crypto/aes_generic.c

   1 /*
   2  * Cryptographic API.
   3  *
   4  * AES Cipher Algorithm.
   5  *
   6  * Based on Brian Gladman's code.
   7  *
   8  * Linux developers:
   9  *  Alexander Kjeldaas <astor@fast.no>
  10  *  Herbert Valerio Riedel <hvr@hvrlab.org>
  11  *  Kyle McMartin <kyle@debian.org>
  12  *  Adam J. Richter <adam@yggdrasil.com> (conversion to 2.5 API).
  13  *
  14  * This program is free software; you can redistribute it and/or modify
  15  * it under the terms of the GNU General Public License as published by
  16  * the Free Software Foundation; either version 2 of the License, or
  17  * (at your option) any later version.
  18  *
  19  * ---------------------------------------------------------------------------
  20  * Copyright (c) 2002, Dr Brian Gladman <brg@gladman.me.uk>, Worcester, UK.
  21  * All rights reserved.
  22  *
  23  * LICENSE TERMS
  24  *
  25  * The free distribution and use of this software in both source and binary
  26  * form is allowed (with or without changes) provided that:
  27  *
  28  *   1. distributions of this source code include the above copyright
  29  *      notice, this list of conditions and the following disclaimer;
  30  *
  31  *   2. distributions in binary form include the above copyright
  32  *      notice, this list of conditions and the following disclaimer
  33  *      in the documentation and/or other associated materials;
  34  *
  35  *   3. the copyright holder's name is not used to endorse products
  36  *      built using this software without specific written permission.
  37  *
  38  * ALTERNATIVELY, provided that this notice is retained in full, this product
  39  * may be distributed under the terms of the GNU General Public License (GPL),
  40  * in which case the provisions of the GPL apply INSTEAD OF those given above.
  41  *
  42  * DISCLAIMER
  43  *
  44  * This software is provided 'as is' with no explicit or implied warranties
  45  * in respect of its properties, including, but not limited to, correctness
  46  * and/or fitness for purpose.
  47  * ---------------------------------------------------------------------------
  48  */
  49
  50 #include <crypto/aes.h>
  51 #include <linux/module.h>
  52 #include <linux/init.h>
  53 #include <linux/types.h>
  54 #include <linux/errno.h>
  55 #include <linux/crypto.h>
  56 #include <asm/byteorder.h>
  57
  58 static inline u8 byte(const u32 x, const unsigned n)
  59 {
  60         return x >> (n << 3);
  61 }
  62
  63 static u8 pow_tab[256] __initdata;
  64 static u8 log_tab[256] __initdata;
  65 static u8 sbx_tab[256] __initdata;
  66 static u8 isb_tab[256] __initdata;
  67 static u32 rco_tab[10];
  68
  69 u32 crypto_ft_tab[4][256];
  70 u32 crypto_fl_tab[4][256];
  71 u32 crypto_it_tab[4][256];
  72 u32 crypto_il_tab[4][256];
  73
  74 EXPORT_SYMBOL_GPL(crypto_ft_tab);
  75 EXPORT_SYMBOL_GPL(crypto_fl_tab);
  76 EXPORT_SYMBOL_GPL(crypto_it_tab);
  77 EXPORT_SYMBOL_GPL(crypto_il_tab);
  78
  79 static inline u8 __init f_mult(u8 a, u8 b)
  80 {
  81         u8 aa = log_tab[a], cc = aa + log_tab[b];
  82
  83         return pow_tab[cc + (cc < aa ? 1 : 0)];
  84 }
  85
  86 #define ff_mult(a, b)   (a && b ? f_mult(a, b) : 0)
  87
  88 static void __init gen_tabs(void)
  89 {
  90         u32 i, t;
  91         u8 p, q;
  92
  93         /*
