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