1 /* 2 * Copyright (C) 2003 Jana Saout <jana@saout.de> 3 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org> 4 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved. 5 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com> 6 * 7 * This file is released under the GPL. 8 */ 9 10 #include <linux/completion.h> 11 #include <linux/err.h> 12 #include <linux/module.h> 13 #include <linux/init.h> 14 #include <linux/kernel.h> 15 #include <linux/key.h> 16 #include <linux/bio.h> 17 #include <linux/blkdev.h> 18 #include <linux/blk-integrity.h> 19 #include <linux/mempool.h> 20 #include <linux/slab.h> 21 #include <linux/crypto.h> 22 #include <linux/workqueue.h> 23 #include <linux/kthread.h> 24 #include <linux/backing-dev.h> 25 #include <linux/atomic.h> 26 #include <linux/scatterlist.h> 27 #include <linux/rbtree.h> 28 #include <linux/ctype.h> 29 #include <asm/page.h> 30 #include <asm/unaligned.h> 31 #include <crypto/hash.h> 32 #include <crypto/md5.h> 33 #include <crypto/algapi.h> 34 #include <crypto/skcipher.h> 35 #include <crypto/aead.h> 36 #include <crypto/authenc.h> 37 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */ 38 #include <linux/key-type.h> 39 #include <keys/user-type.h> 40 #include <keys/encrypted-type.h> 41 #include <keys/trusted-type.h> 42 43 #include <linux/device-mapper.h> 44 45 #include "dm-audit.h" 46 47 #define DM_MSG_PREFIX "crypt" 48 49 /* 50 * context holding the current state of a multi-part conversion 51 */ 52 struct convert_context { 53 struct completion restart; 54 struct bio *bio_in; 55 struct bio *bio_out; 56 struct bvec_iter iter_in; 57 struct bvec_iter iter_out; 58 u64 cc_sector; 59 atomic_t cc_pending; 60 union { 61 struct skcipher_request *req; 62 struct aead_request *req_aead; 63 } r; 64 65 }; 66 67 /* 68 * per bio private data 69 */ 70 struct dm_crypt_io { 71 struct crypt_config *cc; 72 struct bio *base_bio; 73 u8 *integrity_metadata; 74 bool integrity_metadata_from_pool; 75 struct work_struct work; 76 struct tasklet_struct tasklet; 77 78 struct convert_context ctx; 79 80 atomic_t io_pending; 81 blk_status_t error; 82 sector_t sector; 83 84 struct rb_node rb_node; 85 } CRYPTO_MINALIGN_ATTR; 86 87 struct dm_crypt_request { 88 struct convert_context *ctx; 89 struct scatterlist sg_in[4]; 90 struct scatterlist sg_out[4]; 91 u64 iv_sector; 92 }; 93 94 struct crypt_config; 95 96 struct crypt_iv_operations { 97 int (*ctr)(struct crypt_config *cc, struct dm_target *ti, 98 const char *opts); 99 void (*dtr)(struct crypt_config *cc); 100 int (*init)(struct crypt_config *cc); 101 int (*wipe)(struct crypt_config *cc); 102 int (*generator)(struct crypt_config *cc, u8 *iv, 103 struct dm_crypt_request *dmreq); 104 int (*post)(struct crypt_config *cc, u8 *iv, 105 struct dm_crypt_request *dmreq); 106 }; 107 108 struct iv_benbi_private { 109 int shift; 110 }; 111 112 #define LMK_SEED_SIZE 64 /* hash + 0 */ 113 struct iv_lmk_private { 114 struct crypto_shash *hash_tfm; 115 u8 *seed; 116 }; 117 118 #define TCW_WHITENING_SIZE 16 119 struct iv_tcw_private { 120 struct crypto_shash *crc32_tfm; 121 u8 *iv_seed; 122 u8 *whitening; 123 }; 124 125 #define ELEPHANT_MAX_KEY_SIZE 32 126 struct iv_elephant_private { 127 struct crypto_skcipher *tfm; 128 }; 129 130 /* 131 * Crypt: maps a linear range of a block device 132 * and encrypts / decrypts at the same time. 133 */ 134 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID, 135 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD, 136 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE, 137 DM_CRYPT_WRITE_INLINE }; 138 139 enum cipher_flags { 140 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */ 141 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */ 142 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */ 143 }; 144 145 /* 146 * The fields in here must be read only after initialization. 147 */ 148 struct crypt_config { 149 struct dm_dev *dev; 150 sector_t start; 151 152 struct percpu_counter n_allocated_pages; 153 154 struct workqueue_struct *io_queue; 155 struct workqueue_struct *crypt_queue; 156 157 spinlock_t write_thread_lock; 158 struct task_struct *write_thread; 159 struct rb_root write_tree; 160 161 char *cipher_string; 162 char *cipher_auth; 163 char *key_string; 164 165 const struct crypt_iv_operations *iv_gen_ops; 166 union { 167 struct iv_benbi_private benbi; 168 struct iv_lmk_private lmk; 169 struct iv_tcw_private tcw; 170 struct iv_elephant_private elephant; 171 } iv_gen_private; 172 u64 iv_offset; 173 unsigned int iv_size; 174 unsigned short int sector_size; 175 unsigned char sector_shift; 176 177 union { 178 struct crypto_skcipher **tfms; 179 struct crypto_aead **tfms_aead; 180 } cipher_tfm; 181 unsigned tfms_count; 182 unsigned long cipher_flags; 183 184 /* 185 * Layout of each crypto request: 186 * 187 * struct skcipher_request 188 * context 189 * padding 190 * struct dm_crypt_request 191 * padding 192 * IV 193 * 194 * The padding is added so that dm_crypt_request and the IV are 195 * correctly aligned. 196 */ 197 unsigned int dmreq_start; 198 199 unsigned int per_bio_data_size; 200 201 unsigned long flags; 202 unsigned int key_size; 203 unsigned int key_parts; /* independent parts in key buffer */ 204 unsigned int key_extra_size; /* additional keys length */ 205 unsigned int key_mac_size; /* MAC key size for authenc(...) */ 206 207 unsigned int integrity_tag_size; 208 unsigned int integrity_iv_size; 209 unsigned int on_disk_tag_size; 210 211 /* 212 * pool for per bio private data, crypto requests, 213 * encryption requeusts/buffer pages and integrity tags 214 */ 215 unsigned tag_pool_max_sectors; 216 mempool_t tag_pool; 217 mempool_t req_pool; 218 mempool_t page_pool; 219 220 struct bio_set bs; 221 struct mutex bio_alloc_lock; 222 223 u8 *authenc_key; /* space for keys in authenc() format (if used) */ 224 u8 key[]; 225 }; 226 227 #define MIN_IOS 64 228 #define MAX_TAG_SIZE 480 229 #define POOL_ENTRY_SIZE 512 230 231 static DEFINE_SPINLOCK(dm_crypt_clients_lock); 232 static unsigned dm_crypt_clients_n = 0; 233 static volatile unsigned long dm_crypt_pages_per_client; 234 #define DM_CRYPT_MEMORY_PERCENT 2 235 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16) 236 237 static void clone_init(struct dm_crypt_io *, struct bio *); 238 static void kcryptd_queue_crypt(struct dm_crypt_io *io); 239 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 240 struct scatterlist *sg); 241 242 static bool crypt_integrity_aead(struct crypt_config *cc); 243 244 /* 245 * Use this to access cipher attributes that are independent of the key. 246 */ 247 static struct crypto_skcipher *any_tfm(struct crypt_config *cc) 248 { 249 return cc->cipher_tfm.tfms[0]; 250 } 251 252 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc) 253 { 254 return cc->cipher_tfm.tfms_aead[0]; 255 } 256 257 /* 258 * Different IV generation algorithms: 259 * 260 * plain: the initial vector is the 32-bit little-endian version of the sector 261 * number, padded with zeros if necessary. 262 * 263 * plain64: the initial vector is the 64-bit little-endian version of the sector 264 * number, padded with zeros if necessary. 265 * 266 * plain64be: the initial vector is the 64-bit big-endian version of the sector 267 * number, padded with zeros if necessary. 268 * 269 * essiv: "encrypted sector|salt initial vector", the sector number is 270 * encrypted with the bulk cipher using a salt as key. The salt 271 * should be derived from the bulk cipher's key via hashing. 272 * 273 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1 274 * (needed for LRW-32-AES and possible other narrow block modes) 275 * 276 * null: the initial vector is always zero. Provides compatibility with 277 * obsolete loop_fish2 devices. Do not use for new devices. 278 * 279 * lmk: Compatible implementation of the block chaining mode used 280 * by the Loop-AES block device encryption system 281 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/ 282 * It operates on full 512 byte sectors and uses CBC 283 * with an IV derived from the sector number, the data and 284 * optionally extra IV seed. 285 * This means that after decryption the first block 286 * of sector must be tweaked according to decrypted data. 287 * Loop-AES can use three encryption schemes: 288 * version 1: is plain aes-cbc mode 289 * version 2: uses 64 multikey scheme with lmk IV generator 290 * version 3: the same as version 2 with additional IV seed 291 * (it uses 65 keys, last key is used as IV seed) 292 * 293 * tcw: Compatible implementation of the block chaining mode used 294 * by the TrueCrypt device encryption system (prior to version 4.1). 295 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat 296 * It operates on full 512 byte sectors and uses CBC 297 * with an IV derived from initial key and the sector number. 298 * In addition, whitening value is applied on every sector, whitening 299 * is calculated from initial key, sector number and mixed using CRC32. 300 * Note that this encryption scheme is vulnerable to watermarking attacks 301 * and should be used for old compatible containers access only. 302 * 303 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode) 304 * The IV is encrypted little-endian byte-offset (with the same key 305 * and cipher as the volume). 306 * 307 * elephant: The extended version of eboiv with additional Elephant diffuser 308 * used with Bitlocker CBC mode. 309 * This mode was used in older Windows systems 310 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf 311 */ 312 313 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, 314 struct dm_crypt_request *dmreq) 315 { 316 memset(iv, 0, cc->iv_size); 317 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff); 318 319 return 0; 320 } 321 322 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv, 323 struct dm_crypt_request *dmreq) 324 { 325 memset(iv, 0, cc->iv_size); 326 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 327 328 return 0; 329 } 330 331 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv, 332 struct dm_crypt_request *dmreq) 333 { 334 memset(iv, 0, cc->iv_size); 335 /* iv_size is at least of size u64; usually it is 16 bytes */ 336 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector); 337 338 return 0; 339 } 340 341 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, 342 struct dm_crypt_request *dmreq) 343 { 344 /* 345 * ESSIV encryption of the IV is now handled by the crypto API, 346 * so just pass the plain sector number here. 347 */ 348 memset(iv, 0, cc->iv_size); 349 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 350 351 return 0; 352 } 353 354 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti, 355 const char *opts) 356 { 357 unsigned bs; 358 int log; 359 360 if (crypt_integrity_aead(cc)) 361 bs = crypto_aead_blocksize(any_tfm_aead(cc)); 362 else 363 bs = crypto_skcipher_blocksize(any_tfm(cc)); 364 log = ilog2(bs); 365 366 /* we need to calculate how far we must shift the sector count 367 * to get the cipher block count, we use this shift in _gen */ 368 369 if (1 << log != bs) { 370 ti->error = "cypher blocksize is not a power of 2"; 371 return -EINVAL; 372 } 373 374 if (log > 9) { 375 ti->error = "cypher blocksize is > 512"; 376 return -EINVAL; 377 } 378 379 cc->iv_gen_private.benbi.shift = 9 - log; 380 381 return 0; 382 } 383 384 static void crypt_iv_benbi_dtr(struct crypt_config *cc) 385 { 386 } 387 388 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv, 389 struct dm_crypt_request *dmreq) 390 { 391 __be64 val; 392 393 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */ 394 395 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1); 396 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64))); 397 398 return 0; 399 } 400 401 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv, 402 struct dm_crypt_request *dmreq) 403 { 404 memset(iv, 0, cc->iv_size); 405 406 return 0; 407 } 408 409 static void crypt_iv_lmk_dtr(struct crypt_config *cc) 410 { 411 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 412 413 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm)) 414 crypto_free_shash(lmk->hash_tfm); 415 lmk->hash_tfm = NULL; 416 417 kfree_sensitive(lmk->seed); 418 lmk->seed = NULL; 419 } 420 421 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti, 422 const char *opts) 423 { 424 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 425 426 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 427 ti->error = "Unsupported sector size for LMK"; 428 return -EINVAL; 429 } 430 431 lmk->hash_tfm = crypto_alloc_shash("md5", 0, 432 CRYPTO_ALG_ALLOCATES_MEMORY); 433 if (IS_ERR(lmk->hash_tfm)) { 434 ti->error = "Error initializing LMK hash"; 435 return PTR_ERR(lmk->hash_tfm); 436 } 437 438 /* No seed in LMK version 2 */ 439 if (cc->key_parts == cc->tfms_count) { 440 lmk->seed = NULL; 441 return 0; 442 } 443 444 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL); 445 if (!lmk->seed) { 446 crypt_iv_lmk_dtr(cc); 447 ti->error = "Error kmallocing seed storage in LMK"; 448 return -ENOMEM; 449 } 450 451 return 0; 452 } 453 454 static int crypt_iv_lmk_init(struct crypt_config *cc) 455 { 456 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 457 int subkey_size = cc->key_size / cc->key_parts; 458 459 /* LMK seed is on the position of LMK_KEYS + 1 key */ 460 if (lmk->seed) 461 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size), 462 crypto_shash_digestsize(lmk->hash_tfm)); 463 464 return 0; 465 } 466 467 static int crypt_iv_lmk_wipe(struct crypt_config *cc) 468 { 469 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 470 471 if (lmk->seed) 472 memset(lmk->seed, 0, LMK_SEED_SIZE); 473 474 return 0; 475 } 476 477 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv, 478 struct dm_crypt_request *dmreq, 479 u8 *data) 480 { 481 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 482 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm); 483 struct md5_state md5state; 484 __le32 buf[4]; 485 int i, r; 486 487 desc->tfm = lmk->hash_tfm; 488 489 r = crypto_shash_init(desc); 490 if (r) 491 return r; 492 493 if (lmk->seed) { 494 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE); 495 if (r) 496 return r; 497 } 498 499 /* Sector is always 512B, block size 16, add data of blocks 1-31 */ 500 r = crypto_shash_update(desc, data + 16, 16 * 31); 501 if (r) 502 return r; 503 504 /* Sector is cropped to 56 bits here */ 505 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF); 506 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000); 507 buf[2] = cpu_to_le32(4024); 508 buf[3] = 0; 509 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf)); 510 if (r) 511 return r; 512 513 /* No MD5 padding here */ 514 r = crypto_shash_export(desc, &md5state); 515 if (r) 516 return r; 517 518 for (i = 0; i < MD5_HASH_WORDS; i++) 519 __cpu_to_le32s(&md5state.hash[i]); 520 memcpy(iv, &md5state.hash, cc->iv_size); 521 522 return 0; 523 } 524 525 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv, 526 struct dm_crypt_request *dmreq) 527 { 528 struct scatterlist *sg; 529 u8 *src; 530 int r = 0; 531 532 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 533 sg = crypt_get_sg_data(cc, dmreq->sg_in); 534 src = kmap_atomic(sg_page(sg)); 535 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset); 536 kunmap_atomic(src); 537 } else 538 memset(iv, 0, cc->iv_size); 539 540 return r; 541 } 542 543 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv, 544 struct dm_crypt_request *dmreq) 545 { 546 struct scatterlist *sg; 547 u8 *dst; 548 int r; 549 550 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) 551 return 0; 552 553 sg = crypt_get_sg_data(cc, dmreq->sg_out); 554 dst = kmap_atomic(sg_page(sg)); 555 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset); 556 557 /* Tweak the first block of plaintext sector */ 558 if (!r) 559 crypto_xor(dst + sg->offset, iv, cc->iv_size); 560 561 kunmap_atomic(dst); 562 return r; 563 } 564 565 static void crypt_iv_tcw_dtr(struct crypt_config *cc) 566 { 567 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 568 569 kfree_sensitive(tcw->iv_seed); 570 tcw->iv_seed = NULL; 571 kfree_sensitive(tcw->whitening); 572 tcw->whitening = NULL; 573 574 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm)) 575 crypto_free_shash(tcw->crc32_tfm); 576 tcw->crc32_tfm = NULL; 577 } 578 579 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti, 580 const char *opts) 581 { 582 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 583 584 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 585 ti->error = "Unsupported sector size for TCW"; 586 return -EINVAL; 587 } 588 589 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) { 590 ti->error = "Wrong key size for TCW"; 591 return -EINVAL; 592 } 593 594 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0, 595 CRYPTO_ALG_ALLOCATES_MEMORY); 596 if (IS_ERR(tcw->crc32_tfm)) { 597 ti->error = "Error initializing CRC32 in TCW"; 598 return PTR_ERR(tcw->crc32_tfm); 599 } 600 601 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL); 602 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL); 603 if (!tcw->iv_seed || !tcw->whitening) { 604 crypt_iv_tcw_dtr(cc); 605 ti->error = "Error allocating seed storage in TCW"; 606 return -ENOMEM; 607 } 608 609 return 0; 610 } 611 612 static int crypt_iv_tcw_init(struct crypt_config *cc) 613 { 614 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 615 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE; 616 617 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size); 618 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size], 619 TCW_WHITENING_SIZE); 620 621 return 0; 622 } 623 624 static int crypt_iv_tcw_wipe(struct crypt_config *cc) 625 { 626 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 627 628 memset(tcw->iv_seed, 0, cc->iv_size); 629 memset(tcw->whitening, 0, TCW_WHITENING_SIZE); 630 631 return 0; 632 } 633 634 static int crypt_iv_tcw_whitening(struct crypt_config *cc, 635 struct dm_crypt_request *dmreq, 636 u8 *data) 637 { 638 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 639 __le64 sector = cpu_to_le64(dmreq->iv_sector); 640 u8 buf[TCW_WHITENING_SIZE]; 641 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm); 642 int i, r; 643 644 /* xor whitening with sector number */ 645 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8); 646 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8); 647 648 /* calculate crc32 for every 32bit part and xor it */ 649 desc->tfm = tcw->crc32_tfm; 650 for (i = 0; i < 4; i++) { 651 r = crypto_shash_init(desc); 652 if (r) 653 goto out; 654 r = crypto_shash_update(desc, &buf[i * 4], 4); 655 if (r) 656 goto out; 657 r = crypto_shash_final(desc, &buf[i * 4]); 658 if (r) 659 goto out; 660 } 661 crypto_xor(&buf[0], &buf[12], 4); 662 crypto_xor(&buf[4], &buf[8], 4); 663 664 /* apply whitening (8 bytes) to whole sector */ 665 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++) 666 crypto_xor(data + i * 8, buf, 8); 667 out: 668 memzero_explicit(buf, sizeof(buf)); 669 return r; 670 } 671 672 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv, 673 struct dm_crypt_request *dmreq) 674 { 675 struct scatterlist *sg; 676 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 677 __le64 sector = cpu_to_le64(dmreq->iv_sector); 678 u8 *src; 679 int r = 0; 680 681 /* Remove whitening from ciphertext */ 682 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 683 sg = crypt_get_sg_data(cc, dmreq->sg_in); 684 src = kmap_atomic(sg_page(sg)); 685 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset); 686 kunmap_atomic(src); 687 } 688 689 /* Calculate IV */ 690 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8); 691 if (cc->iv_size > 8) 692 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or, 693 cc->iv_size - 8); 694 695 return r; 696 } 697 698 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv, 699 struct dm_crypt_request *dmreq) 700 { 701 struct scatterlist *sg; 702 u8 *dst; 703 int r; 704 705 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 706 return 0; 707 708 /* Apply whitening on ciphertext */ 709 sg = crypt_get_sg_data(cc, dmreq->sg_out); 710 dst = kmap_atomic(sg_page(sg)); 711 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset); 712 kunmap_atomic(dst); 713 714 return r; 715 } 716 717 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv, 718 struct dm_crypt_request *dmreq) 719 { 720 /* Used only for writes, there must be an additional space to store IV */ 721 get_random_bytes(iv, cc->iv_size); 722 return 0; 723 } 724 725 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti, 726 const char *opts) 727 { 728 if (crypt_integrity_aead(cc)) { 729 ti->error = "AEAD transforms not supported for EBOIV"; 730 return -EINVAL; 731 } 732 733 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) { 734 ti->error = "Block size of EBOIV cipher does " 735 "not match IV size of block cipher"; 736 return -EINVAL; 737 } 738 739 return 0; 740 } 741 742 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv, 743 struct dm_crypt_request *dmreq) 744 { 745 u8 buf[MAX_CIPHER_BLOCKSIZE] __aligned(__alignof__(__le64)); 746 struct skcipher_request *req; 747 struct scatterlist src, dst; 748 DECLARE_CRYPTO_WAIT(wait); 749 int err; 750 751 req = skcipher_request_alloc(any_tfm(cc), GFP_NOIO); 752 if (!req) 753 return -ENOMEM; 754 755 memset(buf, 0, cc->iv_size); 756 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 757 758 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size); 759 sg_init_one(&dst, iv, cc->iv_size); 760 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf); 761 skcipher_request_set_callback(req, 0, crypto_req_done, &wait); 762 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 763 skcipher_request_free(req); 764 765 return err; 766 } 767 768 static void crypt_iv_elephant_dtr(struct crypt_config *cc) 769 { 770 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 771 772 crypto_free_skcipher(elephant->tfm); 773 elephant->tfm = NULL; 774 } 775 776 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti, 777 const char *opts) 778 { 779 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 780 int r; 781 782 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0, 783 CRYPTO_ALG_ALLOCATES_MEMORY); 784 if (IS_ERR(elephant->tfm)) { 785 r = PTR_ERR(elephant->tfm); 786 elephant->tfm = NULL; 787 return r; 788 } 789 790 r = crypt_iv_eboiv_ctr(cc, ti, NULL); 791 if (r) 792 crypt_iv_elephant_dtr(cc); 793 return r; 794 } 795 796 static void diffuser_disk_to_cpu(u32 *d, size_t n) 797 { 798 #ifndef __LITTLE_ENDIAN 799 int i; 800 801 for (i = 0; i < n; i++) 802 d[i] = le32_to_cpu((__le32)d[i]); 803 #endif 804 } 805 806 static void diffuser_cpu_to_disk(__le32 *d, size_t n) 807 { 808 #ifndef __LITTLE_ENDIAN 809 int i; 810 811 for (i = 0; i < n; i++) 812 d[i] = cpu_to_le32((u32)d[i]); 813 #endif 814 } 815 816 static void diffuser_a_decrypt(u32 *d, size_t n) 817 { 818 int i, i1, i2, i3; 819 820 for (i = 0; i < 5; i++) { 821 i1 = 0; 822 i2 = n - 2; 823 i3 = n - 5; 824 825 while (i1 < (n - 1)) { 826 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 827 i1++; i2++; i3++; 828 829 if (i3 >= n) 830 i3 -= n; 831 832 d[i1] += d[i2] ^ d[i3]; 833 i1++; i2++; i3++; 834 835 if (i2 >= n) 836 i2 -= n; 837 838 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 839 i1++; i2++; i3++; 840 841 d[i1] += d[i2] ^ d[i3]; 842 i1++; i2++; i3++; 843 } 844 } 845 } 846 847 static void diffuser_a_encrypt(u32 *d, size_t n) 848 { 849 int i, i1, i2, i3; 850 851 for (i = 0; i < 5; i++) { 852 i1 = n - 1; 853 i2 = n - 2 - 1; 854 i3 = n - 5 - 1; 855 856 while (i1 > 0) { 857 d[i1] -= d[i2] ^ d[i3]; 858 i1--; i2--; i3--; 859 860 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 861 i1--; i2--; i3--; 862 863 if (i2 < 0) 864 i2 += n; 865 866 d[i1] -= d[i2] ^ d[i3]; 867 i1--; i2--; i3--; 868 869 if (i3 < 0) 870 i3 += n; 871 872 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 873 i1--; i2--; i3--; 874 } 875 } 876 } 877 878 static void diffuser_b_decrypt(u32 *d, size_t n) 879 { 880 int i, i1, i2, i3; 881 882 for (i = 0; i < 3; i++) { 883 i1 = 0; 884 i2 = 2; 885 i3 = 5; 886 887 while (i1 < (n - 1)) { 888 d[i1] += d[i2] ^ d[i3]; 889 i1++; i2++; i3++; 890 891 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 892 i1++; i2++; i3++; 893 894 if (i2 >= n) 895 i2 -= n; 896 897 d[i1] += d[i2] ^ d[i3]; 898 i1++; i2++; i3++; 899 900 if (i3 >= n) 901 i3 -= n; 902 903 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 904 i1++; i2++; i3++; 905 } 906 } 907 } 908 909 static void diffuser_b_encrypt(u32 *d, size_t n) 910 { 911 int i, i1, i2, i3; 912 913 for (i = 0; i < 3; i++) { 914 i1 = n - 1; 915 i2 = 2 - 1; 916 i3 = 5 - 1; 917 918 while (i1 > 0) { 919 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 920 i1--; i2--; i3--; 921 922 if (i3 < 0) 923 i3 += n; 924 925 d[i1] -= d[i2] ^ d[i3]; 926 i1--; i2--; i3--; 927 928 if (i2 < 0) 929 i2 += n; 930 931 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 932 i1--; i2--; i3--; 933 934 d[i1] -= d[i2] ^ d[i3]; 935 i1--; i2--; i3--; 936 } 937 } 938 } 939 940 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq) 941 { 942 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 943 u8 *es, *ks, *data, *data2, *data_offset; 944 struct skcipher_request *req; 945 struct scatterlist *sg, *sg2, src, dst; 946 DECLARE_CRYPTO_WAIT(wait); 947 int i, r; 948 949 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO); 950 es = kzalloc(16, GFP_NOIO); /* Key for AES */ 951 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */ 952 953 if (!req || !es || !ks) { 954 r = -ENOMEM; 955 goto out; 956 } 957 958 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 959 960 /* E(Ks, e(s)) */ 961 sg_init_one(&src, es, 16); 962 sg_init_one(&dst, ks, 16); 963 skcipher_request_set_crypt(req, &src, &dst, 16, NULL); 964 skcipher_request_set_callback(req, 0, crypto_req_done, &wait); 965 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 966 if (r) 967 goto out; 968 969 /* E(Ks, e'(s)) */ 970 es[15] = 0x80; 971 sg_init_one(&dst, &ks[16], 16); 972 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 973 if (r) 974 goto out; 975 976 sg = crypt_get_sg_data(cc, dmreq->sg_out); 977 data = kmap_atomic(sg_page(sg)); 978 data_offset = data + sg->offset; 979 980 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */ 981 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 982 sg2 = crypt_get_sg_data(cc, dmreq->sg_in); 