1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2003 Jana Saout <jana@saout.de> 4 * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org> 5 * Copyright (C) 2006-2020 Red Hat, Inc. All rights reserved. 6 * Copyright (C) 2013-2020 Milan Broz <gmazyland@gmail.com> 7 * 8 * This file is released under the GPL. 9 */ 10 11 #include <linux/completion.h> 12 #include <linux/err.h> 13 #include <linux/module.h> 14 #include <linux/init.h> 15 #include <linux/kernel.h> 16 #include <linux/key.h> 17 #include <linux/bio.h> 18 #include <linux/blkdev.h> 19 #include <linux/blk-integrity.h> 20 #include <linux/mempool.h> 21 #include <linux/slab.h> 22 #include <linux/crypto.h> 23 #include <linux/workqueue.h> 24 #include <linux/kthread.h> 25 #include <linux/backing-dev.h> 26 #include <linux/atomic.h> 27 #include <linux/scatterlist.h> 28 #include <linux/rbtree.h> 29 #include <linux/ctype.h> 30 #include <asm/page.h> 31 #include <asm/unaligned.h> 32 #include <crypto/hash.h> 33 #include <crypto/md5.h> 34 #include <crypto/skcipher.h> 35 #include <crypto/aead.h> 36 #include <crypto/authenc.h> 37 #include <crypto/utils.h> 38 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */ 39 #include <linux/key-type.h> 40 #include <keys/user-type.h> 41 #include <keys/encrypted-type.h> 42 #include <keys/trusted-type.h> 43 44 #include <linux/device-mapper.h> 45 46 #include "dm-audit.h" 47 48 #define DM_MSG_PREFIX "crypt" 49 50 /* 51 * context holding the current state of a multi-part conversion 52 */ 53 struct convert_context { 54 struct completion restart; 55 struct bio *bio_in; 56 struct bio *bio_out; 57 struct bvec_iter iter_in; 58 struct bvec_iter iter_out; 59 u64 cc_sector; 60 atomic_t cc_pending; 61 union { 62 struct skcipher_request *req; 63 struct aead_request *req_aead; 64 } r; 65 66 }; 67 68 /* 69 * per bio private data 70 */ 71 struct dm_crypt_io { 72 struct crypt_config *cc; 73 struct bio *base_bio; 74 u8 *integrity_metadata; 75 bool integrity_metadata_from_pool:1; 76 bool in_tasklet:1; 77 78 struct work_struct work; 79 struct tasklet_struct tasklet; 80 81 struct convert_context ctx; 82 83 atomic_t io_pending; 84 blk_status_t error; 85 sector_t sector; 86 87 struct rb_node rb_node; 88 } CRYPTO_MINALIGN_ATTR; 89 90 struct dm_crypt_request { 91 struct convert_context *ctx; 92 struct scatterlist sg_in[4]; 93 struct scatterlist sg_out[4]; 94 u64 iv_sector; 95 }; 96 97 struct crypt_config; 98 99 struct crypt_iv_operations { 100 int (*ctr)(struct crypt_config *cc, struct dm_target *ti, 101 const char *opts); 102 void (*dtr)(struct crypt_config *cc); 103 int (*init)(struct crypt_config *cc); 104 int (*wipe)(struct crypt_config *cc); 105 int (*generator)(struct crypt_config *cc, u8 *iv, 106 struct dm_crypt_request *dmreq); 107 int (*post)(struct crypt_config *cc, u8 *iv, 108 struct dm_crypt_request *dmreq); 109 }; 110 111 struct iv_benbi_private { 112 int shift; 113 }; 114 115 #define LMK_SEED_SIZE 64 /* hash + 0 */ 116 struct iv_lmk_private { 117 struct crypto_shash *hash_tfm; 118 u8 *seed; 119 }; 120 121 #define TCW_WHITENING_SIZE 16 122 struct iv_tcw_private { 123 struct crypto_shash *crc32_tfm; 124 u8 *iv_seed; 125 u8 *whitening; 126 }; 127 128 #define ELEPHANT_MAX_KEY_SIZE 32 129 struct iv_elephant_private { 130 struct crypto_skcipher *tfm; 131 }; 132 133 /* 134 * Crypt: maps a linear range of a block device 135 * and encrypts / decrypts at the same time. 136 */ 137 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID, 138 DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD, 139 DM_CRYPT_NO_READ_WORKQUEUE, DM_CRYPT_NO_WRITE_WORKQUEUE, 140 DM_CRYPT_WRITE_INLINE }; 141 142 enum cipher_flags { 143 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */ 144 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */ 145 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */ 146 }; 147 148 /* 149 * The fields in here must be read only after initialization. 150 */ 151 struct crypt_config { 152 struct dm_dev *dev; 153 sector_t start; 154 155 struct percpu_counter n_allocated_pages; 156 157 struct workqueue_struct *io_queue; 158 struct workqueue_struct *crypt_queue; 159 160 spinlock_t write_thread_lock; 161 struct task_struct *write_thread; 162 struct rb_root write_tree; 163 164 char *cipher_string; 165 char *cipher_auth; 166 char *key_string; 167 168 const struct crypt_iv_operations *iv_gen_ops; 169 union { 170 struct iv_benbi_private benbi; 171 struct iv_lmk_private lmk; 172 struct iv_tcw_private tcw; 173 struct iv_elephant_private elephant; 174 } iv_gen_private; 175 u64 iv_offset; 176 unsigned int iv_size; 177 unsigned short sector_size; 178 unsigned char sector_shift; 179 180 union { 181 struct crypto_skcipher **tfms; 182 struct crypto_aead **tfms_aead; 183 } cipher_tfm; 184 unsigned int tfms_count; 185 unsigned long cipher_flags; 186 187 /* 188 * Layout of each crypto request: 189 * 190 * struct skcipher_request 191 * context 192 * padding 193 * struct dm_crypt_request 194 * padding 195 * IV 196 * 197 * The padding is added so that dm_crypt_request and the IV are 198 * correctly aligned. 199 */ 200 unsigned int dmreq_start; 201 202 unsigned int per_bio_data_size; 203 204 unsigned long flags; 205 unsigned int key_size; 206 unsigned int key_parts; /* independent parts in key buffer */ 207 unsigned int key_extra_size; /* additional keys length */ 208 unsigned int key_mac_size; /* MAC key size for authenc(...) */ 209 210 unsigned int integrity_tag_size; 211 unsigned int integrity_iv_size; 212 unsigned int on_disk_tag_size; 213 214 /* 215 * pool for per bio private data, crypto requests, 216 * encryption requeusts/buffer pages and integrity tags 217 */ 218 unsigned int tag_pool_max_sectors; 219 mempool_t tag_pool; 220 mempool_t req_pool; 221 mempool_t page_pool; 222 223 struct bio_set bs; 224 struct mutex bio_alloc_lock; 225 226 u8 *authenc_key; /* space for keys in authenc() format (if used) */ 227 u8 key[]; 228 }; 229 230 #define MIN_IOS 64 231 #define MAX_TAG_SIZE 480 232 #define POOL_ENTRY_SIZE 512 233 234 static DEFINE_SPINLOCK(dm_crypt_clients_lock); 235 static unsigned int dm_crypt_clients_n; 236 static volatile unsigned long dm_crypt_pages_per_client; 237 #define DM_CRYPT_MEMORY_PERCENT 2 238 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16) 239 240 static void crypt_endio(struct bio *clone); 241 static void kcryptd_queue_crypt(struct dm_crypt_io *io); 242 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 243 struct scatterlist *sg); 244 245 static bool crypt_integrity_aead(struct crypt_config *cc); 246 247 /* 248 * Use this to access cipher attributes that are independent of the key. 249 */ 250 static struct crypto_skcipher *any_tfm(struct crypt_config *cc) 251 { 252 return cc->cipher_tfm.tfms[0]; 253 } 254 255 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc) 256 { 257 return cc->cipher_tfm.tfms_aead[0]; 258 } 259 260 /* 261 * Different IV generation algorithms: 262 * 263 * plain: the initial vector is the 32-bit little-endian version of the sector 264 * number, padded with zeros if necessary. 265 * 266 * plain64: the initial vector is the 64-bit little-endian version of the sector 267 * number, padded with zeros if necessary. 268 * 269 * plain64be: the initial vector is the 64-bit big-endian version of the sector 270 * number, padded with zeros if necessary. 271 * 272 * essiv: "encrypted sector|salt initial vector", the sector number is 273 * encrypted with the bulk cipher using a salt as key. The salt 274 * should be derived from the bulk cipher's key via hashing. 275 * 276 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1 277 * (needed for LRW-32-AES and possible other narrow block modes) 278 * 279 * null: the initial vector is always zero. Provides compatibility with 280 * obsolete loop_fish2 devices. Do not use for new devices. 281 * 282 * lmk: Compatible implementation of the block chaining mode used 283 * by the Loop-AES block device encryption system 284 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/ 285 * It operates on full 512 byte sectors and uses CBC 286 * with an IV derived from the sector number, the data and 287 * optionally extra IV seed. 288 * This means that after decryption the first block 289 * of sector must be tweaked according to decrypted data. 290 * Loop-AES can use three encryption schemes: 291 * version 1: is plain aes-cbc mode 292 * version 2: uses 64 multikey scheme with lmk IV generator 293 * version 3: the same as version 2 with additional IV seed 294 * (it uses 65 keys, last key is used as IV seed) 295 * 296 * tcw: Compatible implementation of the block chaining mode used 297 * by the TrueCrypt device encryption system (prior to version 4.1). 298 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat 299 * It operates on full 512 byte sectors and uses CBC 300 * with an IV derived from initial key and the sector number. 301 * In addition, whitening value is applied on every sector, whitening 302 * is calculated from initial key, sector number and mixed using CRC32. 303 * Note that this encryption scheme is vulnerable to watermarking attacks 304 * and should be used for old compatible containers access only. 305 * 306 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode) 307 * The IV is encrypted little-endian byte-offset (with the same key 308 * and cipher as the volume). 309 * 310 * elephant: The extended version of eboiv with additional Elephant diffuser 311 * used with Bitlocker CBC mode. 312 * This mode was used in older Windows systems 313 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf 314 */ 315 316 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, 317 struct dm_crypt_request *dmreq) 318 { 319 memset(iv, 0, cc->iv_size); 320 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff); 321 322 return 0; 323 } 324 325 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv, 326 struct dm_crypt_request *dmreq) 327 { 328 memset(iv, 0, cc->iv_size); 329 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 330 331 return 0; 332 } 333 334 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv, 335 struct dm_crypt_request *dmreq) 336 { 337 memset(iv, 0, cc->iv_size); 338 /* iv_size is at least of size u64; usually it is 16 bytes */ 339 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector); 340 341 return 0; 342 } 343 344 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, 345 struct dm_crypt_request *dmreq) 346 { 347 /* 348 * ESSIV encryption of the IV is now handled by the crypto API, 349 * so just pass the plain sector number here. 350 */ 351 memset(iv, 0, cc->iv_size); 352 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 353 354 return 0; 355 } 356 357 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti, 358 const char *opts) 359 { 360 unsigned int bs; 361 int log; 362 363 if (crypt_integrity_aead(cc)) 364 bs = crypto_aead_blocksize(any_tfm_aead(cc)); 365 else 366 bs = crypto_skcipher_blocksize(any_tfm(cc)); 367 log = ilog2(bs); 368 369 /* 370 * We need to calculate how far we must shift the sector count 371 * to get the cipher block count, we use this shift in _gen. 372 */ 373 if (1 << log != bs) { 374 ti->error = "cypher blocksize is not a power of 2"; 375 return -EINVAL; 376 } 377 378 if (log > 9) { 379 ti->error = "cypher blocksize is > 512"; 380 return -EINVAL; 381 } 382 383 cc->iv_gen_private.benbi.shift = 9 - log; 384 385 return 0; 386 } 387 388 static void crypt_iv_benbi_dtr(struct crypt_config *cc) 389 { 390 } 391 392 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv, 393 struct dm_crypt_request *dmreq) 394 { 395 __be64 val; 396 397 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */ 398 399 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1); 400 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64))); 401 402 return 0; 403 } 404 405 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv, 406 struct dm_crypt_request *dmreq) 407 { 408 memset(iv, 0, cc->iv_size); 409 410 return 0; 411 } 412 413 static void crypt_iv_lmk_dtr(struct crypt_config *cc) 414 { 415 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 416 417 if (lmk->hash_tfm && !IS_ERR(lmk->hash_tfm)) 418 crypto_free_shash(lmk->hash_tfm); 419 lmk->hash_tfm = NULL; 420 421 kfree_sensitive(lmk->seed); 422 lmk->seed = NULL; 423 } 424 425 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti, 426 const char *opts) 427 { 428 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 429 430 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 431 ti->error = "Unsupported sector size for LMK"; 432 return -EINVAL; 433 } 434 435 lmk->hash_tfm = crypto_alloc_shash("md5", 0, 436 CRYPTO_ALG_ALLOCATES_MEMORY); 437 if (IS_ERR(lmk->hash_tfm)) { 438 ti->error = "Error initializing LMK hash"; 439 return PTR_ERR(lmk->hash_tfm); 440 } 441 442 /* No seed in LMK version 2 */ 443 if (cc->key_parts == cc->tfms_count) { 444 lmk->seed = NULL; 445 return 0; 446 } 447 448 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL); 449 if (!lmk->seed) { 450 crypt_iv_lmk_dtr(cc); 451 ti->error = "Error kmallocing seed storage in LMK"; 452 return -ENOMEM; 453 } 454 455 return 0; 456 } 457 458 static int crypt_iv_lmk_init(struct crypt_config *cc) 459 { 460 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 461 int subkey_size = cc->key_size / cc->key_parts; 462 463 /* LMK seed is on the position of LMK_KEYS + 1 key */ 464 if (lmk->seed) 465 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size), 466 crypto_shash_digestsize(lmk->hash_tfm)); 467 