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/hex.h> 15 #include <linux/init.h> 16 #include <linux/kernel.h> 17 #include <linux/key.h> 18 #include <linux/bio.h> 19 #include <linux/blkdev.h> 20 #include <linux/blk-integrity.h> 21 #include <linux/crc32.h> 22 #include <linux/mempool.h> 23 #include <linux/slab.h> 24 #include <linux/crypto.h> 25 #include <linux/fips.h> 26 #include <linux/workqueue.h> 27 #include <linux/kthread.h> 28 #include <linux/backing-dev.h> 29 #include <linux/atomic.h> 30 #include <linux/scatterlist.h> 31 #include <linux/rbtree.h> 32 #include <linux/ctype.h> 33 #include <asm/page.h> 34 #include <linux/unaligned.h> 35 #include <crypto/aes.h> 36 #include <crypto/hash.h> 37 #include <crypto/md5.h> 38 #include <crypto/skcipher.h> 39 #include <crypto/aead.h> 40 #include <crypto/authenc.h> 41 #include <crypto/utils.h> 42 #include <linux/rtnetlink.h> /* for struct rtattr and RTA macros only */ 43 #include <linux/key-type.h> 44 #include <keys/user-type.h> 45 #include <keys/encrypted-type.h> 46 #include <keys/trusted-type.h> 47 48 #include <linux/device-mapper.h> 49 50 #include "dm-audit.h" 51 52 #define DM_MSG_PREFIX "crypt" 53 54 static DEFINE_IDA(workqueue_ida); 55 56 /* 57 * context holding the current state of a multi-part conversion 58 */ 59 struct convert_context { 60 struct completion restart; 61 struct bio *bio_in; 62 struct bvec_iter iter_in; 63 struct bio *bio_out; 64 struct bvec_iter iter_out; 65 atomic_t cc_pending; 66 unsigned int tag_offset; 67 u64 cc_sector; 68 union { 69 struct skcipher_request *req; 70 struct aead_request *req_aead; 71 } r; 72 bool aead_recheck; 73 bool aead_failed; 74 75 }; 76 77 /* 78 * per bio private data 79 */ 80 struct dm_crypt_io { 81 struct crypt_config *cc; 82 struct bio *base_bio; 83 u8 *integrity_metadata; 84 bool integrity_metadata_from_pool:1; 85 86 struct work_struct work; 87 88 struct convert_context ctx; 89 90 atomic_t io_pending; 91 blk_status_t error; 92 sector_t sector; 93 94 struct bvec_iter saved_bi_iter; 95 96 struct rb_node rb_node; 97 } CRYPTO_MINALIGN_ATTR; 98 99 struct dm_crypt_request { 100 struct convert_context *ctx; 101 struct scatterlist sg_in[4]; 102 struct scatterlist sg_out[4]; 103 u64 iv_sector; 104 }; 105 106 struct crypt_config; 107 108 struct crypt_iv_operations { 109 int (*ctr)(struct crypt_config *cc, struct dm_target *ti, 110 const char *opts); 111 void (*dtr)(struct crypt_config *cc); 112 int (*init)(struct crypt_config *cc); 113 void (*wipe)(struct crypt_config *cc); 114 int (*generator)(struct crypt_config *cc, u8 *iv, 115 struct dm_crypt_request *dmreq); 116 void (*post)(struct crypt_config *cc, u8 *iv, 117 struct dm_crypt_request *dmreq); 118 }; 119 120 struct iv_benbi_private { 121 int shift; 122 }; 123 124 #define LMK_SEED_SIZE 64 /* hash + 0 */ 125 struct iv_lmk_private { 126 u8 *seed; 127 }; 128 129 #define TCW_WHITENING_SIZE 16 130 struct iv_tcw_private { 131 u8 *iv_seed; 132 u8 *whitening; 133 }; 134 135 #define ELEPHANT_MAX_KEY_SIZE 32 136 struct iv_elephant_private { 137 struct aes_enckey *key; 138 }; 139 140 /* 141 * Crypt: maps a linear range of a block device 142 * and encrypts / decrypts at the same time. 143 */ 144 enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID, 145 DM_CRYPT_SAME_CPU, DM_CRYPT_HIGH_PRIORITY, 146 DM_CRYPT_NO_OFFLOAD, DM_CRYPT_NO_READ_WORKQUEUE, 147 DM_CRYPT_NO_WRITE_WORKQUEUE, DM_CRYPT_WRITE_INLINE }; 148 149 enum cipher_flags { 150 CRYPT_MODE_INTEGRITY_AEAD, /* Use authenticated mode for cipher */ 151 CRYPT_IV_LARGE_SECTORS, /* Calculate IV from sector_size, not 512B sectors */ 152 CRYPT_ENCRYPT_PREPROCESS, /* Must preprocess data for encryption (elephant) */ 153 CRYPT_KEY_MAC_SIZE_SET, /* The integrity_key_size option was used */ 154 }; 155 156 /* 157 * The fields in here must be read only after initialization. 158 */ 159 struct crypt_config { 160 struct dm_dev *dev; 161 sector_t start; 162 163 struct percpu_counter n_allocated_pages; 164 165 struct workqueue_struct *io_queue; 166 struct workqueue_struct *crypt_queue; 167 168 spinlock_t write_thread_lock; 169 struct task_struct *write_thread; 170 struct rb_root write_tree; 171 172 char *cipher_string; 173 char *cipher_auth; 174 char *key_string; 175 176 const struct crypt_iv_operations *iv_gen_ops; 177 union { 178 struct iv_benbi_private benbi; 179 struct iv_lmk_private lmk; 180 struct iv_tcw_private tcw; 181 struct iv_elephant_private elephant; 182 } iv_gen_private; 183 u64 iv_offset; 184 unsigned int iv_size; 185 unsigned short sector_size; 186 unsigned char sector_shift; 187 188 union { 189 struct crypto_skcipher **tfms; 190 struct crypto_aead **tfms_aead; 191 } cipher_tfm; 192 unsigned int tfms_count; 193 int workqueue_id; 194 unsigned long cipher_flags; 195 196 /* 197 * Layout of each crypto request: 198 * 199 * struct skcipher_request 200 * context 201 * padding 202 * struct dm_crypt_request 203 * padding 204 * IV 205 * 206 * The padding is added so that dm_crypt_request and the IV are 207 * correctly aligned. 208 */ 209 unsigned int dmreq_start; 210 211 unsigned int per_bio_data_size; 212 213 unsigned long flags; 214 unsigned int key_size; 215 unsigned int key_parts; /* independent parts in key buffer */ 216 unsigned int key_extra_size; /* additional keys length */ 217 unsigned int key_mac_size; /* MAC key size for authenc(...) */ 218 219 unsigned int integrity_tag_size; 220 unsigned int integrity_iv_size; 221 unsigned int used_tag_size; 222 unsigned int tuple_size; 223 224 /* 225 * pool for per bio private data, crypto requests, 226 * encryption requeusts/buffer pages and integrity tags 227 */ 228 unsigned int tag_pool_max_sectors; 229 mempool_t tag_pool; 230 mempool_t req_pool; 231 mempool_t page_pool; 232 233 struct bio_set bs; 234 struct mutex bio_alloc_lock; 235 236 u8 *authenc_key; /* space for keys in authenc() format (if used) */ 237 u8 key[] __counted_by(key_size); 238 }; 239 240 #define MIN_IOS 64 241 #define MAX_TAG_SIZE 480 242 #define POOL_ENTRY_SIZE 512 243 244 static DEFINE_SPINLOCK(dm_crypt_clients_lock); 245 static unsigned int dm_crypt_clients_n; 246 static volatile unsigned long dm_crypt_pages_per_client; 247 #define DM_CRYPT_MEMORY_PERCENT 2 248 #define DM_CRYPT_MIN_PAGES_PER_CLIENT (BIO_MAX_VECS * 16) 249 #define DM_CRYPT_DEFAULT_MAX_READ_SIZE 131072 250 #define DM_CRYPT_DEFAULT_MAX_WRITE_SIZE 131072 251 252 static unsigned int max_read_size = 0; 253 module_param(max_read_size, uint, 0644); 254 MODULE_PARM_DESC(max_read_size, "Maximum size of a read request"); 255 static unsigned int max_write_size = 0; 256 module_param(max_write_size, uint, 0644); 257 MODULE_PARM_DESC(max_write_size, "Maximum size of a write request"); 258 259 static unsigned get_max_request_sectors(struct dm_target *ti, struct bio *bio, bool no_split) 260 { 261 struct crypt_config *cc = ti->private; 262 unsigned val, sector_align; 263 bool wrt = op_is_write(bio_op(bio)); 264 265 if (no_split) { 266 val = -1; 267 } else if (wrt) { 268 val = min_not_zero(READ_ONCE(max_write_size), 269 DM_CRYPT_DEFAULT_MAX_WRITE_SIZE); 270 } else { 271 val = min_not_zero(READ_ONCE(max_read_size), 272 DM_CRYPT_DEFAULT_MAX_READ_SIZE); 273 } 274 275 if (wrt || cc->used_tag_size) 276 val = min(val, BIO_MAX_VECS << PAGE_SHIFT); 277 278 sector_align = max(bdev_logical_block_size(cc->dev->bdev), 279 (unsigned)cc->sector_size); 280 val = round_down(val, sector_align); 281 if (unlikely(!val)) 282 val = sector_align; 283 return val >> SECTOR_SHIFT; 284 } 285 286 static void crypt_endio(struct bio *clone); 287 static void kcryptd_queue_crypt(struct dm_crypt_io *io); 288 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 289 struct scatterlist *sg); 290 291 static bool crypt_integrity_aead(struct crypt_config *cc); 292 293 /* 294 * Use this to access cipher attributes that are independent of the key. 295 */ 296 static struct crypto_skcipher *any_tfm(struct crypt_config *cc) 297 { 298 return cc->cipher_tfm.tfms[0]; 299 } 300 301 static struct crypto_aead *any_tfm_aead(struct crypt_config *cc) 302 { 303 return cc->cipher_tfm.tfms_aead[0]; 304 } 305 306 /* 307 * Different IV generation algorithms: 308 * 309 * plain: the initial vector is the 32-bit little-endian version of the sector 310 * number, padded with zeros if necessary. 311 * 312 * plain64: the initial vector is the 64-bit little-endian version of the sector 313 * number, padded with zeros if necessary. 314 * 315 * plain64be: the initial vector is the 64-bit big-endian version of the sector 316 * number, padded with zeros if necessary. 317 * 318 * essiv: "encrypted sector|salt initial vector", the sector number is 319 * encrypted with the bulk cipher using a salt as key. The salt 320 * should be derived from the bulk cipher's key via hashing. 321 * 322 * benbi: the 64-bit "big-endian 'narrow block'-count", starting at 1 323 * (needed for LRW-32-AES and possible other narrow block modes) 324 * 325 * null: the initial vector is always zero. Provides compatibility with 326 * obsolete loop_fish2 devices. Do not use for new devices. 327 * 328 * lmk: Compatible implementation of the block chaining mode used 329 * by the Loop-AES block device encryption system 330 * designed by Jari Ruusu. See http://loop-aes.sourceforge.net/ 331 * It operates on full 512 byte sectors and uses CBC 332 * with an IV derived from the sector number, the data and 333 * optionally extra IV seed. 334 * This means that after decryption the first block 335 * of sector must be tweaked according to decrypted data. 336 * Loop-AES can use three encryption schemes: 337 * version 1: is plain aes-cbc mode 338 * version 2: uses 64 multikey scheme with lmk IV generator 339 * version 3: the same as version 2 with additional IV seed 340 * (it uses 65 keys, last key is used as IV seed) 341 * 342 * tcw: Compatible implementation of the block chaining mode used 343 * by the TrueCrypt device encryption system (prior to version 4.1). 344 * For more info see: https://gitlab.com/cryptsetup/cryptsetup/wikis/TrueCryptOnDiskFormat 345 * It operates on full 512 byte sectors and uses CBC 346 * with an IV derived from initial key and the sector number. 347 * In addition, whitening value is applied on every sector, whitening 348 * is calculated from initial key, sector number and mixed using CRC32. 349 * Note that this encryption scheme is vulnerable to watermarking attacks 350 * and should be used for old compatible containers access only. 351 * 352 * eboiv: Encrypted byte-offset IV (used in Bitlocker in CBC mode) 353 * The IV is encrypted little-endian byte-offset (with the same key 354 * and cipher as the volume). 355 * 356 * elephant: The extended version of eboiv with additional Elephant diffuser 357 * used with Bitlocker CBC mode. 358 * This mode was used in older Windows systems 359 * https://download.microsoft.com/download/0/2/3/0238acaf-d3bf-4a6d-b3d6-0a0be4bbb36e/bitlockercipher200608.pdf 360 */ 361 362 static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, 363 struct dm_crypt_request *dmreq) 364 { 365 memset(iv, 0, cc->iv_size); 366 *(__le32 *)iv = cpu_to_le32(dmreq->iv_sector & 0xffffffff); 367 368 return 0; 369 } 370 371 static int crypt_iv_plain64_gen(struct crypt_config *cc, u8 *iv, 372 struct dm_crypt_request *dmreq) 373 { 374 memset(iv, 0, cc->iv_size); 375 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 376 377 return 0; 378 } 379 380 static int crypt_iv_plain64be_gen(struct crypt_config *cc, u8 *iv, 381 struct dm_crypt_request *dmreq) 382 { 383 memset(iv, 0, cc->iv_size); 384 /* iv_size is at least of size u64; usually it is 16 bytes */ 385 *(__be64 *)&iv[cc->iv_size - sizeof(u64)] = cpu_to_be64(dmreq->iv_sector); 386 387 return 0; 388 } 389 390 static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, 391 struct dm_crypt_request *dmreq) 392 { 393 /* 394 * ESSIV encryption of the IV is now handled by the crypto API, 395 * so just pass the plain sector number here. 396 */ 397 memset(iv, 0, cc->iv_size); 398 *(__le64 *)iv = cpu_to_le64(dmreq->iv_sector); 399 400 return 0; 401 } 402 403 static int crypt_iv_benbi_ctr(struct crypt_config *cc, struct dm_target *ti, 404 const char *opts) 405 { 406 unsigned int bs; 407 int log; 408 409 if (crypt_integrity_aead(cc)) 410 bs = crypto_aead_blocksize(any_tfm_aead(cc)); 411 else 412 bs = crypto_skcipher_blocksize(any_tfm(cc)); 413 log = ilog2(bs); 414 415 /* 416 * We need to calculate how far we must shift the sector count 417 * to get the cipher block count, we use this shift in _gen. 418 */ 419 if (1 << log != bs) { 420 ti->error = "cypher blocksize is not a power of 2"; 421 return -EINVAL; 422 } 423 424 if (log > 9) { 425 ti->error = "cypher blocksize is > 512"; 426 return -EINVAL; 427 } 428 429 cc->iv_gen_private.benbi.shift = 9 - log; 430 431 return 0; 432 } 433 434 static void crypt_iv_benbi_dtr(struct crypt_config *cc) 435 { 436 } 437 438 static int crypt_iv_benbi_gen(struct crypt_config *cc, u8 *iv, 439 struct dm_crypt_request *dmreq) 440 { 441 __be64 val; 442 443 memset(iv, 0, cc->iv_size - sizeof(u64)); /* rest is cleared below */ 444 445 val = cpu_to_be64(((u64)dmreq->iv_sector << cc->iv_gen_private.benbi.shift) + 1); 446 put_unaligned(val, (__be64 *)(iv + cc->iv_size - sizeof(u64))); 447 448 return 0; 449 } 450 451 static int crypt_iv_null_gen(struct crypt_config *cc, u8 *iv, 452 struct dm_crypt_request *dmreq) 453 { 454 memset(iv, 0, cc->iv_size); 455 456 return 0; 457 } 458 459 static void crypt_iv_lmk_dtr(struct crypt_config *cc) 460 { 461 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 462 463 kfree_sensitive(lmk->seed); 464 lmk->seed = NULL; 465 } 466 467 static int crypt_iv_lmk_ctr(struct crypt_config *cc, struct dm_target *ti, 468 const char *opts) 469 { 470 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 471 472 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 473 ti->error = "Unsupported sector size for LMK"; 474 return -EINVAL; 475 } 476 477 if (fips_enabled) { 478 ti->error = "LMK