| OLD | NEW |
|---|
| (Empty) |
| 1 /* This Source Code Form is subject to the terms of the Mozilla Public | |
| 2 * License, v. 2.0. If a copy of the MPL was not distributed with this | |
| 3 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ | |
| 4 /* $Id: mpmontg.c,v 1.25 2012/11/14 01:14:11 wtc%google.com Exp $ */ | |
| 5 | |
| 6 /* This file implements moduluar exponentiation using Montgomery's | |
| 7 * method for modular reduction. This file implements the method | |
| 8 * described as "Improvement 2" in the paper "A Cryptogrpahic Library for | |
| 9 * the Motorola DSP56000" by Stephen R. Dusse' and Burton S. Kaliski Jr. | |
| 10 * published in "Advances in Cryptology: Proceedings of EUROCRYPT '90" | |
| 11 * "Lecture Notes in Computer Science" volume 473, 1991, pg 230-244, | |
| 12 * published by Springer Verlag. | |
| 13 */ | |
| 14 | |
| 15 #define MP_USING_CACHE_SAFE_MOD_EXP 1 | |
| 16 #include <string.h> | |
| 17 #include "mpi-priv.h" | |
| 18 #include "mplogic.h" | |
| 19 #include "mpprime.h" | |
| 20 #ifdef MP_USING_MONT_MULF | |
| 21 #include "montmulf.h" | |
| 22 #endif | |
| 23 #include <stddef.h> /* ptrdiff_t */ | |
| 24 | |
| 25 /* if MP_CHAR_STORE_SLOW is defined, we */ | |
| 26 /* need to know endianness of this platform. */ | |
| 27 #ifdef MP_CHAR_STORE_SLOW | |
| 28 #if !defined(MP_IS_BIG_ENDIAN) && !defined(MP_IS_LITTLE_ENDIAN) | |
| 29 #error "You must define MP_IS_BIG_ENDIAN or MP_IS_LITTLE_ENDIAN\n" \ | |
| 30 " if you define MP_CHAR_STORE_SLOW." | |
| 31 #endif | |
| 32 #endif | |
| 33 | |
| 34 #define STATIC | |
| 35 | |
| 36 #define MAX_ODD_INTS 32 /* 2 ** (WINDOW_BITS - 1) */ | |
| 37 | |
| 38 /*! computes T = REDC(T), 2^b == R | |
| 39 \param T < RN | |
| 40 */ | |
| 41 mp_err s_mp_redc(mp_int *T, mp_mont_modulus *mmm) | |
| 42 { | |
| 43 mp_err res; | |
| 44 mp_size i; | |
| 45 | |
| 46 i = (MP_USED(&mmm->N) << 1) + 1; | |
| 47 MP_CHECKOK( s_mp_pad(T, i) ); | |
| 48 for (i = 0; i < MP_USED(&mmm->N); ++i ) { | |
| 49 mp_digit m_i = MP_DIGIT(T, i) * mmm->n0prime; | |
| 50 /* T += N * m_i * (MP_RADIX ** i); */ | |
| 51 MP_CHECKOK( s_mp_mul_d_add_offset(&mmm->N, m_i, T, i) ); | |
| 52 } | |
| 53 s_mp_clamp(T); | |
| 54 | |
| 55 /* T /= R */ | |
| 56 s_mp_rshd( T, MP_USED(&mmm->N) ); | |
| 57 | |
| 58 if ((res = s_mp_cmp(T, &mmm->N)) >= 0) { | |
| 59 /* T = T - N */ | |
| 60 MP_CHECKOK( s_mp_sub(T, &mmm->N) ); | |
| 61 #ifdef DEBUG | |
| 62 if ((res = mp_cmp(T, &mmm->N)) >= 0) { | |
| 63 res = MP_UNDEF; | |
| 64 goto CLEANUP; | |
| 65 } | |
| 66 #endif | |
| 67 } | |
| 68 res = MP_OKAY; | |
| 69 CLEANUP: | |
| 70 return res; | |
| 71 } | |
| 72 | |
| 73 #if !defined(MP_MONT_USE_MP_MUL) | |
| 74 | |
| 75 /*! c <- REDC( a * b ) mod N | |
| 76 \param a < N i.e. "reduced" | |
| 77 \param b < N i.e. "reduced" | |
| 78 \param mmm modulus N and n0' of N | |
| 79 */ | |
| 80 mp_err s_mp_mul_mont(const mp_int *a, const mp_int *b, mp_int *c, | |
| 81 mp_mont_modulus *mmm) | |
| 82 { | |
| 83 mp_digit *pb; | |
| 84 mp_digit m_i; | |
| 85 mp_err res; | |
| 86 mp_size ib; /* "index b": index of current digit of B */ | |
| 87 mp_size useda, usedb; | |
| 88 | |
| 89 ARGCHK(a != NULL && b != NULL && c != NULL, MP_BADARG); | |
| 90 | |
| 91 if (MP_USED(a) < MP_USED(b)) { | |
| 92 const mp_int *xch = b; /* switch a and b, to do fewer outer loops */ | |
| 93 b = a; | |
| 94 a = xch; | |
| 95 } | |
| 96 | |
| 97 MP_USED(c) = 1; MP_DIGIT(c, 0) = 0; | |
| 98 ib = (MP_USED(&mmm->N) << 1) + 1; | |
| 99 if((res = s_mp_pad(c, ib)) != MP_OKAY) | |
| 100 goto CLEANUP; | |
| 101 | |
| 102 useda = MP_USED(a); | |
| 103 pb = MP_DIGITS(b); | |
| 104 s_mpv_mul_d(MP_DIGITS(a), useda, *pb++, MP_DIGITS(c)); | |
| 105 s_mp_setz(MP_DIGITS(c) + useda + 1, ib - (useda + 1)); | |
| 106 m_i = MP_DIGIT(c, 0) * mmm->n0prime; | |
| 107 s_mp_mul_d_add_offset(&mmm->N, m_i, c, 0); | |
| 108 | |
| 109 /* Outer loop: Digits of b */ | |
| 110 usedb = MP_USED(b); | |
| 111 for (ib = 1; ib < usedb; ib++) { | |
| 112 mp_digit b_i = *pb++; | |
| 113 | |
| 114 /* Inner product: Digits of a */ | |
| 115 if (b_i) | |
| 116 s_mpv_mul_d_add_prop(MP_DIGITS(a), useda, b_i, MP_DIGITS(c) + ib); | |
| 117 m_i = MP_DIGIT(c, ib) * mmm->n0prime; | |
| 118 s_mp_mul_d_add_offset(&mmm->N, m_i, c, ib); | |
| 119 } | |
| 120 if (usedb < MP_USED(&mmm->N)) { | |
| 121 for (usedb = MP_USED(&mmm->N); ib < usedb; ++ib ) { | |
| 122 m_i = MP_DIGIT(c, ib) * mmm->n0prime; | |
| 123 s_mp_mul_d_add_offset(&mmm->N, m_i, c, ib); | |
| 124 } | |
| 125 } | |
| 126 s_mp_clamp(c); | |
| 127 s_mp_rshd( c, MP_USED(&mmm->N) ); /* c /= R */ | |
| 128 if (s_mp_cmp(c, &mmm->N) >= 0) { | |
| 129 MP_CHECKOK( s_mp_sub(c, &mmm->N) ); | |
| 130 } | |
| 131 res = MP_OKAY; | |
| 132 | |
| 133 CLEANUP: | |
| 134 return res; | |
| 135 } | |
| 136 #endif | |
| 137 | |
| 138 STATIC | |
| 139 mp_err s_mp_to_mont(const mp_int *x, mp_mont_modulus *mmm, mp_int *xMont) | |
| 140 { | |
| 141 mp_err res; | |
| 142 | |
| 143 /* xMont = x * R mod N where N is modulus */ | |
| 144 MP_CHECKOK( mp_copy( x, xMont ) ); | |
| 145 MP_CHECKOK( s_mp_lshd( xMont, MP_USED(&mmm->N) ) ); /* xMont = x << b */ | |
| 146 MP_CHECKOK( mp_div(xMont, &mmm->N, 0, xMont) ); /* mod N */ | |
| 147 CLEANUP: | |
| 148 return res; | |
| 149 } | |
| 150 | |
| 151 #ifdef MP_USING_MONT_MULF | |
| 152 | |
| 153 /* the floating point multiply is already cache safe, | |
| 154 * don't turn on cache safe unless we specifically | |
| 155 * force it */ | |
| 156 #ifndef MP_FORCE_CACHE_SAFE | |
| 157 #undef MP_USING_CACHE_SAFE_MOD_EXP | |
| 158 #endif | |
| 159 | |
