1/* $OpenBSD: aes.c,v 1.2 2020/07/22 13:54:30 tobhe Exp $ */ 2/* 3 * Copyright (c) 2016 Thomas Pornin <pornin@bolet.org> 4 * 5 * Modified for OpenBSD by Thomas Pornin and Mike Belopuhov. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining 8 * a copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sublicense, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice shall be 16 * included in all copies or substantial portions of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 21 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 22 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 23 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 24 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 25 * SOFTWARE. 26 */ 27 28#include <sys/types.h> 29#include <sys/systm.h> 30#include <sys/stdint.h> 31 32#include "aes.h" 33 34static inline void 35enc32le(void *dst, uint32_t x) 36{ 37 unsigned char *buf = dst; 38 39 buf[0] = (unsigned char)x; 40 buf[1] = (unsigned char)(x >> 8); 41 buf[2] = (unsigned char)(x >> 16); 42 buf[3] = (unsigned char)(x >> 24); 43} 44 45static inline uint32_t 46dec32le(const void *src) 47{ 48 const unsigned char *buf = src; 49 50 return (uint32_t)buf[0] 51 | ((uint32_t)buf[1] << 8) 52 | ((uint32_t)buf[2] << 16) 53 | ((uint32_t)buf[3] << 24); 54} 55 56/* 57 * This constant-time implementation is "bitsliced": the 128-bit state is 58 * split over eight 32-bit words q* in the following way: 59 * 60 * -- Input block consists in 16 bytes: 61 * a00 a10 a20 a30 a01 a11 a21 a31 a02 a12 a22 a32 a03 a13 a23 a33 62 * In the terminology of FIPS 197, this is a 4x4 matrix which is read 63 * column by column. 64 * 65 * -- Each byte is split into eight bits which are distributed over the 66 * eight words, at the same rank. Thus, for a byte x at rank k, bit 0 67 * (least significant) of x will be at rank k in q0 (if that bit is b, 68 * then it contributes "b << k" to the value of q0), bit 1 of x will be 69 * at rank k in q1, and so on. 70 * 71 * -- Ranks given to bits are in "row order" and are either all even, or 72 * all odd. Two independent AES states are thus interleaved, one using 73 * the even ranks, the other the odd ranks. Row order means: 74 * a00 a01 a02 a03 a10 a11 a12 a13 a20 a21 a22 a23 a30 a31 a32 a33 75 * 76 * Converting input bytes from two AES blocks to bitslice representation 77 * is done in the following way: 78 * -- Decode first block into the four words q0 q2 q4 q6, in that order, 79 * using little-endian convention. 80 * -- Decode second block into the four words q1 q3 q5 q7, in that order, 81 * using little-endian convention. 82 * -- Call aes_ct_ortho(). 83 * 84 * Converting back to bytes is done by using the reverse operations. Note 85 * that aes_ct_ortho() is its own inverse. 