| 1 |
/* crypto/bn/bn_exp.c */ |
| 2 |
/* Copyright (C) 1995-1998 Eric Young ([email protected]) |
| 3 |
* All rights reserved. |
| 4 |
* |
| 5 |
* This package is an SSL implementation written |
| 6 |
* by Eric Young ([email protected]). |
| 7 |
* The implementation was written so as to conform with Netscapes SSL. |
| 8 |
* |
| 9 |
* This library is free for commercial and non-commercial use as long as |
| 10 |
* the following conditions are aheared to. The following conditions |
| 11 |
* apply to all code found in this distribution, be it the RC4, RSA, |
| 12 |
* lhash, DES, etc., code; not just the SSL code. The SSL documentation |
| 13 |
* included with this distribution is covered by the same copyright terms |
| 14 |
* except that the holder is Tim Hudson ([email protected]). |
| 15 |
* |
| 16 |
* Copyright remains Eric Young's, and as such any Copyright notices in |
| 17 |
* the code are not to be removed. |
| 18 |
* If this package is used in a product, Eric Young should be given attribution |
| 19 |
* as the author of the parts of the library used. |
| 20 |
* This can be in the form of a textual message at program startup or |
| 21 |
* in documentation (online or textual) provided with the package. |
| 22 |
* |
| 23 |
* Redistribution and use in source and binary forms, with or without |
| 24 |
* modification, are permitted provided that the following conditions |
| 25 |
* are met: |
| 26 |
* 1. Redistributions of source code must retain the copyright |
| 27 |
* notice, this list of conditions and the following disclaimer. |
| 28 |
* 2. Redistributions in binary form must reproduce the above copyright |
| 29 |
* notice, this list of conditions and the following disclaimer in the |
| 30 |
* documentation and/or other materials provided with the distribution. |
| 31 |
* 3. All advertising materials mentioning features or use of this software |
| 32 |
* must display the following acknowledgement: |
| 33 |
* "This product includes cryptographic software written by |
| 34 |
* Eric Young ([email protected])" |
| 35 |
* The word 'cryptographic' can be left out if the rouines from the library |
| 36 |
* being used are not cryptographic related :-). |
| 37 |
* 4. If you include any Windows specific code (or a derivative thereof) from |
| 38 |
* the apps directory (application code) you must include an acknowledgement: |
| 39 |
* "This product includes software written by Tim Hudson ([email protected])" |
| 40 |
* |
| 41 |
* THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND |
| 42 |
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 43 |
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 44 |
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
| 45 |
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 46 |
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 47 |
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 48 |
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 49 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 50 |
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 51 |
* SUCH DAMAGE. |
| 52 |
* |
| 53 |
* The licence and distribution terms for any publically available version or |
| 54 |
* derivative of this code cannot be changed. i.e. this code cannot simply be |
| 55 |
* copied and put under another distribution licence |
| 56 |
* [including the GNU Public Licence.] |
| 57 |
*/ |
| 58 |
/* ==================================================================== |
| 59 |
* Copyright (c) 1998-2005 The OpenSSL Project. All rights reserved. |
| 60 |
* |
| 61 |
* Redistribution and use in source and binary forms, with or without |
| 62 |
* modification, are permitted provided that the following conditions |
| 63 |
* are met: |
| 64 |
* |
| 65 |
* 1. Redistributions of source code must retain the above copyright |
| 66 |
* notice, this list of conditions and the following disclaimer. |
| 67 |
* |
| 68 |
* 2. Redistributions in binary form must reproduce the above copyright |
| 69 |
* notice, this list of conditions and the following disclaimer in |
| 70 |
* the documentation and/or other materials provided with the |
| 71 |
* distribution. |
| 72 |
* |
| 73 |
* 3. All advertising materials mentioning features or use of this |
| 74 |
* software must display the following acknowledgment: |
| 75 |
* "This product includes software developed by the OpenSSL Project |
| 76 |
* for use in the OpenSSL Toolkit. (http://www.openssl.org/)" |
| 77 |
* |
| 78 |
* 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to |
| 79 |
* endorse or promote products derived from this software without |
| 80 |
* prior written permission. For written permission, please contact |
| 81 |
* [email protected]. |
| 82 |
* |
| 83 |
* 5. Products derived from this software may not be called "OpenSSL" |
| 84 |
* nor may "OpenSSL" appear in their names without prior written |
| 85 |
* permission of the OpenSSL Project. |
| 86 |
* |
| 87 |
* 6. Redistributions of any form whatsoever must retain the following |
| 88 |
* acknowledgment: |
| 89 |
* "This product includes software developed by the OpenSSL Project |
| 90 |
* for use in the OpenSSL Toolkit (http://www.openssl.org/)" |
| 91 |
* |
| 92 |
* THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY |
| 93 |
* EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 94 |
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 95 |
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR |
| 96 |
* ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 97 |
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
| 98 |
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 99 |
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 100 |
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, |
| 101 |
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 102 |
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
| 103 |
* OF THE POSSIBILITY OF SUCH DAMAGE. |
| 104 |
* ==================================================================== |
| 105 |
* |
| 106 |
* This product includes cryptographic software written by Eric Young |
| 107 |
* ([email protected]). This product includes software written by Tim |
| 108 |
* Hudson ([email protected]). |
| 109 |
* |
| 110 |
*/ |
| 111 |
|
| 112 |
#include "cryptlib.h" |
| 113 |
#include "constant_time_locl.h" |
| 114 |
#include "bn_lcl.h" |
| 115 |
|
| 116 |
#include <stdlib.h> |
| 117 |
#ifdef _WIN32 |
| 118 |
# include <malloc.h> |
| 119 |
# ifndef alloca |
| 120 |
# define alloca _alloca |
| 121 |