468 return 0; 469 } 470 471 static int crypt_iv_lmk_wipe(struct crypt_config *cc) 472 { 473 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 474 475 if (lmk->seed) 476 memset(lmk->seed, 0, LMK_SEED_SIZE); 477 478 return 0; 479 } 480 481 static int crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv, 482 struct dm_crypt_request *dmreq, 483 u8 *data) 484 { 485 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 486 SHASH_DESC_ON_STACK(desc, lmk->hash_tfm); 487 struct md5_state md5state; 488 __le32 buf[4]; 489 int i, r; 490 491 desc->tfm = lmk->hash_tfm; 492 493 r = crypto_shash_init(desc); 494 if (r) 495 return r; 496 497 if (lmk->seed) { 498 r = crypto_shash_update(desc, lmk->seed, LMK_SEED_SIZE); 499 if (r) 500 return r; 501 } 502 503 /* Sector is always 512B, block size 16, add data of blocks 1-31 */ 504 r = crypto_shash_update(desc, data + 16, 16 * 31); 505 if (r) 506 return r; 507 508 /* Sector is cropped to 56 bits here */ 509 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF); 510 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000); 511 buf[2] = cpu_to_le32(4024); 512 buf[3] = 0; 513 r = crypto_shash_update(desc, (u8 *)buf, sizeof(buf)); 514 if (r) 515 return r; 516 517 /* No MD5 padding here */ 518 r = crypto_shash_export(desc, &md5state); 519 if (r) 520 return r; 521 522 for (i = 0; i < MD5_HASH_WORDS; i++) 523 __cpu_to_le32s(&md5state.hash[i]); 524 memcpy(iv, &md5state.hash, cc->iv_size); 525 526 return 0; 527 } 528 529 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv, 530 struct dm_crypt_request *dmreq) 531 { 532 struct scatterlist *sg; 533 u8 *src; 534 int r = 0; 535 536 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 537 sg = crypt_get_sg_data(cc, dmreq->sg_in); 538 src = kmap_local_page(sg_page(sg)); 539 r = crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset); 540 kunmap_local(src); 541 } else 542 memset(iv, 0, cc->iv_size); 543 544 return r; 545 } 546 547 static int crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv, 548 struct dm_crypt_request *dmreq) 549 { 550 struct scatterlist *sg; 551 u8 *dst; 552 int r; 553 554 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) 555 return 0; 556 557 sg = crypt_get_sg_data(cc, dmreq->sg_out); 558 dst = kmap_local_page(sg_page(sg)); 559 r = crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset); 560 561 /* Tweak the first block of plaintext sector */ 562 if (!r) 563 crypto_xor(dst + sg->offset, iv, cc->iv_size); 564 565 kunmap_local(dst); 566 return r; 567 } 568 569 static void crypt_iv_tcw_dtr(struct crypt_config *cc) 570 { 571 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 572 573 kfree_sensitive(tcw->iv_seed); 574 tcw->iv_seed = NULL; 575 kfree_sensitive(tcw->whitening); 576 tcw->whitening = NULL; 577 578 if (tcw->crc32_tfm && !IS_ERR(tcw->crc32_tfm)) 579 crypto_free_shash(tcw->crc32_tfm); 580 tcw->crc32_tfm = NULL; 581 } 582 583 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti, 584 const char *opts) 585 { 586 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 587 588 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 589 ti->error = "Unsupported sector size for TCW"; 590 return -EINVAL; 591 } 592 593 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) { 594 ti->error = "Wrong key size for TCW"; 595 return -EINVAL; 596 } 597 598 tcw->crc32_tfm = crypto_alloc_shash("crc32", 0, 599 CRYPTO_ALG_ALLOCATES_MEMORY); 600 if (IS_ERR(tcw->crc32_tfm)) { 601 ti->error = "Error initializing CRC32 in TCW"; 602 return PTR_ERR(tcw->crc32_tfm); 603 } 604 605 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL); 606 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL); 607 if (!tcw->iv_seed || !tcw->whitening) { 608 crypt_iv_tcw_dtr(cc); 609 ti->error = "Error allocating seed storage in TCW"; 610 return -ENOMEM; 611 } 612 613 return 0; 614 } 615 616 static int crypt_iv_tcw_init(struct crypt_config *cc) 617 { 618 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 619 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE; 620 621 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size); 622 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size], 623 TCW_WHITENING_SIZE); 624 625 return 0; 626 } 627 628 static int crypt_iv_tcw_wipe(struct crypt_config *cc) 629 { 630 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 631 632 memset(tcw->iv_seed, 0, cc->iv_size); 633 memset(tcw->whitening, 0, TCW_WHITENING_SIZE); 634 635 return 0; 636 } 637 638 static int crypt_iv_tcw_whitening(struct crypt_config *cc, 639 struct dm_crypt_request *dmreq, 640 u8 *data) 641 { 642 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 643 __le64 sector = cpu_to_le64(dmreq->iv_sector); 644 u8 buf[TCW_WHITENING_SIZE]; 645 SHASH_DESC_ON_STACK(desc, tcw->crc32_tfm); 646 int i, r; 647 648 /* xor whitening with sector number */ 649 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8); 650 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8); 651 652 /* calculate crc32 for every 32bit part and xor it */ 653 desc->tfm = tcw->crc32_tfm; 654 for (i = 0; i < 4; i++) { 655 r = crypto_shash_init(desc); 656 if (r) 657 goto out; 658 r = crypto_shash_update(desc, &buf[i * 4], 4); 659 if (r) 660 goto out; 661 r = crypto_shash_final(desc, &buf[i * 4]); 662 if (r) 663 goto out; 664 } 665 crypto_xor(&buf[0], &buf[12], 4); 666 crypto_xor(&buf[4], &buf[8], 4); 667 668 /* apply whitening (8 bytes) to whole sector */ 669 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++) 670 crypto_xor(data + i * 8, buf, 8); 671 out: 672 memzero_explicit(buf, sizeof(buf)); 673 return r; 674 } 675 676 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv, 677 struct dm_crypt_request *dmreq) 678 { 679 struct scatterlist *sg; 680 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 681 __le64 sector = cpu_to_le64(dmreq->iv_sector); 682 u8 *src; 683 int r = 0; 684 685 /* Remove whitening from ciphertext */ 686 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 687 sg = crypt_get_sg_data(cc, dmreq->sg_in); 688 src = kmap_local_page(sg_page(sg)); 689 r = crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset); 690 kunmap_local(src); 691 } 692 693 /* Calculate IV */ 694 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8); 695 if (cc->iv_size > 8) 696 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or, 697 cc->iv_size - 8); 698 699 return r; 700 } 701 702 static int crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv, 703 struct dm_crypt_request *dmreq) 704 { 705 struct scatterlist *sg; 706 u8 *dst; 707 int r; 708 709 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 710 return 0; 711 712 /* Apply whitening on ciphertext */ 713 sg = crypt_get_sg_data(cc, dmreq->sg_out); 714 dst = kmap_local_page(sg_page(sg)); 715 r = crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset); 716 kunmap_local(dst); 717 718 return r; 719 } 720 721 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv, 722 struct dm_crypt_request *dmreq) 723 { 724 /* Used only for writes, there must be an additional space to store IV */ 725 get_random_bytes(iv, cc->iv_size); 726 return 0; 727 } 728 729 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti, 730 const char *opts) 731 { 732 if (crypt_integrity_aead(cc)) { 733 ti->error = "AEAD transforms not supported for EBOIV"; 734 return -EINVAL; 735 } 736 737 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) { 738 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher"; 739 return -EINVAL; 740 } 741 742 return 0; 743 } 744 745 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv, 746 struct dm_crypt_request *dmreq) 747 { 748 struct crypto_skcipher *tfm = any_tfm(cc); 749 struct skcipher_request *req; 750 struct scatterlist src, dst; 751 DECLARE_CRYPTO_WAIT(wait); 752 unsigned int reqsize; 753 int err; 754 u8 *buf; 755 756 reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm); 757 reqsize = ALIGN(reqsize, __alignof__(__le64)); 758 759 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO); 760 if (!req) 761 return -ENOMEM; 762 763 skcipher_request_set_tfm(req, tfm); 764 765 buf = (u8 *)req + reqsize; 766 memset(buf, 0, cc->iv_size); 767 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 768 769 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size); 770 sg_init_one(&dst, iv, cc->iv_size); 771 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf); 772 skcipher_request_set_callback(req, 0, crypto_req_done, &wait); 773 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 774 kfree_sensitive(req); 775 776 return err; 777 } 778 779 static void crypt_iv_elephant_dtr(struct crypt_config *cc) 780 { 781 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 782 783 crypto_free_skcipher(elephant->tfm); 784 elephant->tfm = NULL; 785 } 786 787 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti, 788 const char *opts) 789 { 790 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 791 int r; 792 793 elephant->tfm = crypto_alloc_skcipher("ecb(aes)", 0, 794 CRYPTO_ALG_ALLOCATES_MEMORY); 795 if (IS_ERR(elephant->tfm)) { 796 r = PTR_ERR(elephant->tfm); 797 elephant->tfm = NULL; 798 return r; 799 } 800 801 r = crypt_iv_eboiv_ctr(cc, ti, NULL); 802 if (r) 803 crypt_iv_elephant_dtr(cc); 804 return r; 805 } 806 807 static void diffuser_disk_to_cpu(u32 *d, size_t n) 808 { 809 #ifndef __LITTLE_ENDIAN 810 int i; 811 812 for (i = 0; i < n; i++) 813 d[i] = le32_to_cpu((__le32)d[i]); 814 #endif 815 } 816 817 static void diffuser_cpu_to_disk(__le32 *d, size_t n) 818 { 819 #ifndef __LITTLE_ENDIAN 820 int i; 821 822 for (i = 0; i < n; i++) 823 d[i] = cpu_to_le32((u32)d[i]); 824 #endif 825 } 826 827 static void diffuser_a_decrypt(u32 *d, size_t n) 828 { 829 int i, i1, i2, i3; 830 831 for (i = 0; i < 5; i++) { 832 i1 = 0; 833 i2 = n - 2; 834 i3 = n - 5; 835 836 while (i1 < (n - 1)) { 837 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 838 i1++; i2++; i3++; 839 840 if (i3 >= n) 841 i3 -= n; 842 843 d[i1] += d[i2] ^ d[i3]; 844 i1++; i2++; i3++; 845 846 if (i2 >= n) 847 i2 -= n; 848 849 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 850 i1++; i2++; i3++; 851 852 d[i1] += d[i2] ^ d[i3]; 853 i1++; i2++; i3++; 854 } 855 } 856 } 857 858 static void diffuser_a_encrypt(u32 *d, size_t n) 859 { 860 int i, i1, i2, i3; 861 862 for (i = 0; i < 5; i++) { 863 i1 = n - 1; 864 i2 = n - 2 - 1; 865 i3 = n - 5 - 1; 866 867 while (i1 > 0) { 868 d[i1] -= d[i2] ^ d[i3]; 869 i1--; i2--; i3--; 870 871 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 872 i1--; i2--; i3--; 873 874 if (i2 < 0) 875 i2 += n; 876 877 d[i1] -= d[i2] ^ d[i3]; 878 i1--; i2--; i3--; 879 880 if (i3 < 0) 881 i3 += n; 882 883 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 884 i1--; i2--; i3--; 885 } 886 } 887 } 888 889 static void diffuser_b_decrypt(u32 *d, size_t n) 890 { 891 int i, i1, i2, i3; 892 893 for (i = 0; i < 3; i++) { 894 i1 = 0; 895 i2 = 2; 896 i3 = 5; 897 898 while (i1 < (n - 1)) { 899 d[i1] += d[i2] ^ d[i3]; 900 i1++; i2++; i3++; 901 902 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 903 i1++; i2++; i3++; 904 905 if (i2 >= n) 906 i2 -= n; 907 908 d[i1] += d[i2] ^ d[i3]; 909 i1++; i2++; i3++; 910 911 if (i3 >= n) 912 i3 -= n; 913 914 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 915 i1++; i2++; i3++; 916 } 917 } 918 } 919 920 static void diffuser_b_encrypt(u32 *d, size_t n) 921 { 922 int i, i1, i2, i3; 923 924 for (i = 0; i < 3; i++) { 925 i1 = n - 1; 926 i2 = 2 - 1; 927 i3 = 5 - 1; 928 929 while (i1 > 0) { 930 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 931 i1--; i2--; i3--; 932 933 if (i3 < 0) 934 i3 += n; 935 936 d[i1] -= d[i2] ^ d[i3]; 937 i1--; i2--; i3--; 938 939 if (i2 < 0) 940 i2 += n; 941 942 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 943 i1--; i2--; i3--; 944 945 d[i1] -= d[i2] ^ d[i3]; 946 i1--; i2--; i3--; 947 } 948 } 949 } 950 951 static int crypt_iv_elephant(struct crypt_config *cc, struct dm_crypt_request *dmreq) 952 { 953 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 954 u8 *es, *ks, *data, *data2, *data_offset; 955 struct skcipher_request *req; 956 struct scatterlist *sg, *sg2, src, dst; 957 DECLARE_CRYPTO_WAIT(wait); 958 int i, r; 959 960 req = skcipher_request_alloc(elephant->tfm, GFP_NOIO); 961 es = kzalloc(16, GFP_NOIO); /* Key for AES */ 962 ks = kzalloc(32, GFP_NOIO); /* Elephant sector key */ 963 964 if (!req || !es || !ks) { 965 r = -ENOMEM; 966 goto out; 967 } 968 969 *(__le64 *)es = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 970 971 /* E(Ks, e(s)) */ 972 sg_init_one(&src, es, 16); 973 sg_init_one(&dst, ks, 16); 974 skcipher_request_set_crypt(req, &src, &dst, 16, NULL); 975 skcipher_request_set_callback(req, 0, crypto_req_done, &wait); 976 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 977 if (r) 978 goto out; 979 980 /* E(Ks, e'(s)) */ 981 es[15] = 0x80; 982 sg_init_one(&dst, &ks[16], 16); 983 r = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 984 if (r) 985 goto out; 986 987 sg = crypt_get_sg_data(cc, dmreq->sg_out); 988 data = kmap_local_page(sg_page(sg)); 989 data_offset = data + sg->offset; 990 991 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */ 992 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 993 sg2 = crypt_get_sg_data(cc, dmreq->sg_in); 994 data2 = kmap_local_page(sg_page(sg2)); 995 memcpy(data_offset, data2 + sg2->offset, cc->sector_size); 996 kunmap_local(data2); 997 } 998 999 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 1000 