86 */ 87 88/* 89 * The AES S-box, as a bitsliced constant-time version. The input array 90 * consists in eight 32-bit words; 32 S-box instances are computed in 91 * parallel. Bits 0 to 7 of each S-box input (bit 0 is least significant) 92 * are spread over the words 0 to 7, at the same rank. 93 */ 94static void 95aes_ct_bitslice_Sbox(uint32_t *q) 96{ 97 /* 98 * This S-box implementation is a straightforward translation of 99 * the circuit described by Boyar and Peralta in "A new 100 * combinational logic minimization technique with applications 101 * to cryptology" (https://eprint.iacr.org/2009/191.pdf). 102 * 103 * Note that variables x* (input) and s* (output) are numbered 104 * in "reverse" order (x0 is the high bit, x7 is the low bit). 105 */ 106 107 uint32_t x0, x1, x2, x3, x4, x5, x6, x7; 108 uint32_t y1, y2, y3, y4, y5, y6, y7, y8, y9; 109 uint32_t y10, y11, y12, y13, y14, y15, y16, y17, y18, y19; 110 uint32_t y20, y21; 111 uint32_t z0, z1, z2, z3, z4, z5, z6, z7, z8, z9; 112 uint32_t z10, z11, z12, z13, z14, z15, z16, z17; 113 uint32_t t0, t1, t2, t3, t4, t5, t6, t7, t8, t9; 114 uint32_t t10, t11, t12, t13, t14, t15, t16, t17, t18, t19; 115 uint32_t t20, t21, t22, t23, t24, t25, t26, t27, t28, t29; 116 uint32_t t30, t31, t32, t33, t34, t35, t36, t37, t38, t39; 117 uint32_t t40, t41, t42, t43, t44, t45, t46, t47, t48, t49; 118 uint32_t t50, t51, t52, t53, t54, t55, t56, t57, t58, t59; 119 uint32_t t60, t61, t62, t63, t64, t65, t66, t67; 120 uint32_t s0, s1, s2, s3, s4, s5, s6, s7; 121 122 x0 = q[7]; 123 x1 = q[6]; 124 x2 = q[5]; 125 x3 = q[4]; 126 x4 = q[3]; 127 x5 = q[2]; 128 x6 = q[1]; 129 x7 = q[0]; 130 131 /* 132 * Top linear transformation. 133 */ 134 y14 = x3 ^ x5; 135 y13 = x0 ^ x6; 136 y9 = x0 ^ x3; 137 y8 = x0 ^ x5; 138 t0 = x1 ^ x2; 139 y1 = t0 ^ x7; 140 y4 = y1 ^ x3; 141 y12 = y13 ^ y14; 142 y2 = y1 ^ x0; 143 y5 = y1 ^ x6; 144 y3 = y5 ^ y8; 145 t1 = x4 ^ y12; 146 y15 = t1 ^ x5; 147 y20 = t1 ^ x1; 148 y6 = y15 ^ x7; 149 y10 = y15 ^ t0; 150 y11 = y20 ^ y9; 151 y7 = x7 ^ y11; 152 y17 = y10 ^ y11; 153 y19 = y10 ^ y8; 154 y16 = t0 ^ y11; 155 y21 = y13 ^ y16; 156 y18 = x0 ^ y16; 157 158 /* 159 * Non-linear section. 160 */ 161 t2 = y12 & y15; 162 t3 = y3 & y6; 163 t4 = t3 ^ t2; 164 t5 = y4 & x7; 165 t6 = t5 ^ t2; 166 t7 = y13 & y16; 167 t8 = y5 & y1; 168 t9 = t8 ^ t7; 169 t10 = y2 & y7; 170 t11 = t10 ^ t7; 171 t12 = y9 & y11; 172 t13 = y14 & y17; 173 t14 = t13 ^ t12; 174 t15 = y8 & y10; 175 t16 = t15 ^ t12; 176 t17 = t4 ^ t14; 177 t18 = t6 ^ t16; 178 t19 = t9 ^ t14; 179 t20 = t11 ^ t16; 180 t21 = t17 ^ y20; 181 t22 = t18 ^ y19; 182 t23 = t19 ^ y21; 183 t24 = t20 ^ y18; 184 185 t25 = t21 ^ t22; 186 t26 = t21 & t23; 187 t27 = t24 ^ t26; 188 t28 = t25 & t27; 189 t29 = t28 ^ t22; 190 t30 = t23 ^ t24; 191 t31 = t22 ^ t26; 