# endif |
| 122 |
#elif defined(__GNUC__) |
| 123 |
# ifndef alloca |
| 124 |
# define alloca(s) __builtin_alloca((s)) |
| 125 |
# endif |
| 126 |
#elif defined(__sun) |
| 127 |
# include <alloca.h> |
| 128 |
#endif |
| 129 |
|
| 130 |
#include "rsaz_exp.h" |
| 131 |
|
| 132 |
#undef SPARC_T4_MONT |
| 133 |
#if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc)) |
| 134 |
# include "sparc_arch.h" |
| 135 |
extern unsigned int OPENSSL_sparcv9cap_P[]; |
| 136 |
# define SPARC_T4_MONT |
| 137 |
#endif |
| 138 |
|
| 139 |
/* maximum precomputation table size for *variable* sliding windows */ |
| 140 |
#define TABLE_SIZE 32 |
| 141 |
|
| 142 |
/* this one works - simple but works */ |
| 143 |
int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) |
| 144 |
{ |
| 145 |
int i, bits, ret = 0; |
| 146 |
BIGNUM *v, *rr; |
| 147 |
|
| 148 |
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) { |
| 149 |
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
| 150 |
BNerr(BN_F_BN_EXP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
| 151 |
return -1; |
| 152 |
} |
| 153 |
|
| 154 |
BN_CTX_start(ctx); |
| 155 |
if ((r == a) || (r == p)) |
| 156 |
rr = BN_CTX_get(ctx); |
| 157 |
else |
| 158 |
rr = r; |
| 159 |
v = BN_CTX_get(ctx); |
| 160 |
if (rr == NULL || v == NULL) |
| 161 |
goto err; |
| 162 |
|
| 163 |
if (BN_copy(v, a) == NULL) |
| 164 |
goto err; |
| 165 |
bits = BN_num_bits(p); |
| 166 |
|
| 167 |
if (BN_is_odd(p)) { |
| 168 |
if (BN_copy(rr, a) == NULL) |
| 169 |
goto err; |
| 170 |
} else { |
| 171 |
if (!BN_one(rr)) |
| 172 |
goto err; |
| 173 |
} |
| 174 |
|
| 175 |
for (i = 1; i < bits; i++) { |
| 176 |
if (!BN_sqr(v, v, ctx)) |
| 177 |
goto err; |
| 178 |
if (BN_is_bit_set(p, i)) { |
| 179 |
if (!BN_mul(rr, rr, v, ctx)) |
| 180 |
goto err; |
| 181 |
} |
| 182 |
} |
| 183 |
if (r != rr && BN_copy(r, rr) == NULL) |
| 184 |
goto err; |
| 185 |
|
| 186 |
ret = 1; |
| 187 |
err: |
| 188 |
BN_CTX_end(ctx); |
| 189 |
bn_check_top(r); |
| 190 |
return (ret); |
| 191 |
} |
| 192 |
|
| 193 |
int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m, |
| 194 |
BN_CTX *ctx) |
| 195 |
{ |
| 196 |
int ret; |
| 197 |
|
| 198 |
bn_check_top(a); |
| 199 |
bn_check_top(p); |
| 200 |
bn_check_top(m); |
| 201 |
|
| 202 |
/*- |
| 203 |
* For even modulus m = 2^k*m_odd, it might make sense to compute |
| 204 |
* a^p mod m_odd and a^p mod 2^k separately (with Montgomery |
| 205 |
* exponentiation for the odd part), using appropriate exponent |
| 206 |
* reductions, and combine the results using the CRT. |
| 207 |
* |
| 208 |
* For now, we use Montgomery only if the modulus is odd; otherwise, |
| 209 |
* exponentiation using the reciprocal-based quick remaindering |
| 210 |
* algorithm is used. |
| 211 |
* |
| 212 |
* (Timing obtained with expspeed.c [computations a^p mod m |
| 213 |
* where a, p, m are of the same length: 256, 512, 1024, 2048, |
| 214 |
* 4096, 8192 bits], compared to the running time of the |
| 215 |
* standard algorithm: |
| 216 |
* |
| 217 |
* BN_mod_exp_mont 33 .. 40 % [AMD K6-2, Linux, debug configuration] |
| 218 |
* 55 .. 77 % [UltraSparc processor, but |
| 219 |
* debug-solaris-sparcv8-gcc conf.] |
| 220 |
* |
| 221 |
* BN_mod_exp_recp 50 .. 70 % [AMD K6-2, Linux, debug configuration] |
| 222 |
* 62 .. 118 % [UltraSparc, debug-solaris-sparcv8-gcc] |
| 223 |
* |
| 224 |
* On the Sparc, BN_mod_exp_recp was faster than BN_mod_exp_mont |
| 225 |
* at 2048 and more bits, but at 512 and 1024 bits, it was |
| 226 |
* slower even than the standard algorithm! |
| 227 |
* |
| 228 |
* "Real" timings [linux-elf, solaris-sparcv9-gcc configurations] |
| 229 |
* should be obtained when the new Montgomery reduction code |
| 230 |
* has been integrated into OpenSSL.) |
| 231 |
*/ |
| 232 |
|
| 233 |
#define MONT_MUL_MOD |
| 234 |
#define MONT_EXP_WORD |
| 235 |
#define RECP_MUL_MOD |
| 236 |
|
| 237 |
#ifdef MONT_MUL_MOD |
| 238 |
/* |
| 239 |
* I have finally been able to take out this pre-condition of the top bit |
| 240 |
* being set. It was caused by an error in BN_div with negatives. There |
| 241 |
* was also another problem when for a^b%m a >= m. eay 07-May-97 |
| 242 |
*/ |
| 243 |
/* if ((m->d[m->top-1]&BN_TBIT) && BN_is_odd(m)) */ |
| 244 |
|
| 245 |
if (BN_is_odd(m)) { |
| 246 |
# ifdef MONT_EXP_WORD |
| 247 |
if (a->top == 1 && !a->neg |
| 248 |
&& (BN_get_flags(p, BN_FLG_CONSTTIME) == 0)) { |
| 249 |
BN_ULONG A = a->d[0]; |
| 250 |
ret = BN_mod_exp_mont_word(r, A, p, m, ctx, NULL); |
| 251 |
} else |
| 252 |
# endif |
| 253 |
ret = BN_mod_exp_mont(r, a, p, m, ctx, NULL); |
| 254 |
} else |
| 255 |
#endif |
| 256 |
#ifdef RECP_MUL_MOD |
| 257 |
{ |
| 258 |
ret = BN_mod_exp_recp(r, a, p, m, ctx); |
| 259 |
} |
| 260 |
#else |
| 261 |
{ |
| 262 |
ret = BN_mod_exp_simple(r, a, p, m, ctx); |
| 263 |
} |
| 264 |
#endif |
| 265 |
|
| 266 |
bn_check_top(r); |
| 267 |
return (ret); |
| 268 |
} |
| 269 |
|
| 270 |
int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
| 271 |
const BIGNUM *m, BN_CTX *ctx) |
| 272 |
{ |
| 273 |
int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
| 274 |
int start = 1; |
| 275 |
BIGNUM *aa; |
| 276 |
/* Table of variables obtained from 'ctx' */ |
| 277 |
BIGNUM *val[TABLE_SIZE]; |
| 278 |
BN_RECP_CTX recp; |
| 279 |
|
| 280 |
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) { |
| 281 |
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
| 282 |
BNerr(BN_F_BN_MOD_EXP_RECP, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
| 283 |
return -1; |
| 284 |
} |
| 285 |
|
| 286 |
bits = BN_num_bits(p); |
| 287 |
if (bits == 0) { |
| 288 |
/* x**0 mod 1 is still zero. */ |
| 289 |
if (BN_is_one(m)) { |
| 290 |
ret = 1; |
| 291 |
BN_zero(r); |
| 292 |
} else { |
| 293 |
ret = BN_one(r); |
| 294 |
} |
| 295 |
return ret; |
| 296 |
} |
| 297 |
|
| 298 |
BN_CTX_start(ctx); |
| 299 |
aa = BN_CTX_get(ctx); |
| 300 |
val[0] = BN_CTX_get(ctx); |
| 301 |
if (!aa || !val[0]) |
| 302 |
goto err; |
| 303 |
|
| 304 |
BN_RECP_CTX_init(&recp); |
| 305 |
if (m->neg) { |
| 306 |
/* ignore sign of 'm' */ |
| 307 |
if (!BN_copy(aa, m)) |
| 308 |
goto err; |
| 309 |
aa->neg = 0; |
| 310 |
if (BN_RECP_CTX_set(&recp, aa, ctx) <= 0) |
| 311 |
goto err; |
| 312 |
} else { |
| 313 |
if (BN_RECP_CTX_set(&recp, m, ctx) <= 0) |
| 314 |
goto err; |
| 315 |
} |
| 316 |
|
| 317 |
if (!BN_nnmod(val[0], a, m, ctx)) |
| 318 |
goto err; /* 1 */ |
| 319 |
if (BN_is_zero(val[0])) { |
| 320 |
BN_zero(r); |
| 321 |
ret = 1; |
| 322 |
goto err; |
| 323 |
} |
| 324 |
|
| 325 |