  94          * log and power tables for GF(2**8) finite field with
  95          * 0x011b as modular polynomial - the simplest primitive
  96          * root is 0x03, used here to generate the tables
  97          */
  98
  99         for (i = 0, p = 1; i < 256; ++i) {
 100                 pow_tab[i] = (u8) p;
 101                 log_tab[p] = (u8) i;
 102
 103                 p ^= (p << 1) ^ (p & 0x80 ? 0x01b : 0);
 104         }
 105
 106         log_tab[1] = 0;
 107
 108         for (i = 0, p = 1; i < 10; ++i) {
 109                 rco_tab[i] = p;
 110
 111                 p = (p << 1) ^ (p & 0x80 ? 0x01b : 0);
 112         }
 113
 114         for (i = 0; i < 256; ++i) {
 115                 p = (i ? pow_tab[255 - log_tab[i]] : 0);
 116                 q = ((p >> 7) | (p << 1)) ^ ((p >> 6) | (p << 2));
 117                 p ^= 0x63 ^ q ^ ((q >> 6) | (q << 2));
 118                 sbx_tab[i] = p;
 119                 isb_tab[p] = (u8) i;
 120         }
 121
 122         for (i = 0; i < 256; ++i) {
 123                 p = sbx_tab[i];
 124
 125                 t = p;
 126                 crypto_fl_tab[0][i] = t;
 127                 crypto_fl_tab[1][i] = rol32(t, 8);
 128                 crypto_fl_tab[2][i] = rol32(t, 16);
 129                 crypto_fl_tab[3][i] = rol32(t, 24);
 130
 131                 t = ((u32) ff_mult(2, p)) |
 132                     ((u32) p << 8) |
 133                     ((u32) p << 16) | ((u32) ff_mult(3, p) << 24);
 134
 135                 crypto_ft_tab[0][i] = t;
 136                 crypto_ft_tab[1][i] = rol32(t, 8);
 137                 crypto_ft_tab[2][i] = rol32(t, 16);
 138                 crypto_ft_tab[3][i] = rol32(t, 24);
 139
 140                 p = isb_tab[i];
 141
 142                 t = p;
 143                 crypto_il_tab[0][i] = t;
 144                 crypto_il_tab[1][i] = rol32(t, 8);
 145                 crypto_il_tab[2][i] = rol32(t, 16);
 146                 crypto_il_tab[3][i] = rol32(t, 24);
 147
 148                 t = ((u32) ff_mult(14, p)) |
 149                     ((u32) ff_mult(9, p) << 8) |
 150                     ((u32) ff_mult(13, p) << 16) |
 151                     ((u32) ff_mult(11, p) << 24);
 152
 153                 crypto_it_tab[0][i] = t;
 154                 crypto_it_tab[1][i] = rol32(t, 8);
 155                 crypto_it_tab[2][i] = rol32(t, 16);
 156                 crypto_it_tab[3][i] = rol32(t, 24);
 157         }
 158 }
 159
 160 /* initialise the key schedule from the user supplied key */
 161
 162 #define star_x(x) (((x) & 0x7f7f7f7f) << 1) ^ ((((x) & 0x80808080) >> 7) * 0x1b)
 163
 164 #define imix_col(y,x)   do {            \
 165         u       = star_x(x);            \
 166         v       = star_x(u);            \
 167         w       = star_x(v);            \
 168         t       = w ^ (x);              \
 169         (y)     = u ^ v ^ w;            \
 170         (y)     ^= ror32(u ^ t, 8) ^    \
 171                 ror32(v ^ t, 16) ^      \
 172                 ror32(t, 24);           \
 173 } while (0)
 174
 175 #define ls_box(x)               \
 176         crypto_fl_tab[0][byte(x, 0)] ^  \
 177         crypto_fl_tab[1][byte(x, 1)] ^  \
 178         crypto_fl_tab[2][byte(x, 2)] ^  \
 179         crypto_fl_tab[3][byte(x, 3)]
 180
 181 #define loop4(i)        do {            \