983 data2 = kmap_atomic(sg_page(sg2)); 984 memcpy(data_offset, data2 + sg2->offset, cc->sector_size); 985 kunmap_atomic(data2); 986 } 987 988 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 989 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32)); 990 diffuser_b_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32)); 991 diffuser_a_decrypt((u32*)data_offset, cc->sector_size / sizeof(u32)); 992 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32)); 993 } 994 995 for (i = 0; i < (cc->sector_size / 32); i++) 996 crypto_xor(data_offset + i * 32, ks, 32); 997 998 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 999 diffuser_disk_to_cpu((u32*)data_offset, cc->sector_size / sizeof(u32)); 1000 diffuser_a_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32)); 1001 diffuser_b_encrypt((u32*)data_offset, cc->sector_size / sizeof(u32)); 1002 diffuser_cpu_to_disk((__le32*)data_offset, cc->sector_size / sizeof(u32)); 1003 } 1004 1005 kunmap_atomic(data); 1006 out: 1007 kfree_sensitive(ks); 1008 kfree_sensitive(es); 1009 skcipher_request_free(req); 1010 return r; 1011 } 1012 1013 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv, 1014 struct dm_crypt_request *dmreq) 1015 { 1016 int r; 1017 1018 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 1019 r = crypt_iv_elephant(cc, dmreq); 1020 if (r) 1021 return r; 1022 } 1023 1024 return crypt_iv_eboiv_gen(cc, iv, dmreq); 1025 } 1026 1027 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv, 1028 struct dm_crypt_request *dmreq) 1029 { 1030 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 1031 return crypt_iv_elephant(cc, dmreq); 1032 1033 return 0; 1034 } 1035 1036 static int crypt_iv_elephant_init(struct crypt_config *cc) 1037 { 1038 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1039 int key_offset = cc->key_size - cc->key_extra_size; 1040 1041 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size); 1042 } 1043 1044 static int crypt_iv_elephant_wipe(struct crypt_config *cc) 1045 { 1046 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1047 u8 key[ELEPHANT_MAX_KEY_SIZE]; 1048 1049 memset(key, 0, cc->key_extra_size); 1050 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size); 1051 } 1052 1053 static const struct crypt_iv_operations crypt_iv_plain_ops = { 1054 .generator = crypt_iv_plain_gen 1055 }; 1056 1057 static const struct crypt_iv_operations crypt_iv_plain64_ops = { 1058 .generator = crypt_iv_plain64_gen 1059 }; 1060 1061 static const struct crypt_iv_operations crypt_iv_plain64be_ops = { 1062 .generator = crypt_iv_plain64be_gen 1063 }; 1064 1065 static const struct crypt_iv_operations crypt_iv_essiv_ops = { 1066 .generator = crypt_iv_essiv_gen 1067 }; 1068 1069 static const struct crypt_iv_operations crypt_iv_benbi_ops = { 1070 .ctr = crypt_iv_benbi_ctr, 1071 .dtr = crypt_iv_benbi_dtr, 1072 .generator = crypt_iv_benbi_gen 1073 }; 1074 1075 static const struct crypt_iv_operations crypt_iv_null_ops = { 1076 .generator = crypt_iv_null_gen 1077 }; 1078 1079 static const struct crypt_iv_operations crypt_iv_lmk_ops = { 1080 .ctr = crypt_iv_lmk_ctr, 1081 .dtr = crypt_iv_lmk_dtr, 1082 .init = crypt_iv_lmk_init, 1083 .wipe = crypt_iv_lmk_wipe, 1084 .generator = crypt_iv_lmk_gen, 1085 .post = crypt_iv_lmk_post 1086 }; 1087 1088 static const struct crypt_iv_operations crypt_iv_tcw_ops = { 1089 .ctr = crypt_iv_tcw_ctr, 1090 .dtr = crypt_iv_tcw_dtr, 1091 .init = crypt_iv_tcw_init, 1092 .wipe = crypt_iv_tcw_wipe, 1093 .generator = crypt_iv_tcw_gen, 1094 .post = crypt_iv_tcw_post 1095 }; 1096 1097 static const struct crypt_iv_operations crypt_iv_random_ops = { 1098 .generator = crypt_iv_random_gen 1099 }; 1100 1101 static const struct crypt_iv_operations crypt_iv_eboiv_ops = { 1102 .ctr = crypt_iv_eboiv_ctr, 1103 .generator = crypt_iv_eboiv_gen 1104 }; 1105 1106 static const struct crypt_iv_operations crypt_iv_elephant_ops = { 1107 .ctr = crypt_iv_elephant_ctr, 1108 .dtr = crypt_iv_elephant_dtr, 1109 .init = crypt_iv_elephant_init, 1110 .wipe = crypt_iv_elephant_wipe, 1111 .generator = crypt_iv_elephant_gen, 1112 .post = crypt_iv_elephant_post 1113 }; 1114 1115 /* 1116 * Integrity extensions 1117 */ 1118 static bool crypt_integrity_aead(struct crypt_config *cc) 1119 { 1120 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 1121 } 1122 1123 static bool crypt_integrity_hmac(struct crypt_config *cc) 1124 { 1125 return crypt_integrity_aead(cc) && cc->key_mac_size; 1126 } 1127 1128 /* Get sg containing data */ 1129 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 1130 struct scatterlist *sg) 1131 { 1132 if (unlikely(crypt_integrity_aead(cc))) 1133 return &sg[2]; 1134 1135 return sg; 1136 } 1137 1138 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio) 1139 { 1140 struct bio_integrity_payload *bip; 1141 unsigned int tag_len; 1142 int ret; 1143 1144 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size) 1145 return 0; 1146 1147 bip = bio_integrity_alloc(bio, GFP_NOIO, 1); 1148 if (IS_ERR(bip)) 1149 return PTR_ERR(bip); 1150 1151 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift); 1152 1153 bip->bip_iter.bi_size = tag_len; 1154 bip->bip_iter.bi_sector = io->cc->start + io->sector; 1155 1156 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata), 1157 tag_len, offset_in_page(io->integrity_metadata)); 1158 if (unlikely(ret != tag_len)) 1159 return -ENOMEM; 1160 1161 return 0; 1162 } 1163 1164 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti) 1165 { 1166 #ifdef CONFIG_BLK_DEV_INTEGRITY 1167 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk); 1168 struct mapped_device *md = dm_table_get_md(ti->table); 1169 1170 /* From now we require underlying device with our integrity profile */ 1171 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) { 1172 ti->error = "Integrity profile not supported."; 1173 return -EINVAL; 1174 } 1175 1176 if (bi->tag_size != cc->on_disk_tag_size || 1177 bi->tuple_size != cc->on_disk_tag_size) { 1178 ti->error = "Integrity profile tag size mismatch."; 1179 return -EINVAL; 1180 } 1181 if (1 << bi->interval_exp != cc->sector_size) { 1182 ti->error = "Integrity profile sector size mismatch."; 1183 return -EINVAL; 1184 } 1185 1186 if (crypt_integrity_aead(cc)) { 1187 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size; 1188 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md), 1189 cc->integrity_tag_size, cc->integrity_iv_size); 1190 1191 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) { 1192 ti->error = "Integrity AEAD auth tag size is not supported."; 1193 return -EINVAL; 1194 } 1195 } else if (cc->integrity_iv_size) 1196 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md), 1197 cc->integrity_iv_size); 1198 1199 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) { 1200 ti->error = "Not enough space for integrity tag in the profile."; 1201 return -EINVAL; 1202 } 1203 1204 return 0; 1205 #else 1206 ti->error = "Integrity profile not supported."; 1207 return -EINVAL; 1208 #endif 1209 } 1210 1211 static void crypt_convert_init(struct crypt_config *cc, 1212 struct convert_context *ctx, 1213 struct bio *bio_out, struct bio *bio_in, 1214 sector_t sector) 1215 { 1216 ctx->bio_in = bio_in; 1217 ctx->bio_out = bio_out; 1218 if (bio_in) 1219 ctx->iter_in = bio_in->bi_iter; 1220 if (bio_out) 1221 ctx->iter_out = bio_out->bi_iter; 1222 ctx->cc_sector = sector + cc->iv_offset; 1223 init_completion(&ctx->restart); 1224 } 1225 1226 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc, 1227 void *req) 1228 { 1229 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start); 1230 } 1231 1232 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq) 1233 { 1234 return (void *)((char *)dmreq - cc->dmreq_start); 1235 } 1236 1237 static u8 *iv_of_dmreq(struct crypt_config *cc, 1238 struct dm_crypt_request *dmreq) 1239 { 1240 if (crypt_integrity_aead(cc)) 1241 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1242 crypto_aead_alignmask(any_tfm_aead(cc)) + 1); 1243 else 1244 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1245 crypto_skcipher_alignmask(any_tfm(cc)) + 1); 1246 } 1247 1248 static u8 *org_iv_of_dmreq(struct crypt_config *cc, 1249 struct dm_crypt_request *dmreq) 1250 { 1251 return iv_of_dmreq(cc, dmreq) + cc->iv_size; 1252 } 1253 1254 static __le64 *org_sector_of_dmreq(struct crypt_config *cc, 1255 struct dm_crypt_request *dmreq) 1256 { 1257 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size; 1258 return (__le64 *) ptr; 1259 } 1260 1261 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc, 1262 struct dm_crypt_request *dmreq) 1263 { 1264 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + 1265 cc->iv_size + sizeof(uint64_t); 1266 return (unsigned int*)ptr; 1267 } 1268 1269 static void *tag_from_dmreq(struct crypt_config *cc, 1270 struct dm_crypt_request *dmreq) 1271 { 1272 struct convert_context *ctx = dmreq->ctx; 1273 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 1274 1275 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) * 1276 cc->on_disk_tag_size]; 1277 } 1278 1279 static void *iv_tag_from_dmreq(struct crypt_config *cc, 1280 struct dm_crypt_request *dmreq) 1281 { 1282 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size; 1283 } 1284 1285 static int crypt_convert_block_aead(struct crypt_config *cc, 1286 struct convert_context *ctx, 1287 struct aead_request *req, 1288 unsigned int tag_offset) 1289 { 1290 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1291 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1292 struct dm_crypt_request *dmreq; 1293 u8 *iv, *org_iv, *tag_iv, *tag; 1294 __le64 *sector; 1295 int r = 0; 1296 1297 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size); 1298 1299 /* Reject unexpected unaligned bio. */ 1300 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1301 return -EIO; 1302 1303 dmreq = dmreq_of_req(cc, req); 1304 dmreq->iv_sector = ctx->cc_sector; 1305 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1306 dmreq->iv_sector >>= cc->sector_shift; 1307 dmreq->ctx = ctx; 1308 1309 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1310 1311 sector = org_sector_of_dmreq(cc, dmreq); 1312 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1313 1314 iv = iv_of_dmreq(cc, dmreq); 1315 org_iv = org_iv_of_dmreq(cc, dmreq); 1316 tag = tag_from_dmreq(cc, dmreq); 1317 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1318 1319 /* AEAD request: 1320 * |----- AAD -------|------ DATA -------|-- AUTH TAG --| 1321 * | (authenticated) | (auth+encryption) | | 1322 * | sector_LE | IV | sector in/out | tag in/out | 1323 */ 1324 sg_init_table(dmreq->sg_in, 4); 1325 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t)); 1326 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size); 1327 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1328 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size); 1329 1330 sg_init_table(dmreq->sg_out, 4); 1331 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t)); 1332 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size); 1333 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1334 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size); 1335 1336 if (cc->iv_gen_ops) { 1337 /* For READs use IV stored in integrity metadata */ 1338 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1339 memcpy(org_iv, tag_iv, cc->iv_size); 1340 } else { 1341 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1342 if (r < 0) 1343 return r; 1344 /* Store generated IV in integrity metadata */ 1345 if (cc->integrity_iv_size) 1346 memcpy(tag_iv, org_iv, cc->iv_size); 1347 } 1348 /* Working copy of IV, to be modified in crypto API */ 1349 memcpy(iv, org_iv, cc->iv_size); 1350 } 1351 1352 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size); 1353 if (bio_data_dir(ctx->bio_in) == WRITE) { 1354 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1355 cc->sector_size, iv); 1356 r = crypto_aead_encrypt(req); 1357 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size) 1358 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0, 1359 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size)); 1360 } else { 1361 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1362 cc->sector_size + cc->integrity_tag_size, iv); 1363 r = crypto_aead_decrypt(req); 1364 } 1365 1366 if (r == -EBADMSG) { 1367 char b[BDEVNAME_SIZE]; 1368 sector_t s = le64_to_cpu(*sector); 1369 1370 DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", 1371 bio_devname(ctx->bio_in, b), s); 1372 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead", 1373 ctx->bio_in, s, 0); 1374 } 1375 1376 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1377 r = cc->iv_gen_ops->post(cc, org_iv, dmreq); 1378 1379 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1380 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1381 1382 return r; 1383 } 1384 1385 static int crypt_convert_block_skcipher(struct crypt_config *cc, 1386 struct convert_context *ctx, 1387 struct skcipher_request *req, 1388 unsigned int tag_offset) 1389 { 1390 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1391 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1392 struct scatterlist *sg_in, *sg_out; 1393 struct