192 t32 = t31 & t30; 193 t33 = t32 ^ t24; 194 t34 = t23 ^ t33; 195 t35 = t27 ^ t33; 196 t36 = t24 & t35; 197 t37 = t36 ^ t34; 198 t38 = t27 ^ t36; 199 t39 = t29 & t38; 200 t40 = t25 ^ t39; 201 202 t41 = t40 ^ t37; 203 t42 = t29 ^ t33; 204 t43 = t29 ^ t40; 205 t44 = t33 ^ t37; 206 t45 = t42 ^ t41; 207 z0 = t44 & y15; 208 z1 = t37 & y6; 209 z2 = t33 & x7; 210 z3 = t43 & y16; 211 z4 = t40 & y1; 212 z5 = t29 & y7; 213 z6 = t42 & y11; 214 z7 = t45 & y17; 215 z8 = t41 & y10; 216 z9 = t44 & y12; 217 z10 = t37 & y3; 218 z11 = t33 & y4; 219 z12 = t43 & y13; 220 z13 = t40 & y5; 221 z14 = t29 & y2; 222 z15 = t42 & y9; 223 z16 = t45 & y14; 224 z17 = t41 & y8; 225 226 /* 227 * Bottom linear transformation. 228 */ 229 t46 = z15 ^ z16; 230 t47 = z10 ^ z11; 231 t48 = z5 ^ z13; 232 t49 = z9 ^ z10; 233 t50 = z2 ^ z12; 234 t51 = z2 ^ z5; 235 t52 = z7 ^ z8; 236 t53 = z0 ^ z3; 237 t54 = z6 ^ z7; 238 t55 = z16 ^ z17; 239 t56 = z12 ^ t48; 240 t57 = t50 ^ t53; 241 t58 = z4 ^ t46; 242 t59 = z3 ^ t54; 243 t60 = t46 ^ t57; 244 t61 = z14 ^ t57; 245 t62 = t52 ^ t58; 246 t63 = t49 ^ t58; 247 t64 = z4 ^ t59; 248 t65 = t61 ^ t62; 249 t66 = z1 ^ t63; 250 s0 = t59 ^ t63; 251 s6 = t56 ^ ~t62; 252 s7 = t48 ^ ~t60; 253 t67 = t64 ^ t65; 254 s3 = t53 ^ t66; 255 s4 = t51 ^ t66; 256 s5 = t47 ^ t65; 257 s1 = t64 ^ ~s3; 258 s2 = t55 ^ ~t67; 259 260 q[7] = s0; 261 q[6] = s1; 262 q[5] = s2; 263 q[4] = s3; 264 q[3] = s4; 265 q[2] = s5; 266 q[1] = s6; 267 q[0] = s7; 268} 269 270/* 271 * Perform bytewise orthogonalization of eight 32-bit words. Bytes 272 * of q0..q7 are spread over all words: for a byte x that occurs 273 * at rank i in q[j] (byte x uses bits 8*i to 8*i+7 in q[j]), the bit 274 * of rank k in x (0 <= k <= 7) goes to q[k] at rank 8*i+j. 275 * 276 * This operation is an involution. 277 */ 278static void 279aes_ct_ortho(uint32_t *q) 280{ 281#define SWAPN(cl, ch, s, x, y) do { \ 282 uint32_t a, b; \ 283 a = (x); \ 284 b = (y); \ 285 (x) = (a & (uint32_t)cl) | ((b & (uint32_t)cl) << (s)); \ 286 (y) = ((a & (uint32_t)ch) >> (s)) | (b & (uint32_t)ch); \ 287 } while (0) 288 289#define SWAP2(x, y) SWAPN(0x55555555, 0xAAAAAAAA, 1, x, y) 290#define SWAP4(x, y) SWAPN(0x33333333, 0xCCCCCCCC, 2, x, y) 291#define SWAP8(x, y) SWAPN(0x0F0F0F0F, 0xF0F0F0F0, 4, x, y) 292 293 SWAP2(q[0], q[1]); 294 SWAP2(q[2], q[3]); 295 SWAP2(q[4], q[5]); 296 SWAP2(q[6], q[7]); 297 298 SWAP4(q[0], q[2]); 299 SWAP4(q[1], q[3]); 300 SWAP4(q[4], q[6]); 301 SWAP4(q[5], q[7]); 302 303 SWAP8(q[0], q[4]); 304 SWAP8(q[1], q[5]); 305 SWAP8(q[2], q[6]); 306 SWAP8(q[3], q[7]); 307} 308 309static inline uint32_t 310sub_word(uint32_t x) 311{ 312 uint32_t q[8]; 313 int i; 314 315 for (i = 0; i < 8; i ++) { 316 q[i] = x; 317 } 318 aes_ct_ortho(q); 319 aes_ct_bitslice_Sbox(q); 320 aes_ct_ortho(q); 321 return q[0]; 322} 323 324static const unsigned char Rcon[] = { 325 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36 326}; 327 328/* 329 * Base key schedule code. The function sub_word() must be defined 330 * below. Subkeys are produced in little-endian convention (but not 331 * bitsliced). Key length is expressed in bytes. 