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1001 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1002 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1003 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32)); 1004 } 1005 1006 for (i = 0; i < (cc->sector_size / 32); i++) 1007 crypto_xor(data_offset + i * 32, ks, 32); 1008 1009 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 1010 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1011 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1012 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 1013 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32)); 1014 } 1015 1016 kunmap_local(data); 1017 out: 1018 kfree_sensitive(ks); 1019 kfree_sensitive(es); 1020 skcipher_request_free(req); 1021 return r; 1022 } 1023 1024 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv, 1025 struct dm_crypt_request *dmreq) 1026 { 1027 int r; 1028 1029 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 1030 r = crypt_iv_elephant(cc, dmreq); 1031 if (r) 1032 return r; 1033 } 1034 1035 return crypt_iv_eboiv_gen(cc, iv, dmreq); 1036 } 1037 1038 static int crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv, 1039 struct dm_crypt_request *dmreq) 1040 { 1041 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 1042 return crypt_iv_elephant(cc, dmreq); 1043 1044 return 0; 1045 } 1046 1047 static int crypt_iv_elephant_init(struct crypt_config *cc) 1048 { 1049 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1050 int key_offset = cc->key_size - cc->key_extra_size; 1051 1052 return crypto_skcipher_setkey(elephant->tfm, &cc->key[key_offset], cc->key_extra_size); 1053 } 1054 1055 static int crypt_iv_elephant_wipe(struct crypt_config *cc) 1056 { 1057 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1058 u8 key[ELEPHANT_MAX_KEY_SIZE]; 1059 1060 memset(key, 0, cc->key_extra_size); 1061 return crypto_skcipher_setkey(elephant->tfm, key, cc->key_extra_size); 1062 } 1063 1064 static const struct crypt_iv_operations crypt_iv_plain_ops = { 1065 .generator = crypt_iv_plain_gen 1066 }; 1067 1068 static const struct crypt_iv_operations crypt_iv_plain64_ops = { 1069 .generator = crypt_iv_plain64_gen 1070 }; 1071 1072 static const struct crypt_iv_operations crypt_iv_plain64be_ops = { 1073 .generator = crypt_iv_plain64be_gen 1074 }; 1075 1076 static const struct crypt_iv_operations crypt_iv_essiv_ops = { 1077 .generator = crypt_iv_essiv_gen 1078 }; 1079 1080 static const struct crypt_iv_operations crypt_iv_benbi_ops = { 1081 .ctr = crypt_iv_benbi_ctr, 1082 .dtr = crypt_iv_benbi_dtr, 1083 .generator = crypt_iv_benbi_gen 1084 }; 1085 1086 static const struct crypt_iv_operations crypt_iv_null_ops = { 1087 .generator = crypt_iv_null_gen 1088 }; 1089 1090 static const struct crypt_iv_operations crypt_iv_lmk_ops = { 1091 .ctr = crypt_iv_lmk_ctr, 1092 .dtr = crypt_iv_lmk_dtr, 1093 .init = crypt_iv_lmk_init, 1094 .wipe = crypt_iv_lmk_wipe, 1095 .generator = crypt_iv_lmk_gen, 1096 .post = crypt_iv_lmk_post 1097 }; 1098 1099 static const struct crypt_iv_operations crypt_iv_tcw_ops = { 1100 .ctr = crypt_iv_tcw_ctr, 1101 .dtr = crypt_iv_tcw_dtr, 1102 .init = crypt_iv_tcw_init, 1103 .wipe = crypt_iv_tcw_wipe, 1104 .generator = crypt_iv_tcw_gen, 1105 .post = crypt_iv_tcw_post 1106 }; 1107 1108 static const struct crypt_iv_operations crypt_iv_random_ops = { 1109 .generator = crypt_iv_random_gen 1110 }; 1111 1112 static const struct crypt_iv_operations crypt_iv_eboiv_ops = { 1113 .ctr = crypt_iv_eboiv_ctr, 1114 .generator = crypt_iv_eboiv_gen 1115 }; 1116 1117 static const struct crypt_iv_operations crypt_iv_elephant_ops = { 1118 .ctr = crypt_iv_elephant_ctr, 1119 .dtr = crypt_iv_elephant_dtr, 1120 .init = crypt_iv_elephant_init, 1121 .wipe = crypt_iv_elephant_wipe, 1122 .generator = crypt_iv_elephant_gen, 1123 .post = crypt_iv_elephant_post 1124 }; 1125 1126 /* 1127 * Integrity extensions 1128 */ 1129 static bool crypt_integrity_aead(struct crypt_config *cc) 1130 { 1131 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 1132 } 1133 1134 static bool crypt_integrity_hmac(struct crypt_config *cc) 1135 { 1136 return crypt_integrity_aead(cc) && cc->key_mac_size; 1137 } 1138 1139 /* Get sg containing data */ 1140 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 1141 struct scatterlist *sg) 1142 { 1143 if (unlikely(crypt_integrity_aead(cc))) 1144 return &sg[2]; 1145 1146 return sg; 1147 } 1148 1149 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio) 1150 { 1151 struct bio_integrity_payload *bip; 1152 unsigned int tag_len; 1153 int ret; 1154 1155 if (!bio_sectors(bio) || !io->cc->on_disk_tag_size) 1156 return 0; 1157 1158 bip = bio_integrity_alloc(bio, GFP_NOIO, 1); 1159 if (IS_ERR(bip)) 1160 return PTR_ERR(bip); 1161 1162 tag_len = io->cc->on_disk_tag_size * (bio_sectors(bio) >> io->cc->sector_shift); 1163 1164 bip->bip_iter.bi_sector = io->cc->start + io->sector; 1165 1166 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata), 1167 tag_len, offset_in_page(io->integrity_metadata)); 1168 if (unlikely(ret != tag_len)) 1169 return -ENOMEM; 1170 1171 return 0; 1172 } 1173 1174 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti) 1175 { 1176 #ifdef CONFIG_BLK_DEV_INTEGRITY 1177 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk); 1178 struct mapped_device *md = dm_table_get_md(ti->table); 1179 1180 /* From now we require underlying device with our integrity profile */ 1181 if (!bi || strcasecmp(bi->profile->name, "DM-DIF-EXT-TAG")) { 1182 ti->error = "Integrity profile not supported."; 1183 return -EINVAL; 1184 } 1185 1186 if (bi->tag_size != cc->on_disk_tag_size || 1187 bi->tuple_size != cc->on_disk_tag_size) { 1188 ti->error = "Integrity profile tag size mismatch."; 1189 return -EINVAL; 1190 } 1191 if (1 << bi->interval_exp != cc->sector_size) { 1192 ti->error = "Integrity profile sector size mismatch."; 1193 return -EINVAL; 1194 } 1195 1196 if (crypt_integrity_aead(cc)) { 1197 cc->integrity_tag_size = cc->on_disk_tag_size - cc->integrity_iv_size; 1198 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md), 1199 cc->integrity_tag_size, cc->integrity_iv_size); 1200 1201 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) { 1202 ti->error = "Integrity AEAD auth tag size is not supported."; 1203 return -EINVAL; 1204 } 1205 } else if (cc->integrity_iv_size) 1206 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md), 1207 cc->integrity_iv_size); 1208 1209 if ((cc->integrity_tag_size + cc->integrity_iv_size) != bi->tag_size) { 1210 ti->error = "Not enough space for integrity tag in the profile."; 1211 return -EINVAL; 1212 } 1213 1214 return 0; 1215 #else 1216 ti->error = "Integrity profile not supported."; 1217 return -EINVAL; 1218 #endif 1219 } 1220 1221 static void crypt_convert_init(struct crypt_config *cc, 1222 struct convert_context *ctx, 1223 struct bio *bio_out, struct bio *bio_in, 1224 sector_t sector) 1225 { 1226 ctx->bio_in = bio_in; 1227 ctx->bio_out = bio_out; 1228 if (bio_in) 1229 ctx->iter_in = bio_in->bi_iter; 1230 if (bio_out) 1231 ctx->iter_out = bio_out->bi_iter; 1232 ctx->cc_sector = sector + cc->iv_offset; 1233 init_completion(&ctx->restart); 1234 } 1235 1236 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc, 1237 void *req) 1238 { 1239 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start); 1240 } 1241 1242 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq) 1243 { 1244 return (void *)((char *)dmreq - cc->dmreq_start); 1245 } 1246 1247 static u8 *iv_of_dmreq(struct crypt_config *cc, 1248 struct dm_crypt_request *dmreq) 1249 { 1250 if (crypt_integrity_aead(cc)) 1251 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1252 crypto_aead_alignmask(any_tfm_aead(cc)) + 1); 1253 else 1254 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1255 crypto_skcipher_alignmask(any_tfm(cc)) + 1); 1256 } 1257 1258 static u8 *org_iv_of_dmreq(struct crypt_config *cc, 1259 struct dm_crypt_request *dmreq) 1260 { 1261 return iv_of_dmreq(cc, dmreq) + cc->iv_size; 1262 } 1263 1264 static __le64 *org_sector_of_dmreq(struct crypt_config *cc, 1265 struct dm_crypt_request *dmreq) 1266 { 1267 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size; 1268 1269 return (__le64 *) ptr; 1270 } 1271 1272 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc, 1273 struct dm_crypt_request *dmreq) 1274 { 1275 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + 1276 cc->iv_size + sizeof(uint64_t); 1277 1278 return (unsigned int *)ptr; 1279 } 1280 1281 static void *tag_from_dmreq(struct crypt_config *cc, 1282 struct dm_crypt_request *dmreq) 1283 { 1284 struct convert_context *ctx = dmreq->ctx; 1285 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 1286 1287 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) * 1288 cc->on_disk_tag_size]; 1289 } 1290 1291 static void *iv_tag_from_dmreq(struct crypt_config *cc, 1292 struct dm_crypt_request *dmreq) 1293 { 1294 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size; 1295 } 1296 1297 static int crypt_convert_block_aead(struct crypt_config *cc, 1298 struct convert_context *ctx, 1299 struct aead_request *req, 1300 unsigned int tag_offset) 1301 { 1302 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1303 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1304 struct dm_crypt_request *dmreq; 1305 u8 *iv, *org_iv, *tag_iv, *tag; 1306 __le64 *sector; 1307 int r = 0; 1308 1309 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size); 1310 1311 /* Reject unexpected unaligned bio. */ 1312 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1313 return -EIO; 1314 1315 dmreq = dmreq_of_req(cc, req); 1316 dmreq->iv_sector = ctx->cc_sector; 1317 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1318 dmreq->iv_sector >>= cc->sector_shift; 1319 dmreq->ctx = ctx; 1320 1321 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1322 1323 sector = org_sector_of_dmreq(cc, dmreq); 1324 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1325 1326 iv = iv_of_dmreq(cc, dmreq); 1327 org_iv = org_iv_of_dmreq(cc, dmreq); 1328 tag = tag_from_dmreq(cc, dmreq); 1329 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1330 1331 /* AEAD request: 1332 * |----- AAD -------|------ DATA -------|-- AUTH TAG --| 1333 * | (authenticated) | (auth+encryption) | | 1334 * | sector_LE | IV | sector in/out | tag in/out | 1335 */ 1336 sg_init_table(dmreq->sg_in, 4); 1337 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t)); 1338 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size); 1339 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1340 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size); 1341 1342 sg_init_table(dmreq->sg_out, 4); 1343 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t)); 1344 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size); 1345 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1346 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size); 1347 1348 if (cc->iv_gen_ops) { 1349 /* For READs use IV stored in integrity metadata */ 1350 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1351 memcpy(org_iv, tag_iv, cc->iv_size); 1352 } else { 1353 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1354 if (r < 0) 1355 return r; 1356 /* Store generated IV in integrity metadata */ 1357 if (cc->integrity_iv_size) 1358 memcpy(tag_iv, org_iv, cc->iv_size); 1359 } 1360 /* Working copy of IV, to be modified in crypto API */ 1361 memcpy(iv, org_iv, cc->iv_size); 1362 } 1363 1364 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size); 1365 if (bio_data_dir(ctx->bio_in) == WRITE) { 1366 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1367 cc->sector_size, iv); 1368 r = crypto_aead_encrypt(req); 1369 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->on_disk_tag_size) 1370 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0, 1371 cc->on_disk_tag_size - (cc->integrity_tag_size + cc->integrity_iv_size)); 1372 } else { 1373 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1374 cc->sector_size + cc->integrity_tag_size, iv); 1375 r = crypto_aead_decrypt(req); 1376 } 1377 1378 if (r == -EBADMSG) { 1379 sector_t s = le64_to_cpu(*sector); 1380 1381 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu", 1382 ctx->bio_in->bi_bdev, s); 1383 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead", 1384 ctx->bio_in, s, 0); 1385 } 1386 1387 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1388 r = cc->iv_gen_ops->post(cc, org_iv, dmreq); 1389 1390 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1391 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1392 1393 return r; 1394 } 1395 1396 static int crypt_convert_block_skcipher(struct crypt_config *cc, 1397 struct convert_context *ctx, 1398 struct skcipher_request *req, 1399 unsigned int tag_offset) 1400 { 1401 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1402 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1403 struct scatterlist *sg_in, *sg_out; 1404 struct dm_crypt_request *dmreq; 1405 u8 *iv, *org_iv, *tag_iv; 1406 __le64 *sector; 1407 int r = 0; 1408 1409 /* Reject unexpected unaligned bio. */ 1410 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1411 return -EIO; 1412 1413 dmreq = dmreq_of_req(cc, req); 1414 dmreq->iv_sector = ctx->cc_sector; 1415 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1416 dmreq->iv_sector >>= cc->sector_shift; 1417 dmreq->ctx = ctx; 1418 1419 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1420 1421 iv = iv_of_dmreq(cc, dmreq); 1422 org_iv = org_iv_of_dmreq(cc, dmreq); 1423 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1424 1425 sector = org_sector_of_dmreq(cc, dmreq); 1426 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1427 1428 /* For skcipher we use only the first sg item */ 1429 sg_in = &dmreq->sg_in[0]; 1430 sg_out = &dmreq->sg_out[0]; 1431 1432 sg_init_table(sg_in, 1); 1433 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1434 1435 sg_init_table(sg_out, 1); 1436 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1437 1438 if (cc->iv_gen_ops) { 1439 /* For READs use IV stored in integrity metadata */ 1440 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1441 memcpy(org_iv, tag_iv, cc->integrity_iv_size); 1442 } else { 1443 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1444 if (r < 0) 1445 return r; 1446 /* Data can be already preprocessed in generator */ 1447 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags)) 1448 sg_in = sg_out; 1449 /* Store generated IV in integrity metadata */ 1450 if (cc->integrity_iv_size) 1451 memcpy(tag_iv, org_iv, cc->integrity_iv_size); 1452 } 1453 /* Working copy of IV, to be modified in crypto API */ 1454 memcpy(iv, org_iv, cc->iv_size); 1455 } 1456 1457 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv); 1458 1459 if (bio_data_dir(ctx->bio_in) == WRITE) 1460 r = crypto_skcipher_encrypt(req); 1461 else 1462 r = crypto_skcipher_decrypt(req); 1463 1464 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1465 r = cc->iv_gen_ops->post(cc, org_iv, dmreq); 1466 1467 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1468 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1469 1470 return r; 1471 } 1472 1473 static void kcryptd_async_done(void *async_req, int error); 1474 1475 static int crypt_alloc_req_skcipher(struct crypt_config *cc, 1476 struct convert_context *ctx) 1477 { 1478 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1); 1479 1480 if (!ctx->r.req) { 1481 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1482 if (!ctx->r.req) 1483 return -ENOMEM; 1484 } 1485 1486 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]); 1487 1488 /* 1489 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1490 * requests if driver request queue is full. 1491 */ 1492 skcipher_request_set_callback(ctx->r.req, 1493 CRYPTO_TFM_REQ_MAY_BACKLOG, 1494 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req)); 1495 1496 return 0; 1497 } 1498 1499 static int crypt_alloc_req_aead(struct crypt_config *cc, 1500 struct convert_context *ctx) 1501 { 1502 if (!ctx->r.req_aead) { 1503 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1504 if (!ctx->r.req_aead) 1505 return -ENOMEM; 1506 } 1507 1508 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]); 1509 1510 /* 1511 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1512 * requests if driver request queue is full. 1513 */ 1514 aead_request_set_callback(ctx->r.req_aead, 1515 CRYPTO_TFM_REQ_MAY_BACKLOG, 1516 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead)); 1517 1518 return 0; 1519 } 1520 1521 static int crypt_alloc_req(struct crypt_config *cc, 1522 struct convert_context *ctx) 1523 { 1524 if (crypt_integrity_aead(cc)) 1525 return crypt_alloc_req_aead(cc, ctx); 1526 else 1527 return crypt_alloc_req_skcipher(cc, ctx); 1528 } 1529 1530 static void crypt_free_req_skcipher(struct crypt_config *cc, 1531 struct skcipher_request *req, struct bio *base_bio) 1532 { 1533 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1534 1535 if ((struct skcipher_request *)(io + 1) != req) 1536 mempool_free(req, &cc->req_pool); 1537 } 1538 1539 static void crypt_free_req_aead(struct crypt_config *cc, 1540 struct aead_request *req, struct bio *base_bio) 1541 { 1542 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1543 1544 if ((struct aead_request *)(io + 1) != req) 1545 mempool_free(req, &cc->req_pool); 1546 } 1547 1548 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio) 1549 { 1550 if (crypt_integrity_aead(cc)) 1551 crypt_free_req_aead(cc, req, base_bio); 1552 else 1553 crypt_free_req_skcipher(cc, req, base_bio); 1554 } 1555 1556 /* 1557 * Encrypt / decrypt data from one bio to another one (can be the same one) 1558 */ 1559 static blk_status_t crypt_convert(struct crypt_config *cc, 1560 struct convert_context *ctx, bool atomic, bool reset_pending) 1561 { 1562 unsigned int tag_offset = 0; 1563 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT; 1564 int r; 1565 1566 /* 1567 * if reset_pending is set we are dealing with the bio for the first time, 1568 * else we're continuing to work on the previous bio, so don't mess with 1569 * the cc_pending counter 1570 */ 1571 if (reset_pending) 1572 atomic_set(&ctx->cc_pending, 1); 1573 1574 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) { 1575 1576 r = crypt_alloc_req(cc, ctx); 1577 if (r) { 1578 complete(&ctx->restart); 1579 return BLK_STS_DEV_RESOURCE; 1580 } 1581 1582 atomic_inc(&ctx->cc_pending); 1583 1584 if (crypt_integrity_aead(cc)) 1585 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, tag_offset); 1586 else 1587 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, tag_offset); 1588 1589 switch (r) { 1590 /* 1591 * The request was queued by a crypto driver 1592 * but the driver request queue is full, let's wait. 1593 */ 1594 case -EBUSY: 1595 if (in_interrupt()) { 1596 if (try_wait_for_completion(&ctx->restart)) { 1597 /* 1598 * we don't have to block to wait for completion, 1599 * so proceed 1600 */ 1601 } else { 1602 /* 1603 * we can't wait for completion without blocking 1604 * exit and continue processing in a workqueue 1605 */ 1606 ctx->r.req = NULL; 1607 ctx->cc_sector += sector_step; 1608 tag_offset++; 1609 return BLK_STS_DEV_RESOURCE; 1610 } 1611 } else { 1612 wait_for_completion(&ctx->restart); 1613 } 1614 reinit_completion(&ctx->restart); 1615 fallthrough; 1616 /* 1617 * The request is queued and processed asynchronously, 1618 * completion function kcryptd_async_done() will be called. 1619 */ 1620 case -EINPROGRESS: 1621 ctx->r.req = NULL; 1622 ctx->cc_sector += sector_step; 1623 tag_offset++; 1624 continue; 1625 /* 1626 * The request was already processed (synchronously). 1627 */ 1628 case 0: 1629 atomic_dec(&ctx->cc_pending); 1630 ctx->cc_sector += sector_step; 1631 tag_offset++; 1632 if (!atomic) 1633 cond_resched(); 1634 continue; 1635 /* 1636 * There was a data integrity error. 1637 */ 1638 case -EBADMSG: 1639 atomic_dec(&ctx->cc_pending); 1640 return BLK_STS_PROTECTION; 1641 /* 1642 * There was an error while processing the request. 1643 */ 1644 default: 1645 atomic_dec(&ctx->cc_pending); 1646 return BLK_STS_IOERR; 1647 } 1648 } 1649 1650 return 0; 1651 } 1652 1653 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone); 1654 1655 /* 1656 * Generate a new unfragmented bio with the given size 1657 * This should never violate the device limitations (but only because 1658 * max_segment_size is being constrained to PAGE_SIZE). 1659 * 1660 * This function may be called concurrently. If we allocate from the mempool 1661 * concurrently, there is a possibility of deadlock. For example, if we have 1662 * mempool of 256 pages, two processes, each wanting 256, pages allocate from 1663 * the mempool concurrently, it may deadlock in a situation where both processes 1664 * have allocated 128 pages and the mempool is exhausted. 1665 * 1666 * In order to avoid this scenario we allocate the pages under a mutex. 1667 * 1668 * In order to not degrade performance with excessive locking, we try 1669 * non-blocking allocations without a mutex first but on failure we fallback 1670 * to blocking allocations with a mutex. 1671 * 1672 * In order to reduce allocation overhead, we try to allocate compound pages in 1673 * the first pass. If they are not available, we fall back to the mempool. 1674 */ 1675 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size) 1676 { 1677 struct crypt_config *cc = io->cc; 1678 struct bio *clone; 1679 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 1680 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM; 1681 unsigned int remaining_size; 1682 unsigned int order = MAX_ORDER; 1683 1684 retry: 1685 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1686 mutex_lock(&cc->bio_alloc_lock); 1687 1688 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf, 1689 GFP_NOIO, &cc->bs); 1690 clone->bi_private = io; 1691 clone->bi_end_io = crypt_endio; 1692 1693 remaining_size = size; 1694 1695 while (remaining_size) { 1696 struct page *pages; 1697 unsigned size_to_add; 1698 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT); 1699 order = min(order, remaining_order); 1700 1701 while (order > 0) { 1702 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) + 1703 (1 << order) > dm_crypt_pages_per_client)) 1704 goto decrease_order; 1705 pages = alloc_pages(gfp_mask 1706 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP, 1707 order); 1708 if (likely(pages != NULL)) { 1709 percpu_counter_add(&cc->n_allocated_pages, 1 << order); 1710 goto have_pages; 1711 } 1712 decrease_order: 1713 order--; 1714 } 1715 1716 pages = mempool_alloc(&cc->page_pool, gfp_mask); 1717 if (!pages) { 1718 crypt_free_buffer_pages(cc, clone); 1719 bio_put(clone); 1720 gfp_mask |= __GFP_DIRECT_RECLAIM; 1721 order = 0; 1722 goto retry; 1723 } 1724 1725 have_pages: 1726 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size); 1727 __bio_add_page(clone, pages, size_to_add, 0); 1728 remaining_size -= size_to_add; 1729 } 1730 1731 /* Allocate space for integrity tags */ 1732 if (dm_crypt_integrity_io_alloc(io, clone)) { 1733 crypt_free_buffer_pages(cc, clone); 1734 bio_put(clone); 1735 clone = NULL; 1736 } 1737 1738 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1739 mutex_unlock(&cc->bio_alloc_lock); 1740 1741 return clone; 1742 } 1743 1744 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone) 1745 { 1746 struct folio_iter fi; 1747 1748 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */ 1749 bio_for_each_folio_all(fi, clone) { 1750 if (folio_test_large(fi.folio)) { 1751 percpu_counter_sub(&cc->n_allocated_pages, 1752 1 << folio_order(fi.folio)); 1753 folio_put(fi.folio); 1754 } else { 1755 mempool_free(&fi.folio->page, &cc->page_pool); 1756 } 1757 } 1758 } 1759 } 1760 1761 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc, 1762 struct bio *bio, sector_t sector) 1763 { 1764 io->cc = cc; 1765 io->base_bio = bio; 1766 io->sector = sector; 1767 io->error = 0; 1768 io->ctx.r.req = NULL; 1769 io->integrity_metadata = NULL; 1770 io->integrity_metadata_from_pool = false; 1771 io->in_tasklet = false; 1772 atomic_set(&io->io_pending, 0); 1773 } 1774 1775 static void crypt_inc_pending(struct dm_crypt_io *io) 1776 { 1777 atomic_inc(&io->io_pending); 1778 } 1779 1780 static void kcryptd_io_bio_endio(struct work_struct *work) 1781 { 1782 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 1783 1784 bio_endio(io->base_bio); 1785 } 1786 1787 /* 1788 * One of the bios was finished. Check for completion of 1789 * the whole request and correctly clean up the buffer. 1790 */ 1791 static void crypt_dec_pending(struct dm_crypt_io *io) 1792 { 1793 struct crypt_config *cc = io->cc; 1794 struct bio *base_bio = io->base_bio; 1795 blk_status_t error = io->error; 1796 1797 if (!atomic_dec_and_test(&io->io_pending)) 1798 return; 1799 1800 if (io->ctx.r.req) 1801 crypt_free_req(cc, io->ctx.r.req, base_bio); 1802 1803 if (unlikely(io->integrity_metadata_from_pool)) 1804 mempool_free(io->integrity_metadata, &io->cc->tag_pool); 1805 else 1806 kfree(io->integrity_metadata); 1807 1808 base_bio->bi_status = error; 1809 1810 /* 1811 * If we are running this function from our tasklet, 1812 * we can't call bio_endio() here, because it will call 1813 * clone_endio() from dm.c, which in turn will 1814 * free the current struct dm_crypt_io structure with 1815 * our tasklet. In this case we need to delay bio_endio() 1816 * execution to after the tasklet is done and dequeued. 1817 */ 1818 if (io->in_tasklet) { 1819 INIT_WORK(&io->work, kcryptd_io_bio_endio); 1820 queue_work(cc->io_queue, &io->work); 1821 return; 1822 } 1823 1824 bio_endio(base_bio); 1825 } 1826 1827 /* 1828 * kcryptd/kcryptd_io: 1829 * 1830 * Needed because it would be very unwise to do decryption in an 1831 * interrupt context. 1832 * 1833 * kcryptd performs the actual encryption or decryption. 1834 * 1835 * kcryptd_io performs the IO submission. 1836 * 1837 * They must be separated as otherwise the final stages could be 1838 * starved by new requests which can block in the first stages due 1839 * to memory allocation. 1840 * 1841 * The work is done per CPU global for all dm-crypt instances. 1842 * They should not depend on each other and do not block. 1843 */ 1844 static void crypt_endio(struct bio *clone) 1845 { 1846 struct dm_crypt_io *io = clone->bi_private; 1847 struct crypt_config *cc = io->cc; 1848 unsigned int rw = bio_data_dir(clone); 1849 blk_status_t error; 1850 1851 /* 1852 * free the processed pages 1853 */ 1854 if (rw == WRITE) 1855 crypt_free_buffer_pages(cc, clone); 1856 1857 error = clone->bi_status; 1858 bio_put(clone); 1859 1860 if (rw == READ && !error) { 1861 kcryptd_queue_crypt(io); 1862 return; 1863 } 1864 1865 if (unlikely(error)) 1866 io->error = error; 1867 1868 crypt_dec_pending(io); 1869 } 1870 1871 #define CRYPT_MAP_READ_GFP GFP_NOWAIT 1872 1873 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp) 1874 { 1875 struct crypt_config *cc = io->cc; 1876 struct bio *clone; 1877 1878 /* 1879 * We need the original biovec array in order to decrypt the whole bio 1880 * data *afterwards* -- thanks to immutable biovecs we don't need to 1881 * worry about the block layer modifying the biovec array; so leverage 1882 * bio_alloc_clone(). 