dm_crypt_request *dmreq; 1394 u8 *iv, *org_iv, *tag_iv; 1395 __le64 *sector; 1396 int r = 0; 1397 1398 /* Reject unexpected unaligned bio. */ 1399 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1400 return -EIO; 1401 1402 dmreq = dmreq_of_req(cc, req); 1403 dmreq->iv_sector = ctx->cc_sector; 1404 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1405 dmreq->iv_sector >>= cc->sector_shift; 1406 dmreq->ctx = ctx; 1407 1408 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1409 1410 iv = iv_of_dmreq(cc, dmreq); 1411 org_iv = org_iv_of_dmreq(cc, dmreq); 1412 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1413 1414 sector = org_sector_of_dmreq(cc, dmreq); 1415 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1416 1417 /* For skcipher we use only the first sg item */ 1418 sg_in = &dmreq->sg_in[0]; 1419 sg_out = &dmreq->sg_out[0]; 1420 1421 sg_init_table(sg_in, 1); 1422 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1423 1424 sg_init_table(sg_out, 1); 1425 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1426 1427 if (cc->iv_gen_ops) { 1428 /* For READs use IV stored in integrity metadata */ 1429 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1430 memcpy(org_iv, tag_iv, cc->integrity_iv_size); 1431 } else { 1432 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1433 if (r < 0) 1434 return r; 1435 /* Data can be already preprocessed in generator */ 1436 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags)) 1437 sg_in = sg_out; 1438 /* Store generated IV in integrity metadata */ 1439 if (cc->integrity_iv_size) 1440 memcpy(tag_iv, org_iv, cc->integrity_iv_size); 1441 } 1442 /* Working copy of IV, to be modified in crypto API */ 1443 memcpy(iv, org_iv, cc->iv_size); 1444 } 1445 1446 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv); 1447 1448 if (bio_data_dir(ctx->bio_in) == WRITE) 1449 r = crypto_skcipher_encrypt(req); 1450 else 1451 r = crypto_skcipher_decrypt(req); 1452 1453 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1454 r = cc->iv_gen_ops->post(cc, org_iv, dmreq); 1455 1456 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1457 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1458 1459 return r; 1460 } 1461 1462 static void kcryptd_async_done(struct crypto_async_request *async_req, 1463 int error); 1464 1465 static int crypt_alloc_req_skcipher(struct crypt_config *cc, 1466 struct convert_context *ctx) 1467 { 1468 unsigned key_index = ctx->cc_sector & (cc->tfms_count - 1); 1469 1470 if (!ctx->r.req) { 1471 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1472 if (!ctx->r.req) 1473 return -ENOMEM; 1474 } 1475 1476 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]); 1477 1478 /* 1479 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1480 * requests if driver request queue is full. 1481 */ 1482 skcipher_request_set_callback(ctx->r.req, 1483 CRYPTO_TFM_REQ_MAY_BACKLOG, 1484 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req)); 1485 1486 return 0; 1487 } 1488 1489 static int crypt_alloc_req_aead(struct crypt_config *cc, 1490 struct convert_context *ctx) 1491 { 1492 if (!ctx->r.req_aead) { 1493 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1494 if (!ctx->r.req_aead) 1495 return -ENOMEM; 1496 } 1497 1498 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]); 1499 1500 /* 1501 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1502 * requests if driver request queue is full. 1503 */ 1504 aead_request_set_callback(ctx->r.req_aead, 1505 CRYPTO_TFM_REQ_MAY_BACKLOG, 1506 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead)); 1507 1508 return 0; 1509 } 1510 1511 static int crypt_alloc_req(struct crypt_config *cc, 1512 struct convert_context *ctx) 1513 { 1514 if (crypt_integrity_aead(cc)) 1515 return crypt_alloc_req_aead(cc, ctx); 1516 else 1517 return crypt_alloc_req_skcipher(cc, ctx); 1518 } 1519 1520 static void crypt_free_req_skcipher(struct crypt_config *cc, 1521 struct skcipher_request *req, struct bio *base_bio) 1522 { 1523 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1524 1525 if ((struct skcipher_request *)(io + 1) != req) 1526 mempool_free(req, &cc->req_pool); 1527 } 1528 1529 static void crypt_free_req_aead(struct crypt_config *cc, 1530 struct aead_request *req, struct bio *base_bio) 1531 { 1532 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1533 1534 if ((struct aead_request *)(io + 1) != req) 1535 mempool_free(req, &cc->req_pool); 1536 } 1537 1538 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio) 1539 { 1540 if (crypt_integrity_aead(cc)) 1541 crypt_free_req_aead(cc, req, base_bio); 1542 else 1543 crypt_free_req_skcipher(cc, req, base_bio); 1544 } 1545 1546 /* 1547 * Encrypt / decrypt data from one bio to another one (can be the same one) 1548 */ 1549 static blk_status_t crypt_convert(struct crypt_config *cc, 1550 struct convert_context *ctx, bool atomic, bool reset_pending) 1551 { 1552 unsigned int tag_offset = 0; 1553 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT; 1554 int r; 1555 1556 /* 1557 * if reset_pending is set we are dealing with the bio for the first time, 1558 * else we're continuing to work on the previous bio, so don't mess with 1559 * the cc_pending counter 1560 */ 1561 if (reset_pending) 1562 atomic_set(&ctx->cc_pending, 1); 1563 1564 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) { 1565 1566 r = crypt_alloc_req(cc, ctx); 1567 if (r) { 1568 complete(&ctx->restart); 1569 return BLK_STS_DEV_RESOURCE; 1570 } 1571 1572 atomic_inc(&ctx->cc_pending); 1573 1574 if (crypt_integrity_aead(cc)) 1575 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset); 1576 else 1577 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset); 1578 1579 switch (r) { 1580 /* 1581 * The request was queued by a crypto driver 1582 * but the driver request queue is full, let's wait. 1583 */ 1584 case -EBUSY: 1585 if (in_interrupt()) { 1586 if (try_wait_for_completion(&ctx->restart)) { 1587 /* 1588 * we don't have to block to wait for completion, 1589 * so proceed 1590 */ 1591 } else { 1592 /* 1593 * we can't wait for completion without blocking 1594 * exit and continue processing in a workqueue 1595 */ 1596 ctx->r.req = NULL; 1597 ctx->cc_sector += sector_step; 1598 tag_offset++; 1599 return BLK_STS_DEV_RESOURCE; 1600 } 1601 } else { 1602 wait_for_completion(&ctx->restart); 1603 } 1604 reinit_completion(&ctx->restart); 1605 fallthrough; 1606 /* 1607 * The request is queued and processed asynchronously, 1608 * completion function kcryptd_async_done() will be called. 1609 */ 1610 case -EINPROGRESS: 1611 ctx->r.req = NULL; 1612 ctx->cc_sector += sector_step; 1613 tag_offset++; 1614 continue; 1615 /* 1616 * The request was already processed (synchronously). 1617 */ 1618 case 0: 1619 atomic_dec(&ctx->cc_pending); 1620 ctx->cc_sector += sector_step; 1621 tag_offset++; 1622 if (!atomic) 1623 cond_resched(); 1624 continue; 1625 /* 1626 * There was a data integrity error. 1627 */ 1628 case -EBADMSG: 1629 atomic_dec(&ctx->cc_pending); 1630 return BLK_STS_PROTECTION; 1631 /* 1632 * There was an error while processing the request. 1633 */ 1634 default: 1635 atomic_dec(&ctx->cc_pending); 1636 return BLK_STS_IOERR; 1637 } 1638 } 1639 1640 return 0; 1641 } 1642 1643 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone); 1644 1645 /* 1646 * Generate a new unfragmented bio with the given size 1647 * This should never violate the device limitations (but only because 1648 * max_segment_size is being constrained to PAGE_SIZE). 1649 * 1650 * This function may be called concurrently. If we allocate from the mempool 1651 * concurrently, there is a possibility of deadlock. For example, if we have 1652 * mempool of 256 pages, two processes, each wanting 256, pages allocate from 1653 * the mempool concurrently, it may deadlock in a situation where both processes 1654 * have allocated 128 pages and the mempool is exhausted. 1655 * 1656 * In order to avoid this scenario we allocate the pages under a mutex. 1657 * 1658 * In order to not degrade performance with excessive locking, we try 1659 * non-blocking allocations without a mutex first but on failure we fallback 1660 * to blocking allocations with a mutex. 1661 */ 1662 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned size) 1663 { 1664 struct crypt_config *cc = io->cc; 1665 struct bio *clone; 1666 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 1667 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM; 1668 unsigned i, len, remaining_size; 1669 struct page *page; 1670 1671 retry: 1672 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1673 mutex_lock(&cc->bio_alloc_lock); 1674 1675 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, &cc->bs); 1676 if (!clone) 1677 goto out; 1678 1679 clone_init(io, clone); 1680 1681 remaining_size = size; 1682 1683 for (i = 0; i < nr_iovecs; i++) { 1684 page = mempool_alloc(&cc->page_pool, gfp_mask); 1685 if (!page) { 1686 crypt_free_buffer_pages(cc, clone); 1687 bio_put(clone); 1688 gfp_mask |= __GFP_DIRECT_RECLAIM; 1689 goto retry; 1690 } 1691 1692 len = (remaining_size > PAGE_SIZE) ? PAGE_SIZE : remaining_size; 1693 1694 bio_add_page(clone, page, len, 0); 1695 1696 remaining_size -= len; 1697 } 1698 1699 /* Allocate space for integrity tags */ 1700 if (dm_crypt_integrity_io_alloc(io, clone)) { 1701 crypt_free_buffer_pages(cc, clone); 1702 bio_put(clone); 1703 clone = NULL; 1704 } 1705 out: 1706 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1707 mutex_unlock(&cc->bio_alloc_lock); 1708 1709 return clone; 1710 } 1711 1712 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone) 1713 { 1714 struct bio_vec *bv; 1715 struct bvec_iter_all iter_all; 1716 1717 bio_for_each_segment_all(bv, clone, iter_all) { 1718 BUG_ON(!bv->bv_page); 1719 mempool_free(bv->bv_page, &cc->page_pool); 1720 } 1721 } 1722 1723 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc, 1724 struct bio *bio, sector_t sector) 1725 { 1726 io->cc = cc; 1727 io->base_bio = bio; 1728 io->sector = sector; 1729 io->error = 0; 1730 io->ctx.r.req = NULL; 1731 io->integrity_metadata = NULL; 1732 io->integrity_metadata_from_pool = false; 1733 atomic_set(&io->io_pending, 0); 1734 } 1735 1736 static void crypt_inc_pending(struct dm_crypt_io *io) 1737 { 1738 atomic_inc(&io->io_pending); 1739 } 1740 1741 static void kcryptd_io_bio_endio(struct work_struct *work) 1742 { 1743 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 1744 bio_endio(io->base_bio); 1745 } 1746 1747 /* 1748 * One of the bios was finished. Check for completion of 1749 * the whole request and correctly clean up the buffer. 1750 */ 1751 static void crypt_dec_pending(struct dm_crypt_io *io) 1752 { 1753 struct crypt_config *cc = io->cc; 1754 struct bio *base_bio = io->base_bio; 1755 blk_status_t error = io->error; 1756 1757 if (!atomic_dec_and_test(&io->io_pending)) 1758 return; 1759 1760 if (io->ctx.r.req) 1761 crypt_free_req(cc, io->ctx.r.req, base_bio); 1762 1763 if (unlikely(io->integrity_metadata_from_pool)) 1764 mempool_free(io->integrity_metadata, &io->cc->tag_pool); 1765 else 1766 kfree(io->integrity_metadata); 1767 1768 base_bio->bi_status = error; 1769 1770 /* 1771 * If we are running this function from our tasklet, 1772 * we can't call bio_endio() here, because it will call 1773 * clone_endio() from dm.c, which in turn will 1774 * free the current struct dm_crypt_io structure with 1775 * our tasklet. In this case we need to delay bio_endio() 1776 * execution to after the tasklet is done and dequeued. 1777 */ 1778 if (tasklet_trylock(&io->tasklet)) { 1779 tasklet_unlock(&io->tasklet); 1780 bio_endio(base_bio); 1781 return; 1782 } 1783 1784 INIT_WORK(&io->work, kcryptd_io_bio_endio); 1785 queue_work(cc->io_queue, &io->work); 1786 } 1787 1788 /* 1789 * kcryptd/kcryptd_io: 1790 * 1791 * Needed because it would be very unwise to do decryption in an 1792 * interrupt context. 1793 * 1794 * kcryptd performs the actual encryption or decryption. 1795 * 1796 * kcryptd_io performs the IO submission. 1797 * 1798 * They must be separated as otherwise the final stages could be 1799 * starved by new requests which can block in the first stages due 1800 * to memory allocation. 1801 * 1802 * The work is done per CPU global for all dm-crypt instances. 1803 * They should not depend on each other and do not block. 1804 */ 1805 static void crypt_endio(struct bio *clone) 1806 { 1807 struct dm_crypt_io *io = clone->bi_private; 1808 struct crypt_config *cc = io->cc; 1809 unsigned rw = bio_data_dir(clone); 1810 blk_status_t error; 1811 1812 /* 1813 * free the processed pages 1814 */ 1815 if (rw == WRITE) 1816 crypt_free_buffer_pages(cc, clone); 1817 1818 error = clone->bi_status; 1819 bio_put(clone); 1820 1821 if (rw == READ && !error) { 1822 kcryptd_queue_crypt(io); 1823 return; 1824 } 1825 1826 if (unlikely(error)) 1827 io->error = error; 1828 1829 crypt_dec_pending(io); 1830 } 1831 1832 static void clone_init(struct dm_crypt_io *io, struct bio *clone) 1833 { 1834 struct crypt_config *cc = io->cc; 1835 1836 clone->bi_private = io; 1837 clone->bi_end_io = crypt_endio; 1838 bio_set_dev(clone, cc->dev->bdev); 1839 clone->bi_opf = io->base_bio->bi_opf; 1840 } 1841 1842 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp) 1843 { 1844 struct crypt_config *cc = io->cc; 1845 struct bio *clone; 1846 1847 /* 1848 * We need the original biovec array in order to decrypt 1849 * the whole bio data *afterwards* -- thanks to immutable 1850 * biovecs we don't need to worry about the block layer 1851 * modifying the biovec array; so leverage bio_clone_fast(). 