| 160 unsigned int mp_using_mont_mulf = 1; | |
| 161 | |
| 162 /* computes montgomery square of the integer in mResult */ | |
| 163 #define SQR \ | |
| 164 conv_i32_to_d32_and_d16(dm1, d16Tmp, mResult, nLen); \ | |
| 165 mont_mulf_noconv(mResult, dm1, d16Tmp, \ | |
| 166 dTmp, dn, MP_DIGITS(modulus), nLen, dn0) | |
| 167 | |
| 168 /* computes montgomery product of x and the integer in mResult */ | |
| 169 #define MUL(x) \ | |
| 170 conv_i32_to_d32(dm1, mResult, nLen); \ | |
| 171 mont_mulf_noconv(mResult, dm1, oddPowers[x], \ | |
| 172 dTmp, dn, MP_DIGITS(modulus), nLen, dn0) | |
| 173 | |
| 174 /* Do modular exponentiation using floating point multiply code. */ | |
| 175 mp_err mp_exptmod_f(const mp_int * montBase, | |
| 176 const mp_int * exponent, | |
| 177 const mp_int * modulus, | |
| 178 mp_int * result, | |
| 179 mp_mont_modulus *mmm, | |
| 180 int nLen, | |
| 181 mp_size bits_in_exponent, | |
| 182 mp_size window_bits, | |
| 183 mp_size odd_ints) | |
| 184 { | |
| 185 mp_digit *mResult; | |
| 186 double *dBuf = 0, *dm1, *dn, *dSqr, *d16Tmp, *dTmp; | |
| 187 double dn0; | |
| 188 mp_size i; | |
| 189 mp_err res; | |
| 190 int expOff; | |
| 191 int dSize = 0, oddPowSize, dTmpSize; | |
| 192 mp_int accum1; | |
| 193 double *oddPowers[MAX_ODD_INTS]; | |
| 194 | |
| 195 /* function for computing n0prime only works if n0 is odd */ | |
| 196 | |
| 197 MP_DIGITS(&accum1) = 0; | |
| 198 | |
| 199 for (i = 0; i < MAX_ODD_INTS; ++i) | |
| 200 oddPowers[i] = 0; | |
| 201 | |
| 202 MP_CHECKOK( mp_init_size(&accum1, 3 * nLen + 2) ); | |
| 203 | |
| 204 mp_set(&accum1, 1); | |
| 205 MP_CHECKOK( s_mp_to_mont(&accum1, mmm, &accum1) ); | |
| 206 MP_CHECKOK( s_mp_pad(&accum1, nLen) ); | |
| 207 | |
| 208 oddPowSize = 2 * nLen + 1; | |
| 209 dTmpSize = 2 * oddPowSize; | |
| 210 dSize = sizeof(double) * (nLen * 4 + 1 + | |
| 211 ((odd_ints + 1) * oddPowSize) + dTmpSize); | |
| 212 dBuf = (double *)malloc(dSize); | |
| 213 dm1 = dBuf; /* array of d32 */ | |
| 214 dn = dBuf + nLen; /* array of d32 */ | |
| 215 dSqr = dn + nLen; /* array of d32 */ | |
| 216 d16Tmp = dSqr + nLen; /* array of d16 */ | |
| 217 dTmp = d16Tmp + oddPowSize; | |
| 218 | |
| 219 for (i = 0; i < odd_ints; ++i) { | |
| 220 oddPowers[i] = dTmp; | |
| 221 dTmp += oddPowSize; | |
| 222 } | |
| 223 mResult = (mp_digit *)(dTmp + dTmpSize); /* size is nLen + 1 */ | |
| 224 | |
| 225 /* Make dn and dn0 */ | |
| 226 conv_i32_to_d32(dn, MP_DIGITS(modulus), nLen); | |
| 227 dn0 = (double)(mmm->n0prime & 0xffff); | |
| 228 | |
| 229 /* Make dSqr */ | |
| 230 conv_i32_to_d32_and_d16(dm1, oddPowers[0], MP_DIGITS(montBase), nLen); | |
| 231 mont_mulf_noconv(mResult, dm1, oddPowers[0], | |
| 232 dTmp, dn, MP_DIGITS(modulus), nLen, dn0); | |
| 233 conv_i32_to_d32(dSqr, mResult, nLen); | |
| 234 | |
| 235 for (i = 1; i < odd_ints; ++i) { | |
| 236 mont_mulf_noconv(mResult, dSqr, oddPowers[i - 1], | |
| 237 dTmp, dn, MP_DIGITS(modulus), nLen, dn0); | |
| 238 conv_i32_to_d16(oddPowers[i], mResult, nLen); | |
| 239 } | |
| 240 | |
| 241 s_mp_copy(MP_DIGITS(&accum1), mResult, nLen); /* from, to, len */ | |
| 242 | |
| 243 for (expOff = bits_in_exponent - window_bits; expOff >= 0; expOff -= window_bi
ts) { | |
| 244 mp_size smallExp; | |
| 245 MP_CHECKOK( mpl_get_bits(exponent, expOff, window_bits) ); | |
| 246 smallExp = (mp_size)res; | |
| 247 | |
| 248 if (window_bits == 1) { | |
| 249 if (!smallExp) { | |
| 250 SQR; | |
| 251 } else if (smallExp & 1) { | |
| 252 SQR; MUL(0); | |
| 253 } else { | |
| 254 abort(); | |
| 255 } | |
| 256 } else if (window_bits == 4) { | |
| 257 if (!smallExp) { | |
| 258 SQR; SQR; SQR; SQR; | |
| 259 } else if (smallExp & 1) { | |
| 260 SQR; SQR; SQR; SQR; MUL(smallExp/2); | |
| 261 } else if (smallExp & 2) { | |
| 262 SQR; SQR; SQR; MUL(smallExp/4); SQR; | |
| 263 } else if (smallExp & 4) { | |
| 264 SQR; SQR; MUL(smallExp/8); SQR; SQR; | |
| 265 } else if (smallExp & 8) { | |
| 266 SQR; MUL(smallExp/16); SQR; SQR; SQR; | |
| 267 } else { | |
| 268 abort(); | |
| 269 } | |
| 270 } else if (window_bits == 5) { | |
| 271 if (!smallExp) { | |
| 272 SQR; SQR; SQR; SQR; SQR; | |
| 273 } else if (smallExp & 1) { | |
| 274 SQR; SQR; SQR; SQR; SQR; MUL(smallExp/2); | |
| 275 } else if (smallExp & 2) { | |
| 276 SQR; SQR; SQR; SQR; MUL(smallExp/4); SQR; | |
| 277 } else if (smallExp & 4) { | |
| 278 SQR; SQR; SQR; MUL(smallExp/8); SQR; SQR; | |
| 279 } else if (smallExp & 8) { | |
| 280 SQR; SQR; MUL(smallExp/16); SQR; SQR; SQR; | |
| 281 } else if (smallExp & 0x10) { | |
| 282 SQR; MUL(smallExp/32); SQR; SQR; SQR; SQR; | |
| 283 } else { | |
| 284 abort(); | |
| 285 } | |
| 286 } else if (window_bits == 6) { | |
| 287 if (!smallExp) { | |
| 288 SQR; SQR; SQR; SQR; SQR; SQR; | |
| 289 } else if (smallExp & 1) { | |
| 290 SQR; SQR; SQR; SQR; SQR; SQR; MUL(smallExp/2); | |
| 291 } else if (smallExp & 2) { | |
| 292 SQR; SQR; SQR; SQR; SQR; MUL(smallExp/4); SQR; | |
| 293 } else if (smallExp & 4) { | |
| 294 SQR; SQR; SQR; SQR; MUL(smallExp/8); SQR; SQR; | |
| 295 } else if (smallExp & 8) { | |
| 296 SQR; SQR; SQR; MUL(smallExp/16); SQR; SQR; SQR; | |
| 297 } else if (smallExp & 0x10) { | |
| 298 SQR; SQR; MUL(smallExp/32); SQR; SQR; SQR; SQR; | |
| 299 } else if (smallExp & 0x20) { | |
| 300 SQR; MUL(smallExp/64); SQR; SQR; SQR; SQR; SQR; | |
| 301 } else { | |
| 302 abort(); | |
| 303 } | |
| 304 } else { | |
| 305 abort(); | |
| 306 } | |
| 307 } | |
| 308 | |
| 309 s_mp_copy(mResult, MP_DIGITS(&accum1), nLen); /* from, to, len */ | |
| 310 | |
| 311 res = s_mp_redc(&accum1, mmm); | |