 182         t = ror32(t, 8);                \
 183         t = ls_box(t) ^ rco_tab[i];     \
 184         t ^= ctx->key_enc[4 * i];               \
 185         ctx->key_enc[4 * i + 4] = t;            \
 186         t ^= ctx->key_enc[4 * i + 1];           \
 187         ctx->key_enc[4 * i + 5] = t;            \
 188         t ^= ctx->key_enc[4 * i + 2];           \
 189         ctx->key_enc[4 * i + 6] = t;            \
 190         t ^= ctx->key_enc[4 * i + 3];           \
 191         ctx->key_enc[4 * i + 7] = t;            \
 192 } while (0)
 193
 194 #define loop6(i)        do {            \
 195         t = ror32(t, 8);                \
 196         t = ls_box(t) ^ rco_tab[i];     \
 197         t ^= ctx->key_enc[6 * i];               \
 198         ctx->key_enc[6 * i + 6] = t;            \
 199         t ^= ctx->key_enc[6 * i + 1];           \
 200         ctx->key_enc[6 * i + 7] = t;            \
 201         t ^= ctx->key_enc[6 * i + 2];           \
 202         ctx->key_enc[6 * i + 8] = t;            \
 203         t ^= ctx->key_enc[6 * i + 3];           \
 204         ctx->key_enc[6 * i + 9] = t;            \
 205         t ^= ctx->key_enc[6 * i + 4];           \
 206         ctx->key_enc[6 * i + 10] = t;           \
 207         t ^= ctx->key_enc[6 * i + 5];           \
 208         ctx->key_enc[6 * i + 11] = t;           \
 209 } while (0)
 210
 211 #define loop8(i)        do {                    \
 212         t = ror32(t, 8);                        \
 213         t = ls_box(t) ^ rco_tab[i];             \
 214         t ^= ctx->key_enc[8 * i];                       \
 215         ctx->key_enc[8 * i + 8] = t;                    \
 216         t ^= ctx->key_enc[8 * i + 1];                   \
 217         ctx->key_enc[8 * i + 9] = t;                    \
 218         t ^= ctx->key_enc[8 * i + 2];                   \
 219         ctx->key_enc[8 * i + 10] = t;                   \
 220         t ^= ctx->key_enc[8 * i + 3];                   \
 221         ctx->key_enc[8 * i + 11] = t;                   \
 222         t  = ctx->key_enc[8 * i + 4] ^ ls_box(t);       \
 223         ctx->key_enc[8 * i + 12] = t;                   \
 224         t ^= ctx->key_enc[8 * i + 5];                   \
 225         ctx->key_enc[8 * i + 13] = t;                   \
 226         t ^= ctx->key_enc[8 * i + 6];                   \
 227         ctx->key_enc[8 * i + 14] = t;                   \
 228         t ^= ctx->key_enc[8 * i + 7];                   \
 229         ctx->key_enc[8 * i + 15] = t;                   \
 230 } while (0)
 231
 232 /**
 233  * crypto_aes_expand_key - Expands the AES key as described in FIPS-197
 234  * @ctx:        The location where the computed key will be stored.
 235  * @in_key:     The supplied key.
 236  * @key_len:    The length of the supplied key.
 237  *
 238  * Returns 0 on success. The function fails only if an invalid key size (or
 239  * pointer) is supplied.
 240  * The expanded key size is 240 bytes (max of 14 rounds with a unique 16 bytes
 241  * key schedule plus a 16 bytes key which is used before the first round).
 242  * The decryption key is prepared for the "Equivalent Inverse Cipher" as
 243  * described in FIPS-197. The first slot (16 bytes) of each key (enc or dec) is
 244  * for the initial combination, the second slot for the first round and so on.