1852 */ 1853 clone = bio_clone_fast(io->base_bio, gfp, &cc->bs); 1854 if (!clone) 1855 return 1; 1856 1857 crypt_inc_pending(io); 1858 1859 clone_init(io, clone); 1860 clone->bi_iter.bi_sector = cc->start + io->sector; 1861 1862 if (dm_crypt_integrity_io_alloc(io, clone)) { 1863 crypt_dec_pending(io); 1864 bio_put(clone); 1865 return 1; 1866 } 1867 1868 submit_bio_noacct(clone); 1869 return 0; 1870 } 1871 1872 static void kcryptd_io_read_work(struct work_struct *work) 1873 { 1874 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 1875 1876 crypt_inc_pending(io); 1877 if (kcryptd_io_read(io, GFP_NOIO)) 1878 io->error = BLK_STS_RESOURCE; 1879 crypt_dec_pending(io); 1880 } 1881 1882 static void kcryptd_queue_read(struct dm_crypt_io *io) 1883 { 1884 struct crypt_config *cc = io->cc; 1885 1886 INIT_WORK(&io->work, kcryptd_io_read_work); 1887 queue_work(cc->io_queue, &io->work); 1888 } 1889 1890 static void kcryptd_io_write(struct dm_crypt_io *io) 1891 { 1892 struct bio *clone = io->ctx.bio_out; 1893 1894 submit_bio_noacct(clone); 1895 } 1896 1897 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node) 1898 1899 static int dmcrypt_write(void *data) 1900 { 1901 struct crypt_config *cc = data; 1902 struct dm_crypt_io *io; 1903 1904 while (1) { 1905 struct rb_root write_tree; 1906 struct blk_plug plug; 1907 1908 spin_lock_irq(&cc->write_thread_lock); 1909 continue_locked: 1910 1911 if (!RB_EMPTY_ROOT(&cc->write_tree)) 1912 goto pop_from_list; 1913 1914 set_current_state(TASK_INTERRUPTIBLE); 1915 1916 spin_unlock_irq(&cc->write_thread_lock); 1917 1918 if (unlikely(kthread_should_stop())) { 1919 set_current_state(TASK_RUNNING); 1920 break; 1921 } 1922 1923 schedule(); 1924 1925 set_current_state(TASK_RUNNING); 1926 spin_lock_irq(&cc->write_thread_lock); 1927 goto continue_locked; 1928 1929 pop_from_list: 1930 write_tree = cc->write_tree; 1931 cc->write_tree = RB_ROOT; 1932 spin_unlock_irq(&cc->write_thread_lock); 1933 1934 BUG_ON(rb_parent(write_tree.rb_node)); 1935 1936 /* 1937 * Note: we cannot walk the tree here with rb_next because 1938 * the structures may be freed when kcryptd_io_write is called. 1939 */ 1940 blk_start_plug(&plug); 1941 do { 1942 io = crypt_io_from_node(rb_first(&write_tree)); 1943 rb_erase(&io->rb_node, &write_tree); 1944 kcryptd_io_write(io); 1945 } while (!RB_EMPTY_ROOT(&write_tree)); 1946 blk_finish_plug(&plug); 1947 } 1948 return 0; 1949 } 1950 1951 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async) 1952 { 1953 struct bio *clone = io->ctx.bio_out; 1954 struct crypt_config *cc = io->cc; 1955 unsigned long flags; 1956 sector_t sector; 1957 struct rb_node **rbp, *parent; 1958 1959 if (unlikely(io->error)) { 1960 crypt_free_buffer_pages(cc, clone); 1961 bio_put(clone); 1962 crypt_dec_pending(io); 1963 return; 1964 } 1965 1966 /* crypt_convert should have filled the clone bio */ 1967 BUG_ON(io->ctx.iter_out.bi_size); 1968 1969 clone->bi_iter.bi_sector = cc->start + io->sector; 1970 1971 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) || 1972 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) { 1973 submit_bio_noacct(clone); 1974 return; 1975 } 1976 1977 spin_lock_irqsave(&cc->write_thread_lock, flags); 1978 if (RB_EMPTY_ROOT(&cc->write_tree)) 1979 wake_up_process(cc->write_thread); 1980 rbp = &cc->write_tree.rb_node; 1981 parent = NULL; 1982 sector = io->sector; 1983 while (*rbp) { 1984 parent = *rbp; 1985 if (sector < crypt_io_from_node(parent)->sector) 1986 rbp = &(*rbp)->rb_left; 1987 else 1988 rbp = &(*rbp)->rb_right; 1989 } 1990 rb_link_node(&io->rb_node, parent, rbp); 1991 rb_insert_color(&io->rb_node, &cc->write_tree); 1992 spin_unlock_irqrestore(&cc->write_thread_lock, flags); 1993 } 1994 1995 static bool kcryptd_crypt_write_inline(struct crypt_config *cc, 1996 struct convert_context *ctx) 1997 1998 { 1999 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags)) 2000 return false; 2001 2002 /* 2003 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering 2004 * constraints so they do not need to be issued inline by 2005 * kcryptd_crypt_write_convert(). 2006 */ 2007 switch (bio_op(ctx->bio_in)) { 2008 case REQ_OP_WRITE: 2009 case REQ_OP_WRITE_SAME: 2010 case REQ_OP_WRITE_ZEROES: 2011 return true; 2012 default: 2013 return false; 2014 } 2015 } 2016 2017 static void kcryptd_crypt_write_continue(struct work_struct *work) 2018 { 2019 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2020 struct crypt_config *cc = io->cc; 2021 struct convert_context *ctx = &io->ctx; 2022 int crypt_finished; 2023 sector_t sector = io->sector; 2024 blk_status_t r; 2025 2026 wait_for_completion(&ctx->restart); 2027 reinit_completion(&ctx->restart); 2028 2029 r = crypt_convert(cc, &io->ctx, true, false); 2030 if (r) 2031 io->error = r; 2032 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2033 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2034 /* Wait for completion signaled by kcryptd_async_done() */ 2035 wait_for_completion(&ctx->restart); 2036 crypt_finished = 1; 2037 } 2038 2039 /* Encryption was already finished, submit io now */ 2040 if (crypt_finished) { 2041 kcryptd_crypt_write_io_submit(io, 0); 2042 io->sector = sector; 2043 } 2044 2045 crypt_dec_pending(io); 2046 } 2047 2048 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io) 2049 { 2050 struct crypt_config *cc = io->cc; 2051 struct convert_context *ctx = &io->ctx; 2052 struct bio *clone; 2053 int crypt_finished; 2054 sector_t sector = io->sector; 2055 blk_status_t r; 2056 2057 /* 2058 * Prevent io from disappearing until this function completes. 2059 */ 2060 crypt_inc_pending(io); 2061 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector); 2062 2063 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size); 2064 if (unlikely(!clone)) { 2065 io->error = BLK_STS_IOERR; 2066 goto dec; 2067 } 2068 2069 io->ctx.bio_out = clone; 2070 io->ctx.iter_out = clone->bi_iter; 2071 2072 sector += bio_sectors(clone); 2073 2074 crypt_inc_pending(io); 2075 r = crypt_convert(cc, ctx, 2076 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true); 2077 /* 2078 * Crypto API backlogged the request, because its queue was full 2079 * and we're in softirq context, so continue from a workqueue 2080 * (TODO: is it actually possible to be in softirq in the write path?) 2081 */ 2082 if (r == BLK_STS_DEV_RESOURCE) { 2083 INIT_WORK(&io->work, kcryptd_crypt_write_continue); 2084 queue_work(cc->crypt_queue, &io->work); 2085 return; 2086 } 2087 if (r) 2088 io->error = r; 2089 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2090 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2091 /* Wait for completion signaled by kcryptd_async_done() */ 2092 wait_for_completion(&ctx->restart); 2093 crypt_finished = 1; 2094 } 2095 2096 /* Encryption was already finished, submit io now */ 2097 if (crypt_finished) { 2098 kcryptd_crypt_write_io_submit(io, 0); 2099 io->sector = sector; 2100 } 2101 2102 dec: 2103 crypt_dec_pending(io); 2104 } 2105 2106 static void kcryptd_crypt_read_done(struct dm_crypt_io *io) 2107 { 2108 crypt_dec_pending(io); 2109 } 2110 2111 static void kcryptd_crypt_read_continue(struct work_struct *work) 2112 { 2113 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2114 struct crypt_config *cc = io->cc; 2115 blk_status_t r; 2116 2117 wait_for_completion(&io->ctx.restart); 2118 reinit_completion(&io->ctx.restart); 2119 2120 r = crypt_convert(cc, &io->ctx, true, false); 2121 if (r) 2122 io->error = r; 2123 2124 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2125 kcryptd_crypt_read_done(io); 2126 2127 crypt_dec_pending(io); 2128 } 2129 2130 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io) 2131 { 2132 struct crypt_config *cc = io->cc; 2133 blk_status_t r; 2134 2135 crypt_inc_pending(io); 2136 2137 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio, 2138 io->sector); 2139 2140 r = crypt_convert(cc, &io->ctx, 2141 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true); 2142 /* 2143 * Crypto API backlogged the request, because its queue was full 2144 * and we're in softirq context, so continue from a workqueue 2145 */ 2146 if (r == BLK_STS_DEV_RESOURCE) { 2147 INIT_WORK(&io->work, kcryptd_crypt_read_continue); 2148 queue_work(cc->crypt_queue, &io->work); 2149 return; 2150 } 2151 if (r) 2152 io->error = r; 2153 2154 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2155 kcryptd_crypt_read_done(io); 2156 2157 crypt_dec_pending(io); 2158 } 2159 2160 static void kcryptd_async_done(struct crypto_async_request *async_req, 2161 int error) 2162 { 2163 struct dm_crypt_request *dmreq = async_req->data; 2164 struct convert_context *ctx = dmreq->ctx; 2165 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 2166 struct crypt_config *cc = io->cc; 2167 2168 /* 2169 * A request from crypto driver backlog is going to be processed now, 2170 * finish the completion and continue in crypt_convert(). 2171 * (Callback will be called for the second time for this request.) 2172 */ 2173 if (error == -EINPROGRESS) { 2174 complete(&ctx->restart); 2175 return; 2176 } 2177 2178 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post) 2179 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq); 2180 2181 if (error == -EBADMSG) { 2182 char b[BDEVNAME_SIZE]; 2183 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)); 2184 2185 DMERR_LIMIT("%s: INTEGRITY AEAD ERROR, sector %llu", 2186 bio_devname(ctx->bio_in, b), 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_crypt_tasklet(unsigned long work) 2226 { 2227 kcryptd_crypt((struct work_struct *)work); 2228 } 2229 2230 static void kcryptd_queue_crypt(struct dm_crypt_io *io) 2231 { 2232 struct crypt_config *cc = io->cc; 2233 2234 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) || 2235 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) { 2236 /* 2237 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context. 2238 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but 2239 * it is being executed with irqs disabled. 2240 */ 2241 if (in_hardirq() || irqs_disabled()) { 2242 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work); 2243 tasklet_schedule(&io->tasklet); 2244 return; 2245 } 2246 2247 kcryptd_crypt(&io->work); 2248 return; 2249 } 2250 2251 INIT_WORK(&io->work, kcryptd_crypt); 2252 queue_work(cc->crypt_queue, &io->work); 2253 } 2254 2255 static void crypt_free_tfms_aead(struct crypt_config *cc) 2256 { 2257 if (!cc->cipher_tfm.tfms_aead) 2258 return; 2259 2260 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2261 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]); 2262 cc->cipher_tfm.tfms_aead[0] = NULL; 2263 } 2264 2265 kfree(cc->cipher_tfm.tfms_aead); 2266 cc->cipher_tfm.tfms_aead = NULL; 2267 } 2268 2269 static void crypt_free_tfms_skcipher(struct crypt_config *cc) 2270 { 2271 unsigned i; 2272 2273 if (!cc->cipher_tfm.tfms) 2274 return; 2275 2276 for (i = 0; i < cc->tfms_count; i++) 2277 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) { 2278 crypto_free_skcipher(cc->cipher_tfm.tfms[i]); 2279 cc->cipher_tfm.tfms[i] = NULL; 2280 } 2281 2282 kfree(cc->cipher_tfm.tfms); 2283 cc->cipher_tfm.tfms = NULL; 2284 } 2285 2286 static void crypt_free_tfms(struct crypt_config *cc) 2287 { 2288 if (crypt_integrity_aead(cc)) 2289 crypt_free_tfms_aead(cc); 2290 else 2291 crypt_free_tfms_skcipher(cc); 2292 } 2293 2294 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode) 2295 { 2296 unsigned i; 2297 int err; 2298 2299 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count, 2300 sizeof(struct crypto_skcipher *), 2301 GFP_KERNEL); 2302 if (!cc->cipher_tfm.tfms) 2303 return -ENOMEM; 2304 2305 for (i = 0; i < cc->tfms_count; i++) { 2306 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 2307 CRYPTO_ALG_ALLOCATES_MEMORY); 2308 if (IS_ERR(cc->cipher_tfm.tfms[i])) { 2309 err = PTR_ERR(cc->cipher_tfm.tfms[i]); 2310 crypt_free_tfms(cc); 2311 return err; 2312 } 2313 } 2314 2315 /* 2316 * dm-crypt performance can vary greatly depending on which crypto 2317 * algorithm implementation is used. Help people debug performance 2318 * problems by logging the ->cra_driver_name. 2319 */ 2320 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2321 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name); 2322 return 0; 2323 } 2324 2325 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode) 2326 { 2327 int err; 2328 2329 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL); 2330 if (!cc->cipher_tfm.tfms) 2331 return -ENOMEM; 2332 2333 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0, 2334 CRYPTO_ALG_ALLOCATES_MEMORY); 2335 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2336 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]); 2337 crypt_free_tfms(cc); 2338 return err; 2339 } 2340 2341 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2342 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name); 2343 return 0; 2344 } 2345 2346 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode) 2347 { 2348 if (crypt_integrity_aead(cc)) 2349 return crypt_alloc_tfms_aead(cc, ciphermode); 2350 else 2351 return crypt_alloc_tfms_skcipher(cc, ciphermode); 2352 } 2353 2354 static unsigned crypt_subkey_size(struct crypt_config *cc) 2355 { 2356 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count); 2357 } 2358 2359 static unsigned crypt_authenckey_size(struct crypt_config *cc) 2360 { 2361 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param)); 2362 } 2363 2364 /* 2365 * If AEAD is composed like authenc(hmac(sha256),xts(aes)), 2366 * the key must be for some reason in special format. 