1883 */ 1884 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs); 1885 if (!clone) 1886 return 1; 1887 clone->bi_private = io; 1888 clone->bi_end_io = crypt_endio; 1889 1890 crypt_inc_pending(io); 1891 1892 clone->bi_iter.bi_sector = cc->start + io->sector; 1893 1894 if (dm_crypt_integrity_io_alloc(io, clone)) { 1895 crypt_dec_pending(io); 1896 bio_put(clone); 1897 return 1; 1898 } 1899 1900 dm_submit_bio_remap(io->base_bio, clone); 1901 return 0; 1902 } 1903 1904 static void kcryptd_io_read_work(struct work_struct *work) 1905 { 1906 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 1907 1908 crypt_inc_pending(io); 1909 if (kcryptd_io_read(io, GFP_NOIO)) 1910 io->error = BLK_STS_RESOURCE; 1911 crypt_dec_pending(io); 1912 } 1913 1914 static void kcryptd_queue_read(struct dm_crypt_io *io) 1915 { 1916 struct crypt_config *cc = io->cc; 1917 1918 INIT_WORK(&io->work, kcryptd_io_read_work); 1919 queue_work(cc->io_queue, &io->work); 1920 } 1921 1922 static void kcryptd_io_write(struct dm_crypt_io *io) 1923 { 1924 struct bio *clone = io->ctx.bio_out; 1925 1926 dm_submit_bio_remap(io->base_bio, clone); 1927 } 1928 1929 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node) 1930 1931 static int dmcrypt_write(void *data) 1932 { 1933 struct crypt_config *cc = data; 1934 struct dm_crypt_io *io; 1935 1936 while (1) { 1937 struct rb_root write_tree; 1938 struct blk_plug plug; 1939 1940 spin_lock_irq(&cc->write_thread_lock); 1941 continue_locked: 1942 1943 if (!RB_EMPTY_ROOT(&cc->write_tree)) 1944 goto pop_from_list; 1945 1946 set_current_state(TASK_INTERRUPTIBLE); 1947 1948 spin_unlock_irq(&cc->write_thread_lock); 1949 1950 if (unlikely(kthread_should_stop())) { 1951 set_current_state(TASK_RUNNING); 1952 break; 1953 } 1954 1955 schedule(); 1956 1957 set_current_state(TASK_RUNNING); 1958 spin_lock_irq(&cc->write_thread_lock); 1959 goto continue_locked; 1960 1961 pop_from_list: 1962 write_tree = cc->write_tree; 1963 cc->write_tree = RB_ROOT; 1964 spin_unlock_irq(&cc->write_thread_lock); 1965 1966 BUG_ON(rb_parent(write_tree.rb_node)); 1967 1968 /* 1969 * Note: we cannot walk the tree here with rb_next because 1970 * the structures may be freed when kcryptd_io_write is called. 1971 */ 1972 blk_start_plug(&plug); 1973 do { 1974 io = crypt_io_from_node(rb_first(&write_tree)); 1975 rb_erase(&io->rb_node, &write_tree); 1976 kcryptd_io_write(io); 1977 cond_resched(); 1978 } while (!RB_EMPTY_ROOT(&write_tree)); 1979 blk_finish_plug(&plug); 1980 } 1981 return 0; 1982 } 1983 1984 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async) 1985 { 1986 struct bio *clone = io->ctx.bio_out; 1987 struct crypt_config *cc = io->cc; 1988 unsigned long flags; 1989 sector_t sector; 1990 struct rb_node **rbp, *parent; 1991 1992 if (unlikely(io->error)) { 1993 crypt_free_buffer_pages(cc, clone); 1994 bio_put(clone); 1995 crypt_dec_pending(io); 1996 return; 1997 } 1998 1999 /* crypt_convert should have filled the clone bio */ 2000 BUG_ON(io->ctx.iter_out.bi_size); 2001 2002 clone->bi_iter.bi_sector = cc->start + io->sector; 2003 2004 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) || 2005 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) { 2006 dm_submit_bio_remap(io->base_bio, clone); 2007 return; 2008 } 2009 2010 spin_lock_irqsave(&cc->write_thread_lock, flags); 2011 if (RB_EMPTY_ROOT(&cc->write_tree)) 2012 wake_up_process(cc->write_thread); 2013 rbp = &cc->write_tree.rb_node; 2014 parent = NULL; 2015 sector = io->sector; 2016 while (*rbp) { 2017 parent = *rbp; 2018 if (sector < crypt_io_from_node(parent)->sector) 2019 rbp = &(*rbp)->rb_left; 2020 else 2021 rbp = &(*rbp)->rb_right; 2022 } 2023 rb_link_node(&io->rb_node, parent, rbp); 2024 rb_insert_color(&io->rb_node, &cc->write_tree); 2025 spin_unlock_irqrestore(&cc->write_thread_lock, flags); 2026 } 2027 2028 static bool kcryptd_crypt_write_inline(struct crypt_config *cc, 2029 struct convert_context *ctx) 2030 2031 { 2032 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags)) 2033 return false; 2034 2035 /* 2036 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering 2037 * constraints so they do not need to be issued inline by 2038 * kcryptd_crypt_write_convert(). 2039 */ 2040 switch (bio_op(ctx->bio_in)) { 2041 case REQ_OP_WRITE: 2042 case REQ_OP_WRITE_ZEROES: 2043 return true; 2044 default: 2045 return false; 2046 } 2047 } 2048 2049 static void kcryptd_crypt_write_continue(struct work_struct *work) 2050 { 2051 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2052 struct crypt_config *cc = io->cc; 2053 struct convert_context *ctx = &io->ctx; 2054 int crypt_finished; 2055 sector_t sector = io->sector; 2056 blk_status_t r; 2057 2058 wait_for_completion(&ctx->restart); 2059 reinit_completion(&ctx->restart); 2060 2061 r = crypt_convert(cc, &io->ctx, true, false); 2062 if (r) 2063 io->error = r; 2064 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2065 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2066 /* Wait for completion signaled by kcryptd_async_done() */ 2067 wait_for_completion(&ctx->restart); 2068 crypt_finished = 1; 2069 } 2070 2071 /* Encryption was already finished, submit io now */ 2072 if (crypt_finished) { 2073 kcryptd_crypt_write_io_submit(io, 0); 2074 io->sector = sector; 2075 } 2076 2077 crypt_dec_pending(io); 2078 } 2079 2080 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io) 2081 { 2082 struct crypt_config *cc = io->cc; 2083 struct convert_context *ctx = &io->ctx; 2084 struct bio *clone; 2085 int crypt_finished; 2086 sector_t sector = io->sector; 2087 blk_status_t r; 2088 2089 /* 2090 * Prevent io from disappearing until this function completes. 2091 */ 2092 crypt_inc_pending(io); 2093 crypt_convert_init(cc, ctx, NULL, io->base_bio, sector); 2094 2095 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size); 2096 if (unlikely(!clone)) { 2097 io->error = BLK_STS_IOERR; 2098 goto dec; 2099 } 2100 2101 io->ctx.bio_out = clone; 2102 io->ctx.iter_out = clone->bi_iter; 2103 2104 sector += bio_sectors(clone); 2105 2106 crypt_inc_pending(io); 2107 r = crypt_convert(cc, ctx, 2108 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true); 2109 /* 2110 * Crypto API backlogged the request, because its queue was full 2111 * and we're in softirq context, so continue from a workqueue 2112 * (TODO: is it actually possible to be in softirq in the write path?) 2113 */ 2114 if (r == BLK_STS_DEV_RESOURCE) { 2115 INIT_WORK(&io->work, kcryptd_crypt_write_continue); 2116 queue_work(cc->crypt_queue, &io->work); 2117 return; 2118 } 2119 if (r) 2120 io->error = r; 2121 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2122 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2123 /* Wait for completion signaled by kcryptd_async_done() */ 2124 wait_for_completion(&ctx->restart); 2125 crypt_finished = 1; 2126 } 2127 2128 /* Encryption was already finished, submit io now */ 2129 if (crypt_finished) { 2130 kcryptd_crypt_write_io_submit(io, 0); 2131 io->sector = sector; 2132 } 2133 2134 dec: 2135 crypt_dec_pending(io); 2136 } 2137 2138 static void kcryptd_crypt_read_done(struct dm_crypt_io *io) 2139 { 2140 crypt_dec_pending(io); 2141 } 2142 2143 static void kcryptd_crypt_read_continue(struct work_struct *work) 2144 { 2145 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2146 struct crypt_config *cc = io->cc; 2147 blk_status_t r; 2148 2149 wait_for_completion(&io->ctx.restart); 2150 reinit_completion(&io->ctx.restart); 2151 2152 r = crypt_convert(cc, &io->ctx, true, false); 2153 if (r) 2154 io->error = r; 2155 2156 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2157 kcryptd_crypt_read_done(io); 2158 2159 crypt_dec_pending(io); 2160 } 2161 2162 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io) 2163 { 2164 struct crypt_config *cc = io->cc; 2165 blk_status_t r; 2166 2167 crypt_inc_pending(io); 2168 2169 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio, 2170 io->sector); 2171 2172 r = crypt_convert(cc, &io->ctx, 2173 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true); 2174 /* 2175 * Crypto API backlogged the request, because its queue was full 2176 * and we're in softirq context, so continue from a workqueue 2177 */ 2178 if (r == BLK_STS_DEV_RESOURCE) { 2179 INIT_WORK(&io->work, kcryptd_crypt_read_continue); 2180 queue_work(cc->crypt_queue, &io->work); 2181 return; 2182 } 2183 if (r) 2184 io->error = r; 2185 2186 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2187 kcryptd_crypt_read_done(io); 2188 2189 crypt_dec_pending(io); 2190 } 2191 2192 static void kcryptd_async_done(void *data, int error) 2193 { 2194 struct dm_crypt_request *dmreq = data; 2195 struct convert_context *ctx = dmreq->ctx; 2196 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 2197 struct crypt_config *cc = io->cc; 2198 2199 /* 2200 * A request from crypto driver backlog is going to be processed now, 2201 * finish the completion and continue in crypt_convert(). 2202 * (Callback will be called for the second time for this request.) 2203 */ 2204 if (error == -EINPROGRESS) { 2205 complete(&ctx->restart); 2206 return; 2207 } 2208 2209 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post) 2210 error = cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq); 2211 2212 if (error == -EBADMSG) { 2213 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)); 2214 2215 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu", 2216 ctx->bio_in->bi_bdev, s); 2217 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead", 2218 ctx->bio_in, s, 0); 2219 io->error = BLK_STS_PROTECTION; 2220 } else if (error < 0) 2221 io->error = BLK_STS_IOERR; 2222 2223 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio); 2224 2225 if (!atomic_dec_and_test(&ctx->cc_pending)) 2226 return; 2227 2228 /* 2229 * The request is fully completed: for inline writes, let 2230 * kcryptd_crypt_write_convert() do the IO submission. 2231 */ 2232 if (bio_data_dir(io->base_bio) == READ) { 2233 kcryptd_crypt_read_done(io); 2234 return; 2235 } 2236 2237 if (kcryptd_crypt_write_inline(cc, ctx)) { 2238 complete(&ctx->restart); 2239 return; 2240 } 2241 2242 kcryptd_crypt_write_io_submit(io, 1); 2243 } 2244 2245 static void kcryptd_crypt(struct work_struct *work) 2246 { 2247 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2248 2249 if (bio_data_dir(io->base_bio) == READ) 2250 kcryptd_crypt_read_convert(io); 2251 else 2252 kcryptd_crypt_write_convert(io); 2253 } 2254 2255 static void kcryptd_crypt_tasklet(unsigned long work) 2256 { 2257 kcryptd_crypt((struct work_struct *)work); 2258 } 2259 2260 static void kcryptd_queue_crypt(struct dm_crypt_io *io) 2261 { 2262 struct crypt_config *cc = io->cc; 2263 2264 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) || 2265 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) { 2266 /* 2267 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context. 2268 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but 2269 * it is being executed with irqs disabled. 2270 */ 2271 if (in_hardirq() || irqs_disabled()) { 2272 io->in_tasklet = true; 2273 tasklet_init(&io->tasklet, kcryptd_crypt_tasklet, (unsigned long)&io->work); 2274 tasklet_schedule(&io->tasklet); 2275 return; 2276 } 2277 2278 kcryptd_crypt(&io->work); 2279 return; 2280 } 2281 2282 INIT_WORK(&io->work, kcryptd_crypt); 2283 queue_work(cc->crypt_queue, &io->work); 2284 } 2285 2286 static void crypt_free_tfms_aead(struct crypt_config *cc) 2287 { 2288 if (!cc->cipher_tfm.tfms_aead) 2289 return; 2290 2291 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2292 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]); 2293 cc->cipher_tfm.tfms_aead[0] = NULL; 2294 } 2295 2296 kfree(cc->cipher_tfm.tfms_aead); 2297 cc->cipher_tfm.tfms_aead = NULL; 2298 } 2299 2300 static void crypt_free_tfms_skcipher(struct crypt_config *cc) 2301 { 2302 unsigned int i; 2303 2304 if (!cc->cipher_tfm.tfms) 2305 return; 2306 2307 for (i = 0; i < cc->tfms_count; i++) 2308 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) { 2309 crypto_free_skcipher(cc->cipher_tfm.tfms[i]); 2310 cc->cipher_tfm.tfms[i] = NULL; 2311 } 2312 2313 kfree(cc->cipher_tfm.tfms); 2314 cc->cipher_tfm.tfms = NULL; 2315 } 2316 2317 static void crypt_free_tfms(struct crypt_config *cc) 2318 { 2319 if (crypt_integrity_aead(cc)) 2320 crypt_free_tfms_aead(cc); 2321 else 2322 crypt_free_tfms_skcipher(cc); 2323 } 2324 2325 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode) 2326 { 2327 unsigned int i; 2328 int err; 2329 2330 cc->cipher_tfm.tfms = kcalloc(cc->tfms_count, 2331 sizeof(struct crypto_skcipher *), 2332 GFP_KERNEL); 2333 if (!cc->cipher_tfm.tfms) 2334 return -ENOMEM; 2335 2336 for (i = 0; i < cc->tfms_count; i++) { 2337 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 2338 CRYPTO_ALG_ALLOCATES_MEMORY); 2339 if (IS_ERR(cc->cipher_tfm.tfms[i])) { 2340 err = PTR_ERR(cc->cipher_tfm.tfms[i]); 2341 crypt_free_tfms(cc); 2342 return err; 2343 } 2344 } 2345 2346 /* 2347 * dm-crypt performance can vary greatly depending on which crypto 2348 * algorithm implementation is used. Help people debug performance 2349 * problems by logging the ->cra_driver_name. 2350 */ 2351 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2352 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name); 2353 return 0; 2354 } 2355 2356 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode) 2357 { 2358 int err; 2359 2360 cc->cipher_tfm.tfms = kmalloc(sizeof(struct crypto_aead *), GFP_KERNEL); 2361 if (!cc->cipher_tfm.tfms) 2362 return -ENOMEM; 2363 2364 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0, 2365 CRYPTO_ALG_ALLOCATES_MEMORY); 2366 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2367 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]); 2368 crypt_free_tfms(cc); 2369 return err; 2370 } 2371 2372 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2373 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name); 2374 return 0; 2375 } 2376 2377 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode) 2378 { 2379 if (crypt_integrity_aead(cc)) 2380 return crypt_alloc_tfms_aead(cc, ciphermode); 2381 else 2382 return crypt_alloc_tfms_skcipher(cc, ciphermode); 2383 } 2384 2385 static unsigned int crypt_subkey_size(struct crypt_config *cc) 2386 { 2387 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count); 2388 } 2389 2390 static unsigned int crypt_authenckey_size(struct crypt_config *cc) 2391 { 2392 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param)); 2393 } 2394 2395 /* 2396 * If AEAD is composed like authenc(hmac(sha256),xts(aes)), 2397 * the key must be for some reason in special format. 