| 312 mp_exch(&accum1, result); | |
| 313 | |
| 314 CLEANUP: | |
| 315 mp_clear(&accum1); | |
| 316 if (dBuf) { | |
| 317 if (dSize) | |
| 318 memset(dBuf, 0, dSize); | |
| 319 free(dBuf); | |
| 320 } | |
| 321 | |
| 322 return res; | |
| 323 } | |
| 324 #undef SQR | |
| 325 #undef MUL | |
| 326 #endif | |
| 327 | |
| 328 #define SQR(a,b) \ | |
| 329 MP_CHECKOK( mp_sqr(a, b) );\ | |
| 330 MP_CHECKOK( s_mp_redc(b, mmm) ) | |
| 331 | |
| 332 #if defined(MP_MONT_USE_MP_MUL) | |
| 333 #define MUL(x,a,b) \ | |
| 334 MP_CHECKOK( mp_mul(a, oddPowers + (x), b) ); \ | |
| 335 MP_CHECKOK( s_mp_redc(b, mmm) ) | |
| 336 #else | |
| 337 #define MUL(x,a,b) \ | |
| 338 MP_CHECKOK( s_mp_mul_mont(a, oddPowers + (x), b, mmm) ) | |
| 339 #endif | |
| 340 | |
| 341 #define SWAPPA ptmp = pa1; pa1 = pa2; pa2 = ptmp | |
| 342 | |
| 343 /* Do modular exponentiation using integer multiply code. */ | |
| 344 mp_err mp_exptmod_i(const mp_int * montBase, | |
| 345 const mp_int * exponent, | |
| 346 const mp_int * modulus, | |
| 347 mp_int * result, | |
| 348 mp_mont_modulus *mmm, | |
| 349 int nLen, | |
| 350 mp_size bits_in_exponent, | |
| 351 mp_size window_bits, | |
| 352 mp_size odd_ints) | |
| 353 { | |
| 354 mp_int *pa1, *pa2, *ptmp; | |
| 355 mp_size i; | |
| 356 mp_err res; | |
| 357 int expOff; | |
| 358 mp_int accum1, accum2, power2, oddPowers[MAX_ODD_INTS]; | |
| 359 | |
| 360 /* power2 = base ** 2; oddPowers[i] = base ** (2*i + 1); */ | |
| 361 /* oddPowers[i] = base ** (2*i + 1); */ | |
| 362 | |
| 363 MP_DIGITS(&accum1) = 0; | |
| 364 MP_DIGITS(&accum2) = 0; | |
| 365 MP_DIGITS(&power2) = 0; | |
| 366 for (i = 0; i < MAX_ODD_INTS; ++i) { | |
| 367 MP_DIGITS(oddPowers + i) = 0; | |
| 368 } | |
| 369 | |
| 370 MP_CHECKOK( mp_init_size(&accum1, 3 * nLen + 2) ); | |
| 371 MP_CHECKOK( mp_init_size(&accum2, 3 * nLen + 2) ); | |
| 372 | |
| 373 MP_CHECKOK( mp_init_copy(&oddPowers[0], montBase) ); | |
| 374 | |
| 375 mp_init_size(&power2, nLen + 2 * MP_USED(montBase) + 2); | |
| 376 MP_CHECKOK( mp_sqr(montBase, &power2) ); /* power2 = montBase ** 2 */ | |
| 377 MP_CHECKOK( s_mp_redc(&power2, mmm) ); | |
| 378 | |
| 379 for (i = 1; i < odd_ints; ++i) { | |
| 380 mp_init_size(oddPowers + i, nLen + 2 * MP_USED(&power2) + 2); | |
| 381 MP_CHECKOK( mp_mul(oddPowers + (i - 1), &power2, oddPowers + i) ); | |
| 382 MP_CHECKOK( s_mp_redc(oddPowers + i, mmm) ); | |
| 383 } | |
| 384 | |
| 385 /* set accumulator to montgomery residue of 1 */ | |
| 386 mp_set(&accum1, 1); | |
| 387 MP_CHECKOK( s_mp_to_mont(&accum1, mmm, &accum1) ); | |
| 388 pa1 = &accum1; | |
| 389 pa2 = &accum2; | |
| 390 | |
| 391 for (expOff = bits_in_exponent - window_bits; expOff >= 0; expOff -= window_bi
ts) { | |
| 392 mp_size smallExp; | |
| 393 MP_CHECKOK( mpl_get_bits(exponent, expOff, window_bits) ); | |
| 394 smallExp = (mp_size)res; | |
| 395 | |
| 396 if (window_bits == 1) { | |
| 397 if (!smallExp) { | |
| 398 SQR(pa1,pa2); SWAPPA; | |
| 399 } else if (smallExp & 1) { | |
| 400 SQR(pa1,pa2); MUL(0,pa2,pa1); | |
| 401 } else { | |
| 402 abort(); | |
| 403 } | |
| 404 } else if (window_bits == 4) { | |
| 405 if (!smallExp) { | |
| 406 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 407 } else if (smallExp & 1) { | |
| 408 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 409 MUL(smallExp/2, pa1,pa2); SWAPPA; | |
| 410 } else if (smallExp & 2) { | |
| 411 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); | |
| 412 MUL(smallExp/4,pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 413 } else if (smallExp & 4) { | |
| 414 SQR(pa1,pa2); SQR(pa2,pa1); MUL(smallExp/8,pa1,pa2); | |
| 415 SQR(pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 416 } else if (smallExp & 8) { | |
| 417 SQR(pa1,pa2); MUL(smallExp/16,pa2,pa1); SQR(pa1,pa2); | |
| 418 SQR(pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 419 } else { | |
| 420 abort(); | |
| 421 } | |
| 422 } else if (window_bits == 5) { | |
| 423 if (!smallExp) { | |
| 424 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 425 SQR(pa1,pa2); SWAPPA; | |
| 426 } else if (smallExp & 1) { | |
| 427 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 428 SQR(pa1,pa2); MUL(smallExp/2,pa2,pa1); | |
| 429 } else if (smallExp & 2) { | |
| 430 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 431 MUL(smallExp/4,pa1,pa2); SQR(pa2,pa1); | |
| 432 } else if (smallExp & 4) { | |
| 433 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); | |
| 434 MUL(smallExp/8,pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 435 } else if (smallExp & 8) { | |
| 436 SQR(pa1,pa2); SQR(pa2,pa1); MUL(smallExp/16,pa1,pa2); | |
| 437 SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 438 } else if (smallExp & 0x10) { | |
| 439 SQR(pa1,pa2); MUL(smallExp/32,pa2,pa1); SQR(pa1,pa2); | |
| 440 SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 441 } else { | |
| 442 abort(); | |
| 443 } | |
| 444 } else if (window_bits == 6) { | |
| 445 if (!smallExp) { | |
| 446 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 447 SQR(pa1,pa2); SQR(pa2,pa1); | |
| 448 } else if (smallExp & 1) { | |
| 449 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 450 SQR(pa1,pa2); SQR(pa2,pa1); MUL(smallExp/2,pa1,pa2); SWAPPA; | |
| 451 } else if (smallExp & 2) { | |
| 452 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 453 SQR(pa1,pa2); MUL(smallExp/4,pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 454 } else if (smallExp & 4) { | |
| 455 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 456 MUL(smallExp/8,pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 457 } else if (smallExp & 8) { | |