window = BN_window_bits_for_exponent_size(bits); |
| 326 |
if (window > 1) { |
| 327 |
if (!BN_mod_mul_reciprocal(aa, val[0], val[0], &recp, ctx)) |
| 328 |
goto err; /* 2 */ |
| 329 |
j = 1 << (window - 1); |
| 330 |
for (i = 1; i < j; i++) { |
| 331 |
if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
| 332 |
!BN_mod_mul_reciprocal(val[i], val[i - 1], aa, &recp, ctx)) |
| 333 |
goto err; |
| 334 |
} |
| 335 |
} |
| 336 |
|
| 337 |
start = 1; /* This is used to avoid multiplication etc |
| 338 |
* when there is only the value '1' in the |
| 339 |
* buffer. */ |
| 340 |
wvalue = 0; /* The 'value' of the window */ |
| 341 |
wstart = bits - 1; /* The top bit of the window */ |
| 342 |
wend = 0; /* The bottom bit of the window */ |
| 343 |
|
| 344 |
if (!BN_one(r)) |
| 345 |
goto err; |
| 346 |
|
| 347 |
for (;;) { |
| 348 |
if (BN_is_bit_set(p, wstart) == 0) { |
| 349 |
if (!start) |
| 350 |
if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
| 351 |
goto err; |
| 352 |
if (wstart == 0) |
| 353 |
break; |
| 354 |
wstart--; |
| 355 |
continue; |
| 356 |
} |
| 357 |
/* |
| 358 |
* We now have wstart on a 'set' bit, we now need to work out how bit |
| 359 |
* a window to do. To do this we need to scan forward until the last |
| 360 |
* set bit before the end of the window |
| 361 |
*/ |
| 362 |
j = wstart; |
| 363 |
wvalue = 1; |
| 364 |
wend = 0; |
| 365 |
for (i = 1; i < window; i++) { |
| 366 |
if (wstart - i < 0) |
| 367 |
break; |
| 368 |
if (BN_is_bit_set(p, wstart - i)) { |
| 369 |
wvalue <<= (i - wend); |
| 370 |
wvalue |= 1; |
| 371 |
wend = i; |
| 372 |
} |
| 373 |
} |
| 374 |
|
| 375 |
/* wend is the size of the current window */ |
| 376 |
j = wend + 1; |
| 377 |
/* add the 'bytes above' */ |
| 378 |
if (!start) |
| 379 |
for (i = 0; i < j; i++) { |
| 380 |
if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx)) |
| 381 |
goto err; |
| 382 |
} |
| 383 |
|
| 384 |
/* wvalue will be an odd number < 2^window */ |
| 385 |
if (!BN_mod_mul_reciprocal(r, r, val[wvalue >> 1], &recp, ctx)) |
| 386 |
goto err; |
| 387 |
|
| 388 |
/* move the 'window' down further */ |
| 389 |
wstart -= wend + 1; |
| 390 |
wvalue = 0; |
| 391 |
start = 0; |
| 392 |
if (wstart < 0) |
| 393 |
break; |
| 394 |
} |
| 395 |
ret = 1; |
| 396 |
err: |
| 397 |
BN_CTX_end(ctx); |
| 398 |
BN_RECP_CTX_free(&recp); |
| 399 |
bn_check_top(r); |
| 400 |
return (ret); |
| 401 |
} |
| 402 |
|
| 403 |
int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
| 404 |
const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
| 405 |
{ |
| 406 |
int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
| 407 |
int start = 1; |
| 408 |
BIGNUM *d, *r; |
| 409 |
const BIGNUM *aa; |
| 410 |
/* Table of variables obtained from 'ctx' */ |
| 411 |
BIGNUM *val[TABLE_SIZE]; |
| 412 |
BN_MONT_CTX *mont = NULL; |
| 413 |
|
| 414 |
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) { |
| 415 |
return BN_mod_exp_mont_consttime(rr, a, p, m, ctx, in_mont); |
| 416 |
} |
| 417 |
|
| 418 |
bn_check_top(a); |
| 419 |
bn_check_top(p); |
| 420 |
bn_check_top(m); |
| 421 |
|
| 422 |
if (!BN_is_odd(m)) { |
| 423 |
BNerr(BN_F_BN_MOD_EXP_MONT, BN_R_CALLED_WITH_EVEN_MODULUS); |
| 424 |
return (0); |
| 425 |
} |
| 426 |
bits = BN_num_bits(p); |
| 427 |
if (bits == 0) { |
| 428 |
/* x**0 mod 1 is still zero. */ |
| 429 |
if (BN_is_one(m)) { |
| 430 |
ret = 1; |
| 431 |
BN_zero(rr); |
| 432 |
} else { |
| 433 |
ret = BN_one(rr); |
| 434 |
} |
| 435 |
return ret; |
| 436 |
} |
| 437 |
|
| 438 |
BN_CTX_start(ctx); |
| 439 |
d = BN_CTX_get(ctx); |
| 440 |
r = BN_CTX_get(ctx); |
| 441 |
val[0] = BN_CTX_get(ctx); |
| 442 |
if (!d || !r || !val[0]) |
| 443 |
goto err; |
| 444 |
|
| 445 |
/* |
| 446 |
* If this is not done, things will break in the montgomery part |
| 447 |
*/ |
| 448 |
|
| 449 |
if (in_mont != NULL) |
| 450 |
mont = in_mont; |
| 451 |
else { |
| 452 |
if ((mont = BN_MONT_CTX_new()) == NULL) |
| 453 |
goto err; |
| 454 |
if (!BN_MONT_CTX_set(mont, m, ctx)) |
| 455 |
goto err; |
| 456 |
} |
| 457 |
|
| 458 |
if (a->neg || BN_ucmp(a, m) >= 0) { |
| 459 |
if (!BN_nnmod(val[0], a, m, ctx)) |
| 460 |
goto err; |
| 461 |
aa = val[0]; |
| 462 |
} else |
| 463 |
aa = a; |
| 464 |
if (BN_is_zero(aa)) { |
| 465 |
BN_zero(rr); |
| 466 |
ret = 1; |
| 467 |
goto err; |
| 468 |
} |
| 469 |
if (!BN_to_montgomery(val[0], aa, mont, ctx)) |
| 470 |
goto err; /* 1 */ |
| 471 |
|
| 472 |
window = BN_window_bits_for_exponent_size(bits); |
| 473 |
if (window > 1) { |
| 474 |
if (!BN_mod_mul_montgomery(d, val[0], val[0], mont, ctx)) |
| 475 |
goto err; /* 2 */ |
| 476 |
j = 1 << (window - 1); |
| 477 |
for (i = 1; i < j; i++) { |
| 478 |
if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
| 479 |
!BN_mod_mul_montgomery(val[i], val[i - 1], d, mont, ctx)) |
| 480 |
goto err; |
| 481 |
} |
| 482 |
} |
| 483 |
|
| 484 |
start = 1; /* This is used to avoid multiplication etc |
| 485 |
* when there is only the value '1' in the |
| 486 |
* buffer. */ |
| 487 |
wvalue = 0; /* The 'value' of the window */ |
| 488 |
wstart = bits - 1; /* The top bit of the window */ |
| 489 |
wend = 0; /* The bottom bit of the window */ |
| 490 |
|
| 491 |
#if 1 /* by Shay Gueron's suggestion */ |
| 492 |
j = m->top; /* borrow j */ |
| 493 |
if (m->d[j - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) { |
| 494 |
if (bn_wexpand(r, j) == NULL) |
| 495 |
goto err; |
| 496 |
/* 2^(top*BN_BITS2) - m */ |
| 497 |
r->d[0] = (0 - m->d[0]) & BN_MASK2; |
| 498 |
for (i = 1; i < j; i++) |
| 499 |
r->d[i] = (~m->d[i]) & BN_MASK2; |
| 500 |
r->top = j; |
| 501 |
/* |
| 502 |
* Upper words will be zero if the corresponding words of 'm' were |
| 503 |
* 0xfff[...], so decrement r->top accordingly. |
| 504 |
*/ |
| 505 |
bn_correct_top(r); |
| 506 |
} else |
| 507 |
#endif |
| 508 |
if (!BN_to_montgomery(r, BN_value_one(), mont, ctx)) |
| 509 |
goto err; |
| 510 |
for (;;) { |
| 511 |
if (BN_is_bit_set(p, wstart) == 0) { |
| 512 |
if (!start) { |
| 513 |
if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
| 514 |
goto err; |
| 515 |
} |
| 516 |
if (wstart == 0) |
| 517 |
break; |
| 518 |
wstart--; |
| 519 |
continue; |
| 520 |
} |
| 521 |
/* |
| 522 |
* We now have wstart on a 'set' bit, we now need to work out how bit |
| 523 |
* a window to do. To do this we need to scan forward until the last |
| 524 |
* set bit before the end of the window |
| 525 |
*/ |
| 526 |
j = wstart; |
| 527 |
wvalue = 1; |
| 528 |
wend = 0; |
| 529 |
for (i = 1; i < window; i++) { |
| 530 |
if (wstart - i < 0) |
