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