332 */ 333static unsigned 334aes_keysched_base(uint32_t *skey, const void *key, size_t key_len) 335{ 336 unsigned num_rounds; 337 int i, j, k, nk, nkf; 338 uint32_t tmp; 339 340 switch (key_len) { 341 case 16: 342 num_rounds = 10; 343 break; 344 case 24: 345 num_rounds = 12; 346 break; 347 case 32: 348 num_rounds = 14; 349 break; 350 default: 351 return 0; 352 } 353 nk = (int)(key_len >> 2); 354 nkf = (int)((num_rounds + 1) << 2); 355 for (i = 0; i < nk; i ++) { 356 tmp = dec32le((const unsigned char *)key + (i << 2)); 357 skey[i] = tmp; 358 } 359 tmp = skey[(key_len >> 2) - 1]; 360 for (i = nk, j = 0, k = 0; i < nkf; i ++) { 361 if (j == 0) { 362 tmp = (tmp << 24) | (tmp >> 8); 363 tmp = sub_word(tmp) ^ Rcon[k]; 364 } else if (nk > 6 && j == 4) { 365 tmp = sub_word(tmp); 366 } 367 tmp ^= skey[i - nk]; 368 skey[i] = tmp; 369 if (++ j == nk) { 370 j = 0; 371 k ++; 372 } 373 } 374 return num_rounds; 375} 376 377/* 378 * AES key schedule, constant-time version. skey[] is filled with n+1 379 * 128-bit subkeys, where n is the number of rounds (10 to 14, depending 380 * on key size). The number of rounds is returned. If the key size is 381 * invalid (not 16, 24 or 32), then 0 is returned. 382 */ 383unsigned 384aes_ct_keysched(uint32_t *comp_skey, const void *key, size_t key_len) 385{ 386 uint32_t skey[60]; 387 unsigned u, num_rounds; 388 389 num_rounds = aes_keysched_base(skey, key, key_len); 390 for (u = 0; u <= num_rounds; u ++) { 391 uint32_t q[8]; 392 393 q[0] = q[1] = skey[(u << 2) + 0]; 394 q[2] = q[3] = skey[(u << 2) + 1]; 395 q[4] = q[5] = skey[(u << 2) + 2]; 396 q[6] = q[7] = skey[(u << 2) + 3]; 397 aes_ct_ortho(q); 398 comp_skey[(u << 2) + 0] = 399 (q[0] & 0x55555555) | (q[1] & 0xAAAAAAAA); 400 comp_skey[(u << 2) + 1] = 401 (q[2] & 0x55555555) | (q[3] & 0xAAAAAAAA); 402 comp_skey[(u << 2) + 2] = 403 (q[4] & 0x55555555) | (q[5] & 0xAAAAAAAA); 404 comp_skey[(u << 2) + 3] = 405 (q[6] & 0x55555555) | (q[7] & 0xAAAAAAAA); 406 } 407 return num_rounds; 408} 409 410/* 411 * Expand AES subkeys as produced by aes_ct_keysched(), into 412 * a larger array suitable for aes_ct_bitslice_encrypt() and 413 * aes_ct_bitslice_decrypt(). 