| 458 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); | |
| 459 MUL(smallExp/16,pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 460 SQR(pa1,pa2); SWAPPA; | |
| 461 } else if (smallExp & 0x10) { | |
| 462 SQR(pa1,pa2); SQR(pa2,pa1); MUL(smallExp/32,pa1,pa2); | |
| 463 SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 464 } else if (smallExp & 0x20) { | |
| 465 SQR(pa1,pa2); MUL(smallExp/64,pa2,pa1); SQR(pa1,pa2); | |
| 466 SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SWAPPA; | |
| 467 } else { | |
| 468 abort(); | |
| 469 } | |
| 470 } else { | |
| 471 abort(); | |
| 472 } | |
| 473 } | |
| 474 | |
| 475 res = s_mp_redc(pa1, mmm); | |
| 476 mp_exch(pa1, result); | |
| 477 | |
| 478 CLEANUP: | |
| 479 mp_clear(&accum1); | |
| 480 mp_clear(&accum2); | |
| 481 mp_clear(&power2); | |
| 482 for (i = 0; i < odd_ints; ++i) { | |
| 483 mp_clear(oddPowers + i); | |
| 484 } | |
| 485 return res; | |
| 486 } | |
| 487 #undef SQR | |
| 488 #undef MUL | |
| 489 | |
| 490 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 491 unsigned int mp_using_cache_safe_exp = 1; | |
| 492 #endif | |
| 493 | |
| 494 mp_err mp_set_safe_modexp(int value) | |
| 495 { | |
| 496 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 497 mp_using_cache_safe_exp = value; | |
| 498 return MP_OKAY; | |
| 499 #else | |
| 500 if (value == 0) { | |
| 501 return MP_OKAY; | |
| 502 } | |
| 503 return MP_BADARG; | |
| 504 #endif | |
| 505 } | |
| 506 | |
| 507 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 508 #define WEAVE_WORD_SIZE 4 | |
| 509 | |
| 510 #ifndef MP_CHAR_STORE_SLOW | |
| 511 /* | |
| 512 * mpi_to_weave takes an array of bignums, a matrix in which each bignum | |
| 513 * occupies all the columns of a row, and transposes it into a matrix in | |
| 514 * which each bignum occupies a column of every row. The first row of the | |
| 515 * input matrix becomes the first column of the output matrix. The n'th | |
| 516 * row of input becomes the n'th column of output. The input data is said | |
| 517 * to be "interleaved" or "woven" into the output matrix. | |
| 518 * | |
| 519 * The array of bignums is left in this woven form. Each time a single | |
| 520 * bignum value is needed, it is recreated by fetching the n'th column, | |
| 521 * forming a single row which is the new bignum. | |
| 522 * | |
| 523 * The purpose of this interleaving is make it impossible to determine which | |
| 524 * of the bignums is being used in any one operation by examining the pattern | |
| 525 * of cache misses. | |
| 526 * | |
| 527 * The weaving function does not transpose the entire input matrix in one call. | |
| 528 * It transposes 4 rows of mp_ints into their respective columns of output. | |
| 529 * | |
| 530 * There are two different implementations of the weaving and unweaving code | |
| 531 * in this file. One uses byte loads and stores. The second uses loads and | |
| 532 * stores of mp_weave_word size values. The weaved forms of these two | |
| 533 * implementations differ. Consequently, each one has its own explanation. | |
| 534 * | |
| 535 * Here is the explanation for the byte-at-a-time implementation. | |
| 536 * | |
| 537 * This implementation treats each mp_int bignum as an array of bytes, | |
| 538 * rather than as an array of mp_digits. It stores those bytes as a | |
| 539 * column of bytes in the output matrix. It doesn't care if the machine | |
| 540 * uses big-endian or little-endian byte ordering within mp_digits. | |
| 541 * The first byte of the mp_digit array becomes the first byte in the output | |
| 542 * column, regardless of whether that byte is the MSB or LSB of the mp_digit. | |
| 543 * | |
| 544 * "bignums" is an array of mp_ints. | |
| 545 * It points to four rows, four mp_ints, a subset of a larger array of mp_ints. | |
| 546 * | |
| 547 * "weaved" is the weaved output matrix. | |
| 548 * The first byte of bignums[0] is stored in weaved[0]. | |
| 549 * | |
| 550 * "nBignums" is the total number of bignums in the array of which "bignums" | |
| 551 * is a part. | |
| 552 * | |
| 553 * "nDigits" is the size in mp_digits of each mp_int in the "bignums" array. | |
| 554 * mp_ints that use less than nDigits digits are logically padded with zeros | |
| 555 * while being stored in the weaved array. | |
| 556 */ | |
| 557 mp_err mpi_to_weave(const mp_int *bignums, | |
| 558 unsigned char *weaved, | |
| 559 mp_size nDigits, /* in each mp_int of input */ | |
| 560 mp_size nBignums) /* in the entire source array */ | |
| 561 { | |
| 562 mp_size i; | |
| 563 unsigned char * endDest = weaved + (nDigits * nBignums * sizeof(mp_digit)); | |
| 564 | |
| 565 for (i=0; i < WEAVE_WORD_SIZE; i++) { | |
| 566 mp_size used = MP_USED(&bignums[i]); | |
| 567 unsigned char *pSrc = (unsigned char *)MP_DIGITS(&bignums[i]); | |
| 568 unsigned char *endSrc = pSrc + (used * sizeof(mp_digit)); | |
| 569 unsigned char *pDest = weaved + i; | |
| 570 | |
| 571 ARGCHK(MP_SIGN(&bignums[i]) == MP_ZPOS, MP_BADARG); | |
| 572 ARGCHK(used <= nDigits, MP_BADARG); | |
| 573 | |
| 574 for (; pSrc < endSrc; pSrc++) { | |
| 575 *pDest = *pSrc; | |
| 576 pDest += nBignums; | |
| 577 } | |
| 578 while (pDest < endDest) { | |
| 579 *pDest = 0; | |
| 580 pDest += nBignums; | |
| 581 } | |
| 582 } | |
| 583 | |
| 584 return MP_OKAY; | |
| 585 } | |
| 586 | |
| 587 /* Reverse the operation above for one mp_int. | |
| 588 * Reconstruct one mp_int from its column in the weaved array. | |
| 589 * "pSrc" points to the offset into the weave array of the bignum we | |