support is disabled due to FIPS"; 479 /* ... because it uses MD5. */ 480 return -EINVAL; 481 } 482 483 /* No seed in LMK version 2 */ 484 if (cc->key_parts == cc->tfms_count) { 485 lmk->seed = NULL; 486 return 0; 487 } 488 489 lmk->seed = kzalloc(LMK_SEED_SIZE, GFP_KERNEL); 490 if (!lmk->seed) { 491 ti->error = "Error kmallocing seed storage in LMK"; 492 return -ENOMEM; 493 } 494 495 return 0; 496 } 497 498 static int crypt_iv_lmk_init(struct crypt_config *cc) 499 { 500 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 501 int subkey_size = cc->key_size / cc->key_parts; 502 503 /* LMK seed is on the position of LMK_KEYS + 1 key */ 504 if (lmk->seed) 505 memcpy(lmk->seed, cc->key + (cc->tfms_count * subkey_size), 506 MD5_DIGEST_SIZE); 507 508 return 0; 509 } 510 511 static void crypt_iv_lmk_wipe(struct crypt_config *cc) 512 { 513 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 514 515 if (lmk->seed) 516 memset(lmk->seed, 0, LMK_SEED_SIZE); 517 } 518 519 static void crypt_iv_lmk_one(struct crypt_config *cc, u8 *iv, 520 struct dm_crypt_request *dmreq, u8 *data) 521 { 522 struct iv_lmk_private *lmk = &cc->iv_gen_private.lmk; 523 struct md5_ctx ctx; 524 __le32 buf[4]; 525 526 md5_init(&ctx); 527 528 if (lmk->seed) 529 md5_update(&ctx, lmk->seed, LMK_SEED_SIZE); 530 531 /* Sector is always 512B, block size 16, add data of blocks 1-31 */ 532 md5_update(&ctx, data + 16, 16 * 31); 533 534 /* Sector is cropped to 56 bits here */ 535 buf[0] = cpu_to_le32(dmreq->iv_sector & 0xFFFFFFFF); 536 buf[1] = cpu_to_le32((((u64)dmreq->iv_sector >> 32) & 0x00FFFFFF) | 0x80000000); 537 buf[2] = cpu_to_le32(4024); 538 buf[3] = 0; 539 md5_update(&ctx, (u8 *)buf, sizeof(buf)); 540 541 /* No MD5 padding here */ 542 cpu_to_le32_array(ctx.state.h, ARRAY_SIZE(ctx.state.h)); 543 memcpy(iv, ctx.state.h, cc->iv_size); 544 } 545 546 static int crypt_iv_lmk_gen(struct crypt_config *cc, u8 *iv, 547 struct dm_crypt_request *dmreq) 548 { 549 struct scatterlist *sg; 550 u8 *src; 551 552 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 553 sg = crypt_get_sg_data(cc, dmreq->sg_in); 554 src = kmap_local_page(sg_page(sg)); 555 crypt_iv_lmk_one(cc, iv, dmreq, src + sg->offset); 556 kunmap_local(src); 557 } else 558 memset(iv, 0, cc->iv_size); 559 return 0; 560 } 561 562 static void crypt_iv_lmk_post(struct crypt_config *cc, u8 *iv, 563 struct dm_crypt_request *dmreq) 564 { 565 struct scatterlist *sg; 566 u8 *dst; 567 568 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) 569 return; 570 571 sg = crypt_get_sg_data(cc, dmreq->sg_out); 572 dst = kmap_local_page(sg_page(sg)); 573 crypt_iv_lmk_one(cc, iv, dmreq, dst + sg->offset); 574 575 /* Tweak the first block of plaintext sector */ 576 crypto_xor(dst + sg->offset, iv, cc->iv_size); 577 578 kunmap_local(dst); 579 } 580 581 static void crypt_iv_tcw_dtr(struct crypt_config *cc) 582 { 583 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 584 585 kfree_sensitive(tcw->iv_seed); 586 tcw->iv_seed = NULL; 587 kfree_sensitive(tcw->whitening); 588 tcw->whitening = NULL; 589 } 590 591 static int crypt_iv_tcw_ctr(struct crypt_config *cc, struct dm_target *ti, 592 const char *opts) 593 { 594 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 595 596 if (cc->sector_size != (1 << SECTOR_SHIFT)) { 597 ti->error = "Unsupported sector size for TCW"; 598 return -EINVAL; 599 } 600 601 if (cc->key_size <= (cc->iv_size + TCW_WHITENING_SIZE)) { 602 ti->error = "Wrong key size for TCW"; 603 return -EINVAL; 604 } 605 606 tcw->iv_seed = kzalloc(cc->iv_size, GFP_KERNEL); 607 tcw->whitening = kzalloc(TCW_WHITENING_SIZE, GFP_KERNEL); 608 if (!tcw->iv_seed || !tcw->whitening) { 609 crypt_iv_tcw_dtr(cc); 610 ti->error = "Error allocating seed storage in TCW"; 611 return -ENOMEM; 612 } 613 614 return 0; 615 } 616 617 static int crypt_iv_tcw_init(struct crypt_config *cc) 618 { 619 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 620 int key_offset = cc->key_size - cc->iv_size - TCW_WHITENING_SIZE; 621 622 memcpy(tcw->iv_seed, &cc->key[key_offset], cc->iv_size); 623 memcpy(tcw->whitening, &cc->key[key_offset + cc->iv_size], 624 TCW_WHITENING_SIZE); 625 626 return 0; 627 } 628 629 static void crypt_iv_tcw_wipe(struct crypt_config *cc) 630 { 631 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 632 633 memset(tcw->iv_seed, 0, cc->iv_size); 634 memset(tcw->whitening, 0, TCW_WHITENING_SIZE); 635 } 636 637 static void crypt_iv_tcw_whitening(struct crypt_config *cc, 638 struct dm_crypt_request *dmreq, u8 *data) 639 { 640 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 641 __le64 sector = cpu_to_le64(dmreq->iv_sector); 642 u8 buf[TCW_WHITENING_SIZE]; 643 int i; 644 645 /* xor whitening with sector number */ 646 crypto_xor_cpy(buf, tcw->whitening, (u8 *)§or, 8); 647 crypto_xor_cpy(&buf[8], tcw->whitening + 8, (u8 *)§or, 8); 648 649 /* calculate crc32 for every 32bit part and xor it */ 650 for (i = 0; i < 4; i++) 651 put_unaligned_le32(crc32(0, &buf[i * 4], 4), &buf[i * 4]); 652 crypto_xor(&buf[0], &buf[12], 4); 653 crypto_xor(&buf[4], &buf[8], 4); 654 655 /* apply whitening (8 bytes) to whole sector */ 656 for (i = 0; i < ((1 << SECTOR_SHIFT) / 8); i++) 657 crypto_xor(data + i * 8, buf, 8); 658 memzero_explicit(buf, sizeof(buf)); 659 } 660 661 static int crypt_iv_tcw_gen(struct crypt_config *cc, u8 *iv, 662 struct dm_crypt_request *dmreq) 663 { 664 struct scatterlist *sg; 665 struct iv_tcw_private *tcw = &cc->iv_gen_private.tcw; 666 __le64 sector = cpu_to_le64(dmreq->iv_sector); 667 u8 *src; 668 669 /* Remove whitening from ciphertext */ 670 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 671 sg = crypt_get_sg_data(cc, dmreq->sg_in); 672 src = kmap_local_page(sg_page(sg)); 673 crypt_iv_tcw_whitening(cc, dmreq, src + sg->offset); 674 kunmap_local(src); 675 } 676 677 /* Calculate IV */ 678 crypto_xor_cpy(iv, tcw->iv_seed, (u8 *)§or, 8); 679 if (cc->iv_size > 8) 680 crypto_xor_cpy(&iv[8], tcw->iv_seed + 8, (u8 *)§or, 681 cc->iv_size - 8); 682 683 return 0; 684 } 685 686 static void crypt_iv_tcw_post(struct crypt_config *cc, u8 *iv, 687 struct dm_crypt_request *dmreq) 688 { 689 struct scatterlist *sg; 690 u8 *dst; 691 692 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 693 return; 694 695 /* Apply whitening on ciphertext */ 696 sg = crypt_get_sg_data(cc, dmreq->sg_out); 697 dst = kmap_local_page(sg_page(sg)); 698 crypt_iv_tcw_whitening(cc, dmreq, dst + sg->offset); 699 kunmap_local(dst); 700 } 701 702 static int crypt_iv_random_gen(struct crypt_config *cc, u8 *iv, 703 struct dm_crypt_request *dmreq) 704 { 705 /* Used only for writes, there must be an additional space to store IV */ 706 get_random_bytes(iv, cc->iv_size); 707 return 0; 708 } 709 710 static int crypt_iv_eboiv_ctr(struct crypt_config *cc, struct dm_target *ti, 711 const char *opts) 712 { 713 if (crypt_integrity_aead(cc)) { 714 ti->error = "AEAD transforms not supported for EBOIV"; 715 return -EINVAL; 716 } 717 718 if (crypto_skcipher_blocksize(any_tfm(cc)) != cc->iv_size) { 719 ti->error = "Block size of EBOIV cipher does not match IV size of block cipher"; 720 return -EINVAL; 721 } 722 723 return 0; 724 } 725 726 static int crypt_iv_eboiv_gen(struct crypt_config *cc, u8 *iv, 727 struct dm_crypt_request *dmreq) 728 { 729 struct crypto_skcipher *tfm = any_tfm(cc); 730 struct skcipher_request *req; 731 struct scatterlist src, dst; 732 DECLARE_CRYPTO_WAIT(wait); 733 unsigned int reqsize; 734 int err; 735 u8 *buf; 736 737 reqsize = sizeof(*req) + crypto_skcipher_reqsize(tfm); 738 reqsize = ALIGN(reqsize, __alignof__(__le64)); 739 740 req = kmalloc(reqsize + cc->iv_size, GFP_NOIO); 741 if (!req) 742 return -ENOMEM; 743 744 skcipher_request_set_tfm(req, tfm); 745 746 buf = (u8 *)req + reqsize; 747 memset(buf, 0, cc->iv_size); 748 *(__le64 *)buf = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 749 750 sg_init_one(&src, page_address(ZERO_PAGE(0)), cc->iv_size); 751 sg_init_one(&dst, iv, cc->iv_size); 752 skcipher_request_set_crypt(req, &src, &dst, cc->iv_size, buf); 753 skcipher_request_set_callback(req, 0, crypto_req_done, &wait); 754 err = crypto_wait_req(crypto_skcipher_encrypt(req), &wait); 755 kfree_sensitive(req); 756 757 return err; 758 } 759 760 static void crypt_iv_elephant_dtr(struct crypt_config *cc) 761 { 762 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 763 764 kfree_sensitive(elephant->key); 765 elephant->key = NULL; 766 } 767 768 static int crypt_iv_elephant_ctr(struct crypt_config *cc, struct dm_target *ti, 769 const char *opts) 770 { 771 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 772 int r; 773 774 elephant->key = kmalloc_obj(*elephant->key); 775 if (!elephant->key) 776 return -ENOMEM; 777 778 r = crypt_iv_eboiv_ctr(cc, ti, NULL); 779 if (r) 780 crypt_iv_elephant_dtr(cc); 781 return r; 782 } 783 784 static void diffuser_disk_to_cpu(u32 *d, size_t n) 785 { 786 #ifndef __LITTLE_ENDIAN 787 int i; 788 789 for (i = 0; i < n; i++) 790 d[i] = le32_to_cpu((__le32)d[i]); 791 #endif 792 } 793 794 static void diffuser_cpu_to_disk(__le32 *d, size_t n) 795 { 796 #ifndef __LITTLE_ENDIAN 797 int i; 798 799 for (i = 0; i < n; i++) 800 d[i] = cpu_to_le32((u32)d[i]); 801 #endif 802 } 803 804 static void diffuser_a_decrypt(u32 *d, size_t n) 805 { 806 int i, i1, i2, i3; 807 808 for (i = 0; i < 5; i++) { 809 i1 = 0; 810 i2 = n - 2; 811 i3 = n - 5; 812 813 while (i1 < (n - 1)) { 814 d[i1] += d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 815 i1++; i2++; i3++; 816 817 if (i3 >= n) 818 i3 -= n; 819 820 d[i1] += d[i2] ^ d[i3]; 821 i1++; i2++; i3++; 822 823 if (i2 >= n) 824 i2 -= n; 825 826 d[i1] += d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 827 i1++; i2++; i3++; 828 829 d[i1] += d[i2] ^ d[i3]; 830 i1++; i2++; i3++; 831 } 832 } 833 } 834 835 static void diffuser_a_encrypt(u32 *d, size_t n) 836 { 837 int i, i1, i2, i3; 838 839 for (i = 0; i < 5; i++) { 840 i1 = n - 1; 841 i2 = n - 2 - 1; 842 i3 = n - 5 - 1; 843 844 while (i1 > 0) { 845 d[i1] -= d[i2] ^ d[i3]; 846 i1--; i2--; i3--; 847 848 d[i1] -= d[i2] ^ (d[i3] << 13 | d[i3] >> 19); 849 i1--; i2--; i3--; 850 851 if (i2 < 0) 852 i2 += n; 853 854 d[i1] -= d[i2] ^ d[i3]; 855 i1--; i2--; i3--; 856 857 if (i3 < 0) 858 i3 += n; 859 860 d[i1] -= d[i2] ^ (d[i3] << 9 | d[i3] >> 23); 861 i1--; i2--; i3--; 862 } 863 } 864 } 865 866 static void diffuser_b_decrypt(u32 *d, size_t n) 867 { 868 int i, i1, i2, i3; 869 870 for (i = 0; i < 3; i++) { 871 i1 = 0; 872 i2 = 2; 873 i3 = 5; 874 875 while (i1 < (n - 1)) { 876 d[i1] += d[i2] ^ d[i3]; 877 i1++; i2++; i3++; 878 879 d[i1] += d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 880 i1++; i2++; i3++; 881 882 if (i2 >= n) 883 i2 -= n; 884 885 d[i1] += d[i2] ^ d[i3]; 886 i1++; i2++; i3++; 887 888 if (i3 >= n) 889 i3 -= n; 890 891 d[i1] += d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 892 i1++; i2++; i3++; 893 } 894 } 895 } 896 897 static void diffuser_b_encrypt(u32 *d, size_t n) 898 { 899 int i, i1, i2, i3; 900 901 for (i = 0; i < 3; i++) { 902 i1 = n - 1; 903 i2 = 2 - 1; 904 i3 = 5 - 1; 905 906 while (i1 > 0) { 907 d[i1] -= d[i2] ^ (d[i3] << 25 | d[i3] >> 7); 908 i1--; i2--; i3--; 909 910 if (i3 < 0) 911 i3 += n; 912 913 d[i1] -= d[i2] ^ d[i3]; 914 i1--; i2--; i3--; 915 916 if (i2 < 0) 917 i2 += n; 918 919 d[i1] -= d[i2] ^ (d[i3] << 10 | d[i3] >> 22); 920 i1--; i2--; i3--; 921 922 d[i1] -= d[i2] ^ d[i3]; 923 i1--; i2--; i3--; 924 } 925 } 926 } 927 928 static void crypt_iv_elephant(struct crypt_config *cc, 929 struct dm_crypt_request *dmreq) 930 { 931 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 932 u8 *data, *data2, *data_offset; 933 struct scatterlist *sg, *sg2; 934 union { 935 __le64 w[2]; 936 u8 b[16]; 937 } es; 938 u8 ks[32] __aligned(__alignof(long)); /* Elephant sector key */ 939 int i; 940 941 es.w[0] = cpu_to_le64(dmreq->iv_sector * cc->sector_size); 942 es.w[1] = 0; 943 944 /* E(Ks, e(s)) */ 945 aes_encrypt(elephant->key, &ks[0], es.b); 946 947 /* E(Ks, e'(s)) */ 948 es.b[15] = 0x80; 949 aes_encrypt(elephant->key, &ks[16], es.b); 950 951 sg = crypt_get_sg_data(cc, dmreq->sg_out); 952 data = kmap_local_page(sg_page(sg)); 953 data_offset = data + sg->offset; 954 955 /* Cannot modify original bio, copy to sg_out and apply Elephant to it */ 956 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 957 sg2 = crypt_get_sg_data(cc, dmreq->sg_in); 958 data2 = kmap_local_page(sg_page(sg2)); 959 memcpy(data_offset, data2 + sg2->offset, cc->sector_size); 960 kunmap_local(data2); 961 } 962 963 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) { 964 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32)); 965 diffuser_b_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 966 diffuser_a_decrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 967 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32)); 968 } 969 970 for (i = 0; i < (cc->sector_size / 32); i++) 971 crypto_xor(data_offset + i * 32, ks, 32); 972 973 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) { 974 diffuser_disk_to_cpu((u32 *)data_offset, cc->sector_size / sizeof(u32)); 975 diffuser_a_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 976 diffuser_b_encrypt((u32 *)data_offset, cc->sector_size / sizeof(u32)); 977 diffuser_cpu_to_disk((__le32 *)data_offset, cc->sector_size / sizeof(u32)); 978 } 979 980 kunmap_local(data); 981 memzero_explicit(ks, sizeof(ks)); 982 memzero_explicit(&es, sizeof(es)); 983 } 984 985 static int crypt_iv_elephant_gen(struct crypt_config *cc, u8 *iv, 986 struct dm_crypt_request *dmreq) 987 { 988 if (bio_data_dir(dmreq->ctx->bio_in) == WRITE) 989 crypt_iv_elephant(cc, dmreq); 990 991 return crypt_iv_eboiv_gen(cc, iv, dmreq); 992 } 993 994 static void crypt_iv_elephant_post(struct crypt_config *cc, u8 *iv, 995 struct dm_crypt_request *dmreq) 996 { 997 if (bio_data_dir(dmreq->ctx->bio_in) != WRITE) 998 crypt_iv_elephant(cc, dmreq); 999 } 1000 1001 static int crypt_iv_elephant_init(struct crypt_config *cc) 1002 { 1003 