2367 * This funcion converts cc->key to this special format. 2368 */ 2369 static void crypt_copy_authenckey(char *p, const void *key, 2370 unsigned enckeylen, unsigned authkeylen) 2371 { 2372 struct crypto_authenc_key_param *param; 2373 struct rtattr *rta; 2374 2375 rta = (struct rtattr *)p; 2376 param = RTA_DATA(rta); 2377 param->enckeylen = cpu_to_be32(enckeylen); 2378 rta->rta_len = RTA_LENGTH(sizeof(*param)); 2379 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM; 2380 p += RTA_SPACE(sizeof(*param)); 2381 memcpy(p, key + enckeylen, authkeylen); 2382 p += authkeylen; 2383 memcpy(p, key, enckeylen); 2384 } 2385 2386 static int crypt_setkey(struct crypt_config *cc) 2387 { 2388 unsigned subkey_size; 2389 int err = 0, i, r; 2390 2391 /* Ignore extra keys (which are used for IV etc) */ 2392 subkey_size = crypt_subkey_size(cc); 2393 2394 if (crypt_integrity_hmac(cc)) { 2395 if (subkey_size < cc->key_mac_size) 2396 return -EINVAL; 2397 2398 crypt_copy_authenckey(cc->authenc_key, cc->key, 2399 subkey_size - cc->key_mac_size, 2400 cc->key_mac_size); 2401 } 2402 2403 for (i = 0; i < cc->tfms_count; i++) { 2404 if (crypt_integrity_hmac(cc)) 2405 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2406 cc->authenc_key, crypt_authenckey_size(cc)); 2407 else if (crypt_integrity_aead(cc)) 2408 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2409 cc->key + (i * subkey_size), 2410 subkey_size); 2411 else 2412 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i], 2413 cc->key + (i * subkey_size), 2414 subkey_size); 2415 if (r) 2416 err = r; 2417 } 2418 2419 if (crypt_integrity_hmac(cc)) 2420 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc)); 2421 2422 return err; 2423 } 2424 2425 #ifdef CONFIG_KEYS 2426 2427 static bool contains_whitespace(const char *str) 2428 { 2429 while (*str) 2430 if (isspace(*str++)) 2431 return true; 2432 return false; 2433 } 2434 2435 static int set_key_user(struct crypt_config *cc, struct key *key) 2436 { 2437 const struct user_key_payload *ukp; 2438 2439 ukp = user_key_payload_locked(key); 2440 if (!ukp) 2441 return -EKEYREVOKED; 2442 2443 if (cc->key_size != ukp->datalen) 2444 return -EINVAL; 2445 2446 memcpy(cc->key, ukp->data, cc->key_size); 2447 2448 return 0; 2449 } 2450 2451 static int set_key_encrypted(struct crypt_config *cc, struct key *key) 2452 { 2453 const struct encrypted_key_payload *ekp; 2454 2455 ekp = key->payload.data[0]; 2456 if (!ekp) 2457 return -EKEYREVOKED; 2458 2459 if (cc->key_size != ekp->decrypted_datalen) 2460 return -EINVAL; 2461 2462 memcpy(cc->key, ekp->decrypted_data, cc->key_size); 2463 2464 return 0; 2465 } 2466 2467 static int set_key_trusted(struct crypt_config *cc, struct key *key) 2468 { 2469 const struct trusted_key_payload *tkp; 2470 2471 tkp = key->payload.data[0]; 2472 if (!tkp) 2473 return -EKEYREVOKED; 2474 2475 if (cc->key_size != tkp->key_len) 2476 return -EINVAL; 2477 2478 memcpy(cc->key, tkp->key, cc->key_size); 2479 2480 return 0; 2481 } 2482 2483 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2484 { 2485 char *new_key_string, *key_desc; 2486 int ret; 2487 struct key_type *type; 2488 struct key *key; 2489 int (*set_key)(struct crypt_config *cc, struct key *key); 2490 2491 /* 2492 * Reject key_string with whitespace. dm core currently lacks code for 2493 * proper whitespace escaping in arguments on DM_TABLE_STATUS path. 2494 */ 2495 if (contains_whitespace(key_string)) { 2496 DMERR("whitespace chars not allowed in key string"); 2497 return -EINVAL; 2498 } 2499 2500 /* look for next ':' separating key_type from key_description */ 2501 key_desc = strpbrk(key_string, ":"); 2502 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1)) 2503 return -EINVAL; 2504 2505 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) { 2506 type = &key_type_logon; 2507 set_key = set_key_user; 2508 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) { 2509 type = &key_type_user; 2510 set_key = set_key_user; 2511 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) && 2512 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) { 2513 type = &key_type_encrypted; 2514 set_key = set_key_encrypted; 2515 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) && 2516 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) { 2517 type = &key_type_trusted; 2518 set_key = set_key_trusted; 2519 } else { 2520 return -EINVAL; 2521 } 2522 2523 new_key_string = kstrdup(key_string, GFP_KERNEL); 2524 if (!new_key_string) 2525 return -ENOMEM; 2526 2527 key = request_key(type, key_desc + 1, NULL); 2528 if (IS_ERR(key)) { 2529 kfree_sensitive(new_key_string); 2530 return PTR_ERR(key); 2531 } 2532 2533 down_read(&key->sem); 2534 2535 ret = set_key(cc, key); 2536 if (ret < 0) { 2537 up_read(&key->sem); 2538 key_put(key); 2539 kfree_sensitive(new_key_string); 2540 return ret; 2541 } 2542 2543 up_read(&key->sem); 2544 key_put(key); 2545 2546 /* clear the flag since following operations may invalidate previously valid key */ 2547 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2548 2549 ret = crypt_setkey(cc); 2550 2551 if (!ret) { 2552 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2553 kfree_sensitive(cc->key_string); 2554 cc->key_string = new_key_string; 2555 } else 2556 kfree_sensitive(new_key_string); 2557 2558 return ret; 2559 } 2560 2561 static int get_key_size(char **key_string) 2562 { 2563 char *colon, dummy; 2564 int ret; 2565 2566 if (*key_string[0] != ':') 2567 return strlen(*key_string) >> 1; 2568 2569 /* look for next ':' in key string */ 2570 colon = strpbrk(*key_string + 1, ":"); 2571 if (!colon) 2572 return -EINVAL; 2573 2574 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':') 2575 return -EINVAL; 2576 2577 *key_string = colon; 2578 2579 /* remaining key string should be :<logon|user>:<key_desc> */ 2580 2581 return ret; 2582 } 2583 2584 #else 2585 2586 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2587 { 2588 return -EINVAL; 2589 } 2590 2591 static int get_key_size(char **key_string) 2592 { 2593 return (*key_string[0] == ':') ? -EINVAL : strlen(*key_string) >> 1; 2594 } 2595 2596 #endif /* CONFIG_KEYS */ 2597 2598 static int crypt_set_key(struct crypt_config *cc, char *key) 2599 { 2600 int r = -EINVAL; 2601 int key_string_len = strlen(key); 2602 2603 /* Hyphen (which gives a key_size of zero) means there is no key. */ 2604 if (!cc->key_size && strcmp(key, "-")) 2605 goto out; 2606 2607 /* ':' means the key is in kernel keyring, short-circuit normal key processing */ 2608 if (key[0] == ':') { 2609 r = crypt_set_keyring_key(cc, key + 1); 2610 goto out; 2611 } 2612 2613 /* clear the flag since following operations may invalidate previously valid key */ 2614 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2615 2616 /* wipe references to any kernel keyring key */ 2617 kfree_sensitive(cc->key_string); 2618 cc->key_string = NULL; 2619 2620 /* Decode key from its hex representation. */ 2621 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0) 2622 goto out; 2623 2624 r = crypt_setkey(cc); 2625 if (!r) 2626 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2627 2628 out: 2629 /* Hex key string not needed after here, so wipe it. */ 2630 memset(key, '0', key_string_len); 2631 2632 return r; 2633 } 2634 2635 static int crypt_wipe_key(struct crypt_config *cc) 2636 { 2637 int r; 2638 2639 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2640 get_random_bytes(&cc->key, cc->key_size); 2641 2642 /* Wipe IV private keys */ 2643 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) { 2644 r = cc->iv_gen_ops->wipe(cc); 2645 if (r) 2646 return r; 2647 } 2648 2649 kfree_sensitive(cc->key_string); 2650 cc->key_string = NULL; 2651 r = crypt_setkey(cc); 2652 memset(&cc->key, 0, cc->key_size * sizeof(u8)); 2653 2654 return r; 2655 } 2656 2657 static void crypt_calculate_pages_per_client(void) 2658 { 2659 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100; 2660 2661 if (!dm_crypt_clients_n) 2662 return; 2663 2664 pages /= dm_crypt_clients_n; 2665 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT) 2666 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT; 2667 dm_crypt_pages_per_client = pages; 2668 } 2669 2670 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data) 2671 { 2672 struct crypt_config *cc = pool_data; 2673 struct page *page; 2674 2675 /* 2676 * Note, percpu_counter_read_positive() may over (and under) estimate 2677 * the current usage by at most (batch - 1) * num_online_cpus() pages, 2678 * but avoids potential spinlock contention of an exact result. 2679 */ 2680 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) && 2681 likely(gfp_mask & __GFP_NORETRY)) 2682 return NULL; 2683 2684 page = alloc_page(gfp_mask); 2685 if (likely(page != NULL)) 2686 percpu_counter_add(&cc->n_allocated_pages, 1); 2687 2688 return page; 2689 } 2690 2691 static void crypt_page_free(void *page, void *pool_data) 2692 { 2693 struct crypt_config *cc = pool_data; 2694 2695 __free_page(page); 2696 percpu_counter_sub(&cc->n_allocated_pages, 1); 2697 } 2698 2699 static void crypt_dtr(struct dm_target *ti) 2700 { 2701 struct crypt_config *cc = ti->private; 2702 2703 ti->private = NULL; 2704 2705 if (!cc) 2706 return; 2707 2708 if (cc->write_thread) 2709 kthread_stop(cc->write_thread); 2710 2711 if (cc->io_queue) 2712 destroy_workqueue(cc->io_queue); 2713 if (cc->crypt_queue) 2714 destroy_workqueue(cc->crypt_queue); 2715 2716 crypt_free_tfms(cc); 2717 2718 bioset_exit(&cc->bs); 2719 2720 mempool_exit(&cc->page_pool); 2721 mempool_exit(&cc->req_pool); 2722 mempool_exit(&cc->tag_pool); 2723 2724 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0); 2725 percpu_counter_destroy(&cc->n_allocated_pages); 2726 2727 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) 2728 cc->iv_gen_ops->dtr(cc); 2729 2730 if (cc->dev) 2731 dm_put_device(ti, cc->dev); 2732 2733 kfree_sensitive(cc->cipher_string); 2734 kfree_sensitive(cc->key_string); 2735 kfree_sensitive(cc->cipher_auth); 2736 kfree_sensitive(cc->authenc_key); 2737 2738 mutex_destroy(&cc->bio_alloc_lock); 2739 2740 /* Must zero key material before freeing */ 2741 kfree_sensitive(cc); 2742 2743 spin_lock(&dm_crypt_clients_lock); 2744 WARN_ON(!dm_crypt_clients_n); 2745 dm_crypt_clients_n--; 2746 crypt_calculate_pages_per_client(); 2747 spin_unlock(&dm_crypt_clients_lock); 2748 2749 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1); 2750 } 2751 2752 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode) 2753 { 2754 struct crypt_config *cc = ti->private; 2755 2756 if (crypt_integrity_aead(cc)) 2757 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2758 else 2759 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2760 2761 if (cc->iv_size) 2762 /* at least a 64 bit sector number should fit in our buffer */ 2763 cc->iv_size = max(cc->iv_size, 2764 (unsigned int)(sizeof(u64) / sizeof(u8))); 2765 else if (ivmode) { 2766 DMWARN("Selected cipher does not support IVs"); 2767 ivmode = NULL; 2768 } 2769 2770 /* Choose ivmode, see comments at iv code. */ 2771 if (ivmode == NULL) 2772 cc->iv_gen_ops = NULL; 2773 else if (strcmp(ivmode, "plain") == 0) 2774 cc->iv_gen_ops = &crypt_iv_plain_ops; 2775 else if (strcmp(ivmode, "plain64") == 0) 2776 cc->iv_gen_ops = &crypt_iv_plain64_ops; 2777 else if (strcmp(ivmode, "plain64be") == 0) 2778 cc->iv_gen_ops = &crypt_iv_plain64be_ops; 2779 else if (strcmp(ivmode, "essiv") == 0) 2780 cc->iv_gen_ops = &crypt_iv_essiv_ops; 2781 else if (strcmp(ivmode, "benbi") == 0) 2782 cc->iv_gen_ops = &crypt_iv_benbi_ops; 2783 else if (strcmp(ivmode, "null") == 0) 2784 cc->iv_gen_ops = &crypt_iv_null_ops; 2785 else if (strcmp(ivmode, "eboiv") == 0) 2786 cc->iv_gen_ops = &crypt_iv_eboiv_ops; 2787 else if (strcmp(ivmode, "elephant") == 0) { 2788 cc->iv_gen_ops = &crypt_iv_elephant_ops; 2789 cc->key_parts = 2; 2790 cc->key_extra_size = cc->key_size / 2; 2791 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE) 2792 return -EINVAL; 2793 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags); 2794 } else if (strcmp(ivmode, "lmk") == 0) { 2795 cc->iv_gen_ops = &crypt_iv_lmk_ops; 2796 /* 2797 * Version 2 and 3 is recognised according 2798 * to length of provided multi-key string. 