2398 * This funcion converts cc->key to this special format. 2399 */ 2400 static void crypt_copy_authenckey(char *p, const void *key, 2401 unsigned int enckeylen, unsigned int authkeylen) 2402 { 2403 struct crypto_authenc_key_param *param; 2404 struct rtattr *rta; 2405 2406 rta = (struct rtattr *)p; 2407 param = RTA_DATA(rta); 2408 param->enckeylen = cpu_to_be32(enckeylen); 2409 rta->rta_len = RTA_LENGTH(sizeof(*param)); 2410 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM; 2411 p += RTA_SPACE(sizeof(*param)); 2412 memcpy(p, key + enckeylen, authkeylen); 2413 p += authkeylen; 2414 memcpy(p, key, enckeylen); 2415 } 2416 2417 static int crypt_setkey(struct crypt_config *cc) 2418 { 2419 unsigned int subkey_size; 2420 int err = 0, i, r; 2421 2422 /* Ignore extra keys (which are used for IV etc) */ 2423 subkey_size = crypt_subkey_size(cc); 2424 2425 if (crypt_integrity_hmac(cc)) { 2426 if (subkey_size < cc->key_mac_size) 2427 return -EINVAL; 2428 2429 crypt_copy_authenckey(cc->authenc_key, cc->key, 2430 subkey_size - cc->key_mac_size, 2431 cc->key_mac_size); 2432 } 2433 2434 for (i = 0; i < cc->tfms_count; i++) { 2435 if (crypt_integrity_hmac(cc)) 2436 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2437 cc->authenc_key, crypt_authenckey_size(cc)); 2438 else if (crypt_integrity_aead(cc)) 2439 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2440 cc->key + (i * subkey_size), 2441 subkey_size); 2442 else 2443 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i], 2444 cc->key + (i * subkey_size), 2445 subkey_size); 2446 if (r) 2447 err = r; 2448 } 2449 2450 if (crypt_integrity_hmac(cc)) 2451 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc)); 2452 2453 return err; 2454 } 2455 2456 #ifdef CONFIG_KEYS 2457 2458 static bool contains_whitespace(const char *str) 2459 { 2460 while (*str) 2461 if (isspace(*str++)) 2462 return true; 2463 return false; 2464 } 2465 2466 static int set_key_user(struct crypt_config *cc, struct key *key) 2467 { 2468 const struct user_key_payload *ukp; 2469 2470 ukp = user_key_payload_locked(key); 2471 if (!ukp) 2472 return -EKEYREVOKED; 2473 2474 if (cc->key_size != ukp->datalen) 2475 return -EINVAL; 2476 2477 memcpy(cc->key, ukp->data, cc->key_size); 2478 2479 return 0; 2480 } 2481 2482 static int set_key_encrypted(struct crypt_config *cc, struct key *key) 2483 { 2484 const struct encrypted_key_payload *ekp; 2485 2486 ekp = key->payload.data[0]; 2487 if (!ekp) 2488 return -EKEYREVOKED; 2489 2490 if (cc->key_size != ekp->decrypted_datalen) 2491 return -EINVAL; 2492 2493 memcpy(cc->key, ekp->decrypted_data, cc->key_size); 2494 2495 return 0; 2496 } 2497 2498 static int set_key_trusted(struct crypt_config *cc, struct key *key) 2499 { 2500 const struct trusted_key_payload *tkp; 2501 2502 tkp = key->payload.data[0]; 2503 if (!tkp) 2504 return -EKEYREVOKED; 2505 2506 if (cc->key_size != tkp->key_len) 2507 return -EINVAL; 2508 2509 memcpy(cc->key, tkp->key, cc->key_size); 2510 2511 return 0; 2512 } 2513 2514 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2515 { 2516 char *new_key_string, *key_desc; 2517 int ret; 2518 struct key_type *type; 2519 struct key *key; 2520 int (*set_key)(struct crypt_config *cc, struct key *key); 2521 2522 /* 2523 * Reject key_string with whitespace. dm core currently lacks code for 2524 * proper whitespace escaping in arguments on DM_TABLE_STATUS path. 2525 */ 2526 if (contains_whitespace(key_string)) { 2527 DMERR("whitespace chars not allowed in key string"); 2528 return -EINVAL; 2529 } 2530 2531 /* look for next ':' separating key_type from key_description */ 2532 key_desc = strchr(key_string, ':'); 2533 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1)) 2534 return -EINVAL; 2535 2536 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) { 2537 type = &key_type_logon; 2538 set_key = set_key_user; 2539 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) { 2540 type = &key_type_user; 2541 set_key = set_key_user; 2542 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) && 2543 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) { 2544 type = &key_type_encrypted; 2545 set_key = set_key_encrypted; 2546 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) && 2547 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) { 2548 type = &key_type_trusted; 2549 set_key = set_key_trusted; 2550 } else { 2551 return -EINVAL; 2552 } 2553 2554 new_key_string = kstrdup(key_string, GFP_KERNEL); 2555 if (!new_key_string) 2556 return -ENOMEM; 2557 2558 key = request_key(type, key_desc + 1, NULL); 2559 if (IS_ERR(key)) { 2560 kfree_sensitive(new_key_string); 2561 return PTR_ERR(key); 2562 } 2563 2564 down_read(&key->sem); 2565 2566 ret = set_key(cc, key); 2567 if (ret < 0) { 2568 up_read(&key->sem); 2569 key_put(key); 2570 kfree_sensitive(new_key_string); 2571 return ret; 2572 } 2573 2574 up_read(&key->sem); 2575 key_put(key); 2576 2577 /* clear the flag since following operations may invalidate previously valid key */ 2578 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2579 2580 ret = crypt_setkey(cc); 2581 2582 if (!ret) { 2583 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2584 kfree_sensitive(cc->key_string); 2585 cc->key_string = new_key_string; 2586 } else 2587 kfree_sensitive(new_key_string); 2588 2589 return ret; 2590 } 2591 2592 static int get_key_size(char **key_string) 2593 { 2594 char *colon, dummy; 2595 int ret; 2596 2597 if (*key_string[0] != ':') 2598 return strlen(*key_string) >> 1; 2599 2600 /* look for next ':' in key string */ 2601 colon = strpbrk(*key_string + 1, ":"); 2602 if (!colon) 2603 return -EINVAL; 2604 2605 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':') 2606 return -EINVAL; 2607 2608 *key_string = colon; 2609 2610 /* remaining key string should be :<logon|user>:<key_desc> */ 2611 2612 return ret; 2613 } 2614 2615 #else 2616 2617 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2618 { 2619 return -EINVAL; 2620 } 2621 2622 static int get_key_size(char **key_string) 2623 { 2624 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1); 2625 } 2626 2627 #endif /* CONFIG_KEYS */ 2628 2629 static int crypt_set_key(struct crypt_config *cc, char *key) 2630 { 2631 int r = -EINVAL; 2632 int key_string_len = strlen(key); 2633 2634 /* Hyphen (which gives a key_size of zero) means there is no key. */ 2635 if (!cc->key_size && strcmp(key, "-")) 2636 goto out; 2637 2638 /* ':' means the key is in kernel keyring, short-circuit normal key processing */ 2639 if (key[0] == ':') { 2640 r = crypt_set_keyring_key(cc, key + 1); 2641 goto out; 2642 } 2643 2644 /* clear the flag since following operations may invalidate previously valid key */ 2645 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2646 2647 /* wipe references to any kernel keyring key */ 2648 kfree_sensitive(cc->key_string); 2649 cc->key_string = NULL; 2650 2651 /* Decode key from its hex representation. */ 2652 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0) 2653 goto out; 2654 2655 r = crypt_setkey(cc); 2656 if (!r) 2657 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2658 2659 out: 2660 /* Hex key string not needed after here, so wipe it. */ 2661 memset(key, '0', key_string_len); 2662 2663 return r; 2664 } 2665 2666 static int crypt_wipe_key(struct crypt_config *cc) 2667 { 2668 int r; 2669 2670 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2671 get_random_bytes(&cc->key, cc->key_size); 2672 2673 /* Wipe IV private keys */ 2674 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) { 2675 r = cc->iv_gen_ops->wipe(cc); 2676 if (r) 2677 return r; 2678 } 2679 2680 kfree_sensitive(cc->key_string); 2681 cc->key_string = NULL; 2682 r = crypt_setkey(cc); 2683 memset(&cc->key, 0, cc->key_size * sizeof(u8)); 2684 2685 return r; 2686 } 2687 2688 static void crypt_calculate_pages_per_client(void) 2689 { 2690 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100; 2691 2692 if (!dm_crypt_clients_n) 2693 return; 2694 2695 pages /= dm_crypt_clients_n; 2696 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT) 2697 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT; 2698 dm_crypt_pages_per_client = pages; 2699 } 2700 2701 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data) 2702 { 2703 struct crypt_config *cc = pool_data; 2704 struct page *page; 2705 2706 /* 2707 * Note, percpu_counter_read_positive() may over (and under) estimate 2708 * the current usage by at most (batch - 1) * num_online_cpus() pages, 2709 * but avoids potential spinlock contention of an exact result. 2710 */ 2711 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) && 2712 likely(gfp_mask & __GFP_NORETRY)) 2713 return NULL; 2714 2715 page = alloc_page(gfp_mask); 2716 if (likely(page != NULL)) 2717 percpu_counter_add(&cc->n_allocated_pages, 1); 2718 2719 return page; 2720 } 2721 2722 static void crypt_page_free(void *page, void *pool_data) 2723 { 2724 struct crypt_config *cc = pool_data; 2725 2726 __free_page(page); 2727 percpu_counter_sub(&cc->n_allocated_pages, 1); 2728 } 2729 2730 static void crypt_dtr(struct dm_target *ti) 2731 { 2732 struct crypt_config *cc = ti->private; 2733 2734 ti->private = NULL; 2735 2736 if (!cc) 2737 return; 2738 2739 if (cc->write_thread) 2740 kthread_stop(cc->write_thread); 2741 2742 if (cc->io_queue) 2743 destroy_workqueue(cc->io_queue); 2744 if (cc->crypt_queue) 2745 destroy_workqueue(cc->crypt_queue); 2746 2747 crypt_free_tfms(cc); 2748 2749 bioset_exit(&cc->bs); 2750 2751 mempool_exit(&cc->page_pool); 2752 mempool_exit(&cc->req_pool); 2753 mempool_exit(&cc->tag_pool); 2754 2755 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0); 2756 percpu_counter_destroy(&cc->n_allocated_pages); 2757 2758 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) 2759 cc->iv_gen_ops->dtr(cc); 2760 2761 if (cc->dev) 2762 dm_put_device(ti, cc->dev); 2763 2764 kfree_sensitive(cc->cipher_string); 2765 kfree_sensitive(cc->key_string); 2766 kfree_sensitive(cc->cipher_auth); 2767 kfree_sensitive(cc->authenc_key); 2768 2769 mutex_destroy(&cc->bio_alloc_lock); 2770 2771 /* Must zero key material before freeing */ 2772 kfree_sensitive(cc); 2773 2774 spin_lock(&dm_crypt_clients_lock); 2775 WARN_ON(!dm_crypt_clients_n); 2776 dm_crypt_clients_n--; 2777 crypt_calculate_pages_per_client(); 2778 spin_unlock(&dm_crypt_clients_lock); 2779 2780 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1); 2781 } 2782 2783 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode) 2784 { 2785 struct crypt_config *cc = ti->private; 2786 2787 if (crypt_integrity_aead(cc)) 2788 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2789 else 2790 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2791 2792 if (cc->iv_size) 2793 /* at least a 64 bit sector number should fit in our buffer */ 2794 cc->iv_size = max(cc->iv_size, 2795 (unsigned int)(sizeof(u64) / sizeof(u8))); 2796 else if (ivmode) { 2797 DMWARN("Selected cipher does not support IVs"); 2798 ivmode = NULL; 2799 } 2800 2801 /* Choose ivmode, see comments at iv code. */ 2802 if (ivmode == NULL) 2803 cc->iv_gen_ops = NULL; 2804 else if (strcmp(ivmode, "plain") == 0) 2805 cc->iv_gen_ops = &crypt_iv_plain_ops; 2806 else if (strcmp(ivmode, "plain64") == 0) 2807 cc->iv_gen_ops = &crypt_iv_plain64_ops; 2808 else if (strcmp(ivmode, "plain64be") == 0) 2809 cc->iv_gen_ops = &crypt_iv_plain64be_ops; 2810 else if (strcmp(ivmode, "essiv") == 0) 2811 cc->iv_gen_ops = &crypt_iv_essiv_ops; 2812 else if (strcmp(ivmode, "benbi") == 0) 2813 cc->iv_gen_ops = &crypt_iv_benbi_ops; 2814 else if (strcmp(ivmode, "null") == 0) 2815 cc->iv_gen_ops = &crypt_iv_null_ops; 2816 else if (strcmp(ivmode, "eboiv") == 0) 2817 cc->iv_gen_ops = &crypt_iv_eboiv_ops; 2818 else if (strcmp(ivmode, "elephant") == 0) { 2819 cc->iv_gen_ops = &crypt_iv_elephant_ops; 2820 cc->key_parts = 2; 2821 cc->key_extra_size = cc->key_size / 2; 2822 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE) 2823 return -EINVAL; 2824 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags); 2825 } else if (strcmp(ivmode, "lmk") == 0) { 2826 cc->iv_gen_ops = &crypt_iv_lmk_ops; 2827 /* 2828 * Version 2 and 3 is recognised according 2829 * to length of provided multi-key string. 