 245  */
 246 int crypto_aes_expand_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
 247                 unsigned int key_len)
 248 {
 249         const __le32 *key = (const __le32 *)in_key;
 250         u32 i, t, u, v, w, j;
 251
 252         if (key_len != AES_KEYSIZE_128 && key_len != AES_KEYSIZE_192 &&
 253                         key_len != AES_KEYSIZE_256)
 254                 return -EINVAL;
 255
 256         ctx->key_length = key_len;
 257
 258         ctx->key_dec[key_len + 24] = ctx->key_enc[0] = le32_to_cpu(key[0]);
 259         ctx->key_dec[key_len + 25] = ctx->key_enc[1] = le32_to_cpu(key[1]);
 260         ctx->key_dec[key_len + 26] = ctx->key_enc[2] = le32_to_cpu(key[2]);
 261         ctx->key_dec[key_len + 27] = ctx->key_enc[3] = le32_to_cpu(key[3]);
 262
 263         switch (key_len) {
 264         case AES_KEYSIZE_128:
 265                 t = ctx->key_enc[3];
 266                 for (i = 0; i < 10; ++i)
 267                         loop4(i);
 268                 break;
 269
 270         case AES_KEYSIZE_192:
 271                 ctx->key_enc[4] = le32_to_cpu(key[4]);
 272                 t = ctx->key_enc[5] = le32_to_cpu(key[5]);
 273                 for (i = 0; i < 8; ++i)
 274                         loop6(i);
 275                 break;
 276
 277         case AES_KEYSIZE_256:
 278                 ctx->key_enc[4] = le32_to_cpu(key[4]);
 279                 ctx->key_enc[5] = le32_to_cpu(key[5]);
 280                 ctx->key_enc[6] = le32_to_cpu(key[6]);
 281                 t = ctx->key_enc[7] = le32_to_cpu(key[7]);
 282                 for (i = 0; i < 7; ++i)
 283                         loop8(i);
 284                 break;
 285         }
 286
 287         ctx->key_dec[0] = ctx->key_enc[key_len + 24];
 288         ctx->key_dec[1] = ctx->key_enc[key_len + 25];
 289         ctx->key_dec[2] = ctx->key_enc[key_len + 26];
 290         ctx->key_dec[3] = ctx->key_enc[key_len + 27];
 291
 292         for (i = 4; i < key_len + 24; ++i) {
 293                 j = key_len + 24 - (i & ~3) + (i & 3);
 294                 imix_col(ctx->key_dec[j], ctx->key_enc[i]);
 295         }
 296         return 0;
 297 }
 298 EXPORT_SYMBOL_GPL(crypto_aes_expand_key);
 299
 300 /**
 301  * crypto_aes_set_key - Set the AES key.
 302  * @tfm:        The %crypto_tfm that is used in the context.
 303  * @in_key:     The input key.
 304  * @key_len:    The size of the key.
 305  *
 306  * Returns 0 on success, on failure the %CRYPTO_TFM_RES_BAD_KEY_LEN flag in tfm
 307  * is set. The function uses crypto_aes_expand_key() to expand the key.
 308  * &crypto_aes_ctx _must_ be the private data embedded in @tfm which is
 309  * retrieved with crypto_tfm_ctx().
 310  */
 311 int crypto_aes_set_key(struct crypto_tfm *tfm, const u8 *in_key,
 312                 unsigned int key_len)
 313 {
 314         struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
 315         u32 *flags = &tfm->crt_flags;
 316         int ret;
 317
 318         ret = crypto_aes_expand_key(ctx, in_key, key_len);
 319         if (!ret)
 320                 return 0;
 321
 322         *flags |= CRYPTO_TFM_RES_BAD_KEY_LEN;
 323         return -EINVAL;
 324 }
 325 EXPORT_SYMBOL_GPL(crypto_aes_set_key);
 326
 327 /* encrypt a block of text */
 328
 329 #define f_rn(bo, bi, n, k)      do {                            \
 330         bo[n] = crypto_ft_tab[0][byte(bi[n], 0)] ^                      \