414 */ 415void 416aes_ct_skey_expand(uint32_t *skey, 417 unsigned num_rounds, const uint32_t *comp_skey) 418{ 419 unsigned u, v, n; 420 421 n = (num_rounds + 1) << 2; 422 for (u = 0, v = 0; u < n; u ++, v += 2) { 423 uint32_t x, y; 424 425 x = y = comp_skey[u]; 426 x &= 0x55555555; 427 skey[v + 0] = x | (x << 1); 428 y &= 0xAAAAAAAA; 429 skey[v + 1] = y | (y >> 1); 430 } 431} 432 433static inline void 434add_round_key(uint32_t *q, const uint32_t *sk) 435{ 436 q[0] ^= sk[0]; 437 q[1] ^= sk[1]; 438 q[2] ^= sk[2]; 439 q[3] ^= sk[3]; 440 q[4] ^= sk[4]; 441 q[5] ^= sk[5]; 442 q[6] ^= sk[6]; 443 q[7] ^= sk[7]; 444} 445 446static inline void 447shift_rows(uint32_t *q) 448{ 449 int i; 450 451 for (i = 0; i < 8; i ++) { 452 uint32_t x; 453 454 x = q[i]; 455 q[i] = (x & 0x000000FF) 456 | ((x & 0x0000FC00) >> 2) | ((x & 0x00000300) << 6) 457 | ((x & 0x00F00000) >> 4) | ((x & 0x000F0000) << 4) 458 | ((x & 0xC0000000) >> 6) | ((x & 0x3F000000) << 2); 459 } 460} 461 462static inline uint32_t 463rotr16(uint32_t x) 464{ 465 return (x << 16) | (x >> 16); 466} 467 468static inline void 469mix_columns(uint32_t *q) 470{ 471 uint32_t q0, q1, q2, q3, q4, q5, q6, q7; 472 uint32_t r0, r1, r2, r3, r4, r5, r6, r7; 473 474 q0 = q[0]; 475 q1 = q[1]; 476 q2 = q[2]; 477 q3 = q[3]; 478 q4 = q[4]; 479 q5 = q[5]; 480 q6 = q[6]; 481 q7 = q[7]; 482 r0 = (q0 >> 8) | (q0 << 24); 483 r1 = (q1 >> 8) | (q1 << 24); 484 r2 = (q2 >> 8) | (q2 << 24); 485 r3 = (q3 >> 8) | (q3 << 24); 486 r4 = (q4 >> 8) | (q4 << 24); 487 r5 = (q5 >> 8) | (q5 << 24); 488 r6 = (q6 >> 8) | (q6 << 24); 489 r7 = (q7 >> 8) | (q7 << 24); 490 491 q[0] = q7 ^ r7 ^ r0 ^ rotr16(q0 ^ r0); 492 q[1] = q0 ^ r0 ^ q7 ^ r7 ^ r1 ^ rotr16(q1 ^ r1); 493 q[2] = q1 ^ r1 ^ r2 ^ rotr16(q2 ^ r2); 494 q[3] = q2 ^ r2 ^ q7 ^ r7 ^ r3 ^ rotr16(q3 ^ r3); 495 q[4] = q3 ^ r3 ^ q7 ^ r7 ^ r4 ^ rotr16(q4 ^ r4); 496 q[5] = q4 ^ r4 ^ r5 ^ rotr16(q5 ^ r5); 497 q[6] = q5 ^ r5 ^ r6 ^ rotr16(q6 ^ r6); 498 q[7] = q6 ^ r6 ^ r7 ^ rotr16(q7 ^ r7); 499} 500 501/* 502 * Compute AES encryption on bitsliced data. Since input is stored on 503 * eight 32-bit words, two block encryptions are actually performed 504 * in parallel. 505 */ 506void 507aes_ct_bitslice_encrypt(unsigned num_rounds, 508 const uint32_t *skey, uint32_t *q) 509{ 510 unsigned u; 511 512 add_round_key(q, skey); 513 for (u = 1; u < num_rounds; u ++) { 514 aes_ct_bitslice_Sbox(q); 515 shift_rows(q); 516 mix_columns(q); 517 add_round_key(q, skey + (u << 3)); 518 } 519 aes_ct_bitslice_Sbox(q); 520 shift_rows(q); 521 add_round_key(q, skey + (num_rounds << 3)); 522} 523 524/* 525 * Like aes_ct_bitslice_Sbox(), but for the inverse S-box. 526 */ 527void 528aes_ct_bitslice_invSbox(uint32_t *q) 529{ 530 /* 531 * AES S-box is: 532 * S(x) = A(I(x)) ^ 0x63 533 * where I() is inversion in GF(256), and A() is a linear 534 * transform (0 is formally defined to be its own inverse). 