2830 * If present (version 3), last key is used as IV seed. 2831 * All keys (including IV seed) are always the same size. 2832 */ 2833 if (cc->key_size % cc->key_parts) { 2834 cc->key_parts++; 2835 cc->key_extra_size = cc->key_size / cc->key_parts; 2836 } 2837 } else if (strcmp(ivmode, "tcw") == 0) { 2838 cc->iv_gen_ops = &crypt_iv_tcw_ops; 2839 cc->key_parts += 2; /* IV + whitening */ 2840 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE; 2841 } else if (strcmp(ivmode, "random") == 0) { 2842 cc->iv_gen_ops = &crypt_iv_random_ops; 2843 /* Need storage space in integrity fields. */ 2844 cc->integrity_iv_size = cc->iv_size; 2845 } else { 2846 ti->error = "Invalid IV mode"; 2847 return -EINVAL; 2848 } 2849 2850 return 0; 2851 } 2852 2853 /* 2854 * Workaround to parse HMAC algorithm from AEAD crypto API spec. 2855 * The HMAC is needed to calculate tag size (HMAC digest size). 2856 * This should be probably done by crypto-api calls (once available...) 2857 */ 2858 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api) 2859 { 2860 char *start, *end, *mac_alg = NULL; 2861 struct crypto_ahash *mac; 2862 2863 if (!strstarts(cipher_api, "authenc(")) 2864 return 0; 2865 2866 start = strchr(cipher_api, '('); 2867 end = strchr(cipher_api, ','); 2868 if (!start || !end || ++start > end) 2869 return -EINVAL; 2870 2871 mac_alg = kzalloc(end - start + 1, GFP_KERNEL); 2872 if (!mac_alg) 2873 return -ENOMEM; 2874 strncpy(mac_alg, start, end - start); 2875 2876 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY); 2877 kfree(mac_alg); 2878 2879 if (IS_ERR(mac)) 2880 return PTR_ERR(mac); 2881 2882 cc->key_mac_size = crypto_ahash_digestsize(mac); 2883 crypto_free_ahash(mac); 2884 2885 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL); 2886 if (!cc->authenc_key) 2887 return -ENOMEM; 2888 2889 return 0; 2890 } 2891 2892 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key, 2893 char **ivmode, char **ivopts) 2894 { 2895 struct crypt_config *cc = ti->private; 2896 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME]; 2897 int ret = -EINVAL; 2898 2899 cc->tfms_count = 1; 2900 2901 /* 2902 * New format (capi: prefix) 2903 * capi:cipher_api_spec-iv:ivopts 2904 */ 2905 tmp = &cipher_in[strlen("capi:")]; 2906 2907 /* Separate IV options if present, it can contain another '-' in hash name */ 2908 *ivopts = strrchr(tmp, ':'); 2909 if (*ivopts) { 2910 **ivopts = '\0'; 2911 (*ivopts)++; 2912 } 2913 /* Parse IV mode */ 2914 *ivmode = strrchr(tmp, '-'); 2915 if (*ivmode) { 2916 **ivmode = '\0'; 2917 (*ivmode)++; 2918 } 2919 /* The rest is crypto API spec */ 2920 cipher_api = tmp; 2921 2922 /* Alloc AEAD, can be used only in new format. */ 2923 if (crypt_integrity_aead(cc)) { 2924 ret = crypt_ctr_auth_cipher(cc, cipher_api); 2925 if (ret < 0) { 2926 ti->error = "Invalid AEAD cipher spec"; 2927 return ret; 2928 } 2929 } 2930 2931 if (*ivmode && !strcmp(*ivmode, "lmk")) 2932 cc->tfms_count = 64; 2933 2934 if (*ivmode && !strcmp(*ivmode, "essiv")) { 2935 if (!*ivopts) { 2936 ti->error = "Digest algorithm missing for ESSIV mode"; 2937 return -EINVAL; 2938 } 2939 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)", 2940 cipher_api, *ivopts); 2941 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 2942 ti->error = "Cannot allocate cipher string"; 2943 return -ENOMEM; 2944 } 2945 cipher_api = buf; 2946 } 2947 2948 cc->key_parts = cc->tfms_count; 2949 2950 /* Allocate cipher */ 2951 ret = crypt_alloc_tfms(cc, cipher_api); 2952 if (ret < 0) { 2953 ti->error = "Error allocating crypto tfm"; 2954 return ret; 2955 } 2956 2957 if (crypt_integrity_aead(cc)) 2958 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2959 else 2960 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2961 2962 return 0; 2963 } 2964 2965 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key, 2966 char **ivmode, char **ivopts) 2967 { 2968 struct crypt_config *cc = ti->private; 2969 char *tmp, *cipher, *chainmode, *keycount; 2970 char *cipher_api = NULL; 2971 int ret = -EINVAL; 2972 char dummy; 2973 2974 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) { 2975 ti->error = "Bad cipher specification"; 2976 return -EINVAL; 2977 } 2978 2979 /* 2980 * Legacy dm-crypt cipher specification 2981 * cipher[:keycount]-mode-iv:ivopts 2982 */ 2983 tmp = cipher_in; 2984 keycount = strsep(&tmp, "-"); 2985 cipher = strsep(&keycount, ":"); 2986 2987 if (!keycount) 2988 cc->tfms_count = 1; 2989 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 || 2990 !is_power_of_2(cc->tfms_count)) { 2991 ti->error = "Bad cipher key count specification"; 2992 return -EINVAL; 2993 } 2994 cc->key_parts = cc->tfms_count; 2995 2996 chainmode = strsep(&tmp, "-"); 2997 *ivmode = strsep(&tmp, ":"); 2998 *ivopts = tmp; 2999 3000 /* 3001 * For compatibility with the original dm-crypt mapping format, if 3002 * only the cipher name is supplied, use cbc-plain. 3003 */ 3004 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) { 3005 chainmode = "cbc"; 3006 *ivmode = "plain"; 3007 } 3008 3009 if (strcmp(chainmode, "ecb") && !*ivmode) { 3010 ti->error = "IV mechanism required"; 3011 return -EINVAL; 3012 } 3013 3014 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL); 3015 if (!cipher_api) 3016 goto bad_mem; 3017 3018 if (*ivmode && !strcmp(*ivmode, "essiv")) { 3019 if (!*ivopts) { 3020 ti->error = "Digest algorithm missing for ESSIV mode"; 3021 kfree(cipher_api); 3022 return -EINVAL; 3023 } 3024 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 3025 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts); 3026 } else { 3027 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 3028 "%s(%s)", chainmode, cipher); 3029 } 3030 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 3031 kfree(cipher_api); 3032 goto bad_mem; 3033 } 3034 3035 /* Allocate cipher */ 3036 ret = crypt_alloc_tfms(cc, cipher_api); 3037 if (ret < 0) { 3038 ti->error = "Error allocating crypto tfm"; 3039 kfree(cipher_api); 3040 return ret; 3041 } 3042 kfree(cipher_api); 3043 3044 return 0; 3045 bad_mem: 3046 ti->error = "Cannot allocate cipher strings"; 3047 return -ENOMEM; 3048 } 3049 3050 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key) 3051 { 3052 struct crypt_config *cc = ti->private; 3053 char *ivmode = NULL, *ivopts = NULL; 3054 int ret; 3055 3056 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL); 3057 if (!cc->cipher_string) { 3058 ti->error = "Cannot allocate cipher strings"; 3059 return -ENOMEM; 3060 } 3061 3062 if (strstarts(cipher_in, "capi:")) 3063 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts); 3064 else 3065 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts); 3066 if (ret) 3067 return ret; 3068 3069 /* Initialize IV */ 3070 ret = crypt_ctr_ivmode(ti, ivmode); 3071 if (ret < 0) 3072 return ret; 3073 3074 /* Initialize and set key */ 3075 ret = crypt_set_key(cc, key); 3076 if (ret < 0) { 3077 ti->error = "Error decoding and setting key"; 3078 return ret; 3079 } 3080 3081 /* Allocate IV */ 3082 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) { 3083 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts); 3084 if (ret < 0) { 3085 ti->error = "Error creating IV"; 3086 return ret; 3087 } 3088 } 3089 3090 /* Initialize IV (set keys for ESSIV etc) */ 3091 if (cc->iv_gen_ops && cc->iv_gen_ops->init) { 3092 ret = cc->iv_gen_ops->init(cc); 3093 if (ret < 0) { 3094 ti->error = "Error initialising IV"; 3095 return ret; 3096 } 3097 } 3098 3099 /* wipe the kernel key payload copy */ 3100 if (cc->key_string) 3101 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3102 3103 return ret; 3104 } 3105 3106 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv) 3107 { 3108 struct crypt_config *cc = ti->private; 3109 struct dm_arg_set as; 3110 static const struct dm_arg _args[] = { 3111 {0, 8, "Invalid number of feature args"}, 3112 }; 3113 unsigned int opt_params, val; 3114 const char *opt_string, *sval; 3115 char dummy; 3116 int ret; 3117 3118 /* Optional parameters */ 3119 as.argc = argc; 3120 as.argv = argv; 3121 3122 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error); 3123 if (ret) 3124 return ret; 3125 3126 while (opt_params--) { 3127 opt_string = dm_shift_arg(&as); 3128 if (!opt_string) { 3129 ti->error = "Not enough feature arguments"; 3130 return -EINVAL; 3131 } 3132 3133 if (!strcasecmp(opt_string, "allow_discards")) 3134 ti->num_discard_bios = 1; 3135 3136 else if (!strcasecmp(opt_string, "same_cpu_crypt")) 3137 set_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3138 3139 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus")) 3140 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3141 else if (!strcasecmp(opt_string, "no_read_workqueue")) 3142 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3143 else if (!strcasecmp(opt_string, "no_write_workqueue")) 3144 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3145 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) { 3146 if (val == 0 || val > MAX_TAG_SIZE) { 3147 ti->error = "Invalid integrity arguments"; 3148 return -EINVAL; 3149 } 3150 cc->on_disk_tag_size = val; 3151 sval = strchr(opt_string + strlen("integrity:"), ':') + 1; 3152 if (!strcasecmp(sval, "aead")) { 3153 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 3154 } else if (strcasecmp(sval, "none")) { 3155 ti->error = "Unknown integrity profile"; 3156 return -EINVAL; 3157 } 3158 3159 cc->cipher_auth = kstrdup(sval, GFP_KERNEL); 3160 if (!cc->cipher_auth) 3161 return -ENOMEM; 3162 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) { 3163 if (cc->sector_size < (1 << SECTOR_SHIFT) || 3164 cc->sector_size > 4096 || 3165 (cc->sector_size & (cc->sector_size - 1))) { 3166 ti->error = "Invalid feature value for sector_size"; 3167 return -EINVAL; 3168 } 3169 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) { 3170 ti->error = "Device size is not multiple of sector_size feature"; 3171 return -EINVAL; 3172 } 3173 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT; 3174 } else if (!strcasecmp(opt_string, "iv_large_sectors")) 3175 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3176 else { 3177 ti->error = "Invalid feature arguments"; 3178 return -EINVAL; 3179 } 3180 } 3181 3182 return 0; 3183 } 3184 3185 #ifdef CONFIG_BLK_DEV_ZONED 3186 static int crypt_report_zones(struct dm_target *ti, 3187 struct dm_report_zones_args *args, unsigned int nr_zones) 3188 { 3189 struct crypt_config *cc = ti->private; 3190 3191 return dm_report_zones(cc->dev->bdev, cc->start, 3192 cc->start + dm_target_offset(ti, args->next_sector), 3193 args, nr_zones); 3194 } 3195 #else 3196 #define crypt_report_zones NULL 3197 #endif 3198 3199 /* 3200 * Construct an encryption mapping: 3201 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start> 3202 */ 3203 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) 3204 { 3205 struct crypt_config *cc; 3206 const char *devname = dm_table_device_name(ti->table); 3207 int key_size; 3208 unsigned int align_mask; 3209 unsigned long long tmpll; 3210 int ret; 3211 size_t iv_size_padding, additional_req_size; 3212 char dummy; 3213 3214 if (argc < 5) { 3215 ti->error = "Not enough arguments"; 3216 return -EINVAL; 3217 } 3218 3219 key_size = get_key_size(&argv[1]); 3220 if (key_size < 0) { 3221 ti->error = "Cannot parse key size"; 3222 return -EINVAL; 3223 } 3224 3225 cc = kzalloc(struct_size(cc, key, key_size), GFP_KERNEL); 3226 if (!cc) { 3227 ti->error = "Cannot allocate encryption context"; 3228 return -ENOMEM; 3229 } 3230 cc->key_size = key_size; 3231 cc->sector_size = (1 << SECTOR_SHIFT); 3232 cc->sector_shift = 0; 3233 3234 ti->private = cc; 3235 3236 spin_lock(&dm_crypt_clients_lock); 3237 dm_crypt_clients_n++; 3238 crypt_calculate_pages_per_client(); 3239 spin_unlock(&dm_crypt_clients_lock); 3240 3241 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL); 3242 if (ret < 0) 3243 goto bad; 3244 3245 /* Optional parameters need to be read before cipher constructor */ 3246 if (argc > 5) { 3247 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]); 3248 if (ret) 3249 goto bad; 3250 } 3251 3252 ret = crypt_ctr_cipher(ti, argv[0], argv[1]); 3253 if (ret < 0) 3254 goto bad; 3255 3256 if (crypt_integrity_aead(cc)) { 3257 cc->dmreq_start = sizeof(struct aead_request); 3258 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc)); 3259 align_mask = crypto_aead_alignmask(any_tfm_aead(cc)); 3260 } else { 3261 cc->dmreq_start = sizeof(struct skcipher_request); 3262 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc)); 3263 align_mask = crypto_skcipher_alignmask(any_tfm(cc)); 3264 } 3265 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request)); 3266 3267 if (align_mask < CRYPTO_MINALIGN) { 3268 /* Allocate the padding exactly */ 3269 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request)) 3270 & align_mask; 3271 } else { 3272 /* 3273 * If the cipher requires greater alignment than kmalloc 3274 * alignment, we don't know the exact position of the 3275 * initialization vector. We must assume worst case. 3276 */ 3277 iv_size_padding = align_mask; 3278 } 3279 3280 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */ 3281 additional_req_size = sizeof(struct dm_crypt_request) + 3282 iv_size_padding + cc->iv_size + 3283 cc->iv_size + 3284 sizeof(uint64_t) + 3285 sizeof(unsigned int); 3286 3287 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size); 3288 if (ret) { 3289 ti->error = "Cannot allocate crypt request mempool"; 3290 goto bad; 3291 } 3292 3293 cc->per_bio_data_size = ti->per_io_data_size = 3294 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size, 3295 ARCH_DMA_MINALIGN); 3296 3297 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc); 3298 if (ret) { 3299 ti->error = "Cannot allocate page mempool"; 3300 goto bad; 3301 } 3302 3303 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS); 3304 if (ret) { 3305 ti->error = "Cannot allocate crypt bioset"; 3306 goto bad; 3307 } 3308 3309 mutex_init(&cc->bio_alloc_lock); 3310 3311 ret = -EINVAL; 3312 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) || 3313 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) { 3314 ti->error = "Invalid iv_offset sector"; 3315 goto bad; 3316 } 3317 cc->iv_offset = tmpll; 3318 3319 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev); 3320 if (ret) { 3321 ti->error = "Device lookup failed"; 3322 goto bad; 3323 } 3324 3325 ret = -EINVAL; 3326 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) { 3327 ti->error = "Invalid device sector"; 3328 goto bad; 3329 } 3330 cc->start = tmpll; 3331 3332 if (bdev_is_zoned(cc->dev->bdev)) { 3333 /* 3334 * For zoned block devices, we need to preserve the issuer write 3335 * ordering. To do so, disable write workqueues and force inline 3336 * encryption completion. 