 331                 crypto_ft_tab[1][byte(bi[(n + 1) & 3], 1)] ^            \
 332                 crypto_ft_tab[2][byte(bi[(n + 2) & 3], 2)] ^            \
 333                 crypto_ft_tab[3][byte(bi[(n + 3) & 3], 3)] ^ *(k + n);  \
 334 } while (0)
 335
 336 #define f_nround(bo, bi, k)     do {\
 337         f_rn(bo, bi, 0, k);     \
 338         f_rn(bo, bi, 1, k);     \
 339         f_rn(bo, bi, 2, k);     \
 340         f_rn(bo, bi, 3, k);     \
 341         k += 4;                 \
 342 } while (0)
 343
 344 #define f_rl(bo, bi, n, k)      do {                            \
 345         bo[n] = crypto_fl_tab[0][byte(bi[n], 0)] ^                      \
 346                 crypto_fl_tab[1][byte(bi[(n + 1) & 3], 1)] ^            \
 347                 crypto_fl_tab[2][byte(bi[(n + 2) & 3], 2)] ^            \
 348                 crypto_fl_tab[3][byte(bi[(n + 3) & 3], 3)] ^ *(k + n);  \
 349 } while (0)
 350
 351 #define f_lround(bo, bi, k)     do {\
 352         f_rl(bo, bi, 0, k);     \
 353         f_rl(bo, bi, 1, k);     \
 354         f_rl(bo, bi, 2, k);     \
 355         f_rl(bo, bi, 3, k);     \
 356 } while (0)
 357
 358 static void aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
 359 {
 360         const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
 361         const __le32 *src = (const __le32 *)in;
 362         __le32 *dst = (__le32 *)out;
 363         u32 b0[4], b1[4];
 364         const u32 *kp = ctx->key_enc + 4;
 365         const int key_len = ctx->key_length;
 366
 367         b0[0] = le32_to_cpu(src[0]) ^ ctx->key_enc[0];
 368         b0[1] = le32_to_cpu(src[1]) ^ ctx->key_enc[1];
 369         b0[2] = le32_to_cpu(src[2]) ^ ctx->key_enc[2];
 370         b0[3] = le32_to_cpu(src[3]) ^ ctx->key_enc[3];
 371
 372         if (key_len > 24) {
 373                 f_nround(b1, b0, kp);
 374                 f_nround(b0, b1, kp);
 375         }
 376
 377         if (key_len > 16) {
 378                 f_nround(b1, b0, kp);
 379                 f_nround(b0, b1, kp);
 380         }
 381
 382         f_nround(b1, b0, kp);
 383         f_nround(b0, b1, kp);
 384         f_nround(b1, b0, kp);
 385         f_nround(b0, b1, kp);
 386         f_nround(b1, b0, kp);
 387         f_nround(b0, b1, kp);
 388         f_nround(b1, b0, kp);
 389         f_nround(b0, b1, kp);
 390         f_nround(b1, b0, kp);
 391         f_lround(b0, b1, kp);
 392
 393         dst[0] = cpu_to_le32(b0[0]);
 394         dst[1] = cpu_to_le32(b0[1]);
 395         dst[2] = cpu_to_le32(b0[2]);
 396         dst[3] = cpu_to_le32(b0[3]);
 397 }
 398
 399 /* decrypt a block of text */
 400
 401 #define i_rn(bo, bi, n, k)      do {                            \
 402         bo[n] = crypto_it_tab[0][byte(bi[n], 0)] ^                      \
 403                 crypto_it_tab[1][byte(bi[(n + 3) & 3], 1)] ^            \
 404                 crypto_it_tab[2][byte(bi[(n + 2) & 3], 2)] ^            \
 405                 crypto_it_tab[3][byte(bi[(n + 1) & 3], 3)] ^ *(k + n);  \
 406 } while (0)
 407
 408 #define i_nround(bo, bi, k)     do {\
 409         i_rn(bo, bi, 0, k);     \
 410         i_rn(bo, bi, 1, k);     \
 411         i_rn(bo, bi, 2, k);     \
 412         i_rn(bo, bi, 3, k);     \
 413         k += 4;                 \
 414 } while (0)
 415
 416 #define i_rl(bo, bi, n, k)      do {                    \
 417         bo[n] = crypto_il_tab[0][byte(bi[n], 0)] ^              \