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1004 int key_offset = cc->key_size - cc->key_extra_size; 1005 1006 return aes_prepareenckey(elephant->key, &cc->key[key_offset], cc->key_extra_size); 1007 } 1008 1009 static void crypt_iv_elephant_wipe(struct crypt_config *cc) 1010 { 1011 struct iv_elephant_private *elephant = &cc->iv_gen_private.elephant; 1012 1013 memzero_explicit(elephant->key, sizeof(*elephant->key)); 1014 } 1015 1016 static const struct crypt_iv_operations crypt_iv_plain_ops = { 1017 .generator = crypt_iv_plain_gen 1018 }; 1019 1020 static const struct crypt_iv_operations crypt_iv_plain64_ops = { 1021 .generator = crypt_iv_plain64_gen 1022 }; 1023 1024 static const struct crypt_iv_operations crypt_iv_plain64be_ops = { 1025 .generator = crypt_iv_plain64be_gen 1026 }; 1027 1028 static const struct crypt_iv_operations crypt_iv_essiv_ops = { 1029 .generator = crypt_iv_essiv_gen 1030 }; 1031 1032 static const struct crypt_iv_operations crypt_iv_benbi_ops = { 1033 .ctr = crypt_iv_benbi_ctr, 1034 .dtr = crypt_iv_benbi_dtr, 1035 .generator = crypt_iv_benbi_gen 1036 }; 1037 1038 static const struct crypt_iv_operations crypt_iv_null_ops = { 1039 .generator = crypt_iv_null_gen 1040 }; 1041 1042 static const struct crypt_iv_operations crypt_iv_lmk_ops = { 1043 .ctr = crypt_iv_lmk_ctr, 1044 .dtr = crypt_iv_lmk_dtr, 1045 .init = crypt_iv_lmk_init, 1046 .wipe = crypt_iv_lmk_wipe, 1047 .generator = crypt_iv_lmk_gen, 1048 .post = crypt_iv_lmk_post 1049 }; 1050 1051 static const struct crypt_iv_operations crypt_iv_tcw_ops = { 1052 .ctr = crypt_iv_tcw_ctr, 1053 .dtr = crypt_iv_tcw_dtr, 1054 .init = crypt_iv_tcw_init, 1055 .wipe = crypt_iv_tcw_wipe, 1056 .generator = crypt_iv_tcw_gen, 1057 .post = crypt_iv_tcw_post 1058 }; 1059 1060 static const struct crypt_iv_operations crypt_iv_random_ops = { 1061 .generator = crypt_iv_random_gen 1062 }; 1063 1064 static const struct crypt_iv_operations crypt_iv_eboiv_ops = { 1065 .ctr = crypt_iv_eboiv_ctr, 1066 .generator = crypt_iv_eboiv_gen 1067 }; 1068 1069 static const struct crypt_iv_operations crypt_iv_elephant_ops = { 1070 .ctr = crypt_iv_elephant_ctr, 1071 .dtr = crypt_iv_elephant_dtr, 1072 .init = crypt_iv_elephant_init, 1073 .wipe = crypt_iv_elephant_wipe, 1074 .generator = crypt_iv_elephant_gen, 1075 .post = crypt_iv_elephant_post 1076 }; 1077 1078 /* 1079 * Integrity extensions 1080 */ 1081 static bool crypt_integrity_aead(struct crypt_config *cc) 1082 { 1083 return test_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 1084 } 1085 1086 static bool crypt_integrity_hmac(struct crypt_config *cc) 1087 { 1088 return crypt_integrity_aead(cc) && cc->key_mac_size; 1089 } 1090 1091 /* Get sg containing data */ 1092 static struct scatterlist *crypt_get_sg_data(struct crypt_config *cc, 1093 struct scatterlist *sg) 1094 { 1095 if (unlikely(crypt_integrity_aead(cc))) 1096 return &sg[2]; 1097 1098 return sg; 1099 } 1100 1101 static int dm_crypt_integrity_io_alloc(struct dm_crypt_io *io, struct bio *bio) 1102 { 1103 struct bio_integrity_payload *bip; 1104 unsigned int tag_len; 1105 int ret; 1106 1107 if (!bio_sectors(bio) || !io->cc->tuple_size) 1108 return 0; 1109 1110 bip = bio_integrity_alloc(bio, GFP_NOIO, 1); 1111 if (IS_ERR(bip)) 1112 return PTR_ERR(bip); 1113 1114 tag_len = io->cc->tuple_size * (bio_sectors(bio) >> io->cc->sector_shift); 1115 1116 bip->bip_iter.bi_sector = bio->bi_iter.bi_sector; 1117 1118 ret = bio_integrity_add_page(bio, virt_to_page(io->integrity_metadata), 1119 tag_len, offset_in_page(io->integrity_metadata)); 1120 if (unlikely(ret != tag_len)) 1121 return -ENOMEM; 1122 1123 return 0; 1124 } 1125 1126 static int crypt_integrity_ctr(struct crypt_config *cc, struct dm_target *ti) 1127 { 1128 #ifdef CONFIG_BLK_DEV_INTEGRITY 1129 struct blk_integrity *bi = blk_get_integrity(cc->dev->bdev->bd_disk); 1130 struct mapped_device *md = dm_table_get_md(ti->table); 1131 1132 /* We require an underlying device with non-PI metadata */ 1133 if (!bi || bi->csum_type != BLK_INTEGRITY_CSUM_NONE) { 1134 ti->error = "Integrity profile not supported."; 1135 return -EINVAL; 1136 } 1137 1138 if (bi->metadata_size < cc->used_tag_size) { 1139 ti->error = "Integrity profile tag size mismatch."; 1140 return -EINVAL; 1141 } 1142 cc->tuple_size = bi->metadata_size; 1143 if (1 << bi->interval_exp != cc->sector_size) { 1144 ti->error = "Integrity profile sector size mismatch."; 1145 return -EINVAL; 1146 } 1147 1148 if (crypt_integrity_aead(cc)) { 1149 cc->integrity_tag_size = cc->used_tag_size - cc->integrity_iv_size; 1150 DMDEBUG("%s: Integrity AEAD, tag size %u, IV size %u.", dm_device_name(md), 1151 cc->integrity_tag_size, cc->integrity_iv_size); 1152 1153 if (crypto_aead_setauthsize(any_tfm_aead(cc), cc->integrity_tag_size)) { 1154 ti->error = "Integrity AEAD auth tag size is not supported."; 1155 return -EINVAL; 1156 } 1157 } else if (cc->integrity_iv_size) 1158 DMDEBUG("%s: Additional per-sector space %u bytes for IV.", dm_device_name(md), 1159 cc->integrity_iv_size); 1160 1161 if ((cc->integrity_tag_size + cc->integrity_iv_size) > cc->tuple_size) { 1162 ti->error = "Not enough space for integrity tag in the profile."; 1163 return -EINVAL; 1164 } 1165 1166 return 0; 1167 #else 1168 ti->error = "Integrity profile not supported."; 1169 return -EINVAL; 1170 #endif 1171 } 1172 1173 static void crypt_convert_init(struct crypt_config *cc, 1174 struct convert_context *ctx, 1175 struct bio *bio_out, struct bio *bio_in, 1176 sector_t sector) 1177 { 1178 ctx->bio_in = bio_in; 1179 ctx->bio_out = bio_out; 1180 if (bio_in) 1181 ctx->iter_in = bio_in->bi_iter; 1182 if (bio_out) 1183 ctx->iter_out = bio_out->bi_iter; 1184 ctx->cc_sector = sector + cc->iv_offset; 1185 ctx->tag_offset = 0; 1186 init_completion(&ctx->restart); 1187 } 1188 1189 static struct dm_crypt_request *dmreq_of_req(struct crypt_config *cc, 1190 void *req) 1191 { 1192 return (struct dm_crypt_request *)((char *)req + cc->dmreq_start); 1193 } 1194 1195 static void *req_of_dmreq(struct crypt_config *cc, struct dm_crypt_request *dmreq) 1196 { 1197 return (void *)((char *)dmreq - cc->dmreq_start); 1198 } 1199 1200 static u8 *iv_of_dmreq(struct crypt_config *cc, 1201 struct dm_crypt_request *dmreq) 1202 { 1203 if (crypt_integrity_aead(cc)) 1204 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1205 crypto_aead_alignmask(any_tfm_aead(cc)) + 1); 1206 else 1207 return (u8 *)ALIGN((unsigned long)(dmreq + 1), 1208 crypto_skcipher_alignmask(any_tfm(cc)) + 1); 1209 } 1210 1211 static u8 *org_iv_of_dmreq(struct crypt_config *cc, 1212 struct dm_crypt_request *dmreq) 1213 { 1214 return iv_of_dmreq(cc, dmreq) + cc->iv_size; 1215 } 1216 1217 static __le64 *org_sector_of_dmreq(struct crypt_config *cc, 1218 struct dm_crypt_request *dmreq) 1219 { 1220 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + cc->iv_size; 1221 1222 return (__le64 *) ptr; 1223 } 1224 1225 static unsigned int *org_tag_of_dmreq(struct crypt_config *cc, 1226 struct dm_crypt_request *dmreq) 1227 { 1228 u8 *ptr = iv_of_dmreq(cc, dmreq) + cc->iv_size + 1229 cc->iv_size + sizeof(uint64_t); 1230 1231 return (unsigned int *)ptr; 1232 } 1233 1234 static void *tag_from_dmreq(struct crypt_config *cc, 1235 struct dm_crypt_request *dmreq) 1236 { 1237 struct convert_context *ctx = dmreq->ctx; 1238 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 1239 1240 return &io->integrity_metadata[*org_tag_of_dmreq(cc, dmreq) * 1241 cc->tuple_size]; 1242 } 1243 1244 static void *iv_tag_from_dmreq(struct crypt_config *cc, 1245 struct dm_crypt_request *dmreq) 1246 { 1247 return tag_from_dmreq(cc, dmreq) + cc->integrity_tag_size; 1248 } 1249 1250 static int crypt_convert_block_aead(struct crypt_config *cc, 1251 struct convert_context *ctx, 1252 struct aead_request *req, 1253 unsigned int tag_offset) 1254 { 1255 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1256 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1257 struct dm_crypt_request *dmreq; 1258 u8 *iv, *org_iv, *tag_iv, *tag; 1259 __le64 *sector; 1260 int r = 0; 1261 1262 BUG_ON(cc->integrity_iv_size && cc->integrity_iv_size != cc->iv_size); 1263 1264 /* Reject unexpected unaligned bio. */ 1265 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1266 return -EIO; 1267 1268 dmreq = dmreq_of_req(cc, req); 1269 dmreq->iv_sector = ctx->cc_sector; 1270 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1271 dmreq->iv_sector >>= cc->sector_shift; 1272 dmreq->ctx = ctx; 1273 1274 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1275 1276 sector = org_sector_of_dmreq(cc, dmreq); 1277 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1278 1279 iv = iv_of_dmreq(cc, dmreq); 1280 org_iv = org_iv_of_dmreq(cc, dmreq); 1281 tag = tag_from_dmreq(cc, dmreq); 1282 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1283 1284 /* AEAD request: 1285 * |----- AAD -------|------ DATA -------|-- AUTH TAG --| 1286 * | (authenticated) | (auth+encryption) | | 1287 * | sector_LE | IV | sector in/out | tag in/out | 1288 */ 1289 sg_init_table(dmreq->sg_in, 4); 1290 sg_set_buf(&dmreq->sg_in[0], sector, sizeof(uint64_t)); 1291 sg_set_buf(&dmreq->sg_in[1], org_iv, cc->iv_size); 1292 sg_set_page(&dmreq->sg_in[2], bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1293 sg_set_buf(&dmreq->sg_in[3], tag, cc->integrity_tag_size); 1294 1295 sg_init_table(dmreq->sg_out, 4); 1296 sg_set_buf(&dmreq->sg_out[0], sector, sizeof(uint64_t)); 1297 sg_set_buf(&dmreq->sg_out[1], org_iv, cc->iv_size); 1298 sg_set_page(&dmreq->sg_out[2], bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1299 sg_set_buf(&dmreq->sg_out[3], tag, cc->integrity_tag_size); 1300 1301 if (cc->iv_gen_ops) { 1302 /* For READs use IV stored in integrity metadata */ 1303 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1304 memcpy(org_iv, tag_iv, cc->iv_size); 1305 } else { 1306 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1307 if (r < 0) 1308 return r; 1309 /* Store generated IV in integrity metadata */ 1310 if (cc->integrity_iv_size) 1311 memcpy(tag_iv, org_iv, cc->iv_size); 1312 } 1313 /* Working copy of IV, to be modified in crypto API */ 1314 memcpy(iv, org_iv, cc->iv_size); 1315 } 1316 1317 aead_request_set_ad(req, sizeof(uint64_t) + cc->iv_size); 1318 if (bio_data_dir(ctx->bio_in) == WRITE) { 1319 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1320 cc->sector_size, iv); 1321 r = crypto_aead_encrypt(req); 1322 if (cc->integrity_tag_size + cc->integrity_iv_size != cc->tuple_size) 1323 memset(tag + cc->integrity_tag_size + cc->integrity_iv_size, 0, 1324 cc->tuple_size - (cc->integrity_tag_size + cc->integrity_iv_size)); 1325 } else { 1326 aead_request_set_crypt(req, dmreq->sg_in, dmreq->sg_out, 1327 cc->sector_size + cc->integrity_tag_size, iv); 1328 r = crypto_aead_decrypt(req); 1329 } 1330 1331 if (r == -EBADMSG) { 1332 sector_t s = le64_to_cpu(*sector); 1333 1334 ctx->aead_failed = true; 1335 if (ctx->aead_recheck) { 1336 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu", 1337 ctx->bio_in->bi_bdev, s); 1338 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead", 1339 ctx->bio_in, s, 0); 1340 } 1341 } 1342 1343 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1344 cc->iv_gen_ops->post(cc, org_iv, dmreq); 1345 1346 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1347 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1348 1349 return r; 1350 } 1351 1352 static int crypt_convert_block_skcipher(struct crypt_config *cc, 1353 struct convert_context *ctx, 1354 struct skcipher_request *req, 1355 unsigned int tag_offset) 1356 { 1357 struct bio_vec bv_in = bio_iter_iovec(ctx->bio_in, ctx->iter_in); 1358 struct bio_vec bv_out = bio_iter_iovec(ctx->bio_out, ctx->iter_out); 1359 struct scatterlist *sg_in, *sg_out; 1360 struct dm_crypt_request *dmreq; 1361 u8 *iv, *org_iv, *tag_iv; 1362 __le64 *sector; 1363 int r = 0; 1364 1365 /* Reject unexpected unaligned bio. */ 1366 if (unlikely(bv_in.bv_len & (cc->sector_size - 1))) 1367 return -EIO; 1368 1369 dmreq = dmreq_of_req(cc, req); 1370 dmreq->iv_sector = ctx->cc_sector; 1371 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 1372 dmreq->iv_sector >>= cc->sector_shift; 1373 dmreq->ctx = ctx; 1374 1375 *org_tag_of_dmreq(cc, dmreq) = tag_offset; 1376 1377 iv = iv_of_dmreq(cc, dmreq); 1378 org_iv = org_iv_of_dmreq(cc, dmreq); 1379 tag_iv = iv_tag_from_dmreq(cc, dmreq); 1380 1381 sector = org_sector_of_dmreq(cc, dmreq); 1382 *sector = cpu_to_le64(ctx->cc_sector - cc->iv_offset); 1383 1384 /* For skcipher we use only the first sg item */ 1385 sg_in = &dmreq->sg_in[0]; 1386 sg_out = &dmreq->sg_out[0]; 1387 1388 sg_init_table(sg_in, 1); 1389 sg_set_page(sg_in, bv_in.bv_page, cc->sector_size, bv_in.bv_offset); 1390 1391 sg_init_table(sg_out, 1); 1392 sg_set_page(sg_out, bv_out.bv_page, cc->sector_size, bv_out.bv_offset); 1393 1394 if (cc->iv_gen_ops) { 1395 /* For READs use IV stored in integrity metadata */ 1396 if (cc->integrity_iv_size && bio_data_dir(ctx->bio_in) != WRITE) { 1397 memcpy(org_iv, tag_iv, cc->integrity_iv_size); 1398 } else { 1399 r = cc->iv_gen_ops->generator(cc, org_iv, dmreq); 1400 if (r < 0) 1401 return r; 1402 /* Data can be already preprocessed in generator */ 1403 if (test_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags)) 1404 sg_in = sg_out; 1405 /* Store generated IV in integrity metadata */ 1406 if (cc->integrity_iv_size) 1407 memcpy(tag_iv, org_iv, cc->integrity_iv_size); 1408 } 1409 /* Working copy of IV, to be modified in crypto API */ 1410 memcpy(iv, org_iv, cc->iv_size); 1411 } 1412 1413 skcipher_request_set_crypt(req, sg_in, sg_out, cc->sector_size, iv); 1414 1415 if (bio_data_dir(ctx->bio_in) == WRITE) 1416 r = crypto_skcipher_encrypt(req); 1417 else 1418 r = crypto_skcipher_decrypt(req); 1419 1420 if (!r && cc->iv_gen_ops && cc->iv_gen_ops->post) 1421 cc->iv_gen_ops->post(cc, org_iv, dmreq); 1422 1423 bio_advance_iter(ctx->bio_in, &ctx->iter_in, cc->sector_size); 1424 bio_advance_iter(ctx->bio_out, &ctx->iter_out, cc->sector_size); 