535 * Since inversion is an involution, the inverse S-box can be 536 * computed from the S-box as: 537 * iS(x) = B(S(B(x ^ 0x63)) ^ 0x63) 538 * where B() is the inverse of A(). Indeed, for any y in GF(256): 539 * iS(S(y)) = B(A(I(B(A(I(y)) ^ 0x63 ^ 0x63))) ^ 0x63 ^ 0x63) = y 540 * 541 * Note: we reuse the implementation of the forward S-box, 542 * instead of duplicating it here, so that total code size is 543 * lower. By merging the B() transforms into the S-box circuit 544 * we could make faster CBC decryption, but CBC decryption is 545 * already quite faster than CBC encryption because we can 546 * process two blocks in parallel. 547 */ 548 uint32_t q0, q1, q2, q3, q4, q5, q6, q7; 549 550 q0 = ~q[0]; 551 q1 = ~q[1]; 552 q2 = q[2]; 553 q3 = q[3]; 554 q4 = q[4]; 555 q5 = ~q[5]; 556 q6 = ~q[6]; 557 q7 = q[7]; 558 q[7] = q1 ^ q4 ^ q6; 559 q[6] = q0 ^ q3 ^ q5; 560 q[5] = q7 ^ q2 ^ q4; 561 q[4] = q6 ^ q1 ^ q3; 562 q[3] = q5 ^ q0 ^ q2; 563 q[2] = q4 ^ q7 ^ q1; 564 q[1] = q3 ^ q6 ^ q0; 565 q[0] = q2 ^ q5 ^ q7; 566 567 aes_ct_bitslice_Sbox(q); 568 569 q0 = ~q[0]; 570 q1 = ~q[1]; 571 q2 = q[2]; 572 q3 = q[3]; 573 q4 = q[4]; 574 q5 = ~q[5]; 575 q6 = ~q[6]; 576 q7 = q[7]; 577 q[7] = q1 ^ q4 ^ q6; 578 q[6] = q0 ^ q3 ^ q5; 579 q[5] = q7 ^ q2 ^ q4; 580 q[4] = q6 ^ q1 ^ q3; 581 q[3] = q5 ^ q0 ^ q2; 582 q[2] = q4 ^ q7 ^ q1; 583 q[1] = q3 ^ q6 ^ q0; 584 q[0] = q2 ^ q5 ^ q7; 585} 586 587static inline void 588inv_shift_rows(uint32_t *q) 589{ 590 int i; 591 592 for (i = 0; i < 8; i ++) { 593 uint32_t x; 594 595 x = q[i]; 596 q[i] = (x & 0x000000FF) 597 | ((x & 0x00003F00) << 2) | ((x & 0x0000C000) >> 6) 598 | ((x & 0x000F0000) << 4) | ((x & 0x00F00000) >> 4) 599 | ((x & 0x03000000) << 6) | ((x & 0xFC000000) >> 2); 600 } 601} 602 603static void 604inv_mix_columns(uint32_t *q) 605{ 606 uint32_t q0, q1, q2, q3, q4, q5, q6, q7; 607 uint32_t r0, r1, r2, r3, r4, r5, r6, r7; 608 609 q0 = q[0]; 610 q1 = q[1]; 611 q2 = q[2]; 612 q3 = q[3]; 613 q4 = q[4]; 614 q5 = q[5]; 615 q6 = q[6]; 616 q7 = q[7]; 617 r0 = (q0 >> 8) | (q0 << 24); 618 r1 = (q1 >> 8) | (q1 << 24); 619 r2 = (q2 >> 8) | (q2 << 24); 620 r3 = (q3 >> 8) | (q3 << 24); 621 r4 = (q4 >> 8) | (q4 << 24); 622 r5 = (q5 >> 8) | (q5 << 24); 623 r6 = (q6 >> 8) | (q6 << 24); 624 r7 = (q7 >> 8) | (q7 << 24); 625 626 q[0] = q5 ^ q6 ^ q7 ^ r0 ^ r5 ^ r7 ^ rotr16(q0 ^ q5 ^ q6 ^ r0 ^ r5); 627 q[1] = q0 ^ q5 ^ r0 ^ r1 ^ r5 ^ r6 ^ r7 ^ rotr16(q1 ^ q5 ^ q7 ^ r1 ^ r5 ^ r6); 628 q[2] = q0 ^ q1 ^ q6 ^ r1 ^ r2 ^ r6 ^ r7 ^ rotr16(q0 ^ q2 ^ q6 ^ r2 ^ r6 ^ r7); 629 q[3] = q0 ^ q1 ^ q2 ^ q5 ^ q6 ^ r0 ^ r2 ^ r3 ^ r5 ^ rotr16(q0 ^ q1 ^ q3 ^ q5 ^ q6 ^ q7 ^ r0 ^ r3 ^ r5 ^ r7); 630 q[4] = q1 ^ q2 ^ q3 ^ q5 ^ r1 ^ r3 ^ r4 ^ r5 ^ r6 ^ r7 ^ rotr16(q1 ^ q2 ^ q4 ^ q5 ^ q7 ^ r1 ^ r4 ^ r5 ^ r6); 631 q[5] = q2 ^ q3 ^ q4 ^ q6 ^ r2 ^ r4 ^ r5 ^ r6 ^ r7 ^ rotr16(q2 ^ q3 ^ q5 ^ q6 ^ r2 ^ r5 ^ r6 ^ r7); 632 q[6] = q3 ^ q4 ^ q5 ^ q7 ^ r3 ^ r5 ^ r6 ^ r7 ^ rotr16(q3 ^ q4 ^ q6 ^ q7 ^ r3 ^ r6 ^ r7); 633 q[7] = q4 ^ q5 ^ q6 ^ r4 ^ r6 ^ r7 ^ rotr16(q4 ^ q5 ^ q7 ^ r4 ^ r7); 634} 635 636/* 637 * Compute AES decryption on bitsliced data. Since input is stored on 638 * eight 32-bit words, two block decryptions are actually performed 639 * in parallel. 