| 590 * are going to reconstruct. | |
| 591 */ | |
| 592 mp_err weave_to_mpi(mp_int *a, /* output, result */ | |
| 593 const unsigned char *pSrc, /* input, byte matrix */ | |
| 594 mp_size nDigits, /* per mp_int output */ | |
| 595 mp_size nBignums) /* bignums in weaved matrix */ | |
| 596 { | |
| 597 unsigned char *pDest = (unsigned char *)MP_DIGITS(a); | |
| 598 unsigned char *endDest = pDest + (nDigits * sizeof(mp_digit)); | |
| 599 | |
| 600 MP_SIGN(a) = MP_ZPOS; | |
| 601 MP_USED(a) = nDigits; | |
| 602 | |
| 603 for (; pDest < endDest; pSrc += nBignums, pDest++) { | |
| 604 *pDest = *pSrc; | |
| 605 } | |
| 606 s_mp_clamp(a); | |
| 607 return MP_OKAY; | |
| 608 } | |
| 609 | |
| 610 #else | |
| 611 | |
| 612 /* Need a primitive that we know is 32 bits long... */ | |
| 613 /* this is true on all modern processors we know of today*/ | |
| 614 typedef unsigned int mp_weave_word; | |
| 615 | |
| 616 /* | |
| 617 * on some platforms character stores into memory is very expensive since they | |
| 618 * generate a read/modify/write operation on the bus. On those platforms | |
| 619 * we need to do integer writes to the bus. Because of some unrolled code, | |
| 620 * in this current code the size of mp_weave_word must be four. The code that | |
| 621 * makes this assumption explicity is called out. (on some platforms a write | |
| 622 * of 4 bytes still requires a single read-modify-write operation. | |
| 623 * | |
| 624 * This function is takes the identical parameters as the function above, | |
| 625 * however it lays out the final array differently. Where the previous function | |
| 626 * treats the mpi_int as an byte array, this function treats it as an array of | |
| 627 * mp_digits where each digit is stored in big endian order. | |
| 628 * | |
| 629 * since we need to interleave on a byte by byte basis, we need to collect | |
| 630 * several mpi structures together into a single uint32 before we write. We | |
| 631 * also need to make sure the uint32 is arranged so that the first value of | |
| 632 * the first array winds up in b[0]. This means construction of that uint32 | |
| 633 * is endian specific (even though the layout of the mp_digits in the array | |
| 634 * is always big endian). | |
| 635 * | |
| 636 * The final data is stored as follows : | |
| 637 * | |
| 638 * Our same logical array p array, m is sizeof(mp_digit), | |
| 639 * N is still count and n is now b_size. If we define p[i].digit[j]0 as the | |
| 640 * most significant byte of the word p[i].digit[j], p[i].digit[j]1 as | |
| 641 * the next most significant byte of p[i].digit[j], ... and p[i].digit[j]m-1 | |
| 642 * is the least significant byte. | |
| 643 * Our array would look like: | |
| 644 * p[0].digit[0]0 p[1].digit[0]0 ... p[N-2].digit[0]0 p[N-1].digit[0]
0 | |
| 645 * p[0].digit[0]1 p[1].digit[0]1 ... p[N-2].digit[0]1 p[N-1].digit[0]
1 | |
| 646 * . . | |
| 647 * p[0].digit[0]m-1 p[1].digit[0]m-1 ... p[N-2].digit[0]m-1 p[N-1].digit[0]
m-1 | |
| 648 * p[0].digit[1]0 p[1].digit[1]0 ... p[N-2].digit[1]0 p[N-1].digit[1]
0 | |
| 649 * . . | |
| 650 * . . | |
| 651 * p[0].digit[n-1]m-2 p[1].digit[n-1]m-2 ... p[N-2].digit[n-1]m-2 p[N-1].digit[n
-1]m-2 | |
| 652 * p[0].digit[n-1]m-1 p[1].digit[n-1]m-1 ... p[N-2].digit[n-1]m-1 p[N-1].digit[n
-1]m-1 | |
| 653 * | |
| 654 */ | |
| 655 mp_err mpi_to_weave(const mp_int *a, unsigned char *b, | |
| 656 mp_size b_size, mp_size count) | |
| 657 { | |
| 658 mp_size i; | |
| 659 mp_digit *digitsa0; | |
| 660 mp_digit *digitsa1; | |
| 661 mp_digit *digitsa2; | |
| 662 mp_digit *digitsa3; | |
| 663 mp_size useda0; | |
| 664 mp_size useda1; | |
| 665 mp_size useda2; | |
| 666 mp_size useda3; | |
| 667 mp_weave_word *weaved = (mp_weave_word *)b; | |
| 668 | |
| 669 count = count/sizeof(mp_weave_word); | |
| 670 | |
| 671 /* this code pretty much depends on this ! */ | |
| 672 #if MP_ARGCHK == 2 | |
| 673 assert(WEAVE_WORD_SIZE == 4); | |
| 674 assert(sizeof(mp_weave_word) == 4); | |
| 675 #endif | |
| 676 | |
| 677 digitsa0 = MP_DIGITS(&a[0]); | |
| 678 digitsa1 = MP_DIGITS(&a[1]); | |
| 679 digitsa2 = MP_DIGITS(&a[2]); | |
| 680 digitsa3 = MP_DIGITS(&a[3]); | |
| 681 useda0 = MP_USED(&a[0]); | |
| 682 useda1 = MP_USED(&a[1]); | |
| 683 useda2 = MP_USED(&a[2]); | |
| 684 useda3 = MP_USED(&a[3]); | |
| 685 | |
| 686 ARGCHK(MP_SIGN(&a[0]) == MP_ZPOS, MP_BADARG); | |
| 687 ARGCHK(MP_SIGN(&a[1]) == MP_ZPOS, MP_BADARG); | |
| 688 ARGCHK(MP_SIGN(&a[2]) == MP_ZPOS, MP_BADARG); | |
| 689 ARGCHK(MP_SIGN(&a[3]) == MP_ZPOS, MP_BADARG); | |
| 690 ARGCHK(useda0 <= b_size, MP_BADARG); | |
| 691 ARGCHK(useda1 <= b_size, MP_BADARG); | |
| 692 ARGCHK(useda2 <= b_size, MP_BADARG); | |
| 693 ARGCHK(useda3 <= b_size, MP_BADARG); | |
| 694 | |
| 695 #define SAFE_FETCH(digit, used, word) ((word) < (used) ? (digit[word]) : 0) | |
| 696 | |
| 697 for (i=0; i < b_size; i++) { | |
| 698 mp_digit d0 = SAFE_FETCH(digitsa0,useda0,i); | |
| 699 mp_digit d1 = SAFE_FETCH(digitsa1,useda1,i); | |
| 700 mp_digit d2 = SAFE_FETCH(digitsa2,useda2,i); | |
| 701 mp_digit d3 = SAFE_FETCH(digitsa3,useda3,i); | |
| 702 register mp_weave_word acc; | |
| 703 | |
| 704 /* | |
| 705 * ONE_STEP takes the MSB of each of our current digits and places that | |
| 706 * byte in the appropriate position for writing to the weaved array. | |
| 707 * On little endian: | |
| 708 * b3 b2 b1 b0 | |
| 709 * On big endian: | |
| 710 * b0 b1 b2 b3 | |
| 711 * When the data is written it would always wind up: | |
| 712 * b[0] = b0 | |
| 713 * b[1] = b1 | |
| 714 * b[2] = b2 | |
| 715 * b[3] = b3 | |