| 531 |
break; |
| 532 |
if (BN_is_bit_set(p, wstart - i)) { |
| 533 |
wvalue <<= (i - wend); |
| 534 |
wvalue |= 1; |
| 535 |
wend = i; |
| 536 |
} |
| 537 |
} |
| 538 |
|
| 539 |
/* wend is the size of the current window */ |
| 540 |
j = wend + 1; |
| 541 |
/* add the 'bytes above' */ |
| 542 |
if (!start) |
| 543 |
for (i = 0; i < j; i++) { |
| 544 |
if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
| 545 |
goto err; |
| 546 |
} |
| 547 |
|
| 548 |
/* wvalue will be an odd number < 2^window */ |
| 549 |
if (!BN_mod_mul_montgomery(r, r, val[wvalue >> 1], mont, ctx)) |
| 550 |
goto err; |
| 551 |
|
| 552 |
/* move the 'window' down further */ |
| 553 |
wstart -= wend + 1; |
| 554 |
wvalue = 0; |
| 555 |
start = 0; |
| 556 |
if (wstart < 0) |
| 557 |
break; |
| 558 |
} |
| 559 |
#if defined(SPARC_T4_MONT) |
| 560 |
if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
| 561 |
j = mont->N.top; /* borrow j */ |
| 562 |
val[0]->d[0] = 1; /* borrow val[0] */ |
| 563 |
for (i = 1; i < j; i++) |
| 564 |
val[0]->d[i] = 0; |
| 565 |
val[0]->top = j; |
| 566 |
if (!BN_mod_mul_montgomery(rr, r, val[0], mont, ctx)) |
| 567 |
goto err; |
| 568 |
} else |
| 569 |
#endif |
| 570 |
if (!BN_from_montgomery(rr, r, mont, ctx)) |
| 571 |
goto err; |
| 572 |
ret = 1; |
| 573 |
err: |
| 574 |
if ((in_mont == NULL) && (mont != NULL)) |
| 575 |
BN_MONT_CTX_free(mont); |
| 576 |
BN_CTX_end(ctx); |
| 577 |
bn_check_top(rr); |
| 578 |
return (ret); |
| 579 |
} |
| 580 |
|
| 581 |
#if defined(SPARC_T4_MONT) |
| 582 |
static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos) |
| 583 |
{ |
| 584 |
BN_ULONG ret = 0; |
| 585 |
int wordpos; |
| 586 |
|
| 587 |
wordpos = bitpos / BN_BITS2; |
| 588 |
bitpos %= BN_BITS2; |
| 589 |
if (wordpos >= 0 && wordpos < a->top) { |
| 590 |
ret = a->d[wordpos] & BN_MASK2; |
| 591 |
if (bitpos) { |
| 592 |
ret >>= bitpos; |
| 593 |
if (++wordpos < a->top) |
| 594 |
ret |= a->d[wordpos] << (BN_BITS2 - bitpos); |
| 595 |
} |
| 596 |
} |
| 597 |
|
| 598 |
return ret & BN_MASK2; |
| 599 |
} |
| 600 |
#endif |
| 601 |
|
| 602 |
/* |
| 603 |
* BN_mod_exp_mont_consttime() stores the precomputed powers in a specific |
| 604 |
* layout so that accessing any of these table values shows the same access |
| 605 |
* pattern as far as cache lines are concerned. The following functions are |
| 606 |
* used to transfer a BIGNUM from/to that table. |
| 607 |
*/ |
| 608 |
|
| 609 |
static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top, |
| 610 |
unsigned char *buf, int idx, |
| 611 |
int window) |
| 612 |
{ |
| 613 |
int i, j; |
| 614 |
int width = 1 << window; |
| 615 |
BN_ULONG *table = (BN_ULONG *)buf; |
| 616 |
|
| 617 |
if (top > b->top) |
| 618 |
top = b->top; /* this works because 'buf' is explicitly |
| 619 |
* zeroed */ |
| 620 |
for (i = 0, j = idx; i < top; i++, j += width) { |
| 621 |
table[j] = b->d[i]; |
| 622 |
} |
| 623 |
|
| 624 |
return 1; |
| 625 |
} |
| 626 |
|
| 627 |
static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top, |
| 628 |
unsigned char *buf, int idx, |
| 629 |
int window) |
| 630 |
{ |
| 631 |
int i, j; |
| 632 |
int width = 1 << window; |
| 633 |
volatile BN_ULONG *table = (volatile BN_ULONG *)buf; |
| 634 |
|
| 635 |
if (bn_wexpand(b, top) == NULL) |
| 636 |
return 0; |
| 637 |
|
| 638 |
if (window <= 3) { |
| 639 |
for (i = 0; i < top; i++, table += width) { |
| 640 |
BN_ULONG acc = 0; |
| 641 |
|
| 642 |
for (j = 0; j < width; j++) { |
| 643 |
acc |= table[j] & |
| 644 |
((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1)); |
| 645 |
} |
| 646 |
|
| 647 |
b->d[i] = acc; |
| 648 |
} |
| 649 |
} else { |
| 650 |
int xstride = 1 << (window - 2); |
| 651 |
BN_ULONG y0, y1, y2, y3; |
| 652 |
|
| 653 |
i = idx >> (window - 2); /* equivalent of idx / xstride */ |
| 654 |
idx &= xstride - 1; /* equivalent of idx % xstride */ |
| 655 |
|
| 656 |
y0 = (BN_ULONG)0 - (constant_time_eq_int(i,0)&1); |
| 657 |
y1 = (BN_ULONG)0 - (constant_time_eq_int(i,1)&1); |
| 658 |
y2 = (BN_ULONG)0 - (constant_time_eq_int(i,2)&1); |
| 659 |
y3 = (BN_ULONG)0 - (constant_time_eq_int(i,3)&1); |
| 660 |
|
| 661 |
for (i = 0; i < top; i++, table += width) { |
| 662 |
BN_ULONG acc = 0; |
| 663 |
|
| 664 |
for (j = 0; j < xstride; j++) { |
| 665 |
acc |= ( (table[j + 0 * xstride] & y0) | |
| 666 |
(table[j + 1 * xstride] & y1) | |
| 667 |
(table[j + 2 * xstride] & y2) | |
| 668 |
(table[j + 3 * xstride] & y3) ) |
| 669 |
& ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1)); |
| 670 |
} |
| 671 |
|
| 672 |
b->d[i] = acc; |
| 673 |
} |
| 674 |
} |
| 675 |
|
| 676 |
b->top = top; |
| 677 |
bn_correct_top(b); |
| 678 |
return 1; |
| 679 |
} |
| 680 |
|
| 681 |
/* |
| 682 |
* Given a pointer value, compute the next address that is a cache line |
| 683 |
* multiple. |
| 684 |
*/ |
| 685 |
#define MOD_EXP_CTIME_ALIGN(x_) \ |
| 686 |
((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK)))) |
| 687 |
|
| 688 |
/* |
| 689 |
* This variant of BN_mod_exp_mont() uses fixed windows and the special |
| 690 |
* precomputation memory layout to limit data-dependency to a minimum to |
| 691 |
* protect secret exponents (cf. the hyper-threading timing attacks pointed |
| 692 |
* out by Colin Percival, |
| 693 |
* http://www.daemonology.net/hyperthreading-considered-harmful/) |
| 694 |
*/ |
| 695 |
int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p, |
| 696 |
const BIGNUM *m, BN_CTX *ctx, |
| 697 |
BN_MONT_CTX *in_mont) |
| 698 |
{ |
| 699 |
int i, bits, ret = 0, window, wvalue; |
| 700 |
int top; |
| 701 |
BN_MONT_CTX *mont = NULL; |
| 702 |
|
| 703 |
int numPowers; |
| 704 |
unsigned char *powerbufFree = NULL; |
| 705 |
int powerbufLen = 0; |
| 706 |
unsigned char *powerbuf = NULL; |
| 707 |
BIGNUM tmp, am; |
| 708 |
#if defined(SPARC_T4_MONT) |
| 709 |
unsigned int t4 = 0; |
| 710 |
#endif |
| 711 |
|
| 712 |
bn_check_top(a); |
| 713 |
bn_check_top(p); |
| 714 |
bn_check_top(m); |
| 715 |
|
| 716 |
if (!BN_is_odd(m)) { |
| 717 |
BNerr(BN_F_BN_MOD_EXP_MONT_CONSTTIME, BN_R_CALLED_WITH_EVEN_MODULUS); |
| 718 |
return (0); |
| 719 |
} |
| 720 |
|
| 721 |
top = m->top; |
| 722 |
|
| 723 |
bits = BN_num_bits(p); |
| 724 |
if (bits == 0) { |
| 725 |
/* x**0 mod 1 is still zero. */ |
| 726 |
if (BN_is_one(m)) { |
| 727 |
ret = 1; |
| 728 |
BN_zero(rr); |
| 729 |
} else { |
| 730 |
ret = BN_one(rr); |
| 731 |
} |
| 732 |
return ret; |
| 733 |
} |
| 734 |
|
| 735 |
BN_CTX_start(ctx); |
| 736 |
|
| 737 |
/* |
| 738 |
* Allocate a montgomery context if it was not supplied by the caller. If |