640 */ 641void 642aes_ct_bitslice_decrypt(unsigned num_rounds, 643 const uint32_t *skey, uint32_t *q) 644{ 645 unsigned u; 646 647 add_round_key(q, skey + (num_rounds << 3)); 648 for (u = num_rounds - 1; u > 0; u --) { 649 inv_shift_rows(q); 650 aes_ct_bitslice_invSbox(q); 651 add_round_key(q, skey + (u << 3)); 652 inv_mix_columns(q); 653 } 654 inv_shift_rows(q); 655 aes_ct_bitslice_invSbox(q); 656 add_round_key(q, skey); 657} 658 659 660int 661AES_Setkey(AES_CTX *ctx, const uint8_t *key, int len) 662{ 663 ctx->num_rounds = aes_ct_keysched(ctx->sk, key, len); 664 if (ctx->num_rounds == 0) 665 return -1; 666 aes_ct_skey_expand(ctx->sk_exp, ctx->num_rounds, ctx->sk); 667 return 0; 668} 669 670void 671AES_Encrypt_ECB(AES_CTX *ctx, const uint8_t *src, 672 uint8_t *dst, size_t num_blocks) 673{ 674 while (num_blocks > 0) { 675 uint32_t q[8]; 676 677 q[0] = dec32le(src); 678 q[2] = dec32le(src + 4); 679 q[4] = dec32le(src + 8); 680 q[6] = dec32le(src + 12); 681 if (num_blocks > 1) { 682 q[1] = dec32le(src + 16); 683 q[3] = dec32le(src + 20); 684 q[5] = dec32le(src + 24); 685 q[7] = dec32le(src + 28); 686 } else { 687 q[1] = 0; 688 q[3] = 0; 689 q[5] = 0; 690 q[7] = 0; 691 } 692 aes_ct_ortho(q); 693 aes_ct_bitslice_encrypt(ctx->num_rounds, ctx->sk_exp, q); 694 aes_ct_ortho(q); 695 enc32le(dst, q[0]); 696 enc32le(dst + 4, q[2]); 697 enc32le(dst + 8, q[4]); 698 enc32le(dst + 12, q[6]); 699 if (num_blocks > 1) { 700 enc32le(dst + 16, q[1]); 701 enc32le(dst + 20, q[3]); 702 enc32le(dst + 24, q[5]); 703 enc32le(dst + 28, q[7]); 704 src += 32; 705 dst += 32; 706 num_blocks -= 2; 707 } else { 708 break; 709 } 710 } 711} 712 713void 714AES_Decrypt_ECB(AES_CTX *ctx, const uint8_t *src, 715 uint8_t *dst, size_t num_blocks) 716{ 717 while (num_blocks > 0) { 718 uint32_t q[8]; 719 720 q[0] = dec32le(src); 721 q[2] = dec32le(src + 4); 722 q[4] = dec32le(src + 8); 723 q[6] = dec32le(src + 12); 724 if (num_blocks > 1) { 725 q[1] = dec32le(src + 16); 726 q[3] = dec32le(src + 20); 727 q[5] = dec32le(src + 24); 728 q[7] = dec32le(src + 28); 729 } else { 730 q[1] = 0; 731 q[3] = 0; 732 q[5] = 0; 733 q[7] = 0; 734 } 735 aes_ct_ortho(q); 736 aes_ct_bitslice_decrypt(ctx->num_rounds, ctx->sk_exp, q); 737 aes_ct_ortho(q); 738 enc32le(dst, q[0]); 739 enc32le(dst + 4, q[2]); 740 enc32le(dst + 8, q[4]); 741 enc32le(dst + 12, q[6]); 742 if (num_blocks > 1) { 743 enc32le(dst + 16, q[1]); 744 enc32le(dst + 20, q[3]); 745 enc32le(dst + 24, q[5]); 746 enc32le(dst + 28, q[7]); 747 src += 32; 748 dst += 32; 749 num_blocks -= 2; 750 } else { 751 break; 752 } 753 } 754} 755 756void 757AES_Encrypt(AES_CTX *ctx, const uint8_t *src, uint8_t *dst) 758{ 759 AES_Encrypt_ECB(ctx, src, dst, 1); 760} 761 762void 763AES_Decrypt(AES_CTX *ctx, const uint8_t *src, uint8_t *dst) 764{ 765 AES_Decrypt_ECB(ctx, src, dst, 1); 766} 767 768int 769AES_KeySetup_Encrypt(uint32_t *skey, const uint8_t *key, int len) 770{ 771 unsigned r, u; 772 uint32_t tkey[60]; 773 774 r = aes_keysched_base(tkey, key, len); 775 if (r == 0) { 776 return 0; 777 } 778 for (u = 0; u < ((r + 1) << 2); u ++) { 779 uint32_t w; 780 781 w = tkey[u]; 782 skey[u] = (w << 24) 783 | ((w & 0x0000FF00) << 8) 784 | ((w & 0x00FF0000) >> 8) 785 | (w >> 24); 786 } 787 return r; 788} 789 790/* 791 * Reduce value x modulo polynomial x^8+x^4+x^3+x+1. This works as 792 * long as x fits on 12 bits at most. 793 */ 794static inline uint32_t 795redgf256(uint32_t x) 796{ 797 uint32_t h; 798 799 h = x >> 8; 800 return (x ^ h ^ (h << 1) ^ (h << 3) ^ (h << 4)) & 0xFF; 801} 802 803/* 804 * Multiplication by 0x09 in GF(256). 805 */ 806static inline uint32_t 807mul9(uint32_t x) 808{ 809 return redgf256(x ^ (x << 3)); 810} 811 812/* 813 * Multiplication by 0x0B in GF(256). 814 */ 815static inline uint32_t 816mulb(uint32_t x) 817{ 818 return redgf256(x ^ (x << 1) ^ (x << 3)); 819} 820 821/* 822 * Multiplication by 0x0D in GF(256). 823 */ 824static inline uint32_t 825muld(uint32_t x) 826{ 827 return redgf256(x ^ (x << 2) ^ (x << 3)); 828} 829 830/* 831 * Multiplication by 0x0E in GF(256). 832 */ 833static inline uint32_t 834mule(uint32_t x) 835{ 836 return redgf256((x << 1) ^ (x << 2) ^ (x << 3)); 837} 838 839int 840AES_KeySetup_Decrypt(uint32_t *skey, const uint8_t *key, int len) 841{ 842 unsigned r, u; 843 uint32_t tkey[60]; 844 845 /* 846 * Compute encryption subkeys. We get them in big-endian 847 * notation. 848 */ 849 r = AES_KeySetup_Encrypt(tkey, key, len); 850 if (r == 0) { 851 return 0; 852 } 853 854 /* 855 * Copy the subkeys in reverse order. Also, apply InvMixColumns() 856 * on the subkeys (except first and last). 857 */ 858 memcpy(skey + (r << 2), tkey, 4 * sizeof(uint32_t)); 859 memcpy(skey, tkey + (r << 2), 4 * sizeof(uint32_t)); 860 for (u = 4; u < (r << 2); u ++) { 861 uint32_t sk, sk0, sk1, sk2, sk3; 862 uint32_t tk, tk0, tk1, tk2, tk3; 863 864 sk = tkey[u]; 865 sk0 = sk >> 24; 866 sk1 = (sk >> 16) & 0xFF; 867 sk2 = (sk >> 8) & 0xFF; 868 sk3 = sk & 0xFF; 869 tk0 = mule(sk0) ^ mulb(sk1) ^ muld(sk2) ^ mul9(sk3); 870 tk1 = mul9(sk0) ^ mule(sk1) ^ mulb(sk2) ^ muld(sk3); 871 tk2 = muld(sk0) ^ mul9(sk1) ^ mule(sk2) ^ mulb(sk3); 872 tk3 = mulb(sk0) ^ muld(sk1) ^ mul9(sk2) ^ mule(sk3); 873 tk = (tk0 << 24) ^ (tk1 << 16) ^ (tk2 << 8) ^ tk3; 874 skey[((r - (u >> 2)) << 2) + (u & 3)] = tk; 875 } 876 877 return r; 878} 879