3337 */ 3338 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3339 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags); 3340 3341 /* 3342 * All zone append writes to a zone of a zoned block device will 3343 * have the same BIO sector, the start of the zone. When the 3344 * cypher IV mode uses sector values, all data targeting a 3345 * zone will be encrypted using the first sector numbers of the 3346 * zone. This will not result in write errors but will 3347 * cause most reads to fail as reads will use the sector values 3348 * for the actual data locations, resulting in IV mismatch. 3349 * To avoid this problem, ask DM core to emulate zone append 3350 * operations with regular writes. 3351 */ 3352 DMDEBUG("Zone append operations will be emulated"); 3353 ti->emulate_zone_append = true; 3354 } 3355 3356 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) { 3357 ret = crypt_integrity_ctr(cc, ti); 3358 if (ret) 3359 goto bad; 3360 3361 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->on_disk_tag_size; 3362 if (!cc->tag_pool_max_sectors) 3363 cc->tag_pool_max_sectors = 1; 3364 3365 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS, 3366 cc->tag_pool_max_sectors * cc->on_disk_tag_size); 3367 if (ret) { 3368 ti->error = "Cannot allocate integrity tags mempool"; 3369 goto bad; 3370 } 3371 3372 cc->tag_pool_max_sectors <<= cc->sector_shift; 3373 } 3374 3375 ret = -ENOMEM; 3376 cc->io_queue = alloc_workqueue("kcryptd_io/%s", WQ_MEM_RECLAIM, 1, devname); 3377 if (!cc->io_queue) { 3378 ti->error = "Couldn't create kcryptd io queue"; 3379 goto bad; 3380 } 3381 3382 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) 3383 cc->crypt_queue = alloc_workqueue("kcryptd/%s", WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 3384 1, devname); 3385 else 3386 cc->crypt_queue = alloc_workqueue("kcryptd/%s", 3387 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, 3388 num_online_cpus(), devname); 3389 if (!cc->crypt_queue) { 3390 ti->error = "Couldn't create kcryptd queue"; 3391 goto bad; 3392 } 3393 3394 spin_lock_init(&cc->write_thread_lock); 3395 cc->write_tree = RB_ROOT; 3396 3397 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname); 3398 if (IS_ERR(cc->write_thread)) { 3399 ret = PTR_ERR(cc->write_thread); 3400 cc->write_thread = NULL; 3401 ti->error = "Couldn't spawn write thread"; 3402 goto bad; 3403 } 3404 3405 ti->num_flush_bios = 1; 3406 ti->limit_swap_bios = true; 3407 ti->accounts_remapped_io = true; 3408 3409 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1); 3410 return 0; 3411 3412 bad: 3413 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0); 3414 crypt_dtr(ti); 3415 return ret; 3416 } 3417 3418 static int crypt_map(struct dm_target *ti, struct bio *bio) 3419 { 3420 struct dm_crypt_io *io; 3421 struct crypt_config *cc = ti->private; 3422 3423 /* 3424 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues. 3425 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight 3426 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters 3427 */ 3428 if (unlikely(bio->bi_opf & REQ_PREFLUSH || 3429 bio_op(bio) == REQ_OP_DISCARD)) { 3430 bio_set_dev(bio, cc->dev->bdev); 3431 if (bio_sectors(bio)) 3432 bio->bi_iter.bi_sector = cc->start + 3433 dm_target_offset(ti, bio->bi_iter.bi_sector); 3434 return DM_MAPIO_REMAPPED; 3435 } 3436 3437 /* 3438 * Check if bio is too large, split as needed. 3439 */ 3440 if (unlikely(bio->bi_iter.bi_size > (BIO_MAX_VECS << PAGE_SHIFT)) && 3441 (bio_data_dir(bio) == WRITE || cc->on_disk_tag_size)) 3442 dm_accept_partial_bio(bio, ((BIO_MAX_VECS << PAGE_SHIFT) >> SECTOR_SHIFT)); 3443 3444 /* 3445 * Ensure that bio is a multiple of internal sector encryption size 3446 * and is aligned to this size as defined in IO hints. 3447 */ 3448 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0)) 3449 return DM_MAPIO_KILL; 3450 3451 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1))) 3452 return DM_MAPIO_KILL; 3453 3454 io = dm_per_bio_data(bio, cc->per_bio_data_size); 3455 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector)); 3456 3457 if (cc->on_disk_tag_size) { 3458 unsigned int tag_len = cc->on_disk_tag_size * (bio_sectors(bio) >> cc->sector_shift); 3459 3460 if (unlikely(tag_len > KMALLOC_MAX_SIZE)) 3461 io->integrity_metadata = NULL; 3462 else 3463 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); 3464 3465 if (unlikely(!io->integrity_metadata)) { 3466 if (bio_sectors(bio) > cc->tag_pool_max_sectors) 3467 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors); 3468 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO); 3469 io->integrity_metadata_from_pool = true; 3470 } 3471 } 3472 3473 if (crypt_integrity_aead(cc)) 3474 io->ctx.r.req_aead = (struct aead_request *)(io + 1); 3475 else 3476 io->ctx.r.req = (struct skcipher_request *)(io + 1); 3477 3478 if (bio_data_dir(io->base_bio) == READ) { 3479 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP)) 3480 kcryptd_queue_read(io); 3481 } else 3482 kcryptd_queue_crypt(io); 3483 3484 return DM_MAPIO_SUBMITTED; 3485 } 3486 3487 static char hex2asc(unsigned char c) 3488 { 3489 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27); 3490 } 3491 3492 static void crypt_status(struct dm_target *ti, status_type_t type, 3493 unsigned int status_flags, char *result, unsigned int maxlen) 3494 { 3495 struct crypt_config *cc = ti->private; 3496 unsigned int i, sz = 0; 3497 int num_feature_args = 0; 3498 3499 switch (type) { 3500 case STATUSTYPE_INFO: 3501 result[0] = '\0'; 3502 break; 3503 3504 case STATUSTYPE_TABLE: 3505 DMEMIT("%s ", cc->cipher_string); 3506 3507 if (cc->key_size > 0) { 3508 if (cc->key_string) 3509 DMEMIT(":%u:%s", cc->key_size, cc->key_string); 3510 else { 3511 for (i = 0; i < cc->key_size; i++) { 3512 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4), 3513 hex2asc(cc->key[i] & 0xf)); 3514 } 3515 } 3516 } else 3517 DMEMIT("-"); 3518 3519 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset, 3520 cc->dev->name, (unsigned long long)cc->start); 3521 3522 num_feature_args += !!ti->num_discard_bios; 3523 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3524 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3525 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3526 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3527 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT); 3528 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3529 if (cc->on_disk_tag_size) 3530 num_feature_args++; 3531 if (num_feature_args) { 3532 DMEMIT(" %d", num_feature_args); 3533 if (ti->num_discard_bios) 3534 DMEMIT(" allow_discards"); 3535 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) 3536 DMEMIT(" same_cpu_crypt"); 3537 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) 3538 DMEMIT(" submit_from_crypt_cpus"); 3539 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) 3540 DMEMIT(" no_read_workqueue"); 3541 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) 3542 DMEMIT(" no_write_workqueue"); 3543 if (cc->on_disk_tag_size) 3544 DMEMIT(" integrity:%u:%s", cc->on_disk_tag_size, cc->cipher_auth); 3545 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3546 DMEMIT(" sector_size:%d", cc->sector_size); 3547 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 3548 DMEMIT(" iv_large_sectors"); 3549 } 3550 break; 3551 3552 case STATUSTYPE_IMA: 3553 DMEMIT_TARGET_NAME_VERSION(ti->type); 3554 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n'); 3555 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n'); 3556 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ? 3557 'y' : 'n'); 3558 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ? 3559 'y' : 'n'); 3560 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ? 3561 'y' : 'n'); 3562 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ? 3563 'y' : 'n'); 3564 3565 if (cc->on_disk_tag_size) 3566 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s", 3567 cc->on_disk_tag_size, cc->cipher_auth); 3568 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3569 DMEMIT(",sector_size=%d", cc->sector_size); 3570 if (cc->cipher_string) 3571 DMEMIT(",cipher_string=%s", cc->cipher_string); 3572 3573 DMEMIT(",key_size=%u", cc->key_size); 3574 DMEMIT(",key_parts=%u", cc->key_parts); 3575 DMEMIT(",key_extra_size=%u", cc->key_extra_size); 3576 DMEMIT(",key_mac_size=%u", cc->key_mac_size); 3577 DMEMIT(";"); 3578 break; 3579 } 3580 } 3581 3582 static void crypt_postsuspend(struct dm_target *ti) 3583 { 3584 struct crypt_config *cc = ti->private; 3585 3586 set_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3587 } 3588 3589 static int crypt_preresume(struct dm_target *ti) 3590 { 3591 struct crypt_config *cc = ti->private; 3592 3593 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) { 3594 DMERR("aborting resume - crypt key is not set."); 3595 return -EAGAIN; 3596 } 3597 3598 return 0; 3599 } 3600 3601 static void crypt_resume(struct dm_target *ti) 3602 { 3603 struct crypt_config *cc = ti->private; 3604 3605 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3606 } 3607 3608 /* Message interface 3609 * key set <key> 3610 * key wipe 3611 */ 3612 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv, 3613 char *result, unsigned int maxlen) 3614 { 3615 struct crypt_config *cc = ti->private; 3616 int key_size, ret = -EINVAL; 3617 3618 if (argc < 2) 3619 goto error; 3620 3621 if (!strcasecmp(argv[0], "key")) { 3622 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) { 3623 DMWARN("not suspended during key manipulation."); 3624 return -EINVAL; 3625 } 3626 if (argc == 3 && !strcasecmp(argv[1], "set")) { 3627 /* The key size may not be changed. */ 3628 key_size = get_key_size(&argv[2]); 3629 if (key_size < 0 || cc->key_size != key_size) { 3630 memset(argv[2], '0', strlen(argv[2])); 3631 return -EINVAL; 3632 } 3633 3634 ret = crypt_set_key(cc, argv[2]); 3635 if (ret) 3636 return ret; 3637 if (cc->iv_gen_ops && cc->iv_gen_ops->init) 3638 ret = cc->iv_gen_ops->init(cc); 3639 /* wipe the kernel key payload copy */ 3640 if (cc->key_string) 3641 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3642 return ret; 3643 } 3644 if (argc == 2 && !strcasecmp(argv[1], "wipe")) 3645 return crypt_wipe_key(cc); 3646 } 3647 3648 error: 3649 DMWARN("unrecognised message received."); 3650 return -EINVAL; 3651 } 3652 3653 static int crypt_iterate_devices(struct dm_target *ti, 3654 iterate_devices_callout_fn fn, void *data) 3655 { 3656 struct crypt_config *cc = ti->private; 3657 3658 return fn(ti, cc->dev, cc->start, ti->len, data); 3659 } 3660 3661 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits) 3662 { 3663 struct crypt_config *cc = ti->private; 3664 3665 /* 3666 * Unfortunate constraint that is required to avoid the potential 3667 * for exceeding underlying device's max_segments limits -- due to 3668 * crypt_alloc_buffer() possibly allocating pages for the encryption 3669 * bio that are not as physically contiguous as the original bio. 3670 */ 3671 limits->max_segment_size = PAGE_SIZE; 3672 3673 limits->logical_block_size = 3674 max_t(unsigned int, limits->logical_block_size, cc->sector_size); 3675 limits->physical_block_size = 3676 max_t(unsigned int, limits->physical_block_size, cc->sector_size); 3677 limits->io_min = max_t(unsigned int, limits->io_min, cc->sector_size); 3678 limits->dma_alignment = limits->logical_block_size - 1; 3679 } 3680 3681 static struct target_type crypt_target = { 3682 .name = "crypt", 3683 .version = {1, 24, 0}, 3684 .module = THIS_MODULE, 3685 .ctr = crypt_ctr, 3686 .dtr = crypt_dtr, 3687 .features = DM_TARGET_ZONED_HM, 3688 .report_zones = crypt_report_zones, 3689 .map = crypt_map, 3690 .status = crypt_status, 3691 .postsuspend = crypt_postsuspend, 3692 .preresume = crypt_preresume, 3693 .resume = crypt_resume, 3694 .message = crypt_message, 3695 .iterate_devices = crypt_iterate_devices, 3696 .io_hints = crypt_io_hints, 3697 }; 3698 module_dm(crypt); 3699 3700 MODULE_AUTHOR("Jana Saout <jana@saout.de>"); 3701 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption"); 3702 MODULE_LICENSE("GPL"); 3703