2799 * If present (version 3), last key is used as IV seed. 2800 * All keys (including IV seed) are always the same size. 2801 */ 2802 if (cc->key_size % cc->key_parts) { 2803 cc->key_parts++; 2804 cc->key_extra_size = cc->key_size / cc->key_parts; 2805 } 2806 } else if (strcmp(ivmode, "tcw") == 0) { 2807 cc->iv_gen_ops = &crypt_iv_tcw_ops; 2808 cc->key_parts += 2; /* IV + whitening */ 2809 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE; 2810 } else if (strcmp(ivmode, "random") == 0) { 2811 cc->iv_gen_ops = &crypt_iv_random_ops; 2812 /* Need storage space in integrity fields. */ 2813 cc->integrity_iv_size = cc->iv_size; 2814 } else { 2815 ti->error = "Invalid IV mode"; 2816 return -EINVAL; 2817 } 2818 2819 return 0; 2820 } 2821 2822 /* 2823 * Workaround to parse HMAC algorithm from AEAD crypto API spec. 2824 * The HMAC is needed to calculate tag size (HMAC digest size). 2825 * This should be probably done by crypto-api calls (once available...) 2826 */ 2827 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api) 2828 { 2829 char *start, *end, *mac_alg = NULL; 2830 struct crypto_ahash *mac; 2831 2832 if (!strstarts(cipher_api, "authenc(")) 2833 return 0; 2834 2835 start = strchr(cipher_api, '('); 2836 end = strchr(cipher_api, ','); 2837 if (!start || !end || ++start > end) 2838 return -EINVAL; 2839 2840 mac_alg = kzalloc(end - start + 1, GFP_KERNEL); 2841 if (!mac_alg) 2842 return -ENOMEM; 2843 strncpy(mac_alg, start, end - start); 2844 2845 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY); 2846 kfree(mac_alg); 2847 2848 if (IS_ERR(mac)) 2849 return PTR_ERR(mac); 2850 2851 cc->key_mac_size = crypto_ahash_digestsize(mac); 2852 crypto_free_ahash(mac); 2853 2854 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL); 2855 if (!cc->authenc_key) 2856 return -ENOMEM; 2857 2858 return 0; 2859 } 2860 2861 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key, 2862 char **ivmode, char **ivopts) 2863 { 2864 struct crypt_config *cc = ti->private; 2865 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME]; 2866 int ret = -EINVAL; 2867 2868 cc->tfms_count = 1; 2869 2870 /* 2871 * New format (capi: prefix) 2872 * capi:cipher_api_spec-iv:ivopts 2873 */ 2874 tmp = &cipher_in[strlen("capi:")]; 2875 2876 /* Separate IV options if present, it can contain another '-' in hash name */ 2877 *ivopts = strrchr(tmp, ':'); 2878 if (*ivopts) { 2879 **ivopts = '\0'; 2880 (*ivopts)++; 2881 } 2882 /* Parse IV mode */ 2883 *ivmode = strrchr(tmp, '-'); 2884 if (*ivmode) { 2885 **ivmode = '\0'; 2886 (*ivmode)++; 2887 } 2888 /* The rest is crypto API spec */ 2889 cipher_api = tmp; 2890 2891 /* Alloc AEAD, can be used only in new format. */ 2892 if (crypt_integrity_aead(cc)) { 2893 ret = crypt_ctr_auth_cipher(cc, cipher_api); 2894 if (ret < 0) { 2895 ti->error = "Invalid AEAD cipher spec"; 2896 return -ENOMEM; 2897 } 2898 } 2899 2900 if (*ivmode && !strcmp(*ivmode, "lmk")) 2901 cc->tfms_count = 64; 2902 2903 if (*ivmode && !strcmp(*ivmode, "essiv")) { 2904 if (!*ivopts) { 2905 ti->error = "Digest algorithm missing for ESSIV mode"; 2906 return -EINVAL; 2907 } 2908 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)", 2909 cipher_api, *ivopts); 2910 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 2911 ti->error = "Cannot allocate cipher string"; 2912 return -ENOMEM; 2913 } 2914 cipher_api = buf; 2915 } 2916 2917 cc->key_parts = cc->tfms_count; 2918 2919 /* Allocate cipher */ 2920 ret = crypt_alloc_tfms(cc, cipher_api); 2921 if (ret < 0) { 2922 ti->error = "Error allocating crypto tfm"; 2923 return ret; 2924 } 2925 2926 if (crypt_integrity_aead(cc)) 2927 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2928 else 2929 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2930 2931 return 0; 2932 } 2933 2934 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key, 2935 char **ivmode, char **ivopts) 2936 { 2937 struct crypt_config *cc = ti->private; 2938 char *tmp, *cipher, *chainmode, *keycount; 2939 char *cipher_api = NULL; 2940 int ret = -EINVAL; 2941 char dummy; 2942 2943 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) { 2944 ti->error = "Bad cipher specification"; 2945 return -EINVAL; 2946 } 2947 2948 /* 2949 * Legacy dm-crypt cipher specification 2950 * cipher[:keycount]-mode-iv:ivopts 2951 */ 2952 tmp = cipher_in; 2953 keycount = strsep(&tmp, "-"); 2954 cipher = strsep(&keycount, ":"); 2955 2956 if (!keycount) 2957 cc->tfms_count = 1; 2958 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 || 2959 !is_power_of_2(cc->tfms_count)) { 2960 ti->error = "Bad cipher key count specification"; 2961 return -EINVAL; 2962 } 2963 cc->key_parts = cc->tfms_count; 2964 2965 chainmode = strsep(&tmp, "-"); 2966 *ivmode = strsep(&tmp, ":"); 2967 *ivopts = tmp; 2968 2969 /* 2970 * For compatibility with the original dm-crypt mapping format, if 2971 * only the cipher name is supplied, use cbc-plain. 2972 */ 2973 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) { 2974 chainmode = "cbc"; 2975 *ivmode = "plain"; 2976 } 2977 2978 if (strcmp(chainmode, "ecb") && !*ivmode) { 2979 ti->error = "IV mechanism required"; 2980 return -EINVAL; 2981 } 2982 2983 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL); 2984 if (!cipher_api) 2985 goto bad_mem; 2986 2987 if (*ivmode && !strcmp(*ivmode, "essiv")) { 2988 if (!*ivopts) { 2989 ti->error = "Digest algorithm missing for ESSIV mode"; 2990 kfree(cipher_api); 2991 return -EINVAL; 2992 } 2993 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 2994 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts); 2995 } else { 2996 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 2997 "%s(%s)", chainmode, cipher); 2998 } 2999 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 3000 kfree(cipher_api); 3001 goto bad_mem; 3002 } 3003 3004 /* Allocate cipher */ 3005 ret = crypt_alloc_tfms(cc, cipher_api); 3006 if (ret < 0) { 3007 ti->error = "Error allocating crypto tfm"; 3008 kfree(cipher_api); 3009 return ret; 3010 } 3011 kfree(cipher_api); 3012 3013 return 0; 3014 bad_mem: 3015 ti->error = "Cannot allocate cipher strings"; 3016 return -ENOMEM; 3017 } 3018 3019 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key) 3020 { 3021 struct crypt_config *cc = ti->private; 3022 char *ivmode = NULL, *ivopts = NULL; 3023 int ret; 3024 3025 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL); 3026 if (!cc->cipher_string) { 3027 ti->error = "Cannot allocate cipher strings"; 3028 return -ENOMEM; 3029 } 3030 3031 if (strstarts(cipher_in, "capi:")) 3032 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts); 3033 else 3034 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts); 3035 if (ret) 3036 return ret; 3037 3038 /* Initialize IV */ 3039 ret = crypt_ctr_ivmode(ti, ivmode); 3040 if (ret < 0) 3041 return ret; 3042 3043 /* Initialize and set key */ 3044 ret = crypt_set_key(cc, key); 3045 if (ret < 0) { 3046 ti->error = "Error decoding and setting key"; 3047 return ret; 3048 } 3049 3050 /* Allocate IV */ 3051 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) { 3052 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts); 3053 if (ret < 0) { 3054 ti->error = "Error creating IV"; 3055 return ret; 3056 } 3057 } 3058 3059 /* Initialize IV (set keys for ESSIV etc) */ 3060 if (cc->iv_gen_ops && cc->iv_gen_ops->init) { 3061 ret = cc->iv_gen_ops->init(cc); 3062 if (ret < 0) { 3063 ti->error = "Error initialising IV"; 3064 return ret; 3065 } 3066 } 3067 3068 /* wipe the kernel key payload copy */ 3069 if (cc->key_string) 3070 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3071 3072 return ret; 3073 } 3074 3075 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv) 3076 { 3077 struct crypt_config *cc = ti->private; 3078 struct dm_arg_set as; 3079 static const struct dm_arg _args[] = { 3080 {0, 8, "Invalid number of feature args"}, 3081 }; 3082 unsigned int opt_params, val; 3083 const char *opt_string, *sval; 3084 char dummy; 3085 int ret; 3086 3087 /* Optional parameters */ 3088 as.argc = argc; 3089 as.argv = argv; 3090 3091 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error); 3092 if (ret) 3093 return ret; 3094 3095 while (opt_params--) { 3096 opt_string = dm_shift_arg(&as); 3097 if (!opt_string) { 3098 ti->error = "Not enough feature arguments"; 3099 return -EINVAL; 3100 } 3101 3102 if (!strcasecmp(opt_string, "allow_discards")) 3103 ti->num_discard_bios = 1; 3104 3105 else if (!strcasecmp(opt_string, "same_cpu_crypt")) 3106 set_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3107 3108 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus")) 3109 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3110 else if (!strcasecmp(opt_string, "no_read_workqueue")) 3111 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3112 else if (!strcasecmp(opt_string, "no_write_workqueue")) 3113 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3114 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) { 3115 if (val == 0 || val > MAX_TAG_SIZE) { 3116 ti->error = "Invalid integrity arguments"; 3117 return -EINVAL; 3118 } 3119 cc->on_disk_tag_size = val; 3120 sval = strchr(opt_string + strlen("integrity:"), ':') + 1; 3121 if (!strcasecmp(sval, "aead")) { 3122 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 3123 } else if (strcasecmp(sval, "none")) { 3124 ti->error = "Unknown integrity profile"; 3125 return -EINVAL; 3126 } 3127 3128 cc->cipher_auth = kstrdup(sval, GFP_KERNEL); 3129 if (!cc->cipher_auth) 3130 return -ENOMEM; 3131 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) { 3132 if (cc->sector_size < (1 << SECTOR_SHIFT) || 3133 cc->sector_size > 4096 || 3134 (cc->sector_size & (cc->sector_size - 1))) { 3135 ti->error = "Invalid feature value for sector_size"; 3136 return -EINVAL; 3137 } 3138 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) { 3139 ti->error = "Device size is not multiple of sector_size feature"; 3140 return -EINVAL; 3141 } 3142 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT; 3143 } else if (!strcasecmp(opt_string, "iv_large_sectors")) 3144 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3145 else { 3146 ti->error = "Invalid feature arguments"; 3147 return -EINVAL; 3148 } 3149 } 3150 3151 return 0; 3152 } 3153 3154 #ifdef CONFIG_BLK_DEV_ZONED 3155 static int crypt_report_zones(struct dm_target *ti, 3156 struct dm_report_zones_args *args, unsigned int nr_zones) 3157 { 3158 struct crypt_config *cc = ti->private; 3159 3160 return dm_report_zones(cc->dev->bdev, cc->start, 3161 cc->start + dm_target_offset(ti, args->next_sector), 3162 args, nr_zones); 3163 } 3164 #else 3165 #define crypt_report_zones NULL 3166 #endif 3167 3168 /* 3169 * Construct an encryption mapping: 3170 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start> 3171 */ 3172 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) 3173 { 3174 struct crypt_config *cc; 3175 const char *devname = dm_table_device_name(ti->table); 3176 int key_size; 3177 unsigned int align_mask; 3178 unsigned long long tmpll; 3179 int ret; 3180 size_t iv_size_padding, additional_req_size; 3181 char dummy; 3182 3183 if (argc < 5) { 3184 ti->error = "Not enough arguments"; 3185 return -EINVAL; 3186 } 3187 3188 key_size = get_key_size(&argv[1]); 3189 if (key_size < 0) { 3190 ti->error = "Cannot parse key size"; 3191 return -EINVAL; 3192 } 3193 3194 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL); 3195 if (!cc) { 3196 ti->error = "Cannot allocate encryption context"; 3197 return -ENOMEM; 3198 } 3199 cc->key_size = key_size; 3200 cc->sector_size = (1 << SECTOR_SHIFT); 3201 cc->sector_shift = 0; 3202 3203 ti->private = cc; 3204 3205 spin_lock(&dm_crypt_clients_lock); 3206 dm_crypt_clients_n++; 3207 crypt_calculate_pages_per_client(); 3208 spin_unlock(&dm_crypt_clients_lock); 3209 3210 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL); 3211 if (ret < 0) 3212 goto bad; 3213 3214 /* Optional parameters need to be read before cipher constructor */ 3215 if (argc > 5) { 3216 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]); 3217 if (ret) 3218 goto bad; 3219 } 3220 3221 ret = crypt_ctr_cipher(ti, argv[0], argv[1]); 3222 if (ret < 0) 3223 goto bad; 3224 3225 if (crypt_integrity_aead(cc)) { 3226 cc->dmreq_start = sizeof(struct aead_request); 3227 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc)); 3228 align_mask = crypto_aead_alignmask(any_tfm_aead(cc)); 3229 } else { 3230 cc->dmreq_start = sizeof(struct skcipher_request); 3231 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc)); 3232 align_mask = crypto_skcipher_alignmask(any_tfm(cc)); 3233 } 3234 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request)); 3235 3236 if (align_mask < CRYPTO_MINALIGN) { 3237 /* Allocate the padding exactly */ 3238 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request)) 3239 & align_mask; 3240 } else { 3241 /* 3242 * If the cipher requires greater alignment than kmalloc 3243 * alignment, we don't know the exact position of the 3244 * initialization vector. We must assume worst case. 3245 */ 3246 iv_size_padding = align_mask; 3247 } 3248 3249 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */ 3250 additional_req_size = sizeof(struct dm_crypt_request) + 3251 iv_size_padding + cc->iv_size + 3252 cc->iv_size + 3253 sizeof(uint64_t) + 3254 sizeof(unsigned int); 3255 3256 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size); 3257 if (ret) { 3258 ti->error = "Cannot allocate crypt request mempool"; 3259 goto bad; 3260 } 3261 3262 cc->per_bio_data_size = ti->per_io_data_size = 3263 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size, 3264 ARCH_KMALLOC_MINALIGN); 3265 3266 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc); 3267 if (ret) { 3268 ti->error = "Cannot allocate page mempool"; 3269 goto bad; 3270 } 3271 3272 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS); 3273 if (ret) { 3274 ti->error = "Cannot allocate crypt bioset"; 3275 goto bad; 3276 } 3277 3278 mutex_init(&cc->bio_alloc_lock); 3279 3280 ret = -EINVAL; 3281 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) || 3282 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) { 3283 ti->error = "Invalid iv_offset sector"; 3284 goto bad; 3285 } 3286 cc->iv_offset = tmpll; 3287 3288 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev); 3289 if (ret) { 3290 ti->error = "Device lookup failed"; 3291 goto bad; 3292 } 3293 3294 ret = -EINVAL; 3295 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) { 3296 ti->error = "Invalid device sector"; 3297 goto bad; 3298 } 3299 cc->start = tmpll; 3300 3301 if (bdev_is_zoned(cc->dev->bdev)) { 3302 /* 3303 * For zoned block devices, we need to preserve the issuer write 3304 * ordering. To do so, disable write workqueues and force inline 3305 * encryption completion. 