1425 1426 return r; 1427 } 1428 1429 static void kcryptd_async_done(void *async_req, int error); 1430 1431 static int crypt_alloc_req_skcipher(struct crypt_config *cc, 1432 struct convert_context *ctx) 1433 { 1434 unsigned int key_index = ctx->cc_sector & (cc->tfms_count - 1); 1435 1436 if (!ctx->r.req) { 1437 ctx->r.req = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1438 if (!ctx->r.req) 1439 return -ENOMEM; 1440 } 1441 1442 skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]); 1443 1444 /* 1445 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1446 * requests if driver request queue is full. 1447 */ 1448 skcipher_request_set_callback(ctx->r.req, 1449 CRYPTO_TFM_REQ_MAY_BACKLOG, 1450 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req)); 1451 1452 return 0; 1453 } 1454 1455 static int crypt_alloc_req_aead(struct crypt_config *cc, 1456 struct convert_context *ctx) 1457 { 1458 if (!ctx->r.req_aead) { 1459 ctx->r.req_aead = mempool_alloc(&cc->req_pool, in_interrupt() ? GFP_ATOMIC : GFP_NOIO); 1460 if (!ctx->r.req_aead) 1461 return -ENOMEM; 1462 } 1463 1464 aead_request_set_tfm(ctx->r.req_aead, cc->cipher_tfm.tfms_aead[0]); 1465 1466 /* 1467 * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs 1468 * requests if driver request queue is full. 1469 */ 1470 aead_request_set_callback(ctx->r.req_aead, 1471 CRYPTO_TFM_REQ_MAY_BACKLOG, 1472 kcryptd_async_done, dmreq_of_req(cc, ctx->r.req_aead)); 1473 1474 return 0; 1475 } 1476 1477 static int crypt_alloc_req(struct crypt_config *cc, 1478 struct convert_context *ctx) 1479 { 1480 if (crypt_integrity_aead(cc)) 1481 return crypt_alloc_req_aead(cc, ctx); 1482 else 1483 return crypt_alloc_req_skcipher(cc, ctx); 1484 } 1485 1486 static void crypt_free_req_skcipher(struct crypt_config *cc, 1487 struct skcipher_request *req, struct bio *base_bio) 1488 { 1489 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1490 1491 if ((struct skcipher_request *)(io + 1) != req) 1492 mempool_free(req, &cc->req_pool); 1493 } 1494 1495 static void crypt_free_req_aead(struct crypt_config *cc, 1496 struct aead_request *req, struct bio *base_bio) 1497 { 1498 struct dm_crypt_io *io = dm_per_bio_data(base_bio, cc->per_bio_data_size); 1499 1500 if ((struct aead_request *)(io + 1) != req) 1501 mempool_free(req, &cc->req_pool); 1502 } 1503 1504 static void crypt_free_req(struct crypt_config *cc, void *req, struct bio *base_bio) 1505 { 1506 if (crypt_integrity_aead(cc)) 1507 crypt_free_req_aead(cc, req, base_bio); 1508 else 1509 crypt_free_req_skcipher(cc, req, base_bio); 1510 } 1511 1512 /* 1513 * Encrypt / decrypt data from one bio to another one (can be the same one) 1514 */ 1515 static blk_status_t crypt_convert(struct crypt_config *cc, 1516 struct convert_context *ctx, bool atomic, bool reset_pending) 1517 { 1518 unsigned int sector_step = cc->sector_size >> SECTOR_SHIFT; 1519 int r; 1520 1521 /* 1522 * if reset_pending is set we are dealing with the bio for the first time, 1523 * else we're continuing to work on the previous bio, so don't mess with 1524 * the cc_pending counter 1525 */ 1526 if (reset_pending) 1527 atomic_set(&ctx->cc_pending, 1); 1528 1529 while (ctx->iter_in.bi_size && ctx->iter_out.bi_size) { 1530 1531 r = crypt_alloc_req(cc, ctx); 1532 if (r) { 1533 complete(&ctx->restart); 1534 return BLK_STS_DEV_RESOURCE; 1535 } 1536 1537 atomic_inc(&ctx->cc_pending); 1538 1539 if (crypt_integrity_aead(cc)) 1540 r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead, ctx->tag_offset); 1541 else 1542 r = crypt_convert_block_skcipher(cc, ctx, ctx->r.req, ctx->tag_offset); 1543 1544 switch (r) { 1545 /* 1546 * The request was queued by a crypto driver 1547 * but the driver request queue is full, let's wait. 1548 */ 1549 case -EBUSY: 1550 if (in_interrupt()) { 1551 if (try_wait_for_completion(&ctx->restart)) { 1552 /* 1553 * we don't have to block to wait for completion, 1554 * so proceed 1555 */ 1556 } else { 1557 /* 1558 * we can't wait for completion without blocking 1559 * exit and continue processing in a workqueue 1560 */ 1561 ctx->r.req = NULL; 1562 ctx->tag_offset++; 1563 ctx->cc_sector += sector_step; 1564 return BLK_STS_DEV_RESOURCE; 1565 } 1566 } else { 1567 wait_for_completion(&ctx->restart); 1568 } 1569 reinit_completion(&ctx->restart); 1570 fallthrough; 1571 /* 1572 * The request is queued and processed asynchronously, 1573 * completion function kcryptd_async_done() will be called. 1574 */ 1575 case -EINPROGRESS: 1576 ctx->r.req = NULL; 1577 ctx->tag_offset++; 1578 ctx->cc_sector += sector_step; 1579 continue; 1580 /* 1581 * The request was already processed (synchronously). 1582 */ 1583 case 0: 1584 atomic_dec(&ctx->cc_pending); 1585 ctx->cc_sector += sector_step; 1586 ctx->tag_offset++; 1587 if (!atomic) 1588 cond_resched(); 1589 continue; 1590 /* 1591 * There was a data integrity error. 1592 */ 1593 case -EBADMSG: 1594 atomic_dec(&ctx->cc_pending); 1595 return BLK_STS_PROTECTION; 1596 /* 1597 * There was an error while processing the request. 1598 */ 1599 default: 1600 atomic_dec(&ctx->cc_pending); 1601 return BLK_STS_IOERR; 1602 } 1603 } 1604 1605 return 0; 1606 } 1607 1608 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone); 1609 1610 /* 1611 * Generate a new unfragmented bio with the given size 1612 * This should never violate the device limitations (but if it did then block 1613 * core should split the bio as needed). 1614 * 1615 * This function may be called concurrently. If we allocate from the mempool 1616 * concurrently, there is a possibility of deadlock. For example, if we have 1617 * mempool of 256 pages, two processes, each wanting 256, pages allocate from 1618 * the mempool concurrently, it may deadlock in a situation where both processes 1619 * have allocated 128 pages and the mempool is exhausted. 1620 * 1621 * In order to avoid this scenario we allocate the pages under a mutex. 1622 * 1623 * In order to not degrade performance with excessive locking, we try 1624 * non-blocking allocations without a mutex first but on failure we fallback 1625 * to blocking allocations with a mutex. 1626 * 1627 * In order to reduce allocation overhead, we try to allocate compound pages in 1628 * the first pass. If they are not available, we fall back to the mempool. 1629 */ 1630 static struct bio *crypt_alloc_buffer(struct dm_crypt_io *io, unsigned int size) 1631 { 1632 struct crypt_config *cc = io->cc; 1633 struct bio *clone; 1634 unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 1635 gfp_t gfp_mask = GFP_NOWAIT | __GFP_HIGHMEM; 1636 unsigned int remaining_size; 1637 unsigned int order = MAX_PAGE_ORDER; 1638 1639 retry: 1640 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1641 mutex_lock(&cc->bio_alloc_lock); 1642 1643 clone = bio_alloc_bioset(cc->dev->bdev, nr_iovecs, io->base_bio->bi_opf, 1644 GFP_NOIO, &cc->bs); 1645 clone->bi_private = io; 1646 clone->bi_end_io = crypt_endio; 1647 clone->bi_ioprio = io->base_bio->bi_ioprio; 1648 clone->bi_iter.bi_sector = cc->start + io->sector; 1649 1650 remaining_size = size; 1651 1652 while (remaining_size) { 1653 struct page *pages; 1654 unsigned size_to_add; 1655 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT); 1656 order = min(order, remaining_order); 1657 1658 while (order > 0) { 1659 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) + 1660 (1 << order) > dm_crypt_pages_per_client)) 1661 goto decrease_order; 1662 pages = alloc_pages(gfp_mask 1663 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP, 1664 order); 1665 if (likely(pages != NULL)) { 1666 percpu_counter_add(&cc->n_allocated_pages, 1 << order); 1667 goto have_pages; 1668 } 1669 decrease_order: 1670 order--; 1671 } 1672 1673 pages = mempool_alloc(&cc->page_pool, gfp_mask); 1674 if (!pages) { 1675 crypt_free_buffer_pages(cc, clone); 1676 bio_put(clone); 1677 gfp_mask |= __GFP_DIRECT_RECLAIM; 1678 order = 0; 1679 goto retry; 1680 } 1681 1682 have_pages: 1683 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size); 1684 __bio_add_page(clone, pages, size_to_add, 0); 1685 remaining_size -= size_to_add; 1686 } 1687 1688 /* Allocate space for integrity tags */ 1689 if (dm_crypt_integrity_io_alloc(io, clone)) { 1690 crypt_free_buffer_pages(cc, clone); 1691 bio_put(clone); 1692 clone = NULL; 1693 } 1694 1695 if (unlikely(gfp_mask & __GFP_DIRECT_RECLAIM)) 1696 mutex_unlock(&cc->bio_alloc_lock); 1697 1698 return clone; 1699 } 1700 1701 static void crypt_free_buffer_pages(struct crypt_config *cc, struct bio *clone) 1702 { 1703 struct folio_iter fi; 1704 1705 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */ 1706 bio_for_each_folio_all(fi, clone) { 1707 if (folio_test_large(fi.folio)) { 1708 percpu_counter_sub(&cc->n_allocated_pages, 1709 folio_nr_pages(fi.folio)); 1710 folio_put(fi.folio); 1711 } else { 1712 mempool_free(&fi.folio->page, &cc->page_pool); 1713 } 1714 } 1715 } 1716 } 1717 1718 static void crypt_io_init(struct dm_crypt_io *io, struct crypt_config *cc, 1719 struct bio *bio, sector_t sector) 1720 { 1721 io->cc = cc; 1722 io->base_bio = bio; 1723 io->sector = sector; 1724 io->error = 0; 1725 io->ctx.aead_recheck = false; 1726 io->ctx.aead_failed = false; 1727 io->ctx.r.req = NULL; 1728 io->integrity_metadata = NULL; 1729 io->integrity_metadata_from_pool = false; 1730 atomic_set(&io->io_pending, 0); 1731 } 1732 1733 static void crypt_inc_pending(struct dm_crypt_io *io) 1734 { 1735 atomic_inc(&io->io_pending); 1736 } 1737 1738 static void kcryptd_queue_read(struct dm_crypt_io *io); 1739 1740 /* 1741 * One of the bios was finished. Check for completion of 1742 * the whole request and correctly clean up the buffer. 1743 */ 1744 static void crypt_dec_pending(struct dm_crypt_io *io) 1745 { 1746 struct crypt_config *cc = io->cc; 1747 struct bio *base_bio = io->base_bio; 1748 1749 if (!atomic_dec_and_test(&io->io_pending)) 1750 return; 1751 1752 if (likely(!io->ctx.aead_recheck) && unlikely(io->ctx.aead_failed) && 1753 cc->used_tag_size && bio_data_dir(base_bio) == READ) { 1754 io->ctx.aead_recheck = true; 1755 io->ctx.aead_failed = false; 1756 io->error = 0; 1757 kcryptd_queue_read(io); 1758 return; 1759 } 1760 1761 if (io->ctx.r.req) 1762 crypt_free_req(cc, io->ctx.r.req, base_bio); 1763 1764 if (unlikely(io->integrity_metadata_from_pool)) 1765 mempool_free(io->integrity_metadata, &io->cc->tag_pool); 1766 else 1767 kfree(io->integrity_metadata); 1768 1769 base_bio->bi_status = io->error; 1770 1771 bio_endio(base_bio); 1772 } 1773 1774 /* 1775 * kcryptd/kcryptd_io: 1776 * 1777 * Needed because it would be very unwise to do decryption in an 1778 * interrupt context. 1779 * 1780 * kcryptd performs the actual encryption or decryption. 1781 * 1782 * kcryptd_io performs the IO submission. 1783 * 1784 * They must be separated as otherwise the final stages could be 1785 * starved by new requests which can block in the first stages due 1786 * to memory allocation. 1787 * 1788 * The work is done per CPU global for all dm-crypt instances. 1789 * They should not depend on each other and do not block. 1790 */ 1791 static void crypt_endio(struct bio *clone) 1792 { 1793 struct dm_crypt_io *io = clone->bi_private; 1794 struct crypt_config *cc = io->cc; 1795 unsigned int rw = bio_data_dir(clone); 1796 blk_status_t error = clone->bi_status; 1797 1798 if (io->ctx.aead_recheck && !error) { 1799 kcryptd_queue_crypt(io); 1800 return; 1801 } 1802 1803 /* 1804 * free the processed pages 1805 */ 1806 if (rw == WRITE || io->ctx.aead_recheck) 1807 crypt_free_buffer_pages(cc, clone); 1808 1809 bio_put(clone); 1810 1811 if (rw == READ && !error) { 1812 kcryptd_queue_crypt(io); 1813 return; 1814 } 1815 1816 if (unlikely(error)) 1817 io->error = error; 1818 1819 crypt_dec_pending(io); 1820 } 1821 1822 #define CRYPT_MAP_READ_GFP GFP_NOWAIT 1823 1824 static int kcryptd_io_read(struct dm_crypt_io *io, gfp_t gfp) 1825 { 1826 struct crypt_config *cc = io->cc; 1827 struct bio *clone; 1828 1829 if (io->ctx.aead_recheck) { 1830 if (!(gfp & __GFP_DIRECT_RECLAIM)) 1831 return 1; 1832 crypt_inc_pending(io); 1833 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size); 1834 if (unlikely(!clone)) { 1835 crypt_dec_pending(io); 1836 return 1; 1837 } 1838 crypt_convert_init(cc, &io->ctx, clone, clone, io->sector); 1839 io->saved_bi_iter = clone->bi_iter; 1840 dm_submit_bio_remap(io->base_bio, clone); 1841 return 0; 1842 } 1843 1844 /* 1845 * We need the original biovec array in order to decrypt the whole bio 1846 * data *afterwards* -- thanks to immutable biovecs we don't need to 1847 * worry about the block layer modifying the biovec array; so leverage 1848 * bio_alloc_clone(). 1849 */ 1850 clone = bio_alloc_clone(cc->dev->bdev, io->base_bio, gfp, &cc->bs); 1851 if (!clone) 1852 return 1; 1853 1854 clone->bi_iter.bi_sector = cc->start + io->sector; 1855 clone->bi_private = io; 1856 clone->bi_end_io = crypt_endio; 1857 1858 crypt_inc_pending(io); 1859 1860 if (dm_crypt_integrity_io_alloc(io, clone)) { 1861 crypt_dec_pending(io); 1862 bio_put(clone); 1863 return 1; 1864 } 1865 1866 dm_submit_bio_remap(io->base_bio, clone); 1867 return 0; 1868 } 1869 1870 static void kcryptd_io_read_work(struct work_struct *work) 1871 { 1872 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 1873 1874 crypt_inc_pending(io); 1875 if (kcryptd_io_read(io, GFP_NOIO)) 1876 io->error = BLK_STS_RESOURCE; 1877 crypt_dec_pending(io); 1878 } 1879 1880 static void kcryptd_queue_read(struct dm_crypt_io *io) 1881 { 1882 struct crypt_config *cc = io->cc; 1883 1884 INIT_WORK(&io->work, kcryptd_io_read_work); 1885 queue_work(cc->io_queue, &io->work); 1886 } 1887 1888 static void kcryptd_io_write(struct dm_crypt_io *io) 1889 { 1890 struct bio *clone = io->ctx.bio_out; 1891 1892 dm_submit_bio_remap(io->base_bio, clone); 1893 } 1894 1895 #define crypt_io_from_node(node) rb_entry((node), struct dm_crypt_io, rb_node) 1896 1897 static int dmcrypt_write(void *data) 1898 { 1899 struct crypt_config *cc = data; 1900 struct dm_crypt_io *io; 1901 1902 while (1) { 1903 struct rb_root write_tree; 1904 struct blk_plug plug; 1905 1906 spin_lock_irq(&cc->write_thread_lock); 1907 continue_locked: 1908 1909 if (!RB_EMPTY_ROOT(&cc->write_tree)) 1910 goto pop_from_list; 1911 1912 set_current_state(TASK_INTERRUPTIBLE); 1913 1914 spin_unlock_irq(&cc->write_thread_lock); 1915 1916 if (unlikely(kthread_should_stop())) { 1917 set_current_state(TASK_RUNNING); 1918 break; 1919 } 1920 1921 schedule(); 1922 1923 spin_lock_irq(&cc->write_thread_lock); 1924 goto continue_locked; 1925 1926 pop_from_list: 1927 write_tree = cc->write_tree; 1928 cc->write_tree = RB_ROOT; 1929 spin_unlock_irq(&cc->write_thread_lock); 1930 1931 BUG_ON(rb_parent(write_tree.rb_node)); 1932 1933 /* 1934 * Note: we cannot walk the tree here with rb_next because 1935 * the structures may be freed when kcryptd_io_write is called. 