| 739 |
* this is not done, things will break in the montgomery part. |
| 740 |
*/ |
| 741 |
if (in_mont != NULL) |
| 742 |
mont = in_mont; |
| 743 |
else { |
| 744 |
if ((mont = BN_MONT_CTX_new()) == NULL) |
| 745 |
goto err; |
| 746 |
if (!BN_MONT_CTX_set(mont, m, ctx)) |
| 747 |
goto err; |
| 748 |
} |
| 749 |
|
| 750 |
#ifdef RSAZ_ENABLED |
| 751 |
/* |
| 752 |
* If the size of the operands allow it, perform the optimized |
| 753 |
* RSAZ exponentiation. For further information see |
| 754 |
* crypto/bn/rsaz_exp.c and accompanying assembly modules. |
| 755 |
*/ |
| 756 |
if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024) |
| 757 |
&& rsaz_avx2_eligible()) { |
| 758 |
if (NULL == bn_wexpand(rr, 16)) |
| 759 |
goto err; |
| 760 |
RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d, |
| 761 |
mont->n0[0]); |
| 762 |
rr->top = 16; |
| 763 |
rr->neg = 0; |
| 764 |
bn_correct_top(rr); |
| 765 |
ret = 1; |
| 766 |
goto err; |
| 767 |
} else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) { |
| 768 |
if (NULL == bn_wexpand(rr, 8)) |
| 769 |
goto err; |
| 770 |
RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d); |
| 771 |
rr->top = 8; |
| 772 |
rr->neg = 0; |
| 773 |
bn_correct_top(rr); |
| 774 |
ret = 1; |
| 775 |
goto err; |
| 776 |
} |
| 777 |
#endif |
| 778 |
|
| 779 |
/* Get the window size to use with size of p. */ |
| 780 |
window = BN_window_bits_for_ctime_exponent_size(bits); |
| 781 |
#if defined(SPARC_T4_MONT) |
| 782 |
if (window >= 5 && (top & 15) == 0 && top <= 64 && |
| 783 |
(OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) == |
| 784 |
(CFR_MONTMUL | CFR_MONTSQR) && (t4 = OPENSSL_sparcv9cap_P[0])) |
| 785 |
window = 5; |
| 786 |
else |
| 787 |
#endif |
| 788 |
#if defined(OPENSSL_BN_ASM_MONT5) |
| 789 |
if (window >= 5) { |
| 790 |
window = 5; /* ~5% improvement for RSA2048 sign, and even |
| 791 |
* for RSA4096 */ |
| 792 |
/* reserve space for mont->N.d[] copy */ |
| 793 |
powerbufLen += top * sizeof(mont->N.d[0]); |
| 794 |
} |
| 795 |
#endif |
| 796 |
(void)0; |
| 797 |
|
| 798 |
/* |
| 799 |
* Allocate a buffer large enough to hold all of the pre-computed powers |
| 800 |
* of am, am itself and tmp. |
| 801 |
*/ |
| 802 |
numPowers = 1 << window; |
| 803 |
powerbufLen += sizeof(m->d[0]) * (top * numPowers + |
| 804 |
((2 * top) > |
| 805 |
numPowers ? (2 * top) : numPowers)); |
| 806 |
#ifdef alloca |
| 807 |
if (powerbufLen < 3072) |
| 808 |
powerbufFree = |
| 809 |
alloca(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH); |
| 810 |
else |
| 811 |
#endif |
| 812 |
if ((powerbufFree = |
| 813 |
(unsigned char *)OPENSSL_malloc(powerbufLen + |
| 814 |
MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH)) |
| 815 |
== NULL) |
| 816 |
goto err; |
| 817 |
|
| 818 |
powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree); |
| 819 |
memset(powerbuf, 0, powerbufLen); |
| 820 |
|
| 821 |
#ifdef alloca |
| 822 |
if (powerbufLen < 3072) |
| 823 |
powerbufFree = NULL; |
| 824 |
#endif |
| 825 |
|
| 826 |
/* lay down tmp and am right after powers table */ |
| 827 |
tmp.d = (BN_ULONG *)(powerbuf + sizeof(m->d[0]) * top * numPowers); |
| 828 |
am.d = tmp.d + top; |
| 829 |
tmp.top = am.top = 0; |
| 830 |
tmp.dmax = am.dmax = top; |
| 831 |
tmp.neg = am.neg = 0; |
| 832 |
tmp.flags = am.flags = BN_FLG_STATIC_DATA; |
| 833 |
|
| 834 |
/* prepare a^0 in Montgomery domain */ |
| 835 |
#if 1 /* by Shay Gueron's suggestion */ |
| 836 |
if (m->d[top - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) { |
| 837 |
/* 2^(top*BN_BITS2) - m */ |
| 838 |
tmp.d[0] = (0 - m->d[0]) & BN_MASK2; |
| 839 |
for (i = 1; i < top; i++) |
| 840 |
tmp.d[i] = (~m->d[i]) & BN_MASK2; |
| 841 |
tmp.top = top; |
| 842 |
} else |
| 843 |
#endif |
| 844 |
if (!BN_to_montgomery(&tmp, BN_value_one(), mont, ctx)) |
| 845 |
goto err; |
| 846 |
|
| 847 |
/* prepare a^1 in Montgomery domain */ |
| 848 |
if (a->neg || BN_ucmp(a, m) >= 0) { |
| 849 |
if (!BN_mod(&am, a, m, ctx)) |
| 850 |
goto err; |
| 851 |
if (!BN_to_montgomery(&am, &am, mont, ctx)) |
| 852 |
goto err; |
| 853 |
} else if (!BN_to_montgomery(&am, a, mont, ctx)) |
| 854 |
goto err; |
| 855 |
|
| 856 |
#if defined(SPARC_T4_MONT) |
| 857 |
if (t4) { |
| 858 |
typedef int (*bn_pwr5_mont_f) (BN_ULONG *tp, const BN_ULONG *np, |
| 859 |
const BN_ULONG *n0, const void *table, |
| 860 |
int power, int bits); |
| 861 |
int bn_pwr5_mont_t4_8(BN_ULONG *tp, const BN_ULONG *np, |
| 862 |
const BN_ULONG *n0, const void *table, |
| 863 |
int power, int bits); |
| 864 |
int bn_pwr5_mont_t4_16(BN_ULONG *tp, const BN_ULONG *np, |
| 865 |
const BN_ULONG *n0, const void *table, |
| 866 |
int power, int bits); |
| 867 |
int bn_pwr5_mont_t4_24(BN_ULONG *tp, const BN_ULONG *np, |
| 868 |
const BN_ULONG *n0, const void *table, |
| 869 |
int power, int bits); |
| 870 |
int bn_pwr5_mont_t4_32(BN_ULONG *tp, const BN_ULONG *np, |
| 871 |
const BN_ULONG *n0, const void *table, |
| 872 |
int power, int bits); |
| 873 |
static const bn_pwr5_mont_f pwr5_funcs[4] = { |
| 874 |
bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16, |
| 875 |
bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32 |
| 876 |
}; |
| 877 |
bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top / 16 - 1]; |
| 878 |
|
| 879 |
typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap, |
| 880 |
const void *bp, const BN_ULONG *np, |
| 881 |
const BN_ULONG *n0); |
| 882 |
int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap, const void *bp, |
| 883 |
const BN_ULONG *np, const BN_ULONG *n0); |
| 884 |
int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap, |
| 885 |
const void *bp, const BN_ULONG *np, |
| 886 |
const BN_ULONG *n0); |
| 887 |
int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap, |
| 888 |
const void *bp, const BN_ULONG *np, |
| 889 |
const BN_ULONG *n0); |
| 890 |
int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap, |
| 891 |
const void *bp, const BN_ULONG *np, |
| 892 |
const BN_ULONG *n0); |
| 893 |
static const bn_mul_mont_f mul_funcs[4] = { |
| 894 |
bn_mul_mont_t4_8, bn_mul_mont_t4_16, |
| 895 |
bn_mul_mont_t4_24, bn_mul_mont_t4_32 |
| 896 |
}; |
| 897 |
bn_mul_mont_f mul_worker = mul_funcs[top / 16 - 1]; |
| 898 |
|
| 899 |
void bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap, |
| 900 |
const void *bp, const BN_ULONG *np, |
| 901 |
const BN_ULONG *n0, int num); |
| 902 |
void bn_mul_mont_t4(BN_ULONG *rp, const BN_ULONG *ap, |
| 903 |
const void *bp, const BN_ULONG *np, |
| 904 |