3306 */ 3307 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3308 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags); 3309 3310 /* 3311 * All zone append writes to a zone of a zoned block device will 3312 * have the same BIO sector, the start of the zone. When the 3313 * cypher IV mode uses sector values, all data targeting a 3314 * zone will be encrypted using the first sector numbers of the 3315 * zone. This will not result in write errors but will 3316 * cause most reads to fail as reads will use the sector values 3317 * for the actual data locations, resulting in IV mismatch. 3318 * To avoid this problem, ask DM core to emulate zone append 3319 * operations with regular writes. 3320 */ 3321 DMDEBUG("Zone append operations will be emulated"); 3322 ti->emulate_zone_append = true; 3323 } 3324 3325 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) { 3326 ret = crypt_integrity_ctr(cc, ti); 3327 if (ret) 3328 goto bad; 3329 3330 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size; 3331 if (!cc->tag_pool_max_sectors) 3332 cc->tag_pool_max_sectors = 1; 3333 3334 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS, 3335 cc->tag_pool_max_sectors * cc->on_disk_tag_size); 3336 if (ret) { 3337 ti->error = "Cannot allocate integrity tags mempool"; 3338 goto bad; 3339 } 3340 3341 cc->tag_pool_max_sectors <<= cc->sector_shift; 3342 } 3343 3344 ret = -ENOMEM; 3345 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname); 3346 if (!cc->io_queue) { 3347 ti->error = "Couldn't create kcryptd io queue"; 3348 goto bad; 3349 } 3350 3351 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) 3352 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 3353 1, devname); 3354 else 3355 cc->crypt_queue = alloc_workqueue("kcryptd/%s", 3356 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, 3357 num_online_cpus(), devname); 3358 if (!cc->crypt_queue) { 3359 ti->error = "Couldn't create kcryptd queue"; 3360 goto bad; 3361 } 3362 3363 spin_lock_init(&cc->write_thread_lock); 3364 cc->write_tree = RB_ROOT; 3365 3366 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname); 3367 if (IS_ERR(cc->write_thread)) { 3368 ret = PTR_ERR(cc->write_thread); 3369 cc->write_thread = NULL; 3370 ti->error = "Couldn't spawn write thread"; 3371 goto bad; 3372 } 3373 3374 ti->num_flush_bios = 1; 3375 ti->limit_swap_bios = true; 3376 3377 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1); 3378 return 0; 3379 3380 bad: 3381 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0); 3382 crypt_dtr(ti); 3383 return ret; 3384 } 3385 3386 static int crypt_map(struct dm_target *ti, struct bio *bio) 3387 { 3388 struct dm_crypt_io *io; 3389 struct crypt_config *cc = ti->private; 3390 3391 /* 3392 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues. 3393 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight 3394 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters 3395 */ 3396 if (unlikely(bio->bi_opf & REQ_PREFLUSH || 3397 bio_op(bio) == REQ_OP_DISCARD)) { 3398 bio_set_dev(bio, cc->dev->bdev); 3399 if (bio_sectors(bio)) 3400 bio->bi_iter.bi_sector = cc->start + 3401 dm_target_offset(ti, bio->bi_iter.bi_sector); 3402 return DM_MAPIO_REMAPPED; 3403 } 3404 3405 /* 3406 * Check if bio is too large, split as needed. 3407 */ 3408 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) && 3409 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size)) 3410 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT)); 3411 3412 /* 3413 * Ensure that bio is a multiple of internal sector encryption size 3414 * and is aligned to this size as defined in IO hints. 3415 */ 3416 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0)) 3417 return DM_MAPIO_KILL; 3418 3419 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1))) 3420 return DM_MAPIO_KILL; 3421 3422 io = dm_per_bio_data(bio, cc->per_bio_data_size); 3423 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector)); 3424 3425 if (cc->on_disk_tag_size) { 3426 unsigned tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift); 3427 3428 if (unlikely(tag_len > KMALLOC_MAX_SIZE) || 3429 unlikely(!(io->integrity_metadata = kmalloc(tag_len, 3430 GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN)))) { 3431 if (bio_sectors(bio) > cc->tag_pool_max_sectors) 3432 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors); 3433 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO); 3434 io->integrity_metadata_from_pool = true; 3435 } 3436 } 3437 3438 if (crypt_integrity_aead(cc)) 3439 io->ctx.r.req_aead = (struct aead_request *)(io + 1); 3440 else 3441 io->ctx.r.req = (struct skcipher_request *)(io + 1); 3442 3443 if (bio_data_dir(io->base_bio) == READ) { 3444 if (kcryptd_io_read(io, GFP_NOWAIT)) 3445 kcryptd_queue_read(io); 3446 } else 3447 kcryptd_queue_crypt(io); 3448 3449 return DM_MAPIO_SUBMITTED; 3450 } 3451 3452 static void crypt_status(struct dm_target *ti, status_type_t type, 3453 unsigned status_flags, char *result, unsigned maxlen) 3454 { 3455 struct crypt_config *cc = ti->private; 3456 unsigned i, sz = 0; 3457 int num_feature_args = 0; 3458 3459 switch (type) { 3460 case STATUSTYPE_INFO: 3461 result[0] = '\0'; 3462 break; 3463 3464 case STATUSTYPE_TABLE: 3465 DMEMIT("%s ", cc->cipher_string); 3466 3467 if (cc->key_size > 0) { 3468 if (cc->key_string) 3469 DMEMIT(":%u:%s", cc->key_size, cc->key_string); 3470 else 3471 for (i = 0; i < cc->key_size; i++) 3472 DMEMIT("%02x", cc->key[i]); 3473 } else 3474 DMEMIT("-"); 3475 3476 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset, 3477 cc->dev->name, (unsigned long long)cc->start); 3478 3479 num_feature_args += !!ti->num_discard_bios; 3480 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3481 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3482 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3483 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3484 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT); 3485 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3486 if (cc->on_disk_tag_size) 3487 num_feature_args++; 3488 if (num_feature_args) { 3489 DMEMIT(" %d", num_feature_args); 3490 if (ti->num_discard_bios) 3491 DMEMIT(" allow_discards"); 3492 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) 3493 DMEMIT(" same_cpu_crypt"); 3494 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) 3495 DMEMIT(" submit_from_crypt_cpus"); 3496 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) 3497 DMEMIT(" no_read_workqueue"); 3498 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) 3499 DMEMIT(" no_write_workqueue"); 3500 if (cc->on_disk_tag_size) 3501 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth); 3502 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3503 DMEMIT(" sector_size:%d", cc->sector_size); 3504 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 3505 DMEMIT(" iv_large_sectors"); 3506 } 3507 break; 3508 3509 case STATUSTYPE_IMA: 3510 DMEMIT_TARGET_NAME_VERSION(ti->type); 3511 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n'); 3512 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n'); 3513 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ? 3514 'y' : 'n'); 3515 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ? 3516 'y' : 'n'); 3517 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ? 3518 'y' : 'n'); 3519 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ? 3520 'y' : 'n'); 3521 3522 if (cc->on_disk_tag_size) 3523 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s", 3524 cc->on_disk_tag_size, cc->cipher_auth); 3525 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3526 DMEMIT(",sector_size=%d", cc->sector_size); 3527 if (cc->cipher_string) 3528 DMEMIT(",cipher_string=%s", cc->cipher_string); 3529 3530 DMEMIT(",key_size=%u", cc->key_size); 3531 DMEMIT(",key_parts=%u", cc->key_parts); 3532 DMEMIT(",key_extra_size=%u", cc->key_extra_size); 3533 DMEMIT(",key_mac_size=%u", cc->key_mac_size); 3534 DMEMIT(";"); 3535 break; 3536 } 3537 } 3538 3539 static void crypt_postsuspend(struct dm_target *ti) 3540 { 3541 struct crypt_config *cc = ti->private; 3542 3543 set_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3544 } 3545 3546 static int crypt_preresume(struct dm_target *ti) 3547 { 3548 struct crypt_config *cc = ti->private; 3549 3550 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) { 3551 DMERR("aborting resume - crypt key is not set."); 3552 return -EAGAIN; 3553 } 3554 3555 return 0; 3556 } 3557 3558 static void crypt_resume(struct dm_target *ti) 3559 { 3560 struct crypt_config *cc = ti->private; 3561 3562 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3563 } 3564 3565 /* Message interface 3566 * key set <key> 3567 * key wipe 3568 */ 3569 static int crypt_message(struct dm_target *ti, unsigned argc, char **argv, 3570 char *result, unsigned maxlen) 3571 { 3572 struct crypt_config *cc = ti->private; 3573 int key_size, ret = -EINVAL; 3574 3575 if (argc < 2) 3576 goto error; 3577 3578 if (!strcasecmp(argv[0], "key")) { 3579 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) { 3580 DMWARN("not suspended during key manipulation."); 3581 return -EINVAL; 3582 } 3583 if (argc == 3 && !strcasecmp(argv[1], "set")) { 3584 /* The key size may not be changed. */ 3585 key_size = get_key_size(&argv[2]); 3586 if (key_size < 0 || cc->key_size != key_size) { 3587 memset(argv[2], '0', strlen(argv[2])); 3588 return -EINVAL; 3589 } 3590 3591 ret = crypt_set_key(cc, argv[2]); 3592 if (ret) 3593 return ret; 3594 if (cc->iv_gen_ops && cc->iv_gen_ops->init) 3595 ret = cc->iv_gen_ops->init(cc); 3596 /* wipe the kernel key payload copy */ 3597 if (cc->key_string) 3598 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3599 return ret; 3600 } 3601 if (argc == 2 && !strcasecmp(argv[1], "wipe")) 3602 return crypt_wipe_key(cc); 3603 } 3604 3605 error: 3606 DMWARN("unrecognised message received."); 3607 return -EINVAL; 3608 } 3609 3610 static int crypt_iterate_devices(struct dm_target *ti, 3611 iterate_devices_callout_fn fn, void *data) 3612 { 3613 struct crypt_config *cc = ti->private; 3614 3615 return fn(ti, cc->dev, cc->start, ti->len, data); 3616 } 3617 3618 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits) 3619 { 3620 struct crypt_config *cc = ti->private; 3621 3622 /* 3623 * Unfortunate constraint that is required to avoid the potential 3624 * for exceeding underlying device's max_segments limits -- due to 3625 * crypt_alloc_buffer() possibly allocating pages for the encryption 3626 * bio that are not as physically contiguous as the original bio. 3627 */ 3628 limits->max_segment_size = PAGE_SIZE; 3629 3630 limits->logical_block_size = 3631 max_t(unsigned, limits->logical_block_size, cc->sector_size); 3632 limits->physical_block_size = 3633 max_t(unsigned, limits->physical_block_size, cc->sector_size); 3634 limits->io_min = max_t(unsigned, limits->io_min, cc->sector_size); 3635 } 3636 3637 static struct target_type crypt_target = { 3638 .name = "crypt", 3639 .version = {1, 23, 0}, 3640 .module = THIS_MODULE, 3641 .ctr = crypt_ctr, 3642 .dtr = crypt_dtr, 3643 .features = DM_TARGET_ZONED_HM, 3644 .report_zones = crypt_report_zones, 3645 .map = crypt_map, 3646 .status = crypt_status, 3647 .postsuspend = crypt_postsuspend, 3648 .preresume = crypt_preresume, 3649 .resume = crypt_resume, 3650 .message = crypt_message, 3651 .iterate_devices = crypt_iterate_devices, 3652 .io_hints = crypt_io_hints, 3653 }; 3654 3655 static int __init dm_crypt_init(void) 3656 { 3657 int r; 3658 3659 r = dm_register_target(&crypt_target); 3660 if (r < 0) 3661 DMERR("register failed %d", r); 3662 3663 return r; 3664 } 3665 3666 static void __exit dm_crypt_exit(void) 3667 { 3668 dm_unregister_target(&crypt_target); 3669 } 3670 3671 module_init(dm_crypt_init); 3672 module_exit(dm_crypt_exit); 3673 3674 MODULE_AUTHOR("Jana Saout <jana@saout.de>"); 3675 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption"); 3676 MODULE_LICENSE("GPL"); 3677