1936 */ 1937 blk_start_plug(&plug); 1938 do { 1939 io = crypt_io_from_node(rb_first(&write_tree)); 1940 rb_erase(&io->rb_node, &write_tree); 1941 kcryptd_io_write(io); 1942 cond_resched(); 1943 } while (!RB_EMPTY_ROOT(&write_tree)); 1944 blk_finish_plug(&plug); 1945 } 1946 return 0; 1947 } 1948 1949 static void kcryptd_crypt_write_io_submit(struct dm_crypt_io *io, int async) 1950 { 1951 struct bio *clone = io->ctx.bio_out; 1952 struct crypt_config *cc = io->cc; 1953 unsigned long flags; 1954 sector_t sector; 1955 struct rb_node **rbp, *parent; 1956 1957 if (unlikely(io->error)) { 1958 crypt_free_buffer_pages(cc, clone); 1959 bio_put(clone); 1960 crypt_dec_pending(io); 1961 return; 1962 } 1963 1964 /* crypt_convert should have filled the clone bio */ 1965 BUG_ON(io->ctx.iter_out.bi_size); 1966 1967 if ((likely(!async) && test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) || 1968 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) { 1969 dm_submit_bio_remap(io->base_bio, clone); 1970 return; 1971 } 1972 1973 spin_lock_irqsave(&cc->write_thread_lock, flags); 1974 if (RB_EMPTY_ROOT(&cc->write_tree)) 1975 wake_up_process(cc->write_thread); 1976 rbp = &cc->write_tree.rb_node; 1977 parent = NULL; 1978 sector = io->sector; 1979 while (*rbp) { 1980 parent = *rbp; 1981 if (sector < crypt_io_from_node(parent)->sector) 1982 rbp = &(*rbp)->rb_left; 1983 else 1984 rbp = &(*rbp)->rb_right; 1985 } 1986 rb_link_node(&io->rb_node, parent, rbp); 1987 rb_insert_color(&io->rb_node, &cc->write_tree); 1988 spin_unlock_irqrestore(&cc->write_thread_lock, flags); 1989 } 1990 1991 static bool kcryptd_crypt_write_inline(struct crypt_config *cc, 1992 struct convert_context *ctx) 1993 1994 { 1995 if (!test_bit(DM_CRYPT_WRITE_INLINE, &cc->flags)) 1996 return false; 1997 1998 /* 1999 * Note: zone append writes (REQ_OP_ZONE_APPEND) do not have ordering 2000 * constraints so they do not need to be issued inline by 2001 * kcryptd_crypt_write_convert(). 2002 */ 2003 switch (bio_op(ctx->bio_in)) { 2004 case REQ_OP_WRITE: 2005 case REQ_OP_WRITE_ZEROES: 2006 return true; 2007 default: 2008 return false; 2009 } 2010 } 2011 2012 static void kcryptd_crypt_write_continue(struct work_struct *work) 2013 { 2014 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2015 struct crypt_config *cc = io->cc; 2016 struct convert_context *ctx = &io->ctx; 2017 int crypt_finished; 2018 blk_status_t r; 2019 2020 wait_for_completion(&ctx->restart); 2021 reinit_completion(&ctx->restart); 2022 2023 r = crypt_convert(cc, &io->ctx, false, false); 2024 if (r) 2025 io->error = r; 2026 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2027 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2028 /* Wait for completion signaled by kcryptd_async_done() */ 2029 wait_for_completion(&ctx->restart); 2030 crypt_finished = 1; 2031 } 2032 2033 /* Encryption was already finished, submit io now */ 2034 if (crypt_finished) 2035 kcryptd_crypt_write_io_submit(io, 0); 2036 2037 crypt_dec_pending(io); 2038 } 2039 2040 static void kcryptd_crypt_write_convert(struct dm_crypt_io *io) 2041 { 2042 struct crypt_config *cc = io->cc; 2043 struct convert_context *ctx = &io->ctx; 2044 struct bio *clone; 2045 int crypt_finished; 2046 blk_status_t r; 2047 2048 /* 2049 * Prevent io from disappearing until this function completes. 2050 */ 2051 crypt_inc_pending(io); 2052 crypt_convert_init(cc, ctx, NULL, io->base_bio, io->sector); 2053 2054 clone = crypt_alloc_buffer(io, io->base_bio->bi_iter.bi_size); 2055 if (unlikely(!clone)) { 2056 io->error = BLK_STS_IOERR; 2057 goto dec; 2058 } 2059 2060 io->ctx.bio_out = clone; 2061 io->ctx.iter_out = clone->bi_iter; 2062 2063 if (crypt_integrity_aead(cc)) { 2064 bio_copy_data(clone, io->base_bio); 2065 io->ctx.bio_in = clone; 2066 io->ctx.iter_in = clone->bi_iter; 2067 } 2068 2069 crypt_inc_pending(io); 2070 r = crypt_convert(cc, ctx, 2071 test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags), true); 2072 /* 2073 * Crypto API backlogged the request, because its queue was full 2074 * and we're in softirq context, so continue from a workqueue 2075 * (TODO: is it actually possible to be in softirq in the write path?) 2076 */ 2077 if (r == BLK_STS_DEV_RESOURCE) { 2078 INIT_WORK(&io->work, kcryptd_crypt_write_continue); 2079 queue_work(cc->crypt_queue, &io->work); 2080 return; 2081 } 2082 if (r) 2083 io->error = r; 2084 crypt_finished = atomic_dec_and_test(&ctx->cc_pending); 2085 if (!crypt_finished && kcryptd_crypt_write_inline(cc, ctx)) { 2086 /* Wait for completion signaled by kcryptd_async_done() */ 2087 wait_for_completion(&ctx->restart); 2088 crypt_finished = 1; 2089 } 2090 2091 /* Encryption was already finished, submit io now */ 2092 if (crypt_finished) 2093 kcryptd_crypt_write_io_submit(io, 0); 2094 2095 dec: 2096 crypt_dec_pending(io); 2097 } 2098 2099 static void kcryptd_crypt_read_done(struct dm_crypt_io *io) 2100 { 2101 if (io->ctx.aead_recheck) { 2102 if (!io->error) { 2103 io->ctx.bio_in->bi_iter = io->saved_bi_iter; 2104 bio_copy_data(io->base_bio, io->ctx.bio_in); 2105 } 2106 crypt_free_buffer_pages(io->cc, io->ctx.bio_in); 2107 bio_put(io->ctx.bio_in); 2108 } 2109 crypt_dec_pending(io); 2110 } 2111 2112 static void kcryptd_crypt_read_continue(struct work_struct *work) 2113 { 2114 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2115 struct crypt_config *cc = io->cc; 2116 blk_status_t r; 2117 2118 wait_for_completion(&io->ctx.restart); 2119 reinit_completion(&io->ctx.restart); 2120 2121 r = crypt_convert(cc, &io->ctx, false, false); 2122 if (r) 2123 io->error = r; 2124 2125 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2126 kcryptd_crypt_read_done(io); 2127 2128 crypt_dec_pending(io); 2129 } 2130 2131 static void kcryptd_crypt_read_convert(struct dm_crypt_io *io) 2132 { 2133 struct crypt_config *cc = io->cc; 2134 blk_status_t r; 2135 2136 crypt_inc_pending(io); 2137 2138 if (io->ctx.aead_recheck) { 2139 r = crypt_convert(cc, &io->ctx, 2140 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true); 2141 } else { 2142 crypt_convert_init(cc, &io->ctx, io->base_bio, io->base_bio, 2143 io->sector); 2144 2145 r = crypt_convert(cc, &io->ctx, 2146 test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags), true); 2147 } 2148 /* 2149 * Crypto API backlogged the request, because its queue was full 2150 * and we're in softirq context, so continue from a workqueue 2151 */ 2152 if (r == BLK_STS_DEV_RESOURCE) { 2153 INIT_WORK(&io->work, kcryptd_crypt_read_continue); 2154 queue_work(cc->crypt_queue, &io->work); 2155 return; 2156 } 2157 if (r) 2158 io->error = r; 2159 2160 if (atomic_dec_and_test(&io->ctx.cc_pending)) 2161 kcryptd_crypt_read_done(io); 2162 2163 crypt_dec_pending(io); 2164 } 2165 2166 static void kcryptd_async_done(void *data, int error) 2167 { 2168 struct dm_crypt_request *dmreq = data; 2169 struct convert_context *ctx = dmreq->ctx; 2170 struct dm_crypt_io *io = container_of(ctx, struct dm_crypt_io, ctx); 2171 struct crypt_config *cc = io->cc; 2172 2173 /* 2174 * A request from crypto driver backlog is going to be processed now, 2175 * finish the completion and continue in crypt_convert(). 2176 * (Callback will be called for the second time for this request.) 2177 */ 2178 if (error == -EINPROGRESS) { 2179 complete(&ctx->restart); 2180 return; 2181 } 2182 2183 if (!error && cc->iv_gen_ops && cc->iv_gen_ops->post) 2184 cc->iv_gen_ops->post(cc, org_iv_of_dmreq(cc, dmreq), dmreq); 2185 2186 if (error == -EBADMSG) { 2187 sector_t s = le64_to_cpu(*org_sector_of_dmreq(cc, dmreq)); 2188 2189 ctx->aead_failed = true; 2190 if (ctx->aead_recheck) { 2191 DMERR_LIMIT("%pg: INTEGRITY AEAD ERROR, sector %llu", 2192 ctx->bio_in->bi_bdev, s); 2193 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-aead", 2194 ctx->bio_in, s, 0); 2195 } 2196 io->error = BLK_STS_PROTECTION; 2197 } else if (error < 0) 2198 io->error = BLK_STS_IOERR; 2199 2200 crypt_free_req(cc, req_of_dmreq(cc, dmreq), io->base_bio); 2201 2202 if (!atomic_dec_and_test(&ctx->cc_pending)) 2203 return; 2204 2205 /* 2206 * The request is fully completed: for inline writes, let 2207 * kcryptd_crypt_write_convert() do the IO submission. 2208 */ 2209 if (bio_data_dir(io->base_bio) == READ) { 2210 kcryptd_crypt_read_done(io); 2211 return; 2212 } 2213 2214 if (kcryptd_crypt_write_inline(cc, ctx)) { 2215 complete(&ctx->restart); 2216 return; 2217 } 2218 2219 kcryptd_crypt_write_io_submit(io, 1); 2220 } 2221 2222 static void kcryptd_crypt(struct work_struct *work) 2223 { 2224 struct dm_crypt_io *io = container_of(work, struct dm_crypt_io, work); 2225 2226 if (bio_data_dir(io->base_bio) == READ) 2227 kcryptd_crypt_read_convert(io); 2228 else 2229 kcryptd_crypt_write_convert(io); 2230 } 2231 2232 static void kcryptd_queue_crypt(struct dm_crypt_io *io) 2233 { 2234 struct crypt_config *cc = io->cc; 2235 2236 if ((bio_data_dir(io->base_bio) == READ && test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) || 2237 (bio_data_dir(io->base_bio) == WRITE && test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags))) { 2238 /* 2239 * in_hardirq(): Crypto API's skcipher_walk_first() refuses to work in hard IRQ context. 2240 * irqs_disabled(): the kernel may run some IO completion from the idle thread, but 2241 * it is being executed with irqs disabled. 2242 */ 2243 if (in_hardirq() || irqs_disabled()) { 2244 INIT_WORK(&io->work, kcryptd_crypt); 2245 queue_work(system_bh_wq, &io->work); 2246 return; 2247 } else { 2248 kcryptd_crypt(&io->work); 2249 return; 2250 } 2251 } 2252 2253 INIT_WORK(&io->work, kcryptd_crypt); 2254 queue_work(cc->crypt_queue, &io->work); 2255 } 2256 2257 static void crypt_free_tfms_aead(struct crypt_config *cc) 2258 { 2259 if (!cc->cipher_tfm.tfms_aead) 2260 return; 2261 2262 if (cc->cipher_tfm.tfms_aead[0] && !IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2263 crypto_free_aead(cc->cipher_tfm.tfms_aead[0]); 2264 cc->cipher_tfm.tfms_aead[0] = NULL; 2265 } 2266 2267 kfree(cc->cipher_tfm.tfms_aead); 2268 cc->cipher_tfm.tfms_aead = NULL; 2269 } 2270 2271 static void crypt_free_tfms_skcipher(struct crypt_config *cc) 2272 { 2273 unsigned int i; 2274 2275 if (!cc->cipher_tfm.tfms) 2276 return; 2277 2278 for (i = 0; i < cc->tfms_count; i++) 2279 if (cc->cipher_tfm.tfms[i] && !IS_ERR(cc->cipher_tfm.tfms[i])) { 2280 crypto_free_skcipher(cc->cipher_tfm.tfms[i]); 2281 cc->cipher_tfm.tfms[i] = NULL; 2282 } 2283 2284 kfree(cc->cipher_tfm.tfms); 2285 cc->cipher_tfm.tfms = NULL; 2286 } 2287 2288 static void crypt_free_tfms(struct crypt_config *cc) 2289 { 2290 if (crypt_integrity_aead(cc)) 2291 crypt_free_tfms_aead(cc); 2292 else 2293 crypt_free_tfms_skcipher(cc); 2294 } 2295 2296 static int crypt_alloc_tfms_skcipher(struct crypt_config *cc, char *ciphermode) 2297 { 2298 unsigned int i; 2299 int err; 2300 2301 cc->cipher_tfm.tfms = kzalloc_objs(struct crypto_skcipher *, 2302 cc->tfms_count); 2303 if (!cc->cipher_tfm.tfms) 2304 return -ENOMEM; 2305 2306 for (i = 0; i < cc->tfms_count; i++) { 2307 cc->cipher_tfm.tfms[i] = crypto_alloc_skcipher(ciphermode, 0, 2308 CRYPTO_ALG_ALLOCATES_MEMORY); 2309 if (IS_ERR(cc->cipher_tfm.tfms[i])) { 2310 err = PTR_ERR(cc->cipher_tfm.tfms[i]); 2311 crypt_free_tfms(cc); 2312 return err; 2313 } 2314 } 2315 2316 /* 2317 * dm-crypt performance can vary greatly depending on which crypto 2318 * algorithm implementation is used. Help people debug performance 2319 * problems by logging the ->cra_driver_name. 2320 */ 2321 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2322 crypto_skcipher_alg(any_tfm(cc))->base.cra_driver_name); 2323 return 0; 2324 } 2325 2326 static int crypt_alloc_tfms_aead(struct crypt_config *cc, char *ciphermode) 2327 { 2328 int err; 2329 2330 cc->cipher_tfm.tfms = kmalloc_obj(struct crypto_skcipher *); 2331 if (!cc->cipher_tfm.tfms) 2332 return -ENOMEM; 2333 2334 cc->cipher_tfm.tfms_aead[0] = crypto_alloc_aead(ciphermode, 0, 2335 CRYPTO_ALG_ALLOCATES_MEMORY); 2336 if (IS_ERR(cc->cipher_tfm.tfms_aead[0])) { 2337 err = PTR_ERR(cc->cipher_tfm.tfms_aead[0]); 2338 crypt_free_tfms(cc); 2339 return err; 2340 } 2341 2342 DMDEBUG_LIMIT("%s using implementation \"%s\"", ciphermode, 2343 crypto_aead_alg(any_tfm_aead(cc))->base.cra_driver_name); 2344 return 0; 2345 } 2346 2347 static int crypt_alloc_tfms(struct crypt_config *cc, char *ciphermode) 2348 { 2349 if (crypt_integrity_aead(cc)) 2350 return crypt_alloc_tfms_aead(cc, ciphermode); 2351 else 2352 return crypt_alloc_tfms_skcipher(cc, ciphermode); 2353 } 2354 2355 static unsigned int crypt_subkey_size(struct crypt_config *cc) 2356 { 2357 return (cc->key_size - cc->key_extra_size) >> ilog2(cc->tfms_count); 2358 } 2359 2360 static unsigned int crypt_authenckey_size(struct crypt_config *cc) 2361 { 2362 return crypt_subkey_size(cc) + RTA_SPACE(sizeof(struct crypto_authenc_key_param)); 2363 } 2364 2365 /* 2366 * If AEAD is composed like authenc(hmac(sha256),xts(aes)), 2367 * the key must be for some reason in special format. 2368 * This funcion converts cc->key to this special format. 