const BN_ULONG *n0, int num); |
| 905 |
void bn_mul_mont_gather5_t4(BN_ULONG *rp, const BN_ULONG *ap, |
| 906 |
const void *table, const BN_ULONG *np, |
| 907 |
const BN_ULONG *n0, int num, int power); |
| 908 |
void bn_flip_n_scatter5_t4(const BN_ULONG *inp, size_t num, |
| 909 |
void *table, size_t power); |
| 910 |
void bn_gather5_t4(BN_ULONG *out, size_t num, |
| 911 |
void *table, size_t power); |
| 912 |
void bn_flip_t4(BN_ULONG *dst, BN_ULONG *src, size_t num); |
| 913 |
|
| 914 |
BN_ULONG *np = mont->N.d, *n0 = mont->n0; |
| 915 |
int stride = 5 * (6 - (top / 16 - 1)); /* multiple of 5, but less |
| 916 |
* than 32 */ |
| 917 |
|
| 918 |
/* |
| 919 |
* BN_to_montgomery can contaminate words above .top [in |
| 920 |
* BN_DEBUG[_DEBUG] build]... |
| 921 |
*/ |
| 922 |
for (i = am.top; i < top; i++) |
| 923 |
am.d[i] = 0; |
| 924 |
for (i = tmp.top; i < top; i++) |
| 925 |
tmp.d[i] = 0; |
| 926 |
|
| 927 |
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 0); |
| 928 |
bn_flip_n_scatter5_t4(am.d, top, powerbuf, 1); |
| 929 |
if (!(*mul_worker) (tmp.d, am.d, am.d, np, n0) && |
| 930 |
!(*mul_worker) (tmp.d, am.d, am.d, np, n0)) |
| 931 |
bn_mul_mont_vis3(tmp.d, am.d, am.d, np, n0, top); |
| 932 |
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 2); |
| 933 |
|
| 934 |
for (i = 3; i < 32; i++) { |
| 935 |
/* Calculate a^i = a^(i-1) * a */ |
| 936 |
if (!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0) && |
| 937 |
!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0)) |
| 938 |
bn_mul_mont_vis3(tmp.d, tmp.d, am.d, np, n0, top); |
| 939 |
bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, i); |
| 940 |
} |
| 941 |
|
| 942 |
/* switch to 64-bit domain */ |
| 943 |
np = alloca(top * sizeof(BN_ULONG)); |
| 944 |
top /= 2; |
| 945 |
bn_flip_t4(np, mont->N.d, top); |
| 946 |
|
| 947 |
bits--; |
| 948 |
for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--) |
| 949 |
wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
| 950 |
bn_gather5_t4(tmp.d, top, powerbuf, wvalue); |
| 951 |
|
| 952 |
/* |
| 953 |
* Scan the exponent one window at a time starting from the most |
| 954 |
* significant bits. |
| 955 |
*/ |
| 956 |
while (bits >= 0) { |
| 957 |
if (bits < stride) |
| 958 |
stride = bits + 1; |
| 959 |
bits -= stride; |
| 960 |
wvalue = bn_get_bits(p, bits + 1); |
| 961 |
|
| 962 |
if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
| 963 |
continue; |
| 964 |
/* retry once and fall back */ |
| 965 |
if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride)) |
| 966 |
continue; |
| 967 |
|
| 968 |
bits += stride - 5; |
| 969 |
wvalue >>= stride - 5; |
| 970 |
wvalue &= 31; |
| 971 |
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 972 |
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 973 |
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 974 |
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 975 |
bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 976 |
bn_mul_mont_gather5_t4(tmp.d, tmp.d, powerbuf, np, n0, top, |
| 977 |
wvalue); |
| 978 |
} |
| 979 |
|
| 980 |
bn_flip_t4(tmp.d, tmp.d, top); |
| 981 |
top *= 2; |
| 982 |
/* back to 32-bit domain */ |
| 983 |
tmp.top = top; |
| 984 |
bn_correct_top(&tmp); |
| 985 |
OPENSSL_cleanse(np, top * sizeof(BN_ULONG)); |
| 986 |
} else |
| 987 |
#endif |
| 988 |
#if defined(OPENSSL_BN_ASM_MONT5) |
| 989 |
if (window == 5 && top > 1) { |
| 990 |
/* |
| 991 |
* This optimization uses ideas from http://eprint.iacr.org/2011/239, |
| 992 |
* specifically optimization of cache-timing attack countermeasures |
| 993 |
* and pre-computation optimization. |
| 994 |
*/ |
| 995 |
|
| 996 |
/* |
| 997 |
* Dedicated window==4 case improves 512-bit RSA sign by ~15%, but as |
| 998 |
* 512-bit RSA is hardly relevant, we omit it to spare size... |
| 999 |
*/ |
| 1000 |
void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap, |
| 1001 |
const void *table, const BN_ULONG *np, |
| 1002 |
const BN_ULONG *n0, int num, int power); |
| 1003 |
void bn_scatter5(const BN_ULONG *inp, size_t num, |
| 1004 |
void *table, size_t power); |
| 1005 |
void bn_gather5(BN_ULONG *out, size_t num, void *table, size_t power); |
| 1006 |
void bn_power5(BN_ULONG *rp, const BN_ULONG *ap, |
| 1007 |
const void *table, const BN_ULONG *np, |
| 1008 |
const BN_ULONG *n0, int num, int power); |
| 1009 |
int bn_get_bits5(const BN_ULONG *ap, int off); |
| 1010 |
int bn_from_montgomery(BN_ULONG *rp, const BN_ULONG *ap, |
| 1011 |
const BN_ULONG *not_used, const BN_ULONG *np, |
| 1012 |
const BN_ULONG *n0, int num); |
| 1013 |
|
| 1014 |
BN_ULONG *n0 = mont->n0, *np; |
| 1015 |
|
| 1016 |
/* |
| 1017 |
* BN_to_montgomery can contaminate words above .top [in |
| 1018 |
* BN_DEBUG[_DEBUG] build]... |
| 1019 |
*/ |
| 1020 |
for (i = am.top; i < top; i++) |
| 1021 |
am.d[i] = 0; |
| 1022 |
for (i = tmp.top; i < top; i++) |
| 1023 |
tmp.d[i] = 0; |
| 1024 |
|
| 1025 |
/* |
| 1026 |
* copy mont->N.d[] to improve cache locality |
| 1027 |
*/ |
| 1028 |
for (np = am.d + top, i = 0; i < top; i++) |
| 1029 |
np[i] = mont->N.d[i]; |
| 1030 |
|
| 1031 |
bn_scatter5(tmp.d, top, powerbuf, 0); |
| 1032 |
bn_scatter5(am.d, am.top, powerbuf, 1); |
| 1033 |
bn_mul_mont(tmp.d, am.d, am.d, np, n0, top); |
| 1034 |
bn_scatter5(tmp.d, top, powerbuf, 2); |
| 1035 |
|
| 1036 |
# if 0 |
| 1037 |
for (i = 3; i < 32; i++) { |
| 1038 |
/* Calculate a^i = a^(i-1) * a */ |
| 1039 |
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
| 1040 |
bn_scatter5(tmp.d, top, powerbuf, i); |
| 1041 |
} |
| 1042 |
# else |
| 1043 |
/* same as above, but uses squaring for 1/2 of operations */ |
| 1044 |
for (i = 4; i < 32; i *= 2) { |
| 1045 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1046 |
bn_scatter5(tmp.d, top, powerbuf, i); |
| 1047 |
} |
| 1048 |
for (i = 3; i < 8; i += 2) { |
| 1049 |
int j; |
| 1050 |
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
| 1051 |
bn_scatter5(tmp.d, top, powerbuf, i); |
| 1052 |
for (j = 2 * i; j < 32; j *= 2) { |
| 1053 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1054 |
bn_scatter5(tmp.d, top, powerbuf, j); |
| 1055 |
} |
| 1056 |
} |
| 1057 |
for (; i < 16; i += 2) { |
| 1058 |
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
| 1059 |
bn_scatter5(tmp.d, top, powerbuf, i); |
| 1060 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1061 |
bn_scatter5(tmp.d, top, powerbuf, 2 * i); |
| 1062 |
} |
| 1063 |
for (; i < 32; i += 2) { |
| 1064 |
bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1); |
| 1065 |