| 716 * | |
| 717 * Once we've written the MSB, we shift the whole digit up left one | |
| 718 * byte, putting the Next Most Significant Byte in the MSB position, | |
| 719 * so we we repeat the next one step that byte will be written. | |
| 720 * NOTE: This code assumes sizeof(mp_weave_word) and MP_WEAVE_WORD_SIZE | |
| 721 * is 4. | |
| 722 */ | |
| 723 #ifdef MP_IS_LITTLE_ENDIAN | |
| 724 #define MPI_WEAVE_ONE_STEP \ | |
| 725 acc = (d0 >> (MP_DIGIT_BIT-8)) & 0x000000ff; d0 <<= 8; /*b0*/ \ | |
| 726 acc |= (d1 >> (MP_DIGIT_BIT-16)) & 0x0000ff00; d1 <<= 8; /*b1*/ \ | |
| 727 acc |= (d2 >> (MP_DIGIT_BIT-24)) & 0x00ff0000; d2 <<= 8; /*b2*/ \ | |
| 728 acc |= (d3 >> (MP_DIGIT_BIT-32)) & 0xff000000; d3 <<= 8; /*b3*/ \ | |
| 729 *weaved = acc; weaved += count; | |
| 730 #else | |
| 731 #define MPI_WEAVE_ONE_STEP \ | |
| 732 acc = (d0 >> (MP_DIGIT_BIT-32)) & 0xff000000; d0 <<= 8; /*b0*/ \ | |
| 733 acc |= (d1 >> (MP_DIGIT_BIT-24)) & 0x00ff0000; d1 <<= 8; /*b1*/ \ | |
| 734 acc |= (d2 >> (MP_DIGIT_BIT-16)) & 0x0000ff00; d2 <<= 8; /*b2*/ \ | |
| 735 acc |= (d3 >> (MP_DIGIT_BIT-8)) & 0x000000ff; d3 <<= 8; /*b3*/ \ | |
| 736 *weaved = acc; weaved += count; | |
| 737 #endif | |
| 738 switch (sizeof(mp_digit)) { | |
| 739 case 32: | |
| 740 MPI_WEAVE_ONE_STEP | |
| 741 MPI_WEAVE_ONE_STEP | |
| 742 MPI_WEAVE_ONE_STEP | |
| 743 MPI_WEAVE_ONE_STEP | |
| 744 MPI_WEAVE_ONE_STEP | |
| 745 MPI_WEAVE_ONE_STEP | |
| 746 MPI_WEAVE_ONE_STEP | |
| 747 MPI_WEAVE_ONE_STEP | |
| 748 MPI_WEAVE_ONE_STEP | |
| 749 MPI_WEAVE_ONE_STEP | |
| 750 MPI_WEAVE_ONE_STEP | |
| 751 MPI_WEAVE_ONE_STEP | |
| 752 MPI_WEAVE_ONE_STEP | |
| 753 MPI_WEAVE_ONE_STEP | |
| 754 MPI_WEAVE_ONE_STEP | |
| 755 MPI_WEAVE_ONE_STEP | |
| 756 case 16: | |
| 757 MPI_WEAVE_ONE_STEP | |
| 758 MPI_WEAVE_ONE_STEP | |
| 759 MPI_WEAVE_ONE_STEP | |
| 760 MPI_WEAVE_ONE_STEP | |
| 761 MPI_WEAVE_ONE_STEP | |
| 762 MPI_WEAVE_ONE_STEP | |
| 763 MPI_WEAVE_ONE_STEP | |
| 764 MPI_WEAVE_ONE_STEP | |
| 765 case 8: | |
| 766 MPI_WEAVE_ONE_STEP | |
| 767 MPI_WEAVE_ONE_STEP | |
| 768 MPI_WEAVE_ONE_STEP | |
| 769 MPI_WEAVE_ONE_STEP | |
| 770 case 4: | |
| 771 MPI_WEAVE_ONE_STEP | |
| 772 MPI_WEAVE_ONE_STEP | |
| 773 case 2: | |
| 774 MPI_WEAVE_ONE_STEP | |
| 775 case 1: | |
| 776 MPI_WEAVE_ONE_STEP | |
| 777 break; | |
| 778 } | |
| 779 } | |
| 780 | |
| 781 return MP_OKAY; | |
| 782 } | |
| 783 | |
| 784 /* reverse the operation above for one entry. | |
| 785 * b points to the offset into the weave array of the power we are | |
| 786 * calculating */ | |
| 787 mp_err weave_to_mpi(mp_int *a, const unsigned char *b, | |
| 788 mp_size b_size, mp_size count) | |
| 789 { | |
| 790 mp_digit *pb = MP_DIGITS(a); | |
| 791 mp_digit *end = &pb[b_size]; | |
| 792 | |
| 793 MP_SIGN(a) = MP_ZPOS; | |
| 794 MP_USED(a) = b_size; | |
| 795 | |
| 796 for (; pb < end; pb++) { | |
| 797 register mp_digit digit; | |
| 798 | |
| 799 digit = *b << 8; b += count; | |
| 800 #define MPI_UNWEAVE_ONE_STEP digit |= *b; b += count; digit = digit << 8; | |
| 801 switch (sizeof(mp_digit)) { | |
| 802 case 32: | |
| 803 MPI_UNWEAVE_ONE_STEP | |
| 804 MPI_UNWEAVE_ONE_STEP | |
| 805 MPI_UNWEAVE_ONE_STEP | |
| 806 MPI_UNWEAVE_ONE_STEP | |
| 807 MPI_UNWEAVE_ONE_STEP | |
| 808 MPI_UNWEAVE_ONE_STEP | |
| 809 MPI_UNWEAVE_ONE_STEP | |
| 810 MPI_UNWEAVE_ONE_STEP | |
| 811 MPI_UNWEAVE_ONE_STEP | |
| 812 MPI_UNWEAVE_ONE_STEP | |
| 813 MPI_UNWEAVE_ONE_STEP | |
| 814 MPI_UNWEAVE_ONE_STEP | |
| 815 MPI_UNWEAVE_ONE_STEP | |
| 816 MPI_UNWEAVE_ONE_STEP | |
| 817 MPI_UNWEAVE_ONE_STEP | |
| 818 MPI_UNWEAVE_ONE_STEP | |
| 819 case 16: | |
| 820 MPI_UNWEAVE_ONE_STEP | |
| 821 MPI_UNWEAVE_ONE_STEP | |
| 822 MPI_UNWEAVE_ONE_STEP | |
| 823 MPI_UNWEAVE_ONE_STEP | |
| 824 MPI_UNWEAVE_ONE_STEP | |
| 825 MPI_UNWEAVE_ONE_STEP | |
| 826 MPI_UNWEAVE_ONE_STEP | |
| 827 MPI_UNWEAVE_ONE_STEP | |
| 828 case 8: | |
| 829 MPI_UNWEAVE_ONE_STEP | |
| 830 MPI_UNWEAVE_ONE_STEP | |
| 831 MPI_UNWEAVE_ONE_STEP | |
| 832 MPI_UNWEAVE_ONE_STEP | |
| 833 case 4: | |
| 834 MPI_UNWEAVE_ONE_STEP | |
| 835 MPI_UNWEAVE_ONE_STEP | |
| 836 case 2: | |
| 837 break; | |
| 838 } | |
| 839 digit |= *b; b += count; | |
| 840 | |
| 841 *pb = digit; | |
| 842 } | |
| 843 s_mp_clamp(a); | |
| 844 return MP_OKAY; | |
| 845 } | |
| 846 #endif | |
| 847 | |
| 848 | |
| 849 #define SQR(a,b) \ | |
| 850 MP_CHECKOK( mp_sqr(a, b) );\ | |
| 851 MP_CHECKOK( s_mp_redc(b, mmm) ) | |
| 852 | |
| 853 #if defined(MP_MONT_USE_MP_MUL) | |
| 854 #define MUL_NOWEAVE(x,a,b) \ | |
| 855 MP_CHECKOK( mp_mul(a, x, b) ); \ | |
| 856 MP_CHECKOK( s_mp_redc(b, mmm) ) | |
| 857 #else | |
| 858 #define MUL_NOWEAVE(x,a,b) \ | |
| 859 MP_CHECKOK( s_mp_mul_mont(a, x, b, mmm) ) | |
| 860 #endif | |
| 861 | |
| 862 #define MUL(x,a,b) \ | |
| 863 MP_CHECKOK( weave_to_mpi(&tmp, powers + (x), nLen, num_powers) ); \ | |
| 864 MUL_NOWEAVE(&tmp,a,b) | |
| 865 | |
| 866 #define SWAPPA ptmp = pa1; pa1 = pa2; pa2 = ptmp | |
| 867 #define MP_ALIGN(x,y) ((((ptrdiff_t)(x))+((y)-1))&(((ptrdiff_t)0)-(y))) | |
| 868 | |
| 869 /* Do modular exponentiation using integer multiply code. */ | |
| 870 mp_err mp_exptmod_safe_i(const mp_int * montBase, | |
| 871 const mp_int * exponent, | |
| 872 const mp_int * modulus, | |
| 873 mp_int * result, | |
| 874 mp_mont_modulus *mmm, | |
| 875 int nLen, | |
| 876 mp_size bits_in_exponent, | |
| 877 mp_size window_bits, | |
| 878 mp_size num_powers) | |
| 879 { | |
| 880 mp_int *pa1, *pa2, *ptmp; | |
| 881 mp_size i; | |
| 882 mp_size first_window; | |
| 883 mp_err res; | |
| 884 int expOff; | |
| 885 mp_int accum1, accum2, accum[WEAVE_WORD_SIZE]; | |