2369 */ 2370 static void crypt_copy_authenckey(char *p, const void *key, 2371 unsigned int enckeylen, unsigned int authkeylen) 2372 { 2373 struct crypto_authenc_key_param *param; 2374 struct rtattr *rta; 2375 2376 rta = (struct rtattr *)p; 2377 param = RTA_DATA(rta); 2378 param->enckeylen = cpu_to_be32(enckeylen); 2379 rta->rta_len = RTA_LENGTH(sizeof(*param)); 2380 rta->rta_type = CRYPTO_AUTHENC_KEYA_PARAM; 2381 p += RTA_SPACE(sizeof(*param)); 2382 memcpy(p, key + enckeylen, authkeylen); 2383 p += authkeylen; 2384 memcpy(p, key, enckeylen); 2385 } 2386 2387 static int crypt_setkey(struct crypt_config *cc) 2388 { 2389 unsigned int subkey_size; 2390 int err = 0, i, r; 2391 2392 /* Ignore extra keys (which are used for IV etc) */ 2393 subkey_size = crypt_subkey_size(cc); 2394 2395 if (crypt_integrity_hmac(cc)) { 2396 if (subkey_size < cc->key_mac_size) 2397 return -EINVAL; 2398 2399 crypt_copy_authenckey(cc->authenc_key, cc->key, 2400 subkey_size - cc->key_mac_size, 2401 cc->key_mac_size); 2402 } 2403 2404 for (i = 0; i < cc->tfms_count; i++) { 2405 if (crypt_integrity_hmac(cc)) 2406 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2407 cc->authenc_key, crypt_authenckey_size(cc)); 2408 else if (crypt_integrity_aead(cc)) 2409 r = crypto_aead_setkey(cc->cipher_tfm.tfms_aead[i], 2410 cc->key + (i * subkey_size), 2411 subkey_size); 2412 else 2413 r = crypto_skcipher_setkey(cc->cipher_tfm.tfms[i], 2414 cc->key + (i * subkey_size), 2415 subkey_size); 2416 if (r) 2417 err = r; 2418 } 2419 2420 if (crypt_integrity_hmac(cc)) 2421 memzero_explicit(cc->authenc_key, crypt_authenckey_size(cc)); 2422 2423 return err; 2424 } 2425 2426 #ifdef CONFIG_KEYS 2427 2428 static bool contains_whitespace(const char *str) 2429 { 2430 while (*str) 2431 if (isspace(*str++)) 2432 return true; 2433 return false; 2434 } 2435 2436 static int set_key_user(struct crypt_config *cc, struct key *key) 2437 { 2438 const struct user_key_payload *ukp; 2439 2440 ukp = user_key_payload_locked(key); 2441 if (!ukp) 2442 return -EKEYREVOKED; 2443 2444 if (cc->key_size != ukp->datalen) 2445 return -EINVAL; 2446 2447 memcpy(cc->key, ukp->data, cc->key_size); 2448 2449 return 0; 2450 } 2451 2452 static int set_key_encrypted(struct crypt_config *cc, struct key *key) 2453 { 2454 const struct encrypted_key_payload *ekp; 2455 2456 ekp = key->payload.data[0]; 2457 if (!ekp) 2458 return -EKEYREVOKED; 2459 2460 if (cc->key_size != ekp->decrypted_datalen) 2461 return -EINVAL; 2462 2463 memcpy(cc->key, ekp->decrypted_data, cc->key_size); 2464 2465 return 0; 2466 } 2467 2468 static int set_key_trusted(struct crypt_config *cc, struct key *key) 2469 { 2470 const struct trusted_key_payload *tkp; 2471 2472 tkp = key->payload.data[0]; 2473 if (!tkp) 2474 return -EKEYREVOKED; 2475 2476 if (cc->key_size != tkp->key_len) 2477 return -EINVAL; 2478 2479 memcpy(cc->key, tkp->key, cc->key_size); 2480 2481 return 0; 2482 } 2483 2484 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2485 { 2486 char *new_key_string, *key_desc; 2487 int ret; 2488 struct key_type *type; 2489 struct key *key; 2490 int (*set_key)(struct crypt_config *cc, struct key *key); 2491 2492 /* 2493 * Reject key_string with whitespace. dm core currently lacks code for 2494 * proper whitespace escaping in arguments on DM_TABLE_STATUS path. 2495 */ 2496 if (contains_whitespace(key_string)) { 2497 DMERR("whitespace chars not allowed in key string"); 2498 return -EINVAL; 2499 } 2500 2501 /* look for next ':' separating key_type from key_description */ 2502 key_desc = strchr(key_string, ':'); 2503 if (!key_desc || key_desc == key_string || !strlen(key_desc + 1)) 2504 return -EINVAL; 2505 2506 if (!strncmp(key_string, "logon:", key_desc - key_string + 1)) { 2507 type = &key_type_logon; 2508 set_key = set_key_user; 2509 } else if (!strncmp(key_string, "user:", key_desc - key_string + 1)) { 2510 type = &key_type_user; 2511 set_key = set_key_user; 2512 } else if (IS_ENABLED(CONFIG_ENCRYPTED_KEYS) && 2513 !strncmp(key_string, "encrypted:", key_desc - key_string + 1)) { 2514 type = &key_type_encrypted; 2515 set_key = set_key_encrypted; 2516 } else if (IS_ENABLED(CONFIG_TRUSTED_KEYS) && 2517 !strncmp(key_string, "trusted:", key_desc - key_string + 1)) { 2518 type = &key_type_trusted; 2519 set_key = set_key_trusted; 2520 } else { 2521 return -EINVAL; 2522 } 2523 2524 new_key_string = kstrdup(key_string, GFP_KERNEL); 2525 if (!new_key_string) 2526 return -ENOMEM; 2527 2528 key = request_key(type, key_desc + 1, NULL); 2529 if (IS_ERR(key)) { 2530 ret = PTR_ERR(key); 2531 goto free_new_key_string; 2532 } 2533 2534 down_read(&key->sem); 2535 ret = set_key(cc, key); 2536 up_read(&key->sem); 2537 key_put(key); 2538 if (ret < 0) 2539 goto free_new_key_string; 2540 2541 /* clear the flag since following operations may invalidate previously valid key */ 2542 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2543 2544 ret = crypt_setkey(cc); 2545 if (ret) 2546 goto free_new_key_string; 2547 2548 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2549 kfree_sensitive(cc->key_string); 2550 cc->key_string = new_key_string; 2551 return 0; 2552 2553 free_new_key_string: 2554 kfree_sensitive(new_key_string); 2555 return ret; 2556 } 2557 2558 static int get_key_size(char **key_string) 2559 { 2560 char *colon, dummy; 2561 int ret; 2562 2563 if (*key_string[0] != ':') 2564 return strlen(*key_string) >> 1; 2565 2566 /* look for next ':' in key string */ 2567 colon = strpbrk(*key_string + 1, ":"); 2568 if (!colon) 2569 return -EINVAL; 2570 2571 if (sscanf(*key_string + 1, "%u%c", &ret, &dummy) != 2 || dummy != ':') 2572 return -EINVAL; 2573 2574 *key_string = colon; 2575 2576 /* remaining key string should be :<logon|user>:<key_desc> */ 2577 2578 return ret; 2579 } 2580 2581 #else 2582 2583 static int crypt_set_keyring_key(struct crypt_config *cc, const char *key_string) 2584 { 2585 return -EINVAL; 2586 } 2587 2588 static int get_key_size(char **key_string) 2589 { 2590 return (*key_string[0] == ':') ? -EINVAL : (int)(strlen(*key_string) >> 1); 2591 } 2592 2593 #endif /* CONFIG_KEYS */ 2594 2595 static int crypt_set_key(struct crypt_config *cc, char *key) 2596 { 2597 int r = -EINVAL; 2598 int key_string_len = strlen(key); 2599 2600 /* Hyphen (which gives a key_size of zero) means there is no key. */ 2601 if (!cc->key_size && strcmp(key, "-")) 2602 goto out; 2603 2604 /* ':' means the key is in kernel keyring, short-circuit normal key processing */ 2605 if (key[0] == ':') { 2606 r = crypt_set_keyring_key(cc, key + 1); 2607 goto out; 2608 } 2609 2610 /* clear the flag since following operations may invalidate previously valid key */ 2611 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2612 2613 /* wipe references to any kernel keyring key */ 2614 kfree_sensitive(cc->key_string); 2615 cc->key_string = NULL; 2616 2617 /* Decode key from its hex representation. */ 2618 if (cc->key_size && hex2bin(cc->key, key, cc->key_size) < 0) 2619 goto out; 2620 2621 r = crypt_setkey(cc); 2622 if (!r) 2623 set_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2624 2625 out: 2626 /* Hex key string not needed after here, so wipe it. */ 2627 memset(key, '0', key_string_len); 2628 2629 return r; 2630 } 2631 2632 static int crypt_wipe_key(struct crypt_config *cc) 2633 { 2634 int r; 2635 2636 clear_bit(DM_CRYPT_KEY_VALID, &cc->flags); 2637 get_random_bytes(&cc->key, cc->key_size); 2638 2639 /* Wipe IV private keys */ 2640 if (cc->iv_gen_ops && cc->iv_gen_ops->wipe) 2641 cc->iv_gen_ops->wipe(cc); 2642 2643 kfree_sensitive(cc->key_string); 2644 cc->key_string = NULL; 2645 r = crypt_setkey(cc); 2646 memset(&cc->key, 0, cc->key_size * sizeof(u8)); 2647 2648 return r; 2649 } 2650 2651 static void crypt_calculate_pages_per_client(void) 2652 { 2653 unsigned long pages = (totalram_pages() - totalhigh_pages()) * DM_CRYPT_MEMORY_PERCENT / 100; 2654 2655 if (!dm_crypt_clients_n) 2656 return; 2657 2658 pages /= dm_crypt_clients_n; 2659 if (pages < DM_CRYPT_MIN_PAGES_PER_CLIENT) 2660 pages = DM_CRYPT_MIN_PAGES_PER_CLIENT; 2661 dm_crypt_pages_per_client = pages; 2662 } 2663 2664 static void *crypt_page_alloc(gfp_t gfp_mask, void *pool_data) 2665 { 2666 struct crypt_config *cc = pool_data; 2667 struct page *page; 2668 2669 /* 2670 * Note, percpu_counter_read_positive() may over (and under) estimate 2671 * the current usage by at most (batch - 1) * num_online_cpus() pages, 2672 * but avoids potential spinlock contention of an exact result. 2673 */ 2674 if (unlikely(percpu_counter_read_positive(&cc->n_allocated_pages) >= dm_crypt_pages_per_client) && 2675 likely(gfp_mask & __GFP_NORETRY)) 2676 return NULL; 2677 2678 page = alloc_page(gfp_mask); 2679 if (likely(page != NULL)) 2680 percpu_counter_add(&cc->n_allocated_pages, 1); 2681 2682 return page; 2683 } 2684 2685 static void crypt_page_free(void *page, void *pool_data) 2686 { 2687 struct crypt_config *cc = pool_data; 2688 2689 __free_page(page); 2690 percpu_counter_sub(&cc->n_allocated_pages, 1); 2691 } 2692 2693 static void crypt_dtr(struct dm_target *ti) 2694 { 2695 struct crypt_config *cc = ti->private; 2696 2697 ti->private = NULL; 2698 2699 if (!cc) 2700 return; 2701 2702 if (cc->write_thread) 2703 kthread_stop(cc->write_thread); 2704 2705 if (cc->io_queue) 2706 destroy_workqueue(cc->io_queue); 2707 if (cc->crypt_queue) 2708 destroy_workqueue(cc->crypt_queue); 2709 2710 if (cc->workqueue_id) 2711 ida_free(&workqueue_ida, cc->workqueue_id); 2712 2713 crypt_free_tfms(cc); 2714 2715 bioset_exit(&cc->bs); 2716 2717 mempool_exit(&cc->page_pool); 2718 mempool_exit(&cc->req_pool); 2719 mempool_exit(&cc->tag_pool); 2720 2721 WARN_ON(percpu_counter_sum(&cc->n_allocated_pages) != 0); 2722 percpu_counter_destroy(&cc->n_allocated_pages); 2723 2724 if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) 2725 cc->iv_gen_ops->dtr(cc); 2726 2727 if (cc->dev) 2728 dm_put_device(ti, cc->dev); 2729 2730 kfree_sensitive(cc->cipher_string); 2731 kfree_sensitive(cc->key_string); 2732 kfree_sensitive(cc->cipher_auth); 2733 kfree_sensitive(cc->authenc_key); 2734 2735 mutex_destroy(&cc->bio_alloc_lock); 2736 2737 /* Must zero key material before freeing */ 2738 kfree_sensitive(cc); 2739 2740 spin_lock(&dm_crypt_clients_lock); 2741 WARN_ON(!dm_crypt_clients_n); 2742 dm_crypt_clients_n--; 2743 crypt_calculate_pages_per_client(); 2744 spin_unlock(&dm_crypt_clients_lock); 2745 2746 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1); 2747 } 2748 2749 static int crypt_ctr_ivmode(struct dm_target *ti, const char *ivmode) 2750 { 2751 struct crypt_config *cc = ti->private; 2752 2753 if (crypt_integrity_aead(cc)) 2754 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2755 else 2756 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2757 2758 if (cc->iv_size) 2759 /* at least a 64 bit sector number should fit in our buffer */ 2760 cc->iv_size = max(cc->iv_size, 2761 (unsigned int)(sizeof(u64) / sizeof(u8))); 2762 else if (ivmode) { 2763 DMWARN("Selected cipher does not support IVs"); 2764 ivmode = NULL; 2765 } 2766 2767 /* Choose ivmode, see comments at iv code. */ 2768 if (ivmode == NULL) 2769 cc->iv_gen_ops = NULL; 2770 else if (strcmp(ivmode, "plain") == 0) 2771 cc->iv_gen_ops = &crypt_iv_plain_ops; 2772 else if (strcmp(ivmode, "plain64") == 0) 2773 cc->iv_gen_ops = &crypt_iv_plain64_ops; 2774 else if (strcmp(ivmode, "plain64be") == 0) 2775 cc->iv_gen_ops = &crypt_iv_plain64be_ops; 2776 else if (strcmp(ivmode, "essiv") == 0) 2777 cc->iv_gen_ops = &crypt_iv_essiv_ops; 2778 else if (strcmp(ivmode, "benbi") == 0) 2779 cc->iv_gen_ops = &crypt_iv_benbi_ops; 2780 else if (strcmp(ivmode, "null") == 0) 2781 cc->iv_gen_ops = &crypt_iv_null_ops; 2782 else if (strcmp(ivmode, "eboiv") == 0) 2783 cc->iv_gen_ops = &crypt_iv_eboiv_ops; 2784 else if (strcmp(ivmode, "elephant") == 0) { 2785 cc->iv_gen_ops = &crypt_iv_elephant_ops; 2786 cc->key_parts = 2; 2787 cc->key_extra_size = cc->key_size / 2; 2788 if (cc->key_extra_size > ELEPHANT_MAX_KEY_SIZE) 2789 return -EINVAL; 2790 set_bit(CRYPT_ENCRYPT_PREPROCESS, &cc->cipher_flags); 2791 } else if (strcmp(ivmode, "lmk") == 0) { 2792 cc->iv_gen_ops = &crypt_iv_lmk_ops; 2793 /* 2794 * Version 2 and 3 is recognised according 2795 * to length of provided multi-key string. 2796 * If present (version 3), last key is used as IV seed. 2797 * All keys (including IV seed) are always the same size. 2798 */ 2799 if (cc->key_size % cc->key_parts) { 2800 cc->key_parts++; 2801 cc->key_extra_size = cc->key_size / cc->key_parts; 2802 } 2803 } else if (strcmp(ivmode, "tcw") == 0) { 2804 cc->iv_gen_ops = &crypt_iv_tcw_ops; 2805 cc->key_parts += 2; /* IV + whitening */ 2806 cc->key_extra_size = cc->iv_size + TCW_WHITENING_SIZE; 2807 } else if (strcmp(ivmode, "random") == 0) { 2808 cc->iv_gen_ops = &crypt_iv_random_ops; 2809 /* Need storage space in integrity fields. */ 2810 cc->integrity_iv_size = cc->iv_size; 2811 } else { 2812 ti->error = "Invalid IV mode"; 2813 return -EINVAL; 2814 } 2815 2816 return 0; 2817 } 2818 2819 /* 2820 * Workaround to parse HMAC algorithm from AEAD crypto API spec. 2821 * The HMAC is needed to calculate tag size (HMAC digest size). 2822 * This should be probably done by crypto-api calls (once available...) 