bn_scatter5(tmp.d, top, powerbuf, i); |
| 1066 |
} |
| 1067 |
# endif |
| 1068 |
bits--; |
| 1069 |
for (wvalue = 0, i = bits % 5; i >= 0; i--, bits--) |
| 1070 |
wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
| 1071 |
bn_gather5(tmp.d, top, powerbuf, wvalue); |
| 1072 |
|
| 1073 |
/* |
| 1074 |
* Scan the exponent one window at a time starting from the most |
| 1075 |
* significant bits. |
| 1076 |
*/ |
| 1077 |
if (top & 7) |
| 1078 |
while (bits >= 0) { |
| 1079 |
for (wvalue = 0, i = 0; i < 5; i++, bits--) |
| 1080 |
wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
| 1081 |
|
| 1082 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1083 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1084 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1085 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1086 |
bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top); |
| 1087 |
bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top, |
| 1088 |
wvalue); |
| 1089 |
} else { |
| 1090 |
while (bits >= 0) { |
| 1091 |
wvalue = bn_get_bits5(p->d, bits - 4); |
| 1092 |
bits -= 5; |
| 1093 |
bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top, wvalue); |
| 1094 |
} |
| 1095 |
} |
| 1096 |
|
| 1097 |
ret = bn_from_montgomery(tmp.d, tmp.d, NULL, np, n0, top); |
| 1098 |
tmp.top = top; |
| 1099 |
bn_correct_top(&tmp); |
| 1100 |
if (ret) { |
| 1101 |
if (!BN_copy(rr, &tmp)) |
| 1102 |
ret = 0; |
| 1103 |
goto err; /* non-zero ret means it's not error */ |
| 1104 |
} |
| 1105 |
} else |
| 1106 |
#endif |
| 1107 |
{ |
| 1108 |
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, window)) |
| 1109 |
goto err; |
| 1110 |
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, window)) |
| 1111 |
goto err; |
| 1112 |
|
| 1113 |
/* |
| 1114 |
* If the window size is greater than 1, then calculate |
| 1115 |
* val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even |
| 1116 |
* powers could instead be computed as (a^(i/2))^2 to use the slight |
| 1117 |
* performance advantage of sqr over mul). |
| 1118 |
*/ |
| 1119 |
if (window > 1) { |
| 1120 |
if (!BN_mod_mul_montgomery(&tmp, &am, &am, mont, ctx)) |
| 1121 |
goto err; |
| 1122 |
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2, |
| 1123 |
window)) |
| 1124 |
goto err; |
| 1125 |
for (i = 3; i < numPowers; i++) { |
| 1126 |
/* Calculate a^i = a^(i-1) * a */ |
| 1127 |
if (!BN_mod_mul_montgomery(&tmp, &am, &tmp, mont, ctx)) |
| 1128 |
goto err; |
| 1129 |
if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, i, |
| 1130 |
window)) |
| 1131 |
goto err; |
| 1132 |
} |
| 1133 |
} |
| 1134 |
|
| 1135 |
bits--; |
| 1136 |
for (wvalue = 0, i = bits % window; i >= 0; i--, bits--) |
| 1137 |
wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
| 1138 |
if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue, |
| 1139 |
window)) |
| 1140 |
goto err; |
| 1141 |
|
| 1142 |
/* |
| 1143 |
* Scan the exponent one window at a time starting from the most |
| 1144 |
* significant bits. |
| 1145 |
*/ |
| 1146 |
while (bits >= 0) { |
| 1147 |
wvalue = 0; /* The 'value' of the window */ |
| 1148 |
|
| 1149 |
/* Scan the window, squaring the result as we go */ |
| 1150 |
for (i = 0; i < window; i++, bits--) { |
| 1151 |
if (!BN_mod_mul_montgomery(&tmp, &tmp, &tmp, mont, ctx)) |
| 1152 |
goto err; |
| 1153 |
wvalue = (wvalue << 1) + BN_is_bit_set(p, bits); |
| 1154 |
} |
| 1155 |
|
| 1156 |
/* |
| 1157 |
* Fetch the appropriate pre-computed value from the pre-buf |
| 1158 |
*/ |
| 1159 |
if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue, |
| 1160 |
window)) |
| 1161 |
goto err; |
| 1162 |
|
| 1163 |
/* Multiply the result into the intermediate result */ |
| 1164 |
if (!BN_mod_mul_montgomery(&tmp, &tmp, &am, mont, ctx)) |
| 1165 |
goto err; |
| 1166 |
} |
| 1167 |
} |
| 1168 |
|
| 1169 |
/* Convert the final result from montgomery to standard format */ |
| 1170 |
#if defined(SPARC_T4_MONT) |
| 1171 |
if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) { |
| 1172 |
am.d[0] = 1; /* borrow am */ |
| 1173 |
for (i = 1; i < top; i++) |
| 1174 |
am.d[i] = 0; |
| 1175 |
if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx)) |
| 1176 |
goto err; |
| 1177 |
} else |
| 1178 |
#endif |
| 1179 |
if (!BN_from_montgomery(rr, &tmp, mont, ctx)) |
| 1180 |
goto err; |
| 1181 |
ret = 1; |
| 1182 |
err: |
| 1183 |
if ((in_mont == NULL) && (mont != NULL)) |
| 1184 |
BN_MONT_CTX_free(mont); |
| 1185 |
if (powerbuf != NULL) { |
| 1186 |
OPENSSL_cleanse(powerbuf, powerbufLen); |
| 1187 |
if (powerbufFree) |
| 1188 |
OPENSSL_free(powerbufFree); |
| 1189 |
} |
| 1190 |
BN_CTX_end(ctx); |
| 1191 |
return (ret); |
| 1192 |
} |
| 1193 |
|
| 1194 |
int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONG a, const BIGNUM *p, |
| 1195 |
const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont) |
| 1196 |
{ |
| 1197 |
BN_MONT_CTX *mont = NULL; |
| 1198 |
int b, bits, ret = 0; |
| 1199 |
int r_is_one; |
| 1200 |
BN_ULONG w, next_w; |
| 1201 |
BIGNUM *d, *r, *t; |
| 1202 |
BIGNUM *swap_tmp; |
| 1203 |
#define BN_MOD_MUL_WORD(r, w, m) \ |
| 1204 |
(BN_mul_word(r, (w)) && \ |
| 1205 |
(/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \ |
| 1206 |
(BN_mod(t, r, m, ctx) && (swap_tmp = r, r = t, t = swap_tmp, 1)))) |
| 1207 |
/* |
| 1208 |
* BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is |
| 1209 |
* probably more overhead than always using BN_mod (which uses BN_copy if |
| 1210 |
* a similar test returns true). |
| 1211 |
*/ |
| 1212 |
/* |
| 1213 |
* We can use BN_mod and do not need BN_nnmod because our accumulator is |
| 1214 |
* never negative (the result of BN_mod does not depend on the sign of |
| 1215 |
* the modulus). |
| 1216 |
*/ |
| 1217 |
#define BN_TO_MONTGOMERY_WORD(r, w, mont) \ |
| 1218 |
(BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx)) |
| 1219 |
|
| 1220 |
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) { |
| 1221 |
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
| 1222 |
BNerr(BN_F_BN_MOD_EXP_MONT_WORD, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
| 1223 |
return -1; |
| 1224 |
} |
| 1225 |
|
| 1226 |
bn_check_top(p); |
| 1227 |
bn_check_top(m); |
| 1228 |
|
| 1229 |
if (!BN_is_odd(m)) { |
| 1230 |
BNerr(BN_F_BN_MOD_EXP_MONT_WORD, BN_R_CALLED_WITH_EVEN_MODULUS); |
| 1231 |
return (0); |
| 1232 |
} |
| 1233 |
if (m->top == 1) |
| 1234 |
a %= m->d[0]; /* make sure that 'a' is reduced */ |
| 1235 |
|
| 1236 |
bits = BN_num_bits(p); |
| 1237 |
if (bits == 0) { |
| 1238 |
/* x**0 mod 1 is still zero. */ |
