}
3602		break;
3603
3604	case STATUSTYPE_IMA:
3605		DMEMIT_TARGET_NAME_VERSION(ti->type);
3606		DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n');
3607		DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n');
3608		DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n');
3609		DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ?
3610		       'y' : 'n');
3611		DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ?
3612		       'y' : 'n');
3613		DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ?
3614		       'y' : 'n');
3615		DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ?
3616		       'y' : 'n');
3617
3618		if (cc->used_tag_size)
3619			DMEMIT(",integrity_tag_size=%u,cipher_auth=%s",
3620			       cc->used_tag_size, cc->cipher_auth);
3621		if (cc->sector_size != (1 << SECTOR_SHIFT))
3622			DMEMIT(",sector_size=%d", cc->sector_size);
3623		if (cc->cipher_string)
3624			DMEMIT(",cipher_string=%s", cc->cipher_string);
3625
3626		DMEMIT(",key_size=%u", cc->key_size);
3627		DMEMIT(",key_parts=%u", cc->key_parts);
3628		DMEMIT(",key_extra_size=%u", cc->key_extra_size);
3629		DMEMIT(",key_mac_size=%u", cc->key_mac_size);
3630		DMEMIT(";");
3631		break;
3632	}
3633}
3634
3635static void crypt_postsuspend(struct dm_target *ti)
3636{
3637	struct crypt_config *cc = ti->private;
3638
3639	set_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3640}
3641
3642static int crypt_preresume(struct dm_target *ti)
3643{
3644	struct crypt_config *cc = ti->private;
3645
3646	if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) {
3647		DMERR("aborting resume - crypt key is not set.");
3648		return -EAGAIN;
3649	}
3650
3651	return 0;
3652}
3653
3654static void crypt_resume(struct dm_target *ti)
3655{
3656	struct crypt_config *cc = ti->private;
3657
3658	clear_bit(DM_CRYPT_SUSPENDED, &cc->flags);
3659}
3660
3661/* Message interface
3662 *	key set <key>
3663 *	key wipe
3664 */
3665static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv,
3666			 char *result, unsigned int maxlen)
3667{
3668	struct crypt_config *cc = ti->private;
3669	int key_size, ret = -EINVAL;
3670
3671	if (argc < 2)
3672		goto error;
3673
3674	if (!strcasecmp(argv[0], "key")) {
3675		if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) {
3676			DMWARN("not suspended during key manipulation.");
3677			return -EINVAL;
3678		}
3679		if (argc == 3 && !strcasecmp(argv[1], "set")) {
3680			/* The key size may not be changed. */
3681			key_size = get_key_size(&argv[2]);
3682			if (key_size < 0 || cc->key_size != key_size) {
3683				memset(argv[2], '0', strlen(argv[2]));
3684				return -EINVAL;
3685			}
3686
3687			ret = crypt_set_key(cc, argv[2]);
3688			if (ret)
3689				return ret;
3690			if (cc->iv_gen_ops && cc->iv_gen_ops->init)
3691				ret = cc->iv_gen_ops->init(cc);
3692			/* wipe the kernel key payload copy */
3693			if (cc->key_string)
3694				memset(cc->key, 0, cc->key_size * sizeof(u8));
3695			return ret;
3696		}
3697		if (argc == 2 && !strcasecmp(argv[1], "wipe"))
3698			return crypt_wipe_key(cc);
3699	}
3700
3701error:
3702	DMWARN("unrecognised message received.");
3703	return -EINVAL;
3704}
3705
3706static int crypt_iterate_devices(struct dm_target *ti,
3707				 iterate_devices_callout_fn fn, void *data)
3708{
3709	struct crypt_config *cc = ti->private;
3710
3711	return fn(ti, cc->dev, cc->start, ti->len, data);
3712}
3713
3714static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits)
3715{
3716	struct crypt_config *cc = ti->private;
3717
3718	limits->logical_block_size =
3719		max_t(unsigned int, limits->logical_block_size, cc->sector_size);
3720	limits->physical_block_size =
3721		max_t(unsigned int, limits->physical_block_size, cc->sector_size);
3722	limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size);
3723	limits->dma_alignment = limits->logical_block_size - 1;
3724
3725	/*
3726	 * For zoned dm-crypt targets, there will be no internal splitting of
3727	 * write BIOs to avoid exceeding BIO_MAX_VECS vectors per BIO. But
3728	 * without respecting this limit, crypt_alloc_buffer() will trigger a
3729	 * BUG(). Avoid this by forcing DM core to split write BIOs to this
3730	 * limit.
3731	 */
3732	if (ti->emulate_zone_append)
3733		limits->max_hw_sectors = min(limits->max_hw_sectors,
3734					     BIO_MAX_VECS << PAGE_SECTORS_SHIFT);
3735
3736	limits->atomic_write_hw_unit_max = min(limits->atomic_write_hw_unit_max,
3737					       BIO_MAX_VECS << PAGE_SHIFT);
3738	limits->atomic_write_hw_max = min(limits->atomic_write_hw_max,
3739					  BIO_MAX_VECS << PAGE_SHIFT);
3740}
3741
3742static struct target_type crypt_target = {
3743	.name   = "crypt",
3744	.version = {1, 29, 0},
3745	.module = THIS_MODULE,
3746	.ctr    = crypt_ctr,
3747	.dtr    = crypt_dtr,
3748	.features = DM_TARGET_ZONED_HM | DM_TARGET_ATOMIC_WRITES,
3749	.report_zones = crypt_report_zones,
3750	.map    = crypt_map,
3751	.status = crypt_status,
3752	.postsuspend = crypt_postsuspend,
3753	.preresume = crypt_preresume,
3754	.resume = crypt_resume,
3755	.message = crypt_message,
3756	.iterate_devices = crypt_iterate_devices,
3757	.io_hints = crypt_io_hints,
3758};
3759module_dm(crypt);
3760
3761MODULE_AUTHOR("Jana Saout <jana@saout.de>");
3762MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption");
3763MODULE_LICENSE("GPL");
~/ Read-only file preview.