2823 */ 2824 static int crypt_ctr_auth_cipher(struct crypt_config *cc, char *cipher_api) 2825 { 2826 char *start, *end, *mac_alg = NULL; 2827 struct crypto_ahash *mac; 2828 2829 if (!strstarts(cipher_api, "authenc(")) 2830 return 0; 2831 2832 start = strchr(cipher_api, '('); 2833 end = strchr(cipher_api, ','); 2834 if (!start || !end || ++start > end) 2835 return -EINVAL; 2836 2837 mac_alg = kmemdup_nul(start, end - start, GFP_KERNEL); 2838 if (!mac_alg) 2839 return -ENOMEM; 2840 2841 mac = crypto_alloc_ahash(mac_alg, 0, CRYPTO_ALG_ALLOCATES_MEMORY); 2842 kfree(mac_alg); 2843 2844 if (IS_ERR(mac)) 2845 return PTR_ERR(mac); 2846 2847 if (!test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags)) 2848 cc->key_mac_size = crypto_ahash_digestsize(mac); 2849 crypto_free_ahash(mac); 2850 2851 cc->authenc_key = kmalloc(crypt_authenckey_size(cc), GFP_KERNEL); 2852 if (!cc->authenc_key) 2853 return -ENOMEM; 2854 2855 return 0; 2856 } 2857 2858 static int crypt_ctr_cipher_new(struct dm_target *ti, char *cipher_in, char *key, 2859 char **ivmode, char **ivopts) 2860 { 2861 struct crypt_config *cc = ti->private; 2862 char *tmp, *cipher_api, buf[CRYPTO_MAX_ALG_NAME]; 2863 int ret = -EINVAL; 2864 2865 cc->tfms_count = 1; 2866 2867 /* 2868 * New format (capi: prefix) 2869 * capi:cipher_api_spec-iv:ivopts 2870 */ 2871 tmp = &cipher_in[strlen("capi:")]; 2872 2873 /* Separate IV options if present, it can contain another '-' in hash name */ 2874 *ivopts = strrchr(tmp, ':'); 2875 if (*ivopts) { 2876 **ivopts = '\0'; 2877 (*ivopts)++; 2878 } 2879 /* Parse IV mode */ 2880 *ivmode = strrchr(tmp, '-'); 2881 if (*ivmode) { 2882 **ivmode = '\0'; 2883 (*ivmode)++; 2884 } 2885 /* The rest is crypto API spec */ 2886 cipher_api = tmp; 2887 2888 /* Alloc AEAD, can be used only in new format. */ 2889 if (crypt_integrity_aead(cc)) { 2890 ret = crypt_ctr_auth_cipher(cc, cipher_api); 2891 if (ret < 0) { 2892 ti->error = "Invalid AEAD cipher spec"; 2893 return ret; 2894 } 2895 } 2896 2897 if (*ivmode && !strcmp(*ivmode, "lmk")) 2898 cc->tfms_count = 64; 2899 2900 if (*ivmode && !strcmp(*ivmode, "essiv")) { 2901 if (!*ivopts) { 2902 ti->error = "Digest algorithm missing for ESSIV mode"; 2903 return -EINVAL; 2904 } 2905 ret = snprintf(buf, CRYPTO_MAX_ALG_NAME, "essiv(%s,%s)", 2906 cipher_api, *ivopts); 2907 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 2908 ti->error = "Cannot allocate cipher string"; 2909 return -ENOMEM; 2910 } 2911 cipher_api = buf; 2912 } 2913 2914 cc->key_parts = cc->tfms_count; 2915 2916 /* Allocate cipher */ 2917 ret = crypt_alloc_tfms(cc, cipher_api); 2918 if (ret < 0) { 2919 ti->error = "Error allocating crypto tfm"; 2920 return ret; 2921 } 2922 2923 if (crypt_integrity_aead(cc)) 2924 cc->iv_size = crypto_aead_ivsize(any_tfm_aead(cc)); 2925 else 2926 cc->iv_size = crypto_skcipher_ivsize(any_tfm(cc)); 2927 2928 return 0; 2929 } 2930 2931 static int crypt_ctr_cipher_old(struct dm_target *ti, char *cipher_in, char *key, 2932 char **ivmode, char **ivopts) 2933 { 2934 struct crypt_config *cc = ti->private; 2935 char *tmp, *cipher, *chainmode, *keycount; 2936 char *cipher_api = NULL; 2937 int ret = -EINVAL; 2938 char dummy; 2939 2940 if (strchr(cipher_in, '(') || crypt_integrity_aead(cc)) { 2941 ti->error = "Bad cipher specification"; 2942 return -EINVAL; 2943 } 2944 2945 /* 2946 * Legacy dm-crypt cipher specification 2947 * cipher[:keycount]-mode-iv:ivopts 2948 */ 2949 tmp = cipher_in; 2950 keycount = strsep(&tmp, "-"); 2951 cipher = strsep(&keycount, ":"); 2952 2953 if (!keycount) 2954 cc->tfms_count = 1; 2955 else if (sscanf(keycount, "%u%c", &cc->tfms_count, &dummy) != 1 || 2956 !is_power_of_2(cc->tfms_count)) { 2957 ti->error = "Bad cipher key count specification"; 2958 return -EINVAL; 2959 } 2960 cc->key_parts = cc->tfms_count; 2961 2962 chainmode = strsep(&tmp, "-"); 2963 *ivmode = strsep(&tmp, ":"); 2964 *ivopts = tmp; 2965 2966 /* 2967 * For compatibility with the original dm-crypt mapping format, if 2968 * only the cipher name is supplied, use cbc-plain. 2969 */ 2970 if (!chainmode || (!strcmp(chainmode, "plain") && !*ivmode)) { 2971 chainmode = "cbc"; 2972 *ivmode = "plain"; 2973 } 2974 2975 if (strcmp(chainmode, "ecb") && !*ivmode) { 2976 ti->error = "IV mechanism required"; 2977 return -EINVAL; 2978 } 2979 2980 cipher_api = kmalloc(CRYPTO_MAX_ALG_NAME, GFP_KERNEL); 2981 if (!cipher_api) 2982 goto bad_mem; 2983 2984 if (*ivmode && !strcmp(*ivmode, "essiv")) { 2985 if (!*ivopts) { 2986 ti->error = "Digest algorithm missing for ESSIV mode"; 2987 kfree(cipher_api); 2988 return -EINVAL; 2989 } 2990 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 2991 "essiv(%s(%s),%s)", chainmode, cipher, *ivopts); 2992 } else { 2993 ret = snprintf(cipher_api, CRYPTO_MAX_ALG_NAME, 2994 "%s(%s)", chainmode, cipher); 2995 } 2996 if (ret < 0 || ret >= CRYPTO_MAX_ALG_NAME) { 2997 kfree(cipher_api); 2998 goto bad_mem; 2999 } 3000 3001 /* Allocate cipher */ 3002 ret = crypt_alloc_tfms(cc, cipher_api); 3003 if (ret < 0) { 3004 ti->error = "Error allocating crypto tfm"; 3005 kfree(cipher_api); 3006 return ret; 3007 } 3008 kfree(cipher_api); 3009 3010 return 0; 3011 bad_mem: 3012 ti->error = "Cannot allocate cipher strings"; 3013 return -ENOMEM; 3014 } 3015 3016 static int crypt_ctr_cipher(struct dm_target *ti, char *cipher_in, char *key) 3017 { 3018 struct crypt_config *cc = ti->private; 3019 char *ivmode = NULL, *ivopts = NULL; 3020 int ret; 3021 3022 cc->cipher_string = kstrdup(cipher_in, GFP_KERNEL); 3023 if (!cc->cipher_string) { 3024 ti->error = "Cannot allocate cipher strings"; 3025 return -ENOMEM; 3026 } 3027 3028 if (strstarts(cipher_in, "capi:")) 3029 ret = crypt_ctr_cipher_new(ti, cipher_in, key, &ivmode, &ivopts); 3030 else 3031 ret = crypt_ctr_cipher_old(ti, cipher_in, key, &ivmode, &ivopts); 3032 if (ret) 3033 return ret; 3034 3035 /* Initialize IV */ 3036 ret = crypt_ctr_ivmode(ti, ivmode); 3037 if (ret < 0) 3038 return ret; 3039 3040 /* Initialize and set key */ 3041 ret = crypt_set_key(cc, key); 3042 if (ret < 0) { 3043 ti->error = "Error decoding and setting key"; 3044 return ret; 3045 } 3046 3047 /* Allocate IV */ 3048 if (cc->iv_gen_ops && cc->iv_gen_ops->ctr) { 3049 ret = cc->iv_gen_ops->ctr(cc, ti, ivopts); 3050 if (ret < 0) { 3051 ti->error = "Error creating IV"; 3052 return ret; 3053 } 3054 } 3055 3056 /* Initialize IV (set keys for ESSIV etc) */ 3057 if (cc->iv_gen_ops && cc->iv_gen_ops->init) { 3058 ret = cc->iv_gen_ops->init(cc); 3059 if (ret < 0) { 3060 ti->error = "Error initialising IV"; 3061 return ret; 3062 } 3063 } 3064 3065 /* wipe the kernel key payload copy */ 3066 if (cc->key_string) 3067 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3068 3069 return ret; 3070 } 3071 3072 static int crypt_ctr_optional(struct dm_target *ti, unsigned int argc, char **argv) 3073 { 3074 struct crypt_config *cc = ti->private; 3075 struct dm_arg_set as; 3076 static const struct dm_arg _args[] = { 3077 {0, 9, "Invalid number of feature args"}, 3078 }; 3079 unsigned int opt_params, val; 3080 const char *opt_string, *sval; 3081 char dummy; 3082 int ret; 3083 3084 /* Optional parameters */ 3085 as.argc = argc; 3086 as.argv = argv; 3087 3088 ret = dm_read_arg_group(_args, &as, &opt_params, &ti->error); 3089 if (ret) 3090 return ret; 3091 3092 while (opt_params--) { 3093 opt_string = dm_shift_arg(&as); 3094 if (!opt_string) { 3095 ti->error = "Not enough feature arguments"; 3096 return -EINVAL; 3097 } 3098 3099 if (!strcasecmp(opt_string, "allow_discards")) 3100 ti->num_discard_bios = 1; 3101 3102 else if (!strcasecmp(opt_string, "same_cpu_crypt")) 3103 set_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3104 else if (!strcasecmp(opt_string, "high_priority")) 3105 set_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags); 3106 3107 else if (!strcasecmp(opt_string, "submit_from_crypt_cpus")) 3108 set_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3109 else if (!strcasecmp(opt_string, "no_read_workqueue")) 3110 set_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3111 else if (!strcasecmp(opt_string, "no_write_workqueue")) 3112 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3113 else if (sscanf(opt_string, "integrity:%u:", &val) == 1) { 3114 if (val == 0 || val > MAX_TAG_SIZE) { 3115 ti->error = "Invalid integrity arguments"; 3116 return -EINVAL; 3117 } 3118 cc->used_tag_size = val; 3119 sval = strchr(opt_string + strlen("integrity:"), ':') + 1; 3120 if (!strcasecmp(sval, "aead")) { 3121 set_bit(CRYPT_MODE_INTEGRITY_AEAD, &cc->cipher_flags); 3122 } else if (strcasecmp(sval, "none")) { 3123 ti->error = "Unknown integrity profile"; 3124 return -EINVAL; 3125 } 3126 3127 cc->cipher_auth = kstrdup(sval, GFP_KERNEL); 3128 if (!cc->cipher_auth) 3129 return -ENOMEM; 3130 } else if (sscanf(opt_string, "integrity_key_size:%u%c", &val, &dummy) == 1) { 3131 if (!val) { 3132 ti->error = "Invalid integrity_key_size argument"; 3133 return -EINVAL; 3134 } 3135 cc->key_mac_size = val; 3136 set_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags); 3137 } else if (sscanf(opt_string, "sector_size:%hu%c", &cc->sector_size, &dummy) == 1) { 3138 if (cc->sector_size < (1 << SECTOR_SHIFT) || 3139 cc->sector_size > 4096 || 3140 (cc->sector_size & (cc->sector_size - 1))) { 3141 ti->error = "Invalid feature value for sector_size"; 3142 return -EINVAL; 3143 } 3144 if (ti->len & ((cc->sector_size >> SECTOR_SHIFT) - 1)) { 3145 ti->error = "Device size is not multiple of sector_size feature"; 3146 return -EINVAL; 3147 } 3148 cc->sector_shift = __ffs(cc->sector_size) - SECTOR_SHIFT; 3149 } else if (!strcasecmp(opt_string, "iv_large_sectors")) 3150 set_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3151 else { 3152 ti->error = "Invalid feature arguments"; 3153 return -EINVAL; 3154 } 3155 } 3156 3157 return 0; 3158 } 3159 3160 #ifdef CONFIG_BLK_DEV_ZONED 3161 static int crypt_report_zones(struct dm_target *ti, 3162 struct dm_report_zones_args *args, unsigned int nr_zones) 3163 { 3164 struct crypt_config *cc = ti->private; 3165 3166 return dm_report_zones(cc->dev->bdev, cc->start, 3167 cc->start + dm_target_offset(ti, args->next_sector), 3168 args, nr_zones); 3169 } 3170 #else 3171 #define crypt_report_zones NULL 3172 #endif 3173 3174 /* 3175 * Construct an encryption mapping: 3176 * <cipher> [<key>|:<key_size>:<user|logon>:<key_description>] <iv_offset> <dev_path> <start> 3177 */ 3178 static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) 3179 { 3180 struct crypt_config *cc; 3181 const char *devname = dm_table_device_name(ti->table); 3182 int key_size, wq_id; 3183 unsigned int align_mask; 3184 unsigned int common_wq_flags; 3185 unsigned long long tmpll; 3186 int ret; 3187 size_t iv_size_padding, additional_req_size; 3188 char dummy; 3189 3190 if (argc < 5) { 3191 ti->error = "Not enough arguments"; 3192 return -EINVAL; 3193 } 3194 3195 key_size = get_key_size(&argv[1]); 3196 if (key_size < 0) { 3197 ti->error = "Cannot parse key size"; 3198 return -EINVAL; 3199 } 3200 3201 cc = kzalloc_flex(*cc, key, key_size); 3202 if (!cc) { 3203 ti->error = "Cannot allocate encryption context"; 3204 return -ENOMEM; 3205 } 3206 cc->key_size = key_size; 3207 cc->sector_size = (1 << SECTOR_SHIFT); 3208 cc->sector_shift = 0; 3209 3210 ti->private = cc; 3211 3212 spin_lock(&dm_crypt_clients_lock); 3213 dm_crypt_clients_n++; 3214 crypt_calculate_pages_per_client(); 3215 spin_unlock(&dm_crypt_clients_lock); 3216 3217 ret = percpu_counter_init(&cc->n_allocated_pages, 0, GFP_KERNEL); 3218 if (ret < 0) 3219 goto bad; 3220 3221 /* Optional parameters need to be read before cipher constructor */ 3222 if (argc > 5) { 3223 ret = crypt_ctr_optional(ti, argc - 5, &argv[5]); 3224 if (ret) 3225 goto bad; 3226 } 3227 3228 ret = crypt_ctr_cipher(ti, argv[0], argv[1]); 3229 if (ret < 0) 3230 goto bad; 3231 3232 if (crypt_integrity_aead(cc)) { 3233 cc->dmreq_start = sizeof(struct aead_request); 3234 cc->dmreq_start += crypto_aead_reqsize(any_tfm_aead(cc)); 3235 align_mask = crypto_aead_alignmask(any_tfm_aead(cc)); 3236 } else { 3237 cc->dmreq_start = sizeof(struct skcipher_request); 3238 cc->dmreq_start += crypto_skcipher_reqsize(any_tfm(cc)); 3239 align_mask = crypto_skcipher_alignmask(any_tfm(cc)); 3240 } 3241 cc->dmreq_start = ALIGN(cc->dmreq_start, __alignof__(struct dm_crypt_request)); 3242 3243 if (align_mask < CRYPTO_MINALIGN) { 3244 /* Allocate the padding exactly */ 3245 iv_size_padding = -(cc->dmreq_start + sizeof(struct dm_crypt_request)) 3246 & align_mask; 3247 } else { 3248 /* 3249 * If the cipher requires greater alignment than kmalloc 3250 * alignment, we don't know the exact position of the 3251 * initialization vector. We must assume worst case. 3252 */ 3253 iv_size_padding = align_mask; 3254 } 3255 3256 /* ...| IV + padding | original IV | original sec. number | bio tag offset | */ 3257 additional_req_size = sizeof(struct dm_crypt_request) + 3258 iv_size_padding + cc->iv_size + 3259 cc->iv_size + 3260 sizeof(uint64_t) + 3261 sizeof(unsigned int); 3262 3263 ret = mempool_init_kmalloc_pool(&cc->req_pool, MIN_IOS, cc->dmreq_start + additional_req_size); 3264 if (ret) { 3265 ti->error = "Cannot allocate crypt request mempool"; 3266 goto bad; 3267 } 3268 3269 cc->per_bio_data_size = ti->per_io_data_size = 3270 ALIGN(sizeof(struct dm_crypt_io) + cc->dmreq_start + additional_req_size, 3271 ARCH_DMA_MINALIGN); 3272 3273 ret = mempool_init(&cc->page_pool, BIO_MAX_VECS, crypt_page_alloc, crypt_page_free, cc); 3274 if (ret) { 3275 ti->error = "Cannot allocate page mempool"; 3276 goto bad; 3277 } 3278 3279 ret = bioset_init(&cc->bs, MIN_IOS, 0, BIOSET_NEED_BVECS); 3280 if (ret) { 3281 ti->error = "Cannot allocate crypt bioset"; 3282 goto bad; 3283 } 3284 3285 mutex_init(&cc->bio_alloc_lock); 3286 3287 ret = -EINVAL; 3288 if ((sscanf(argv[2], "%llu%c", &tmpll, &dummy) != 1) || 3289 (tmpll & ((cc->sector_size >> SECTOR_SHIFT) - 1))) { 3290 ti->error = "Invalid iv_offset sector"; 3291 goto bad; 3292 } 3293 cc->iv_offset = tmpll; 3294 3295 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table), &cc->dev); 3296 if (ret) { 3297 ti->error = "Device lookup failed"; 3298 goto bad; 3299 } 3300 3301 ret = -EINVAL; 3302 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) { 3303 ti->error = "Invalid device sector"; 3304 goto bad; 3305 } 3306 cc->start = tmpll; 3307 3308 if (bdev_is_zoned(cc->dev->bdev)) { 3309 /* 3310 * For zoned block devices, we need to preserve the issuer write 3311 * ordering. To do so, disable write workqueues and force inline 3312 * encryption completion. 