| 886 mp_int tmp; | |
| 887 unsigned char *powersArray; | |
| 888 unsigned char *powers; | |
| 889 | |
| 890 MP_DIGITS(&accum1) = 0; | |
| 891 MP_DIGITS(&accum2) = 0; | |
| 892 MP_DIGITS(&accum[0]) = 0; | |
| 893 MP_DIGITS(&accum[1]) = 0; | |
| 894 MP_DIGITS(&accum[2]) = 0; | |
| 895 MP_DIGITS(&accum[3]) = 0; | |
| 896 MP_DIGITS(&tmp) = 0; | |
| 897 | |
| 898 powersArray = (unsigned char *)malloc(num_powers*(nLen*sizeof(mp_digit)+1)); | |
| 899 if (powersArray == NULL) { | |
| 900 res = MP_MEM; | |
| 901 goto CLEANUP; | |
| 902 } | |
| 903 | |
| 904 /* powers[i] = base ** (i); */ | |
| 905 powers = (unsigned char *)MP_ALIGN(powersArray,num_powers); | |
| 906 | |
| 907 /* grab the first window value. This allows us to preload accumulator1 | |
| 908 * and save a conversion, some squares and a multiple*/ | |
| 909 MP_CHECKOK( mpl_get_bits(exponent, | |
| 910 bits_in_exponent-window_bits, window_bits) ); | |
| 911 first_window = (mp_size)res; | |
| 912 | |
| 913 MP_CHECKOK( mp_init_size(&accum1, 3 * nLen + 2) ); | |
| 914 MP_CHECKOK( mp_init_size(&accum2, 3 * nLen + 2) ); | |
| 915 MP_CHECKOK( mp_init_size(&tmp, 3 * nLen + 2) ); | |
| 916 | |
| 917 /* build the first WEAVE_WORD powers inline */ | |
| 918 /* if WEAVE_WORD_SIZE is not 4, this code will have to change */ | |
| 919 if (num_powers > 2) { | |
| 920 MP_CHECKOK( mp_init_size(&accum[0], 3 * nLen + 2) ); | |
| 921 MP_CHECKOK( mp_init_size(&accum[1], 3 * nLen + 2) ); | |
| 922 MP_CHECKOK( mp_init_size(&accum[2], 3 * nLen + 2) ); | |
| 923 MP_CHECKOK( mp_init_size(&accum[3], 3 * nLen + 2) ); | |
| 924 mp_set(&accum[0], 1); | |
| 925 MP_CHECKOK( s_mp_to_mont(&accum[0], mmm, &accum[0]) ); | |
| 926 MP_CHECKOK( mp_copy(montBase, &accum[1]) ); | |
| 927 SQR(montBase, &accum[2]); | |
| 928 MUL_NOWEAVE(montBase, &accum[2], &accum[3]); | |
| 929 MP_CHECKOK( mpi_to_weave(accum, powers, nLen, num_powers) ); | |
| 930 if (first_window < 4) { | |
| 931 MP_CHECKOK( mp_copy(&accum[first_window], &accum1) ); | |
| 932 first_window = num_powers; | |
| 933 } | |
| 934 } else { | |
| 935 if (first_window == 0) { | |
| 936 mp_set(&accum1, 1); | |
| 937 MP_CHECKOK( s_mp_to_mont(&accum1, mmm, &accum1) ); | |
| 938 } else { | |
| 939 /* assert first_window == 1? */ | |
| 940 MP_CHECKOK( mp_copy(montBase, &accum1) ); | |
| 941 } | |
| 942 } | |
| 943 | |
| 944 /* | |
| 945 * calculate all the powers in the powers array. | |
| 946 * this adds 2**(k-1)-2 square operations over just calculating the | |
| 947 * odd powers where k is the window size in the two other mp_modexpt | |
| 948 * implementations in this file. We will get some of that | |
| 949 * back by not needing the first 'k' squares and one multiply for the | |
| 950 * first window */ | |
| 951 for (i = WEAVE_WORD_SIZE; i < num_powers; i++) { | |
| 952 int acc_index = i & (WEAVE_WORD_SIZE-1); /* i % WEAVE_WORD_SIZE */ | |
| 953 if ( i & 1 ) { | |
| 954 MUL_NOWEAVE(montBase, &accum[acc_index-1] , &accum[acc_index]); | |
| 955 /* we've filled the array do our 'per array' processing */ | |
| 956 if (acc_index == (WEAVE_WORD_SIZE-1)) { | |
| 957 MP_CHECKOK( mpi_to_weave(accum, powers + i - (WEAVE_WORD_SIZE-1), | |
| 958 nLen, num_powers) ); | |
| 959 | |
| 960 if (first_window <= i) { | |
| 961 MP_CHECKOK( mp_copy(&accum[first_window & (WEAVE_WORD_SIZE-1)], | |
| 962 &accum1) ); | |
| 963 first_window = num_powers; | |
| 964 } | |
| 965 } | |
| 966 } else { | |
| 967 /* up to 8 we can find 2^i-1 in the accum array, but at 8 we our source | |
| 968 * and target are the same so we need to copy.. After that, the | |
| 969 * value is overwritten, so we need to fetch it from the stored | |
| 970 * weave array */ | |
| 971 if (i > 2* WEAVE_WORD_SIZE) { | |
| 972 MP_CHECKOK(weave_to_mpi(&accum2, powers+i/2, nLen, num_powers)); | |
| 973 SQR(&accum2, &accum[acc_index]); | |
| 974 } else { | |
| 975 int half_power_index = (i/2) & (WEAVE_WORD_SIZE-1); | |
| 976 if (half_power_index == acc_index) { | |
| 977 /* copy is cheaper than weave_to_mpi */ | |
| 978 MP_CHECKOK(mp_copy(&accum[half_power_index], &accum2)); | |
| 979 SQR(&accum2,&accum[acc_index]); | |
| 980 } else { | |
| 981 SQR(&accum[half_power_index],&accum[acc_index]); | |
| 982 } | |
| 983 } | |
| 984 } | |
| 985 } | |
| 986 /* if the accum1 isn't set, Then there is something wrong with our logic | |
| 987 * above and is an internal programming error. | |
| 988 */ | |
| 989 #if MP_ARGCHK == 2 | |
| 990 assert(MP_USED(&accum1) != 0); | |
| 991 #endif | |
| 992 | |
| 993 /* set accumulator to montgomery residue of 1 */ | |
| 994 pa1 = &accum1; | |
| 995 pa2 = &accum2; | |
| 996 | |
| 997 for (expOff = bits_in_exponent - window_bits*2; expOff >= 0; expOff -= window_
bits) { | |
| 998 mp_size smallExp; | |
| 999 MP_CHECKOK( mpl_get_bits(exponent, expOff, window_bits) ); | |
| 1000 smallExp = (mp_size)res; | |
| 1001 | |
| 1002 /* handle unroll the loops */ | |
| 1003 switch (window_bits) { | |
| 1004 case 1: | |
| 1005 if (!smallExp) { | |
| 1006 SQR(pa1,pa2); SWAPPA; | |
| 1007 } else if (smallExp & 1) { | |
| 1008 SQR(pa1,pa2); MUL_NOWEAVE(montBase,pa2,pa1); | |
| 1009 } else { | |
| 1010 abort(); | |
| 1011 } | |
| 1012 break; | |
| 1013 case 6: | |
| 1014 SQR(pa1,pa2); SQR(pa2,pa1); | |
| 1015 /* fall through */ | |
| 1016 case 4: | |
| 1017 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 1018 MUL(smallExp, pa1,pa2); SWAPPA; | |
| 1019 break; | |
| 1020 case 5: | |
| 1021 SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1); | |
| 1022 SQR(pa1,pa2); MUL(smallExp,pa2,pa1); | |
| 1023 break; | |