 418         crypto_il_tab[1][byte(bi[(n + 3) & 3], 1)] ^            \
 419         crypto_il_tab[2][byte(bi[(n + 2) & 3], 2)] ^            \
 420         crypto_il_tab[3][byte(bi[(n + 1) & 3], 3)] ^ *(k + n);  \
 421 } while (0)
 422
 423 #define i_lround(bo, bi, k)     do {\
 424         i_rl(bo, bi, 0, k);     \
 425         i_rl(bo, bi, 1, k);     \
 426         i_rl(bo, bi, 2, k);     \
 427         i_rl(bo, bi, 3, k);     \
 428 } while (0)
 429
 430 static void aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
 431 {
 432         const struct crypto_aes_ctx *ctx = crypto_tfm_ctx(tfm);
 433         const __le32 *src = (const __le32 *)in;
 434         __le32 *dst = (__le32 *)out;
 435         u32 b0[4], b1[4];
 436         const int key_len = ctx->key_length;
 437         const u32 *kp = ctx->key_dec + 4;
 438
 439         b0[0] = le32_to_cpu(src[0]) ^  ctx->key_dec[0];
 440         b0[1] = le32_to_cpu(src[1]) ^  ctx->key_dec[1];
 441         b0[2] = le32_to_cpu(src[2]) ^  ctx->key_dec[2];
 442         b0[3] = le32_to_cpu(src[3]) ^  ctx->key_dec[3];
 443
 444         if (key_len > 24) {
 445                 i_nround(b1, b0, kp);
 446                 i_nround(b0, b1, kp);
 447         }
 448
 449         if (key_len > 16) {
 450                 i_nround(b1, b0, kp);
 451                 i_nround(b0, b1, kp);
 452         }
 453
 454         i_nround(b1, b0, kp);
 455         i_nround(b0, b1, kp);
 456         i_nround(b1, b0, kp);
 457         i_nround(b0, b1, kp);
 458         i_nround(b1, b0, kp);
 459         i_nround(b0, b1, kp);
 460         i_nround(b1, b0, kp);
 461         i_nround(b0, b1, kp);
 462         i_nround(b1, b0, kp);
 463         i_lround(b0, b1, kp);
 464
 465         dst[0] = cpu_to_le32(b0[0]);
 466         dst[1] = cpu_to_le32(b0[1]);
 467         dst[2] = cpu_to_le32(b0[2]);
 468         dst[3] = cpu_to_le32(b0[3]);
 469 }
 470
 471 static struct crypto_alg aes_alg = {
 472         .cra_name               =       "aes",
 473         .cra_driver_name        =       "aes-generic",
 474         .cra_priority           =       100,
 475         .cra_flags              =       CRYPTO_ALG_TYPE_CIPHER,
 476         .cra_blocksize          =       AES_BLOCK_SIZE,
 477         .cra_ctxsize            =       sizeof(struct crypto_aes_ctx),
 478         .cra_alignmask          =       3,
 479         .cra_module             =       THIS_MODULE,
 480         .cra_list               =       LIST_HEAD_INIT(aes_alg.cra_list),
 481         .cra_u                  =       {
 482                 .cipher = {
 483                         .cia_min_keysize        =       AES_MIN_KEY_SIZE,
 484                         .cia_max_keysize        =       AES_MAX_KEY_SIZE,
 485                         .cia_setkey             =       crypto_aes_set_key,
 486                         .cia_encrypt            =       aes_encrypt,
 487                         .cia_decrypt            =       aes_decrypt
 488                 }
 489         }
 490 };
 491
 492 static int __init aes_init(void)
 493 {
 494         gen_tabs();
 495         return crypto_register_alg(&aes_alg);
 496 }
 497
 498 static void __exit aes_fini(void)
 499 {
 500         crypto_unregister_alg(&aes_alg);
 501 }
 502
 503 module_init(aes_init);
 504 module_exit(aes_fini);
 505
 506 MODULE_DESCRIPTION("Rijndael (AES) Cipher Algorithm");
 507 MODULE_LICENSE("Dual BSD/GPL");
 508 MODULE_ALIAS("aes");