3313 */ 3314 set_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3315 set_bit(DM_CRYPT_WRITE_INLINE, &cc->flags); 3316 3317 /* 3318 * All zone append writes to a zone of a zoned block device will 3319 * have the same BIO sector, the start of the zone. When the 3320 * cypher IV mode uses sector values, all data targeting a 3321 * zone will be encrypted using the first sector numbers of the 3322 * zone. This will not result in write errors but will 3323 * cause most reads to fail as reads will use the sector values 3324 * for the actual data locations, resulting in IV mismatch. 3325 * To avoid this problem, ask DM core to emulate zone append 3326 * operations with regular writes. 3327 */ 3328 DMDEBUG("Zone append operations will be emulated"); 3329 ti->emulate_zone_append = true; 3330 } 3331 3332 if (crypt_integrity_aead(cc) || cc->integrity_iv_size) { 3333 ret = crypt_integrity_ctr(cc, ti); 3334 if (ret) 3335 goto bad; 3336 3337 cc->tag_pool_max_sectors = POOL_ENTRY_SIZE / cc->tuple_size; 3338 if (!cc->tag_pool_max_sectors) 3339 cc->tag_pool_max_sectors = 1; 3340 3341 ret = mempool_init_kmalloc_pool(&cc->tag_pool, MIN_IOS, 3342 cc->tag_pool_max_sectors * cc->tuple_size); 3343 if (ret) { 3344 ti->error = "Cannot allocate integrity tags mempool"; 3345 goto bad; 3346 } 3347 3348 cc->tag_pool_max_sectors <<= cc->sector_shift; 3349 } 3350 3351 wq_id = ida_alloc_min(&workqueue_ida, 1, GFP_KERNEL); 3352 if (wq_id < 0) { 3353 ti->error = "Couldn't get workqueue id"; 3354 ret = wq_id; 3355 goto bad; 3356 } 3357 cc->workqueue_id = wq_id; 3358 3359 ret = -ENOMEM; 3360 common_wq_flags = WQ_MEM_RECLAIM | WQ_SYSFS; 3361 if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags)) 3362 common_wq_flags |= WQ_HIGHPRI; 3363 3364 cc->io_queue = alloc_workqueue("kcryptd_io-%s-%d", 3365 common_wq_flags | WQ_PERCPU, 1, 3366 devname, wq_id); 3367 if (!cc->io_queue) { 3368 ti->error = "Couldn't create kcryptd io queue"; 3369 goto bad; 3370 } 3371 3372 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) { 3373 cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d", 3374 common_wq_flags | WQ_CPU_INTENSIVE | WQ_PERCPU, 3375 1, devname, wq_id); 3376 } else { 3377 /* 3378 * While crypt_queue is certainly CPU intensive, the use of 3379 * WQ_CPU_INTENSIVE is meaningless with WQ_UNBOUND. 3380 */ 3381 cc->crypt_queue = alloc_workqueue("kcryptd-%s-%d", 3382 common_wq_flags | WQ_UNBOUND, 3383 num_online_cpus(), devname, wq_id); 3384 } 3385 if (!cc->crypt_queue) { 3386 ti->error = "Couldn't create kcryptd queue"; 3387 goto bad; 3388 } 3389 3390 spin_lock_init(&cc->write_thread_lock); 3391 cc->write_tree = RB_ROOT; 3392 3393 cc->write_thread = kthread_run(dmcrypt_write, cc, "dmcrypt_write/%s", devname); 3394 if (IS_ERR(cc->write_thread)) { 3395 ret = PTR_ERR(cc->write_thread); 3396 cc->write_thread = NULL; 3397 ti->error = "Couldn't spawn write thread"; 3398 goto bad; 3399 } 3400 if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags)) 3401 set_user_nice(cc->write_thread, MIN_NICE); 3402 3403 ti->num_flush_bios = 1; 3404 ti->limit_swap_bios = true; 3405 ti->accounts_remapped_io = true; 3406 3407 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1); 3408 return 0; 3409 3410 bad: 3411 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0); 3412 crypt_dtr(ti); 3413 return ret; 3414 } 3415 3416 static int crypt_map(struct dm_target *ti, struct bio *bio) 3417 { 3418 struct dm_crypt_io *io; 3419 struct crypt_config *cc = ti->private; 3420 unsigned max_sectors; 3421 bool no_split; 3422 3423 /* 3424 * If bio is REQ_PREFLUSH or REQ_OP_DISCARD, just bypass crypt queues. 3425 * - for REQ_PREFLUSH device-mapper core ensures that no IO is in-flight 3426 * - for REQ_OP_DISCARD caller must use flush if IO ordering matters 3427 */ 3428 if (unlikely(bio->bi_opf & REQ_PREFLUSH || 3429 bio_op(bio) == REQ_OP_DISCARD)) { 3430 bio_set_dev(bio, cc->dev->bdev); 3431 if (bio_sectors(bio)) 3432 bio->bi_iter.bi_sector = cc->start + 3433 dm_target_offset(ti, bio->bi_iter.bi_sector); 3434 return DM_MAPIO_REMAPPED; 3435 } 3436 3437 /* 3438 * Check if bio is too large, split as needed. 3439 * 3440 * For zoned devices, splitting write operations creates the 3441 * risk of deadlocking queue freeze operations with zone write 3442 * plugging BIO work when the reminder of a split BIO is 3443 * issued. So always allow the entire BIO to proceed. 3444 */ 3445 no_split = (ti->emulate_zone_append && op_is_write(bio_op(bio))) || 3446 (bio->bi_opf & REQ_ATOMIC); 3447 max_sectors = get_max_request_sectors(ti, bio, no_split); 3448 if (unlikely(bio_sectors(bio) > max_sectors)) { 3449 if (unlikely(no_split)) 3450 return DM_MAPIO_KILL; 3451 dm_accept_partial_bio(bio, max_sectors); 3452 } 3453 3454 /* 3455 * Ensure that bio is a multiple of internal sector encryption size 3456 * and is aligned to this size as defined in IO hints. 3457 */ 3458 if (unlikely((bio->bi_iter.bi_sector & ((cc->sector_size >> SECTOR_SHIFT) - 1)) != 0)) 3459 return DM_MAPIO_KILL; 3460 3461 if (unlikely(bio->bi_iter.bi_size & (cc->sector_size - 1))) 3462 return DM_MAPIO_KILL; 3463 3464 io = dm_per_bio_data(bio, cc->per_bio_data_size); 3465 crypt_io_init(io, cc, bio, dm_target_offset(ti, bio->bi_iter.bi_sector)); 3466 3467 if (cc->tuple_size) { 3468 unsigned int tag_len = cc->tuple_size * (bio_sectors(bio) >> cc->sector_shift); 3469 3470 if (unlikely(tag_len > KMALLOC_MAX_SIZE)) 3471 io->integrity_metadata = NULL; 3472 else 3473 io->integrity_metadata = kmalloc(tag_len, GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); 3474 3475 if (unlikely(!io->integrity_metadata)) { 3476 if (bio_sectors(bio) > cc->tag_pool_max_sectors) 3477 dm_accept_partial_bio(bio, cc->tag_pool_max_sectors); 3478 io->integrity_metadata = mempool_alloc(&cc->tag_pool, GFP_NOIO); 3479 io->integrity_metadata_from_pool = true; 3480 } 3481 } 3482 3483 if (crypt_integrity_aead(cc)) 3484 io->ctx.r.req_aead = (struct aead_request *)(io + 1); 3485 else 3486 io->ctx.r.req = (struct skcipher_request *)(io + 1); 3487 3488 if (bio_data_dir(io->base_bio) == READ) { 3489 if (kcryptd_io_read(io, CRYPT_MAP_READ_GFP)) 3490 kcryptd_queue_read(io); 3491 } else 3492 kcryptd_queue_crypt(io); 3493 3494 return DM_MAPIO_SUBMITTED; 3495 } 3496 3497 static char hex2asc(unsigned char c) 3498 { 3499 return c + '0' + ((unsigned int)(9 - c) >> 4 & 0x27); 3500 } 3501 3502 static void crypt_status(struct dm_target *ti, status_type_t type, 3503 unsigned int status_flags, char *result, unsigned int maxlen) 3504 { 3505 struct crypt_config *cc = ti->private; 3506 unsigned int i, sz = 0; 3507 int num_feature_args = 0; 3508 3509 switch (type) { 3510 case STATUSTYPE_INFO: 3511 result[0] = '\0'; 3512 break; 3513 3514 case STATUSTYPE_TABLE: 3515 DMEMIT("%s ", cc->cipher_string); 3516 3517 if (cc->key_size > 0) { 3518 if (cc->key_string) 3519 DMEMIT(":%u:%s", cc->key_size, cc->key_string); 3520 else { 3521 for (i = 0; i < cc->key_size; i++) { 3522 DMEMIT("%c%c", hex2asc(cc->key[i] >> 4), 3523 hex2asc(cc->key[i] & 0xf)); 3524 } 3525 } 3526 } else 3527 DMEMIT("-"); 3528 3529 DMEMIT(" %llu %s %llu", (unsigned long long)cc->iv_offset, 3530 cc->dev->name, (unsigned long long)cc->start); 3531 3532 num_feature_args += !!ti->num_discard_bios; 3533 num_feature_args += test_bit(DM_CRYPT_SAME_CPU, &cc->flags); 3534 num_feature_args += test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags); 3535 num_feature_args += test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags); 3536 num_feature_args += test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags); 3537 num_feature_args += test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags); 3538 num_feature_args += !!cc->used_tag_size; 3539 num_feature_args += cc->sector_size != (1 << SECTOR_SHIFT); 3540 num_feature_args += test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags); 3541 num_feature_args += test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags); 3542 if (num_feature_args) { 3543 DMEMIT(" %d", num_feature_args); 3544 if (ti->num_discard_bios) 3545 DMEMIT(" allow_discards"); 3546 if (test_bit(DM_CRYPT_SAME_CPU, &cc->flags)) 3547 DMEMIT(" same_cpu_crypt"); 3548 if (test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags)) 3549 DMEMIT(" high_priority"); 3550 if (test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags)) 3551 DMEMIT(" submit_from_crypt_cpus"); 3552 if (test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags)) 3553 DMEMIT(" no_read_workqueue"); 3554 if (test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags)) 3555 DMEMIT(" no_write_workqueue"); 3556 if (cc->used_tag_size) 3557 DMEMIT(" integrity:%u:%s", cc->used_tag_size, cc->cipher_auth); 3558 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3559 DMEMIT(" sector_size:%d", cc->sector_size); 3560 if (test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags)) 3561 DMEMIT(" iv_large_sectors"); 3562 if (test_bit(CRYPT_KEY_MAC_SIZE_SET, &cc->cipher_flags)) 3563 DMEMIT(" integrity_key_size:%u", cc->key_mac_size); 3564 } 3565 break; 3566 3567 case STATUSTYPE_IMA: 3568 DMEMIT_TARGET_NAME_VERSION(ti->type); 3569 DMEMIT(",allow_discards=%c", ti->num_discard_bios ? 'y' : 'n'); 3570 DMEMIT(",same_cpu_crypt=%c", test_bit(DM_CRYPT_SAME_CPU, &cc->flags) ? 'y' : 'n'); 3571 DMEMIT(",high_priority=%c", test_bit(DM_CRYPT_HIGH_PRIORITY, &cc->flags) ? 'y' : 'n'); 3572 DMEMIT(",submit_from_crypt_cpus=%c", test_bit(DM_CRYPT_NO_OFFLOAD, &cc->flags) ? 3573 'y' : 'n'); 3574 DMEMIT(",no_read_workqueue=%c", test_bit(DM_CRYPT_NO_READ_WORKQUEUE, &cc->flags) ? 3575 'y' : 'n'); 3576 DMEMIT(",no_write_workqueue=%c", test_bit(DM_CRYPT_NO_WRITE_WORKQUEUE, &cc->flags) ? 3577 'y' : 'n'); 3578 DMEMIT(",iv_large_sectors=%c", test_bit(CRYPT_IV_LARGE_SECTORS, &cc->cipher_flags) ? 3579 'y' : 'n'); 3580 3581 if (cc->used_tag_size) 3582 DMEMIT(",integrity_tag_size=%u,cipher_auth=%s", 3583 cc->used_tag_size, cc->cipher_auth); 3584 if (cc->sector_size != (1 << SECTOR_SHIFT)) 3585 DMEMIT(",sector_size=%d", cc->sector_size); 3586 if (cc->cipher_string) 3587 DMEMIT(",cipher_string=%s", cc->cipher_string); 3588 3589 DMEMIT(",key_size=%u", cc->key_size); 3590 DMEMIT(",key_parts=%u", cc->key_parts); 3591 DMEMIT(",key_extra_size=%u", cc->key_extra_size); 3592 DMEMIT(",key_mac_size=%u", cc->key_mac_size); 3593 DMEMIT(";"); 3594 break; 3595 } 3596 } 3597 3598 static void crypt_postsuspend(struct dm_target *ti) 3599 { 3600 struct crypt_config *cc = ti->private; 3601 3602 set_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3603 } 3604 3605 static int crypt_preresume(struct dm_target *ti) 3606 { 3607 struct crypt_config *cc = ti->private; 3608 3609 if (!test_bit(DM_CRYPT_KEY_VALID, &cc->flags)) { 3610 DMERR("aborting resume - crypt key is not set."); 3611 return -EAGAIN; 3612 } 3613 3614 return 0; 3615 } 3616 3617 static void crypt_resume(struct dm_target *ti) 3618 { 3619 struct crypt_config *cc = ti->private; 3620 3621 clear_bit(DM_CRYPT_SUSPENDED, &cc->flags); 3622 } 3623 3624 /* Message interface 3625 * key set <key> 3626 * key wipe 3627 */ 3628 static int crypt_message(struct dm_target *ti, unsigned int argc, char **argv, 3629 char *result, unsigned int maxlen) 3630 { 3631 struct crypt_config *cc = ti->private; 3632 int key_size, ret = -EINVAL; 3633 3634 if (argc < 2) 3635 goto error; 3636 3637 if (!strcasecmp(argv[0], "key")) { 3638 if (!test_bit(DM_CRYPT_SUSPENDED, &cc->flags)) { 3639 DMWARN("not suspended during key manipulation."); 3640 return -EINVAL; 3641 } 3642 if (argc == 3 && !strcasecmp(argv[1], "set")) { 3643 /* The key size may not be changed. */ 3644 key_size = get_key_size(&argv[2]); 3645 if (key_size < 0 || cc->key_size != key_size) { 3646 memset(argv[2], '0', strlen(argv[2])); 3647 return -EINVAL; 3648 } 3649 3650 ret = crypt_set_key(cc, argv[2]); 3651 if (ret) 3652 return ret; 3653 if (cc->iv_gen_ops && cc->iv_gen_ops->init) 3654 ret = cc->iv_gen_ops->init(cc); 3655 /* wipe the kernel key payload copy */ 3656 if (cc->key_string) 3657 memset(cc->key, 0, cc->key_size * sizeof(u8)); 3658 return ret; 3659 } 3660 if (argc == 2 && !strcasecmp(argv[1], "wipe")) 3661 return crypt_wipe_key(cc); 3662 } 3663 3664 error: 3665 DMWARN("unrecognised message received."); 3666 return -EINVAL; 3667 } 3668 3669 static int crypt_iterate_devices(struct dm_target *ti, 3670 iterate_devices_callout_fn fn, void *data) 3671 { 3672 struct crypt_config *cc = ti->private; 3673 3674 return fn(ti, cc->dev, cc->start, ti->len, data); 3675 } 3676 3677 static void crypt_io_hints(struct dm_target *ti, struct queue_limits *limits) 3678 { 3679 struct crypt_config *cc = ti->private; 3680 3681 dm_stack_bs_limits(limits, cc->sector_size); 3682 limits->dma_alignment = limits->logical_block_size - 1; 3683 3684 /* 3685 * For zoned dm-crypt targets, there will be no internal splitting of 3686 * write BIOs to avoid exceeding BIO_MAX_VECS vectors per BIO. But 3687 * without respecting this limit, crypt_alloc_buffer() will trigger a 3688 * BUG(). Avoid this by forcing DM core to split write BIOs to this 3689 * limit. 3690 */ 3691 if (ti->emulate_zone_append) 3692 limits->max_hw_sectors = min(limits->max_hw_sectors, 3693 BIO_MAX_VECS << PAGE_SECTORS_SHIFT); 3694 3695 limits->atomic_write_hw_unit_max = min(limits->atomic_write_hw_unit_max, 3696 BIO_MAX_VECS << PAGE_SHIFT); 3697 limits->atomic_write_hw_max = min(limits->atomic_write_hw_max, 3698 BIO_MAX_VECS << PAGE_SHIFT); 3699 } 3700 3701 static struct target_type crypt_target = { 3702 .name = "crypt", 3703 .version = {1, 29, 0}, 3704 .module = THIS_MODULE, 3705 .ctr = crypt_ctr, 3706 .dtr = crypt_dtr, 3707 .features = DM_TARGET_ZONED_HM | DM_TARGET_ATOMIC_WRITES, 3708 .report_zones = crypt_report_zones, 3709 .map = crypt_map, 3710 .status = crypt_status, 3711 .postsuspend = crypt_postsuspend, 3712 .preresume = crypt_preresume, 3713 .resume = crypt_resume, 3714 .message = crypt_message, 3715 .iterate_devices = crypt_iterate_devices, 3716 .io_hints = crypt_io_hints, 3717 }; 3718 module_dm(crypt); 3719 3720 MODULE_AUTHOR("Jana Saout <jana@saout.de>"); 3721 MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption"); 3722 MODULE_LICENSE("GPL"); 3723