| 1024 default: | |
| 1025 abort(); /* could do a loop? */ | |
| 1026 } | |
| 1027 } | |
| 1028 | |
| 1029 res = s_mp_redc(pa1, mmm); | |
| 1030 mp_exch(pa1, result); | |
| 1031 | |
| 1032 CLEANUP: | |
| 1033 mp_clear(&accum1); | |
| 1034 mp_clear(&accum2); | |
| 1035 mp_clear(&accum[0]); | |
| 1036 mp_clear(&accum[1]); | |
| 1037 mp_clear(&accum[2]); | |
| 1038 mp_clear(&accum[3]); | |
| 1039 mp_clear(&tmp); | |
| 1040 /* PORT_Memset(powers,0,num_powers*nLen*sizeof(mp_digit)); */ | |
| 1041 free(powersArray); | |
| 1042 return res; | |
| 1043 } | |
| 1044 #undef SQR | |
| 1045 #undef MUL | |
| 1046 #endif | |
| 1047 | |
| 1048 mp_err mp_exptmod(const mp_int *inBase, const mp_int *exponent, | |
| 1049 const mp_int *modulus, mp_int *result) | |
| 1050 { | |
| 1051 const mp_int *base; | |
| 1052 mp_size bits_in_exponent, i, window_bits, odd_ints; | |
| 1053 mp_err res; | |
| 1054 int nLen; | |
| 1055 mp_int montBase, goodBase; | |
| 1056 mp_mont_modulus mmm; | |
| 1057 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 1058 static unsigned int max_window_bits; | |
| 1059 #endif | |
| 1060 | |
| 1061 /* function for computing n0prime only works if n0 is odd */ | |
| 1062 if (!mp_isodd(modulus)) | |
| 1063 return s_mp_exptmod(inBase, exponent, modulus, result); | |
| 1064 | |
| 1065 MP_DIGITS(&montBase) = 0; | |
| 1066 MP_DIGITS(&goodBase) = 0; | |
| 1067 | |
| 1068 if (mp_cmp(inBase, modulus) < 0) { | |
| 1069 base = inBase; | |
| 1070 } else { | |
| 1071 MP_CHECKOK( mp_init(&goodBase) ); | |
| 1072 base = &goodBase; | |
| 1073 MP_CHECKOK( mp_mod(inBase, modulus, &goodBase) ); | |
| 1074 } | |
| 1075 | |
| 1076 nLen = MP_USED(modulus); | |
| 1077 MP_CHECKOK( mp_init_size(&montBase, 2 * nLen + 2) ); | |
| 1078 | |
| 1079 mmm.N = *modulus; /* a copy of the mp_int struct */ | |
| 1080 | |
| 1081 /* compute n0', given n0, n0' = -(n0 ** -1) mod MP_RADIX | |
| 1082 ** where n0 = least significant mp_digit of N, the modulus. | |
| 1083 */ | |
| 1084 mmm.n0prime = 0 - s_mp_invmod_radix( MP_DIGIT(modulus, 0) ); | |
| 1085 | |
| 1086 MP_CHECKOK( s_mp_to_mont(base, &mmm, &montBase) ); | |
| 1087 | |
| 1088 bits_in_exponent = mpl_significant_bits(exponent); | |
| 1089 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 1090 if (mp_using_cache_safe_exp) { | |
| 1091 if (bits_in_exponent > 780) | |
| 1092 window_bits = 6; | |
| 1093 else if (bits_in_exponent > 256) | |
| 1094 window_bits = 5; | |
| 1095 else if (bits_in_exponent > 20) | |
| 1096 window_bits = 4; | |
| 1097 /* RSA public key exponents are typically under 20 bits (common values | |
| 1098 * are: 3, 17, 65537) and a 4-bit window is inefficient | |
| 1099 */ | |
| 1100 else | |
| 1101 window_bits = 1; | |
| 1102 } else | |
| 1103 #endif | |
| 1104 if (bits_in_exponent > 480) | |
| 1105 window_bits = 6; | |
| 1106 else if (bits_in_exponent > 160) | |
| 1107 window_bits = 5; | |
| 1108 else if (bits_in_exponent > 20) | |
| 1109 window_bits = 4; | |
| 1110 /* RSA public key exponents are typically under 20 bits (common values | |
| 1111 * are: 3, 17, 65537) and a 4-bit window is inefficient | |
| 1112 */ | |
| 1113 else | |
| 1114 window_bits = 1; | |
| 1115 | |
| 1116 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 1117 /* | |
| 1118 * clamp the window size based on | |
| 1119 * the cache line size. | |
| 1120 */ | |
| 1121 if (!max_window_bits) { | |
| 1122 unsigned long cache_size = s_mpi_getProcessorLineSize(); | |
| 1123 /* processor has no cache, use 'fast' code always */ | |
| 1124 if (cache_size == 0) { | |
| 1125 mp_using_cache_safe_exp = 0; | |
| 1126 } | |
| 1127 if ((cache_size == 0) || (cache_size >= 64)) { | |
| 1128 max_window_bits = 6; | |
| 1129 } else if (cache_size >= 32) { | |
| 1130 max_window_bits = 5; | |
| 1131 } else if (cache_size >= 16) { | |
| 1132 max_window_bits = 4; | |
| 1133 } else max_window_bits = 1; /* should this be an assert? */ | |
| 1134 } | |
| 1135 | |
| 1136 /* clamp the window size down before we caclulate bits_in_exponent */ | |
| 1137 if (mp_using_cache_safe_exp) { | |
| 1138 if (window_bits > max_window_bits) { | |
| 1139 window_bits = max_window_bits; | |
| 1140 } | |
| 1141 } | |
| 1142 #endif | |
| 1143 | |
| 1144 odd_ints = 1 << (window_bits - 1); | |
| 1145 i = bits_in_exponent % window_bits; | |
| 1146 if (i != 0) { | |
| 1147 bits_in_exponent += window_bits - i; | |
| 1148 } | |
| 1149 | |
| 1150 #ifdef MP_USING_MONT_MULF | |
| 1151 if (mp_using_mont_mulf) { | |
| 1152 MP_CHECKOK( s_mp_pad(&montBase, nLen) ); | |
| 1153 res = mp_exptmod_f(&montBase, exponent, modulus, result, &mmm, nLen, | |
| 1154 bits_in_exponent, window_bits, odd_ints); | |
| 1155 } else | |
| 1156 #endif | |
| 1157 #ifdef MP_USING_CACHE_SAFE_MOD_EXP | |
| 1158 if (mp_using_cache_safe_exp) { | |
| 1159 res = mp_exptmod_safe_i(&montBase, exponent, modulus, result, &mmm, nLen, | |
| 1160 bits_in_exponent, window_bits, 1 << window_bits); | |
| 1161 } else | |
| 1162 #endif | |
| 1163 res = mp_exptmod_i(&montBase, exponent, modulus, result, &mmm, nLen, | |
| 1164 bits_in_exponent, window_bits, odd_ints); | |
| 1165 | |
| 1166 CLEANUP: | |
| 1167 mp_clear(&montBase); | |
| 1168 mp_clear(&goodBase); | |
| 1169 /* Don't mp_clear mmm.N because it is merely a copy of modulus. | |
| 1170 ** Just zap it. | |
| 1171 */ | |
| 1172 memset(&mmm, 0, sizeof mmm); | |
| 1173 return res; | |
| 1174 } | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 14249009:
Change the NSS and NSPR source tree to the new directory structure to be (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/nss/