| 1239 |
if (BN_is_one(m)) { |
| 1240 |
ret = 1; |
| 1241 |
BN_zero(rr); |
| 1242 |
} else { |
| 1243 |
ret = BN_one(rr); |
| 1244 |
} |
| 1245 |
return ret; |
| 1246 |
} |
| 1247 |
if (a == 0) { |
| 1248 |
BN_zero(rr); |
| 1249 |
ret = 1; |
| 1250 |
return ret; |
| 1251 |
} |
| 1252 |
|
| 1253 |
BN_CTX_start(ctx); |
| 1254 |
d = BN_CTX_get(ctx); |
| 1255 |
r = BN_CTX_get(ctx); |
| 1256 |
t = BN_CTX_get(ctx); |
| 1257 |
if (d == NULL || r == NULL || t == NULL) |
| 1258 |
goto err; |
| 1259 |
|
| 1260 |
if (in_mont != NULL) |
| 1261 |
mont = in_mont; |
| 1262 |
else { |
| 1263 |
if ((mont = BN_MONT_CTX_new()) == NULL) |
| 1264 |
goto err; |
| 1265 |
if (!BN_MONT_CTX_set(mont, m, ctx)) |
| 1266 |
goto err; |
| 1267 |
} |
| 1268 |
|
| 1269 |
r_is_one = 1; /* except for Montgomery factor */ |
| 1270 |
|
| 1271 |
/* bits-1 >= 0 */ |
| 1272 |
|
| 1273 |
/* The result is accumulated in the product r*w. */ |
| 1274 |
w = a; /* bit 'bits-1' of 'p' is always set */ |
| 1275 |
for (b = bits - 2; b >= 0; b--) { |
| 1276 |
/* First, square r*w. */ |
| 1277 |
next_w = w * w; |
| 1278 |
if ((next_w / w) != w) { /* overflow */ |
| 1279 |
if (r_is_one) { |
| 1280 |
if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
| 1281 |
goto err; |
| 1282 |
r_is_one = 0; |
| 1283 |
} else { |
| 1284 |
if (!BN_MOD_MUL_WORD(r, w, m)) |
| 1285 |
goto err; |
| 1286 |
} |
| 1287 |
next_w = 1; |
| 1288 |
} |
| 1289 |
w = next_w; |
| 1290 |
if (!r_is_one) { |
| 1291 |
if (!BN_mod_mul_montgomery(r, r, r, mont, ctx)) |
| 1292 |
goto err; |
| 1293 |
} |
| 1294 |
|
| 1295 |
/* Second, multiply r*w by 'a' if exponent bit is set. */ |
| 1296 |
if (BN_is_bit_set(p, b)) { |
| 1297 |
next_w = w * a; |
| 1298 |
if ((next_w / a) != w) { /* overflow */ |
| 1299 |
if (r_is_one) { |
| 1300 |
if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
| 1301 |
goto err; |
| 1302 |
r_is_one = 0; |
| 1303 |
} else { |
| 1304 |
if (!BN_MOD_MUL_WORD(r, w, m)) |
| 1305 |
goto err; |
| 1306 |
} |
| 1307 |
next_w = a; |
| 1308 |
} |
| 1309 |
w = next_w; |
| 1310 |
} |
| 1311 |
} |
| 1312 |
|
| 1313 |
/* Finally, set r:=r*w. */ |
| 1314 |
if (w != 1) { |
| 1315 |
if (r_is_one) { |
| 1316 |
if (!BN_TO_MONTGOMERY_WORD(r, w, mont)) |
| 1317 |
goto err; |
| 1318 |
r_is_one = 0; |
| 1319 |
} else { |
| 1320 |
if (!BN_MOD_MUL_WORD(r, w, m)) |
| 1321 |
goto err; |
| 1322 |
} |
| 1323 |
} |
| 1324 |
|
| 1325 |
if (r_is_one) { /* can happen only if a == 1 */ |
| 1326 |
if (!BN_one(rr)) |
| 1327 |
goto err; |
| 1328 |
} else { |
| 1329 |
if (!BN_from_montgomery(rr, r, mont, ctx)) |
| 1330 |
goto err; |
| 1331 |
} |
| 1332 |
ret = 1; |
| 1333 |
err: |
| 1334 |
if ((in_mont == NULL) && (mont != NULL)) |
| 1335 |
BN_MONT_CTX_free(mont); |
| 1336 |
BN_CTX_end(ctx); |
| 1337 |
bn_check_top(rr); |
| 1338 |
return (ret); |
| 1339 |
} |
| 1340 |
|
| 1341 |
/* The old fallback, simple version :-) */ |
| 1342 |
int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, |
| 1343 |
const BIGNUM *m, BN_CTX *ctx) |
| 1344 |
{ |
| 1345 |
int i, j, bits, ret = 0, wstart, wend, window, wvalue; |
| 1346 |
int start = 1; |
| 1347 |
BIGNUM *d; |
| 1348 |
/* Table of variables obtained from 'ctx' */ |
| 1349 |
BIGNUM *val[TABLE_SIZE]; |
| 1350 |
|
| 1351 |
if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0) { |
| 1352 |
/* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */ |
| 1353 |
BNerr(BN_F_BN_MOD_EXP_SIMPLE, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED); |
| 1354 |
return -1; |
| 1355 |
} |
| 1356 |
|
| 1357 |
bits = BN_num_bits(p); |
| 1358 |
if (bits == 0) { |
| 1359 |
/* x**0 mod 1 is still zero. */ |
| 1360 |
if (BN_is_one(m)) { |
| 1361 |
ret = 1; |
| 1362 |
BN_zero(r); |
| 1363 |
} else { |
| 1364 |
ret = BN_one(r); |
| 1365 |
} |
| 1366 |
return ret; |
| 1367 |
} |
| 1368 |
|
| 1369 |
BN_CTX_start(ctx); |
| 1370 |
d = BN_CTX_get(ctx); |
| 1371 |
val[0] = BN_CTX_get(ctx); |
| 1372 |
if (!d || !val[0]) |
| 1373 |
goto err; |
| 1374 |
|
| 1375 |
if (!BN_nnmod(val[0], a, m, ctx)) |
| 1376 |
goto err; /* 1 */ |
| 1377 |
if (BN_is_zero(val[0])) { |
| 1378 |
BN_zero(r); |
| 1379 |
ret = 1; |
| 1380 |
goto err; |
| 1381 |
} |
| 1382 |
|
| 1383 |
window = BN_window_bits_for_exponent_size(bits); |
| 1384 |
if (window > 1) { |
| 1385 |
if (!BN_mod_mul(d, val[0], val[0], m, ctx)) |
| 1386 |
goto err; /* 2 */ |
| 1387 |
j = 1 << (window - 1); |
| 1388 |
for (i = 1; i < j; i++) { |
| 1389 |
if (((val[i] = BN_CTX_get(ctx)) == NULL) || |
| 1390 |
!BN_mod_mul(val[i], val[i - 1], d, m, ctx)) |
| 1391 |
goto err; |
| 1392 |
} |
| 1393 |
} |
| 1394 |
|
| 1395 |
start = 1; /* This is used to avoid multiplication etc |
| 1396 |
* when there is only the value '1' in the |
| 1397 |
* buffer. */ |
| 1398 |
wvalue = 0; /* The 'value' of the window */ |
| 1399 |
wstart = bits - 1; /* The top bit of the window */ |
| 1400 |
wend = 0; /* The bottom bit of the window */ |
| 1401 |
|
| 1402 |
if (!BN_one(r)) |
| 1403 |
goto err; |
| 1404 |
|
| 1405 |
for (;;) { |
| 1406 |
if (BN_is_bit_set(p, wstart) == 0) { |
| 1407 |
if (!start) |
| 1408 |
if (!BN_mod_mul(r, r, r, m, ctx)) |
| 1409 |
goto err; |
| 1410 |
if (wstart == 0) |
| 1411 |
break; |
| 1412 |
wstart--; |
| 1413 |
continue; |
| 1414 |
} |
| 1415 |
/* |
| 1416 |
* We now have wstart on a 'set' bit, we now need to work out how bit |
| 1417 |
* a window to do. To do this we need to scan forward until the last |
| 1418 |
* set bit before the end of the window |
| 1419 |
*/ |
| 1420 |
j = wstart; |
| 1421 |
wvalue = 1; |
| 1422 |
wend = 0; |
| 1423 |
for (i = 1; i < window; i++) { |
| 1424 |
if (wstart - i < 0) |
| 1425 |
break; |
| 1426 |
if (BN_is_bit_set(p, wstart - i)) { |
| 1427 |
wvalue <<= (i - wend); |
| 1428 |
wvalue |= 1; |
| 1429 |
wend = i; |
| 1430 |
} |
| 1431 |
} |
| 1432 |
|
| 1433 |
/* wend is the size of the current window */ |
| 1434 |
j = wend + 1; |
| 1435 |
/* add the 'bytes above' */ |
| 1436 |
if (!start) |
| 1437 |
for (i = 0; i < j; i++) { |
| 1438 |
if (!BN_mod_mul(r, r, r, m, ctx)) |
| 1439 |
goto err; |
| 1440 |
} |
| 1441 |
|
| 1442 |
/* wvalue will be an odd number < 2^window */ |
| 1443 |
if (!BN_mod_mul(r, r, val[wvalue >> 1], m, ctx)) |
| 1444 |
goto err; |
| 1445 |
|
| 1446 |
/* move the 'window' down further */ |
| 1447 |
wstart -= wend + 1; |
| 1448 |
wvalue = 0; |
| 1449 |
start = 0; |
| 1450 |
if (wstart < 0) |
| 1451 |
break; |
| 1452 |
} |
| 1453 |
ret = 1; |
| 1454 |
err: |
| 1455 |
BN_CTX_end(ctx); |
| 1456 |
bn_check_top(r); |
| 1457 |
return (ret); |
| 1458 |
} |